diff options
Diffstat (limited to 'sys')
77 files changed, 1123 insertions, 7925 deletions
diff --git a/sys/alpha/alpha/db_interface.c b/sys/alpha/alpha/db_interface.c index d6b1297..e94a490 100644 --- a/sys/alpha/alpha/db_interface.c +++ b/sys/alpha/alpha/db_interface.c @@ -60,13 +60,13 @@ #include <sys/cons.h> #include <sys/ktr.h> #include <sys/mutex.h> +#include <sys/smp.h> #include <vm/vm.h> #include <machine/db_machdep.h> #include <machine/pal.h> #include <machine/prom.h> -#include <machine/smp.h> #include <alpha/alpha/db_instruction.h> @@ -573,45 +573,46 @@ db_branch_taken(ins, pc, regs) DB_SHOW_COMMAND(pcpu, db_show_pcpu) { struct globaldata *gd; +#ifdef SMP int id; if (have_addr) id = ((addr >> 4) % 16) * 10 + (addr % 16); else id = PCPU_GET(cpuid); - SLIST_FOREACH(gd, &cpuhead, gd_allcpu) { - if (gd->gd_cpuid == id) - break; - } - if (gd == NULL) + gd = globaldata_find(id); + if (gd == NULL) { db_printf("CPU %d not found\n", id); - else { - db_printf("cpuid = %d\n", gd->gd_cpuid); - db_printf("ipis = %lx\n", gd->gd_pending_ipis); - db_printf("next ASN = %d\n", gd->gd_next_asn); - db_printf("curproc = "); - if (gd->gd_curproc != NULL) - db_printf("%p: pid %d \"%s\"\n", gd->gd_curproc, - gd->gd_curproc->p_pid, gd->gd_curproc->p_comm); - else - db_printf("none\n"); - db_printf("curpcb = %p\n", gd->gd_curpcb); - db_printf("fpcurproc = "); - if (gd->gd_fpcurproc != NULL) - db_printf("%p: pid %d \"%s\"\n", gd->gd_fpcurproc, - gd->gd_fpcurproc->p_pid, gd->gd_fpcurproc->p_comm); - else - db_printf("none\n"); - db_printf("idleproc = "); - if (gd->gd_idleproc != NULL) - db_printf("%p: pid %d \"%s\"\n", gd->gd_idleproc, - gd->gd_idleproc->p_pid, gd->gd_idleproc->p_comm); - else - db_printf("none\n"); + return; + } +#else + gd = GLOBALP; +#endif + db_printf("cpuid = %d\n", gd->gd_cpuid); + db_printf("ipis = %lx\n", gd->gd_pending_ipis); + db_printf("next ASN = %d\n", gd->gd_next_asn); + db_printf("curproc = "); + if (gd->gd_curproc != NULL) + db_printf("%p: pid %d \"%s\"\n", gd->gd_curproc, + gd->gd_curproc->p_pid, gd->gd_curproc->p_comm); + else + db_printf("none\n"); + db_printf("curpcb = %p\n", gd->gd_curpcb); + db_printf("fpcurproc = "); + if (gd->gd_fpcurproc != NULL) + db_printf("%p: pid %d \"%s\"\n", gd->gd_fpcurproc, + gd->gd_fpcurproc->p_pid, gd->gd_fpcurproc->p_comm); + else + db_printf("none\n"); + db_printf("idleproc = "); + if (gd->gd_idleproc != NULL) + db_printf("%p: pid %d \"%s\"\n", gd->gd_idleproc, + gd->gd_idleproc->p_pid, gd->gd_idleproc->p_comm); + else + db_printf("none\n"); #ifdef WITNESS - db_printf("spin locks held:\n"); - witness_list_locks(&gd->gd_spinlocks); + db_printf("spin locks held:\n"); + witness_list_locks(&gd->gd_spinlocks); #endif - } } diff --git a/sys/alpha/alpha/interrupt.c b/sys/alpha/alpha/interrupt.c index 346b21c..b28b53d 100644 --- a/sys/alpha/alpha/interrupt.c +++ b/sys/alpha/alpha/interrupt.c @@ -52,6 +52,7 @@ #include <sys/kthread.h> #include <sys/ktr.h> #include <sys/mutex.h> +#include <sys/smp.h> #include <sys/unistd.h> #include <machine/reg.h> @@ -83,7 +84,7 @@ void (*perf_irq)(unsigned long, struct trapframe *) = dummy_perf; static u_int schedclk2; -static driver_intr_t alpha_clock_interrupt; +static void alpha_clock_interrupt(struct trapframe *framep); void interrupt(a0, a1, a2, framep) @@ -453,7 +454,7 @@ alpha_dispatch_intr(void *frame, unsigned long vector) } static void -alpha_clock_interrupt(void *framep) +alpha_clock_interrupt(struct trapframe *framep) { cnt.v_intr++; @@ -462,10 +463,26 @@ alpha_clock_interrupt(void *framep) #else intrcnt[INTRCNT_CLOCK]++; #endif - if (platform.clockintr){ - (*platform.clockintr)((struct trapframe *)framep); - /* divide hz (1024) by 8 to get stathz (128) */ - if((++schedclk2 & 0x7) == 0) - statclock((struct clockframe *)framep); + if (platform.clockintr) { +#ifdef SMP + /* + * Only one processor drives the actual timer. + */ + if (PCPU_GET(cpuid) == boot_cpu_id) { +#endif + (*platform.clockintr)(framep); + /* divide hz (1024) by 8 to get stathz (128) */ + if ((++schedclk2 & 0x7) == 0) + statclock((struct clockframe *)framep); +#ifdef SMP + } else { + mtx_lock_spin(&sched_lock); + hardclock_process(curproc, TRAPF_USERMODE(framep)); + if ((schedclk2 & 0x7) == 0) + statclock_process(curproc, TRAPF_PC(framep), + TRAPF_USERMODE(framep)); + mtx_unlock_spin(&sched_lock); + } +#endif } } diff --git a/sys/alpha/alpha/machdep.c b/sys/alpha/alpha/machdep.c index f623669..177fe8e 100644 --- a/sys/alpha/alpha/machdep.c +++ b/sys/alpha/alpha/machdep.c @@ -112,6 +112,7 @@ #include <sys/vmmeter.h> #include <sys/msgbuf.h> #include <sys/exec.h> +#include <sys/smp.h> #include <sys/sysctl.h> #include <sys/uio.h> #include <sys/linker.h> @@ -130,7 +131,6 @@ #include <machine/reg.h> #include <machine/fpu.h> #include <machine/pal.h> -#include <machine/smp.h> #include <machine/globaldata.h> #include <machine/cpuconf.h> #include <machine/bootinfo.h> @@ -152,8 +152,6 @@ struct platform platform; alpha_chipset_t chipset; struct bootinfo_kernel bootinfo; -struct cpuhead cpuhead; - struct mtx sched_lock; struct mtx Giant; @@ -415,14 +413,6 @@ again: vm_pager_bufferinit(); EVENTHANDLER_REGISTER(shutdown_final, alpha_srm_shutdown, 0, SHUTDOWN_PRI_LAST); - -#ifdef SMP - /* - * OK, enough kmem_alloc/malloc state should be up, lets get on with it! - */ - mp_start(); /* fire up the secondaries */ - mp_announce(); -#endif /* SMP */ } /* @@ -1002,12 +992,6 @@ alpha_init(pfn, ptb, bim, bip, biv) */ globalp->gd_idlepcb.apcb_ptbr = proc0.p_addr->u_pcb.pcb_hw.apcb_ptbr; - /* - * Record all cpus in a list. - */ - SLIST_INIT(&cpuhead); - SLIST_INSERT_HEAD(&cpuhead, GLOBALP, gd_allcpu); - /* Setup curproc so that mutexes work */ PCPU_SET(curproc, &proc0); PCPU_SET(spinlocks, NULL); diff --git a/sys/alpha/alpha/mp_machdep.c b/sys/alpha/alpha/mp_machdep.c index 3715c2f..3024102 100644 --- a/sys/alpha/alpha/mp_machdep.c +++ b/sys/alpha/alpha/mp_machdep.c @@ -35,6 +35,7 @@ #include <sys/malloc.h> #include <sys/mutex.h> #include <sys/kernel.h> +#include <sys/smp.h> #include <sys/sysctl.h> #include <vm/vm.h> @@ -43,55 +44,18 @@ #include <sys/user.h> #include <sys/dkstat.h> -#include <machine/smp.h> #include <machine/atomic.h> #include <machine/globaldata.h> #include <machine/pmap.h> #include <machine/rpb.h> #include <machine/clock.h> -#define CHECKSTATE_USER 0 -#define CHECKSTATE_SYS 1 -#define CHECKSTATE_INTR 2 - -volatile u_int stopped_cpus; -volatile u_int started_cpus; -volatile u_int checkstate_probed_cpus; -volatile u_int checkstate_need_ast; - /* Set to 1 once we're ready to let the APs out of the pen. */ static volatile int aps_ready = 0; static struct mtx ap_boot_mtx; -struct proc* checkstate_curproc[MAXCPU]; -int checkstate_cpustate[MAXCPU]; -u_long checkstate_pc[MAXCPU]; -volatile u_int resched_cpus; -void (*cpustop_restartfunc) __P((void)); -int mp_ncpus; - -volatile int smp_started; -int boot_cpu_id; -u_int32_t all_cpus; - -static struct globaldata *cpuid_to_globaldata[MAXCPU]; - -int smp_active = 0; /* are the APs allowed to run? */ -SYSCTL_INT(_machdep, OID_AUTO, smp_active, CTLFLAG_RW, &smp_active, 0, ""); - -static int smp_cpus = 1; /* how many cpu's running */ -SYSCTL_INT(_machdep, OID_AUTO, smp_cpus, CTLFLAG_RD, &smp_cpus, 0, ""); - -/* Enable forwarding of a signal to a process running on a different CPU */ -static int forward_signal_enabled = 1; -SYSCTL_INT(_machdep, OID_AUTO, forward_signal_enabled, CTLFLAG_RW, - &forward_signal_enabled, 0, ""); - -/* Enable forwarding of roundrobin to all other cpus */ -static int forward_roundrobin_enabled = 1; -SYSCTL_INT(_machdep, OID_AUTO, forward_roundrobin_enabled, CTLFLAG_RW, - &forward_roundrobin_enabled, 0, ""); +u_int boot_cpu_id; /* * Communicate with a console running on a secondary processor. @@ -225,7 +189,6 @@ smp_init_secondary(void) if (smp_cpus == mp_ncpus) { smp_started = 1; smp_active = 1; - ipi_all(0); } mtx_unlock_spin(&ap_boot_mtx); @@ -250,7 +213,7 @@ static int smp_start_secondary(int cpuid) { struct pcs *cpu = LOCATE_PCS(hwrpb, cpuid); - struct pcs *bootcpu = LOCATE_PCS(hwrpb, hwrpb->rpb_primary_cpu_id); + struct pcs *bootcpu = LOCATE_PCS(hwrpb, boot_cpu_id); struct alpha_pcb *pcb = (struct alpha_pcb *) cpu->pcs_hwpcb; struct globaldata *globaldata; int i; @@ -272,7 +235,6 @@ smp_start_secondary(int cpuid) } globaldata_init(globaldata, cpuid, sz); - SLIST_INSERT_HEAD(&cpuhead, globaldata, gd_allcpu); /* * Copy the idle pcb and setup the address to start executing. @@ -329,48 +291,53 @@ smp_start_secondary(int cpuid) return 1; } -/* - * Register a struct globaldata. - */ -void -globaldata_register(struct globaldata *globaldata) -{ - cpuid_to_globaldata[globaldata->gd_cpuid] = globaldata; -} +/* Other stuff */ -struct globaldata * -globaldata_find(int cpuid) +int +cpu_mp_probe(void) { - return cpuid_to_globaldata[cpuid]; -} - -/* Other stuff */ + struct pcs *pcsp; + int i, cpus; -/* lock around the MP rendezvous */ -static struct mtx smp_rv_mtx; + /* XXX: Need to check for valid platforms here. */ -static void -init_locks(void) -{ - mtx_init(&smp_rv_mtx, "smp rendezvous", MTX_SPIN); - mtx_init(&ap_boot_mtx, "ap boot", MTX_SPIN); + /* Make sure we have at least one secondary CPU. */ + cpus = 0; + for (i = 0; i < hwrpb->rpb_pcs_cnt; i++) { + if (i == PCPU_GET(cpuid)) + continue; + pcsp = (struct pcs *)((char *)hwrpb + hwrpb->rpb_pcs_off + + (i * hwrpb->rpb_pcs_size)); + if ((pcsp->pcs_flags & PCS_PP) == 0) + continue; + if ((pcsp->pcs_flags & PCS_PA) == 0) + continue; + if ((pcsp->pcs_flags & PCS_PV) == 0) + continue; + if (i > MAXCPU) + continue; + cpus++; + } + return (cpus); } void -mp_start() +cpu_mp_start() { int i; - int cpuid = PCPU_GET(cpuid); - init_locks(); + mtx_init(&ap_boot_mtx, "ap boot", MTX_SPIN); mp_ncpus = 1; - all_cpus = 1 << cpuid; + boot_cpu_id = PCPU_GET(cpuid); + KASSERT(boot_cpu_id == hwrpb->rpb_primary_cpu_id, + ("mp_start() called on non-primary CPU")); + all_cpus = 1 << boot_cpu_id; for (i = 0; i < hwrpb->rpb_pcs_cnt; i++) { struct pcs *pcsp; - if (i == cpuid) + if (i == boot_cpu_id) continue; pcsp = (struct pcs *)((char *)hwrpb + hwrpb->rpb_pcs_off + (i * hwrpb->rpb_pcs_size)); @@ -397,10 +364,10 @@ mp_start() all_cpus |= 1 << i; mp_ncpus++; } - PCPU_SET(other_cpus, all_cpus & ~(1<<cpuid)); + PCPU_SET(other_cpus, all_cpus & ~(1 << boot_cpu_id)); for (i = 0; i < hwrpb->rpb_pcs_cnt; i++) { - if (i == cpuid) + if (i == boot_cpu_id) continue; if (all_cpus & 1 << i) smp_start_secondary(i); @@ -408,561 +375,8 @@ mp_start() } void -mp_announce() -{ - - printf("SMP System Detected: %d usable CPUs\n", mp_ncpus); -} - -void -smp_invltlb() -{ -} - -#if 0 -#define GD_TO_INDEX(pc, prof) \ - ((int)(((u_quad_t)((pc) - (prof)->pr_off) * \ - (u_quad_t)((prof)->pr_scale)) >> 16) & ~1) - -static void -addupc_intr_forwarded(struct proc *p, int id, int *astmap) -{ - int i; - struct uprof *prof; - u_long pc; - - pc = checkstate_pc[id]; - prof = &p->p_stats->p_prof; - if (pc >= prof->pr_off && - (i = GD_TO_INDEX(pc, prof)) < prof->pr_size) { - if ((p->p_sflag & PS_OWEUPC) == 0) { - prof->pr_addr = pc; - prof->pr_ticks = 1; - p->p_sflag |= PS_OWEUPC; - } - *astmap |= (1 << id); - } -} - -static void -forwarded_statclock(int id, int pscnt, int *astmap) -{ - struct pstats *pstats; - long rss; - struct rusage *ru; - struct vmspace *vm; - int cpustate; - struct proc *p; -#ifdef GPROF - register struct gmonparam *g; - int i; -#endif - - p = checkstate_curproc[id]; - cpustate = checkstate_cpustate[id]; - - /* XXX */ - if (p->p_ithd) - cpustate = CHECKSTATE_INTR; - else if (p == cpuid_to_globaldata[id]->gd_idleproc) - cpustate = CHECKSTATE_SYS; - - switch (cpustate) { - case CHECKSTATE_USER: - if (p->p_sflag & PS_PROFIL) - addupc_intr_forwarded(p, id, astmap); - if (pscnt > 1) - return; - p->p_uticks++; - if (p->p_nice > NZERO) - cp_time[CP_NICE]++; - else - cp_time[CP_USER]++; - break; - case CHECKSTATE_SYS: -#ifdef GPROF - /* - * Kernel statistics are just like addupc_intr, only easier. - */ - g = &_gmonparam; - if (g->state == GMON_PROF_ON) { - i = checkstate_pc[id] - g->lowpc; - if (i < g->textsize) { - i /= HISTFRACTION * sizeof(*g->kcount); - g->kcount[i]++; - } - } -#endif - if (pscnt > 1) - return; - - p->p_sticks++; - if (p == cpuid_to_globaldata[id]->gd_idleproc) - cp_time[CP_IDLE]++; - else - cp_time[CP_SYS]++; - break; - case CHECKSTATE_INTR: - default: -#ifdef GPROF - /* - * Kernel statistics are just like addupc_intr, only easier. - */ - g = &_gmonparam; - if (g->state == GMON_PROF_ON) { - i = checkstate_pc[id] - g->lowpc; - if (i < g->textsize) { - i /= HISTFRACTION * sizeof(*g->kcount); - g->kcount[i]++; - } - } -#endif - if (pscnt > 1) - return; - KASSERT(p != NULL, ("NULL process in interrupt state")); - p->p_iticks++; - cp_time[CP_INTR]++; - } - - schedclock(p); - - /* Update resource usage integrals and maximums. */ - if ((pstats = p->p_stats) != NULL && - (ru = &pstats->p_ru) != NULL && - (vm = p->p_vmspace) != NULL) { - ru->ru_ixrss += pgtok(vm->vm_tsize); - ru->ru_idrss += pgtok(vm->vm_dsize); - ru->ru_isrss += pgtok(vm->vm_ssize); - rss = pgtok(vmspace_resident_count(vm)); - if (ru->ru_maxrss < rss) - ru->ru_maxrss = rss; - } -} - -void -forward_statclock(int pscnt) -{ - int map; - int id; - int i; - - /* Kludge. We don't yet have separate locks for the interrupts - * and the kernel. This means that we cannot let the other processors - * handle complex interrupts while inhibiting them from entering - * the kernel in a non-interrupt context. - * - * What we can do, without changing the locking mechanisms yet, - * is letting the other processors handle a very simple interrupt - * (wich determines the processor states), and do the main - * work ourself. - */ - - CTR1(KTR_SMP, "forward_statclock(%d)", pscnt); - - if (!smp_started || cold || panicstr) - return; - - /* Step 1: Probe state (user, cpu, interrupt, spinlock, idle ) */ - - mtx_lock_spin(&smp_rv_mtx); - map = PCPU_GET(other_cpus) & ~stopped_cpus ; - checkstate_probed_cpus = 0; - if (map != 0) - ipi_selected(map, IPI_CHECKSTATE); - - i = 0; - while (checkstate_probed_cpus != map) { - alpha_mb(); - /* spin */ - i++; - if (i == 100000) { -#ifdef DIAGNOSTIC - printf("forward_statclock: checkstate %x\n", - checkstate_probed_cpus); -#endif - break; - } - } - mtx_unlock_spin(&smp_rv_mtx); - - /* - * Step 2: walk through other processors processes, update ticks and - * profiling info. - */ - - map = 0; - /* XXX: wrong, should walk bits in all_cpus */ - for (id = 0; id < mp_ncpus; id++) { - if (id == PCPU_GET(cpuid)) - continue; - if (((1 << id) & checkstate_probed_cpus) == 0) - continue; - forwarded_statclock(id, pscnt, &map); - } - if (map != 0) { - mtx_lock_spin(&smp_rv_mtx); - checkstate_need_ast |= map; - ipi_selected(map, IPI_AST); - i = 0; - while ((checkstate_need_ast & map) != 0) { - alpha_mb(); - /* spin */ - i++; - if (i > 100000) { -#ifdef DIAGNOSTIC - printf("forward_statclock: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } - mtx_unlock_spin(&smp_rv_mtx); - } -} - -void -forward_hardclock(int pscnt) -{ - int map; - int id; - struct proc *p; - struct pstats *pstats; - int i; - - /* Kludge. We don't yet have separate locks for the interrupts - * and the kernel. This means that we cannot let the other processors - * handle complex interrupts while inhibiting them from entering - * the kernel in a non-interrupt context. - * - * What we can do, without changing the locking mechanisms yet, - * is letting the other processors handle a very simple interrupt - * (wich determines the processor states), and do the main - * work ourself. - */ - - CTR1(KTR_SMP, "forward_hardclock(%d)", pscnt); - - if (!smp_started || cold || panicstr) - return; - - /* Step 1: Probe state (user, cpu, interrupt, spinlock, idle) */ - - mtx_lock_spin(&smp_rv_mtx); - map = PCPU_GET(other_cpus) & ~stopped_cpus ; - checkstate_probed_cpus = 0; - if (map != 0) - ipi_selected(map, IPI_CHECKSTATE); - - i = 0; - while (checkstate_probed_cpus != map) { - alpha_mb(); - /* spin */ - i++; - if (i == 100000) { -#ifdef DIAGNOSTIC - printf("forward_hardclock: checkstate %x\n", - checkstate_probed_cpus); -#endif - breakpoint(); - break; - } - } - mtx_unlock_spin(&smp_rv_mtx); - - /* - * Step 2: walk through other processors processes, update virtual - * timer and profiling timer. If stathz == 0, also update ticks and - * profiling info. - */ - - map = 0; - /* XXX: wrong, should walk bits in all_cpus */ - for (id = 0; id < mp_ncpus; id++) { - if (id == PCPU_GET(cpuid)) - continue; - if (((1 << id) & checkstate_probed_cpus) == 0) - continue; - p = checkstate_curproc[id]; - if (p) { - pstats = p->p_stats; - if (checkstate_cpustate[id] == CHECKSTATE_USER && - timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) && - itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) { - p->p_sflag |= PS_ALRMPEND; - map |= (1 << id); - } - if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) && - itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) { - p->p_sflag |= PS_PROFPEND; - map |= (1 << id); - } - } - if (stathz == 0) { - forwarded_statclock( id, pscnt, &map); - } - } - if (map != 0) { - mtx_lock_spin(&smp_rv_mtx); - checkstate_need_ast |= map; - ipi_selected(map, IPI_AST); - i = 0; - while ((checkstate_need_ast & map) != 0) { - alpha_mb(); - /* spin */ - i++; - if (i > 100000) { -#ifdef DIAGNOSTIC - printf("forward_hardclock: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } - mtx_unlock_spin(&smp_rv_mtx); - } -} -#endif /* 0 */ - -void -forward_signal(struct proc *p) -{ - int map; - int id; - int i; - - /* Kludge. We don't yet have separate locks for the interrupts - * and the kernel. This means that we cannot let the other processors - * handle complex interrupts while inhibiting them from entering - * the kernel in a non-interrupt context. - * - * What we can do, without changing the locking mechanisms yet, - * is letting the other processors handle a very simple interrupt - * (wich determines the processor states), and do the main - * work ourself. - */ - - CTR1(KTR_SMP, "forward_signal(%p)", p); - - if (!smp_started || cold || panicstr) - return; - if (!forward_signal_enabled) - return; - mtx_lock_spin(&sched_lock); - while (1) { - if (p->p_stat != SRUN) { - mtx_unlock_spin(&sched_lock); - return; - } - id = p->p_oncpu; - mtx_unlock_spin(&sched_lock); - if (id == 0xff) - return; - map = (1<<id); - checkstate_need_ast |= map; - ipi_selected(map, IPI_AST); - i = 0; - while ((checkstate_need_ast & map) != 0) { - alpha_mb(); - /* spin */ - i++; - if (i > 100000) { -#if DIAGNOSTIC - printf("forward_signal: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } - mtx_lock_spin(&sched_lock); - if (id == p->p_oncpu) { - mtx_unlock_spin(&sched_lock); - return; - } - } -} - -void -forward_roundrobin(void) -{ - u_int map; - int i; - - CTR0(KTR_SMP, "forward_roundrobin()"); - - if (!smp_started || cold || panicstr) - return; - if (!forward_roundrobin_enabled) - return; - resched_cpus |= PCPU_GET(other_cpus); - map = PCPU_GET(other_cpus) & ~stopped_cpus ; - ipi_selected(map, IPI_AST); - i = 0; - while ((checkstate_need_ast & map) != 0) { - alpha_mb(); - /* spin */ - i++; - if (i > 100000) { -#if DIAGNOSTIC - printf("forward_roundrobin: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } -} - -/* - * When called the executing CPU will send an IPI to all other CPUs - * requesting that they halt execution. - * - * Usually (but not necessarily) called with 'other_cpus' as its arg. - * - * - Signals all CPUs in map to stop. - * - Waits for each to stop. - * - * Returns: - * -1: error - * 0: NA - * 1: ok - * - * XXX FIXME: this is not MP-safe, needs a lock to prevent multiple CPUs - * from executing at same time. - */ -int -stop_cpus(u_int map) -{ - int i; - - if (!smp_started) - return 0; - - CTR1(KTR_SMP, "stop_cpus(%x)", map); - - /* send the stop IPI to all CPUs in map */ - ipi_selected(map, IPI_STOP); - - i = 0; - while ((stopped_cpus & map) != map) { - /* spin */ - i++; -#ifdef DIAGNOSTIC - if (i == 100000) { - printf("timeout stopping cpus\n"); - break; - } -#endif - alpha_mb(); - } - - return 1; -} - - -/* - * Called by a CPU to restart stopped CPUs. - * - * Usually (but not necessarily) called with 'stopped_cpus' as its arg. - * - * - Signals all CPUs in map to restart. - * - Waits for each to restart. - * - * Returns: - * -1: error - * 0: NA - * 1: ok - */ -int -restart_cpus(u_int map) -{ - if (!smp_started) - return 0; - - CTR1(KTR_SMP, "restart_cpus(%x)", map); - - started_cpus = map; /* signal other cpus to restart */ - alpha_mb(); - - while ((stopped_cpus & map) != 0) /* wait for each to clear its bit */ - alpha_mb(); - - return 1; -} - -/* - * All-CPU rendezvous. CPUs are signalled, all execute the setup function - * (if specified), rendezvous, execute the action function (if specified), - * rendezvous again, execute the teardown function (if specified), and then - * resume. - * - * Note that the supplied external functions _must_ be reentrant and aware - * that they are running in parallel and in an unknown lock context. - */ -static void (*smp_rv_setup_func)(void *arg); -static void (*smp_rv_action_func)(void *arg); -static void (*smp_rv_teardown_func)(void *arg); -static void *smp_rv_func_arg; -static volatile int smp_rv_waiters[2]; - -void -smp_rendezvous_action(void) +cpu_mp_announce() { - /* setup function */ - if (smp_rv_setup_func != NULL) - smp_rv_setup_func(smp_rv_func_arg); - /* spin on entry rendezvous */ - atomic_add_int(&smp_rv_waiters[0], 1); - while (smp_rv_waiters[0] < mp_ncpus) - alpha_mb(); - /* action function */ - if (smp_rv_action_func != NULL) - smp_rv_action_func(smp_rv_func_arg); - /* spin on exit rendezvous */ - atomic_add_int(&smp_rv_waiters[1], 1); - while (smp_rv_waiters[1] < mp_ncpus) - alpha_mb(); - /* teardown function */ - if (smp_rv_teardown_func != NULL) - smp_rv_teardown_func(smp_rv_func_arg); -} - -void -smp_rendezvous(void (* setup_func)(void *), - void (* action_func)(void *), - void (* teardown_func)(void *), - void *arg) -{ - - if (!smp_started) { - if (setup_func != NULL) - setup_func(arg); - if (action_func != NULL) - action_func(arg); - if (teardown_func != NULL) - teardown_func(arg); - return; - } - - /* obtain rendezvous lock */ - mtx_lock_spin(&smp_rv_mtx); - - /* set static function pointers */ - smp_rv_setup_func = setup_func; - smp_rv_action_func = action_func; - smp_rv_teardown_func = teardown_func; - smp_rv_func_arg = arg; - smp_rv_waiters[0] = 0; - smp_rv_waiters[1] = 0; - - /* signal other processors, which will enter the IPI with interrupts off */ - ipi_all_but_self(IPI_RENDEZVOUS); - - /* call executor function */ - CTR2(KTR_SMP, "smp_rv: calling action = %p, arg = %p\n", action_func, - arg); - smp_rendezvous_action(); - - /* release lock */ - mtx_unlock_spin(&smp_rv_mtx); } /* @@ -979,7 +393,7 @@ ipi_selected(u_int32_t cpus, u_int64_t ipi) int cpuid = ffs(cpus) - 1; cpus &= ~(1 << cpuid); - globaldata = cpuid_to_globaldata[cpuid]; + globaldata = globaldata_find(cpuid); if (globaldata) { atomic_set_64(&globaldata->gd_pending_ipis, ipi); alpha_mb(); @@ -1048,25 +462,6 @@ smp_handle_ipi(struct trapframe *frame) case IPI_AST: CTR0(KTR_SMP, "IPI_AST"); - atomic_clear_int(&checkstate_need_ast, cpumask); - mtx_lock_spin(&sched_lock); - curproc->p_sflag |= PS_ASTPENDING; - mtx_unlock_spin(&sched_lock); - break; - - case IPI_CHECKSTATE: - CTR0(KTR_SMP, "IPI_CHECKSTATE"); - if (frame->tf_regs[FRAME_PS] & ALPHA_PSL_USERMODE) - checkstate_cpustate[PCPU_GET(cpuid)] = - CHECKSTATE_USER; - else if (curproc->p_intr_nesting_level == 1) - checkstate_cpustate[PCPU_GET(cpuid)] = - CHECKSTATE_SYS; - else - checkstate_cpustate[PCPU_GET(cpuid)] = - CHECKSTATE_INTR; - checkstate_curproc[PCPU_GET(cpuid)] = curproc; - atomic_set_int(&checkstate_probed_cpus, cpumask); break; case IPI_STOP: @@ -1085,7 +480,7 @@ smp_handle_ipi(struct trapframe *frame) * requests to provide PALcode to secondaries and to start up new * secondaries that are added to the system on the fly. */ - if (PCPU_GET(cpuid) == hwrpb->rpb_primary_cpu_id) { + if (PCPU_GET(cpuid) == boot_cpu_id) { u_int cpuid; u_int64_t txrdy; #ifdef DIAGNOSTIC diff --git a/sys/alpha/alpha/pmap.c b/sys/alpha/alpha/pmap.c index d7fe3df..f70bd00 100644 --- a/sys/alpha/alpha/pmap.c +++ b/sys/alpha/alpha/pmap.c @@ -154,6 +154,7 @@ #include <sys/msgbuf.h> #include <sys/vmmeter.h> #include <sys/mman.h> +#include <sys/smp.h> #include <sys/sx.h> #include <vm/vm.h> @@ -172,7 +173,6 @@ #include <machine/md_var.h> #include <machine/rpb.h> -#include <machine/smp.h> #ifndef PMAP_SHPGPERPROC #define PMAP_SHPGPERPROC 200 diff --git a/sys/alpha/alpha/trap.c b/sys/alpha/alpha/trap.c index 228c50b..7238e7a 100644 --- a/sys/alpha/alpha/trap.c +++ b/sys/alpha/alpha/trap.c @@ -42,6 +42,7 @@ #include <sys/exec.h> #include <sys/lock.h> #include <sys/mutex.h> +#include <sys/smp.h> #include <sys/vmmeter.h> #include <sys/sysent.h> #include <sys/syscall.h> @@ -59,7 +60,6 @@ #include <machine/reg.h> #include <machine/pal.h> #include <machine/fpu.h> -#include <machine/smp.h> #ifdef KTRACE #include <sys/uio.h> @@ -150,7 +150,7 @@ userret(p, frame, oticks) mtx_lock_spin(&sched_lock); p->p_pri.pri_level = p->p_pri.pri_user; - if (resched_wanted()) { + if (resched_wanted(p)) { /* * Since we are curproc, a clock interrupt could * change our priority without changing run queues @@ -877,7 +877,7 @@ ast(framep) * acquiring and releasing mutexes in assembly is not fun. */ mtx_lock_spin(&sched_lock); - if (!(astpending(p) || resched_wanted())) { + if (!(astpending(p) || resched_wanted(p))) { mtx_unlock_spin(&sched_lock); return; } diff --git a/sys/alpha/include/cpu.h b/sys/alpha/include/cpu.h index d8ff1cc..f3bd8d2 100644 --- a/sys/alpha/include/cpu.h +++ b/sys/alpha/include/cpu.h @@ -68,7 +68,6 @@ struct clockframe { #define CLKF_USERMODE(framep) TRAPF_USERMODE(&(framep)->cf_tf) #define CLKF_PC(framep) TRAPF_PC(&(framep)->cf_tf) -#define CLKF_INTR(framep) (curproc->p_intr_nesting_level >= 2) /* * Arrange to handle pending profiling ticks before returning to user mode. diff --git a/sys/alpha/include/globaldata.h b/sys/alpha/include/globaldata.h index 15bd8c41a..4d6d02f 100644 --- a/sys/alpha/include/globaldata.h +++ b/sys/alpha/include/globaldata.h @@ -66,14 +66,7 @@ struct globaldata { #endif }; -SLIST_HEAD(cpuhead, globaldata); -extern struct cpuhead cpuhead; - void globaldata_init(struct globaldata *pcpu, int cpuid, size_t sz); -struct globaldata *globaldata_find(int cpuid); -#ifdef SMP -void globaldata_register(struct globaldata *pcpu); -#endif #endif /* _KERNEL */ diff --git a/sys/alpha/include/ipl.h b/sys/alpha/include/ipl.h index 357dc7a..9b94bf6 100644 --- a/sys/alpha/include/ipl.h +++ b/sys/alpha/include/ipl.h @@ -30,23 +30,4 @@ #define _MACHINE_IPL_H_ -#include <machine/cpu.h> /* for pal inlines */ - -#ifdef SMP -/* - * Interprocessor interrupts for SMP. - */ -#define IPI_INVLTLB 0x0001 -#define IPI_RENDEZVOUS 0x0002 -#define IPI_AST 0x0004 -#define IPI_CHECKSTATE 0x0008 -#define IPI_STOP 0x0010 - -void ipi_selected(u_int32_t cpus, u_int64_t ipi); -void ipi_all(u_int64_t ipi); -void ipi_all_but_self(u_int64_t ipi); -void ipi_self(u_int64_t ipi); -void smp_handle_ipi(struct trapframe *frame); -#endif - #endif /* !_MACHINE_IPL_H_ */ diff --git a/sys/alpha/include/pcpu.h b/sys/alpha/include/pcpu.h index 15bd8c41a..4d6d02f 100644 --- a/sys/alpha/include/pcpu.h +++ b/sys/alpha/include/pcpu.h @@ -66,14 +66,7 @@ struct globaldata { #endif }; -SLIST_HEAD(cpuhead, globaldata); -extern struct cpuhead cpuhead; - void globaldata_init(struct globaldata *pcpu, int cpuid, size_t sz); -struct globaldata *globaldata_find(int cpuid); -#ifdef SMP -void globaldata_register(struct globaldata *pcpu); -#endif #endif /* _KERNEL */ diff --git a/sys/alpha/include/smp.h b/sys/alpha/include/smp.h index 2c208c8..05af7f4 100644 --- a/sys/alpha/include/smp.h +++ b/sys/alpha/include/smp.h @@ -15,41 +15,24 @@ #ifdef _KERNEL -#include <sys/ipl.h> -#include <sys/ktr.h> -#include <sys/mutex.h> +/* + * Interprocessor interrupts for SMP. + */ +#define IPI_INVLTLB 0x0001 +#define IPI_RENDEZVOUS 0x0002 +#define IPI_AST 0x0004 +#define IPI_CHECKSTATE 0x0008 +#define IPI_STOP 0x0010 #ifndef LOCORE -#define BETTER_CLOCK /* unconditional on alpha */ - -/* global data in mp_machdep.c */ -extern volatile u_int checkstate_probed_cpus; -extern volatile u_int checkstate_need_ast; -extern volatile u_int resched_cpus; -extern void (*cpustop_restartfunc) __P((void)); - -extern int smp_active; -extern int mp_ncpus; -extern u_int all_cpus; -extern u_int started_cpus; -extern u_int stopped_cpus; +extern u_int boot_cpu_id; -/* functions in mp_machdep.c */ -void mp_start(void); -void mp_announce(void); -void smp_invltlb(void); -void forward_statclock(int pscnt); -void forward_hardclock(int pscnt); -void forward_signal(struct proc *); -void forward_roundrobin(void); -int stop_cpus(u_int); -int restart_cpus(u_int); -void smp_rendezvous_action(void); -void smp_rendezvous(void (*)(void *), - void (*)(void *), - void (*)(void *), - void *arg); +void ipi_selected(u_int cpus, u_int64_t ipi); +void ipi_all(u_int64_t ipi); +void ipi_all_but_self(u_int64_t ipi); +void ipi_self(u_int64_t ipi); +void smp_handle_ipi(struct trapframe *frame); void smp_init_secondary(void); #endif /* !LOCORE */ diff --git a/sys/amd64/amd64/amd64_mem.c b/sys/amd64/amd64/amd64_mem.c index 7c3ae75..2d687af 100644 --- a/sys/amd64/amd64/amd64_mem.c +++ b/sys/amd64/amd64/amd64_mem.c @@ -31,14 +31,11 @@ #include <sys/systm.h> #include <sys/malloc.h> #include <sys/memrange.h> +#include <sys/smp.h> #include <machine/md_var.h> #include <machine/specialreg.h> -#ifdef SMP -#include <machine/smp.h> -#endif - /* * i686 memory range operations * diff --git a/sys/amd64/amd64/apic_vector.S b/sys/amd64/amd64/apic_vector.S index 3f0521f..5c68f81 100644 --- a/sys/amd64/amd64/apic_vector.S +++ b/sys/amd64/amd64/apic_vector.S @@ -182,7 +182,6 @@ Xspuriousint: iret - /* * Handle TLB shootdowns. */ @@ -211,71 +210,61 @@ Xinvltlb: popl %eax iret - /* - * Executed by a CPU when it receives an Xcpucheckstate IPI from another CPU, - * - * - Stores current cpu state in checkstate_cpustate[cpuid] - * 0 == user, 1 == sys, 2 == intr - * - Stores current process in checkstate_curproc[cpuid] - * - * - Signals its receipt by setting bit cpuid in checkstate_probed_cpus. - * - * stack: 0->ds, 4->fs, 8->ebx, 12->eax, 16->eip, 20->cs, 24->eflags + * Forward hardclock to another CPU. Pushes a trapframe and calls + * forwarded_hardclock(). */ - .text SUPERALIGN_TEXT - .globl Xcpucheckstate - .globl checkstate_cpustate - .globl checkstate_curproc - .globl checkstate_pc -Xcpucheckstate: - pushl %eax - pushl %ebx - pushl %ds /* save current data segment */ - pushl %fs - - movl $KDSEL, %eax - mov %ax, %ds /* use KERNEL data segment */ + .globl Xhardclock +Xhardclock: + PUSH_FRAME + movl $KDSEL, %eax /* reload with kernel's data segment */ + mov %ax, %ds + mov %ax, %es movl $KPSEL, %eax mov %ax, %fs movl $0, lapic+LA_EOI /* End Of Interrupt to APIC */ - movl $0, %ebx - movl 20(%esp), %eax - andl $3, %eax - cmpl $3, %eax - je 1f - testl $PSL_VM, 24(%esp) - jne 1f - incl %ebx /* system or interrupt */ -1: - movl PCPU(CPUID), %eax - movl %ebx, checkstate_cpustate(,%eax,4) - movl PCPU(CURPROC), %ebx - movl %ebx, checkstate_curproc(,%eax,4) - - movl 16(%esp), %ebx - movl %ebx, checkstate_pc(,%eax,4) + movl PCPU(CURPROC),%ebx + incl P_INTR_NESTING_LEVEL(%ebx) + call forwarded_hardclock + decl P_INTR_NESTING_LEVEL(%ebx) + MEXITCOUNT + jmp doreti - lock /* checkstate_probed_cpus |= (1<<id) */ - btsl %eax, checkstate_probed_cpus +/* + * Forward statclock to another CPU. Pushes a trapframe and calls + * forwarded_statclock(). + */ + .text + SUPERALIGN_TEXT + .globl Xstatclock +Xstatclock: + PUSH_FRAME + movl $KDSEL, %eax /* reload with kernel's data segment */ + mov %ax, %ds + mov %ax, %es + movl $KPSEL, %eax + mov %ax, %fs - popl %fs - popl %ds /* restore previous data segment */ - popl %ebx - popl %eax - iret + movl $0, lapic+LA_EOI /* End Of Interrupt to APIC */ + FAKE_MCOUNT(13*4(%esp)) + movl PCPU(CURPROC),%ebx + incl P_INTR_NESTING_LEVEL(%ebx) + call forwarded_statclock + decl P_INTR_NESTING_LEVEL(%ebx) + MEXITCOUNT + jmp doreti /* * Executed by a CPU when it receives an Xcpuast IPI from another CPU, * - * - Signals its receipt by clearing bit cpuid in checkstate_need_ast. - * - * - We need a better method of triggering asts on other cpus. + * The other CPU has already executed aston() or need_resched() on our + * current process, so we simply need to ack the interrupt and return + * via doreti to run ast(). */ .text @@ -289,40 +278,12 @@ Xcpuast: movl $KPSEL, %eax mov %ax, %fs - movl PCPU(CPUID), %eax - lock /* checkstate_need_ast &= ~(1<<id) */ - btrl %eax, checkstate_need_ast movl $0, lapic+LA_EOI /* End Of Interrupt to APIC */ - lock - btsl %eax, checkstate_pending_ast - jc 1f - FAKE_MCOUNT(13*4(%esp)) - MTX_LOCK_SPIN(sched_lock, 0) - movl PCPU(CURPROC),%ebx - orl $PS_ASTPENDING, P_SFLAG(%ebx) - - movl PCPU(CPUID), %eax - lock - btrl %eax, checkstate_pending_ast - lock - btrl %eax, CNAME(resched_cpus) - jnc 2f - orl $PS_NEEDRESCHED, P_SFLAG(%ebx) - lock - incl CNAME(want_resched_cnt) -2: - MTX_UNLOCK_SPIN(sched_lock) - lock - incl CNAME(cpuast_cnt) MEXITCOUNT jmp doreti -1: - /* We are already in the process of delivering an ast for this CPU */ - POP_FRAME - iret /* * Executed by a CPU when it receives an Xcpustop IPI from another CPU, @@ -331,7 +292,6 @@ Xcpuast: * - Waits for permission to restart. * - Signals its restart. */ - .text SUPERALIGN_TEXT .globl Xcpustop @@ -357,20 +317,19 @@ Xcpustop: pushl %eax call CNAME(savectx) /* Save process context */ addl $4, %esp - movl PCPU(CPUID), %eax lock - btsl %eax, stopped_cpus /* stopped_cpus |= (1<<id) */ + btsl %eax, CNAME(stopped_cpus) /* stopped_cpus |= (1<<id) */ 1: - btl %eax, started_cpus /* while (!(started_cpus & (1<<id))) */ + btl %eax, CNAME(started_cpus) /* while (!(started_cpus & (1<<id))) */ jnc 1b lock - btrl %eax, started_cpus /* started_cpus &= ~(1<<id) */ + btrl %eax, CNAME(started_cpus) /* started_cpus &= ~(1<<id) */ lock - btrl %eax, stopped_cpus /* stopped_cpus &= ~(1<<id) */ + btrl %eax, CNAME(stopped_cpus) /* stopped_cpus &= ~(1<<id) */ test %eax, %eax jnz 2f @@ -492,34 +451,6 @@ _xhits: .space (NCPU * 4), 0 #endif /* COUNT_XINVLTLB_HITS */ -/* variables used by stop_cpus()/restart_cpus()/Xcpustop */ - .globl stopped_cpus, started_cpus -stopped_cpus: - .long 0 -started_cpus: - .long 0 - - .globl checkstate_probed_cpus -checkstate_probed_cpus: - .long 0 - .globl checkstate_need_ast -checkstate_need_ast: - .long 0 -checkstate_pending_ast: - .long 0 - .globl CNAME(resched_cpus) - .globl CNAME(want_resched_cnt) - .globl CNAME(cpuast_cnt) - .globl CNAME(cpustop_restartfunc) -CNAME(resched_cpus): - .long 0 -CNAME(want_resched_cnt): - .long 0 -CNAME(cpuast_cnt): - .long 0 -CNAME(cpustop_restartfunc): - .long 0 - .globl apic_pin_trigger apic_pin_trigger: .long 0 diff --git a/sys/amd64/amd64/db_interface.c b/sys/amd64/amd64/db_interface.c index 8c7dc0b..cdaefe4 100644 --- a/sys/amd64/amd64/db_interface.c +++ b/sys/amd64/amd64/db_interface.c @@ -37,10 +37,10 @@ #include <sys/linker_set.h> #include <sys/lock.h> #include <sys/proc.h> +#include <sys/smp.h> #include <machine/cpu.h> #ifdef SMP -#include <machine/smp.h> #include <machine/smptests.h> /** CPUSTOP_ON_DDBBREAK */ #endif @@ -334,43 +334,44 @@ Debugger(msg) DB_SHOW_COMMAND(pcpu, db_show_pcpu) { struct globaldata *gd; +#ifdef SMP int id; if (have_addr) id = ((addr >> 4) % 16) * 10 + (addr % 16); else id = PCPU_GET(cpuid); - SLIST_FOREACH(gd, &cpuhead, gd_allcpu) { - if (gd->gd_cpuid == id) - break; - } - if (gd == NULL) + gd = globaldata_find(id); + if (gd == NULL) { db_printf("CPU %d not found\n", id); - else { - db_printf("cpuid = %d\n", gd->gd_cpuid); - db_printf("curproc = "); - if (gd->gd_curproc != NULL) - db_printf("%p: pid %d \"%s\"\n", gd->gd_curproc, - gd->gd_curproc->p_pid, gd->gd_curproc->p_comm); - else - db_printf("none\n"); - db_printf("curpcb = %p\n", gd->gd_curpcb); - db_printf("npxproc = "); - if (gd->gd_npxproc != NULL) - db_printf("%p: pid %d \"%s\"\n", gd->gd_npxproc, - gd->gd_npxproc->p_pid, gd->gd_npxproc->p_comm); - else - db_printf("none\n"); - db_printf("idleproc = "); - if (gd->gd_idleproc != NULL) - db_printf("%p: pid %d \"%s\"\n", gd->gd_idleproc, - gd->gd_idleproc->p_pid, gd->gd_idleproc->p_comm); - else - db_printf("none\n"); + return; + } +#else + gd = GLOBALDATA; +#endif + db_printf("cpuid = %d\n", gd->gd_cpuid); + db_printf("curproc = "); + if (gd->gd_curproc != NULL) + db_printf("%p: pid %d \"%s\"\n", gd->gd_curproc, + gd->gd_curproc->p_pid, gd->gd_curproc->p_comm); + else + db_printf("none\n"); + db_printf("curpcb = %p\n", gd->gd_curpcb); + db_printf("npxproc = "); + if (gd->gd_npxproc != NULL) + db_printf("%p: pid %d \"%s\"\n", gd->gd_npxproc, + gd->gd_npxproc->p_pid, gd->gd_npxproc->p_comm); + else + db_printf("none\n"); + db_printf("idleproc = "); + if (gd->gd_idleproc != NULL) + db_printf("%p: pid %d \"%s\"\n", gd->gd_idleproc, + gd->gd_idleproc->p_pid, gd->gd_idleproc->p_comm); + else + db_printf("none\n"); #ifdef WITNESS - db_printf("spin locks held:\n"); - witness_list_locks(&gd->gd_spinlocks); + db_printf("spin locks held:\n"); + witness_list_locks(&gd->gd_spinlocks); #endif - } } diff --git a/sys/amd64/amd64/machdep.c b/sys/amd64/amd64/machdep.c index dbdc61a..c89852a 100644 --- a/sys/amd64/amd64/machdep.c +++ b/sys/amd64/amd64/machdep.c @@ -66,6 +66,7 @@ #include <sys/bio.h> #include <sys/buf.h> #include <sys/reboot.h> +#include <sys/smp.h> #include <sys/callout.h> #include <sys/msgbuf.h> #include <sys/sysent.h> @@ -104,9 +105,6 @@ #include <machine/globaldata.h> #include <machine/globals.h> #include <machine/intrcnt.h> -#ifdef SMP -#include <machine/smp.h> -#endif #ifdef PERFMON #include <machine/perfmon.h> #endif @@ -249,8 +247,6 @@ static struct trapframe proc0_tf; static struct globaldata __globaldata; #endif -struct cpuhead cpuhead; - struct mtx sched_lock; struct mtx Giant; @@ -441,17 +437,12 @@ again: bufinit(); vm_pager_bufferinit(); - SLIST_INIT(&cpuhead); - SLIST_INSERT_HEAD(&cpuhead, GLOBALDATA, gd_allcpu); - #ifdef SMP - /* - * OK, enough kmem_alloc/malloc state should be up, lets get on with it! - */ - mp_start(); /* fire up the APs and APICs */ - mp_announce(); -#endif /* SMP */ + globaldata_register(GLOBALDATA); +#else + /* For SMP, we delay the cpu_setregs() until after SMP startup. */ cpu_setregs(); +#endif } /* @@ -1631,7 +1622,7 @@ physmap_done: physmap[1] = mp_bootaddress(physmap[1] / 1024); /* look for the MP hardware - needed for apic addresses */ - mp_probe(); + i386_mp_probe(); #endif /* diff --git a/sys/amd64/amd64/mp_machdep.c b/sys/amd64/amd64/mp_machdep.c index 28a5c72..21e6b6e 100644 --- a/sys/amd64/amd64/mp_machdep.c +++ b/sys/amd64/amd64/mp_machdep.c @@ -42,6 +42,7 @@ #include <sys/malloc.h> #include <sys/memrange.h> #include <sys/mutex.h> +#include <sys/smp.h> #include <sys/dkstat.h> #include <sys/cons.h> /* cngetc() */ @@ -57,9 +58,9 @@ #include <sys/gmon.h> #endif -#include <machine/smp.h> #include <machine/apic.h> #include <machine/atomic.h> +#include <machine/cpu.h> #include <machine/cpufunc.h> #include <machine/ipl.h> #include <machine/mpapic.h> @@ -243,7 +244,6 @@ int current_postcode; extern struct region_descriptor r_gdt, r_idt; int bsp_apic_ready = 0; /* flags useability of BSP apic */ -int mp_ncpus; /* # of CPUs, including BSP */ int mp_naps; /* # of Applications processors */ int mp_nbusses; /* # of busses */ int mp_napics; /* # of IO APICs */ @@ -273,9 +273,6 @@ int io_num_to_apic_id[NAPICID]; int apic_id_to_logical[NAPICID]; -/* Bitmap of all available CPUs */ -u_int all_cpus; - /* AP uses this during bootstrap. Do not staticize. */ char *bootSTK; static int bootAP; @@ -288,28 +285,9 @@ extern pt_entry_t *SMPpt; struct pcb stoppcbs[MAXCPU]; -int smp_started; /* has the system started? */ -int smp_active = 0; /* are the APs allowed to run? */ -SYSCTL_INT(_machdep, OID_AUTO, smp_active, CTLFLAG_RW, &smp_active, 0, ""); - -/* XXX maybe should be hw.ncpu */ -static int smp_cpus = 1; /* how many cpu's running */ -SYSCTL_INT(_machdep, OID_AUTO, smp_cpus, CTLFLAG_RD, &smp_cpus, 0, ""); - int invltlb_ok = 0; /* throttle smp_invltlb() till safe */ SYSCTL_INT(_machdep, OID_AUTO, invltlb_ok, CTLFLAG_RW, &invltlb_ok, 0, ""); -/* Enable forwarding of a signal to a process running on a different CPU */ -static int forward_signal_enabled = 1; -SYSCTL_INT(_machdep, OID_AUTO, forward_signal_enabled, CTLFLAG_RW, - &forward_signal_enabled, 0, ""); - -/* Enable forwarding of roundrobin to all other cpus */ -static int forward_roundrobin_enabled = 1; -SYSCTL_INT(_machdep, OID_AUTO, forward_roundrobin_enabled, CTLFLAG_RW, - &forward_roundrobin_enabled, 0, ""); - - /* * Local data and functions. */ @@ -354,9 +332,6 @@ struct mtx mcount_mtx; struct mtx com_mtx; #endif /* USE_COMLOCK */ -/* lock around the MP rendezvous */ -static struct mtx smp_rv_mtx; - static void init_locks(void) { @@ -367,13 +342,9 @@ init_locks(void) */ mtx_init(&mcount_mtx, "mcount", MTX_DEF); - mtx_init(&smp_rv_mtx, "smp rendezvous", MTX_SPIN); - #ifdef USE_COMLOCK mtx_init(&com_mtx, "com", MTX_SPIN); #endif /* USE_COMLOCK */ - - mtx_init(&ap_boot_mtx, "ap boot", MTX_SPIN); } /* @@ -397,8 +368,8 @@ mp_bootaddress(u_int basemem) /* * Look for an Intel MP spec table (ie, SMP capable hardware). */ -int -mp_probe(void) +void +i386_mp_probe(void) { int x; u_long segment; @@ -427,7 +398,7 @@ mp_probe(void) /* nothing found */ mpfps = (mpfps_t)0; mp_capable = 0; - return 0; + return; found: /* calculate needed resources */ @@ -436,15 +407,19 @@ found: /* flag fact that we are running multiple processors */ mp_capable = 1; - return 1; } +int +cpu_mp_probe(void) +{ + return (mp_capable); +} /* * Initialize the SMP hardware and the APIC and start up the AP's. */ void -mp_start(void) +cpu_mp_start(void) { POSTCODE(MP_START_POST); @@ -453,6 +428,8 @@ mp_start(void) mp_enable(boot_address); else panic("MP hardware not found!"); + + cpu_setregs(); } @@ -460,13 +437,12 @@ mp_start(void) * Print various information about the SMP system hardware and setup. */ void -mp_announce(void) +cpu_mp_announce(void) { int x; POSTCODE(MP_ANNOUNCE_POST); - printf("FreeBSD/SMP: Multiprocessor motherboard\n"); printf(" cpu0 (BSP): apic id: %2d", CPU_TO_ID(0)); printf(", version: 0x%08x", cpu_apic_versions[0]); printf(", at 0x%08x\n", cpu_apic_address); @@ -623,8 +599,12 @@ mp_enable(u_int boot_addr) setidt(XINVLTLB_OFFSET, Xinvltlb, SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); - /* install an inter-CPU IPI for reading processor state */ - setidt(XCPUCHECKSTATE_OFFSET, Xcpucheckstate, + /* install an inter-CPU IPI for forwarding hardclock() */ + setidt(XHARDCLOCK_OFFSET, Xhardclock, + SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); + + /* install an inter-CPU IPI for forwarding statclock() */ + setidt(XSTATCLOCK_OFFSET, Xstatclock, SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); /* install an inter-CPU IPI for all-CPU rendezvous */ @@ -1938,6 +1918,8 @@ start_all_aps(u_int boot_addr) POSTCODE(START_ALL_APS_POST); + mtx_init(&ap_boot_mtx, "ap boot", MTX_SPIN); + /* initialize BSP's local APIC */ apic_initialize(); bsp_apic_ready = 1; @@ -1985,8 +1967,8 @@ start_all_aps(u_int boot_addr) (PG_V | PG_RW | vtophys(PAGE_SIZE * i + stack)); /* prime data page for it to use */ - SLIST_INSERT_HEAD(&cpuhead, gd, gd_allcpu); gd->gd_cpuid = x; + globaldata_register(gd); /* setup a vector to our boot code */ *((volatile u_short *) WARMBOOT_OFF) = WARMBOOT_TARGET; @@ -2328,472 +2310,65 @@ ap_init(void) panic("scheduler returned us to ap_init"); } -#define CHECKSTATE_USER 0 -#define CHECKSTATE_SYS 1 -#define CHECKSTATE_INTR 2 - -/* Do not staticize. Used from apic_vector.s */ -struct proc* checkstate_curproc[MAXCPU]; -int checkstate_cpustate[MAXCPU]; -u_long checkstate_pc[MAXCPU]; - -#define PC_TO_INDEX(pc, prof) \ - ((int)(((u_quad_t)((pc) - (prof)->pr_off) * \ - (u_quad_t)((prof)->pr_scale)) >> 16) & ~1) - -static void -addupc_intr_forwarded(struct proc *p, int id, int *astmap) -{ - int i; - struct uprof *prof; - u_long pc; - - pc = checkstate_pc[id]; - prof = &p->p_stats->p_prof; - if (pc >= prof->pr_off && - (i = PC_TO_INDEX(pc, prof)) < prof->pr_size) { - mtx_assert(&sched_lock, MA_OWNED); - if ((p->p_sflag & PS_OWEUPC) == 0) { - prof->pr_addr = pc; - prof->pr_ticks = 1; - p->p_sflag |= PS_OWEUPC; - } - *astmap |= (1 << id); - } -} - -static void -forwarded_statclock(int id, int pscnt, int *astmap) -{ - struct pstats *pstats; - long rss; - struct rusage *ru; - struct vmspace *vm; - int cpustate; - struct proc *p; -#ifdef GPROF - register struct gmonparam *g; - int i; -#endif - - mtx_assert(&sched_lock, MA_OWNED); - p = checkstate_curproc[id]; - cpustate = checkstate_cpustate[id]; - - /* XXX */ - if (p->p_ithd) - cpustate = CHECKSTATE_INTR; - else if (p == SMP_prvspace[id].globaldata.gd_idleproc) - cpustate = CHECKSTATE_SYS; - - switch (cpustate) { - case CHECKSTATE_USER: - if (p->p_sflag & PS_PROFIL) - addupc_intr_forwarded(p, id, astmap); - if (pscnt > 1) - return; - p->p_uticks++; - if (p->p_nice > NZERO) - cp_time[CP_NICE]++; - else - cp_time[CP_USER]++; - break; - case CHECKSTATE_SYS: -#ifdef GPROF - /* - * Kernel statistics are just like addupc_intr, only easier. - */ - g = &_gmonparam; - if (g->state == GMON_PROF_ON) { - i = checkstate_pc[id] - g->lowpc; - if (i < g->textsize) { - i /= HISTFRACTION * sizeof(*g->kcount); - g->kcount[i]++; - } - } -#endif - if (pscnt > 1) - return; - - p->p_sticks++; - if (p == SMP_prvspace[id].globaldata.gd_idleproc) - cp_time[CP_IDLE]++; - else - cp_time[CP_SYS]++; - break; - case CHECKSTATE_INTR: - default: -#ifdef GPROF - /* - * Kernel statistics are just like addupc_intr, only easier. - */ - g = &_gmonparam; - if (g->state == GMON_PROF_ON) { - i = checkstate_pc[id] - g->lowpc; - if (i < g->textsize) { - i /= HISTFRACTION * sizeof(*g->kcount); - g->kcount[i]++; - } - } -#endif - if (pscnt > 1) - return; - KASSERT(p != NULL, ("NULL process in interrupt state")); - p->p_iticks++; - cp_time[CP_INTR]++; - } - - schedclock(p); - - /* Update resource usage integrals and maximums. */ - if ((pstats = p->p_stats) != NULL && - (ru = &pstats->p_ru) != NULL && - (vm = p->p_vmspace) != NULL) { - ru->ru_ixrss += pgtok(vm->vm_tsize); - ru->ru_idrss += pgtok(vm->vm_dsize); - ru->ru_isrss += pgtok(vm->vm_ssize); - rss = pgtok(vmspace_resident_count(vm)); - if (ru->ru_maxrss < rss) - ru->ru_maxrss = rss; - } -} - +/* + * For statclock, we send an IPI to all CPU's to have them call this + * function. + */ void -forward_statclock(int pscnt) +forwarded_statclock(struct trapframe frame) { - int map; - int id; - int i; - /* Kludge. We don't yet have separate locks for the interrupts - * and the kernel. This means that we cannot let the other processors - * handle complex interrupts while inhibiting them from entering - * the kernel in a non-interrupt context. - * - * What we can do, without changing the locking mechanisms yet, - * is letting the other processors handle a very simple interrupt - * (wich determines the processor states), and do the main - * work ourself. - */ - - CTR1(KTR_SMP, "forward_statclock(%d)", pscnt); - - if (!smp_started || !invltlb_ok || cold || panicstr) - return; - - /* Step 1: Probe state (user, cpu, interrupt, spinlock, idle ) */ - - map = PCPU_GET(other_cpus) & ~stopped_cpus ; - checkstate_probed_cpus = 0; - if (map != 0) - ipi_selected(map, IPI_CHECKSTATE); - - i = 0; - while (checkstate_probed_cpus != map) { - /* spin */ - i++; - if (i == 100000) { -#ifdef DIAGNOSTIC - printf("forward_statclock: checkstate %x\n", - checkstate_probed_cpus); -#endif - break; - } - } - - /* - * Step 2: walk through other processors processes, update ticks and - * profiling info. - */ - - map = 0; - for (id = 0; id < mp_ncpus; id++) { - if (id == PCPU_GET(cpuid)) - continue; - if (((1 << id) & checkstate_probed_cpus) == 0) - continue; - forwarded_statclock(id, pscnt, &map); - } - if (map != 0) { - checkstate_need_ast |= map; - ipi_selected(map, IPI_AST); - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#ifdef DIAGNOSTIC - printf("forward_statclock: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } - } + mtx_lock_spin(&sched_lock); + statclock_process(curproc, TRAPF_PC(&frame), TRAPF_USERMODE(&frame)); + mtx_unlock_spin(&sched_lock); } -void -forward_hardclock(int pscnt) +void +forward_statclock(void) { int map; - int id; - struct proc *p; - struct pstats *pstats; - int i; - - /* Kludge. We don't yet have separate locks for the interrupts - * and the kernel. This means that we cannot let the other processors - * handle complex interrupts while inhibiting them from entering - * the kernel in a non-interrupt context. - * - * What we can do, without changing the locking mechanisms yet, - * is letting the other processors handle a very simple interrupt - * (wich determines the processor states), and do the main - * work ourself. - */ - CTR1(KTR_SMP, "forward_hardclock(%d)", pscnt); + CTR0(KTR_SMP, "forward_statclock"); if (!smp_started || !invltlb_ok || cold || panicstr) return; - /* Step 1: Probe state (user, cpu, interrupt, spinlock, idle) */ - map = PCPU_GET(other_cpus) & ~stopped_cpus ; - checkstate_probed_cpus = 0; if (map != 0) - ipi_selected(map, IPI_CHECKSTATE); - - i = 0; - while (checkstate_probed_cpus != map) { - /* spin */ - i++; - if (i == 100000) { -#ifdef DIAGNOSTIC - printf("forward_hardclock: checkstate %x\n", - checkstate_probed_cpus); -#endif - break; - } - } - - /* - * Step 2: walk through other processors processes, update virtual - * timer and profiling timer. If stathz == 0, also update ticks and - * profiling info. - */ - - map = 0; - for (id = 0; id < mp_ncpus; id++) { - if (id == PCPU_GET(cpuid)) - continue; - if (((1 << id) & checkstate_probed_cpus) == 0) - continue; - p = checkstate_curproc[id]; - if (p) { - pstats = p->p_stats; - if (checkstate_cpustate[id] == CHECKSTATE_USER && - timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) && - itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) { - p->p_sflag |= PS_ALRMPEND; - map |= (1 << id); - } - if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) && - itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) { - p->p_sflag |= PS_PROFPEND; - map |= (1 << id); - } - } - if (stathz == 0) { - forwarded_statclock( id, pscnt, &map); - } - } - if (map != 0) { - checkstate_need_ast |= map; - ipi_selected(map, IPI_AST); - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#ifdef DIAGNOSTIC - printf("forward_hardclock: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } - } + ipi_selected(map, IPI_STATCLOCK); } -void -forward_signal(struct proc *p) +/* + * For each hardclock(), we send an IPI to all other CPU's to have them + * execute this function. It would be nice to reduce contention on + * sched_lock if we could simply peek at the CPU to determine the user/kernel + * state and call hardclock_process() on the CPU receiving the clock interrupt + * and then just use a simple IPI to handle any ast's if needed. + */ +void +forwarded_hardclock(struct trapframe frame) { - int map; - int id; - int i; - - /* Kludge. We don't yet have separate locks for the interrupts - * and the kernel. This means that we cannot let the other processors - * handle complex interrupts while inhibiting them from entering - * the kernel in a non-interrupt context. - * - * What we can do, without changing the locking mechanisms yet, - * is letting the other processors handle a very simple interrupt - * (wich determines the processor states), and do the main - * work ourself. - */ - - CTR1(KTR_SMP, "forward_signal(%p)", p); - if (!smp_started || !invltlb_ok || cold || panicstr) - return; - if (!forward_signal_enabled) - return; mtx_lock_spin(&sched_lock); - while (1) { - if (p->p_stat != SRUN) { - mtx_unlock_spin(&sched_lock); - return; - } - id = p->p_oncpu; - mtx_unlock_spin(&sched_lock); - if (id == 0xff) - return; - map = (1<<id); - checkstate_need_ast |= map; - ipi_selected(map, IPI_AST); - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#if 0 - printf("forward_signal: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } - mtx_lock_spin(&sched_lock); - if (id == p->p_oncpu) { - mtx_unlock_spin(&sched_lock); - return; - } - } + hardclock_process(curproc, TRAPF_USERMODE(&frame)); + mtx_unlock_spin(&sched_lock); } -void -forward_roundrobin(void) +void +forward_hardclock(void) { u_int map; - int i; - CTR0(KTR_SMP, "forward_roundrobin()"); + CTR0(KTR_SMP, "forward_hardclock"); if (!smp_started || !invltlb_ok || cold || panicstr) return; - if (!forward_roundrobin_enabled) - return; - resched_cpus |= PCPU_GET(other_cpus); - map = PCPU_GET(other_cpus) & ~stopped_cpus ; -#if 1 - ipi_selected(map, IPI_AST); -#else - ipi_all_but_self(IPI_AST); -#endif - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#if 0 - printf("forward_roundrobin: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } -} - -/* - * When called the executing CPU will send an IPI to all other CPUs - * requesting that they halt execution. - * - * Usually (but not necessarily) called with 'other_cpus' as its arg. - * - * - Signals all CPUs in map to stop. - * - Waits for each to stop. - * - * Returns: - * -1: error - * 0: NA - * 1: ok - * - * XXX FIXME: this is not MP-safe, needs a lock to prevent multiple CPUs - * from executing at same time. - */ -int -stop_cpus(u_int map) -{ - int count = 0; - - if (!smp_started) - return 0; - - /* send the Xcpustop IPI to all CPUs in map */ - ipi_selected(map, IPI_STOP); - - while (count++ < 100000 && (stopped_cpus & map) != map) - /* spin */ ; - -#ifdef DIAGNOSTIC - if ((stopped_cpus & map) != map) - printf("Warning: CPUs 0x%x did not stop!\n", - (~(stopped_cpus & map)) & map); -#endif - - return 1; -} - - -/* - * Called by a CPU to restart stopped CPUs. - * - * Usually (but not necessarily) called with 'stopped_cpus' as its arg. - * - * - Signals all CPUs in map to restart. - * - Waits for each to restart. - * - * Returns: - * -1: error - * 0: NA - * 1: ok - */ -int -restart_cpus(u_int map) -{ - int count = 0; - - if (!smp_started) - return 0; - - started_cpus = map; /* signal other cpus to restart */ - - /* wait for each to clear its bit */ - while (count++ < 100000 && (stopped_cpus & map) != 0) - /* spin */ ; -#ifdef DIAGNOSTIC - if ((stopped_cpus & map) != 0) - printf("Warning: CPUs 0x%x did not restart!\n", - (~(stopped_cpus & map)) & map); -#endif - - return 1; + map = PCPU_GET(other_cpus) & ~stopped_cpus ; + if (map != 0) + ipi_selected(map, IPI_HARDCLOCK); } - #ifdef APIC_INTR_REORDER /* * Maintain mapping from softintr vector to isr bit in local apic. @@ -2811,73 +2386,6 @@ set_lapic_isrloc(int intr, int vector) #endif /* - * All-CPU rendezvous. CPUs are signalled, all execute the setup function - * (if specified), rendezvous, execute the action function (if specified), - * rendezvous again, execute the teardown function (if specified), and then - * resume. - * - * Note that the supplied external functions _must_ be reentrant and aware - * that they are running in parallel and in an unknown lock context. - */ -static void (*smp_rv_setup_func)(void *arg); -static void (*smp_rv_action_func)(void *arg); -static void (*smp_rv_teardown_func)(void *arg); -static void *smp_rv_func_arg; -static volatile int smp_rv_waiters[2]; - -void -smp_rendezvous_action(void) -{ - /* setup function */ - if (smp_rv_setup_func != NULL) - smp_rv_setup_func(smp_rv_func_arg); - /* spin on entry rendezvous */ - atomic_add_int(&smp_rv_waiters[0], 1); - while (smp_rv_waiters[0] < mp_ncpus) - ; - /* action function */ - if (smp_rv_action_func != NULL) - smp_rv_action_func(smp_rv_func_arg); - /* spin on exit rendezvous */ - atomic_add_int(&smp_rv_waiters[1], 1); - while (smp_rv_waiters[1] < mp_ncpus) - ; - /* teardown function */ - if (smp_rv_teardown_func != NULL) - smp_rv_teardown_func(smp_rv_func_arg); -} - -void -smp_rendezvous(void (* setup_func)(void *), - void (* action_func)(void *), - void (* teardown_func)(void *), - void *arg) -{ - - /* obtain rendezvous lock */ - mtx_lock_spin(&smp_rv_mtx); - - /* set static function pointers */ - smp_rv_setup_func = setup_func; - smp_rv_action_func = action_func; - smp_rv_teardown_func = teardown_func; - smp_rv_func_arg = arg; - smp_rv_waiters[0] = 0; - smp_rv_waiters[1] = 0; - - /* - * signal other processors, which will enter the IPI with interrupts off - */ - ipi_all_but_self(IPI_RENDEZVOUS); - - /* call executor function */ - smp_rendezvous_action(); - - /* release lock */ - mtx_unlock_spin(&smp_rv_mtx); -} - -/* * send an IPI to a set of cpus. */ void diff --git a/sys/amd64/amd64/mptable.c b/sys/amd64/amd64/mptable.c index 28a5c72..21e6b6e 100644 --- a/sys/amd64/amd64/mptable.c +++ b/sys/amd64/amd64/mptable.c @@ -42,6 +42,7 @@ #include <sys/malloc.h> #include <sys/memrange.h> #include <sys/mutex.h> +#include <sys/smp.h> #include <sys/dkstat.h> #include <sys/cons.h> /* cngetc() */ @@ -57,9 +58,9 @@ #include <sys/gmon.h> #endif -#include <machine/smp.h> #include <machine/apic.h> #include <machine/atomic.h> +#include <machine/cpu.h> #include <machine/cpufunc.h> #include <machine/ipl.h> #include <machine/mpapic.h> @@ -243,7 +244,6 @@ int current_postcode; extern struct region_descriptor r_gdt, r_idt; int bsp_apic_ready = 0; /* flags useability of BSP apic */ -int mp_ncpus; /* # of CPUs, including BSP */ int mp_naps; /* # of Applications processors */ int mp_nbusses; /* # of busses */ int mp_napics; /* # of IO APICs */ @@ -273,9 +273,6 @@ int io_num_to_apic_id[NAPICID]; int apic_id_to_logical[NAPICID]; -/* Bitmap of all available CPUs */ -u_int all_cpus; - /* AP uses this during bootstrap. Do not staticize. */ char *bootSTK; static int bootAP; @@ -288,28 +285,9 @@ extern pt_entry_t *SMPpt; struct pcb stoppcbs[MAXCPU]; -int smp_started; /* has the system started? */ -int smp_active = 0; /* are the APs allowed to run? */ -SYSCTL_INT(_machdep, OID_AUTO, smp_active, CTLFLAG_RW, &smp_active, 0, ""); - -/* XXX maybe should be hw.ncpu */ -static int smp_cpus = 1; /* how many cpu's running */ -SYSCTL_INT(_machdep, OID_AUTO, smp_cpus, CTLFLAG_RD, &smp_cpus, 0, ""); - int invltlb_ok = 0; /* throttle smp_invltlb() till safe */ SYSCTL_INT(_machdep, OID_AUTO, invltlb_ok, CTLFLAG_RW, &invltlb_ok, 0, ""); -/* Enable forwarding of a signal to a process running on a different CPU */ -static int forward_signal_enabled = 1; -SYSCTL_INT(_machdep, OID_AUTO, forward_signal_enabled, CTLFLAG_RW, - &forward_signal_enabled, 0, ""); - -/* Enable forwarding of roundrobin to all other cpus */ -static int forward_roundrobin_enabled = 1; -SYSCTL_INT(_machdep, OID_AUTO, forward_roundrobin_enabled, CTLFLAG_RW, - &forward_roundrobin_enabled, 0, ""); - - /* * Local data and functions. */ @@ -354,9 +332,6 @@ struct mtx mcount_mtx; struct mtx com_mtx; #endif /* USE_COMLOCK */ -/* lock around the MP rendezvous */ -static struct mtx smp_rv_mtx; - static void init_locks(void) { @@ -367,13 +342,9 @@ init_locks(void) */ mtx_init(&mcount_mtx, "mcount", MTX_DEF); - mtx_init(&smp_rv_mtx, "smp rendezvous", MTX_SPIN); - #ifdef USE_COMLOCK mtx_init(&com_mtx, "com", MTX_SPIN); #endif /* USE_COMLOCK */ - - mtx_init(&ap_boot_mtx, "ap boot", MTX_SPIN); } /* @@ -397,8 +368,8 @@ mp_bootaddress(u_int basemem) /* * Look for an Intel MP spec table (ie, SMP capable hardware). */ -int -mp_probe(void) +void +i386_mp_probe(void) { int x; u_long segment; @@ -427,7 +398,7 @@ mp_probe(void) /* nothing found */ mpfps = (mpfps_t)0; mp_capable = 0; - return 0; + return; found: /* calculate needed resources */ @@ -436,15 +407,19 @@ found: /* flag fact that we are running multiple processors */ mp_capable = 1; - return 1; } +int +cpu_mp_probe(void) +{ + return (mp_capable); +} /* * Initialize the SMP hardware and the APIC and start up the AP's. */ void -mp_start(void) +cpu_mp_start(void) { POSTCODE(MP_START_POST); @@ -453,6 +428,8 @@ mp_start(void) mp_enable(boot_address); else panic("MP hardware not found!"); + + cpu_setregs(); } @@ -460,13 +437,12 @@ mp_start(void) * Print various information about the SMP system hardware and setup. */ void -mp_announce(void) +cpu_mp_announce(void) { int x; POSTCODE(MP_ANNOUNCE_POST); - printf("FreeBSD/SMP: Multiprocessor motherboard\n"); printf(" cpu0 (BSP): apic id: %2d", CPU_TO_ID(0)); printf(", version: 0x%08x", cpu_apic_versions[0]); printf(", at 0x%08x\n", cpu_apic_address); @@ -623,8 +599,12 @@ mp_enable(u_int boot_addr) setidt(XINVLTLB_OFFSET, Xinvltlb, SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); - /* install an inter-CPU IPI for reading processor state */ - setidt(XCPUCHECKSTATE_OFFSET, Xcpucheckstate, + /* install an inter-CPU IPI for forwarding hardclock() */ + setidt(XHARDCLOCK_OFFSET, Xhardclock, + SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); + + /* install an inter-CPU IPI for forwarding statclock() */ + setidt(XSTATCLOCK_OFFSET, Xstatclock, SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); /* install an inter-CPU IPI for all-CPU rendezvous */ @@ -1938,6 +1918,8 @@ start_all_aps(u_int boot_addr) POSTCODE(START_ALL_APS_POST); + mtx_init(&ap_boot_mtx, "ap boot", MTX_SPIN); + /* initialize BSP's local APIC */ apic_initialize(); bsp_apic_ready = 1; @@ -1985,8 +1967,8 @@ start_all_aps(u_int boot_addr) (PG_V | PG_RW | vtophys(PAGE_SIZE * i + stack)); /* prime data page for it to use */ - SLIST_INSERT_HEAD(&cpuhead, gd, gd_allcpu); gd->gd_cpuid = x; + globaldata_register(gd); /* setup a vector to our boot code */ *((volatile u_short *) WARMBOOT_OFF) = WARMBOOT_TARGET; @@ -2328,472 +2310,65 @@ ap_init(void) panic("scheduler returned us to ap_init"); } -#define CHECKSTATE_USER 0 -#define CHECKSTATE_SYS 1 -#define CHECKSTATE_INTR 2 - -/* Do not staticize. Used from apic_vector.s */ -struct proc* checkstate_curproc[MAXCPU]; -int checkstate_cpustate[MAXCPU]; -u_long checkstate_pc[MAXCPU]; - -#define PC_TO_INDEX(pc, prof) \ - ((int)(((u_quad_t)((pc) - (prof)->pr_off) * \ - (u_quad_t)((prof)->pr_scale)) >> 16) & ~1) - -static void -addupc_intr_forwarded(struct proc *p, int id, int *astmap) -{ - int i; - struct uprof *prof; - u_long pc; - - pc = checkstate_pc[id]; - prof = &p->p_stats->p_prof; - if (pc >= prof->pr_off && - (i = PC_TO_INDEX(pc, prof)) < prof->pr_size) { - mtx_assert(&sched_lock, MA_OWNED); - if ((p->p_sflag & PS_OWEUPC) == 0) { - prof->pr_addr = pc; - prof->pr_ticks = 1; - p->p_sflag |= PS_OWEUPC; - } - *astmap |= (1 << id); - } -} - -static void -forwarded_statclock(int id, int pscnt, int *astmap) -{ - struct pstats *pstats; - long rss; - struct rusage *ru; - struct vmspace *vm; - int cpustate; - struct proc *p; -#ifdef GPROF - register struct gmonparam *g; - int i; -#endif - - mtx_assert(&sched_lock, MA_OWNED); - p = checkstate_curproc[id]; - cpustate = checkstate_cpustate[id]; - - /* XXX */ - if (p->p_ithd) - cpustate = CHECKSTATE_INTR; - else if (p == SMP_prvspace[id].globaldata.gd_idleproc) - cpustate = CHECKSTATE_SYS; - - switch (cpustate) { - case CHECKSTATE_USER: - if (p->p_sflag & PS_PROFIL) - addupc_intr_forwarded(p, id, astmap); - if (pscnt > 1) - return; - p->p_uticks++; - if (p->p_nice > NZERO) - cp_time[CP_NICE]++; - else - cp_time[CP_USER]++; - break; - case CHECKSTATE_SYS: -#ifdef GPROF - /* - * Kernel statistics are just like addupc_intr, only easier. - */ - g = &_gmonparam; - if (g->state == GMON_PROF_ON) { - i = checkstate_pc[id] - g->lowpc; - if (i < g->textsize) { - i /= HISTFRACTION * sizeof(*g->kcount); - g->kcount[i]++; - } - } -#endif - if (pscnt > 1) - return; - - p->p_sticks++; - if (p == SMP_prvspace[id].globaldata.gd_idleproc) - cp_time[CP_IDLE]++; - else - cp_time[CP_SYS]++; - break; - case CHECKSTATE_INTR: - default: -#ifdef GPROF - /* - * Kernel statistics are just like addupc_intr, only easier. - */ - g = &_gmonparam; - if (g->state == GMON_PROF_ON) { - i = checkstate_pc[id] - g->lowpc; - if (i < g->textsize) { - i /= HISTFRACTION * sizeof(*g->kcount); - g->kcount[i]++; - } - } -#endif - if (pscnt > 1) - return; - KASSERT(p != NULL, ("NULL process in interrupt state")); - p->p_iticks++; - cp_time[CP_INTR]++; - } - - schedclock(p); - - /* Update resource usage integrals and maximums. */ - if ((pstats = p->p_stats) != NULL && - (ru = &pstats->p_ru) != NULL && - (vm = p->p_vmspace) != NULL) { - ru->ru_ixrss += pgtok(vm->vm_tsize); - ru->ru_idrss += pgtok(vm->vm_dsize); - ru->ru_isrss += pgtok(vm->vm_ssize); - rss = pgtok(vmspace_resident_count(vm)); - if (ru->ru_maxrss < rss) - ru->ru_maxrss = rss; - } -} - +/* + * For statclock, we send an IPI to all CPU's to have them call this + * function. + */ void -forward_statclock(int pscnt) +forwarded_statclock(struct trapframe frame) { - int map; - int id; - int i; - /* Kludge. We don't yet have separate locks for the interrupts - * and the kernel. This means that we cannot let the other processors - * handle complex interrupts while inhibiting them from entering - * the kernel in a non-interrupt context. - * - * What we can do, without changing the locking mechanisms yet, - * is letting the other processors handle a very simple interrupt - * (wich determines the processor states), and do the main - * work ourself. - */ - - CTR1(KTR_SMP, "forward_statclock(%d)", pscnt); - - if (!smp_started || !invltlb_ok || cold || panicstr) - return; - - /* Step 1: Probe state (user, cpu, interrupt, spinlock, idle ) */ - - map = PCPU_GET(other_cpus) & ~stopped_cpus ; - checkstate_probed_cpus = 0; - if (map != 0) - ipi_selected(map, IPI_CHECKSTATE); - - i = 0; - while (checkstate_probed_cpus != map) { - /* spin */ - i++; - if (i == 100000) { -#ifdef DIAGNOSTIC - printf("forward_statclock: checkstate %x\n", - checkstate_probed_cpus); -#endif - break; - } - } - - /* - * Step 2: walk through other processors processes, update ticks and - * profiling info. - */ - - map = 0; - for (id = 0; id < mp_ncpus; id++) { - if (id == PCPU_GET(cpuid)) - continue; - if (((1 << id) & checkstate_probed_cpus) == 0) - continue; - forwarded_statclock(id, pscnt, &map); - } - if (map != 0) { - checkstate_need_ast |= map; - ipi_selected(map, IPI_AST); - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#ifdef DIAGNOSTIC - printf("forward_statclock: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } - } + mtx_lock_spin(&sched_lock); + statclock_process(curproc, TRAPF_PC(&frame), TRAPF_USERMODE(&frame)); + mtx_unlock_spin(&sched_lock); } -void -forward_hardclock(int pscnt) +void +forward_statclock(void) { int map; - int id; - struct proc *p; - struct pstats *pstats; - int i; - - /* Kludge. We don't yet have separate locks for the interrupts - * and the kernel. This means that we cannot let the other processors - * handle complex interrupts while inhibiting them from entering - * the kernel in a non-interrupt context. - * - * What we can do, without changing the locking mechanisms yet, - * is letting the other processors handle a very simple interrupt - * (wich determines the processor states), and do the main - * work ourself. - */ - CTR1(KTR_SMP, "forward_hardclock(%d)", pscnt); + CTR0(KTR_SMP, "forward_statclock"); if (!smp_started || !invltlb_ok || cold || panicstr) return; - /* Step 1: Probe state (user, cpu, interrupt, spinlock, idle) */ - map = PCPU_GET(other_cpus) & ~stopped_cpus ; - checkstate_probed_cpus = 0; if (map != 0) - ipi_selected(map, IPI_CHECKSTATE); - - i = 0; - while (checkstate_probed_cpus != map) { - /* spin */ - i++; - if (i == 100000) { -#ifdef DIAGNOSTIC - printf("forward_hardclock: checkstate %x\n", - checkstate_probed_cpus); -#endif - break; - } - } - - /* - * Step 2: walk through other processors processes, update virtual - * timer and profiling timer. If stathz == 0, also update ticks and - * profiling info. - */ - - map = 0; - for (id = 0; id < mp_ncpus; id++) { - if (id == PCPU_GET(cpuid)) - continue; - if (((1 << id) & checkstate_probed_cpus) == 0) - continue; - p = checkstate_curproc[id]; - if (p) { - pstats = p->p_stats; - if (checkstate_cpustate[id] == CHECKSTATE_USER && - timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) && - itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) { - p->p_sflag |= PS_ALRMPEND; - map |= (1 << id); - } - if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) && - itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) { - p->p_sflag |= PS_PROFPEND; - map |= (1 << id); - } - } - if (stathz == 0) { - forwarded_statclock( id, pscnt, &map); - } - } - if (map != 0) { - checkstate_need_ast |= map; - ipi_selected(map, IPI_AST); - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#ifdef DIAGNOSTIC - printf("forward_hardclock: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } - } + ipi_selected(map, IPI_STATCLOCK); } -void -forward_signal(struct proc *p) +/* + * For each hardclock(), we send an IPI to all other CPU's to have them + * execute this function. It would be nice to reduce contention on + * sched_lock if we could simply peek at the CPU to determine the user/kernel + * state and call hardclock_process() on the CPU receiving the clock interrupt + * and then just use a simple IPI to handle any ast's if needed. + */ +void +forwarded_hardclock(struct trapframe frame) { - int map; - int id; - int i; - - /* Kludge. We don't yet have separate locks for the interrupts - * and the kernel. This means that we cannot let the other processors - * handle complex interrupts while inhibiting them from entering - * the kernel in a non-interrupt context. - * - * What we can do, without changing the locking mechanisms yet, - * is letting the other processors handle a very simple interrupt - * (wich determines the processor states), and do the main - * work ourself. - */ - - CTR1(KTR_SMP, "forward_signal(%p)", p); - if (!smp_started || !invltlb_ok || cold || panicstr) - return; - if (!forward_signal_enabled) - return; mtx_lock_spin(&sched_lock); - while (1) { - if (p->p_stat != SRUN) { - mtx_unlock_spin(&sched_lock); - return; - } - id = p->p_oncpu; - mtx_unlock_spin(&sched_lock); - if (id == 0xff) - return; - map = (1<<id); - checkstate_need_ast |= map; - ipi_selected(map, IPI_AST); - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#if 0 - printf("forward_signal: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } - mtx_lock_spin(&sched_lock); - if (id == p->p_oncpu) { - mtx_unlock_spin(&sched_lock); - return; - } - } + hardclock_process(curproc, TRAPF_USERMODE(&frame)); + mtx_unlock_spin(&sched_lock); } -void -forward_roundrobin(void) +void +forward_hardclock(void) { u_int map; - int i; - CTR0(KTR_SMP, "forward_roundrobin()"); + CTR0(KTR_SMP, "forward_hardclock"); if (!smp_started || !invltlb_ok || cold || panicstr) return; - if (!forward_roundrobin_enabled) - return; - resched_cpus |= PCPU_GET(other_cpus); - map = PCPU_GET(other_cpus) & ~stopped_cpus ; -#if 1 - ipi_selected(map, IPI_AST); -#else - ipi_all_but_self(IPI_AST); -#endif - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#if 0 - printf("forward_roundrobin: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } -} - -/* - * When called the executing CPU will send an IPI to all other CPUs - * requesting that they halt execution. - * - * Usually (but not necessarily) called with 'other_cpus' as its arg. - * - * - Signals all CPUs in map to stop. - * - Waits for each to stop. - * - * Returns: - * -1: error - * 0: NA - * 1: ok - * - * XXX FIXME: this is not MP-safe, needs a lock to prevent multiple CPUs - * from executing at same time. - */ -int -stop_cpus(u_int map) -{ - int count = 0; - - if (!smp_started) - return 0; - - /* send the Xcpustop IPI to all CPUs in map */ - ipi_selected(map, IPI_STOP); - - while (count++ < 100000 && (stopped_cpus & map) != map) - /* spin */ ; - -#ifdef DIAGNOSTIC - if ((stopped_cpus & map) != map) - printf("Warning: CPUs 0x%x did not stop!\n", - (~(stopped_cpus & map)) & map); -#endif - - return 1; -} - - -/* - * Called by a CPU to restart stopped CPUs. - * - * Usually (but not necessarily) called with 'stopped_cpus' as its arg. - * - * - Signals all CPUs in map to restart. - * - Waits for each to restart. - * - * Returns: - * -1: error - * 0: NA - * 1: ok - */ -int -restart_cpus(u_int map) -{ - int count = 0; - - if (!smp_started) - return 0; - - started_cpus = map; /* signal other cpus to restart */ - - /* wait for each to clear its bit */ - while (count++ < 100000 && (stopped_cpus & map) != 0) - /* spin */ ; -#ifdef DIAGNOSTIC - if ((stopped_cpus & map) != 0) - printf("Warning: CPUs 0x%x did not restart!\n", - (~(stopped_cpus & map)) & map); -#endif - - return 1; + map = PCPU_GET(other_cpus) & ~stopped_cpus ; + if (map != 0) + ipi_selected(map, IPI_HARDCLOCK); } - #ifdef APIC_INTR_REORDER /* * Maintain mapping from softintr vector to isr bit in local apic. @@ -2811,73 +2386,6 @@ set_lapic_isrloc(int intr, int vector) #endif /* - * All-CPU rendezvous. CPUs are signalled, all execute the setup function - * (if specified), rendezvous, execute the action function (if specified), - * rendezvous again, execute the teardown function (if specified), and then - * resume. - * - * Note that the supplied external functions _must_ be reentrant and aware - * that they are running in parallel and in an unknown lock context. - */ -static void (*smp_rv_setup_func)(void *arg); -static void (*smp_rv_action_func)(void *arg); -static void (*smp_rv_teardown_func)(void *arg); -static void *smp_rv_func_arg; -static volatile int smp_rv_waiters[2]; - -void -smp_rendezvous_action(void) -{ - /* setup function */ - if (smp_rv_setup_func != NULL) - smp_rv_setup_func(smp_rv_func_arg); - /* spin on entry rendezvous */ - atomic_add_int(&smp_rv_waiters[0], 1); - while (smp_rv_waiters[0] < mp_ncpus) - ; - /* action function */ - if (smp_rv_action_func != NULL) - smp_rv_action_func(smp_rv_func_arg); - /* spin on exit rendezvous */ - atomic_add_int(&smp_rv_waiters[1], 1); - while (smp_rv_waiters[1] < mp_ncpus) - ; - /* teardown function */ - if (smp_rv_teardown_func != NULL) - smp_rv_teardown_func(smp_rv_func_arg); -} - -void -smp_rendezvous(void (* setup_func)(void *), - void (* action_func)(void *), - void (* teardown_func)(void *), - void *arg) -{ - - /* obtain rendezvous lock */ - mtx_lock_spin(&smp_rv_mtx); - - /* set static function pointers */ - smp_rv_setup_func = setup_func; - smp_rv_action_func = action_func; - smp_rv_teardown_func = teardown_func; - smp_rv_func_arg = arg; - smp_rv_waiters[0] = 0; - smp_rv_waiters[1] = 0; - - /* - * signal other processors, which will enter the IPI with interrupts off - */ - ipi_all_but_self(IPI_RENDEZVOUS); - - /* call executor function */ - smp_rendezvous_action(); - - /* release lock */ - mtx_unlock_spin(&smp_rv_mtx); -} - -/* * send an IPI to a set of cpus. */ void diff --git a/sys/amd64/amd64/sys_machdep.c b/sys/amd64/amd64/sys_machdep.c index cf05d73..5ed16ab 100644 --- a/sys/amd64/amd64/sys_machdep.c +++ b/sys/amd64/amd64/sys_machdep.c @@ -41,6 +41,7 @@ #include <sys/ipl.h> #include <sys/malloc.h> #include <sys/proc.h> +#include <sys/smp.h> #include <vm/vm.h> #include <sys/lock.h> @@ -53,9 +54,6 @@ #include <machine/cpu.h> #include <machine/pcb_ext.h> /* pcb.h included by sys/user.h */ #include <machine/sysarch.h> -#ifdef SMP -#include <machine/smp.h> -#endif #include <vm/vm_kern.h> /* for kernel_map */ @@ -154,7 +152,15 @@ i386_extend_pcb(struct proc *p) ssdtosd(&ssd, &ext->ext_tssd); /* switch to the new TSS after syscall completes */ - need_resched(); + /* + * XXX: The sched_lock here needs to be over a slightly larger area. + * I have patches to more properly lock accesses to process ldt's + * and tss's that still need to be reviewed, but this keeps us from + * panic'ing on the mtx_assert() in need_resched() for the time being. + */ + mtx_lock_spin(&sched_lock); + need_resched(p); + mtx_unlock_spin(&sched_lock); return 0; } diff --git a/sys/amd64/amd64/trap.c b/sys/amd64/amd64/trap.c index 7088d0c..d9597df 100644 --- a/sys/amd64/amd64/trap.c +++ b/sys/amd64/amd64/trap.c @@ -177,7 +177,7 @@ userret(p, frame, oticks) mtx_lock_spin(&sched_lock); p->p_pri.pri_level = p->p_pri.pri_user; - if (resched_wanted()) { + if (resched_wanted(p)) { /* * Since we are curproc, clock will normally just change * our priority without moving us from one queue to another @@ -1277,7 +1277,7 @@ ast(framep) * acquiring and releasing mutexes in assembly is not fun. */ mtx_lock_spin(&sched_lock); - if (!(astpending(p) || resched_wanted())) { + if (!(astpending(p) || resched_wanted(p))) { mtx_unlock_spin(&sched_lock); return; } diff --git a/sys/amd64/amd64/tsc.c b/sys/amd64/amd64/tsc.c index 9034f81..ce67390 100644 --- a/sys/amd64/amd64/tsc.c +++ b/sys/amd64/amd64/tsc.c @@ -211,6 +211,10 @@ clkintr(struct clockframe frame) mtx_unlock_spin(&clock_lock); } timer_func(&frame); +#ifdef SMP + if (timer_func == hardclock) + forward_hardclock(); +#endif switch (timer0_state) { case RELEASED: @@ -253,6 +257,9 @@ clkintr(struct clockframe frame) timer_func = hardclock; timer0_state = RELEASED; hardclock(&frame); +#ifdef SMP + forward_hardclock(); +#endif } break; } @@ -374,8 +381,12 @@ release_timer2() static void rtcintr(struct clockframe frame) { - while (rtcin(RTC_INTR) & RTCIR_PERIOD) + while (rtcin(RTC_INTR) & RTCIR_PERIOD) { statclock(&frame); +#ifdef SMP + forward_statclock(); +#endif + } } #include "opt_ddb.h" diff --git a/sys/amd64/amd64/vm_machdep.c b/sys/amd64/amd64/vm_machdep.c index 0a15103..f8fa5bb 100644 --- a/sys/amd64/amd64/vm_machdep.c +++ b/sys/amd64/amd64/vm_machdep.c @@ -59,14 +59,12 @@ #include <sys/kernel.h> #include <sys/ktr.h> #include <sys/mutex.h> +#include <sys/smp.h> #include <sys/sysctl.h> #include <sys/unistd.h> #include <machine/cpu.h> #include <machine/md_var.h> -#ifdef SMP -#include <machine/smp.h> -#endif #include <machine/pcb.h> #include <machine/pcb_ext.h> #include <machine/vm86.h> diff --git a/sys/amd64/include/cpu.h b/sys/amd64/include/cpu.h index 74f21b9..29b34ee 100644 --- a/sys/amd64/include/cpu.h +++ b/sys/amd64/include/cpu.h @@ -65,8 +65,6 @@ #define CLKF_USERMODE(framep) \ ((ISPL((framep)->cf_cs) == SEL_UPL) || ((framep)->cf_eflags & PSL_VM)) - -#define CLKF_INTR(framep) (curproc->p_intr_nesting_level >= 2) #define CLKF_PC(framep) ((framep)->cf_eip) /* diff --git a/sys/amd64/include/mptable.h b/sys/amd64/include/mptable.h index 28a5c72..21e6b6e 100644 --- a/sys/amd64/include/mptable.h +++ b/sys/amd64/include/mptable.h @@ -42,6 +42,7 @@ #include <sys/malloc.h> #include <sys/memrange.h> #include <sys/mutex.h> +#include <sys/smp.h> #include <sys/dkstat.h> #include <sys/cons.h> /* cngetc() */ @@ -57,9 +58,9 @@ #include <sys/gmon.h> #endif -#include <machine/smp.h> #include <machine/apic.h> #include <machine/atomic.h> +#include <machine/cpu.h> #include <machine/cpufunc.h> #include <machine/ipl.h> #include <machine/mpapic.h> @@ -243,7 +244,6 @@ int current_postcode; extern struct region_descriptor r_gdt, r_idt; int bsp_apic_ready = 0; /* flags useability of BSP apic */ -int mp_ncpus; /* # of CPUs, including BSP */ int mp_naps; /* # of Applications processors */ int mp_nbusses; /* # of busses */ int mp_napics; /* # of IO APICs */ @@ -273,9 +273,6 @@ int io_num_to_apic_id[NAPICID]; int apic_id_to_logical[NAPICID]; -/* Bitmap of all available CPUs */ -u_int all_cpus; - /* AP uses this during bootstrap. Do not staticize. */ char *bootSTK; static int bootAP; @@ -288,28 +285,9 @@ extern pt_entry_t *SMPpt; struct pcb stoppcbs[MAXCPU]; -int smp_started; /* has the system started? */ -int smp_active = 0; /* are the APs allowed to run? */ -SYSCTL_INT(_machdep, OID_AUTO, smp_active, CTLFLAG_RW, &smp_active, 0, ""); - -/* XXX maybe should be hw.ncpu */ -static int smp_cpus = 1; /* how many cpu's running */ -SYSCTL_INT(_machdep, OID_AUTO, smp_cpus, CTLFLAG_RD, &smp_cpus, 0, ""); - int invltlb_ok = 0; /* throttle smp_invltlb() till safe */ SYSCTL_INT(_machdep, OID_AUTO, invltlb_ok, CTLFLAG_RW, &invltlb_ok, 0, ""); -/* Enable forwarding of a signal to a process running on a different CPU */ -static int forward_signal_enabled = 1; -SYSCTL_INT(_machdep, OID_AUTO, forward_signal_enabled, CTLFLAG_RW, - &forward_signal_enabled, 0, ""); - -/* Enable forwarding of roundrobin to all other cpus */ -static int forward_roundrobin_enabled = 1; -SYSCTL_INT(_machdep, OID_AUTO, forward_roundrobin_enabled, CTLFLAG_RW, - &forward_roundrobin_enabled, 0, ""); - - /* * Local data and functions. */ @@ -354,9 +332,6 @@ struct mtx mcount_mtx; struct mtx com_mtx; #endif /* USE_COMLOCK */ -/* lock around the MP rendezvous */ -static struct mtx smp_rv_mtx; - static void init_locks(void) { @@ -367,13 +342,9 @@ init_locks(void) */ mtx_init(&mcount_mtx, "mcount", MTX_DEF); - mtx_init(&smp_rv_mtx, "smp rendezvous", MTX_SPIN); - #ifdef USE_COMLOCK mtx_init(&com_mtx, "com", MTX_SPIN); #endif /* USE_COMLOCK */ - - mtx_init(&ap_boot_mtx, "ap boot", MTX_SPIN); } /* @@ -397,8 +368,8 @@ mp_bootaddress(u_int basemem) /* * Look for an Intel MP spec table (ie, SMP capable hardware). */ -int -mp_probe(void) +void +i386_mp_probe(void) { int x; u_long segment; @@ -427,7 +398,7 @@ mp_probe(void) /* nothing found */ mpfps = (mpfps_t)0; mp_capable = 0; - return 0; + return; found: /* calculate needed resources */ @@ -436,15 +407,19 @@ found: /* flag fact that we are running multiple processors */ mp_capable = 1; - return 1; } +int +cpu_mp_probe(void) +{ + return (mp_capable); +} /* * Initialize the SMP hardware and the APIC and start up the AP's. */ void -mp_start(void) +cpu_mp_start(void) { POSTCODE(MP_START_POST); @@ -453,6 +428,8 @@ mp_start(void) mp_enable(boot_address); else panic("MP hardware not found!"); + + cpu_setregs(); } @@ -460,13 +437,12 @@ mp_start(void) * Print various information about the SMP system hardware and setup. */ void -mp_announce(void) +cpu_mp_announce(void) { int x; POSTCODE(MP_ANNOUNCE_POST); - printf("FreeBSD/SMP: Multiprocessor motherboard\n"); printf(" cpu0 (BSP): apic id: %2d", CPU_TO_ID(0)); printf(", version: 0x%08x", cpu_apic_versions[0]); printf(", at 0x%08x\n", cpu_apic_address); @@ -623,8 +599,12 @@ mp_enable(u_int boot_addr) setidt(XINVLTLB_OFFSET, Xinvltlb, SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); - /* install an inter-CPU IPI for reading processor state */ - setidt(XCPUCHECKSTATE_OFFSET, Xcpucheckstate, + /* install an inter-CPU IPI for forwarding hardclock() */ + setidt(XHARDCLOCK_OFFSET, Xhardclock, + SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); + + /* install an inter-CPU IPI for forwarding statclock() */ + setidt(XSTATCLOCK_OFFSET, Xstatclock, SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); /* install an inter-CPU IPI for all-CPU rendezvous */ @@ -1938,6 +1918,8 @@ start_all_aps(u_int boot_addr) POSTCODE(START_ALL_APS_POST); + mtx_init(&ap_boot_mtx, "ap boot", MTX_SPIN); + /* initialize BSP's local APIC */ apic_initialize(); bsp_apic_ready = 1; @@ -1985,8 +1967,8 @@ start_all_aps(u_int boot_addr) (PG_V | PG_RW | vtophys(PAGE_SIZE * i + stack)); /* prime data page for it to use */ - SLIST_INSERT_HEAD(&cpuhead, gd, gd_allcpu); gd->gd_cpuid = x; + globaldata_register(gd); /* setup a vector to our boot code */ *((volatile u_short *) WARMBOOT_OFF) = WARMBOOT_TARGET; @@ -2328,472 +2310,65 @@ ap_init(void) panic("scheduler returned us to ap_init"); } -#define CHECKSTATE_USER 0 -#define CHECKSTATE_SYS 1 -#define CHECKSTATE_INTR 2 - -/* Do not staticize. Used from apic_vector.s */ -struct proc* checkstate_curproc[MAXCPU]; -int checkstate_cpustate[MAXCPU]; -u_long checkstate_pc[MAXCPU]; - -#define PC_TO_INDEX(pc, prof) \ - ((int)(((u_quad_t)((pc) - (prof)->pr_off) * \ - (u_quad_t)((prof)->pr_scale)) >> 16) & ~1) - -static void -addupc_intr_forwarded(struct proc *p, int id, int *astmap) -{ - int i; - struct uprof *prof; - u_long pc; - - pc = checkstate_pc[id]; - prof = &p->p_stats->p_prof; - if (pc >= prof->pr_off && - (i = PC_TO_INDEX(pc, prof)) < prof->pr_size) { - mtx_assert(&sched_lock, MA_OWNED); - if ((p->p_sflag & PS_OWEUPC) == 0) { - prof->pr_addr = pc; - prof->pr_ticks = 1; - p->p_sflag |= PS_OWEUPC; - } - *astmap |= (1 << id); - } -} - -static void -forwarded_statclock(int id, int pscnt, int *astmap) -{ - struct pstats *pstats; - long rss; - struct rusage *ru; - struct vmspace *vm; - int cpustate; - struct proc *p; -#ifdef GPROF - register struct gmonparam *g; - int i; -#endif - - mtx_assert(&sched_lock, MA_OWNED); - p = checkstate_curproc[id]; - cpustate = checkstate_cpustate[id]; - - /* XXX */ - if (p->p_ithd) - cpustate = CHECKSTATE_INTR; - else if (p == SMP_prvspace[id].globaldata.gd_idleproc) - cpustate = CHECKSTATE_SYS; - - switch (cpustate) { - case CHECKSTATE_USER: - if (p->p_sflag & PS_PROFIL) - addupc_intr_forwarded(p, id, astmap); - if (pscnt > 1) - return; - p->p_uticks++; - if (p->p_nice > NZERO) - cp_time[CP_NICE]++; - else - cp_time[CP_USER]++; - break; - case CHECKSTATE_SYS: -#ifdef GPROF - /* - * Kernel statistics are just like addupc_intr, only easier. - */ - g = &_gmonparam; - if (g->state == GMON_PROF_ON) { - i = checkstate_pc[id] - g->lowpc; - if (i < g->textsize) { - i /= HISTFRACTION * sizeof(*g->kcount); - g->kcount[i]++; - } - } -#endif - if (pscnt > 1) - return; - - p->p_sticks++; - if (p == SMP_prvspace[id].globaldata.gd_idleproc) - cp_time[CP_IDLE]++; - else - cp_time[CP_SYS]++; - break; - case CHECKSTATE_INTR: - default: -#ifdef GPROF - /* - * Kernel statistics are just like addupc_intr, only easier. - */ - g = &_gmonparam; - if (g->state == GMON_PROF_ON) { - i = checkstate_pc[id] - g->lowpc; - if (i < g->textsize) { - i /= HISTFRACTION * sizeof(*g->kcount); - g->kcount[i]++; - } - } -#endif - if (pscnt > 1) - return; - KASSERT(p != NULL, ("NULL process in interrupt state")); - p->p_iticks++; - cp_time[CP_INTR]++; - } - - schedclock(p); - - /* Update resource usage integrals and maximums. */ - if ((pstats = p->p_stats) != NULL && - (ru = &pstats->p_ru) != NULL && - (vm = p->p_vmspace) != NULL) { - ru->ru_ixrss += pgtok(vm->vm_tsize); - ru->ru_idrss += pgtok(vm->vm_dsize); - ru->ru_isrss += pgtok(vm->vm_ssize); - rss = pgtok(vmspace_resident_count(vm)); - if (ru->ru_maxrss < rss) - ru->ru_maxrss = rss; - } -} - +/* + * For statclock, we send an IPI to all CPU's to have them call this + * function. + */ void -forward_statclock(int pscnt) +forwarded_statclock(struct trapframe frame) { - int map; - int id; - int i; - /* Kludge. We don't yet have separate locks for the interrupts - * and the kernel. This means that we cannot let the other processors - * handle complex interrupts while inhibiting them from entering - * the kernel in a non-interrupt context. - * - * What we can do, without changing the locking mechanisms yet, - * is letting the other processors handle a very simple interrupt - * (wich determines the processor states), and do the main - * work ourself. - */ - - CTR1(KTR_SMP, "forward_statclock(%d)", pscnt); - - if (!smp_started || !invltlb_ok || cold || panicstr) - return; - - /* Step 1: Probe state (user, cpu, interrupt, spinlock, idle ) */ - - map = PCPU_GET(other_cpus) & ~stopped_cpus ; - checkstate_probed_cpus = 0; - if (map != 0) - ipi_selected(map, IPI_CHECKSTATE); - - i = 0; - while (checkstate_probed_cpus != map) { - /* spin */ - i++; - if (i == 100000) { -#ifdef DIAGNOSTIC - printf("forward_statclock: checkstate %x\n", - checkstate_probed_cpus); -#endif - break; - } - } - - /* - * Step 2: walk through other processors processes, update ticks and - * profiling info. - */ - - map = 0; - for (id = 0; id < mp_ncpus; id++) { - if (id == PCPU_GET(cpuid)) - continue; - if (((1 << id) & checkstate_probed_cpus) == 0) - continue; - forwarded_statclock(id, pscnt, &map); - } - if (map != 0) { - checkstate_need_ast |= map; - ipi_selected(map, IPI_AST); - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#ifdef DIAGNOSTIC - printf("forward_statclock: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } - } + mtx_lock_spin(&sched_lock); + statclock_process(curproc, TRAPF_PC(&frame), TRAPF_USERMODE(&frame)); + mtx_unlock_spin(&sched_lock); } -void -forward_hardclock(int pscnt) +void +forward_statclock(void) { int map; - int id; - struct proc *p; - struct pstats *pstats; - int i; - - /* Kludge. We don't yet have separate locks for the interrupts - * and the kernel. This means that we cannot let the other processors - * handle complex interrupts while inhibiting them from entering - * the kernel in a non-interrupt context. - * - * What we can do, without changing the locking mechanisms yet, - * is letting the other processors handle a very simple interrupt - * (wich determines the processor states), and do the main - * work ourself. - */ - CTR1(KTR_SMP, "forward_hardclock(%d)", pscnt); + CTR0(KTR_SMP, "forward_statclock"); if (!smp_started || !invltlb_ok || cold || panicstr) return; - /* Step 1: Probe state (user, cpu, interrupt, spinlock, idle) */ - map = PCPU_GET(other_cpus) & ~stopped_cpus ; - checkstate_probed_cpus = 0; if (map != 0) - ipi_selected(map, IPI_CHECKSTATE); - - i = 0; - while (checkstate_probed_cpus != map) { - /* spin */ - i++; - if (i == 100000) { -#ifdef DIAGNOSTIC - printf("forward_hardclock: checkstate %x\n", - checkstate_probed_cpus); -#endif - break; - } - } - - /* - * Step 2: walk through other processors processes, update virtual - * timer and profiling timer. If stathz == 0, also update ticks and - * profiling info. - */ - - map = 0; - for (id = 0; id < mp_ncpus; id++) { - if (id == PCPU_GET(cpuid)) - continue; - if (((1 << id) & checkstate_probed_cpus) == 0) - continue; - p = checkstate_curproc[id]; - if (p) { - pstats = p->p_stats; - if (checkstate_cpustate[id] == CHECKSTATE_USER && - timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) && - itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) { - p->p_sflag |= PS_ALRMPEND; - map |= (1 << id); - } - if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) && - itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) { - p->p_sflag |= PS_PROFPEND; - map |= (1 << id); - } - } - if (stathz == 0) { - forwarded_statclock( id, pscnt, &map); - } - } - if (map != 0) { - checkstate_need_ast |= map; - ipi_selected(map, IPI_AST); - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#ifdef DIAGNOSTIC - printf("forward_hardclock: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } - } + ipi_selected(map, IPI_STATCLOCK); } -void -forward_signal(struct proc *p) +/* + * For each hardclock(), we send an IPI to all other CPU's to have them + * execute this function. It would be nice to reduce contention on + * sched_lock if we could simply peek at the CPU to determine the user/kernel + * state and call hardclock_process() on the CPU receiving the clock interrupt + * and then just use a simple IPI to handle any ast's if needed. + */ +void +forwarded_hardclock(struct trapframe frame) { - int map; - int id; - int i; - - /* Kludge. We don't yet have separate locks for the interrupts - * and the kernel. This means that we cannot let the other processors - * handle complex interrupts while inhibiting them from entering - * the kernel in a non-interrupt context. - * - * What we can do, without changing the locking mechanisms yet, - * is letting the other processors handle a very simple interrupt - * (wich determines the processor states), and do the main - * work ourself. - */ - - CTR1(KTR_SMP, "forward_signal(%p)", p); - if (!smp_started || !invltlb_ok || cold || panicstr) - return; - if (!forward_signal_enabled) - return; mtx_lock_spin(&sched_lock); - while (1) { - if (p->p_stat != SRUN) { - mtx_unlock_spin(&sched_lock); - return; - } - id = p->p_oncpu; - mtx_unlock_spin(&sched_lock); - if (id == 0xff) - return; - map = (1<<id); - checkstate_need_ast |= map; - ipi_selected(map, IPI_AST); - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#if 0 - printf("forward_signal: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } - mtx_lock_spin(&sched_lock); - if (id == p->p_oncpu) { - mtx_unlock_spin(&sched_lock); - return; - } - } + hardclock_process(curproc, TRAPF_USERMODE(&frame)); + mtx_unlock_spin(&sched_lock); } -void -forward_roundrobin(void) +void +forward_hardclock(void) { u_int map; - int i; - CTR0(KTR_SMP, "forward_roundrobin()"); + CTR0(KTR_SMP, "forward_hardclock"); if (!smp_started || !invltlb_ok || cold || panicstr) return; - if (!forward_roundrobin_enabled) - return; - resched_cpus |= PCPU_GET(other_cpus); - map = PCPU_GET(other_cpus) & ~stopped_cpus ; -#if 1 - ipi_selected(map, IPI_AST); -#else - ipi_all_but_self(IPI_AST); -#endif - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#if 0 - printf("forward_roundrobin: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } -} - -/* - * When called the executing CPU will send an IPI to all other CPUs - * requesting that they halt execution. - * - * Usually (but not necessarily) called with 'other_cpus' as its arg. - * - * - Signals all CPUs in map to stop. - * - Waits for each to stop. - * - * Returns: - * -1: error - * 0: NA - * 1: ok - * - * XXX FIXME: this is not MP-safe, needs a lock to prevent multiple CPUs - * from executing at same time. - */ -int -stop_cpus(u_int map) -{ - int count = 0; - - if (!smp_started) - return 0; - - /* send the Xcpustop IPI to all CPUs in map */ - ipi_selected(map, IPI_STOP); - - while (count++ < 100000 && (stopped_cpus & map) != map) - /* spin */ ; - -#ifdef DIAGNOSTIC - if ((stopped_cpus & map) != map) - printf("Warning: CPUs 0x%x did not stop!\n", - (~(stopped_cpus & map)) & map); -#endif - - return 1; -} - - -/* - * Called by a CPU to restart stopped CPUs. - * - * Usually (but not necessarily) called with 'stopped_cpus' as its arg. - * - * - Signals all CPUs in map to restart. - * - Waits for each to restart. - * - * Returns: - * -1: error - * 0: NA - * 1: ok - */ -int -restart_cpus(u_int map) -{ - int count = 0; - - if (!smp_started) - return 0; - - started_cpus = map; /* signal other cpus to restart */ - - /* wait for each to clear its bit */ - while (count++ < 100000 && (stopped_cpus & map) != 0) - /* spin */ ; -#ifdef DIAGNOSTIC - if ((stopped_cpus & map) != 0) - printf("Warning: CPUs 0x%x did not restart!\n", - (~(stopped_cpus & map)) & map); -#endif - - return 1; + map = PCPU_GET(other_cpus) & ~stopped_cpus ; + if (map != 0) + ipi_selected(map, IPI_HARDCLOCK); } - #ifdef APIC_INTR_REORDER /* * Maintain mapping from softintr vector to isr bit in local apic. @@ -2811,73 +2386,6 @@ set_lapic_isrloc(int intr, int vector) #endif /* - * All-CPU rendezvous. CPUs are signalled, all execute the setup function - * (if specified), rendezvous, execute the action function (if specified), - * rendezvous again, execute the teardown function (if specified), and then - * resume. - * - * Note that the supplied external functions _must_ be reentrant and aware - * that they are running in parallel and in an unknown lock context. - */ -static void (*smp_rv_setup_func)(void *arg); -static void (*smp_rv_action_func)(void *arg); -static void (*smp_rv_teardown_func)(void *arg); -static void *smp_rv_func_arg; -static volatile int smp_rv_waiters[2]; - -void -smp_rendezvous_action(void) -{ - /* setup function */ - if (smp_rv_setup_func != NULL) - smp_rv_setup_func(smp_rv_func_arg); - /* spin on entry rendezvous */ - atomic_add_int(&smp_rv_waiters[0], 1); - while (smp_rv_waiters[0] < mp_ncpus) - ; - /* action function */ - if (smp_rv_action_func != NULL) - smp_rv_action_func(smp_rv_func_arg); - /* spin on exit rendezvous */ - atomic_add_int(&smp_rv_waiters[1], 1); - while (smp_rv_waiters[1] < mp_ncpus) - ; - /* teardown function */ - if (smp_rv_teardown_func != NULL) - smp_rv_teardown_func(smp_rv_func_arg); -} - -void -smp_rendezvous(void (* setup_func)(void *), - void (* action_func)(void *), - void (* teardown_func)(void *), - void *arg) -{ - - /* obtain rendezvous lock */ - mtx_lock_spin(&smp_rv_mtx); - - /* set static function pointers */ - smp_rv_setup_func = setup_func; - smp_rv_action_func = action_func; - smp_rv_teardown_func = teardown_func; - smp_rv_func_arg = arg; - smp_rv_waiters[0] = 0; - smp_rv_waiters[1] = 0; - - /* - * signal other processors, which will enter the IPI with interrupts off - */ - ipi_all_but_self(IPI_RENDEZVOUS); - - /* call executor function */ - smp_rendezvous_action(); - - /* release lock */ - mtx_unlock_spin(&smp_rv_mtx); -} - -/* * send an IPI to a set of cpus. */ void diff --git a/sys/amd64/include/pcpu.h b/sys/amd64/include/pcpu.h index e713227..ef82a66 100644 --- a/sys/amd64/include/pcpu.h +++ b/sys/amd64/include/pcpu.h @@ -74,9 +74,6 @@ struct globaldata { #endif }; -SLIST_HEAD(cpuhead, globaldata); -extern struct cpuhead cpuhead; - #ifdef SMP /* * This is the upper (0xff800000) address space layout that is per-cpu. diff --git a/sys/amd64/include/smp.h b/sys/amd64/include/smp.h index 9f9149e..f699e1f 100644 --- a/sys/amd64/include/smp.h +++ b/sys/amd64/include/smp.h @@ -41,7 +41,21 @@ extern int current_postcode; /** XXX currently in mp_machdep.c */ outb(0x80, current_postcode) +#include <sys/bus.h> /* XXX */ #include <machine/apic.h> +#include <machine/frame.h> +#include <i386/isa/icu.h> +#include <i386/isa/intr_machdep.h> + +/* + * Interprocessor interrupts for SMP. + */ +#define IPI_INVLTLB XINVLTLB_OFFSET +#define IPI_RENDEZVOUS XRENDEZVOUS_OFFSET +#define IPI_AST XCPUAST_OFFSET +#define IPI_STOP XCPUSTOP_OFFSET +#define IPI_HARDCLOCK XHARDCLOCK_OFFSET +#define IPI_STATCLOCK XSTATCLOCK_OFFSET /* global data in mpboot.s */ extern int bootMP_size; @@ -49,18 +63,8 @@ extern int bootMP_size; /* functions in mpboot.s */ void bootMP __P((void)); -/* global data in apic_vector.s */ -extern volatile u_int stopped_cpus; -extern volatile u_int started_cpus; - -extern volatile u_int checkstate_probed_cpus; -extern volatile u_int checkstate_need_ast; -extern volatile u_int resched_cpus; -extern void (*cpustop_restartfunc) __P((void)); - /* global data in mp_machdep.c */ extern int bsp_apic_ready; -extern int mp_ncpus; extern int mp_naps; extern int mp_nbusses; extern int mp_napics; @@ -81,14 +85,11 @@ struct apic_intmapinfo { int redirindex; }; extern struct apic_intmapinfo int_to_apicintpin[]; -extern u_int all_cpus; extern struct pcb stoppcbs[]; /* functions in mp_machdep.c */ +void i386_mp_probe __P((void)); u_int mp_bootaddress __P((u_int)); -int mp_probe __P((void)); -void mp_start __P((void)); -void mp_announce __P((void)); u_int isa_apic_mask __P((u_int)); int isa_apic_irq __P((int)); int pci_apic_irq __P((int, int, int)); @@ -107,20 +108,17 @@ void revoke_apic_irq __P((int irq)); void bsp_apic_configure __P((void)); void init_secondary __P((void)); void smp_invltlb __P((void)); -int stop_cpus __P((u_int)); -int restart_cpus __P((u_int)); -void forward_statclock __P((int pscnt)); -void forward_hardclock __P((int pscnt)); -void forward_signal __P((struct proc *)); -void forward_roundrobin __P((void)); +void forward_statclock __P((void)); +void forwarded_statclock __P((struct trapframe frame)); +void forward_hardclock __P((void)); +void forwarded_hardclock __P((struct trapframe frame)); +void ipi_selected __P((u_int cpus, u_int ipi)); +void ipi_all __P((u_int ipi)); +void ipi_all_but_self __P((u_int ipi)); +void ipi_self __P((u_int ipi)); #ifdef APIC_INTR_REORDER void set_lapic_isrloc __P((int, int)); #endif /* APIC_INTR_REORDER */ -void smp_rendezvous_action __P((void)); -void smp_rendezvous __P((void (*)(void *), - void (*)(void *), - void (*)(void *), - void *arg)); /* global data in mpapic.c */ extern volatile lapic_t lapic; @@ -147,8 +145,6 @@ void io_apic_write __P((int, int, u_int)); /* global data in init_smp.c */ extern int invltlb_ok; -extern int smp_active; -extern int smp_started; extern volatile int smp_idle_loops; #endif /* !LOCORE */ diff --git a/sys/amd64/isa/clock.c b/sys/amd64/isa/clock.c index 9034f81..ce67390 100644 --- a/sys/amd64/isa/clock.c +++ b/sys/amd64/isa/clock.c @@ -211,6 +211,10 @@ clkintr(struct clockframe frame) mtx_unlock_spin(&clock_lock); } timer_func(&frame); +#ifdef SMP + if (timer_func == hardclock) + forward_hardclock(); +#endif switch (timer0_state) { case RELEASED: @@ -253,6 +257,9 @@ clkintr(struct clockframe frame) timer_func = hardclock; timer0_state = RELEASED; hardclock(&frame); +#ifdef SMP + forward_hardclock(); +#endif } break; } @@ -374,8 +381,12 @@ release_timer2() static void rtcintr(struct clockframe frame) { - while (rtcin(RTC_INTR) & RTCIR_PERIOD) + while (rtcin(RTC_INTR) & RTCIR_PERIOD) { statclock(&frame); +#ifdef SMP + forward_statclock(); +#endif + } } #include "opt_ddb.h" diff --git a/sys/amd64/isa/intr_machdep.h b/sys/amd64/isa/intr_machdep.h index ed58437..6350524 100644 --- a/sys/amd64/isa/intr_machdep.h +++ b/sys/amd64/isa/intr_machdep.h @@ -107,10 +107,11 @@ #define XINVLTLB_OFFSET (ICU_OFFSET + 112) /* inter-cpu clock handling */ -#define XCPUCHECKSTATE_OFFSET (ICU_OFFSET + 113) +#define XHARDCLOCK_OFFSET (ICU_OFFSET + 113) +#define XSTATCLOCK_OFFSET (ICU_OFFSET + 114) /* inter-CPU rendezvous */ -#define XRENDEZVOUS_OFFSET (ICU_OFFSET + 114) +#define XRENDEZVOUS_OFFSET (ICU_OFFSET + 115) /* IPI to generate an additional software trap at the target CPU */ #define XCPUAST_OFFSET (ICU_OFFSET + 48) @@ -173,7 +174,8 @@ inthand_t inthand_t Xinvltlb, /* TLB shootdowns */ - Xcpucheckstate, /* Check cpu state */ + Xhardclock, /* Forward hardclock() */ + Xstatclock, /* Forward statclock() */ Xcpuast, /* Additional software trap on other cpu */ Xcpustop, /* CPU stops & waits for another CPU to restart it */ Xspuriousint, /* handle APIC "spurious INTs" */ diff --git a/sys/conf/files b/sys/conf/files index fe4f634..794d6e6 100644 --- a/sys/conf/files +++ b/sys/conf/files @@ -716,6 +716,7 @@ kern/subr_prof.c standard kern/subr_rman.c standard kern/subr_sbuf.c standard kern/subr_scanf.c standard +kern/subr_smp.c optional smp kern/subr_taskqueue.c standard kern/subr_witness.c optional witness kern/subr_xxx.c standard diff --git a/sys/conf/files.ia64 b/sys/conf/files.ia64 index 70c453f..ec058ef 100644 --- a/sys/conf/files.ia64 +++ b/sys/conf/files.ia64 @@ -36,7 +36,7 @@ ia64/ia64/interrupt.c standard ia64/ia64/machdep.c standard ia64/ia64/mem.c standard ia64/ia64/mountroot.c optional slice -ia64/ia64/mp_machdep.c standard +ia64/ia64/mp_machdep.c optional smp ia64/ia64/pal.s standard ia64/ia64/perfmon.c optional perfmon profiling-routine ia64/ia64/perfmon.c optional perfmon diff --git a/sys/i386/i386/apic_vector.s b/sys/i386/i386/apic_vector.s index 3f0521f..5c68f81 100644 --- a/sys/i386/i386/apic_vector.s +++ b/sys/i386/i386/apic_vector.s @@ -182,7 +182,6 @@ Xspuriousint: iret - /* * Handle TLB shootdowns. */ @@ -211,71 +210,61 @@ Xinvltlb: popl %eax iret - /* - * Executed by a CPU when it receives an Xcpucheckstate IPI from another CPU, - * - * - Stores current cpu state in checkstate_cpustate[cpuid] - * 0 == user, 1 == sys, 2 == intr - * - Stores current process in checkstate_curproc[cpuid] - * - * - Signals its receipt by setting bit cpuid in checkstate_probed_cpus. - * - * stack: 0->ds, 4->fs, 8->ebx, 12->eax, 16->eip, 20->cs, 24->eflags + * Forward hardclock to another CPU. Pushes a trapframe and calls + * forwarded_hardclock(). */ - .text SUPERALIGN_TEXT - .globl Xcpucheckstate - .globl checkstate_cpustate - .globl checkstate_curproc - .globl checkstate_pc -Xcpucheckstate: - pushl %eax - pushl %ebx - pushl %ds /* save current data segment */ - pushl %fs - - movl $KDSEL, %eax - mov %ax, %ds /* use KERNEL data segment */ + .globl Xhardclock +Xhardclock: + PUSH_FRAME + movl $KDSEL, %eax /* reload with kernel's data segment */ + mov %ax, %ds + mov %ax, %es movl $KPSEL, %eax mov %ax, %fs movl $0, lapic+LA_EOI /* End Of Interrupt to APIC */ - movl $0, %ebx - movl 20(%esp), %eax - andl $3, %eax - cmpl $3, %eax - je 1f - testl $PSL_VM, 24(%esp) - jne 1f - incl %ebx /* system or interrupt */ -1: - movl PCPU(CPUID), %eax - movl %ebx, checkstate_cpustate(,%eax,4) - movl PCPU(CURPROC), %ebx - movl %ebx, checkstate_curproc(,%eax,4) - - movl 16(%esp), %ebx - movl %ebx, checkstate_pc(,%eax,4) + movl PCPU(CURPROC),%ebx + incl P_INTR_NESTING_LEVEL(%ebx) + call forwarded_hardclock + decl P_INTR_NESTING_LEVEL(%ebx) + MEXITCOUNT + jmp doreti - lock /* checkstate_probed_cpus |= (1<<id) */ - btsl %eax, checkstate_probed_cpus +/* + * Forward statclock to another CPU. Pushes a trapframe and calls + * forwarded_statclock(). + */ + .text + SUPERALIGN_TEXT + .globl Xstatclock +Xstatclock: + PUSH_FRAME + movl $KDSEL, %eax /* reload with kernel's data segment */ + mov %ax, %ds + mov %ax, %es + movl $KPSEL, %eax + mov %ax, %fs - popl %fs - popl %ds /* restore previous data segment */ - popl %ebx - popl %eax - iret + movl $0, lapic+LA_EOI /* End Of Interrupt to APIC */ + FAKE_MCOUNT(13*4(%esp)) + movl PCPU(CURPROC),%ebx + incl P_INTR_NESTING_LEVEL(%ebx) + call forwarded_statclock + decl P_INTR_NESTING_LEVEL(%ebx) + MEXITCOUNT + jmp doreti /* * Executed by a CPU when it receives an Xcpuast IPI from another CPU, * - * - Signals its receipt by clearing bit cpuid in checkstate_need_ast. - * - * - We need a better method of triggering asts on other cpus. + * The other CPU has already executed aston() or need_resched() on our + * current process, so we simply need to ack the interrupt and return + * via doreti to run ast(). */ .text @@ -289,40 +278,12 @@ Xcpuast: movl $KPSEL, %eax mov %ax, %fs - movl PCPU(CPUID), %eax - lock /* checkstate_need_ast &= ~(1<<id) */ - btrl %eax, checkstate_need_ast movl $0, lapic+LA_EOI /* End Of Interrupt to APIC */ - lock - btsl %eax, checkstate_pending_ast - jc 1f - FAKE_MCOUNT(13*4(%esp)) - MTX_LOCK_SPIN(sched_lock, 0) - movl PCPU(CURPROC),%ebx - orl $PS_ASTPENDING, P_SFLAG(%ebx) - - movl PCPU(CPUID), %eax - lock - btrl %eax, checkstate_pending_ast - lock - btrl %eax, CNAME(resched_cpus) - jnc 2f - orl $PS_NEEDRESCHED, P_SFLAG(%ebx) - lock - incl CNAME(want_resched_cnt) -2: - MTX_UNLOCK_SPIN(sched_lock) - lock - incl CNAME(cpuast_cnt) MEXITCOUNT jmp doreti -1: - /* We are already in the process of delivering an ast for this CPU */ - POP_FRAME - iret /* * Executed by a CPU when it receives an Xcpustop IPI from another CPU, @@ -331,7 +292,6 @@ Xcpuast: * - Waits for permission to restart. * - Signals its restart. */ - .text SUPERALIGN_TEXT .globl Xcpustop @@ -357,20 +317,19 @@ Xcpustop: pushl %eax call CNAME(savectx) /* Save process context */ addl $4, %esp - movl PCPU(CPUID), %eax lock - btsl %eax, stopped_cpus /* stopped_cpus |= (1<<id) */ + btsl %eax, CNAME(stopped_cpus) /* stopped_cpus |= (1<<id) */ 1: - btl %eax, started_cpus /* while (!(started_cpus & (1<<id))) */ + btl %eax, CNAME(started_cpus) /* while (!(started_cpus & (1<<id))) */ jnc 1b lock - btrl %eax, started_cpus /* started_cpus &= ~(1<<id) */ + btrl %eax, CNAME(started_cpus) /* started_cpus &= ~(1<<id) */ lock - btrl %eax, stopped_cpus /* stopped_cpus &= ~(1<<id) */ + btrl %eax, CNAME(stopped_cpus) /* stopped_cpus &= ~(1<<id) */ test %eax, %eax jnz 2f @@ -492,34 +451,6 @@ _xhits: .space (NCPU * 4), 0 #endif /* COUNT_XINVLTLB_HITS */ -/* variables used by stop_cpus()/restart_cpus()/Xcpustop */ - .globl stopped_cpus, started_cpus -stopped_cpus: - .long 0 -started_cpus: - .long 0 - - .globl checkstate_probed_cpus -checkstate_probed_cpus: - .long 0 - .globl checkstate_need_ast -checkstate_need_ast: - .long 0 -checkstate_pending_ast: - .long 0 - .globl CNAME(resched_cpus) - .globl CNAME(want_resched_cnt) - .globl CNAME(cpuast_cnt) - .globl CNAME(cpustop_restartfunc) -CNAME(resched_cpus): - .long 0 -CNAME(want_resched_cnt): - .long 0 -CNAME(cpuast_cnt): - .long 0 -CNAME(cpustop_restartfunc): - .long 0 - .globl apic_pin_trigger apic_pin_trigger: .long 0 diff --git a/sys/i386/i386/db_interface.c b/sys/i386/i386/db_interface.c index 8c7dc0b..cdaefe4 100644 --- a/sys/i386/i386/db_interface.c +++ b/sys/i386/i386/db_interface.c @@ -37,10 +37,10 @@ #include <sys/linker_set.h> #include <sys/lock.h> #include <sys/proc.h> +#include <sys/smp.h> #include <machine/cpu.h> #ifdef SMP -#include <machine/smp.h> #include <machine/smptests.h> /** CPUSTOP_ON_DDBBREAK */ #endif @@ -334,43 +334,44 @@ Debugger(msg) DB_SHOW_COMMAND(pcpu, db_show_pcpu) { struct globaldata *gd; +#ifdef SMP int id; if (have_addr) id = ((addr >> 4) % 16) * 10 + (addr % 16); else id = PCPU_GET(cpuid); - SLIST_FOREACH(gd, &cpuhead, gd_allcpu) { - if (gd->gd_cpuid == id) - break; - } - if (gd == NULL) + gd = globaldata_find(id); + if (gd == NULL) { db_printf("CPU %d not found\n", id); - else { - db_printf("cpuid = %d\n", gd->gd_cpuid); - db_printf("curproc = "); - if (gd->gd_curproc != NULL) - db_printf("%p: pid %d \"%s\"\n", gd->gd_curproc, - gd->gd_curproc->p_pid, gd->gd_curproc->p_comm); - else - db_printf("none\n"); - db_printf("curpcb = %p\n", gd->gd_curpcb); - db_printf("npxproc = "); - if (gd->gd_npxproc != NULL) - db_printf("%p: pid %d \"%s\"\n", gd->gd_npxproc, - gd->gd_npxproc->p_pid, gd->gd_npxproc->p_comm); - else - db_printf("none\n"); - db_printf("idleproc = "); - if (gd->gd_idleproc != NULL) - db_printf("%p: pid %d \"%s\"\n", gd->gd_idleproc, - gd->gd_idleproc->p_pid, gd->gd_idleproc->p_comm); - else - db_printf("none\n"); + return; + } +#else + gd = GLOBALDATA; +#endif + db_printf("cpuid = %d\n", gd->gd_cpuid); + db_printf("curproc = "); + if (gd->gd_curproc != NULL) + db_printf("%p: pid %d \"%s\"\n", gd->gd_curproc, + gd->gd_curproc->p_pid, gd->gd_curproc->p_comm); + else + db_printf("none\n"); + db_printf("curpcb = %p\n", gd->gd_curpcb); + db_printf("npxproc = "); + if (gd->gd_npxproc != NULL) + db_printf("%p: pid %d \"%s\"\n", gd->gd_npxproc, + gd->gd_npxproc->p_pid, gd->gd_npxproc->p_comm); + else + db_printf("none\n"); + db_printf("idleproc = "); + if (gd->gd_idleproc != NULL) + db_printf("%p: pid %d \"%s\"\n", gd->gd_idleproc, + gd->gd_idleproc->p_pid, gd->gd_idleproc->p_comm); + else + db_printf("none\n"); #ifdef WITNESS - db_printf("spin locks held:\n"); - witness_list_locks(&gd->gd_spinlocks); + db_printf("spin locks held:\n"); + witness_list_locks(&gd->gd_spinlocks); #endif - } } diff --git a/sys/i386/i386/i686_mem.c b/sys/i386/i386/i686_mem.c index 7c3ae75..2d687af 100644 --- a/sys/i386/i386/i686_mem.c +++ b/sys/i386/i386/i686_mem.c @@ -31,14 +31,11 @@ #include <sys/systm.h> #include <sys/malloc.h> #include <sys/memrange.h> +#include <sys/smp.h> #include <machine/md_var.h> #include <machine/specialreg.h> -#ifdef SMP -#include <machine/smp.h> -#endif - /* * i686 memory range operations * diff --git a/sys/i386/i386/machdep.c b/sys/i386/i386/machdep.c index dbdc61a..c89852a 100644 --- a/sys/i386/i386/machdep.c +++ b/sys/i386/i386/machdep.c @@ -66,6 +66,7 @@ #include <sys/bio.h> #include <sys/buf.h> #include <sys/reboot.h> +#include <sys/smp.h> #include <sys/callout.h> #include <sys/msgbuf.h> #include <sys/sysent.h> @@ -104,9 +105,6 @@ #include <machine/globaldata.h> #include <machine/globals.h> #include <machine/intrcnt.h> -#ifdef SMP -#include <machine/smp.h> -#endif #ifdef PERFMON #include <machine/perfmon.h> #endif @@ -249,8 +247,6 @@ static struct trapframe proc0_tf; static struct globaldata __globaldata; #endif -struct cpuhead cpuhead; - struct mtx sched_lock; struct mtx Giant; @@ -441,17 +437,12 @@ again: bufinit(); vm_pager_bufferinit(); - SLIST_INIT(&cpuhead); - SLIST_INSERT_HEAD(&cpuhead, GLOBALDATA, gd_allcpu); - #ifdef SMP - /* - * OK, enough kmem_alloc/malloc state should be up, lets get on with it! - */ - mp_start(); /* fire up the APs and APICs */ - mp_announce(); -#endif /* SMP */ + globaldata_register(GLOBALDATA); +#else + /* For SMP, we delay the cpu_setregs() until after SMP startup. */ cpu_setregs(); +#endif } /* @@ -1631,7 +1622,7 @@ physmap_done: physmap[1] = mp_bootaddress(physmap[1] / 1024); /* look for the MP hardware - needed for apic addresses */ - mp_probe(); + i386_mp_probe(); #endif /* diff --git a/sys/i386/i386/mp_machdep.c b/sys/i386/i386/mp_machdep.c index 28a5c72..21e6b6e 100644 --- a/sys/i386/i386/mp_machdep.c +++ b/sys/i386/i386/mp_machdep.c @@ -42,6 +42,7 @@ #include <sys/malloc.h> #include <sys/memrange.h> #include <sys/mutex.h> +#include <sys/smp.h> #include <sys/dkstat.h> #include <sys/cons.h> /* cngetc() */ @@ -57,9 +58,9 @@ #include <sys/gmon.h> #endif -#include <machine/smp.h> #include <machine/apic.h> #include <machine/atomic.h> +#include <machine/cpu.h> #include <machine/cpufunc.h> #include <machine/ipl.h> #include <machine/mpapic.h> @@ -243,7 +244,6 @@ int current_postcode; extern struct region_descriptor r_gdt, r_idt; int bsp_apic_ready = 0; /* flags useability of BSP apic */ -int mp_ncpus; /* # of CPUs, including BSP */ int mp_naps; /* # of Applications processors */ int mp_nbusses; /* # of busses */ int mp_napics; /* # of IO APICs */ @@ -273,9 +273,6 @@ int io_num_to_apic_id[NAPICID]; int apic_id_to_logical[NAPICID]; -/* Bitmap of all available CPUs */ -u_int all_cpus; - /* AP uses this during bootstrap. Do not staticize. */ char *bootSTK; static int bootAP; @@ -288,28 +285,9 @@ extern pt_entry_t *SMPpt; struct pcb stoppcbs[MAXCPU]; -int smp_started; /* has the system started? */ -int smp_active = 0; /* are the APs allowed to run? */ -SYSCTL_INT(_machdep, OID_AUTO, smp_active, CTLFLAG_RW, &smp_active, 0, ""); - -/* XXX maybe should be hw.ncpu */ -static int smp_cpus = 1; /* how many cpu's running */ -SYSCTL_INT(_machdep, OID_AUTO, smp_cpus, CTLFLAG_RD, &smp_cpus, 0, ""); - int invltlb_ok = 0; /* throttle smp_invltlb() till safe */ SYSCTL_INT(_machdep, OID_AUTO, invltlb_ok, CTLFLAG_RW, &invltlb_ok, 0, ""); -/* Enable forwarding of a signal to a process running on a different CPU */ -static int forward_signal_enabled = 1; -SYSCTL_INT(_machdep, OID_AUTO, forward_signal_enabled, CTLFLAG_RW, - &forward_signal_enabled, 0, ""); - -/* Enable forwarding of roundrobin to all other cpus */ -static int forward_roundrobin_enabled = 1; -SYSCTL_INT(_machdep, OID_AUTO, forward_roundrobin_enabled, CTLFLAG_RW, - &forward_roundrobin_enabled, 0, ""); - - /* * Local data and functions. */ @@ -354,9 +332,6 @@ struct mtx mcount_mtx; struct mtx com_mtx; #endif /* USE_COMLOCK */ -/* lock around the MP rendezvous */ -static struct mtx smp_rv_mtx; - static void init_locks(void) { @@ -367,13 +342,9 @@ init_locks(void) */ mtx_init(&mcount_mtx, "mcount", MTX_DEF); - mtx_init(&smp_rv_mtx, "smp rendezvous", MTX_SPIN); - #ifdef USE_COMLOCK mtx_init(&com_mtx, "com", MTX_SPIN); #endif /* USE_COMLOCK */ - - mtx_init(&ap_boot_mtx, "ap boot", MTX_SPIN); } /* @@ -397,8 +368,8 @@ mp_bootaddress(u_int basemem) /* * Look for an Intel MP spec table (ie, SMP capable hardware). */ -int -mp_probe(void) +void +i386_mp_probe(void) { int x; u_long segment; @@ -427,7 +398,7 @@ mp_probe(void) /* nothing found */ mpfps = (mpfps_t)0; mp_capable = 0; - return 0; + return; found: /* calculate needed resources */ @@ -436,15 +407,19 @@ found: /* flag fact that we are running multiple processors */ mp_capable = 1; - return 1; } +int +cpu_mp_probe(void) +{ + return (mp_capable); +} /* * Initialize the SMP hardware and the APIC and start up the AP's. */ void -mp_start(void) +cpu_mp_start(void) { POSTCODE(MP_START_POST); @@ -453,6 +428,8 @@ mp_start(void) mp_enable(boot_address); else panic("MP hardware not found!"); + + cpu_setregs(); } @@ -460,13 +437,12 @@ mp_start(void) * Print various information about the SMP system hardware and setup. */ void -mp_announce(void) +cpu_mp_announce(void) { int x; POSTCODE(MP_ANNOUNCE_POST); - printf("FreeBSD/SMP: Multiprocessor motherboard\n"); printf(" cpu0 (BSP): apic id: %2d", CPU_TO_ID(0)); printf(", version: 0x%08x", cpu_apic_versions[0]); printf(", at 0x%08x\n", cpu_apic_address); @@ -623,8 +599,12 @@ mp_enable(u_int boot_addr) setidt(XINVLTLB_OFFSET, Xinvltlb, SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); - /* install an inter-CPU IPI for reading processor state */ - setidt(XCPUCHECKSTATE_OFFSET, Xcpucheckstate, + /* install an inter-CPU IPI for forwarding hardclock() */ + setidt(XHARDCLOCK_OFFSET, Xhardclock, + SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); + + /* install an inter-CPU IPI for forwarding statclock() */ + setidt(XSTATCLOCK_OFFSET, Xstatclock, SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); /* install an inter-CPU IPI for all-CPU rendezvous */ @@ -1938,6 +1918,8 @@ start_all_aps(u_int boot_addr) POSTCODE(START_ALL_APS_POST); + mtx_init(&ap_boot_mtx, "ap boot", MTX_SPIN); + /* initialize BSP's local APIC */ apic_initialize(); bsp_apic_ready = 1; @@ -1985,8 +1967,8 @@ start_all_aps(u_int boot_addr) (PG_V | PG_RW | vtophys(PAGE_SIZE * i + stack)); /* prime data page for it to use */ - SLIST_INSERT_HEAD(&cpuhead, gd, gd_allcpu); gd->gd_cpuid = x; + globaldata_register(gd); /* setup a vector to our boot code */ *((volatile u_short *) WARMBOOT_OFF) = WARMBOOT_TARGET; @@ -2328,472 +2310,65 @@ ap_init(void) panic("scheduler returned us to ap_init"); } -#define CHECKSTATE_USER 0 -#define CHECKSTATE_SYS 1 -#define CHECKSTATE_INTR 2 - -/* Do not staticize. Used from apic_vector.s */ -struct proc* checkstate_curproc[MAXCPU]; -int checkstate_cpustate[MAXCPU]; -u_long checkstate_pc[MAXCPU]; - -#define PC_TO_INDEX(pc, prof) \ - ((int)(((u_quad_t)((pc) - (prof)->pr_off) * \ - (u_quad_t)((prof)->pr_scale)) >> 16) & ~1) - -static void -addupc_intr_forwarded(struct proc *p, int id, int *astmap) -{ - int i; - struct uprof *prof; - u_long pc; - - pc = checkstate_pc[id]; - prof = &p->p_stats->p_prof; - if (pc >= prof->pr_off && - (i = PC_TO_INDEX(pc, prof)) < prof->pr_size) { - mtx_assert(&sched_lock, MA_OWNED); - if ((p->p_sflag & PS_OWEUPC) == 0) { - prof->pr_addr = pc; - prof->pr_ticks = 1; - p->p_sflag |= PS_OWEUPC; - } - *astmap |= (1 << id); - } -} - -static void -forwarded_statclock(int id, int pscnt, int *astmap) -{ - struct pstats *pstats; - long rss; - struct rusage *ru; - struct vmspace *vm; - int cpustate; - struct proc *p; -#ifdef GPROF - register struct gmonparam *g; - int i; -#endif - - mtx_assert(&sched_lock, MA_OWNED); - p = checkstate_curproc[id]; - cpustate = checkstate_cpustate[id]; - - /* XXX */ - if (p->p_ithd) - cpustate = CHECKSTATE_INTR; - else if (p == SMP_prvspace[id].globaldata.gd_idleproc) - cpustate = CHECKSTATE_SYS; - - switch (cpustate) { - case CHECKSTATE_USER: - if (p->p_sflag & PS_PROFIL) - addupc_intr_forwarded(p, id, astmap); - if (pscnt > 1) - return; - p->p_uticks++; - if (p->p_nice > NZERO) - cp_time[CP_NICE]++; - else - cp_time[CP_USER]++; - break; - case CHECKSTATE_SYS: -#ifdef GPROF - /* - * Kernel statistics are just like addupc_intr, only easier. - */ - g = &_gmonparam; - if (g->state == GMON_PROF_ON) { - i = checkstate_pc[id] - g->lowpc; - if (i < g->textsize) { - i /= HISTFRACTION * sizeof(*g->kcount); - g->kcount[i]++; - } - } -#endif - if (pscnt > 1) - return; - - p->p_sticks++; - if (p == SMP_prvspace[id].globaldata.gd_idleproc) - cp_time[CP_IDLE]++; - else - cp_time[CP_SYS]++; - break; - case CHECKSTATE_INTR: - default: -#ifdef GPROF - /* - * Kernel statistics are just like addupc_intr, only easier. - */ - g = &_gmonparam; - if (g->state == GMON_PROF_ON) { - i = checkstate_pc[id] - g->lowpc; - if (i < g->textsize) { - i /= HISTFRACTION * sizeof(*g->kcount); - g->kcount[i]++; - } - } -#endif - if (pscnt > 1) - return; - KASSERT(p != NULL, ("NULL process in interrupt state")); - p->p_iticks++; - cp_time[CP_INTR]++; - } - - schedclock(p); - - /* Update resource usage integrals and maximums. */ - if ((pstats = p->p_stats) != NULL && - (ru = &pstats->p_ru) != NULL && - (vm = p->p_vmspace) != NULL) { - ru->ru_ixrss += pgtok(vm->vm_tsize); - ru->ru_idrss += pgtok(vm->vm_dsize); - ru->ru_isrss += pgtok(vm->vm_ssize); - rss = pgtok(vmspace_resident_count(vm)); - if (ru->ru_maxrss < rss) - ru->ru_maxrss = rss; - } -} - +/* + * For statclock, we send an IPI to all CPU's to have them call this + * function. + */ void -forward_statclock(int pscnt) +forwarded_statclock(struct trapframe frame) { - int map; - int id; - int i; - /* Kludge. We don't yet have separate locks for the interrupts - * and the kernel. This means that we cannot let the other processors - * handle complex interrupts while inhibiting them from entering - * the kernel in a non-interrupt context. - * - * What we can do, without changing the locking mechanisms yet, - * is letting the other processors handle a very simple interrupt - * (wich determines the processor states), and do the main - * work ourself. - */ - - CTR1(KTR_SMP, "forward_statclock(%d)", pscnt); - - if (!smp_started || !invltlb_ok || cold || panicstr) - return; - - /* Step 1: Probe state (user, cpu, interrupt, spinlock, idle ) */ - - map = PCPU_GET(other_cpus) & ~stopped_cpus ; - checkstate_probed_cpus = 0; - if (map != 0) - ipi_selected(map, IPI_CHECKSTATE); - - i = 0; - while (checkstate_probed_cpus != map) { - /* spin */ - i++; - if (i == 100000) { -#ifdef DIAGNOSTIC - printf("forward_statclock: checkstate %x\n", - checkstate_probed_cpus); -#endif - break; - } - } - - /* - * Step 2: walk through other processors processes, update ticks and - * profiling info. - */ - - map = 0; - for (id = 0; id < mp_ncpus; id++) { - if (id == PCPU_GET(cpuid)) - continue; - if (((1 << id) & checkstate_probed_cpus) == 0) - continue; - forwarded_statclock(id, pscnt, &map); - } - if (map != 0) { - checkstate_need_ast |= map; - ipi_selected(map, IPI_AST); - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#ifdef DIAGNOSTIC - printf("forward_statclock: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } - } + mtx_lock_spin(&sched_lock); + statclock_process(curproc, TRAPF_PC(&frame), TRAPF_USERMODE(&frame)); + mtx_unlock_spin(&sched_lock); } -void -forward_hardclock(int pscnt) +void +forward_statclock(void) { int map; - int id; - struct proc *p; - struct pstats *pstats; - int i; - - /* Kludge. We don't yet have separate locks for the interrupts - * and the kernel. This means that we cannot let the other processors - * handle complex interrupts while inhibiting them from entering - * the kernel in a non-interrupt context. - * - * What we can do, without changing the locking mechanisms yet, - * is letting the other processors handle a very simple interrupt - * (wich determines the processor states), and do the main - * work ourself. - */ - CTR1(KTR_SMP, "forward_hardclock(%d)", pscnt); + CTR0(KTR_SMP, "forward_statclock"); if (!smp_started || !invltlb_ok || cold || panicstr) return; - /* Step 1: Probe state (user, cpu, interrupt, spinlock, idle) */ - map = PCPU_GET(other_cpus) & ~stopped_cpus ; - checkstate_probed_cpus = 0; if (map != 0) - ipi_selected(map, IPI_CHECKSTATE); - - i = 0; - while (checkstate_probed_cpus != map) { - /* spin */ - i++; - if (i == 100000) { -#ifdef DIAGNOSTIC - printf("forward_hardclock: checkstate %x\n", - checkstate_probed_cpus); -#endif - break; - } - } - - /* - * Step 2: walk through other processors processes, update virtual - * timer and profiling timer. If stathz == 0, also update ticks and - * profiling info. - */ - - map = 0; - for (id = 0; id < mp_ncpus; id++) { - if (id == PCPU_GET(cpuid)) - continue; - if (((1 << id) & checkstate_probed_cpus) == 0) - continue; - p = checkstate_curproc[id]; - if (p) { - pstats = p->p_stats; - if (checkstate_cpustate[id] == CHECKSTATE_USER && - timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) && - itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) { - p->p_sflag |= PS_ALRMPEND; - map |= (1 << id); - } - if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) && - itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) { - p->p_sflag |= PS_PROFPEND; - map |= (1 << id); - } - } - if (stathz == 0) { - forwarded_statclock( id, pscnt, &map); - } - } - if (map != 0) { - checkstate_need_ast |= map; - ipi_selected(map, IPI_AST); - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#ifdef DIAGNOSTIC - printf("forward_hardclock: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } - } + ipi_selected(map, IPI_STATCLOCK); } -void -forward_signal(struct proc *p) +/* + * For each hardclock(), we send an IPI to all other CPU's to have them + * execute this function. It would be nice to reduce contention on + * sched_lock if we could simply peek at the CPU to determine the user/kernel + * state and call hardclock_process() on the CPU receiving the clock interrupt + * and then just use a simple IPI to handle any ast's if needed. + */ +void +forwarded_hardclock(struct trapframe frame) { - int map; - int id; - int i; - - /* Kludge. We don't yet have separate locks for the interrupts - * and the kernel. This means that we cannot let the other processors - * handle complex interrupts while inhibiting them from entering - * the kernel in a non-interrupt context. - * - * What we can do, without changing the locking mechanisms yet, - * is letting the other processors handle a very simple interrupt - * (wich determines the processor states), and do the main - * work ourself. - */ - - CTR1(KTR_SMP, "forward_signal(%p)", p); - if (!smp_started || !invltlb_ok || cold || panicstr) - return; - if (!forward_signal_enabled) - return; mtx_lock_spin(&sched_lock); - while (1) { - if (p->p_stat != SRUN) { - mtx_unlock_spin(&sched_lock); - return; - } - id = p->p_oncpu; - mtx_unlock_spin(&sched_lock); - if (id == 0xff) - return; - map = (1<<id); - checkstate_need_ast |= map; - ipi_selected(map, IPI_AST); - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#if 0 - printf("forward_signal: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } - mtx_lock_spin(&sched_lock); - if (id == p->p_oncpu) { - mtx_unlock_spin(&sched_lock); - return; - } - } + hardclock_process(curproc, TRAPF_USERMODE(&frame)); + mtx_unlock_spin(&sched_lock); } -void -forward_roundrobin(void) +void +forward_hardclock(void) { u_int map; - int i; - CTR0(KTR_SMP, "forward_roundrobin()"); + CTR0(KTR_SMP, "forward_hardclock"); if (!smp_started || !invltlb_ok || cold || panicstr) return; - if (!forward_roundrobin_enabled) - return; - resched_cpus |= PCPU_GET(other_cpus); - map = PCPU_GET(other_cpus) & ~stopped_cpus ; -#if 1 - ipi_selected(map, IPI_AST); -#else - ipi_all_but_self(IPI_AST); -#endif - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#if 0 - printf("forward_roundrobin: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } -} - -/* - * When called the executing CPU will send an IPI to all other CPUs - * requesting that they halt execution. - * - * Usually (but not necessarily) called with 'other_cpus' as its arg. - * - * - Signals all CPUs in map to stop. - * - Waits for each to stop. - * - * Returns: - * -1: error - * 0: NA - * 1: ok - * - * XXX FIXME: this is not MP-safe, needs a lock to prevent multiple CPUs - * from executing at same time. - */ -int -stop_cpus(u_int map) -{ - int count = 0; - - if (!smp_started) - return 0; - - /* send the Xcpustop IPI to all CPUs in map */ - ipi_selected(map, IPI_STOP); - - while (count++ < 100000 && (stopped_cpus & map) != map) - /* spin */ ; - -#ifdef DIAGNOSTIC - if ((stopped_cpus & map) != map) - printf("Warning: CPUs 0x%x did not stop!\n", - (~(stopped_cpus & map)) & map); -#endif - - return 1; -} - - -/* - * Called by a CPU to restart stopped CPUs. - * - * Usually (but not necessarily) called with 'stopped_cpus' as its arg. - * - * - Signals all CPUs in map to restart. - * - Waits for each to restart. - * - * Returns: - * -1: error - * 0: NA - * 1: ok - */ -int -restart_cpus(u_int map) -{ - int count = 0; - - if (!smp_started) - return 0; - - started_cpus = map; /* signal other cpus to restart */ - - /* wait for each to clear its bit */ - while (count++ < 100000 && (stopped_cpus & map) != 0) - /* spin */ ; -#ifdef DIAGNOSTIC - if ((stopped_cpus & map) != 0) - printf("Warning: CPUs 0x%x did not restart!\n", - (~(stopped_cpus & map)) & map); -#endif - - return 1; + map = PCPU_GET(other_cpus) & ~stopped_cpus ; + if (map != 0) + ipi_selected(map, IPI_HARDCLOCK); } - #ifdef APIC_INTR_REORDER /* * Maintain mapping from softintr vector to isr bit in local apic. @@ -2811,73 +2386,6 @@ set_lapic_isrloc(int intr, int vector) #endif /* - * All-CPU rendezvous. CPUs are signalled, all execute the setup function - * (if specified), rendezvous, execute the action function (if specified), - * rendezvous again, execute the teardown function (if specified), and then - * resume. - * - * Note that the supplied external functions _must_ be reentrant and aware - * that they are running in parallel and in an unknown lock context. - */ -static void (*smp_rv_setup_func)(void *arg); -static void (*smp_rv_action_func)(void *arg); -static void (*smp_rv_teardown_func)(void *arg); -static void *smp_rv_func_arg; -static volatile int smp_rv_waiters[2]; - -void -smp_rendezvous_action(void) -{ - /* setup function */ - if (smp_rv_setup_func != NULL) - smp_rv_setup_func(smp_rv_func_arg); - /* spin on entry rendezvous */ - atomic_add_int(&smp_rv_waiters[0], 1); - while (smp_rv_waiters[0] < mp_ncpus) - ; - /* action function */ - if (smp_rv_action_func != NULL) - smp_rv_action_func(smp_rv_func_arg); - /* spin on exit rendezvous */ - atomic_add_int(&smp_rv_waiters[1], 1); - while (smp_rv_waiters[1] < mp_ncpus) - ; - /* teardown function */ - if (smp_rv_teardown_func != NULL) - smp_rv_teardown_func(smp_rv_func_arg); -} - -void -smp_rendezvous(void (* setup_func)(void *), - void (* action_func)(void *), - void (* teardown_func)(void *), - void *arg) -{ - - /* obtain rendezvous lock */ - mtx_lock_spin(&smp_rv_mtx); - - /* set static function pointers */ - smp_rv_setup_func = setup_func; - smp_rv_action_func = action_func; - smp_rv_teardown_func = teardown_func; - smp_rv_func_arg = arg; - smp_rv_waiters[0] = 0; - smp_rv_waiters[1] = 0; - - /* - * signal other processors, which will enter the IPI with interrupts off - */ - ipi_all_but_self(IPI_RENDEZVOUS); - - /* call executor function */ - smp_rendezvous_action(); - - /* release lock */ - mtx_unlock_spin(&smp_rv_mtx); -} - -/* * send an IPI to a set of cpus. */ void diff --git a/sys/i386/i386/mptable.c b/sys/i386/i386/mptable.c index 28a5c72..21e6b6e 100644 --- a/sys/i386/i386/mptable.c +++ b/sys/i386/i386/mptable.c @@ -42,6 +42,7 @@ #include <sys/malloc.h> #include <sys/memrange.h> #include <sys/mutex.h> +#include <sys/smp.h> #include <sys/dkstat.h> #include <sys/cons.h> /* cngetc() */ @@ -57,9 +58,9 @@ #include <sys/gmon.h> #endif -#include <machine/smp.h> #include <machine/apic.h> #include <machine/atomic.h> +#include <machine/cpu.h> #include <machine/cpufunc.h> #include <machine/ipl.h> #include <machine/mpapic.h> @@ -243,7 +244,6 @@ int current_postcode; extern struct region_descriptor r_gdt, r_idt; int bsp_apic_ready = 0; /* flags useability of BSP apic */ -int mp_ncpus; /* # of CPUs, including BSP */ int mp_naps; /* # of Applications processors */ int mp_nbusses; /* # of busses */ int mp_napics; /* # of IO APICs */ @@ -273,9 +273,6 @@ int io_num_to_apic_id[NAPICID]; int apic_id_to_logical[NAPICID]; -/* Bitmap of all available CPUs */ -u_int all_cpus; - /* AP uses this during bootstrap. Do not staticize. */ char *bootSTK; static int bootAP; @@ -288,28 +285,9 @@ extern pt_entry_t *SMPpt; struct pcb stoppcbs[MAXCPU]; -int smp_started; /* has the system started? */ -int smp_active = 0; /* are the APs allowed to run? */ -SYSCTL_INT(_machdep, OID_AUTO, smp_active, CTLFLAG_RW, &smp_active, 0, ""); - -/* XXX maybe should be hw.ncpu */ -static int smp_cpus = 1; /* how many cpu's running */ -SYSCTL_INT(_machdep, OID_AUTO, smp_cpus, CTLFLAG_RD, &smp_cpus, 0, ""); - int invltlb_ok = 0; /* throttle smp_invltlb() till safe */ SYSCTL_INT(_machdep, OID_AUTO, invltlb_ok, CTLFLAG_RW, &invltlb_ok, 0, ""); -/* Enable forwarding of a signal to a process running on a different CPU */ -static int forward_signal_enabled = 1; -SYSCTL_INT(_machdep, OID_AUTO, forward_signal_enabled, CTLFLAG_RW, - &forward_signal_enabled, 0, ""); - -/* Enable forwarding of roundrobin to all other cpus */ -static int forward_roundrobin_enabled = 1; -SYSCTL_INT(_machdep, OID_AUTO, forward_roundrobin_enabled, CTLFLAG_RW, - &forward_roundrobin_enabled, 0, ""); - - /* * Local data and functions. */ @@ -354,9 +332,6 @@ struct mtx mcount_mtx; struct mtx com_mtx; #endif /* USE_COMLOCK */ -/* lock around the MP rendezvous */ -static struct mtx smp_rv_mtx; - static void init_locks(void) { @@ -367,13 +342,9 @@ init_locks(void) */ mtx_init(&mcount_mtx, "mcount", MTX_DEF); - mtx_init(&smp_rv_mtx, "smp rendezvous", MTX_SPIN); - #ifdef USE_COMLOCK mtx_init(&com_mtx, "com", MTX_SPIN); #endif /* USE_COMLOCK */ - - mtx_init(&ap_boot_mtx, "ap boot", MTX_SPIN); } /* @@ -397,8 +368,8 @@ mp_bootaddress(u_int basemem) /* * Look for an Intel MP spec table (ie, SMP capable hardware). */ -int -mp_probe(void) +void +i386_mp_probe(void) { int x; u_long segment; @@ -427,7 +398,7 @@ mp_probe(void) /* nothing found */ mpfps = (mpfps_t)0; mp_capable = 0; - return 0; + return; found: /* calculate needed resources */ @@ -436,15 +407,19 @@ found: /* flag fact that we are running multiple processors */ mp_capable = 1; - return 1; } +int +cpu_mp_probe(void) +{ + return (mp_capable); +} /* * Initialize the SMP hardware and the APIC and start up the AP's. */ void -mp_start(void) +cpu_mp_start(void) { POSTCODE(MP_START_POST); @@ -453,6 +428,8 @@ mp_start(void) mp_enable(boot_address); else panic("MP hardware not found!"); + + cpu_setregs(); } @@ -460,13 +437,12 @@ mp_start(void) * Print various information about the SMP system hardware and setup. */ void -mp_announce(void) +cpu_mp_announce(void) { int x; POSTCODE(MP_ANNOUNCE_POST); - printf("FreeBSD/SMP: Multiprocessor motherboard\n"); printf(" cpu0 (BSP): apic id: %2d", CPU_TO_ID(0)); printf(", version: 0x%08x", cpu_apic_versions[0]); printf(", at 0x%08x\n", cpu_apic_address); @@ -623,8 +599,12 @@ mp_enable(u_int boot_addr) setidt(XINVLTLB_OFFSET, Xinvltlb, SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); - /* install an inter-CPU IPI for reading processor state */ - setidt(XCPUCHECKSTATE_OFFSET, Xcpucheckstate, + /* install an inter-CPU IPI for forwarding hardclock() */ + setidt(XHARDCLOCK_OFFSET, Xhardclock, + SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); + + /* install an inter-CPU IPI for forwarding statclock() */ + setidt(XSTATCLOCK_OFFSET, Xstatclock, SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); /* install an inter-CPU IPI for all-CPU rendezvous */ @@ -1938,6 +1918,8 @@ start_all_aps(u_int boot_addr) POSTCODE(START_ALL_APS_POST); + mtx_init(&ap_boot_mtx, "ap boot", MTX_SPIN); + /* initialize BSP's local APIC */ apic_initialize(); bsp_apic_ready = 1; @@ -1985,8 +1967,8 @@ start_all_aps(u_int boot_addr) (PG_V | PG_RW | vtophys(PAGE_SIZE * i + stack)); /* prime data page for it to use */ - SLIST_INSERT_HEAD(&cpuhead, gd, gd_allcpu); gd->gd_cpuid = x; + globaldata_register(gd); /* setup a vector to our boot code */ *((volatile u_short *) WARMBOOT_OFF) = WARMBOOT_TARGET; @@ -2328,472 +2310,65 @@ ap_init(void) panic("scheduler returned us to ap_init"); } -#define CHECKSTATE_USER 0 -#define CHECKSTATE_SYS 1 -#define CHECKSTATE_INTR 2 - -/* Do not staticize. Used from apic_vector.s */ -struct proc* checkstate_curproc[MAXCPU]; -int checkstate_cpustate[MAXCPU]; -u_long checkstate_pc[MAXCPU]; - -#define PC_TO_INDEX(pc, prof) \ - ((int)(((u_quad_t)((pc) - (prof)->pr_off) * \ - (u_quad_t)((prof)->pr_scale)) >> 16) & ~1) - -static void -addupc_intr_forwarded(struct proc *p, int id, int *astmap) -{ - int i; - struct uprof *prof; - u_long pc; - - pc = checkstate_pc[id]; - prof = &p->p_stats->p_prof; - if (pc >= prof->pr_off && - (i = PC_TO_INDEX(pc, prof)) < prof->pr_size) { - mtx_assert(&sched_lock, MA_OWNED); - if ((p->p_sflag & PS_OWEUPC) == 0) { - prof->pr_addr = pc; - prof->pr_ticks = 1; - p->p_sflag |= PS_OWEUPC; - } - *astmap |= (1 << id); - } -} - -static void -forwarded_statclock(int id, int pscnt, int *astmap) -{ - struct pstats *pstats; - long rss; - struct rusage *ru; - struct vmspace *vm; - int cpustate; - struct proc *p; -#ifdef GPROF - register struct gmonparam *g; - int i; -#endif - - mtx_assert(&sched_lock, MA_OWNED); - p = checkstate_curproc[id]; - cpustate = checkstate_cpustate[id]; - - /* XXX */ - if (p->p_ithd) - cpustate = CHECKSTATE_INTR; - else if (p == SMP_prvspace[id].globaldata.gd_idleproc) - cpustate = CHECKSTATE_SYS; - - switch (cpustate) { - case CHECKSTATE_USER: - if (p->p_sflag & PS_PROFIL) - addupc_intr_forwarded(p, id, astmap); - if (pscnt > 1) - return; - p->p_uticks++; - if (p->p_nice > NZERO) - cp_time[CP_NICE]++; - else - cp_time[CP_USER]++; - break; - case CHECKSTATE_SYS: -#ifdef GPROF - /* - * Kernel statistics are just like addupc_intr, only easier. - */ - g = &_gmonparam; - if (g->state == GMON_PROF_ON) { - i = checkstate_pc[id] - g->lowpc; - if (i < g->textsize) { - i /= HISTFRACTION * sizeof(*g->kcount); - g->kcount[i]++; - } - } -#endif - if (pscnt > 1) - return; - - p->p_sticks++; - if (p == SMP_prvspace[id].globaldata.gd_idleproc) - cp_time[CP_IDLE]++; - else - cp_time[CP_SYS]++; - break; - case CHECKSTATE_INTR: - default: -#ifdef GPROF - /* - * Kernel statistics are just like addupc_intr, only easier. - */ - g = &_gmonparam; - if (g->state == GMON_PROF_ON) { - i = checkstate_pc[id] - g->lowpc; - if (i < g->textsize) { - i /= HISTFRACTION * sizeof(*g->kcount); - g->kcount[i]++; - } - } -#endif - if (pscnt > 1) - return; - KASSERT(p != NULL, ("NULL process in interrupt state")); - p->p_iticks++; - cp_time[CP_INTR]++; - } - - schedclock(p); - - /* Update resource usage integrals and maximums. */ - if ((pstats = p->p_stats) != NULL && - (ru = &pstats->p_ru) != NULL && - (vm = p->p_vmspace) != NULL) { - ru->ru_ixrss += pgtok(vm->vm_tsize); - ru->ru_idrss += pgtok(vm->vm_dsize); - ru->ru_isrss += pgtok(vm->vm_ssize); - rss = pgtok(vmspace_resident_count(vm)); - if (ru->ru_maxrss < rss) - ru->ru_maxrss = rss; - } -} - +/* + * For statclock, we send an IPI to all CPU's to have them call this + * function. + */ void -forward_statclock(int pscnt) +forwarded_statclock(struct trapframe frame) { - int map; - int id; - int i; - /* Kludge. We don't yet have separate locks for the interrupts - * and the kernel. This means that we cannot let the other processors - * handle complex interrupts while inhibiting them from entering - * the kernel in a non-interrupt context. - * - * What we can do, without changing the locking mechanisms yet, - * is letting the other processors handle a very simple interrupt - * (wich determines the processor states), and do the main - * work ourself. - */ - - CTR1(KTR_SMP, "forward_statclock(%d)", pscnt); - - if (!smp_started || !invltlb_ok || cold || panicstr) - return; - - /* Step 1: Probe state (user, cpu, interrupt, spinlock, idle ) */ - - map = PCPU_GET(other_cpus) & ~stopped_cpus ; - checkstate_probed_cpus = 0; - if (map != 0) - ipi_selected(map, IPI_CHECKSTATE); - - i = 0; - while (checkstate_probed_cpus != map) { - /* spin */ - i++; - if (i == 100000) { -#ifdef DIAGNOSTIC - printf("forward_statclock: checkstate %x\n", - checkstate_probed_cpus); -#endif - break; - } - } - - /* - * Step 2: walk through other processors processes, update ticks and - * profiling info. - */ - - map = 0; - for (id = 0; id < mp_ncpus; id++) { - if (id == PCPU_GET(cpuid)) - continue; - if (((1 << id) & checkstate_probed_cpus) == 0) - continue; - forwarded_statclock(id, pscnt, &map); - } - if (map != 0) { - checkstate_need_ast |= map; - ipi_selected(map, IPI_AST); - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#ifdef DIAGNOSTIC - printf("forward_statclock: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } - } + mtx_lock_spin(&sched_lock); + statclock_process(curproc, TRAPF_PC(&frame), TRAPF_USERMODE(&frame)); + mtx_unlock_spin(&sched_lock); } -void -forward_hardclock(int pscnt) +void +forward_statclock(void) { int map; - int id; - struct proc *p; - struct pstats *pstats; - int i; - - /* Kludge. We don't yet have separate locks for the interrupts - * and the kernel. This means that we cannot let the other processors - * handle complex interrupts while inhibiting them from entering - * the kernel in a non-interrupt context. - * - * What we can do, without changing the locking mechanisms yet, - * is letting the other processors handle a very simple interrupt - * (wich determines the processor states), and do the main - * work ourself. - */ - CTR1(KTR_SMP, "forward_hardclock(%d)", pscnt); + CTR0(KTR_SMP, "forward_statclock"); if (!smp_started || !invltlb_ok || cold || panicstr) return; - /* Step 1: Probe state (user, cpu, interrupt, spinlock, idle) */ - map = PCPU_GET(other_cpus) & ~stopped_cpus ; - checkstate_probed_cpus = 0; if (map != 0) - ipi_selected(map, IPI_CHECKSTATE); - - i = 0; - while (checkstate_probed_cpus != map) { - /* spin */ - i++; - if (i == 100000) { -#ifdef DIAGNOSTIC - printf("forward_hardclock: checkstate %x\n", - checkstate_probed_cpus); -#endif - break; - } - } - - /* - * Step 2: walk through other processors processes, update virtual - * timer and profiling timer. If stathz == 0, also update ticks and - * profiling info. - */ - - map = 0; - for (id = 0; id < mp_ncpus; id++) { - if (id == PCPU_GET(cpuid)) - continue; - if (((1 << id) & checkstate_probed_cpus) == 0) - continue; - p = checkstate_curproc[id]; - if (p) { - pstats = p->p_stats; - if (checkstate_cpustate[id] == CHECKSTATE_USER && - timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) && - itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) { - p->p_sflag |= PS_ALRMPEND; - map |= (1 << id); - } - if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) && - itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) { - p->p_sflag |= PS_PROFPEND; - map |= (1 << id); - } - } - if (stathz == 0) { - forwarded_statclock( id, pscnt, &map); - } - } - if (map != 0) { - checkstate_need_ast |= map; - ipi_selected(map, IPI_AST); - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#ifdef DIAGNOSTIC - printf("forward_hardclock: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } - } + ipi_selected(map, IPI_STATCLOCK); } -void -forward_signal(struct proc *p) +/* + * For each hardclock(), we send an IPI to all other CPU's to have them + * execute this function. It would be nice to reduce contention on + * sched_lock if we could simply peek at the CPU to determine the user/kernel + * state and call hardclock_process() on the CPU receiving the clock interrupt + * and then just use a simple IPI to handle any ast's if needed. + */ +void +forwarded_hardclock(struct trapframe frame) { - int map; - int id; - int i; - - /* Kludge. We don't yet have separate locks for the interrupts - * and the kernel. This means that we cannot let the other processors - * handle complex interrupts while inhibiting them from entering - * the kernel in a non-interrupt context. - * - * What we can do, without changing the locking mechanisms yet, - * is letting the other processors handle a very simple interrupt - * (wich determines the processor states), and do the main - * work ourself. - */ - - CTR1(KTR_SMP, "forward_signal(%p)", p); - if (!smp_started || !invltlb_ok || cold || panicstr) - return; - if (!forward_signal_enabled) - return; mtx_lock_spin(&sched_lock); - while (1) { - if (p->p_stat != SRUN) { - mtx_unlock_spin(&sched_lock); - return; - } - id = p->p_oncpu; - mtx_unlock_spin(&sched_lock); - if (id == 0xff) - return; - map = (1<<id); - checkstate_need_ast |= map; - ipi_selected(map, IPI_AST); - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#if 0 - printf("forward_signal: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } - mtx_lock_spin(&sched_lock); - if (id == p->p_oncpu) { - mtx_unlock_spin(&sched_lock); - return; - } - } + hardclock_process(curproc, TRAPF_USERMODE(&frame)); + mtx_unlock_spin(&sched_lock); } -void -forward_roundrobin(void) +void +forward_hardclock(void) { u_int map; - int i; - CTR0(KTR_SMP, "forward_roundrobin()"); + CTR0(KTR_SMP, "forward_hardclock"); if (!smp_started || !invltlb_ok || cold || panicstr) return; - if (!forward_roundrobin_enabled) - return; - resched_cpus |= PCPU_GET(other_cpus); - map = PCPU_GET(other_cpus) & ~stopped_cpus ; -#if 1 - ipi_selected(map, IPI_AST); -#else - ipi_all_but_self(IPI_AST); -#endif - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#if 0 - printf("forward_roundrobin: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } -} - -/* - * When called the executing CPU will send an IPI to all other CPUs - * requesting that they halt execution. - * - * Usually (but not necessarily) called with 'other_cpus' as its arg. - * - * - Signals all CPUs in map to stop. - * - Waits for each to stop. - * - * Returns: - * -1: error - * 0: NA - * 1: ok - * - * XXX FIXME: this is not MP-safe, needs a lock to prevent multiple CPUs - * from executing at same time. - */ -int -stop_cpus(u_int map) -{ - int count = 0; - - if (!smp_started) - return 0; - - /* send the Xcpustop IPI to all CPUs in map */ - ipi_selected(map, IPI_STOP); - - while (count++ < 100000 && (stopped_cpus & map) != map) - /* spin */ ; - -#ifdef DIAGNOSTIC - if ((stopped_cpus & map) != map) - printf("Warning: CPUs 0x%x did not stop!\n", - (~(stopped_cpus & map)) & map); -#endif - - return 1; -} - - -/* - * Called by a CPU to restart stopped CPUs. - * - * Usually (but not necessarily) called with 'stopped_cpus' as its arg. - * - * - Signals all CPUs in map to restart. - * - Waits for each to restart. - * - * Returns: - * -1: error - * 0: NA - * 1: ok - */ -int -restart_cpus(u_int map) -{ - int count = 0; - - if (!smp_started) - return 0; - - started_cpus = map; /* signal other cpus to restart */ - - /* wait for each to clear its bit */ - while (count++ < 100000 && (stopped_cpus & map) != 0) - /* spin */ ; -#ifdef DIAGNOSTIC - if ((stopped_cpus & map) != 0) - printf("Warning: CPUs 0x%x did not restart!\n", - (~(stopped_cpus & map)) & map); -#endif - - return 1; + map = PCPU_GET(other_cpus) & ~stopped_cpus ; + if (map != 0) + ipi_selected(map, IPI_HARDCLOCK); } - #ifdef APIC_INTR_REORDER /* * Maintain mapping from softintr vector to isr bit in local apic. @@ -2811,73 +2386,6 @@ set_lapic_isrloc(int intr, int vector) #endif /* - * All-CPU rendezvous. CPUs are signalled, all execute the setup function - * (if specified), rendezvous, execute the action function (if specified), - * rendezvous again, execute the teardown function (if specified), and then - * resume. - * - * Note that the supplied external functions _must_ be reentrant and aware - * that they are running in parallel and in an unknown lock context. - */ -static void (*smp_rv_setup_func)(void *arg); -static void (*smp_rv_action_func)(void *arg); -static void (*smp_rv_teardown_func)(void *arg); -static void *smp_rv_func_arg; -static volatile int smp_rv_waiters[2]; - -void -smp_rendezvous_action(void) -{ - /* setup function */ - if (smp_rv_setup_func != NULL) - smp_rv_setup_func(smp_rv_func_arg); - /* spin on entry rendezvous */ - atomic_add_int(&smp_rv_waiters[0], 1); - while (smp_rv_waiters[0] < mp_ncpus) - ; - /* action function */ - if (smp_rv_action_func != NULL) - smp_rv_action_func(smp_rv_func_arg); - /* spin on exit rendezvous */ - atomic_add_int(&smp_rv_waiters[1], 1); - while (smp_rv_waiters[1] < mp_ncpus) - ; - /* teardown function */ - if (smp_rv_teardown_func != NULL) - smp_rv_teardown_func(smp_rv_func_arg); -} - -void -smp_rendezvous(void (* setup_func)(void *), - void (* action_func)(void *), - void (* teardown_func)(void *), - void *arg) -{ - - /* obtain rendezvous lock */ - mtx_lock_spin(&smp_rv_mtx); - - /* set static function pointers */ - smp_rv_setup_func = setup_func; - smp_rv_action_func = action_func; - smp_rv_teardown_func = teardown_func; - smp_rv_func_arg = arg; - smp_rv_waiters[0] = 0; - smp_rv_waiters[1] = 0; - - /* - * signal other processors, which will enter the IPI with interrupts off - */ - ipi_all_but_self(IPI_RENDEZVOUS); - - /* call executor function */ - smp_rendezvous_action(); - - /* release lock */ - mtx_unlock_spin(&smp_rv_mtx); -} - -/* * send an IPI to a set of cpus. */ void diff --git a/sys/i386/i386/sys_machdep.c b/sys/i386/i386/sys_machdep.c index cf05d73..5ed16ab 100644 --- a/sys/i386/i386/sys_machdep.c +++ b/sys/i386/i386/sys_machdep.c @@ -41,6 +41,7 @@ #include <sys/ipl.h> #include <sys/malloc.h> #include <sys/proc.h> +#include <sys/smp.h> #include <vm/vm.h> #include <sys/lock.h> @@ -53,9 +54,6 @@ #include <machine/cpu.h> #include <machine/pcb_ext.h> /* pcb.h included by sys/user.h */ #include <machine/sysarch.h> -#ifdef SMP -#include <machine/smp.h> -#endif #include <vm/vm_kern.h> /* for kernel_map */ @@ -154,7 +152,15 @@ i386_extend_pcb(struct proc *p) ssdtosd(&ssd, &ext->ext_tssd); /* switch to the new TSS after syscall completes */ - need_resched(); + /* + * XXX: The sched_lock here needs to be over a slightly larger area. + * I have patches to more properly lock accesses to process ldt's + * and tss's that still need to be reviewed, but this keeps us from + * panic'ing on the mtx_assert() in need_resched() for the time being. + */ + mtx_lock_spin(&sched_lock); + need_resched(p); + mtx_unlock_spin(&sched_lock); return 0; } diff --git a/sys/i386/i386/trap.c b/sys/i386/i386/trap.c index 7088d0c..d9597df 100644 --- a/sys/i386/i386/trap.c +++ b/sys/i386/i386/trap.c @@ -177,7 +177,7 @@ userret(p, frame, oticks) mtx_lock_spin(&sched_lock); p->p_pri.pri_level = p->p_pri.pri_user; - if (resched_wanted()) { + if (resched_wanted(p)) { /* * Since we are curproc, clock will normally just change * our priority without moving us from one queue to another @@ -1277,7 +1277,7 @@ ast(framep) * acquiring and releasing mutexes in assembly is not fun. */ mtx_lock_spin(&sched_lock); - if (!(astpending(p) || resched_wanted())) { + if (!(astpending(p) || resched_wanted(p))) { mtx_unlock_spin(&sched_lock); return; } diff --git a/sys/i386/i386/tsc.c b/sys/i386/i386/tsc.c index 9034f81..ce67390 100644 --- a/sys/i386/i386/tsc.c +++ b/sys/i386/i386/tsc.c @@ -211,6 +211,10 @@ clkintr(struct clockframe frame) mtx_unlock_spin(&clock_lock); } timer_func(&frame); +#ifdef SMP + if (timer_func == hardclock) + forward_hardclock(); +#endif switch (timer0_state) { case RELEASED: @@ -253,6 +257,9 @@ clkintr(struct clockframe frame) timer_func = hardclock; timer0_state = RELEASED; hardclock(&frame); +#ifdef SMP + forward_hardclock(); +#endif } break; } @@ -374,8 +381,12 @@ release_timer2() static void rtcintr(struct clockframe frame) { - while (rtcin(RTC_INTR) & RTCIR_PERIOD) + while (rtcin(RTC_INTR) & RTCIR_PERIOD) { statclock(&frame); +#ifdef SMP + forward_statclock(); +#endif + } } #include "opt_ddb.h" diff --git a/sys/i386/i386/vm_machdep.c b/sys/i386/i386/vm_machdep.c index 0a15103..f8fa5bb 100644 --- a/sys/i386/i386/vm_machdep.c +++ b/sys/i386/i386/vm_machdep.c @@ -59,14 +59,12 @@ #include <sys/kernel.h> #include <sys/ktr.h> #include <sys/mutex.h> +#include <sys/smp.h> #include <sys/sysctl.h> #include <sys/unistd.h> #include <machine/cpu.h> #include <machine/md_var.h> -#ifdef SMP -#include <machine/smp.h> -#endif #include <machine/pcb.h> #include <machine/pcb_ext.h> #include <machine/vm86.h> diff --git a/sys/i386/include/cpu.h b/sys/i386/include/cpu.h index 74f21b9..29b34ee 100644 --- a/sys/i386/include/cpu.h +++ b/sys/i386/include/cpu.h @@ -65,8 +65,6 @@ #define CLKF_USERMODE(framep) \ ((ISPL((framep)->cf_cs) == SEL_UPL) || ((framep)->cf_eflags & PSL_VM)) - -#define CLKF_INTR(framep) (curproc->p_intr_nesting_level >= 2) #define CLKF_PC(framep) ((framep)->cf_eip) /* diff --git a/sys/i386/include/globaldata.h b/sys/i386/include/globaldata.h index e713227..ef82a66 100644 --- a/sys/i386/include/globaldata.h +++ b/sys/i386/include/globaldata.h @@ -74,9 +74,6 @@ struct globaldata { #endif }; -SLIST_HEAD(cpuhead, globaldata); -extern struct cpuhead cpuhead; - #ifdef SMP /* * This is the upper (0xff800000) address space layout that is per-cpu. diff --git a/sys/i386/include/ipl.h b/sys/i386/include/ipl.h index a0c9d11..61b5513 100644 --- a/sys/i386/include/ipl.h +++ b/sys/i386/include/ipl.h @@ -42,23 +42,4 @@ #include <i386/isa/icu_ipl.h> #endif -#ifndef LOCORE -#ifdef SMP - -/* - * Interprocessor interrupts for SMP. - */ -#define IPI_INVLTLB XINVLTLB_OFFSET -#define IPI_RENDEZVOUS XRENDEZVOUS_OFFSET -#define IPI_AST XCPUAST_OFFSET -#define IPI_CHECKSTATE XCPUCHECKSTATE_OFFSET -#define IPI_STOP XCPUSTOP_OFFSET - -void ipi_selected(u_int32_t cpus, u_int ipi); -void ipi_all(u_int ipi); -void ipi_all_but_self(u_int ipi); -void ipi_self(u_int ipi); -#endif /* SMP */ -#endif /* !LOCORE */ - #endif /* !_MACHINE_IPL_H_ */ diff --git a/sys/i386/include/mptable.h b/sys/i386/include/mptable.h index 28a5c72..21e6b6e 100644 --- a/sys/i386/include/mptable.h +++ b/sys/i386/include/mptable.h @@ -42,6 +42,7 @@ #include <sys/malloc.h> #include <sys/memrange.h> #include <sys/mutex.h> +#include <sys/smp.h> #include <sys/dkstat.h> #include <sys/cons.h> /* cngetc() */ @@ -57,9 +58,9 @@ #include <sys/gmon.h> #endif -#include <machine/smp.h> #include <machine/apic.h> #include <machine/atomic.h> +#include <machine/cpu.h> #include <machine/cpufunc.h> #include <machine/ipl.h> #include <machine/mpapic.h> @@ -243,7 +244,6 @@ int current_postcode; extern struct region_descriptor r_gdt, r_idt; int bsp_apic_ready = 0; /* flags useability of BSP apic */ -int mp_ncpus; /* # of CPUs, including BSP */ int mp_naps; /* # of Applications processors */ int mp_nbusses; /* # of busses */ int mp_napics; /* # of IO APICs */ @@ -273,9 +273,6 @@ int io_num_to_apic_id[NAPICID]; int apic_id_to_logical[NAPICID]; -/* Bitmap of all available CPUs */ -u_int all_cpus; - /* AP uses this during bootstrap. Do not staticize. */ char *bootSTK; static int bootAP; @@ -288,28 +285,9 @@ extern pt_entry_t *SMPpt; struct pcb stoppcbs[MAXCPU]; -int smp_started; /* has the system started? */ -int smp_active = 0; /* are the APs allowed to run? */ -SYSCTL_INT(_machdep, OID_AUTO, smp_active, CTLFLAG_RW, &smp_active, 0, ""); - -/* XXX maybe should be hw.ncpu */ -static int smp_cpus = 1; /* how many cpu's running */ -SYSCTL_INT(_machdep, OID_AUTO, smp_cpus, CTLFLAG_RD, &smp_cpus, 0, ""); - int invltlb_ok = 0; /* throttle smp_invltlb() till safe */ SYSCTL_INT(_machdep, OID_AUTO, invltlb_ok, CTLFLAG_RW, &invltlb_ok, 0, ""); -/* Enable forwarding of a signal to a process running on a different CPU */ -static int forward_signal_enabled = 1; -SYSCTL_INT(_machdep, OID_AUTO, forward_signal_enabled, CTLFLAG_RW, - &forward_signal_enabled, 0, ""); - -/* Enable forwarding of roundrobin to all other cpus */ -static int forward_roundrobin_enabled = 1; -SYSCTL_INT(_machdep, OID_AUTO, forward_roundrobin_enabled, CTLFLAG_RW, - &forward_roundrobin_enabled, 0, ""); - - /* * Local data and functions. */ @@ -354,9 +332,6 @@ struct mtx mcount_mtx; struct mtx com_mtx; #endif /* USE_COMLOCK */ -/* lock around the MP rendezvous */ -static struct mtx smp_rv_mtx; - static void init_locks(void) { @@ -367,13 +342,9 @@ init_locks(void) */ mtx_init(&mcount_mtx, "mcount", MTX_DEF); - mtx_init(&smp_rv_mtx, "smp rendezvous", MTX_SPIN); - #ifdef USE_COMLOCK mtx_init(&com_mtx, "com", MTX_SPIN); #endif /* USE_COMLOCK */ - - mtx_init(&ap_boot_mtx, "ap boot", MTX_SPIN); } /* @@ -397,8 +368,8 @@ mp_bootaddress(u_int basemem) /* * Look for an Intel MP spec table (ie, SMP capable hardware). */ -int -mp_probe(void) +void +i386_mp_probe(void) { int x; u_long segment; @@ -427,7 +398,7 @@ mp_probe(void) /* nothing found */ mpfps = (mpfps_t)0; mp_capable = 0; - return 0; + return; found: /* calculate needed resources */ @@ -436,15 +407,19 @@ found: /* flag fact that we are running multiple processors */ mp_capable = 1; - return 1; } +int +cpu_mp_probe(void) +{ + return (mp_capable); +} /* * Initialize the SMP hardware and the APIC and start up the AP's. */ void -mp_start(void) +cpu_mp_start(void) { POSTCODE(MP_START_POST); @@ -453,6 +428,8 @@ mp_start(void) mp_enable(boot_address); else panic("MP hardware not found!"); + + cpu_setregs(); } @@ -460,13 +437,12 @@ mp_start(void) * Print various information about the SMP system hardware and setup. */ void -mp_announce(void) +cpu_mp_announce(void) { int x; POSTCODE(MP_ANNOUNCE_POST); - printf("FreeBSD/SMP: Multiprocessor motherboard\n"); printf(" cpu0 (BSP): apic id: %2d", CPU_TO_ID(0)); printf(", version: 0x%08x", cpu_apic_versions[0]); printf(", at 0x%08x\n", cpu_apic_address); @@ -623,8 +599,12 @@ mp_enable(u_int boot_addr) setidt(XINVLTLB_OFFSET, Xinvltlb, SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); - /* install an inter-CPU IPI for reading processor state */ - setidt(XCPUCHECKSTATE_OFFSET, Xcpucheckstate, + /* install an inter-CPU IPI for forwarding hardclock() */ + setidt(XHARDCLOCK_OFFSET, Xhardclock, + SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); + + /* install an inter-CPU IPI for forwarding statclock() */ + setidt(XSTATCLOCK_OFFSET, Xstatclock, SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); /* install an inter-CPU IPI for all-CPU rendezvous */ @@ -1938,6 +1918,8 @@ start_all_aps(u_int boot_addr) POSTCODE(START_ALL_APS_POST); + mtx_init(&ap_boot_mtx, "ap boot", MTX_SPIN); + /* initialize BSP's local APIC */ apic_initialize(); bsp_apic_ready = 1; @@ -1985,8 +1967,8 @@ start_all_aps(u_int boot_addr) (PG_V | PG_RW | vtophys(PAGE_SIZE * i + stack)); /* prime data page for it to use */ - SLIST_INSERT_HEAD(&cpuhead, gd, gd_allcpu); gd->gd_cpuid = x; + globaldata_register(gd); /* setup a vector to our boot code */ *((volatile u_short *) WARMBOOT_OFF) = WARMBOOT_TARGET; @@ -2328,472 +2310,65 @@ ap_init(void) panic("scheduler returned us to ap_init"); } -#define CHECKSTATE_USER 0 -#define CHECKSTATE_SYS 1 -#define CHECKSTATE_INTR 2 - -/* Do not staticize. Used from apic_vector.s */ -struct proc* checkstate_curproc[MAXCPU]; -int checkstate_cpustate[MAXCPU]; -u_long checkstate_pc[MAXCPU]; - -#define PC_TO_INDEX(pc, prof) \ - ((int)(((u_quad_t)((pc) - (prof)->pr_off) * \ - (u_quad_t)((prof)->pr_scale)) >> 16) & ~1) - -static void -addupc_intr_forwarded(struct proc *p, int id, int *astmap) -{ - int i; - struct uprof *prof; - u_long pc; - - pc = checkstate_pc[id]; - prof = &p->p_stats->p_prof; - if (pc >= prof->pr_off && - (i = PC_TO_INDEX(pc, prof)) < prof->pr_size) { - mtx_assert(&sched_lock, MA_OWNED); - if ((p->p_sflag & PS_OWEUPC) == 0) { - prof->pr_addr = pc; - prof->pr_ticks = 1; - p->p_sflag |= PS_OWEUPC; - } - *astmap |= (1 << id); - } -} - -static void -forwarded_statclock(int id, int pscnt, int *astmap) -{ - struct pstats *pstats; - long rss; - struct rusage *ru; - struct vmspace *vm; - int cpustate; - struct proc *p; -#ifdef GPROF - register struct gmonparam *g; - int i; -#endif - - mtx_assert(&sched_lock, MA_OWNED); - p = checkstate_curproc[id]; - cpustate = checkstate_cpustate[id]; - - /* XXX */ - if (p->p_ithd) - cpustate = CHECKSTATE_INTR; - else if (p == SMP_prvspace[id].globaldata.gd_idleproc) - cpustate = CHECKSTATE_SYS; - - switch (cpustate) { - case CHECKSTATE_USER: - if (p->p_sflag & PS_PROFIL) - addupc_intr_forwarded(p, id, astmap); - if (pscnt > 1) - return; - p->p_uticks++; - if (p->p_nice > NZERO) - cp_time[CP_NICE]++; - else - cp_time[CP_USER]++; - break; - case CHECKSTATE_SYS: -#ifdef GPROF - /* - * Kernel statistics are just like addupc_intr, only easier. - */ - g = &_gmonparam; - if (g->state == GMON_PROF_ON) { - i = checkstate_pc[id] - g->lowpc; - if (i < g->textsize) { - i /= HISTFRACTION * sizeof(*g->kcount); - g->kcount[i]++; - } - } -#endif - if (pscnt > 1) - return; - - p->p_sticks++; - if (p == SMP_prvspace[id].globaldata.gd_idleproc) - cp_time[CP_IDLE]++; - else - cp_time[CP_SYS]++; - break; - case CHECKSTATE_INTR: - default: -#ifdef GPROF - /* - * Kernel statistics are just like addupc_intr, only easier. - */ - g = &_gmonparam; - if (g->state == GMON_PROF_ON) { - i = checkstate_pc[id] - g->lowpc; - if (i < g->textsize) { - i /= HISTFRACTION * sizeof(*g->kcount); - g->kcount[i]++; - } - } -#endif - if (pscnt > 1) - return; - KASSERT(p != NULL, ("NULL process in interrupt state")); - p->p_iticks++; - cp_time[CP_INTR]++; - } - - schedclock(p); - - /* Update resource usage integrals and maximums. */ - if ((pstats = p->p_stats) != NULL && - (ru = &pstats->p_ru) != NULL && - (vm = p->p_vmspace) != NULL) { - ru->ru_ixrss += pgtok(vm->vm_tsize); - ru->ru_idrss += pgtok(vm->vm_dsize); - ru->ru_isrss += pgtok(vm->vm_ssize); - rss = pgtok(vmspace_resident_count(vm)); - if (ru->ru_maxrss < rss) - ru->ru_maxrss = rss; - } -} - +/* + * For statclock, we send an IPI to all CPU's to have them call this + * function. + */ void -forward_statclock(int pscnt) +forwarded_statclock(struct trapframe frame) { - int map; - int id; - int i; - /* Kludge. We don't yet have separate locks for the interrupts - * and the kernel. This means that we cannot let the other processors - * handle complex interrupts while inhibiting them from entering - * the kernel in a non-interrupt context. - * - * What we can do, without changing the locking mechanisms yet, - * is letting the other processors handle a very simple interrupt - * (wich determines the processor states), and do the main - * work ourself. - */ - - CTR1(KTR_SMP, "forward_statclock(%d)", pscnt); - - if (!smp_started || !invltlb_ok || cold || panicstr) - return; - - /* Step 1: Probe state (user, cpu, interrupt, spinlock, idle ) */ - - map = PCPU_GET(other_cpus) & ~stopped_cpus ; - checkstate_probed_cpus = 0; - if (map != 0) - ipi_selected(map, IPI_CHECKSTATE); - - i = 0; - while (checkstate_probed_cpus != map) { - /* spin */ - i++; - if (i == 100000) { -#ifdef DIAGNOSTIC - printf("forward_statclock: checkstate %x\n", - checkstate_probed_cpus); -#endif - break; - } - } - - /* - * Step 2: walk through other processors processes, update ticks and - * profiling info. - */ - - map = 0; - for (id = 0; id < mp_ncpus; id++) { - if (id == PCPU_GET(cpuid)) - continue; - if (((1 << id) & checkstate_probed_cpus) == 0) - continue; - forwarded_statclock(id, pscnt, &map); - } - if (map != 0) { - checkstate_need_ast |= map; - ipi_selected(map, IPI_AST); - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#ifdef DIAGNOSTIC - printf("forward_statclock: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } - } + mtx_lock_spin(&sched_lock); + statclock_process(curproc, TRAPF_PC(&frame), TRAPF_USERMODE(&frame)); + mtx_unlock_spin(&sched_lock); } -void -forward_hardclock(int pscnt) +void +forward_statclock(void) { int map; - int id; - struct proc *p; - struct pstats *pstats; - int i; - - /* Kludge. We don't yet have separate locks for the interrupts - * and the kernel. This means that we cannot let the other processors - * handle complex interrupts while inhibiting them from entering - * the kernel in a non-interrupt context. - * - * What we can do, without changing the locking mechanisms yet, - * is letting the other processors handle a very simple interrupt - * (wich determines the processor states), and do the main - * work ourself. - */ - CTR1(KTR_SMP, "forward_hardclock(%d)", pscnt); + CTR0(KTR_SMP, "forward_statclock"); if (!smp_started || !invltlb_ok || cold || panicstr) return; - /* Step 1: Probe state (user, cpu, interrupt, spinlock, idle) */ - map = PCPU_GET(other_cpus) & ~stopped_cpus ; - checkstate_probed_cpus = 0; if (map != 0) - ipi_selected(map, IPI_CHECKSTATE); - - i = 0; - while (checkstate_probed_cpus != map) { - /* spin */ - i++; - if (i == 100000) { -#ifdef DIAGNOSTIC - printf("forward_hardclock: checkstate %x\n", - checkstate_probed_cpus); -#endif - break; - } - } - - /* - * Step 2: walk through other processors processes, update virtual - * timer and profiling timer. If stathz == 0, also update ticks and - * profiling info. - */ - - map = 0; - for (id = 0; id < mp_ncpus; id++) { - if (id == PCPU_GET(cpuid)) - continue; - if (((1 << id) & checkstate_probed_cpus) == 0) - continue; - p = checkstate_curproc[id]; - if (p) { - pstats = p->p_stats; - if (checkstate_cpustate[id] == CHECKSTATE_USER && - timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) && - itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) { - p->p_sflag |= PS_ALRMPEND; - map |= (1 << id); - } - if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) && - itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) { - p->p_sflag |= PS_PROFPEND; - map |= (1 << id); - } - } - if (stathz == 0) { - forwarded_statclock( id, pscnt, &map); - } - } - if (map != 0) { - checkstate_need_ast |= map; - ipi_selected(map, IPI_AST); - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#ifdef DIAGNOSTIC - printf("forward_hardclock: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } - } + ipi_selected(map, IPI_STATCLOCK); } -void -forward_signal(struct proc *p) +/* + * For each hardclock(), we send an IPI to all other CPU's to have them + * execute this function. It would be nice to reduce contention on + * sched_lock if we could simply peek at the CPU to determine the user/kernel + * state and call hardclock_process() on the CPU receiving the clock interrupt + * and then just use a simple IPI to handle any ast's if needed. + */ +void +forwarded_hardclock(struct trapframe frame) { - int map; - int id; - int i; - - /* Kludge. We don't yet have separate locks for the interrupts - * and the kernel. This means that we cannot let the other processors - * handle complex interrupts while inhibiting them from entering - * the kernel in a non-interrupt context. - * - * What we can do, without changing the locking mechanisms yet, - * is letting the other processors handle a very simple interrupt - * (wich determines the processor states), and do the main - * work ourself. - */ - - CTR1(KTR_SMP, "forward_signal(%p)", p); - if (!smp_started || !invltlb_ok || cold || panicstr) - return; - if (!forward_signal_enabled) - return; mtx_lock_spin(&sched_lock); - while (1) { - if (p->p_stat != SRUN) { - mtx_unlock_spin(&sched_lock); - return; - } - id = p->p_oncpu; - mtx_unlock_spin(&sched_lock); - if (id == 0xff) - return; - map = (1<<id); - checkstate_need_ast |= map; - ipi_selected(map, IPI_AST); - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#if 0 - printf("forward_signal: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } - mtx_lock_spin(&sched_lock); - if (id == p->p_oncpu) { - mtx_unlock_spin(&sched_lock); - return; - } - } + hardclock_process(curproc, TRAPF_USERMODE(&frame)); + mtx_unlock_spin(&sched_lock); } -void -forward_roundrobin(void) +void +forward_hardclock(void) { u_int map; - int i; - CTR0(KTR_SMP, "forward_roundrobin()"); + CTR0(KTR_SMP, "forward_hardclock"); if (!smp_started || !invltlb_ok || cold || panicstr) return; - if (!forward_roundrobin_enabled) - return; - resched_cpus |= PCPU_GET(other_cpus); - map = PCPU_GET(other_cpus) & ~stopped_cpus ; -#if 1 - ipi_selected(map, IPI_AST); -#else - ipi_all_but_self(IPI_AST); -#endif - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#if 0 - printf("forward_roundrobin: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } -} - -/* - * When called the executing CPU will send an IPI to all other CPUs - * requesting that they halt execution. - * - * Usually (but not necessarily) called with 'other_cpus' as its arg. - * - * - Signals all CPUs in map to stop. - * - Waits for each to stop. - * - * Returns: - * -1: error - * 0: NA - * 1: ok - * - * XXX FIXME: this is not MP-safe, needs a lock to prevent multiple CPUs - * from executing at same time. - */ -int -stop_cpus(u_int map) -{ - int count = 0; - - if (!smp_started) - return 0; - - /* send the Xcpustop IPI to all CPUs in map */ - ipi_selected(map, IPI_STOP); - - while (count++ < 100000 && (stopped_cpus & map) != map) - /* spin */ ; - -#ifdef DIAGNOSTIC - if ((stopped_cpus & map) != map) - printf("Warning: CPUs 0x%x did not stop!\n", - (~(stopped_cpus & map)) & map); -#endif - - return 1; -} - - -/* - * Called by a CPU to restart stopped CPUs. - * - * Usually (but not necessarily) called with 'stopped_cpus' as its arg. - * - * - Signals all CPUs in map to restart. - * - Waits for each to restart. - * - * Returns: - * -1: error - * 0: NA - * 1: ok - */ -int -restart_cpus(u_int map) -{ - int count = 0; - - if (!smp_started) - return 0; - - started_cpus = map; /* signal other cpus to restart */ - - /* wait for each to clear its bit */ - while (count++ < 100000 && (stopped_cpus & map) != 0) - /* spin */ ; -#ifdef DIAGNOSTIC - if ((stopped_cpus & map) != 0) - printf("Warning: CPUs 0x%x did not restart!\n", - (~(stopped_cpus & map)) & map); -#endif - - return 1; + map = PCPU_GET(other_cpus) & ~stopped_cpus ; + if (map != 0) + ipi_selected(map, IPI_HARDCLOCK); } - #ifdef APIC_INTR_REORDER /* * Maintain mapping from softintr vector to isr bit in local apic. @@ -2811,73 +2386,6 @@ set_lapic_isrloc(int intr, int vector) #endif /* - * All-CPU rendezvous. CPUs are signalled, all execute the setup function - * (if specified), rendezvous, execute the action function (if specified), - * rendezvous again, execute the teardown function (if specified), and then - * resume. - * - * Note that the supplied external functions _must_ be reentrant and aware - * that they are running in parallel and in an unknown lock context. - */ -static void (*smp_rv_setup_func)(void *arg); -static void (*smp_rv_action_func)(void *arg); -static void (*smp_rv_teardown_func)(void *arg); -static void *smp_rv_func_arg; -static volatile int smp_rv_waiters[2]; - -void -smp_rendezvous_action(void) -{ - /* setup function */ - if (smp_rv_setup_func != NULL) - smp_rv_setup_func(smp_rv_func_arg); - /* spin on entry rendezvous */ - atomic_add_int(&smp_rv_waiters[0], 1); - while (smp_rv_waiters[0] < mp_ncpus) - ; - /* action function */ - if (smp_rv_action_func != NULL) - smp_rv_action_func(smp_rv_func_arg); - /* spin on exit rendezvous */ - atomic_add_int(&smp_rv_waiters[1], 1); - while (smp_rv_waiters[1] < mp_ncpus) - ; - /* teardown function */ - if (smp_rv_teardown_func != NULL) - smp_rv_teardown_func(smp_rv_func_arg); -} - -void -smp_rendezvous(void (* setup_func)(void *), - void (* action_func)(void *), - void (* teardown_func)(void *), - void *arg) -{ - - /* obtain rendezvous lock */ - mtx_lock_spin(&smp_rv_mtx); - - /* set static function pointers */ - smp_rv_setup_func = setup_func; - smp_rv_action_func = action_func; - smp_rv_teardown_func = teardown_func; - smp_rv_func_arg = arg; - smp_rv_waiters[0] = 0; - smp_rv_waiters[1] = 0; - - /* - * signal other processors, which will enter the IPI with interrupts off - */ - ipi_all_but_self(IPI_RENDEZVOUS); - - /* call executor function */ - smp_rendezvous_action(); - - /* release lock */ - mtx_unlock_spin(&smp_rv_mtx); -} - -/* * send an IPI to a set of cpus. */ void diff --git a/sys/i386/include/pcpu.h b/sys/i386/include/pcpu.h index e713227..ef82a66 100644 --- a/sys/i386/include/pcpu.h +++ b/sys/i386/include/pcpu.h @@ -74,9 +74,6 @@ struct globaldata { #endif }; -SLIST_HEAD(cpuhead, globaldata); -extern struct cpuhead cpuhead; - #ifdef SMP /* * This is the upper (0xff800000) address space layout that is per-cpu. diff --git a/sys/i386/include/smp.h b/sys/i386/include/smp.h index 9f9149e..f699e1f 100644 --- a/sys/i386/include/smp.h +++ b/sys/i386/include/smp.h @@ -41,7 +41,21 @@ extern int current_postcode; /** XXX currently in mp_machdep.c */ outb(0x80, current_postcode) +#include <sys/bus.h> /* XXX */ #include <machine/apic.h> +#include <machine/frame.h> +#include <i386/isa/icu.h> +#include <i386/isa/intr_machdep.h> + +/* + * Interprocessor interrupts for SMP. + */ +#define IPI_INVLTLB XINVLTLB_OFFSET +#define IPI_RENDEZVOUS XRENDEZVOUS_OFFSET +#define IPI_AST XCPUAST_OFFSET +#define IPI_STOP XCPUSTOP_OFFSET +#define IPI_HARDCLOCK XHARDCLOCK_OFFSET +#define IPI_STATCLOCK XSTATCLOCK_OFFSET /* global data in mpboot.s */ extern int bootMP_size; @@ -49,18 +63,8 @@ extern int bootMP_size; /* functions in mpboot.s */ void bootMP __P((void)); -/* global data in apic_vector.s */ -extern volatile u_int stopped_cpus; -extern volatile u_int started_cpus; - -extern volatile u_int checkstate_probed_cpus; -extern volatile u_int checkstate_need_ast; -extern volatile u_int resched_cpus; -extern void (*cpustop_restartfunc) __P((void)); - /* global data in mp_machdep.c */ extern int bsp_apic_ready; -extern int mp_ncpus; extern int mp_naps; extern int mp_nbusses; extern int mp_napics; @@ -81,14 +85,11 @@ struct apic_intmapinfo { int redirindex; }; extern struct apic_intmapinfo int_to_apicintpin[]; -extern u_int all_cpus; extern struct pcb stoppcbs[]; /* functions in mp_machdep.c */ +void i386_mp_probe __P((void)); u_int mp_bootaddress __P((u_int)); -int mp_probe __P((void)); -void mp_start __P((void)); -void mp_announce __P((void)); u_int isa_apic_mask __P((u_int)); int isa_apic_irq __P((int)); int pci_apic_irq __P((int, int, int)); @@ -107,20 +108,17 @@ void revoke_apic_irq __P((int irq)); void bsp_apic_configure __P((void)); void init_secondary __P((void)); void smp_invltlb __P((void)); -int stop_cpus __P((u_int)); -int restart_cpus __P((u_int)); -void forward_statclock __P((int pscnt)); -void forward_hardclock __P((int pscnt)); -void forward_signal __P((struct proc *)); -void forward_roundrobin __P((void)); +void forward_statclock __P((void)); +void forwarded_statclock __P((struct trapframe frame)); +void forward_hardclock __P((void)); +void forwarded_hardclock __P((struct trapframe frame)); +void ipi_selected __P((u_int cpus, u_int ipi)); +void ipi_all __P((u_int ipi)); +void ipi_all_but_self __P((u_int ipi)); +void ipi_self __P((u_int ipi)); #ifdef APIC_INTR_REORDER void set_lapic_isrloc __P((int, int)); #endif /* APIC_INTR_REORDER */ -void smp_rendezvous_action __P((void)); -void smp_rendezvous __P((void (*)(void *), - void (*)(void *), - void (*)(void *), - void *arg)); /* global data in mpapic.c */ extern volatile lapic_t lapic; @@ -147,8 +145,6 @@ void io_apic_write __P((int, int, u_int)); /* global data in init_smp.c */ extern int invltlb_ok; -extern int smp_active; -extern int smp_started; extern volatile int smp_idle_loops; #endif /* !LOCORE */ diff --git a/sys/i386/isa/apic_vector.s b/sys/i386/isa/apic_vector.s index 3f0521f..5c68f81 100644 --- a/sys/i386/isa/apic_vector.s +++ b/sys/i386/isa/apic_vector.s @@ -182,7 +182,6 @@ Xspuriousint: iret - /* * Handle TLB shootdowns. */ @@ -211,71 +210,61 @@ Xinvltlb: popl %eax iret - /* - * Executed by a CPU when it receives an Xcpucheckstate IPI from another CPU, - * - * - Stores current cpu state in checkstate_cpustate[cpuid] - * 0 == user, 1 == sys, 2 == intr - * - Stores current process in checkstate_curproc[cpuid] - * - * - Signals its receipt by setting bit cpuid in checkstate_probed_cpus. - * - * stack: 0->ds, 4->fs, 8->ebx, 12->eax, 16->eip, 20->cs, 24->eflags + * Forward hardclock to another CPU. Pushes a trapframe and calls + * forwarded_hardclock(). */ - .text SUPERALIGN_TEXT - .globl Xcpucheckstate - .globl checkstate_cpustate - .globl checkstate_curproc - .globl checkstate_pc -Xcpucheckstate: - pushl %eax - pushl %ebx - pushl %ds /* save current data segment */ - pushl %fs - - movl $KDSEL, %eax - mov %ax, %ds /* use KERNEL data segment */ + .globl Xhardclock +Xhardclock: + PUSH_FRAME + movl $KDSEL, %eax /* reload with kernel's data segment */ + mov %ax, %ds + mov %ax, %es movl $KPSEL, %eax mov %ax, %fs movl $0, lapic+LA_EOI /* End Of Interrupt to APIC */ - movl $0, %ebx - movl 20(%esp), %eax - andl $3, %eax - cmpl $3, %eax - je 1f - testl $PSL_VM, 24(%esp) - jne 1f - incl %ebx /* system or interrupt */ -1: - movl PCPU(CPUID), %eax - movl %ebx, checkstate_cpustate(,%eax,4) - movl PCPU(CURPROC), %ebx - movl %ebx, checkstate_curproc(,%eax,4) - - movl 16(%esp), %ebx - movl %ebx, checkstate_pc(,%eax,4) + movl PCPU(CURPROC),%ebx + incl P_INTR_NESTING_LEVEL(%ebx) + call forwarded_hardclock + decl P_INTR_NESTING_LEVEL(%ebx) + MEXITCOUNT + jmp doreti - lock /* checkstate_probed_cpus |= (1<<id) */ - btsl %eax, checkstate_probed_cpus +/* + * Forward statclock to another CPU. Pushes a trapframe and calls + * forwarded_statclock(). + */ + .text + SUPERALIGN_TEXT + .globl Xstatclock +Xstatclock: + PUSH_FRAME + movl $KDSEL, %eax /* reload with kernel's data segment */ + mov %ax, %ds + mov %ax, %es + movl $KPSEL, %eax + mov %ax, %fs - popl %fs - popl %ds /* restore previous data segment */ - popl %ebx - popl %eax - iret + movl $0, lapic+LA_EOI /* End Of Interrupt to APIC */ + FAKE_MCOUNT(13*4(%esp)) + movl PCPU(CURPROC),%ebx + incl P_INTR_NESTING_LEVEL(%ebx) + call forwarded_statclock + decl P_INTR_NESTING_LEVEL(%ebx) + MEXITCOUNT + jmp doreti /* * Executed by a CPU when it receives an Xcpuast IPI from another CPU, * - * - Signals its receipt by clearing bit cpuid in checkstate_need_ast. - * - * - We need a better method of triggering asts on other cpus. + * The other CPU has already executed aston() or need_resched() on our + * current process, so we simply need to ack the interrupt and return + * via doreti to run ast(). */ .text @@ -289,40 +278,12 @@ Xcpuast: movl $KPSEL, %eax mov %ax, %fs - movl PCPU(CPUID), %eax - lock /* checkstate_need_ast &= ~(1<<id) */ - btrl %eax, checkstate_need_ast movl $0, lapic+LA_EOI /* End Of Interrupt to APIC */ - lock - btsl %eax, checkstate_pending_ast - jc 1f - FAKE_MCOUNT(13*4(%esp)) - MTX_LOCK_SPIN(sched_lock, 0) - movl PCPU(CURPROC),%ebx - orl $PS_ASTPENDING, P_SFLAG(%ebx) - - movl PCPU(CPUID), %eax - lock - btrl %eax, checkstate_pending_ast - lock - btrl %eax, CNAME(resched_cpus) - jnc 2f - orl $PS_NEEDRESCHED, P_SFLAG(%ebx) - lock - incl CNAME(want_resched_cnt) -2: - MTX_UNLOCK_SPIN(sched_lock) - lock - incl CNAME(cpuast_cnt) MEXITCOUNT jmp doreti -1: - /* We are already in the process of delivering an ast for this CPU */ - POP_FRAME - iret /* * Executed by a CPU when it receives an Xcpustop IPI from another CPU, @@ -331,7 +292,6 @@ Xcpuast: * - Waits for permission to restart. * - Signals its restart. */ - .text SUPERALIGN_TEXT .globl Xcpustop @@ -357,20 +317,19 @@ Xcpustop: pushl %eax call CNAME(savectx) /* Save process context */ addl $4, %esp - movl PCPU(CPUID), %eax lock - btsl %eax, stopped_cpus /* stopped_cpus |= (1<<id) */ + btsl %eax, CNAME(stopped_cpus) /* stopped_cpus |= (1<<id) */ 1: - btl %eax, started_cpus /* while (!(started_cpus & (1<<id))) */ + btl %eax, CNAME(started_cpus) /* while (!(started_cpus & (1<<id))) */ jnc 1b lock - btrl %eax, started_cpus /* started_cpus &= ~(1<<id) */ + btrl %eax, CNAME(started_cpus) /* started_cpus &= ~(1<<id) */ lock - btrl %eax, stopped_cpus /* stopped_cpus &= ~(1<<id) */ + btrl %eax, CNAME(stopped_cpus) /* stopped_cpus &= ~(1<<id) */ test %eax, %eax jnz 2f @@ -492,34 +451,6 @@ _xhits: .space (NCPU * 4), 0 #endif /* COUNT_XINVLTLB_HITS */ -/* variables used by stop_cpus()/restart_cpus()/Xcpustop */ - .globl stopped_cpus, started_cpus -stopped_cpus: - .long 0 -started_cpus: - .long 0 - - .globl checkstate_probed_cpus -checkstate_probed_cpus: - .long 0 - .globl checkstate_need_ast -checkstate_need_ast: - .long 0 -checkstate_pending_ast: - .long 0 - .globl CNAME(resched_cpus) - .globl CNAME(want_resched_cnt) - .globl CNAME(cpuast_cnt) - .globl CNAME(cpustop_restartfunc) -CNAME(resched_cpus): - .long 0 -CNAME(want_resched_cnt): - .long 0 -CNAME(cpuast_cnt): - .long 0 -CNAME(cpustop_restartfunc): - .long 0 - .globl apic_pin_trigger apic_pin_trigger: .long 0 diff --git a/sys/i386/isa/clock.c b/sys/i386/isa/clock.c index 9034f81..ce67390 100644 --- a/sys/i386/isa/clock.c +++ b/sys/i386/isa/clock.c @@ -211,6 +211,10 @@ clkintr(struct clockframe frame) mtx_unlock_spin(&clock_lock); } timer_func(&frame); +#ifdef SMP + if (timer_func == hardclock) + forward_hardclock(); +#endif switch (timer0_state) { case RELEASED: @@ -253,6 +257,9 @@ clkintr(struct clockframe frame) timer_func = hardclock; timer0_state = RELEASED; hardclock(&frame); +#ifdef SMP + forward_hardclock(); +#endif } break; } @@ -374,8 +381,12 @@ release_timer2() static void rtcintr(struct clockframe frame) { - while (rtcin(RTC_INTR) & RTCIR_PERIOD) + while (rtcin(RTC_INTR) & RTCIR_PERIOD) { statclock(&frame); +#ifdef SMP + forward_statclock(); +#endif + } } #include "opt_ddb.h" diff --git a/sys/i386/isa/intr_machdep.h b/sys/i386/isa/intr_machdep.h index ed58437..6350524 100644 --- a/sys/i386/isa/intr_machdep.h +++ b/sys/i386/isa/intr_machdep.h @@ -107,10 +107,11 @@ #define XINVLTLB_OFFSET (ICU_OFFSET + 112) /* inter-cpu clock handling */ -#define XCPUCHECKSTATE_OFFSET (ICU_OFFSET + 113) +#define XHARDCLOCK_OFFSET (ICU_OFFSET + 113) +#define XSTATCLOCK_OFFSET (ICU_OFFSET + 114) /* inter-CPU rendezvous */ -#define XRENDEZVOUS_OFFSET (ICU_OFFSET + 114) +#define XRENDEZVOUS_OFFSET (ICU_OFFSET + 115) /* IPI to generate an additional software trap at the target CPU */ #define XCPUAST_OFFSET (ICU_OFFSET + 48) @@ -173,7 +174,8 @@ inthand_t inthand_t Xinvltlb, /* TLB shootdowns */ - Xcpucheckstate, /* Check cpu state */ + Xhardclock, /* Forward hardclock() */ + Xstatclock, /* Forward statclock() */ Xcpuast, /* Additional software trap on other cpu */ Xcpustop, /* CPU stops & waits for another CPU to restart it */ Xspuriousint, /* handle APIC "spurious INTs" */ diff --git a/sys/ia64/ia64/db_interface.c b/sys/ia64/ia64/db_interface.c index 3750189..7069f60 100644 --- a/sys/ia64/ia64/db_interface.c +++ b/sys/ia64/ia64/db_interface.c @@ -44,6 +44,7 @@ #include <sys/reboot.h> #include <sys/systm.h> #include <sys/kernel.h> +#include <sys/smp.h> #include <sys/cons.h> #include <sys/ktr.h> @@ -51,7 +52,6 @@ #include <machine/db_machdep.h> #include <machine/mutex.h> -#include <machine/smp.h> #include <machine/inst.h> diff --git a/sys/ia64/ia64/machdep.c b/sys/ia64/ia64/machdep.c index a113c9b..9eac9ed 100644 --- a/sys/ia64/ia64/machdep.c +++ b/sys/ia64/ia64/machdep.c @@ -82,8 +82,6 @@ u_int32_t cycles_per_sec; int cold = 1; struct bootinfo_kernel bootinfo; -struct cpuhead cpuhead; - struct mtx sched_lock; struct mtx Giant; @@ -579,12 +577,6 @@ ia64_init() proc0.p_md.md_tf = (struct trapframe *)(proc0.p_addr->u_pcb.pcb_sp + 16); - /* - * Record all cpus in a list. - */ - SLIST_INIT(&cpuhead); - SLIST_INSERT_HEAD(&cpuhead, GLOBALP, gd_allcpu); - /* Setup curproc so that mutexes work */ PCPU_SET(curproc, &proc0); PCPU_SET(spinlocks, NULL); @@ -1367,3 +1359,16 @@ ia64_fpstate_switch(struct proc *p) p->p_md.md_flags |= MDP_FPUSED; } + +/* + * Initialise a struct globaldata. + */ +void +globaldata_init(struct globaldata *globaldata, int cpuid, size_t sz) +{ + bzero(globaldata, sz); + globaldata->gd_cpuid = cpuid; +#ifdef SMP + globaldata_register(globaldata); +#endif +} diff --git a/sys/ia64/ia64/mp_machdep.c b/sys/ia64/ia64/mp_machdep.c index db73f61..143963a 100644 --- a/sys/ia64/ia64/mp_machdep.c +++ b/sys/ia64/ia64/mp_machdep.c @@ -34,6 +34,7 @@ #include <sys/malloc.h> #include <sys/mutex.h> #include <sys/kernel.h> +#include <sys/smp.h> #include <sys/sysctl.h> #include <vm/vm.h> @@ -42,7 +43,6 @@ #include <sys/user.h> #include <sys/dkstat.h> -#include <machine/smp.h> #include <machine/atomic.h> #include <machine/ipl.h> #include <machine/globaldata.h> @@ -53,8 +53,6 @@ #define CHECKSTATE_SYS 1 #define CHECKSTATE_INTR 2 -volatile u_int stopped_cpus; -volatile u_int started_cpus; volatile u_int checkstate_probed_cpus; volatile u_int checkstate_need_ast; volatile u_int checkstate_pending_ast; @@ -62,76 +60,27 @@ struct proc* checkstate_curproc[MAXCPU]; int checkstate_cpustate[MAXCPU]; u_long checkstate_pc[MAXCPU]; volatile u_int resched_cpus; -void (*cpustop_restartfunc) __P((void)); -int mp_ncpus; -int smp_started; int boot_cpu_id; -u_int32_t all_cpus; - -static struct globaldata *cpuid_to_globaldata[MAXCPU]; - -int smp_active = 0; /* are the APs allowed to run? */ -SYSCTL_INT(_machdep, OID_AUTO, smp_active, CTLFLAG_RW, &smp_active, 0, ""); /* Is forwarding of a interrupt to the CPU holding the ISR lock enabled ? */ int forward_irq_enabled = 1; SYSCTL_INT(_machdep, OID_AUTO, forward_irq_enabled, CTLFLAG_RW, &forward_irq_enabled, 0, ""); -/* Enable forwarding of a signal to a process running on a different CPU */ -static int forward_signal_enabled = 1; -SYSCTL_INT(_machdep, OID_AUTO, forward_signal_enabled, CTLFLAG_RW, - &forward_signal_enabled, 0, ""); - -/* Enable forwarding of roundrobin to all other cpus */ -static int forward_roundrobin_enabled = 1; -SYSCTL_INT(_machdep, OID_AUTO, forward_roundrobin_enabled, CTLFLAG_RW, - &forward_roundrobin_enabled, 0, ""); - -/* - * Initialise a struct globaldata. - */ -void -globaldata_init(struct globaldata *globaldata, int cpuid, size_t sz) -{ - bzero(globaldata, sz); - globaldata->gd_cpuid = cpuid; - globaldata->gd_other_cpus = all_cpus & ~(1 << cpuid); - cpuid_to_globaldata[cpuid] = globaldata; -} - -struct globaldata * -globaldata_find(int cpuid) -{ - return cpuid_to_globaldata[cpuid]; -} - -/* Other stuff */ - -/* lock around the MP rendezvous */ -static struct mtx smp_rv_mtx; - -static void -init_locks(void) -{ - - mtx_init(&smp_rv_mtx, "smp_rendezvous", MTX_SPIN); -} - -void -mp_start() +int +cpu_mp_probe() { - init_locks(); + return (0); } void -mp_announce() +cpu_mp_start() { } void -smp_invltlb() +cpu_mp_announce() { } @@ -426,227 +375,6 @@ forward_hardclock(int pscnt) } } -void -forward_signal(struct proc *p) -{ - int map; - int id; - int i; - - /* Kludge. We don't yet have separate locks for the interrupts - * and the kernel. This means that we cannot let the other processors - * handle complex interrupts while inhibiting them from entering - * the kernel in a non-interrupt context. - * - * What we can do, without changing the locking mechanisms yet, - * is letting the other processors handle a very simple interrupt - * (wich determines the processor states), and do the main - * work ourself. - */ - - CTR1(KTR_SMP, "forward_signal(%p)", p); - - if (!smp_started || cold || panicstr) - return; - if (!forward_signal_enabled) - return; - while (1) { - if (p->p_stat != SRUN) - return; - id = p->p_oncpu; - if (id == 0xff) - return; - map = (1<<id); - checkstate_need_ast |= map; - ipi_selected(map, IPI_AST); - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#if 0 - printf("forward_signal: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } - if (id == p->p_oncpu) - return; - } -} - -void -forward_roundrobin(void) -{ - u_int map; - int i; - - CTR0(KTR_SMP, "forward_roundrobin()"); - - if (!smp_started || cold || panicstr) - return; - if (!forward_roundrobin_enabled) - return; - resched_cpus |= PCPU_GET(other_cpus); - map = PCPU_GET(other_cpus) & ~stopped_cpus ; - ipi_selected(map, IPI_AST); - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#if 0 - printf("forward_roundrobin: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } -} - -/* - * When called the executing CPU will send an IPI to all other CPUs - * requesting that they halt execution. - * - * Usually (but not necessarily) called with 'other_cpus' as its arg. - * - * - Signals all CPUs in map to stop. - * - Waits for each to stop. - * - * Returns: - * -1: error - * 0: NA - * 1: ok - * - * XXX FIXME: this is not MP-safe, needs a lock to prevent multiple CPUs - * from executing at same time. - */ -int -stop_cpus(u_int map) -{ - int i; - - if (!smp_started) - return 0; - - CTR1(KTR_SMP, "stop_cpus(%x)", map); - - /* send the stop IPI to all CPUs in map */ - ipi_selected(map, IPI_STOP); - - i = 0; - while ((stopped_cpus & map) != map) { - /* spin */ - i++; - if (i == 100000) { - printf("timeout stopping cpus\n"); - break; - } - ia64_mf(); - } - - printf("stopped_cpus=%x\n", stopped_cpus); - - return 1; -} - - -/* - * Called by a CPU to restart stopped CPUs. - * - * Usually (but not necessarily) called with 'stopped_cpus' as its arg. - * - * - Signals all CPUs in map to restart. - * - Waits for each to restart. - * - * Returns: - * -1: error - * 0: NA - * 1: ok - */ -int -restart_cpus(u_int map) -{ - if (!smp_started) - return 0; - - CTR1(KTR_SMP, "restart_cpus(%x)", map); - - started_cpus = map; /* signal other cpus to restart */ - ia64_mf(); - - while ((stopped_cpus & map) != 0) /* wait for each to clear its bit */ - ia64_mf(); - - return 1; -} - -/* - * All-CPU rendezvous. CPUs are signalled, all execute the setup function - * (if specified), rendezvous, execute the action function (if specified), - * rendezvous again, execute the teardown function (if specified), and then - * resume. - * - * Note that the supplied external functions _must_ be reentrant and aware - * that they are running in parallel and in an unknown lock context. - */ -static void (*smp_rv_setup_func)(void *arg); -static void (*smp_rv_action_func)(void *arg); -static void (*smp_rv_teardown_func)(void *arg); -static void *smp_rv_func_arg; -static volatile int smp_rv_waiters[2]; - -void -smp_rendezvous_action(void) -{ - /* setup function */ - if (smp_rv_setup_func != NULL) - smp_rv_setup_func(smp_rv_func_arg); - /* spin on entry rendezvous */ - atomic_add_int(&smp_rv_waiters[0], 1); - while (smp_rv_waiters[0] < mp_ncpus) - ; - /* action function */ - if (smp_rv_action_func != NULL) - smp_rv_action_func(smp_rv_func_arg); - /* spin on exit rendezvous */ - atomic_add_int(&smp_rv_waiters[1], 1); - while (smp_rv_waiters[1] < mp_ncpus) - ; - /* teardown function */ - if (smp_rv_teardown_func != NULL) - smp_rv_teardown_func(smp_rv_func_arg); -} - -void -smp_rendezvous(void (* setup_func)(void *), - void (* action_func)(void *), - void (* teardown_func)(void *), - void *arg) -{ - - /* obtain rendezvous lock */ - mtx_lock_spin(&smp_rv_mtx); - - /* set static function pointers */ - smp_rv_setup_func = setup_func; - smp_rv_action_func = action_func; - smp_rv_teardown_func = teardown_func; - smp_rv_func_arg = arg; - smp_rv_waiters[0] = 0; - smp_rv_waiters[1] = 0; - - /* signal other processors, which will enter the IPI with interrupts off */ - ipi_all_but_self(IPI_RENDEZVOUS); - - /* call executor function */ - smp_rendezvous_action(); - - /* release lock */ - mtx_unlock_spin(&smp_rv_mtx); -} - /* * send an IPI to a set of cpus. */ @@ -661,7 +389,7 @@ ipi_selected(u_int32_t cpus, u_int64_t ipi) int cpuid = ffs(cpus) - 1; cpus &= ~(1 << cpuid); - globaldata = cpuid_to_globaldata[cpuid]; + globaldata = globaldata_find(cpuid); if (globaldata) { atomic_set_64(&globaldata->gd_pending_ipis, ipi); ia64_mf(); @@ -733,9 +461,6 @@ smp_handle_ipi(struct trapframe *frame) CTR0(KTR_SMP, "IPI_AST"); atomic_clear_int(&checkstate_need_ast, 1<<cpuid); atomic_set_int(&checkstate_pending_ast, 1<<cpuid); - if ((frame->tf_cr_ipsr & IA64_PSR_CPL) - == IA64_PSR_CPL_USER) - ast(frame); /* XXX */ break; case IPI_CHECKSTATE: diff --git a/sys/ia64/ia64/trap.c b/sys/ia64/ia64/trap.c index 8486877..8f9b258 100644 --- a/sys/ia64/ia64/trap.c +++ b/sys/ia64/ia64/trap.c @@ -41,6 +41,7 @@ #include <sys/exec.h> #include <sys/lock.h> #include <sys/mutex.h> +#include <sys/smp.h> #include <sys/vmmeter.h> #include <sys/sysent.h> #include <sys/syscall.h> @@ -58,7 +59,6 @@ #include <machine/reg.h> #include <machine/pal.h> #include <machine/fpu.h> -#include <machine/smp.h> #ifdef KTRACE #include <sys/uio.h> @@ -90,7 +90,7 @@ userret(register struct proc *p, struct trapframe *frame, u_quad_t oticks) mtx_lock_spin(&sched_lock); p->p_pri.pri_level = p->p_pri.pri_user; - if (resched_wanted()) { + if (resched_wanted(p)) { /* * Since we are curproc, a clock interrupt could * change our priority without changing run queues @@ -638,7 +638,7 @@ ast(framep) * acquiring and release mutexes in assembly is not fun. */ mtx_lock_spin(&sched_lock); - if (!(astpending(p) || resched_wanted())) { + if (!(astpending(p) || resched_wanted(p))) { mtx_unlock_spin(&sched_lock); return; } diff --git a/sys/ia64/include/cpu.h b/sys/ia64/include/cpu.h index 2b15a48..61766d2 100644 --- a/sys/ia64/include/cpu.h +++ b/sys/ia64/include/cpu.h @@ -68,9 +68,6 @@ struct clockframe { #define CLKF_USERMODE(framep) TRAPF_USERMODE(&(framep)->cf_tf) #define CLKF_PC(framep) TRAPF_PC(&(framep)->cf_tf) -#define CLKF_BASEPRI(framep) \ - (((framep)->cf_tf.tf_cr_ipsr & IA64_PSR_I) == 0) -#define CLKF_INTR(framep) (curproc->p_intr_nesting_level >= 2) /* * Give a profiling tick to the current process when the user profiling diff --git a/sys/ia64/include/globaldata.h b/sys/ia64/include/globaldata.h index a92f543..0ec1e19 100644 --- a/sys/ia64/include/globaldata.h +++ b/sys/ia64/include/globaldata.h @@ -66,11 +66,7 @@ struct globaldata { #endif }; -SLIST_HEAD(cpuhead, globaldata); -extern struct cpuhead cpuhead; - void globaldata_init(struct globaldata *pcpu, int cpuid, size_t sz); -struct globaldata *globaldata_find(int cpuid); #endif /* _KERNEL */ diff --git a/sys/ia64/include/ipl.h b/sys/ia64/include/ipl.h index 8432299..cbdecbb 100644 --- a/sys/ia64/include/ipl.h +++ b/sys/ia64/include/ipl.h @@ -29,19 +29,4 @@ #ifndef _MACHINE_IPL_H_ #define _MACHINE_IPL_H_ -/* - * Interprocessor interrupts for SMP. - */ -#define IPI_INVLTLB 0x0001 -#define IPI_RENDEZVOUS 0x0002 -#define IPI_AST 0x0004 -#define IPI_CHECKSTATE 0x0008 -#define IPI_STOP 0x0010 - -void ipi_selected(u_int32_t cpus, u_int64_t ipi); -void ipi_all(u_int64_t ipi); -void ipi_all_but_self(u_int64_t ipi); -void ipi_self(u_int64_t ipi); -void smp_handle_ipi(struct trapframe *frame); - #endif /* !_MACHINE_IPL_H_ */ diff --git a/sys/ia64/include/pcpu.h b/sys/ia64/include/pcpu.h index a92f543..0ec1e19 100644 --- a/sys/ia64/include/pcpu.h +++ b/sys/ia64/include/pcpu.h @@ -66,11 +66,7 @@ struct globaldata { #endif }; -SLIST_HEAD(cpuhead, globaldata); -extern struct cpuhead cpuhead; - void globaldata_init(struct globaldata *pcpu, int cpuid, size_t sz); -struct globaldata *globaldata_find(int cpuid); #endif /* _KERNEL */ diff --git a/sys/ia64/include/smp.h b/sys/ia64/include/smp.h index 65c85ef..e4f2091 100644 --- a/sys/ia64/include/smp.h +++ b/sys/ia64/include/smp.h @@ -6,43 +6,28 @@ #ifdef _KERNEL -#include <machine/mutex.h> -#include <machine/ipl.h> -#include <sys/ktr.h> +/* + * Interprocessor interrupts for SMP. + */ +#define IPI_INVLTLB 0x0001 +#define IPI_RENDEZVOUS 0x0002 +#define IPI_AST 0x0004 +#define IPI_CHECKSTATE 0x0008 +#define IPI_STOP 0x0010 #ifndef LOCORE -#define BETTER_CLOCK /* unconditional on ia64 */ - /* global data in mp_machdep.c */ extern volatile u_int checkstate_probed_cpus; extern volatile u_int checkstate_need_ast; extern volatile u_int resched_cpus; -extern void (*cpustop_restartfunc) __P((void)); - -extern int smp_active; -extern int mp_ncpus; -extern u_int all_cpus; -extern u_int started_cpus; -extern u_int stopped_cpus; -/* functions in mp_machdep.c */ -void mp_start(void); -void mp_announce(void); -void smp_invltlb(void); -void forward_statclock(int pscnt); -void forward_hardclock(int pscnt); -void forward_signal(struct proc *); -void forward_roundrobin(void); -int stop_cpus(u_int); -int restart_cpus(u_int); -void smp_rendezvous_action(void); -void smp_rendezvous(void (*)(void *), - void (*)(void *), - void (*)(void *), - void *arg); +void ipi_selected(u_int cpus, u_int ipi); +void ipi_all(u_int ipi); +void ipi_all_but_self(u_int ipi); +void ipi_self(u_int ipi); void smp_init_secondary(void); #endif /* !LOCORE */ #endif /* _KERNEL */ -#endif +#endif /* !_MACHINE_SMP_H */ diff --git a/sys/isa/atrtc.c b/sys/isa/atrtc.c index 9034f81..ce67390 100644 --- a/sys/isa/atrtc.c +++ b/sys/isa/atrtc.c @@ -211,6 +211,10 @@ clkintr(struct clockframe frame) mtx_unlock_spin(&clock_lock); } timer_func(&frame); +#ifdef SMP + if (timer_func == hardclock) + forward_hardclock(); +#endif switch (timer0_state) { case RELEASED: @@ -253,6 +257,9 @@ clkintr(struct clockframe frame) timer_func = hardclock; timer0_state = RELEASED; hardclock(&frame); +#ifdef SMP + forward_hardclock(); +#endif } break; } @@ -374,8 +381,12 @@ release_timer2() static void rtcintr(struct clockframe frame) { - while (rtcin(RTC_INTR) & RTCIR_PERIOD) + while (rtcin(RTC_INTR) & RTCIR_PERIOD) { statclock(&frame); +#ifdef SMP + forward_statclock(); +#endif + } } #include "opt_ddb.h" diff --git a/sys/kern/kern_clock.c b/sys/kern/kern_clock.c index c7492c7..dca9a46 100644 --- a/sys/kern/kern_clock.c +++ b/sys/kern/kern_clock.c @@ -52,6 +52,7 @@ #include <sys/proc.h> #include <sys/resourcevar.h> #include <sys/signalvar.h> +#include <sys/smp.h> #include <sys/timetc.h> #include <sys/timepps.h> #include <vm/vm.h> @@ -63,7 +64,6 @@ #include <machine/cpu.h> #include <machine/limits.h> -#include <machine/smp.h> #ifdef GPROF #include <sys/gmon.h> @@ -150,46 +150,57 @@ initclocks(dummy) } /* + * Each time the real-time timer fires, this function is called on all CPUs + * with each CPU passing in its curproc as the first argument. If possible + * a nice optimization in the future would be to allow the CPU receiving the + * actual real-time timer interrupt to call this function on behalf of the + * other CPUs rather than sending an IPI to all other CPUs so that they + * can call this function. Note that hardclock() calls hardclock_process() + * for the CPU receiving the timer interrupt, so only the other CPUs in the + * system need to call this function (or have it called on their behalf. + */ +void +hardclock_process(p, user) + struct proc *p; + int user; +{ + struct pstats *pstats; + + /* + * Run current process's virtual and profile time, as needed. + */ + mtx_assert(&sched_lock, MA_OWNED); + pstats = p->p_stats; + if (user && + timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) && + itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) { + p->p_sflag |= PS_ALRMPEND; + aston(p); + } + if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) && + itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) { + p->p_sflag |= PS_PROFPEND; + aston(p); + } +} + +/* * The real-time timer, interrupting hz times per second. */ void hardclock(frame) register struct clockframe *frame; { - register struct proc *p; int need_softclock = 0; - p = curproc; - if (p != PCPU_GET(idleproc)) { - register struct pstats *pstats; - - /* - * Run current process's virtual and profile time, as needed. - */ - pstats = p->p_stats; - if (CLKF_USERMODE(frame) && - timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) && - itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) { - mtx_lock_spin(&sched_lock); - p->p_sflag |= PS_ALRMPEND; - aston(p); - mtx_unlock_spin(&sched_lock); - } - if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) && - itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) { - mtx_lock_spin(&sched_lock); - p->p_sflag |= PS_PROFPEND; - aston(p); - mtx_unlock_spin(&sched_lock); - } - } - -#if defined(SMP) && defined(__i386__) - forward_hardclock(pscnt); -#endif + mtx_lock_spin(&sched_lock); + hardclock_process(curproc, CLKF_USERMODE(frame)); + mtx_unlock_spin(&sched_lock); /* * If no separate statistics clock is available, run it from here. + * + * XXX: this only works for UP */ if (stathz == 0) statclock(frame); @@ -328,43 +339,39 @@ stopprofclock(p) } /* - * Statistics clock. Grab profile sample, and if divider reaches 0, - * do process and kernel statistics. Most of the statistics are only + * Do process and kernel statistics. Most of the statistics are only * used by user-level statistics programs. The main exceptions are - * p->p_uticks, p->p_sticks, p->p_iticks, and p->p_estcpu. + * p->p_uticks, p->p_sticks, p->p_iticks, and p->p_estcpu. This function + * should be called by all CPUs in the system for each statistics clock + * interrupt. See the description of hardclock_process for more detail on + * this function's relationship to statclock. */ void -statclock(frame) - register struct clockframe *frame; +statclock_process(p, pc, user) + struct proc *p; + register_t pc; + int user; { #ifdef GPROF - register struct gmonparam *g; + struct gmonparam *g; int i; #endif - register struct proc *p; struct pstats *pstats; long rss; struct rusage *ru; struct vmspace *vm; - mtx_lock_spin(&sched_lock); - - if (CLKF_USERMODE(frame)) { + KASSERT(p == curproc, ("statclock_process: p != curproc")); + mtx_assert(&sched_lock, MA_OWNED); + if (user) { /* * Came from user mode; CPU was in user state. * If this process is being profiled, record the tick. */ - p = curproc; if (p->p_sflag & PS_PROFIL) - addupc_intr(p, CLKF_PC(frame), 1); -#if defined(SMP) && defined(__i386__) - if (stathz != 0) - forward_statclock(pscnt); -#endif - if (--pscnt > 0) { - mtx_unlock_spin(&sched_lock); + addupc_intr(p, pc, 1); + if (pscnt < psdiv) return; - } /* * Charge the time as appropriate. */ @@ -380,21 +387,15 @@ statclock(frame) */ g = &_gmonparam; if (g->state == GMON_PROF_ON) { - i = CLKF_PC(frame) - g->lowpc; + i = pc - g->lowpc; if (i < g->textsize) { i /= HISTFRACTION * sizeof(*g->kcount); g->kcount[i]++; } } #endif -#if defined(SMP) && defined(__i386__) - if (stathz != 0) - forward_statclock(pscnt); -#endif - if (--pscnt > 0) { - mtx_unlock_spin(&sched_lock); + if (pscnt < psdiv) return; - } /* * Came from kernel mode, so we were: * - handling an interrupt, @@ -407,8 +408,7 @@ statclock(frame) * so that we know how much of its real time was spent * in ``non-process'' (i.e., interrupt) work. */ - p = curproc; - if ((p->p_ithd != NULL) || CLKF_INTR(frame)) { + if ((p->p_ithd != NULL) || p->p_intr_nesting_level >= 2) { p->p_iticks++; cp_time[CP_INTR]++; } else { @@ -419,7 +419,6 @@ statclock(frame) cp_time[CP_IDLE]++; } } - pscnt = psdiv; schedclock(p); @@ -434,7 +433,23 @@ statclock(frame) if (ru->ru_maxrss < rss) ru->ru_maxrss = rss; } +} +/* + * Statistics clock. Grab profile sample, and if divider reaches 0, + * do process and kernel statistics. Most of the statistics are only + * used by user-level statistics programs. The main exceptions are + * p->p_uticks, p->p_sticks, p->p_iticks, and p->p_estcpu. + */ +void +statclock(frame) + register struct clockframe *frame; +{ + + mtx_lock_spin(&sched_lock); + if (--pscnt == 0) + pscnt = psdiv; + statclock_process(curproc, CLKF_PC(frame), CLKF_USERMODE(frame)); mtx_unlock_spin(&sched_lock); } diff --git a/sys/kern/kern_idle.c b/sys/kern/kern_idle.c index c7a013d..f7dd122 100644 --- a/sys/kern/kern_idle.c +++ b/sys/kern/kern_idle.c @@ -12,6 +12,7 @@ #include <sys/kernel.h> #include <sys/ktr.h> #include <sys/signalvar.h> +#include <sys/smp.h> #include <sys/resourcevar.h> #include <sys/vmmeter.h> #include <sys/sysctl.h> @@ -30,7 +31,6 @@ #include <machine/cpu.h> #include <machine/md_var.h> -#include <machine/smp.h> #include <machine/globaldata.h> #include <machine/globals.h> @@ -48,26 +48,32 @@ static void idle_proc(void *dummy); static void idle_setup(void *dummy) { +#ifdef SMP struct globaldata *gd; +#endif + struct proc *p; int error; - SLIST_FOREACH(gd, &cpuhead, gd_allcpu) { #ifdef SMP - error = kthread_create(idle_proc, NULL, &gd->gd_idleproc, - RFSTOPPED|RFHIGHPID, "idle: cpu%d", - gd->gd_cpuid); + SLIST_FOREACH(gd, &cpuhead, gd_allcpu) { + error = kthread_create(idle_proc, NULL, &p, + RFSTOPPED | RFHIGHPID, "idle: cpu%d", gd->gd_cpuid); + gd->gd_idleproc = p; + if (gd->gd_curproc == NULL) + gd->gd_curproc = p; #else - error = kthread_create(idle_proc, NULL, &gd->gd_idleproc, - RFSTOPPED|RFHIGHPID, "idle"); + error = kthread_create(idle_proc, NULL, &p, + RFSTOPPED | RFHIGHPID, "idle"); + PCPU_SET(idleproc, p); #endif if (error) panic("idle_setup: kthread_create error %d\n", error); - gd->gd_idleproc->p_flag |= P_NOLOAD; - gd->gd_idleproc->p_stat = SRUN; - if (gd->gd_curproc == NULL) - gd->gd_curproc = gd->gd_idleproc; + p->p_flag |= P_NOLOAD; + p->p_stat = SRUN; +#ifdef SMP } +#endif } /* diff --git a/sys/kern/kern_intr.c b/sys/kern/kern_intr.c index aa4d84a..70848f6 100644 --- a/sys/kern/kern_intr.c +++ b/sys/kern/kern_intr.c @@ -381,7 +381,7 @@ ithread_schedule(struct ithd *ithread, int do_switch) mi_switch(); sched_lock.mtx_savecrit = savecrit; } else - need_resched(); + need_resched(curproc); } else { CTR3(KTR_INTR, __func__ ": pid %d: it_need %d, state %d", p->p_pid, ithread->it_need, p->p_stat); diff --git a/sys/kern/kern_mib.c b/sys/kern/kern_mib.c index bb9f640..7d8ad0f 100644 --- a/sys/kern/kern_mib.c +++ b/sys/kern/kern_mib.c @@ -46,7 +46,7 @@ #include <sys/sysctl.h> #include <sys/proc.h> #include <sys/jail.h> -#include <machine/smp.h> +#include <sys/smp.h> SYSCTL_NODE(, 0, sysctl, CTLFLAG_RW, 0, "Sysctl internal magic"); diff --git a/sys/kern/kern_shutdown.c b/sys/kern/kern_shutdown.c index 8463991..ffe40c4 100644 --- a/sys/kern/kern_shutdown.c +++ b/sys/kern/kern_shutdown.c @@ -59,6 +59,7 @@ #include <sys/mount.h> #include <sys/mutex.h> #include <sys/queue.h> +#include <sys/smp.h> /* smp_active, cpuid */ #include <sys/sysctl.h> #include <sys/conf.h> #include <sys/sysproto.h> @@ -66,7 +67,6 @@ #include <machine/pcb.h> #include <machine/md_var.h> -#include <machine/smp.h> /* smp_active, cpuid */ #include <sys/signalvar.h> diff --git a/sys/kern/kern_sig.c b/sys/kern/kern_sig.c index b387237..c57074a 100644 --- a/sys/kern/kern_sig.c +++ b/sys/kern/kern_sig.c @@ -62,6 +62,7 @@ #include <sys/ktr.h> #include <sys/ktrace.h> #include <sys/resourcevar.h> +#include <sys/smp.h> #include <sys/stat.h> #include <sys/sx.h> #include <sys/syslog.h> @@ -70,7 +71,6 @@ #include <sys/malloc.h> #include <machine/cpu.h> -#include <machine/smp.h> #define ONSIG 32 /* NSIG for osig* syscalls. XXX. */ @@ -1301,12 +1301,11 @@ psignal(p, sig) */ if (p->p_stat == SRUN) { signotify(p); - mtx_unlock_spin(&sched_lock); #ifdef SMP forward_signal(p); #endif - } else - mtx_unlock_spin(&sched_lock); + } + mtx_unlock_spin(&sched_lock); goto out; } /*NOTREACHED*/ diff --git a/sys/kern/kern_synch.c b/sys/kern/kern_synch.c index 13bb1404d..dc8d25a 100644 --- a/sys/kern/kern_synch.c +++ b/sys/kern/kern_synch.c @@ -52,6 +52,7 @@ #include <sys/proc.h> #include <sys/resourcevar.h> #include <sys/signalvar.h> +#include <sys/smp.h> #include <sys/sx.h> #include <sys/sysctl.h> #include <sys/sysproto.h> @@ -64,7 +65,6 @@ #endif #include <machine/cpu.h> -#include <machine/smp.h> static void sched_setup __P((void *dummy)); SYSINIT(sched_setup, SI_SUB_KICK_SCHEDULER, SI_ORDER_FIRST, sched_setup, NULL) @@ -110,7 +110,7 @@ maybe_resched(p) mtx_assert(&sched_lock, MA_OWNED); if (p->p_pri.pri_level < curproc->p_pri.pri_level) - need_resched(); + need_resched(curproc); } int @@ -129,11 +129,11 @@ roundrobin(arg) { mtx_lock_spin(&sched_lock); - need_resched(); - mtx_unlock_spin(&sched_lock); + need_resched(curproc); #ifdef SMP forward_roundrobin(); #endif + mtx_unlock_spin(&sched_lock); callout_reset(&roundrobin_callout, sched_quantum, roundrobin, NULL); } @@ -911,7 +911,7 @@ mi_switch() sched_nest = sched_lock.mtx_recurse; curproc->p_lastcpu = curproc->p_oncpu; curproc->p_oncpu = NOCPU; - clear_resched(); + clear_resched(curproc); cpu_switch(); curproc->p_oncpu = PCPU_GET(cpuid); sched_lock.mtx_recurse = sched_nest; diff --git a/sys/kern/ksched.c b/sys/kern/ksched.c index b729ccf..dbdc174 100644 --- a/sys/kern/ksched.c +++ b/sys/kern/ksched.c @@ -175,7 +175,7 @@ int ksched_setscheduler(register_t *ret, struct ksched *ksched, mtx_lock_spin(&sched_lock); rtp_to_pri(&rtp, &p->p_pri); - need_resched(); + need_resched(p); mtx_unlock_spin(&sched_lock); } else @@ -197,7 +197,7 @@ int ksched_setscheduler(register_t *ret, struct ksched *ksched, * on the scheduling code: You must leave the * scheduling info alone. */ - need_resched(); + need_resched(p); mtx_unlock_spin(&sched_lock); } break; @@ -216,7 +216,7 @@ int ksched_getscheduler(register_t *ret, struct ksched *ksched, struct proc *p) int ksched_yield(register_t *ret, struct ksched *ksched) { mtx_lock_spin(&sched_lock); - need_resched(); + need_resched(curproc); mtx_unlock_spin(&sched_lock); return 0; } diff --git a/sys/kern/subr_smp.c b/sys/kern/subr_smp.c index 28a5c72..6c73e47 100644 --- a/sys/kern/subr_smp.c +++ b/sys/kern/subr_smp.c @@ -1,2719 +1,199 @@ /* - * Copyright (c) 1996, by Steve Passe - * All rights reserved. + * Copyright (c) 2001 + * John Baldwin <jhb@FreeBSD.org>. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. - * 2. The name of the developer may NOT be used to endorse or promote products - * derived from this software without specific prior written permission. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 4. Neither the name of the author nor the names of any co-contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. * - * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND + * THIS SOFTWARE IS PROVIDED BY JOHN BALDWIN AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE - * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE - * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL - * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS - * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) - * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT - * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY - * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF - * SUCH DAMAGE. + * ARE DISCLAIMED. IN NO EVENT SHALL JOHN BALDWIN OR THE VOICES IN HIS HEAD + * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF + * THE POSSIBILITY OF SUCH DAMAGE. * * $FreeBSD$ */ -#include "opt_cpu.h" - -#ifdef SMP -#include <machine/smptests.h> -#else -#error -#endif +/* + * This module holds the global variables and machine independent functions + * used for the kernel SMP support. It also provides MI support for per-cpu + * data. + */ #include <sys/param.h> -#include <sys/bus.h> #include <sys/systm.h> -#include <sys/kernel.h> +#include <sys/ipl.h> +#include <sys/ktr.h> #include <sys/proc.h> -#include <sys/sysctl.h> +#include <sys/lock.h> #include <sys/malloc.h> -#include <sys/memrange.h> #include <sys/mutex.h> -#include <sys/dkstat.h> -#include <sys/cons.h> /* cngetc() */ +#include <sys/kernel.h> +#include <sys/smp.h> +#include <sys/sysctl.h> #include <vm/vm.h> -#include <vm/vm_param.h> #include <vm/pmap.h> -#include <vm/vm_kern.h> -#include <vm/vm_extern.h> -#include <sys/lock.h> #include <vm/vm_map.h> #include <sys/user.h> -#ifdef GPROF -#include <sys/gmon.h> -#endif +#include <sys/dkstat.h> -#include <machine/smp.h> -#include <machine/apic.h> #include <machine/atomic.h> -#include <machine/cpufunc.h> -#include <machine/ipl.h> -#include <machine/mpapic.h> -#include <machine/psl.h> -#include <machine/segments.h> -#include <machine/smptests.h> /** TEST_DEFAULT_CONFIG, TEST_TEST1 */ -#include <machine/tss.h> -#include <machine/specialreg.h> #include <machine/globaldata.h> +#include <machine/pmap.h> +#include <machine/clock.h> -#if defined(APIC_IO) -#include <machine/md_var.h> /* setidt() */ -#include <i386/isa/icu.h> /* IPIs */ -#include <i386/isa/intr_machdep.h> /* IPIs */ -#endif /* APIC_IO */ - -#if defined(TEST_DEFAULT_CONFIG) -#define MPFPS_MPFB1 TEST_DEFAULT_CONFIG -#else -#define MPFPS_MPFB1 mpfps->mpfb1 -#endif /* TEST_DEFAULT_CONFIG */ - -#define WARMBOOT_TARGET 0 -#define WARMBOOT_OFF (KERNBASE + 0x0467) -#define WARMBOOT_SEG (KERNBASE + 0x0469) - -#ifdef PC98 -#define BIOS_BASE (0xe8000) -#define BIOS_SIZE (0x18000) -#else -#define BIOS_BASE (0xf0000) -#define BIOS_SIZE (0x10000) -#endif -#define BIOS_COUNT (BIOS_SIZE/4) - -#define CMOS_REG (0x70) -#define CMOS_DATA (0x71) -#define BIOS_RESET (0x0f) -#define BIOS_WARM (0x0a) - -#define PROCENTRY_FLAG_EN 0x01 -#define PROCENTRY_FLAG_BP 0x02 -#define IOAPICENTRY_FLAG_EN 0x01 - - -/* MP Floating Pointer Structure */ -typedef struct MPFPS { - char signature[4]; - void *pap; - u_char length; - u_char spec_rev; - u_char checksum; - u_char mpfb1; - u_char mpfb2; - u_char mpfb3; - u_char mpfb4; - u_char mpfb5; -} *mpfps_t; - -/* MP Configuration Table Header */ -typedef struct MPCTH { - char signature[4]; - u_short base_table_length; - u_char spec_rev; - u_char checksum; - u_char oem_id[8]; - u_char product_id[12]; - void *oem_table_pointer; - u_short oem_table_size; - u_short entry_count; - void *apic_address; - u_short extended_table_length; - u_char extended_table_checksum; - u_char reserved; -} *mpcth_t; - - -typedef struct PROCENTRY { - u_char type; - u_char apic_id; - u_char apic_version; - u_char cpu_flags; - u_long cpu_signature; - u_long feature_flags; - u_long reserved1; - u_long reserved2; -} *proc_entry_ptr; - -typedef struct BUSENTRY { - u_char type; - u_char bus_id; - char bus_type[6]; -} *bus_entry_ptr; - -typedef struct IOAPICENTRY { - u_char type; - u_char apic_id; - u_char apic_version; - u_char apic_flags; - void *apic_address; -} *io_apic_entry_ptr; - -typedef struct INTENTRY { - u_char type; - u_char int_type; - u_short int_flags; - u_char src_bus_id; - u_char src_bus_irq; - u_char dst_apic_id; - u_char dst_apic_int; -} *int_entry_ptr; - -/* descriptions of MP basetable entries */ -typedef struct BASETABLE_ENTRY { - u_char type; - u_char length; - char name[16]; -} basetable_entry; - -/* - * this code MUST be enabled here and in mpboot.s. - * it follows the very early stages of AP boot by placing values in CMOS ram. - * it NORMALLY will never be needed and thus the primitive method for enabling. - * -#define CHECK_POINTS - */ - -#if defined(CHECK_POINTS) && !defined(PC98) -#define CHECK_READ(A) (outb(CMOS_REG, (A)), inb(CMOS_DATA)) -#define CHECK_WRITE(A,D) (outb(CMOS_REG, (A)), outb(CMOS_DATA, (D))) - -#define CHECK_INIT(D); \ - CHECK_WRITE(0x34, (D)); \ - CHECK_WRITE(0x35, (D)); \ - CHECK_WRITE(0x36, (D)); \ - CHECK_WRITE(0x37, (D)); \ - CHECK_WRITE(0x38, (D)); \ - CHECK_WRITE(0x39, (D)); - -#define CHECK_PRINT(S); \ - printf("%s: %d, %d, %d, %d, %d, %d\n", \ - (S), \ - CHECK_READ(0x34), \ - CHECK_READ(0x35), \ - CHECK_READ(0x36), \ - CHECK_READ(0x37), \ - CHECK_READ(0x38), \ - CHECK_READ(0x39)); - -#else /* CHECK_POINTS */ - -#define CHECK_INIT(D) -#define CHECK_PRINT(S) - -#endif /* CHECK_POINTS */ - -/* - * Values to send to the POST hardware. - */ -#define MP_BOOTADDRESS_POST 0x10 -#define MP_PROBE_POST 0x11 -#define MPTABLE_PASS1_POST 0x12 - -#define MP_START_POST 0x13 -#define MP_ENABLE_POST 0x14 -#define MPTABLE_PASS2_POST 0x15 - -#define START_ALL_APS_POST 0x16 -#define INSTALL_AP_TRAMP_POST 0x17 -#define START_AP_POST 0x18 - -#define MP_ANNOUNCE_POST 0x19 - -/* used to hold the AP's until we are ready to release them */ -static struct mtx ap_boot_mtx; - -/** XXX FIXME: where does this really belong, isa.h/isa.c perhaps? */ -int current_postcode; - -/** XXX FIXME: what system files declare these??? */ -extern struct region_descriptor r_gdt, r_idt; - -int bsp_apic_ready = 0; /* flags useability of BSP apic */ -int mp_ncpus; /* # of CPUs, including BSP */ -int mp_naps; /* # of Applications processors */ -int mp_nbusses; /* # of busses */ -int mp_napics; /* # of IO APICs */ -int boot_cpu_id; /* designated BSP */ -vm_offset_t cpu_apic_address; -vm_offset_t io_apic_address[NAPICID]; /* NAPICID is more than enough */ -extern int nkpt; - -u_int32_t cpu_apic_versions[MAXCPU]; -u_int32_t *io_apic_versions; - -#ifdef APIC_INTR_REORDER -struct { - volatile int *location; - int bit; -} apic_isrbit_location[32]; -#endif - -struct apic_intmapinfo int_to_apicintpin[APIC_INTMAPSIZE]; - -/* - * APIC ID logical/physical mapping structures. - * We oversize these to simplify boot-time config. - */ -int cpu_num_to_apic_id[NAPICID]; -int io_num_to_apic_id[NAPICID]; -int apic_id_to_logical[NAPICID]; - +volatile u_int stopped_cpus; +volatile u_int started_cpus; -/* Bitmap of all available CPUs */ -u_int all_cpus; +void (*cpustop_restartfunc) __P((void)); +int mp_ncpus; -/* AP uses this during bootstrap. Do not staticize. */ -char *bootSTK; -static int bootAP; +volatile int smp_started; +u_int all_cpus; -/* Hotwire a 0->4MB V==P mapping */ -extern pt_entry_t *KPTphys; +static struct globaldata *cpuid_to_globaldata[MAXCPU]; +struct cpuhead cpuhead = SLIST_HEAD_INITIALIZER(cpuhead); -/* SMP page table page */ -extern pt_entry_t *SMPpt; +SYSCTL_NODE(_kern, OID_AUTO, smp, CTLFLAG_RD, NULL, "Kernel SMP"); -struct pcb stoppcbs[MAXCPU]; +int smp_active = 0; /* are the APs allowed to run? */ +SYSCTL_INT(_kern_smp, OID_AUTO, active, CTLFLAG_RW, &smp_active, 0, ""); -int smp_started; /* has the system started? */ -int smp_active = 0; /* are the APs allowed to run? */ -SYSCTL_INT(_machdep, OID_AUTO, smp_active, CTLFLAG_RW, &smp_active, 0, ""); - -/* XXX maybe should be hw.ncpu */ -static int smp_cpus = 1; /* how many cpu's running */ -SYSCTL_INT(_machdep, OID_AUTO, smp_cpus, CTLFLAG_RD, &smp_cpus, 0, ""); - -int invltlb_ok = 0; /* throttle smp_invltlb() till safe */ -SYSCTL_INT(_machdep, OID_AUTO, invltlb_ok, CTLFLAG_RW, &invltlb_ok, 0, ""); +int smp_cpus = 1; /* how many cpu's running */ +SYSCTL_INT(_kern_smp, OID_AUTO, cpus, CTLFLAG_RD, &smp_cpus, 0, ""); /* Enable forwarding of a signal to a process running on a different CPU */ static int forward_signal_enabled = 1; -SYSCTL_INT(_machdep, OID_AUTO, forward_signal_enabled, CTLFLAG_RW, +SYSCTL_INT(_kern_smp, OID_AUTO, forward_signal_enabled, CTLFLAG_RW, &forward_signal_enabled, 0, ""); /* Enable forwarding of roundrobin to all other cpus */ static int forward_roundrobin_enabled = 1; -SYSCTL_INT(_machdep, OID_AUTO, forward_roundrobin_enabled, CTLFLAG_RW, +SYSCTL_INT(_kern_smp, OID_AUTO, forward_roundrobin_enabled, CTLFLAG_RW, &forward_roundrobin_enabled, 0, ""); +/* Variables needed for SMP rendezvous. */ +static void (*smp_rv_setup_func)(void *arg); +static void (*smp_rv_action_func)(void *arg); +static void (*smp_rv_teardown_func)(void *arg); +static void *smp_rv_func_arg; +static volatile int smp_rv_waiters[2]; +static struct mtx smp_rv_mtx; /* - * Local data and functions. - */ - -/* Set to 1 once we're ready to let the APs out of the pen. */ -static volatile int aps_ready = 0; - -static int mp_capable; -static u_int boot_address; -static u_int base_memory; - -static int picmode; /* 0: virtual wire mode, 1: PIC mode */ -static mpfps_t mpfps; -static int search_for_sig(u_int32_t target, int count); -static void mp_enable(u_int boot_addr); - -static void mptable_pass1(void); -static int mptable_pass2(void); -static void default_mp_table(int type); -static void fix_mp_table(void); -static void setup_apic_irq_mapping(void); -static void init_locks(void); -static int start_all_aps(u_int boot_addr); -static void install_ap_tramp(u_int boot_addr); -static int start_ap(int logicalCpu, u_int boot_addr); -void ap_init(void); -static int apic_int_is_bus_type(int intr, int bus_type); -static void release_aps(void *dummy); - -/* - * initialize all the SMP locks + * Initialize MI SMP variables and call the MD SMP initialization code. */ - -/* critical region around IO APIC, apic_imen */ -struct mtx imen_mtx; - -/* lock region used by kernel profiling */ -struct mtx mcount_mtx; - -#ifdef USE_COMLOCK -/* locks com (tty) data/hardware accesses: a FASTINTR() */ -struct mtx com_mtx; -#endif /* USE_COMLOCK */ - -/* lock around the MP rendezvous */ -static struct mtx smp_rv_mtx; - static void -init_locks(void) -{ - /* - * XXX The mcount mutex probably needs to be statically initialized, - * since it will be used even in the function calls that get us to this - * point. - */ - mtx_init(&mcount_mtx, "mcount", MTX_DEF); - - mtx_init(&smp_rv_mtx, "smp rendezvous", MTX_SPIN); - -#ifdef USE_COMLOCK - mtx_init(&com_mtx, "com", MTX_SPIN); -#endif /* USE_COMLOCK */ - - mtx_init(&ap_boot_mtx, "ap boot", MTX_SPIN); -} - -/* - * Calculate usable address in base memory for AP trampoline code. - */ -u_int -mp_bootaddress(u_int basemem) -{ - POSTCODE(MP_BOOTADDRESS_POST); - - base_memory = basemem * 1024; /* convert to bytes */ - - boot_address = base_memory & ~0xfff; /* round down to 4k boundary */ - if ((base_memory - boot_address) < bootMP_size) - boot_address -= 4096; /* not enough, lower by 4k */ - - return boot_address; -} - - -/* - * Look for an Intel MP spec table (ie, SMP capable hardware). - */ -int -mp_probe(void) -{ - int x; - u_long segment; - u_int32_t target; - - POSTCODE(MP_PROBE_POST); - - /* see if EBDA exists */ - if ((segment = (u_long) * (u_short *) (KERNBASE + 0x40e)) != 0) { - /* search first 1K of EBDA */ - target = (u_int32_t) (segment << 4); - if ((x = search_for_sig(target, 1024 / 4)) >= 0) - goto found; - } else { - /* last 1K of base memory, effective 'top of base' passed in */ - target = (u_int32_t) (base_memory - 0x400); - if ((x = search_for_sig(target, 1024 / 4)) >= 0) - goto found; - } - - /* search the BIOS */ - target = (u_int32_t) BIOS_BASE; - if ((x = search_for_sig(target, BIOS_COUNT)) >= 0) - goto found; - - /* nothing found */ - mpfps = (mpfps_t)0; - mp_capable = 0; - return 0; - -found: - /* calculate needed resources */ - mpfps = (mpfps_t)x; - mptable_pass1(); - - /* flag fact that we are running multiple processors */ - mp_capable = 1; - return 1; -} - - -/* - * Initialize the SMP hardware and the APIC and start up the AP's. - */ -void -mp_start(void) +mp_start(void *dummy) { - POSTCODE(MP_START_POST); - /* look for MP capable motherboard */ - if (mp_capable) - mp_enable(boot_address); - else - panic("MP hardware not found!"); -} - - -/* - * Print various information about the SMP system hardware and setup. - */ -void -mp_announce(void) -{ - int x; - - POSTCODE(MP_ANNOUNCE_POST); - - printf("FreeBSD/SMP: Multiprocessor motherboard\n"); - printf(" cpu0 (BSP): apic id: %2d", CPU_TO_ID(0)); - printf(", version: 0x%08x", cpu_apic_versions[0]); - printf(", at 0x%08x\n", cpu_apic_address); - for (x = 1; x <= mp_naps; ++x) { - printf(" cpu%d (AP): apic id: %2d", x, CPU_TO_ID(x)); - printf(", version: 0x%08x", cpu_apic_versions[x]); - printf(", at 0x%08x\n", cpu_apic_address); - } - -#if defined(APIC_IO) - for (x = 0; x < mp_napics; ++x) { - printf(" io%d (APIC): apic id: %2d", x, IO_TO_ID(x)); - printf(", version: 0x%08x", io_apic_versions[x]); - printf(", at 0x%08x\n", io_apic_address[x]); - } -#else - printf(" Warning: APIC I/O disabled\n"); -#endif /* APIC_IO */ -} - -/* - * AP cpu's call this to sync up protected mode. - */ -void -init_secondary(void) -{ - int gsel_tss; - int x, myid = bootAP; - - gdt_segs[GPRIV_SEL].ssd_base = (int) &SMP_prvspace[myid]; - gdt_segs[GPROC0_SEL].ssd_base = - (int) &SMP_prvspace[myid].globaldata.gd_common_tss; - SMP_prvspace[myid].globaldata.gd_prvspace = - &SMP_prvspace[myid].globaldata; - - for (x = 0; x < NGDT; x++) { - ssdtosd(&gdt_segs[x], &gdt[myid * NGDT + x].sd); - } - - r_gdt.rd_limit = NGDT * sizeof(gdt[0]) - 1; - r_gdt.rd_base = (int) &gdt[myid * NGDT]; - lgdt(&r_gdt); /* does magic intra-segment return */ - - lidt(&r_idt); - - lldt(_default_ldt); - PCPU_SET(currentldt, _default_ldt); - - gsel_tss = GSEL(GPROC0_SEL, SEL_KPL); - gdt[myid * NGDT + GPROC0_SEL].sd.sd_type = SDT_SYS386TSS; - PCPU_SET(common_tss.tss_esp0, 0); /* not used until after switch */ - PCPU_SET(common_tss.tss_ss0, GSEL(GDATA_SEL, SEL_KPL)); - PCPU_SET(common_tss.tss_ioopt, (sizeof (struct i386tss)) << 16); - PCPU_SET(tss_gdt, &gdt[myid * NGDT + GPROC0_SEL].sd); - PCPU_SET(common_tssd, *PCPU_GET(tss_gdt)); - ltr(gsel_tss); + /* Probe for MP hardware. */ + if (cpu_mp_probe() == 0) + return; - pmap_set_opt(); + mtx_init(&smp_rv_mtx, "smp rendezvous", MTX_SPIN); + cpu_mp_start(); + printf("FreeBSD/SMP: Multiprocessor System Detected: %d CPUs\n", + mp_ncpus); + cpu_mp_announce(); } +SYSINIT(cpu_mp, SI_SUB_CPU, SI_ORDER_SECOND, mp_start, NULL) - -#if defined(APIC_IO) /* - * Final configuration of the BSP's local APIC: - * - disable 'pic mode'. - * - disable 'virtual wire mode'. - * - enable NMI. + * Register a struct globaldata. */ void -bsp_apic_configure(void) +globaldata_register(struct globaldata *globaldata) { - u_char byte; - u_int32_t temp; - - /* leave 'pic mode' if necessary */ - if (picmode) { - outb(0x22, 0x70); /* select IMCR */ - byte = inb(0x23); /* current contents */ - byte |= 0x01; /* mask external INTR */ - outb(0x23, byte); /* disconnect 8259s/NMI */ - } - - /* mask lint0 (the 8259 'virtual wire' connection) */ - temp = lapic.lvt_lint0; - temp |= APIC_LVT_M; /* set the mask */ - lapic.lvt_lint0 = temp; - - /* setup lint1 to handle NMI */ - temp = lapic.lvt_lint1; - temp &= ~APIC_LVT_M; /* clear the mask */ - lapic.lvt_lint1 = temp; - - if (bootverbose) - apic_dump("bsp_apic_configure()"); -} -#endif /* APIC_IO */ - - -/******************************************************************* - * local functions and data - */ - -/* - * start the SMP system - */ -static void -mp_enable(u_int boot_addr) -{ - int x; -#if defined(APIC_IO) - int apic; - u_int ux; -#endif /* APIC_IO */ - - POSTCODE(MP_ENABLE_POST); - - /* turn on 4MB of V == P addressing so we can get to MP table */ - *(int *)PTD = PG_V | PG_RW | ((uintptr_t)(void *)KPTphys & PG_FRAME); - invltlb(); - - /* examine the MP table for needed info, uses physical addresses */ - x = mptable_pass2(); - - *(int *)PTD = 0; - invltlb(); - - /* can't process default configs till the CPU APIC is pmapped */ - if (x) - default_mp_table(x); - - /* post scan cleanup */ - fix_mp_table(); - setup_apic_irq_mapping(); - -#if defined(APIC_IO) - - /* fill the LOGICAL io_apic_versions table */ - for (apic = 0; apic < mp_napics; ++apic) { - ux = io_apic_read(apic, IOAPIC_VER); - io_apic_versions[apic] = ux; - io_apic_set_id(apic, IO_TO_ID(apic)); - } - - /* program each IO APIC in the system */ - for (apic = 0; apic < mp_napics; ++apic) - if (io_apic_setup(apic) < 0) - panic("IO APIC setup failure"); - - /* install a 'Spurious INTerrupt' vector */ - setidt(XSPURIOUSINT_OFFSET, Xspuriousint, - SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); - - /* install an inter-CPU IPI for TLB invalidation */ - setidt(XINVLTLB_OFFSET, Xinvltlb, - SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); - - /* install an inter-CPU IPI for reading processor state */ - setidt(XCPUCHECKSTATE_OFFSET, Xcpucheckstate, - SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); - - /* install an inter-CPU IPI for all-CPU rendezvous */ - setidt(XRENDEZVOUS_OFFSET, Xrendezvous, - SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); - - /* install an inter-CPU IPI for forcing an additional software trap */ - setidt(XCPUAST_OFFSET, Xcpuast, - SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); - - /* install an inter-CPU IPI for CPU stop/restart */ - setidt(XCPUSTOP_OFFSET, Xcpustop, - SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); - -#if defined(TEST_TEST1) - /* install a "fake hardware INTerrupt" vector */ - setidt(XTEST1_OFFSET, Xtest1, - SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL)); -#endif /** TEST_TEST1 */ - -#endif /* APIC_IO */ - - /* initialize all SMP locks */ - init_locks(); - - /* start each Application Processor */ - start_all_aps(boot_addr); -} - - -/* - * look for the MP spec signature - */ - -/* string defined by the Intel MP Spec as identifying the MP table */ -#define MP_SIG 0x5f504d5f /* _MP_ */ -#define NEXT(X) ((X) += 4) -static int -search_for_sig(u_int32_t target, int count) -{ - int x; - u_int32_t *addr = (u_int32_t *) (KERNBASE + target); - - for (x = 0; x < count; NEXT(x)) - if (addr[x] == MP_SIG) - /* make array index a byte index */ - return (target + (x * sizeof(u_int32_t))); - - return -1; -} - - -static basetable_entry basetable_entry_types[] = -{ - {0, 20, "Processor"}, - {1, 8, "Bus"}, - {2, 8, "I/O APIC"}, - {3, 8, "I/O INT"}, - {4, 8, "Local INT"} -}; - -typedef struct BUSDATA { - u_char bus_id; - enum busTypes bus_type; -} bus_datum; - -typedef struct INTDATA { - u_char int_type; - u_short int_flags; - u_char src_bus_id; - u_char src_bus_irq; - u_char dst_apic_id; - u_char dst_apic_int; - u_char int_vector; -} io_int, local_int; - -typedef struct BUSTYPENAME { - u_char type; - char name[7]; -} bus_type_name; - -static bus_type_name bus_type_table[] = -{ - {CBUS, "CBUS"}, - {CBUSII, "CBUSII"}, - {EISA, "EISA"}, - {MCA, "MCA"}, - {UNKNOWN_BUSTYPE, "---"}, - {ISA, "ISA"}, - {MCA, "MCA"}, - {UNKNOWN_BUSTYPE, "---"}, - {UNKNOWN_BUSTYPE, "---"}, - {UNKNOWN_BUSTYPE, "---"}, - {UNKNOWN_BUSTYPE, "---"}, - {UNKNOWN_BUSTYPE, "---"}, - {PCI, "PCI"}, - {UNKNOWN_BUSTYPE, "---"}, - {UNKNOWN_BUSTYPE, "---"}, - {UNKNOWN_BUSTYPE, "---"}, - {UNKNOWN_BUSTYPE, "---"}, - {XPRESS, "XPRESS"}, - {UNKNOWN_BUSTYPE, "---"} -}; -/* from MP spec v1.4, table 5-1 */ -static int default_data[7][5] = -{ -/* nbus, id0, type0, id1, type1 */ - {1, 0, ISA, 255, 255}, - {1, 0, EISA, 255, 255}, - {1, 0, EISA, 255, 255}, - {1, 0, MCA, 255, 255}, - {2, 0, ISA, 1, PCI}, - {2, 0, EISA, 1, PCI}, - {2, 0, MCA, 1, PCI} -}; - - -/* the bus data */ -static bus_datum *bus_data; - -/* the IO INT data, one entry per possible APIC INTerrupt */ -static io_int *io_apic_ints; - -static int nintrs; - -static int processor_entry __P((proc_entry_ptr entry, int cpu)); -static int bus_entry __P((bus_entry_ptr entry, int bus)); -static int io_apic_entry __P((io_apic_entry_ptr entry, int apic)); -static int int_entry __P((int_entry_ptr entry, int intr)); -static int lookup_bus_type __P((char *name)); - - -/* - * 1st pass on motherboard's Intel MP specification table. - * - * initializes: - * mp_ncpus = 1 - * - * determines: - * cpu_apic_address (common to all CPUs) - * io_apic_address[N] - * mp_naps - * mp_nbusses - * mp_napics - * nintrs - */ -static void -mptable_pass1(void) -{ - int x; - mpcth_t cth; - int totalSize; - void* position; - int count; - int type; - - POSTCODE(MPTABLE_PASS1_POST); - - /* clear various tables */ - for (x = 0; x < NAPICID; ++x) { - io_apic_address[x] = ~0; /* IO APIC address table */ - } - - /* init everything to empty */ - mp_naps = 0; - mp_nbusses = 0; - mp_napics = 0; - nintrs = 0; - - /* check for use of 'default' configuration */ - if (MPFPS_MPFB1 != 0) { - /* use default addresses */ - cpu_apic_address = DEFAULT_APIC_BASE; - io_apic_address[0] = DEFAULT_IO_APIC_BASE; - - /* fill in with defaults */ - mp_naps = 2; /* includes BSP */ - mp_nbusses = default_data[MPFPS_MPFB1 - 1][0]; -#if defined(APIC_IO) - mp_napics = 1; - nintrs = 16; -#endif /* APIC_IO */ - } - else { - if ((cth = mpfps->pap) == 0) - panic("MP Configuration Table Header MISSING!"); - - cpu_apic_address = (vm_offset_t) cth->apic_address; - - /* walk the table, recording info of interest */ - totalSize = cth->base_table_length - sizeof(struct MPCTH); - position = (u_char *) cth + sizeof(struct MPCTH); - count = cth->entry_count; - - while (count--) { - switch (type = *(u_char *) position) { - case 0: /* processor_entry */ - if (((proc_entry_ptr)position)->cpu_flags - & PROCENTRY_FLAG_EN) - ++mp_naps; - break; - case 1: /* bus_entry */ - ++mp_nbusses; - break; - case 2: /* io_apic_entry */ - if (((io_apic_entry_ptr)position)->apic_flags - & IOAPICENTRY_FLAG_EN) - io_apic_address[mp_napics++] = - (vm_offset_t)((io_apic_entry_ptr) - position)->apic_address; - break; - case 3: /* int_entry */ - ++nintrs; - break; - case 4: /* int_entry */ - break; - default: - panic("mpfps Base Table HOSED!"); - /* NOTREACHED */ - } - - totalSize -= basetable_entry_types[type].length; - (u_char*)position += basetable_entry_types[type].length; - } - } - - /* qualify the numbers */ - if (mp_naps > MAXCPU) { - printf("Warning: only using %d of %d available CPUs!\n", - MAXCPU, mp_naps); - mp_naps = MAXCPU; - } - - /* - * Count the BSP. - * This is also used as a counter while starting the APs. - */ - mp_ncpus = 1; - --mp_naps; /* subtract the BSP */ + KASSERT(globaldata->gd_cpuid >= 0 && globaldata->gd_cpuid < MAXCPU, + ("globaldata_register: invalid cpuid")); + cpuid_to_globaldata[globaldata->gd_cpuid] = globaldata; + SLIST_INSERT_HEAD(&cpuhead, globaldata, gd_allcpu); } - /* - * 2nd pass on motherboard's Intel MP specification table. - * - * sets: - * boot_cpu_id - * ID_TO_IO(N), phy APIC ID to log CPU/IO table - * CPU_TO_ID(N), logical CPU to APIC ID table - * IO_TO_ID(N), logical IO to APIC ID table - * bus_data[N] - * io_apic_ints[N] + * Locate a struct globaldata by cpu id. */ -static int -mptable_pass2(void) +struct globaldata * +globaldata_find(u_int cpuid) { - int x; - mpcth_t cth; - int totalSize; - void* position; - int count; - int type; - int apic, bus, cpu, intr; - int i, j; - int pgeflag; - - POSTCODE(MPTABLE_PASS2_POST); - - pgeflag = 0; /* XXX - Not used under SMP yet. */ - - MALLOC(io_apic_versions, u_int32_t *, sizeof(u_int32_t) * mp_napics, - M_DEVBUF, M_WAITOK); - MALLOC(ioapic, volatile ioapic_t **, sizeof(ioapic_t *) * mp_napics, - M_DEVBUF, M_WAITOK); - MALLOC(io_apic_ints, io_int *, sizeof(io_int) * (nintrs + 1), - M_DEVBUF, M_WAITOK); - MALLOC(bus_data, bus_datum *, sizeof(bus_datum) * mp_nbusses, - M_DEVBUF, M_WAITOK); - - bzero(ioapic, sizeof(ioapic_t *) * mp_napics); - - for (i = 0; i < mp_napics; i++) { - for (j = 0; j < mp_napics; j++) { - /* same page frame as a previous IO apic? */ - if (((vm_offset_t)SMPpt[NPTEPG-2-j] & PG_FRAME) == - (io_apic_address[i] & PG_FRAME)) { - ioapic[i] = (ioapic_t *)((u_int)SMP_prvspace - + (NPTEPG-2-j) * PAGE_SIZE - + (io_apic_address[i] & PAGE_MASK)); - break; - } - /* use this slot if available */ - if (((vm_offset_t)SMPpt[NPTEPG-2-j] & PG_FRAME) == 0) { - SMPpt[NPTEPG-2-j] = (pt_entry_t)(PG_V | PG_RW | - pgeflag | (io_apic_address[i] & PG_FRAME)); - ioapic[i] = (ioapic_t *)((u_int)SMP_prvspace - + (NPTEPG-2-j) * PAGE_SIZE - + (io_apic_address[i] & PAGE_MASK)); - break; - } - } - } - - /* clear various tables */ - for (x = 0; x < NAPICID; ++x) { - ID_TO_IO(x) = -1; /* phy APIC ID to log CPU/IO table */ - CPU_TO_ID(x) = -1; /* logical CPU to APIC ID table */ - IO_TO_ID(x) = -1; /* logical IO to APIC ID table */ - } - - /* clear bus data table */ - for (x = 0; x < mp_nbusses; ++x) - bus_data[x].bus_id = 0xff; - - /* clear IO APIC INT table */ - for (x = 0; x < (nintrs + 1); ++x) { - io_apic_ints[x].int_type = 0xff; - io_apic_ints[x].int_vector = 0xff; - } - - /* setup the cpu/apic mapping arrays */ - boot_cpu_id = -1; - - /* record whether PIC or virtual-wire mode */ - picmode = (mpfps->mpfb2 & 0x80) ? 1 : 0; - - /* check for use of 'default' configuration */ - if (MPFPS_MPFB1 != 0) - return MPFPS_MPFB1; /* return default configuration type */ - - if ((cth = mpfps->pap) == 0) - panic("MP Configuration Table Header MISSING!"); - - /* walk the table, recording info of interest */ - totalSize = cth->base_table_length - sizeof(struct MPCTH); - position = (u_char *) cth + sizeof(struct MPCTH); - count = cth->entry_count; - apic = bus = intr = 0; - cpu = 1; /* pre-count the BSP */ - - while (count--) { - switch (type = *(u_char *) position) { - case 0: - if (processor_entry(position, cpu)) - ++cpu; - break; - case 1: - if (bus_entry(position, bus)) - ++bus; - break; - case 2: - if (io_apic_entry(position, apic)) - ++apic; - break; - case 3: - if (int_entry(position, intr)) - ++intr; - break; - case 4: - /* int_entry(position); */ - break; - default: - panic("mpfps Base Table HOSED!"); - /* NOTREACHED */ - } - - totalSize -= basetable_entry_types[type].length; - (u_char *) position += basetable_entry_types[type].length; - } - - if (boot_cpu_id == -1) - panic("NO BSP found!"); - - /* report fact that its NOT a default configuration */ - return 0; -} - -void -assign_apic_irq(int apic, int intpin, int irq) -{ - int x; - - if (int_to_apicintpin[irq].ioapic != -1) - panic("assign_apic_irq: inconsistent table"); - - int_to_apicintpin[irq].ioapic = apic; - int_to_apicintpin[irq].int_pin = intpin; - int_to_apicintpin[irq].apic_address = ioapic[apic]; - int_to_apicintpin[irq].redirindex = IOAPIC_REDTBL + 2 * intpin; - - for (x = 0; x < nintrs; x++) { - if ((io_apic_ints[x].int_type == 0 || - io_apic_ints[x].int_type == 3) && - io_apic_ints[x].int_vector == 0xff && - io_apic_ints[x].dst_apic_id == IO_TO_ID(apic) && - io_apic_ints[x].dst_apic_int == intpin) - io_apic_ints[x].int_vector = irq; - } + return (cpuid_to_globaldata[cpuid]); } void -revoke_apic_irq(int irq) -{ - int x; - int oldapic; - int oldintpin; - - if (int_to_apicintpin[irq].ioapic == -1) - panic("assign_apic_irq: inconsistent table"); - - oldapic = int_to_apicintpin[irq].ioapic; - oldintpin = int_to_apicintpin[irq].int_pin; - - int_to_apicintpin[irq].ioapic = -1; - int_to_apicintpin[irq].int_pin = 0; - int_to_apicintpin[irq].apic_address = NULL; - int_to_apicintpin[irq].redirindex = 0; - - for (x = 0; x < nintrs; x++) { - if ((io_apic_ints[x].int_type == 0 || - io_apic_ints[x].int_type == 3) && - io_apic_ints[x].int_vector == 0xff && - io_apic_ints[x].dst_apic_id == IO_TO_ID(oldapic) && - io_apic_ints[x].dst_apic_int == oldintpin) - io_apic_ints[x].int_vector = 0xff; - } -} - - -static void -allocate_apic_irq(int intr) -{ - int apic; - int intpin; - int irq; - - if (io_apic_ints[intr].int_vector != 0xff) - return; /* Interrupt handler already assigned */ - - if (io_apic_ints[intr].int_type != 0 && - (io_apic_ints[intr].int_type != 3 || - (io_apic_ints[intr].dst_apic_id == IO_TO_ID(0) && - io_apic_ints[intr].dst_apic_int == 0))) - return; /* Not INT or ExtInt on != (0, 0) */ - - irq = 0; - while (irq < APIC_INTMAPSIZE && - int_to_apicintpin[irq].ioapic != -1) - irq++; - - if (irq >= APIC_INTMAPSIZE) - return; /* No free interrupt handlers */ - - apic = ID_TO_IO(io_apic_ints[intr].dst_apic_id); - intpin = io_apic_ints[intr].dst_apic_int; - - assign_apic_irq(apic, intpin, irq); - io_apic_setup_intpin(apic, intpin); -} - - -static void -swap_apic_id(int apic, int oldid, int newid) -{ - int x; - int oapic; - - - if (oldid == newid) - return; /* Nothing to do */ - - printf("Changing APIC ID for IO APIC #%d from %d to %d in MP table\n", - apic, oldid, newid); - - /* Swap physical APIC IDs in interrupt entries */ - for (x = 0; x < nintrs; x++) { - if (io_apic_ints[x].dst_apic_id == oldid) - io_apic_ints[x].dst_apic_id = newid; - else if (io_apic_ints[x].dst_apic_id == newid) - io_apic_ints[x].dst_apic_id = oldid; - } - - /* Swap physical APIC IDs in IO_TO_ID mappings */ - for (oapic = 0; oapic < mp_napics; oapic++) - if (IO_TO_ID(oapic) == newid) - break; - - if (oapic < mp_napics) { - printf("Changing APIC ID for IO APIC #%d from " - "%d to %d in MP table\n", - oapic, newid, oldid); - IO_TO_ID(oapic) = oldid; - } - IO_TO_ID(apic) = newid; -} - - -static void -fix_id_to_io_mapping(void) -{ - int x; - - for (x = 0; x < NAPICID; x++) - ID_TO_IO(x) = -1; - - for (x = 0; x <= mp_naps; x++) - if (CPU_TO_ID(x) < NAPICID) - ID_TO_IO(CPU_TO_ID(x)) = x; - - for (x = 0; x < mp_napics; x++) - if (IO_TO_ID(x) < NAPICID) - ID_TO_IO(IO_TO_ID(x)) = x; -} - - -static int -first_free_apic_id(void) -{ - int freeid, x; - - for (freeid = 0; freeid < NAPICID; freeid++) { - for (x = 0; x <= mp_naps; x++) - if (CPU_TO_ID(x) == freeid) - break; - if (x <= mp_naps) - continue; - for (x = 0; x < mp_napics; x++) - if (IO_TO_ID(x) == freeid) - break; - if (x < mp_napics) - continue; - return freeid; - } - return freeid; -} - - -static int -io_apic_id_acceptable(int apic, int id) -{ - int cpu; /* Logical CPU number */ - int oapic; /* Logical IO APIC number for other IO APIC */ - - if (id >= NAPICID) - return 0; /* Out of range */ - - for (cpu = 0; cpu <= mp_naps; cpu++) - if (CPU_TO_ID(cpu) == id) - return 0; /* Conflict with CPU */ - - for (oapic = 0; oapic < mp_napics && oapic < apic; oapic++) - if (IO_TO_ID(oapic) == id) - return 0; /* Conflict with other APIC */ - - return 1; /* ID is acceptable for IO APIC */ -} - - -/* - * parse an Intel MP specification table - */ -static void -fix_mp_table(void) -{ - int x; - int id; - int bus_0 = 0; /* Stop GCC warning */ - int bus_pci = 0; /* Stop GCC warning */ - int num_pci_bus; - int apic; /* IO APIC unit number */ - int freeid; /* Free physical APIC ID */ - int physid; /* Current physical IO APIC ID */ - - /* - * Fix mis-numbering of the PCI bus and its INT entries if the BIOS - * did it wrong. The MP spec says that when more than 1 PCI bus - * exists the BIOS must begin with bus entries for the PCI bus and use - * actual PCI bus numbering. This implies that when only 1 PCI bus - * exists the BIOS can choose to ignore this ordering, and indeed many - * MP motherboards do ignore it. This causes a problem when the PCI - * sub-system makes requests of the MP sub-system based on PCI bus - * numbers. So here we look for the situation and renumber the - * busses and associated INTs in an effort to "make it right". - */ - - /* find bus 0, PCI bus, count the number of PCI busses */ - for (num_pci_bus = 0, x = 0; x < mp_nbusses; ++x) { - if (bus_data[x].bus_id == 0) { - bus_0 = x; - } - if (bus_data[x].bus_type == PCI) { - ++num_pci_bus; - bus_pci = x; - } - } - /* - * bus_0 == slot of bus with ID of 0 - * bus_pci == slot of last PCI bus encountered - */ - - /* check the 1 PCI bus case for sanity */ - /* if it is number 0 all is well */ - if (num_pci_bus == 1 && - bus_data[bus_pci].bus_id != 0) { - - /* mis-numbered, swap with whichever bus uses slot 0 */ - - /* swap the bus entry types */ - bus_data[bus_pci].bus_type = bus_data[bus_0].bus_type; - bus_data[bus_0].bus_type = PCI; - - /* swap each relavant INTerrupt entry */ - id = bus_data[bus_pci].bus_id; - for (x = 0; x < nintrs; ++x) { - if (io_apic_ints[x].src_bus_id == id) { - io_apic_ints[x].src_bus_id = 0; - } - else if (io_apic_ints[x].src_bus_id == 0) { - io_apic_ints[x].src_bus_id = id; - } - } - } - - /* Assign IO APIC IDs. - * - * First try the existing ID. If a conflict is detected, try - * the ID in the MP table. If a conflict is still detected, find - * a free id. - * - * We cannot use the ID_TO_IO table before all conflicts has been - * resolved and the table has been corrected. - */ - for (apic = 0; apic < mp_napics; ++apic) { /* For all IO APICs */ - - /* First try to use the value set by the BIOS */ - physid = io_apic_get_id(apic); - if (io_apic_id_acceptable(apic, physid)) { - if (IO_TO_ID(apic) != physid) - swap_apic_id(apic, IO_TO_ID(apic), physid); - continue; - } - - /* Then check if the value in the MP table is acceptable */ - if (io_apic_id_acceptable(apic, IO_TO_ID(apic))) - continue; - - /* Last resort, find a free APIC ID and use it */ - freeid = first_free_apic_id(); - if (freeid >= NAPICID) - panic("No free physical APIC IDs found"); - - if (io_apic_id_acceptable(apic, freeid)) { - swap_apic_id(apic, IO_TO_ID(apic), freeid); - continue; - } - panic("Free physical APIC ID not usable"); - } - fix_id_to_io_mapping(); - - /* detect and fix broken Compaq MP table */ - if (apic_int_type(0, 0) == -1) { - printf("APIC_IO: MP table broken: 8259->APIC entry missing!\n"); - io_apic_ints[nintrs].int_type = 3; /* ExtInt */ - io_apic_ints[nintrs].int_vector = 0xff; /* Unassigned */ - /* XXX fixme, set src bus id etc, but it doesn't seem to hurt */ - io_apic_ints[nintrs].dst_apic_id = IO_TO_ID(0); - io_apic_ints[nintrs].dst_apic_int = 0; /* Pin 0 */ - nintrs++; - } -} - - -/* Assign low level interrupt handlers */ -static void -setup_apic_irq_mapping(void) -{ - int x; - int int_vector; - - /* Clear array */ - for (x = 0; x < APIC_INTMAPSIZE; x++) { - int_to_apicintpin[x].ioapic = -1; - int_to_apicintpin[x].int_pin = 0; - int_to_apicintpin[x].apic_address = NULL; - int_to_apicintpin[x].redirindex = 0; - } - - /* First assign ISA/EISA interrupts */ - for (x = 0; x < nintrs; x++) { - int_vector = io_apic_ints[x].src_bus_irq; - if (int_vector < APIC_INTMAPSIZE && - io_apic_ints[x].int_vector == 0xff && - int_to_apicintpin[int_vector].ioapic == -1 && - (apic_int_is_bus_type(x, ISA) || - apic_int_is_bus_type(x, EISA)) && - io_apic_ints[x].int_type == 0) { - assign_apic_irq(ID_TO_IO(io_apic_ints[x].dst_apic_id), - io_apic_ints[x].dst_apic_int, - int_vector); - } - } - - /* Assign ExtInt entry if no ISA/EISA interrupt 0 entry */ - for (x = 0; x < nintrs; x++) { - if (io_apic_ints[x].dst_apic_int == 0 && - io_apic_ints[x].dst_apic_id == IO_TO_ID(0) && - io_apic_ints[x].int_vector == 0xff && - int_to_apicintpin[0].ioapic == -1 && - io_apic_ints[x].int_type == 3) { - assign_apic_irq(0, 0, 0); - break; - } - } - /* PCI interrupt assignment is deferred */ -} - - -static int -processor_entry(proc_entry_ptr entry, int cpu) -{ - /* check for usability */ - if (!(entry->cpu_flags & PROCENTRY_FLAG_EN)) - return 0; - - if(entry->apic_id >= NAPICID) - panic("CPU APIC ID out of range (0..%d)", NAPICID - 1); - /* check for BSP flag */ - if (entry->cpu_flags & PROCENTRY_FLAG_BP) { - boot_cpu_id = entry->apic_id; - CPU_TO_ID(0) = entry->apic_id; - ID_TO_CPU(entry->apic_id) = 0; - return 0; /* its already been counted */ - } - - /* add another AP to list, if less than max number of CPUs */ - else if (cpu < MAXCPU) { - CPU_TO_ID(cpu) = entry->apic_id; - ID_TO_CPU(entry->apic_id) = cpu; - return 1; - } - - return 0; -} - - -static int -bus_entry(bus_entry_ptr entry, int bus) -{ - int x; - char c, name[8]; - - /* encode the name into an index */ - for (x = 0; x < 6; ++x) { - if ((c = entry->bus_type[x]) == ' ') - break; - name[x] = c; - } - name[x] = '\0'; - - if ((x = lookup_bus_type(name)) == UNKNOWN_BUSTYPE) - panic("unknown bus type: '%s'", name); - - bus_data[bus].bus_id = entry->bus_id; - bus_data[bus].bus_type = x; - - return 1; -} - - -static int -io_apic_entry(io_apic_entry_ptr entry, int apic) -{ - if (!(entry->apic_flags & IOAPICENTRY_FLAG_EN)) - return 0; - - IO_TO_ID(apic) = entry->apic_id; - if (entry->apic_id < NAPICID) - ID_TO_IO(entry->apic_id) = apic; - - return 1; -} - - -static int -lookup_bus_type(char *name) -{ - int x; - - for (x = 0; x < MAX_BUSTYPE; ++x) - if (strcmp(bus_type_table[x].name, name) == 0) - return bus_type_table[x].type; - - return UNKNOWN_BUSTYPE; -} - - -static int -int_entry(int_entry_ptr entry, int intr) -{ - int apic; - - io_apic_ints[intr].int_type = entry->int_type; - io_apic_ints[intr].int_flags = entry->int_flags; - io_apic_ints[intr].src_bus_id = entry->src_bus_id; - io_apic_ints[intr].src_bus_irq = entry->src_bus_irq; - if (entry->dst_apic_id == 255) { - /* This signal goes to all IO APICS. Select an IO APIC - with sufficient number of interrupt pins */ - for (apic = 0; apic < mp_napics; apic++) - if (((io_apic_read(apic, IOAPIC_VER) & - IOART_VER_MAXREDIR) >> MAXREDIRSHIFT) >= - entry->dst_apic_int) - break; - if (apic < mp_napics) - io_apic_ints[intr].dst_apic_id = IO_TO_ID(apic); - else - io_apic_ints[intr].dst_apic_id = entry->dst_apic_id; - } else - io_apic_ints[intr].dst_apic_id = entry->dst_apic_id; - io_apic_ints[intr].dst_apic_int = entry->dst_apic_int; - - return 1; -} - - -static int -apic_int_is_bus_type(int intr, int bus_type) -{ - int bus; - - for (bus = 0; bus < mp_nbusses; ++bus) - if ((bus_data[bus].bus_id == io_apic_ints[intr].src_bus_id) - && ((int) bus_data[bus].bus_type == bus_type)) - return 1; - - return 0; -} - - -/* - * Given a traditional ISA INT mask, return an APIC mask. - */ -u_int -isa_apic_mask(u_int isa_mask) -{ - int isa_irq; - int apic_pin; - -#if defined(SKIP_IRQ15_REDIRECT) - if (isa_mask == (1 << 15)) { - printf("skipping ISA IRQ15 redirect\n"); - return isa_mask; - } -#endif /* SKIP_IRQ15_REDIRECT */ - - isa_irq = ffs(isa_mask); /* find its bit position */ - if (isa_irq == 0) /* doesn't exist */ - return 0; - --isa_irq; /* make it zero based */ - - apic_pin = isa_apic_irq(isa_irq); /* look for APIC connection */ - if (apic_pin == -1) - return 0; - - return (1 << apic_pin); /* convert pin# to a mask */ -} - - -/* - * Determine which APIC pin an ISA/EISA INT is attached to. - */ -#define INTTYPE(I) (io_apic_ints[(I)].int_type) -#define INTPIN(I) (io_apic_ints[(I)].dst_apic_int) -#define INTIRQ(I) (io_apic_ints[(I)].int_vector) -#define INTAPIC(I) (ID_TO_IO(io_apic_ints[(I)].dst_apic_id)) - -#define SRCBUSIRQ(I) (io_apic_ints[(I)].src_bus_irq) -int -isa_apic_irq(int isa_irq) -{ - int intr; - - for (intr = 0; intr < nintrs; ++intr) { /* check each record */ - if (INTTYPE(intr) == 0) { /* standard INT */ - if (SRCBUSIRQ(intr) == isa_irq) { - if (apic_int_is_bus_type(intr, ISA) || - apic_int_is_bus_type(intr, EISA)) { - if (INTIRQ(intr) == 0xff) - return -1; /* unassigned */ - return INTIRQ(intr); /* found */ - } - } - } - } - return -1; /* NOT found */ -} - - -/* - * Determine which APIC pin a PCI INT is attached to. - */ -#define SRCBUSID(I) (io_apic_ints[(I)].src_bus_id) -#define SRCBUSDEVICE(I) ((io_apic_ints[(I)].src_bus_irq >> 2) & 0x1f) -#define SRCBUSLINE(I) (io_apic_ints[(I)].src_bus_irq & 0x03) -int -pci_apic_irq(int pciBus, int pciDevice, int pciInt) -{ - int intr; - - --pciInt; /* zero based */ - - for (intr = 0; intr < nintrs; ++intr) /* check each record */ - if ((INTTYPE(intr) == 0) /* standard INT */ - && (SRCBUSID(intr) == pciBus) - && (SRCBUSDEVICE(intr) == pciDevice) - && (SRCBUSLINE(intr) == pciInt)) /* a candidate IRQ */ - if (apic_int_is_bus_type(intr, PCI)) { - if (INTIRQ(intr) == 0xff) - allocate_apic_irq(intr); - if (INTIRQ(intr) == 0xff) - return -1; /* unassigned */ - return INTIRQ(intr); /* exact match */ - } - - return -1; /* NOT found */ -} - -int -next_apic_irq(int irq) -{ - int intr, ointr; - int bus, bustype; - - bus = 0; - bustype = 0; - for (intr = 0; intr < nintrs; intr++) { - if (INTIRQ(intr) != irq || INTTYPE(intr) != 0) - continue; - bus = SRCBUSID(intr); - bustype = apic_bus_type(bus); - if (bustype != ISA && - bustype != EISA && - bustype != PCI) - continue; - break; - } - if (intr >= nintrs) { - return -1; - } - for (ointr = intr + 1; ointr < nintrs; ointr++) { - if (INTTYPE(ointr) != 0) - continue; - if (bus != SRCBUSID(ointr)) - continue; - if (bustype == PCI) { - if (SRCBUSDEVICE(intr) != SRCBUSDEVICE(ointr)) - continue; - if (SRCBUSLINE(intr) != SRCBUSLINE(ointr)) - continue; - } - if (bustype == ISA || bustype == EISA) { - if (SRCBUSIRQ(intr) != SRCBUSIRQ(ointr)) - continue; - } - if (INTPIN(intr) == INTPIN(ointr)) - continue; - break; - } - if (ointr >= nintrs) { - return -1; - } - return INTIRQ(ointr); -} -#undef SRCBUSLINE -#undef SRCBUSDEVICE -#undef SRCBUSID -#undef SRCBUSIRQ - -#undef INTPIN -#undef INTIRQ -#undef INTAPIC -#undef INTTYPE - - -/* - * Reprogram the MB chipset to NOT redirect an ISA INTerrupt. - * - * XXX FIXME: - * Exactly what this means is unclear at this point. It is a solution - * for motherboards that redirect the MBIRQ0 pin. Generically a motherboard - * could route any of the ISA INTs to upper (>15) IRQ values. But most would - * NOT be redirected via MBIRQ0, thus "undirect()ing" them would NOT be an - * option. - */ -int -undirect_isa_irq(int rirq) -{ -#if defined(READY) - if (bootverbose) - printf("Freeing redirected ISA irq %d.\n", rirq); - /** FIXME: tickle the MB redirector chip */ - return -1; -#else - if (bootverbose) - printf("Freeing (NOT implemented) redirected ISA irq %d.\n", rirq); - return 0; -#endif /* READY */ -} - - -/* - * Reprogram the MB chipset to NOT redirect a PCI INTerrupt - */ -int -undirect_pci_irq(int rirq) -{ -#if defined(READY) - if (bootverbose) - printf("Freeing redirected PCI irq %d.\n", rirq); - - /** FIXME: tickle the MB redirector chip */ - return -1; -#else - if (bootverbose) - printf("Freeing (NOT implemented) redirected PCI irq %d.\n", - rirq); - return 0; -#endif /* READY */ -} - - -/* - * given a bus ID, return: - * the bus type if found - * -1 if NOT found - */ -int -apic_bus_type(int id) -{ - int x; - - for (x = 0; x < mp_nbusses; ++x) - if (bus_data[x].bus_id == id) - return bus_data[x].bus_type; - - return -1; -} - - -/* - * given a LOGICAL APIC# and pin#, return: - * the associated src bus ID if found - * -1 if NOT found - */ -int -apic_src_bus_id(int apic, int pin) -{ - int x; - - /* search each of the possible INTerrupt sources */ - for (x = 0; x < nintrs; ++x) - if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) && - (pin == io_apic_ints[x].dst_apic_int)) - return (io_apic_ints[x].src_bus_id); - - return -1; /* NOT found */ -} - - -/* - * given a LOGICAL APIC# and pin#, return: - * the associated src bus IRQ if found - * -1 if NOT found - */ -int -apic_src_bus_irq(int apic, int pin) -{ - int x; - - for (x = 0; x < nintrs; x++) - if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) && - (pin == io_apic_ints[x].dst_apic_int)) - return (io_apic_ints[x].src_bus_irq); - - return -1; /* NOT found */ -} - - -/* - * given a LOGICAL APIC# and pin#, return: - * the associated INTerrupt type if found - * -1 if NOT found - */ -int -apic_int_type(int apic, int pin) -{ - int x; - - /* search each of the possible INTerrupt sources */ - for (x = 0; x < nintrs; ++x) - if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) && - (pin == io_apic_ints[x].dst_apic_int)) - return (io_apic_ints[x].int_type); - - return -1; /* NOT found */ -} - -int -apic_irq(int apic, int pin) -{ - int x; - int res; - - for (x = 0; x < nintrs; ++x) - if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) && - (pin == io_apic_ints[x].dst_apic_int)) { - res = io_apic_ints[x].int_vector; - if (res == 0xff) - return -1; - if (apic != int_to_apicintpin[res].ioapic) - panic("apic_irq: inconsistent table"); - if (pin != int_to_apicintpin[res].int_pin) - panic("apic_irq inconsistent table (2)"); - return res; - } - return -1; -} - - -/* - * given a LOGICAL APIC# and pin#, return: - * the associated trigger mode if found - * -1 if NOT found - */ -int -apic_trigger(int apic, int pin) -{ - int x; - - /* search each of the possible INTerrupt sources */ - for (x = 0; x < nintrs; ++x) - if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) && - (pin == io_apic_ints[x].dst_apic_int)) - return ((io_apic_ints[x].int_flags >> 2) & 0x03); - - return -1; /* NOT found */ -} - - -/* - * given a LOGICAL APIC# and pin#, return: - * the associated 'active' level if found - * -1 if NOT found - */ -int -apic_polarity(int apic, int pin) -{ - int x; - - /* search each of the possible INTerrupt sources */ - for (x = 0; x < nintrs; ++x) - if ((apic == ID_TO_IO(io_apic_ints[x].dst_apic_id)) && - (pin == io_apic_ints[x].dst_apic_int)) - return (io_apic_ints[x].int_flags & 0x03); - - return -1; /* NOT found */ -} - - -/* - * set data according to MP defaults - * FIXME: probably not complete yet... - */ -static void -default_mp_table(int type) -{ - int ap_cpu_id; -#if defined(APIC_IO) - int io_apic_id; - int pin; -#endif /* APIC_IO */ - -#if 0 - printf(" MP default config type: %d\n", type); - switch (type) { - case 1: - printf(" bus: ISA, APIC: 82489DX\n"); - break; - case 2: - printf(" bus: EISA, APIC: 82489DX\n"); - break; - case 3: - printf(" bus: EISA, APIC: 82489DX\n"); - break; - case 4: - printf(" bus: MCA, APIC: 82489DX\n"); - break; - case 5: - printf(" bus: ISA+PCI, APIC: Integrated\n"); - break; - case 6: - printf(" bus: EISA+PCI, APIC: Integrated\n"); - break; - case 7: - printf(" bus: MCA+PCI, APIC: Integrated\n"); - break; - default: - printf(" future type\n"); - break; - /* NOTREACHED */ - } -#endif /* 0 */ - - boot_cpu_id = (lapic.id & APIC_ID_MASK) >> 24; - ap_cpu_id = (boot_cpu_id == 0) ? 1 : 0; - - /* BSP */ - CPU_TO_ID(0) = boot_cpu_id; - ID_TO_CPU(boot_cpu_id) = 0; - - /* one and only AP */ - CPU_TO_ID(1) = ap_cpu_id; - ID_TO_CPU(ap_cpu_id) = 1; - -#if defined(APIC_IO) - /* one and only IO APIC */ - io_apic_id = (io_apic_read(0, IOAPIC_ID) & APIC_ID_MASK) >> 24; - - /* - * sanity check, refer to MP spec section 3.6.6, last paragraph - * necessary as some hardware isn't properly setting up the IO APIC - */ -#if defined(REALLY_ANAL_IOAPICID_VALUE) - if (io_apic_id != 2) { -#else - if ((io_apic_id == 0) || (io_apic_id == 1) || (io_apic_id == 15)) { -#endif /* REALLY_ANAL_IOAPICID_VALUE */ - io_apic_set_id(0, 2); - io_apic_id = 2; - } - IO_TO_ID(0) = io_apic_id; - ID_TO_IO(io_apic_id) = 0; -#endif /* APIC_IO */ - - /* fill out bus entries */ - switch (type) { - case 1: - case 2: - case 3: - case 4: - case 5: - case 6: - case 7: - bus_data[0].bus_id = default_data[type - 1][1]; - bus_data[0].bus_type = default_data[type - 1][2]; - bus_data[1].bus_id = default_data[type - 1][3]; - bus_data[1].bus_type = default_data[type - 1][4]; - break; - - /* case 4: case 7: MCA NOT supported */ - default: /* illegal/reserved */ - panic("BAD default MP config: %d", type); - /* NOTREACHED */ - } - -#if defined(APIC_IO) - /* general cases from MP v1.4, table 5-2 */ - for (pin = 0; pin < 16; ++pin) { - io_apic_ints[pin].int_type = 0; - io_apic_ints[pin].int_flags = 0x05; /* edge/active-hi */ - io_apic_ints[pin].src_bus_id = 0; - io_apic_ints[pin].src_bus_irq = pin; /* IRQ2 caught below */ - io_apic_ints[pin].dst_apic_id = io_apic_id; - io_apic_ints[pin].dst_apic_int = pin; /* 1-to-1 */ - } - - /* special cases from MP v1.4, table 5-2 */ - if (type == 2) { - io_apic_ints[2].int_type = 0xff; /* N/C */ - io_apic_ints[13].int_type = 0xff; /* N/C */ -#if !defined(APIC_MIXED_MODE) - /** FIXME: ??? */ - panic("sorry, can't support type 2 default yet"); -#endif /* APIC_MIXED_MODE */ - } - else - io_apic_ints[2].src_bus_irq = 0; /* ISA IRQ0 is on APIC INT 2 */ - - if (type == 7) - io_apic_ints[0].int_type = 0xff; /* N/C */ - else - io_apic_ints[0].int_type = 3; /* vectored 8259 */ -#endif /* APIC_IO */ -} - - -/* - * start each AP in our list - */ -static int -start_all_aps(u_int boot_addr) -{ - int x, i, pg; - u_char mpbiosreason; - u_long mpbioswarmvec; - struct globaldata *gd; - char *stack; - uintptr_t kptbase; - - POSTCODE(START_ALL_APS_POST); - - /* initialize BSP's local APIC */ - apic_initialize(); - bsp_apic_ready = 1; - - /* install the AP 1st level boot code */ - install_ap_tramp(boot_addr); - - - /* save the current value of the warm-start vector */ - mpbioswarmvec = *((u_long *) WARMBOOT_OFF); -#ifndef PC98 - outb(CMOS_REG, BIOS_RESET); - mpbiosreason = inb(CMOS_DATA); -#endif - - /* record BSP in CPU map */ - all_cpus = 1; - - /* set up temporary P==V mapping for AP boot */ - /* XXX this is a hack, we should boot the AP on its own stack/PTD */ - kptbase = (uintptr_t)(void *)KPTphys; - for (x = 0; x < NKPT; x++) - PTD[x] = (pd_entry_t)(PG_V | PG_RW | - ((kptbase + x * PAGE_SIZE) & PG_FRAME)); - invltlb(); - - /* start each AP */ - for (x = 1; x <= mp_naps; ++x) { - - /* This is a bit verbose, it will go away soon. */ - - /* first page of AP's private space */ - pg = x * i386_btop(sizeof(struct privatespace)); - - /* allocate a new private data page */ - gd = (struct globaldata *)kmem_alloc(kernel_map, PAGE_SIZE); - - /* wire it into the private page table page */ - SMPpt[pg] = (pt_entry_t)(PG_V | PG_RW | vtophys(gd)); - - /* allocate and set up an idle stack data page */ - stack = (char *)kmem_alloc(kernel_map, UPAGES*PAGE_SIZE); - for (i = 0; i < UPAGES; i++) - SMPpt[pg + 1 + i] = (pt_entry_t) - (PG_V | PG_RW | vtophys(PAGE_SIZE * i + stack)); - - /* prime data page for it to use */ - SLIST_INSERT_HEAD(&cpuhead, gd, gd_allcpu); - gd->gd_cpuid = x; - - /* setup a vector to our boot code */ - *((volatile u_short *) WARMBOOT_OFF) = WARMBOOT_TARGET; - *((volatile u_short *) WARMBOOT_SEG) = (boot_addr >> 4); -#ifndef PC98 - outb(CMOS_REG, BIOS_RESET); - outb(CMOS_DATA, BIOS_WARM); /* 'warm-start' */ -#endif - - bootSTK = &SMP_prvspace[x].idlestack[UPAGES*PAGE_SIZE]; - bootAP = x; - - /* attempt to start the Application Processor */ - CHECK_INIT(99); /* setup checkpoints */ - if (!start_ap(x, boot_addr)) { - printf("AP #%d (PHY# %d) failed!\n", x, CPU_TO_ID(x)); - CHECK_PRINT("trace"); /* show checkpoints */ - /* better panic as the AP may be running loose */ - printf("panic y/n? [y] "); - if (cngetc() != 'n') - panic("bye-bye"); - } - CHECK_PRINT("trace"); /* show checkpoints */ - - /* record its version info */ - cpu_apic_versions[x] = cpu_apic_versions[0]; - - all_cpus |= (1 << x); /* record AP in CPU map */ - } - - /* build our map of 'other' CPUs */ - PCPU_SET(other_cpus, all_cpus & ~(1 << PCPU_GET(cpuid))); - - /* fill in our (BSP) APIC version */ - cpu_apic_versions[0] = lapic.version; - - /* restore the warmstart vector */ - *(u_long *) WARMBOOT_OFF = mpbioswarmvec; -#ifndef PC98 - outb(CMOS_REG, BIOS_RESET); - outb(CMOS_DATA, mpbiosreason); -#endif - - /* - * Set up the idle context for the BSP. Similar to above except - * that some was done by locore, some by pmap.c and some is implicit - * because the BSP is cpu#0 and the page is initially zero, and also - * because we can refer to variables by name on the BSP.. - */ - - /* Allocate and setup BSP idle stack */ - stack = (char *)kmem_alloc(kernel_map, UPAGES * PAGE_SIZE); - for (i = 0; i < UPAGES; i++) - SMPpt[1 + i] = (pt_entry_t) - (PG_V | PG_RW | vtophys(PAGE_SIZE * i + stack)); - - for (x = 0; x < NKPT; x++) - PTD[x] = 0; - pmap_set_opt(); - - /* number of APs actually started */ - return mp_ncpus - 1; -} - - -/* - * load the 1st level AP boot code into base memory. - */ - -/* targets for relocation */ -extern void bigJump(void); -extern void bootCodeSeg(void); -extern void bootDataSeg(void); -extern void MPentry(void); -extern u_int MP_GDT; -extern u_int mp_gdtbase; - -static void -install_ap_tramp(u_int boot_addr) -{ - int x; - int size = *(int *) ((u_long) & bootMP_size); - u_char *src = (u_char *) ((u_long) bootMP); - u_char *dst = (u_char *) boot_addr + KERNBASE; - u_int boot_base = (u_int) bootMP; - u_int8_t *dst8; - u_int16_t *dst16; - u_int32_t *dst32; - - POSTCODE(INSTALL_AP_TRAMP_POST); - - for (x = 0; x < size; ++x) - *dst++ = *src++; - - /* - * modify addresses in code we just moved to basemem. unfortunately we - * need fairly detailed info about mpboot.s for this to work. changes - * to mpboot.s might require changes here. - */ - - /* boot code is located in KERNEL space */ - dst = (u_char *) boot_addr + KERNBASE; - - /* modify the lgdt arg */ - dst32 = (u_int32_t *) (dst + ((u_int) & mp_gdtbase - boot_base)); - *dst32 = boot_addr + ((u_int) & MP_GDT - boot_base); - - /* modify the ljmp target for MPentry() */ - dst32 = (u_int32_t *) (dst + ((u_int) bigJump - boot_base) + 1); - *dst32 = ((u_int) MPentry - KERNBASE); - - /* modify the target for boot code segment */ - dst16 = (u_int16_t *) (dst + ((u_int) bootCodeSeg - boot_base)); - dst8 = (u_int8_t *) (dst16 + 1); - *dst16 = (u_int) boot_addr & 0xffff; - *dst8 = ((u_int) boot_addr >> 16) & 0xff; - - /* modify the target for boot data segment */ - dst16 = (u_int16_t *) (dst + ((u_int) bootDataSeg - boot_base)); - dst8 = (u_int8_t *) (dst16 + 1); - *dst16 = (u_int) boot_addr & 0xffff; - *dst8 = ((u_int) boot_addr >> 16) & 0xff; -} - - -/* - * this function starts the AP (application processor) identified - * by the APIC ID 'physicalCpu'. It does quite a "song and dance" - * to accomplish this. This is necessary because of the nuances - * of the different hardware we might encounter. It ain't pretty, - * but it seems to work. - */ -static int -start_ap(int logical_cpu, u_int boot_addr) -{ - int physical_cpu; - int vector; - int cpus; - u_long icr_lo, icr_hi; - - POSTCODE(START_AP_POST); - - /* get the PHYSICAL APIC ID# */ - physical_cpu = CPU_TO_ID(logical_cpu); - - /* calculate the vector */ - vector = (boot_addr >> 12) & 0xff; - - /* used as a watchpoint to signal AP startup */ - cpus = mp_ncpus; - - /* - * first we do an INIT/RESET IPI this INIT IPI might be run, reseting - * and running the target CPU. OR this INIT IPI might be latched (P5 - * bug), CPU waiting for STARTUP IPI. OR this INIT IPI might be - * ignored. - */ - - /* setup the address for the target AP */ - icr_hi = lapic.icr_hi & ~APIC_ID_MASK; - icr_hi |= (physical_cpu << 24); - lapic.icr_hi = icr_hi; - - /* do an INIT IPI: assert RESET */ - icr_lo = lapic.icr_lo & 0xfff00000; - lapic.icr_lo = icr_lo | 0x0000c500; - - /* wait for pending status end */ - while (lapic.icr_lo & APIC_DELSTAT_MASK) - /* spin */ ; - - /* do an INIT IPI: deassert RESET */ - lapic.icr_lo = icr_lo | 0x00008500; - - /* wait for pending status end */ - u_sleep(10000); /* wait ~10mS */ - while (lapic.icr_lo & APIC_DELSTAT_MASK) - /* spin */ ; - - /* - * next we do a STARTUP IPI: the previous INIT IPI might still be - * latched, (P5 bug) this 1st STARTUP would then terminate - * immediately, and the previously started INIT IPI would continue. OR - * the previous INIT IPI has already run. and this STARTUP IPI will - * run. OR the previous INIT IPI was ignored. and this STARTUP IPI - * will run. - */ - - /* do a STARTUP IPI */ - lapic.icr_lo = icr_lo | 0x00000600 | vector; - while (lapic.icr_lo & APIC_DELSTAT_MASK) - /* spin */ ; - u_sleep(200); /* wait ~200uS */ - - /* - * finally we do a 2nd STARTUP IPI: this 2nd STARTUP IPI should run IF - * the previous STARTUP IPI was cancelled by a latched INIT IPI. OR - * this STARTUP IPI will be ignored, as only ONE STARTUP IPI is - * recognized after hardware RESET or INIT IPI. - */ - - lapic.icr_lo = icr_lo | 0x00000600 | vector; - while (lapic.icr_lo & APIC_DELSTAT_MASK) - /* spin */ ; - u_sleep(200); /* wait ~200uS */ - - /* wait for it to start */ - set_apic_timer(5000000);/* == 5 seconds */ - while (read_apic_timer()) - if (mp_ncpus > cpus) - return 1; /* return SUCCESS */ - - return 0; /* return FAILURE */ -} - -/* - * Flush the TLB on all other CPU's - * - * XXX: Needs to handshake and wait for completion before proceding. - */ -void -smp_invltlb(void) -{ -#if defined(APIC_IO) - if (smp_started && invltlb_ok) - ipi_all_but_self(IPI_INVLTLB); -#endif /* APIC_IO */ -} - -void -invlpg(u_int addr) -{ - __asm __volatile("invlpg (%0)"::"r"(addr):"memory"); - - /* send a message to the other CPUs */ - smp_invltlb(); -} - -void -invltlb(void) -{ - u_long temp; - - /* - * This should be implemented as load_cr3(rcr3()) when load_cr3() is - * inlined. - */ - __asm __volatile("movl %%cr3, %0; movl %0, %%cr3":"=r"(temp) :: "memory"); - - /* send a message to the other CPUs */ - smp_invltlb(); -} - - -/* - * This is called once the rest of the system is up and running and we're - * ready to let the AP's out of the pen. - */ -void -ap_init(void) +forward_signal(struct proc *p) { - u_int apic_id; - - /* spin until all the AP's are ready */ - while (!aps_ready) - /* spin */ ; - - /* - * Set curproc to our per-cpu idleproc so that mutexes have - * something unique to lock with. - */ - PCPU_SET(curproc, PCPU_GET(idleproc)); - PCPU_SET(spinlocks, NULL); - - /* lock against other AP's that are waking up */ - mtx_lock_spin(&ap_boot_mtx); - - /* BSP may have changed PTD while we're waiting for the lock */ - cpu_invltlb(); - - smp_cpus++; - -#if defined(I586_CPU) && !defined(NO_F00F_HACK) - lidt(&r_idt); -#endif - - /* Build our map of 'other' CPUs. */ - PCPU_SET(other_cpus, all_cpus & ~(1 << PCPU_GET(cpuid))); - - printf("SMP: AP CPU #%d Launched!\n", PCPU_GET(cpuid)); - - /* set up CPU registers and state */ - cpu_setregs(); - - /* set up FPU state on the AP */ - npxinit(__INITIAL_NPXCW__); - - /* A quick check from sanity claus */ - apic_id = (apic_id_to_logical[(lapic.id & 0x0f000000) >> 24]); - if (PCPU_GET(cpuid) != apic_id) { - printf("SMP: cpuid = %d\n", PCPU_GET(cpuid)); - printf("SMP: apic_id = %d\n", apic_id); - printf("PTD[MPPTDI] = %p\n", (void *)PTD[MPPTDI]); - panic("cpuid mismatch! boom!!"); - } - - /* Init local apic for irq's */ - apic_initialize(); - - /* Set memory range attributes for this CPU to match the BSP */ - mem_range_AP_init(); + int id; /* - * Activate smp_invltlb, although strictly speaking, this isn't - * quite correct yet. We should have a bitfield for cpus willing - * to accept TLB flush IPI's or something and sync them. + * signotify() has already set PS_ASTPENDING on this process so all + * we need to do is poke it if it is currently executing so that it + * executes ast(). */ - if (smp_cpus == mp_ncpus) { - invltlb_ok = 1; - smp_started = 1; /* enable IPI's, tlb shootdown, freezes etc */ - smp_active = 1; /* historic */ - } - - /* let other AP's wake up now */ - mtx_unlock_spin(&ap_boot_mtx); - - /* wait until all the AP's are up */ - while (smp_started == 0) - ; /* nothing */ - - microuptime(PCPU_PTR(switchtime)); - PCPU_SET(switchticks, ticks); - - /* ok, now grab sched_lock and enter the scheduler */ - enable_intr(); - mtx_lock_spin(&sched_lock); - cpu_throw(); /* doesn't return */ - - panic("scheduler returned us to ap_init"); -} - -#define CHECKSTATE_USER 0 -#define CHECKSTATE_SYS 1 -#define CHECKSTATE_INTR 2 - -/* Do not staticize. Used from apic_vector.s */ -struct proc* checkstate_curproc[MAXCPU]; -int checkstate_cpustate[MAXCPU]; -u_long checkstate_pc[MAXCPU]; - -#define PC_TO_INDEX(pc, prof) \ - ((int)(((u_quad_t)((pc) - (prof)->pr_off) * \ - (u_quad_t)((prof)->pr_scale)) >> 16) & ~1) - -static void -addupc_intr_forwarded(struct proc *p, int id, int *astmap) -{ - int i; - struct uprof *prof; - u_long pc; - - pc = checkstate_pc[id]; - prof = &p->p_stats->p_prof; - if (pc >= prof->pr_off && - (i = PC_TO_INDEX(pc, prof)) < prof->pr_size) { - mtx_assert(&sched_lock, MA_OWNED); - if ((p->p_sflag & PS_OWEUPC) == 0) { - prof->pr_addr = pc; - prof->pr_ticks = 1; - p->p_sflag |= PS_OWEUPC; - } - *astmap |= (1 << id); - } -} - -static void -forwarded_statclock(int id, int pscnt, int *astmap) -{ - struct pstats *pstats; - long rss; - struct rusage *ru; - struct vmspace *vm; - int cpustate; - struct proc *p; -#ifdef GPROF - register struct gmonparam *g; - int i; -#endif - mtx_assert(&sched_lock, MA_OWNED); - p = checkstate_curproc[id]; - cpustate = checkstate_cpustate[id]; - - /* XXX */ - if (p->p_ithd) - cpustate = CHECKSTATE_INTR; - else if (p == SMP_prvspace[id].globaldata.gd_idleproc) - cpustate = CHECKSTATE_SYS; - - switch (cpustate) { - case CHECKSTATE_USER: - if (p->p_sflag & PS_PROFIL) - addupc_intr_forwarded(p, id, astmap); - if (pscnt > 1) - return; - p->p_uticks++; - if (p->p_nice > NZERO) - cp_time[CP_NICE]++; - else - cp_time[CP_USER]++; - break; - case CHECKSTATE_SYS: -#ifdef GPROF - /* - * Kernel statistics are just like addupc_intr, only easier. - */ - g = &_gmonparam; - if (g->state == GMON_PROF_ON) { - i = checkstate_pc[id] - g->lowpc; - if (i < g->textsize) { - i /= HISTFRACTION * sizeof(*g->kcount); - g->kcount[i]++; - } - } -#endif - if (pscnt > 1) - return; + KASSERT(p->p_stat == SRUN, ("forward_signal: process is not SRUN")); - p->p_sticks++; - if (p == SMP_prvspace[id].globaldata.gd_idleproc) - cp_time[CP_IDLE]++; - else - cp_time[CP_SYS]++; - break; - case CHECKSTATE_INTR: - default: -#ifdef GPROF - /* - * Kernel statistics are just like addupc_intr, only easier. - */ - g = &_gmonparam; - if (g->state == GMON_PROF_ON) { - i = checkstate_pc[id] - g->lowpc; - if (i < g->textsize) { - i /= HISTFRACTION * sizeof(*g->kcount); - g->kcount[i]++; - } - } -#endif - if (pscnt > 1) - return; - KASSERT(p != NULL, ("NULL process in interrupt state")); - p->p_iticks++; - cp_time[CP_INTR]++; - } - - schedclock(p); - - /* Update resource usage integrals and maximums. */ - if ((pstats = p->p_stats) != NULL && - (ru = &pstats->p_ru) != NULL && - (vm = p->p_vmspace) != NULL) { - ru->ru_ixrss += pgtok(vm->vm_tsize); - ru->ru_idrss += pgtok(vm->vm_dsize); - ru->ru_isrss += pgtok(vm->vm_ssize); - rss = pgtok(vmspace_resident_count(vm)); - if (ru->ru_maxrss < rss) - ru->ru_maxrss = rss; - } -} - -void -forward_statclock(int pscnt) -{ - int map; - int id; - int i; - - /* Kludge. We don't yet have separate locks for the interrupts - * and the kernel. This means that we cannot let the other processors - * handle complex interrupts while inhibiting them from entering - * the kernel in a non-interrupt context. - * - * What we can do, without changing the locking mechanisms yet, - * is letting the other processors handle a very simple interrupt - * (wich determines the processor states), and do the main - * work ourself. - */ - - CTR1(KTR_SMP, "forward_statclock(%d)", pscnt); + CTR1(KTR_SMP, "forward_signal(%p)", p); - if (!smp_started || !invltlb_ok || cold || panicstr) + if (!smp_started || cold || panicstr) return; - - /* Step 1: Probe state (user, cpu, interrupt, spinlock, idle ) */ - - map = PCPU_GET(other_cpus) & ~stopped_cpus ; - checkstate_probed_cpus = 0; - if (map != 0) - ipi_selected(map, IPI_CHECKSTATE); - - i = 0; - while (checkstate_probed_cpus != map) { - /* spin */ - i++; - if (i == 100000) { -#ifdef DIAGNOSTIC - printf("forward_statclock: checkstate %x\n", - checkstate_probed_cpus); -#endif - break; - } - } - - /* - * Step 2: walk through other processors processes, update ticks and - * profiling info. - */ - - map = 0; - for (id = 0; id < mp_ncpus; id++) { - if (id == PCPU_GET(cpuid)) - continue; - if (((1 << id) & checkstate_probed_cpus) == 0) - continue; - forwarded_statclock(id, pscnt, &map); - } - if (map != 0) { - checkstate_need_ast |= map; - ipi_selected(map, IPI_AST); - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#ifdef DIAGNOSTIC - printf("forward_statclock: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } - } -} - -void -forward_hardclock(int pscnt) -{ - int map; - int id; - struct proc *p; - struct pstats *pstats; - int i; - - /* Kludge. We don't yet have separate locks for the interrupts - * and the kernel. This means that we cannot let the other processors - * handle complex interrupts while inhibiting them from entering - * the kernel in a non-interrupt context. - * - * What we can do, without changing the locking mechanisms yet, - * is letting the other processors handle a very simple interrupt - * (wich determines the processor states), and do the main - * work ourself. - */ - - CTR1(KTR_SMP, "forward_hardclock(%d)", pscnt); - - if (!smp_started || !invltlb_ok || cold || panicstr) + if (!forward_signal_enabled) return; - /* Step 1: Probe state (user, cpu, interrupt, spinlock, idle) */ - - map = PCPU_GET(other_cpus) & ~stopped_cpus ; - checkstate_probed_cpus = 0; - if (map != 0) - ipi_selected(map, IPI_CHECKSTATE); - - i = 0; - while (checkstate_probed_cpus != map) { - /* spin */ - i++; - if (i == 100000) { -#ifdef DIAGNOSTIC - printf("forward_hardclock: checkstate %x\n", - checkstate_probed_cpus); -#endif - break; - } - } - - /* - * Step 2: walk through other processors processes, update virtual - * timer and profiling timer. If stathz == 0, also update ticks and - * profiling info. - */ - - map = 0; - for (id = 0; id < mp_ncpus; id++) { - if (id == PCPU_GET(cpuid)) - continue; - if (((1 << id) & checkstate_probed_cpus) == 0) - continue; - p = checkstate_curproc[id]; - if (p) { - pstats = p->p_stats; - if (checkstate_cpustate[id] == CHECKSTATE_USER && - timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) && - itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) { - p->p_sflag |= PS_ALRMPEND; - map |= (1 << id); - } - if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) && - itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) { - p->p_sflag |= PS_PROFPEND; - map |= (1 << id); - } - } - if (stathz == 0) { - forwarded_statclock( id, pscnt, &map); - } - } - if (map != 0) { - checkstate_need_ast |= map; - ipi_selected(map, IPI_AST); - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#ifdef DIAGNOSTIC - printf("forward_hardclock: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } - } -} - -void -forward_signal(struct proc *p) -{ - int map; - int id; - int i; - - /* Kludge. We don't yet have separate locks for the interrupts - * and the kernel. This means that we cannot let the other processors - * handle complex interrupts while inhibiting them from entering - * the kernel in a non-interrupt context. - * - * What we can do, without changing the locking mechanisms yet, - * is letting the other processors handle a very simple interrupt - * (wich determines the processor states), and do the main - * work ourself. - */ - - CTR1(KTR_SMP, "forward_signal(%p)", p); - - if (!smp_started || !invltlb_ok || cold || panicstr) + /* No need to IPI ourself. */ + if (p == curproc) return; - if (!forward_signal_enabled) + + id = p->p_oncpu; + if (id == NOCPU) return; - mtx_lock_spin(&sched_lock); - while (1) { - if (p->p_stat != SRUN) { - mtx_unlock_spin(&sched_lock); - return; - } - id = p->p_oncpu; - mtx_unlock_spin(&sched_lock); - if (id == 0xff) - return; - map = (1<<id); - checkstate_need_ast |= map; - ipi_selected(map, IPI_AST); - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#if 0 - printf("forward_signal: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; - } - } - mtx_lock_spin(&sched_lock); - if (id == p->p_oncpu) { - mtx_unlock_spin(&sched_lock); - return; - } - } + ipi_selected(1 << id, IPI_AST); } void forward_roundrobin(void) { - u_int map; - int i; + struct globaldata *gd; + struct proc *p; + u_int id, map; + + mtx_assert(&sched_lock, MA_OWNED); CTR0(KTR_SMP, "forward_roundrobin()"); - if (!smp_started || !invltlb_ok || cold || panicstr) + if (!smp_started || cold || panicstr) return; if (!forward_roundrobin_enabled) return; - resched_cpus |= PCPU_GET(other_cpus); - map = PCPU_GET(other_cpus) & ~stopped_cpus ; -#if 1 - ipi_selected(map, IPI_AST); -#else - ipi_all_but_self(IPI_AST); -#endif - i = 0; - while ((checkstate_need_ast & map) != 0) { - /* spin */ - i++; - if (i > 100000) { -#if 0 - printf("forward_roundrobin: dropped ast 0x%x\n", - checkstate_need_ast & map); -#endif - break; + map = 0; + SLIST_FOREACH(gd, &cpuhead, gd_allcpu) { + p = gd->gd_curproc; + id = gd->gd_cpuid; + if (id != PCPU_GET(cpuid) && (id & stopped_cpus) == 0 && + p != gd->gd_idleproc) { + need_resched(p); + map |= id; } } + ipi_selected(map, IPI_AST); } /* @@ -2736,22 +216,27 @@ forward_roundrobin(void) int stop_cpus(u_int map) { - int count = 0; + int i; if (!smp_started) return 0; - /* send the Xcpustop IPI to all CPUs in map */ + CTR1(KTR_SMP, "stop_cpus(%x)", map); + + /* send the stop IPI to all CPUs in map */ ipi_selected(map, IPI_STOP); - while (count++ < 100000 && (stopped_cpus & map) != map) - /* spin */ ; - + i = 0; + while ((atomic_load_acq_int(&stopped_cpus) & map) != map) { + /* spin */ + i++; #ifdef DIAGNOSTIC - if ((stopped_cpus & map) != map) - printf("Warning: CPUs 0x%x did not stop!\n", - (~(stopped_cpus & map)) & map); + if (i == 100000) { + printf("timeout stopping cpus\n"); + break; + } #endif + } return 1; } @@ -2773,43 +258,22 @@ stop_cpus(u_int map) int restart_cpus(u_int map) { - int count = 0; if (!smp_started) return 0; - started_cpus = map; /* signal other cpus to restart */ + CTR1(KTR_SMP, "restart_cpus(%x)", map); - /* wait for each to clear its bit */ - while (count++ < 100000 && (stopped_cpus & map) != 0) - /* spin */ ; + /* signal other cpus to restart */ + atomic_store_rel_int(&started_cpus, map); -#ifdef DIAGNOSTIC - if ((stopped_cpus & map) != 0) - printf("Warning: CPUs 0x%x did not restart!\n", - (~(stopped_cpus & map)) & map); -#endif + /* wait for each to clear its bit */ + while ((atomic_load_acq_int(&stopped_cpus) & map) != 0) + ; /* nothing */ return 1; } - -#ifdef APIC_INTR_REORDER -/* - * Maintain mapping from softintr vector to isr bit in local apic. - */ -void -set_lapic_isrloc(int intr, int vector) -{ - if (intr < 0 || intr > 32) - panic("set_apic_isrloc: bad intr argument: %d",intr); - if (vector < ICU_OFFSET || vector > 255) - panic("set_apic_isrloc: bad vector argument: %d",vector); - apic_isrbit_location[intr].location = &lapic.isr0 + ((vector>>5)<<2); - apic_isrbit_location[intr].bit = (1<<(vector & 31)); -} -#endif - /* * All-CPU rendezvous. CPUs are signalled, all execute the setup function * (if specified), rendezvous, execute the action function (if specified), @@ -2819,29 +283,24 @@ set_lapic_isrloc(int intr, int vector) * Note that the supplied external functions _must_ be reentrant and aware * that they are running in parallel and in an unknown lock context. */ -static void (*smp_rv_setup_func)(void *arg); -static void (*smp_rv_action_func)(void *arg); -static void (*smp_rv_teardown_func)(void *arg); -static void *smp_rv_func_arg; -static volatile int smp_rv_waiters[2]; - void smp_rendezvous_action(void) { + /* setup function */ if (smp_rv_setup_func != NULL) smp_rv_setup_func(smp_rv_func_arg); /* spin on entry rendezvous */ atomic_add_int(&smp_rv_waiters[0], 1); - while (smp_rv_waiters[0] < mp_ncpus) - ; + while (atomic_load_acq_int(&smp_rv_waiters[0]) < mp_ncpus) + ; /* nothing */ /* action function */ if (smp_rv_action_func != NULL) smp_rv_action_func(smp_rv_func_arg); /* spin on exit rendezvous */ atomic_add_int(&smp_rv_waiters[1], 1); - while (smp_rv_waiters[1] < mp_ncpus) - ; + while (atomic_load_acq_int(&smp_rv_waiters[1]) < mp_ncpus) + ; /* nothing */ /* teardown function */ if (smp_rv_teardown_func != NULL) smp_rv_teardown_func(smp_rv_func_arg); @@ -2854,6 +313,16 @@ smp_rendezvous(void (* setup_func)(void *), void *arg) { + if (!smp_started) { + if (setup_func != NULL) + setup_func(arg); + if (action_func != NULL) + action_func(arg); + if (teardown_func != NULL) + teardown_func(arg); + return; + } + /* obtain rendezvous lock */ mtx_lock_spin(&smp_rv_mtx); @@ -2865,9 +334,7 @@ smp_rendezvous(void (* setup_func)(void *), smp_rv_waiters[0] = 0; smp_rv_waiters[1] = 0; - /* - * signal other processors, which will enter the IPI with interrupts off - */ + /* signal other processors, which will enter the IPI with interrupts off */ ipi_all_but_self(IPI_RENDEZVOUS); /* call executor function */ @@ -2876,55 +343,3 @@ smp_rendezvous(void (* setup_func)(void *), /* release lock */ mtx_unlock_spin(&smp_rv_mtx); } - -/* - * send an IPI to a set of cpus. - */ -void -ipi_selected(u_int32_t cpus, u_int ipi) -{ - - CTR2(KTR_SMP, __func__ ": cpus: %x ipi: %x", cpus, ipi); - selected_apic_ipi(cpus, ipi, APIC_DELMODE_FIXED); -} - -/* - * send an IPI INTerrupt containing 'vector' to all CPUs, including myself - */ -void -ipi_all(u_int ipi) -{ - - CTR1(KTR_SMP, __func__ ": ipi: %x", ipi); - apic_ipi(APIC_DEST_ALLISELF, ipi, APIC_DELMODE_FIXED); -} - -/* - * send an IPI to all CPUs EXCEPT myself - */ -void -ipi_all_but_self(u_int ipi) -{ - - CTR1(KTR_SMP, __func__ ": ipi: %x", ipi); - apic_ipi(APIC_DEST_ALLESELF, ipi, APIC_DELMODE_FIXED); -} - -/* - * send an IPI to myself - */ -void -ipi_self(u_int ipi) -{ - - CTR1(KTR_SMP, __func__ ": ipi: %x", ipi); - apic_ipi(APIC_DEST_SELF, ipi, APIC_DELMODE_FIXED); -} - -void -release_aps(void *dummy __unused) -{ - atomic_store_rel_int(&aps_ready, 1); -} - -SYSINIT(start_aps, SI_SUB_SMP, SI_ORDER_FIRST, release_aps, NULL); diff --git a/sys/kern/subr_trap.c b/sys/kern/subr_trap.c index 7088d0c..d9597df 100644 --- a/sys/kern/subr_trap.c +++ b/sys/kern/subr_trap.c @@ -177,7 +177,7 @@ userret(p, frame, oticks) mtx_lock_spin(&sched_lock); p->p_pri.pri_level = p->p_pri.pri_user; - if (resched_wanted()) { + if (resched_wanted(p)) { /* * Since we are curproc, clock will normally just change * our priority without moving us from one queue to another @@ -1277,7 +1277,7 @@ ast(framep) * acquiring and releasing mutexes in assembly is not fun. */ mtx_lock_spin(&sched_lock); - if (!(astpending(p) || resched_wanted())) { + if (!(astpending(p) || resched_wanted(p))) { mtx_unlock_spin(&sched_lock); return; } diff --git a/sys/pc98/i386/machdep.c b/sys/pc98/i386/machdep.c index 1068954..e4395d7 100644 --- a/sys/pc98/i386/machdep.c +++ b/sys/pc98/i386/machdep.c @@ -262,8 +262,6 @@ static struct trapframe proc0_tf; static struct globaldata __globaldata; #endif -struct cpuhead cpuhead; - struct mtx sched_lock; struct mtx Giant; @@ -454,16 +452,10 @@ again: bufinit(); vm_pager_bufferinit(); - SLIST_INIT(&cpuhead); - SLIST_INSERT_HEAD(&cpuhead, GLOBALDATA, gd_allcpu); - #ifdef SMP - /* - * OK, enough kmem_alloc/malloc state should be up, lets get on with it! - */ - mp_start(); /* fire up the APs and APICs */ - mp_announce(); -#endif /* SMP */ + globaldata_register(GLOBALDATA); +#endif + cpu_setregs(); } diff --git a/sys/pc98/pc98/machdep.c b/sys/pc98/pc98/machdep.c index 1068954..e4395d7 100644 --- a/sys/pc98/pc98/machdep.c +++ b/sys/pc98/pc98/machdep.c @@ -262,8 +262,6 @@ static struct trapframe proc0_tf; static struct globaldata __globaldata; #endif -struct cpuhead cpuhead; - struct mtx sched_lock; struct mtx Giant; @@ -454,16 +452,10 @@ again: bufinit(); vm_pager_bufferinit(); - SLIST_INIT(&cpuhead); - SLIST_INSERT_HEAD(&cpuhead, GLOBALDATA, gd_allcpu); - #ifdef SMP - /* - * OK, enough kmem_alloc/malloc state should be up, lets get on with it! - */ - mp_start(); /* fire up the APs and APICs */ - mp_announce(); -#endif /* SMP */ + globaldata_register(GLOBALDATA); +#endif + cpu_setregs(); } diff --git a/sys/posix4/ksched.c b/sys/posix4/ksched.c index b729ccf..dbdc174 100644 --- a/sys/posix4/ksched.c +++ b/sys/posix4/ksched.c @@ -175,7 +175,7 @@ int ksched_setscheduler(register_t *ret, struct ksched *ksched, mtx_lock_spin(&sched_lock); rtp_to_pri(&rtp, &p->p_pri); - need_resched(); + need_resched(p); mtx_unlock_spin(&sched_lock); } else @@ -197,7 +197,7 @@ int ksched_setscheduler(register_t *ret, struct ksched *ksched, * on the scheduling code: You must leave the * scheduling info alone. */ - need_resched(); + need_resched(p); mtx_unlock_spin(&sched_lock); } break; @@ -216,7 +216,7 @@ int ksched_getscheduler(register_t *ret, struct ksched *ksched, struct proc *p) int ksched_yield(register_t *ret, struct ksched *ksched) { mtx_lock_spin(&sched_lock); - need_resched(); + need_resched(curproc); mtx_unlock_spin(&sched_lock); return 0; } diff --git a/sys/powerpc/include/globaldata.h b/sys/powerpc/include/globaldata.h index 15bd8c41a..4d6d02f 100644 --- a/sys/powerpc/include/globaldata.h +++ b/sys/powerpc/include/globaldata.h @@ -66,14 +66,7 @@ struct globaldata { #endif }; -SLIST_HEAD(cpuhead, globaldata); -extern struct cpuhead cpuhead; - void globaldata_init(struct globaldata *pcpu, int cpuid, size_t sz); -struct globaldata *globaldata_find(int cpuid); -#ifdef SMP -void globaldata_register(struct globaldata *pcpu); -#endif #endif /* _KERNEL */ diff --git a/sys/powerpc/include/pcpu.h b/sys/powerpc/include/pcpu.h index 15bd8c41a..4d6d02f 100644 --- a/sys/powerpc/include/pcpu.h +++ b/sys/powerpc/include/pcpu.h @@ -66,14 +66,7 @@ struct globaldata { #endif }; -SLIST_HEAD(cpuhead, globaldata); -extern struct cpuhead cpuhead; - void globaldata_init(struct globaldata *pcpu, int cpuid, size_t sz); -struct globaldata *globaldata_find(int cpuid); -#ifdef SMP -void globaldata_register(struct globaldata *pcpu); -#endif #endif /* _KERNEL */ diff --git a/sys/powerpc/powerpc/mp_machdep.c b/sys/powerpc/powerpc/mp_machdep.c index e8c65f3..c67f4aa 100644 --- a/sys/powerpc/powerpc/mp_machdep.c +++ b/sys/powerpc/powerpc/mp_machdep.c @@ -34,6 +34,7 @@ #include <sys/malloc.h> #include <sys/mutex.h> #include <sys/kernel.h> +#include <sys/smp.h> #include <sys/sysctl.h> #include <vm/vm.h> @@ -42,7 +43,6 @@ #include <sys/user.h> #include <sys/dkstat.h> -#include <machine/smp.h> #include <machine/atomic.h> #include <machine/globaldata.h> #include <machine/pmap.h> diff --git a/sys/sys/proc.h b/sys/sys/proc.h index aa496e0..d1115d8 100644 --- a/sys/sys/proc.h +++ b/sys/sys/proc.h @@ -382,16 +382,16 @@ sigonstack(size_t sp) * Preempt the current process if in interrupt from user mode, * or after the current trap/syscall if in system mode. */ -#define need_resched() do { \ +#define need_resched(p) do { \ mtx_assert(&sched_lock, MA_OWNED); \ - curproc->p_sflag |= PS_NEEDRESCHED; \ + (p)->p_sflag |= PS_NEEDRESCHED; \ } while (0) -#define resched_wanted() (curproc->p_sflag & PS_NEEDRESCHED) +#define resched_wanted(p) ((p)->p_sflag & PS_NEEDRESCHED) -#define clear_resched() do { \ +#define clear_resched(p) do { \ mtx_assert(&sched_lock, MA_OWNED); \ - curproc->p_sflag &= ~PS_NEEDRESCHED; \ + (p)->p_sflag &= ~PS_NEEDRESCHED; \ } while (0) /* diff --git a/sys/sys/smp.h b/sys/sys/smp.h index 9f9149e..dffd914 100644 --- a/sys/sys/smp.h +++ b/sys/sys/smp.h @@ -7,152 +7,61 @@ * ---------------------------------------------------------------------------- * * $FreeBSD$ - * */ -#ifndef _MACHINE_SMP_H_ -#define _MACHINE_SMP_H_ +#ifndef _SYS_SMP_H_ +#define _SYS_SMP_H_ #ifdef _KERNEL +#include <machine/globaldata.h> +#include <machine/smp.h> -#if defined(SMP) && defined(I386_CPU) && !defined(COMPILING_LINT) -#error SMP not supported with I386_CPU -#endif -#if defined(SMP) && !defined(APIC_IO) -# error APIC_IO required for SMP, add "options APIC_IO" to your config file. -#endif /* SMP && !APIC_IO */ +#ifndef LOCORE -#if defined(SMP) || defined(APIC_IO) +#ifdef SMP +SLIST_HEAD(cpuhead, globaldata); -#ifndef LOCORE +extern struct cpuhead cpuhead; +extern void (*cpustop_restartfunc)(void); +extern int mp_ncpus; +extern int smp_active; +extern int smp_started; +extern int smp_cpus; +extern u_int all_cpus; +extern volatile u_int started_cpus; +extern volatile u_int stopped_cpus; /* - * For sending values to POST displays. - * XXX FIXME: where does this really belong, isa.h/isa.c perhaps? + * Machine dependent functions used to initialize MP support. + * + * The cpu_mp_probe() should check to see if MP support is present and return + * zero if it is not or non-zero if it is. If MP support is present, then + * cpu_mp_start() will be called so that MP can be enabled. This function + * should do things such as startup secondary processors. It should also + * setup mp_ncpus, all_cpus, and smp_cpus. It should also ensure that + * smp_active and smp_started are initialized at the appropriate time. + * Once cpu_mp_start() returns, machine independent MP startup code will be + * executed and a simple message will be output to the console. Finally, + * cpu_mp_announce() will be called so that machine dependent messages about + * the MP support may be output to the console if desired. */ -extern int current_postcode; /** XXX currently in mp_machdep.c */ -#define POSTCODE(X) current_postcode = (X), \ - outb(0x80, current_postcode) -#define POSTCODE_LO(X) current_postcode &= 0xf0, \ - current_postcode |= ((X) & 0x0f), \ - outb(0x80, current_postcode) -#define POSTCODE_HI(X) current_postcode &= 0x0f, \ - current_postcode |= (((X) << 4) & 0xf0), \ - outb(0x80, current_postcode) - - -#include <machine/apic.h> - -/* global data in mpboot.s */ -extern int bootMP_size; - -/* functions in mpboot.s */ -void bootMP __P((void)); - -/* global data in apic_vector.s */ -extern volatile u_int stopped_cpus; -extern volatile u_int started_cpus; - -extern volatile u_int checkstate_probed_cpus; -extern volatile u_int checkstate_need_ast; -extern volatile u_int resched_cpus; -extern void (*cpustop_restartfunc) __P((void)); - -/* global data in mp_machdep.c */ -extern int bsp_apic_ready; -extern int mp_ncpus; -extern int mp_naps; -extern int mp_nbusses; -extern int mp_napics; -extern int mp_picmode; -extern int boot_cpu_id; -extern vm_offset_t cpu_apic_address; -extern vm_offset_t io_apic_address[]; -extern u_int32_t cpu_apic_versions[]; -extern u_int32_t *io_apic_versions; -extern int cpu_num_to_apic_id[]; -extern int io_num_to_apic_id[]; -extern int apic_id_to_logical[]; -#define APIC_INTMAPSIZE 32 -struct apic_intmapinfo { - int ioapic; - int int_pin; - volatile void *apic_address; - int redirindex; -}; -extern struct apic_intmapinfo int_to_apicintpin[]; -extern u_int all_cpus; -extern struct pcb stoppcbs[]; - -/* functions in mp_machdep.c */ -u_int mp_bootaddress __P((u_int)); -int mp_probe __P((void)); -void mp_start __P((void)); -void mp_announce __P((void)); -u_int isa_apic_mask __P((u_int)); -int isa_apic_irq __P((int)); -int pci_apic_irq __P((int, int, int)); -int apic_irq __P((int, int)); -int next_apic_irq __P((int)); -int undirect_isa_irq __P((int)); -int undirect_pci_irq __P((int)); -int apic_bus_type __P((int)); -int apic_src_bus_id __P((int, int)); -int apic_src_bus_irq __P((int, int)); -int apic_int_type __P((int, int)); -int apic_trigger __P((int, int)); -int apic_polarity __P((int, int)); -void assign_apic_irq __P((int apic, int intpin, int irq)); -void revoke_apic_irq __P((int irq)); -void bsp_apic_configure __P((void)); -void init_secondary __P((void)); -void smp_invltlb __P((void)); -int stop_cpus __P((u_int)); -int restart_cpus __P((u_int)); -void forward_statclock __P((int pscnt)); -void forward_hardclock __P((int pscnt)); -void forward_signal __P((struct proc *)); -void forward_roundrobin __P((void)); -#ifdef APIC_INTR_REORDER -void set_lapic_isrloc __P((int, int)); -#endif /* APIC_INTR_REORDER */ -void smp_rendezvous_action __P((void)); -void smp_rendezvous __P((void (*)(void *), - void (*)(void *), - void (*)(void *), - void *arg)); - -/* global data in mpapic.c */ -extern volatile lapic_t lapic; -extern volatile ioapic_t **ioapic; - -/* functions in mpapic.c */ -void apic_dump __P((char*)); -void apic_initialize __P((void)); -void imen_dump __P((void)); -int apic_ipi __P((int, int, int)); -int selected_apic_ipi __P((u_int, int, int)); -int io_apic_setup __P((int)); -void io_apic_setup_intpin __P((int, int)); -void io_apic_set_id __P((int, int)); -int io_apic_get_id __P((int)); -int ext_int_setup __P((int, int)); - -void set_apic_timer __P((int)); -int read_apic_timer __P((void)); -void u_sleep __P((int)); -u_int io_apic_read __P((int, int)); -void io_apic_write __P((int, int, u_int)); - - -/* global data in init_smp.c */ -extern int invltlb_ok; -extern int smp_active; -extern int smp_started; -extern volatile int smp_idle_loops; - +void cpu_mp_announce(void); +int cpu_mp_probe(void); +void cpu_mp_start(void); + +void forward_signal(struct proc *); +void forward_roundrobin(void); +void globaldata_register(struct globaldata *gd); +struct globaldata *globaldata_find(u_int cpuid); +int restart_cpus(u_int); +int stop_cpus(u_int); +void smp_rendezvous_action(void); +void smp_rendezvous(void (*)(void *), + void (*)(void *), + void (*)(void *), + void *arg); + +#endif /* SMP */ #endif /* !LOCORE */ -#endif /* SMP && !APIC_IO */ - #endif /* _KERNEL */ -#endif /* _MACHINE_SMP_H_ */ +#endif /* _SYS_SMP_H_ */ diff --git a/sys/sys/systm.h b/sys/sys/systm.h index ef3da27..4f6042d 100644 --- a/sys/sys/systm.h +++ b/sys/sys/systm.h @@ -152,8 +152,10 @@ int susword __P((void *base, int word)); void realitexpire __P((void *)); void hardclock __P((struct clockframe *frame)); +void hardclock_process __P((struct proc *p, int user)); void softclock __P((void *)); void statclock __P((struct clockframe *frame)); +void statclock_process __P((struct proc *p, register_t pc, int user)); void startprofclock __P((struct proc *)); void stopprofclock __P((struct proc *)); |
