diff options
Diffstat (limited to 'sys/vm/vm_glue.c')
-rw-r--r-- | sys/vm/vm_glue.c | 500 |
1 files changed, 290 insertions, 210 deletions
diff --git a/sys/vm/vm_glue.c b/sys/vm/vm_glue.c index 5676ff3..f181ab0 100644 --- a/sys/vm/vm_glue.c +++ b/sys/vm/vm_glue.c @@ -67,16 +67,22 @@ #include <sys/buf.h> #include <sys/user.h> +#include <sys/kernel.h> +#include <sys/dkstat.h> + #include <vm/vm.h> #include <vm/vm_page.h> +#include <vm/vm_pageout.h> #include <vm/vm_kern.h> -#include <machine/cpu.h> +#include <machine/stdarg.h> +extern char kstack[]; int avefree = 0; /* XXX */ -unsigned maxdmap = MAXDSIZ; /* XXX */ int readbuffers = 0; /* XXX allow kgdb to read kernel buffer pool */ +/* vm_map_t upages_map; */ +void swapout(struct proc *p); int kernacc(addr, len, rw) caddr_t addr; @@ -89,18 +95,6 @@ kernacc(addr, len, rw) saddr = trunc_page(addr); eaddr = round_page(addr+len); rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot); - /* - * XXX there are still some things (e.g. the buffer cache) that - * are managed behind the VM system's back so even though an - * address is accessible in the mind of the VM system, there may - * not be physical pages where the VM thinks there is. This can - * lead to bogus allocation of pages in the kernel address space - * or worse, inconsistencies at the pmap level. We only worry - * about the buffer cache for now. - */ - if (!readbuffers && rv && (eaddr > (vm_offset_t)buffers && - saddr < (vm_offset_t)buffers + MAXBSIZE * nbuf)) - rv = FALSE; return(rv == TRUE); } @@ -112,6 +106,23 @@ useracc(addr, len, rw) boolean_t rv; vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE; + /* + * XXX - specially disallow access to user page tables - they are + * in the map. + * + * XXX - don't specially disallow access to the user area - treat + * it as incorrectly as elsewhere. + * + * XXX - VM_MAXUSER_ADDRESS is an end address, not a max. It was + * only used (as an end address) in trap.c. Use it as an end + * address here too. + */ + if ((vm_offset_t) addr >= VM_MAXUSER_ADDRESS + || (vm_offset_t) addr + len > VM_MAXUSER_ADDRESS + || (vm_offset_t) addr + len <= (vm_offset_t) addr) { + return (FALSE); + } + rv = vm_map_check_protection(&curproc->p_vmspace->vm_map, trunc_page(addr), round_page(addr+len), prot); return(rv == TRUE); @@ -121,40 +132,18 @@ useracc(addr, len, rw) /* * Change protections on kernel pages from addr to addr+len * (presumably so debugger can plant a breakpoint). - * - * We force the protection change at the pmap level. If we were - * to use vm_map_protect a change to allow writing would be lazily- - * applied meaning we would still take a protection fault, something - * we really don't want to do. It would also fragment the kernel - * map unnecessarily. We cannot use pmap_protect since it also won't - * enforce a write-enable request. Using pmap_enter is the only way - * we can ensure the change takes place properly. + * All addresses are assumed to reside in the Sysmap, */ -void chgkprot(addr, len, rw) register caddr_t addr; int len, rw; { - vm_prot_t prot; - vm_offset_t pa, sva, eva; - - prot = rw == B_READ ? VM_PROT_READ : VM_PROT_READ|VM_PROT_WRITE; - eva = round_page(addr + len); - for (sva = trunc_page(addr); sva < eva; sva += PAGE_SIZE) { - /* - * Extract physical address for the page. - * We use a cheezy hack to differentiate physical - * page 0 from an invalid mapping, not that it - * really matters... - */ - pa = pmap_extract(kernel_pmap, sva|1); - if (pa == 0) - panic("chgkprot: invalid page"); - pmap_enter(kernel_pmap, sva, pa&~1, prot, TRUE); - } + vm_prot_t prot = rw == B_READ ? VM_PROT_READ : VM_PROT_WRITE; + + vm_map_protect(kernel_map, trunc_page(addr), + round_page(addr+len), prot, FALSE); } #endif - void vslock(addr, len) caddr_t addr; @@ -172,8 +161,8 @@ vsunlock(addr, len, dirtied) { #ifdef lint dirtied++; -#endif - vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page(addr), +#endif lint + vm_map_pageable(&curproc->p_vmspace->vm_map, trunc_page(addr), round_page(addr+len), TRUE); } @@ -194,16 +183,19 @@ vm_fork(p1, p2, isvfork) int isvfork; { register struct user *up; - vm_offset_t addr; + vm_offset_t addr, ptaddr; + int i; + struct vm_map *vp; + + while( cnt.v_free_count < cnt.v_free_min) + VM_WAIT; -#ifdef i386 /* * avoid copying any of the parent's pagetables or other per-process * objects that reside in the map by marking all of them non-inheritable */ (void)vm_map_inherit(&p1->p_vmspace->vm_map, - UPT_MIN_ADDRESS-UPAGES*NBPG, VM_MAX_ADDRESS, VM_INHERIT_NONE); -#endif + UPT_MIN_ADDRESS - UPAGES * NBPG, VM_MAX_ADDRESS, VM_INHERIT_NONE); p2->p_vmspace = vmspace_fork(p1->p_vmspace); #ifdef SYSVSHM @@ -211,23 +203,40 @@ vm_fork(p1, p2, isvfork) shmfork(p1, p2, isvfork); #endif -#ifndef i386 /* * Allocate a wired-down (for now) pcb and kernel stack for the process */ - addr = kmem_alloc_pageable(kernel_map, ctob(UPAGES)); - if (addr == 0) - panic("vm_fork: no more kernel virtual memory"); - vm_map_pageable(kernel_map, addr, addr + ctob(UPAGES), FALSE); -#else -/* XXX somehow, on 386, ocassionally pageout removes active, wired down kstack, -and pagetables, WITHOUT going thru vm_page_unwire! Why this appears to work is -not yet clear, yet it does... */ - addr = kmem_alloc(kernel_map, ctob(UPAGES)); - if (addr == 0) - panic("vm_fork: no more kernel virtual memory"); -#endif - up = (struct user *)addr; + + addr = (vm_offset_t) kstack; + + vp = &p2->p_vmspace->vm_map; + + /* ream out old pagetables and kernel stack */ + (void)vm_deallocate(vp, addr, UPT_MAX_ADDRESS - addr); + + /* get new pagetables and kernel stack */ + (void)vm_allocate(vp, &addr, UPT_MAX_ADDRESS - addr, FALSE); + + /* force in the page table encompassing the UPAGES */ + ptaddr = trunc_page((u_int)vtopte(addr)); + vm_map_pageable(vp, ptaddr, ptaddr + NBPG, FALSE); + + /* and force in (demand-zero) the UPAGES */ + vm_map_pageable(vp, addr, addr + UPAGES * NBPG, FALSE); + + /* get a kernel virtual address for the UPAGES for this proc */ + up = (struct user *)kmem_alloc_pageable(kernel_map, UPAGES * NBPG); + + /* and force-map the upages into the kernel pmap */ + for (i = 0; i < UPAGES; i++) + pmap_enter(vm_map_pmap(kernel_map), + ((vm_offset_t) up) + NBPG * i, + pmap_extract(vp->pmap, addr + NBPG * i), + VM_PROT_READ|VM_PROT_WRITE, 1); + + /* and allow the UPAGES page table entry to be paged (at the vm system level) */ + vm_map_pageable(vp, ptaddr, ptaddr + NBPG, TRUE); + p2->p_addr = up; /* @@ -246,15 +255,7 @@ not yet clear, yet it does... */ ((caddr_t)&up->u_stats.pstat_endcopy - (caddr_t)&up->u_stats.pstat_startcopy)); -#ifdef i386 - { u_int addr = UPT_MIN_ADDRESS - UPAGES*NBPG; struct vm_map *vp; - - vp = &p2->p_vmspace->vm_map; - (void)vm_deallocate(vp, addr, UPT_MAX_ADDRESS - addr); - (void)vm_allocate(vp, &addr, UPT_MAX_ADDRESS - addr, FALSE); - (void)vm_map_inherit(vp, addr, UPT_MAX_ADDRESS, VM_INHERIT_NONE); - } -#endif + /* * cpu_fork will copy and update the kernel stack and pcb, * and make the child ready to run. It marks the child @@ -273,6 +274,7 @@ void vm_init_limits(p) register struct proc *p; { + int tmp; /* * Set up the initial limits on process VM. @@ -285,11 +287,13 @@ vm_init_limits(p) p->p_rlimit[RLIMIT_STACK].rlim_max = MAXSSIZ; p->p_rlimit[RLIMIT_DATA].rlim_cur = DFLDSIZ; p->p_rlimit[RLIMIT_DATA].rlim_max = MAXDSIZ; - p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(cnt.v_free_count); + tmp = ((2 * cnt.v_free_count) / 3) - 32; + if (cnt.v_free_count < 512) + tmp = cnt.v_free_count; + p->p_rlimit[RLIMIT_RSS].rlim_cur = ptoa(tmp); + p->p_rlimit[RLIMIT_RSS].rlim_max = RLIM_INFINITY; } -#include <vm/vm_pageout.h> - #ifdef DEBUG int enableswap = 1; int swapdebug = 0; @@ -298,12 +302,67 @@ int swapdebug = 0; #define SDB_SWAPOUT 4 #endif +void +faultin(p) +struct proc *p; +{ + vm_offset_t i; + vm_offset_t vaddr, ptaddr; + vm_offset_t v, v1; + struct user *up; + int s; + int opflag; + + if ((p->p_flag & P_INMEM) == 0) { + int rv0, rv1; + vm_map_t map; + + ++p->p_lock; + + map = &p->p_vmspace->vm_map; + /* force the page table encompassing the kernel stack (upages) */ + ptaddr = trunc_page((u_int)vtopte(kstack)); + vm_map_pageable(map, ptaddr, ptaddr + NBPG, FALSE); + + /* wire in the UPAGES */ + vm_map_pageable(map, (vm_offset_t) kstack, + (vm_offset_t) kstack + UPAGES * NBPG, FALSE); + + /* and map them nicely into the kernel pmap */ + for (i = 0; i < UPAGES; i++) { + vm_offset_t off = i * NBPG; + vm_offset_t pa = (vm_offset_t) + pmap_extract(&p->p_vmspace->vm_pmap, + (vm_offset_t) kstack + off); + pmap_enter(vm_map_pmap(kernel_map), + ((vm_offset_t)p->p_addr) + off, + pa, VM_PROT_READ|VM_PROT_WRITE, 1); + } + + /* and let the page table pages go (at least above pmap level) */ + vm_map_pageable(map, ptaddr, ptaddr + NBPG, TRUE); + + s = splhigh(); + + if (p->p_stat == SRUN) + setrunqueue(p); + + p->p_flag |= P_INMEM; + + /* undo the effect of setting SLOCK above */ + --p->p_lock; + splx(s); + + } + +} + +int swapinreq; +int percentactive; /* - * Brutally simple: - * 1. Attempt to swapin every swaped-out, runnable process in - * order of priority. - * 2. If not enough memory, wake the pageout daemon and let it - * clear some space. + * This swapin algorithm attempts to swap-in processes only if there + * is enough space for them. Of course, if a process waits for a long + * time, it will be swapped in anyway. */ void scheduler() @@ -313,88 +372,104 @@ scheduler() struct proc *pp; int ppri; vm_offset_t addr; - vm_size_t size; + int lastidle, lastrun; + int curidle, currun; + int forceload; + int percent; + int ntries; + + lastidle = 0; + lastrun = 0; loop: -#ifdef DEBUG - while (!enableswap) - sleep((caddr_t)&proc0, PVM); -#endif + ntries = 0; + vmmeter(); + + curidle = cp_time[CP_IDLE]; + currun = cp_time[CP_USER] + cp_time[CP_SYS] + cp_time[CP_NICE]; + percent = (100*(currun-lastrun)) / ( 1 + (currun-lastrun) + (curidle-lastidle)); + lastrun = currun; + lastidle = curidle; + if( percent > 100) + percent = 100; + percentactive = percent; + + if( percentactive < 25) + forceload = 1; + else + forceload = 0; + +loop1: pp = NULL; ppri = INT_MIN; for (p = (struct proc *)allproc; p != NULL; p = p->p_next) { if (p->p_stat == SRUN && (p->p_flag & P_INMEM) == 0) { + int mempri; pri = p->p_swtime + p->p_slptime - p->p_nice * 8; - if (pri > ppri) { + mempri = pri > 0 ? pri : 0; + /* + * if this process is higher priority and there is + * enough space, then select this process instead + * of the previous selection. + */ + if (pri > ppri && + (((cnt.v_free_count + (mempri * (4*PAGE_SIZE) / PAGE_SIZE) >= (p->p_vmspace->vm_swrss)) || (ntries > 0 && forceload)))) { pp = p; ppri = pri; } } } -#ifdef DEBUG - if (swapdebug & SDB_FOLLOW) - printf("sched: running, procp %x pri %d\n", pp, ppri); -#endif + + if ((pp == NULL) && (ntries == 0) && forceload) { + ++ntries; + goto loop1; + } + /* * Nothing to do, back to sleep */ if ((p = pp) == NULL) { - sleep((caddr_t)&proc0, PVM); + tsleep((caddr_t)&proc0, PVM, "sched", 0); goto loop; } /* - * We would like to bring someone in. - * This part is really bogus cuz we could deadlock on memory - * despite our feeble check. + * We would like to bring someone in. (only if there is space). */ - size = round_page(ctob(UPAGES)); - addr = (vm_offset_t) p->p_addr; - if (cnt.v_free_count > atop(size)) { -#ifdef DEBUG - if (swapdebug & SDB_SWAPIN) - printf("swapin: pid %d(%s)@%x, pri %d free %d\n", - p->p_pid, p->p_comm, p->p_addr, - ppri, cnt.v_free_count); -#endif - vm_map_pageable(kernel_map, addr, addr+size, FALSE); - /* - * Some architectures need to be notified when the - * user area has moved to new physical page(s) (e.g. - * see pmax/pmax/vm_machdep.c). - */ - cpu_swapin(p); - (void) splstatclock(); - if (p->p_stat == SRUN) - setrunqueue(p); - p->p_flag |= P_INMEM; - (void) spl0(); +/* + printf("swapin: %d, free: %d, res: %d, min: %d\n", + p->p_pid, cnt.v_free_count, cnt.v_free_reserved, cnt.v_free_min); +*/ + (void) splhigh(); + if ((forceload && (cnt.v_free_count > (cnt.v_free_reserved + UPAGES + 1))) || + (cnt.v_free_count >= cnt.v_free_min)) { + spl0(); + faultin(p); p->p_swtime = 0; goto loop; - } + } + /* + * log the memory shortage + */ + swapinreq += p->p_vmspace->vm_swrss; /* * Not enough memory, jab the pageout daemon and wait til the * coast is clear. */ -#ifdef DEBUG - if (swapdebug & SDB_FOLLOW) - printf("sched: no room for pid %d(%s), free %d\n", - p->p_pid, p->p_comm, cnt.v_free_count); -#endif - (void) splhigh(); - VM_WAIT; + if( cnt.v_free_count < cnt.v_free_min) { + VM_WAIT; + } else { + tsleep((caddr_t)&proc0, PVM, "sched", 0); + } (void) spl0(); -#ifdef DEBUG - if (swapdebug & SDB_FOLLOW) - printf("sched: room again, free %d\n", cnt.v_free_count); -#endif goto loop; } -#define swappable(p) \ - (((p)->p_flag & \ - (P_SYSTEM | P_INMEM | P_NOSWAP | P_WEXIT | P_PHYSIO)) == P_INMEM) +#define swappable(p) \ + (((p)->p_lock == 0) && \ + ((p)->p_flag & (P_TRACED|P_NOSWAP|P_SYSTEM|P_INMEM|P_WEXIT|P_PHYSIO)) == P_INMEM) +extern int vm_pageout_free_min; /* * Swapout is driven by the pageout daemon. Very simple, we find eligible * procs and unwire their u-areas. We try to always "swap" at least one @@ -409,54 +484,86 @@ swapout_threads() register struct proc *p; struct proc *outp, *outp2; int outpri, outpri2; + int tpri; int didswap = 0; + int swapneeded = swapinreq; extern int maxslp; + int runnablenow; + int s; -#ifdef DEBUG - if (!enableswap) - return; -#endif +swapmore: + runnablenow = 0; outp = outp2 = NULL; - outpri = outpri2 = 0; + outpri = outpri2 = INT_MIN; for (p = (struct proc *)allproc; p != NULL; p = p->p_next) { if (!swappable(p)) continue; switch (p->p_stat) { case SRUN: - if (p->p_swtime > outpri2) { + ++runnablenow; + /* + * count the process as being in a runnable state + */ + if ((tpri = p->p_swtime + p->p_nice * 8) > outpri2) { outp2 = p; - outpri2 = p->p_swtime; + outpri2 = tpri; } continue; case SSLEEP: case SSTOP: - if (p->p_slptime >= maxslp) { + /* + * do not swapout a process that is waiting for VM datastructures + * there is a possible deadlock. + */ + if (!lock_try_write( &p->p_vmspace->vm_map.lock)) { + continue; + } + vm_map_unlock( &p->p_vmspace->vm_map); + if (p->p_slptime > maxslp) { swapout(p); didswap++; - } else if (p->p_slptime > outpri) { + } else if ((tpri = p->p_slptime + p->p_nice * 8) > outpri) { outp = p; - outpri = p->p_slptime; + outpri = tpri ; } continue; } } /* - * If we didn't get rid of any real duds, toss out the next most - * likely sleeping/stopped or running candidate. We only do this - * if we are real low on memory since we don't gain much by doing - * it (UPAGES pages). + * We swapout only if there are more than two runnable processes or if + * another process needs some space to swapin. */ - if (didswap == 0 && - cnt.v_free_count <= atop(round_page(ctob(UPAGES)))) { - if ((p = outp) == 0) - p = outp2; -#ifdef DEBUG - if (swapdebug & SDB_SWAPOUT) - printf("swapout_threads: no duds, try procp %x\n", p); -#endif - if (p) + if ((swapinreq || ((percentactive > 90) && (runnablenow > 2))) && + (((cnt.v_free_count + cnt.v_inactive_count) <= (cnt.v_free_target + cnt.v_inactive_target)) || + (cnt.v_free_count < cnt.v_free_min))) { + if ((p = outp) == 0) { + p = outp2; + } + + if (p) { swapout(p); + didswap = 1; + } + } + + /* + * if we previously had found a process to swapout, and we need to swapout + * more then try again. + */ +#if 0 + if( p && swapinreq) + goto swapmore; +#endif + + /* + * If we swapped something out, and another process needed memory, + * then wakeup the sched process. + */ + if (didswap) { + if (swapneeded) + wakeup((caddr_t)&proc0); + swapinreq = 0; } } @@ -465,59 +572,37 @@ swapout(p) register struct proc *p; { vm_offset_t addr; - vm_size_t size; + struct pmap *pmap = &p->p_vmspace->vm_pmap; + vm_map_t map = &p->p_vmspace->vm_map; + vm_offset_t ptaddr; + int i; -#ifdef DEBUG - if (swapdebug & SDB_SWAPOUT) - printf("swapout: pid %d(%s)@%x, stat %x pri %d free %d\n", - p->p_pid, p->p_comm, p->p_addr, p->p_stat, - p->p_slptime, cnt.v_free_count); -#endif - size = round_page(ctob(UPAGES)); - addr = (vm_offset_t) p->p_addr; -#if defined(hp300) || defined(luna68k) + ++p->p_stats->p_ru.ru_nswap; /* - * Ugh! u-area is double mapped to a fixed address behind the - * back of the VM system and accesses are usually through that - * address rather than the per-process address. Hence reference - * and modify information are recorded at the fixed address and - * lost at context switch time. We assume the u-struct and - * kernel stack are always accessed/modified and force it to be so. + * remember the process resident count */ - { - register int i; - volatile long tmp; - - for (i = 0; i < UPAGES; i++) { - tmp = *(long *)addr; *(long *)addr = tmp; - addr += NBPG; - } - addr = (vm_offset_t) p->p_addr; - } -#endif -#ifdef mips + p->p_vmspace->vm_swrss = + p->p_vmspace->vm_pmap.pm_stats.resident_count; /* - * Be sure to save the floating point coprocessor state before - * paging out the u-struct. + * and decrement the amount of needed space */ - { - extern struct proc *machFPCurProcPtr; + swapinreq -= min(swapinreq, p->p_vmspace->vm_pmap.pm_stats.resident_count); - if (p == machFPCurProcPtr) { - MachSaveCurFPState(p); - machFPCurProcPtr = (struct proc *)0; - } - } -#endif -#ifndef i386 /* temporary measure till we find spontaineous unwire of kstack */ - vm_map_pageable(kernel_map, addr, addr+size, TRUE); - pmap_collect(vm_map_pmap(&p->p_vmspace->vm_map)); -#endif (void) splhigh(); p->p_flag &= ~P_INMEM; if (p->p_stat == SRUN) remrq(p); (void) spl0(); + + ++p->p_lock; +/* let the upages be paged */ + pmap_remove(vm_map_pmap(kernel_map), + (vm_offset_t) p->p_addr, ((vm_offset_t) p->p_addr) + UPAGES * NBPG); + + vm_map_pageable(map, (vm_offset_t) kstack, + (vm_offset_t) kstack + UPAGES * NBPG, TRUE); + + --p->p_lock; p->p_swtime = 0; } @@ -525,6 +610,7 @@ swapout(p) * The rest of these routines fake thread handling */ +#ifndef assert_wait void assert_wait(event, ruptible) int event; @@ -535,44 +621,38 @@ assert_wait(event, ruptible) #endif curproc->p_thread = event; } +#endif void -thread_block() +thread_block(char *msg) { - int s = splhigh(); - if (curproc->p_thread) - sleep((caddr_t)curproc->p_thread, PVM); - splx(s); + tsleep((caddr_t)curproc->p_thread, PVM, msg, 0); } + void -thread_sleep(event, lock, ruptible) +thread_sleep_(event, lock, wmesg) int event; simple_lock_t lock; - boolean_t ruptible; + char *wmesg; { -#ifdef lint - ruptible++; -#endif - int s = splhigh(); curproc->p_thread = event; simple_unlock(lock); - if (curproc->p_thread) - sleep((caddr_t)event, PVM); - splx(s); + if (curproc->p_thread) { + tsleep((caddr_t)event, PVM, wmesg, 0); + } } +#ifndef thread_wakeup void thread_wakeup(event) int event; { - int s = splhigh(); - wakeup((caddr_t)event); - splx(s); } +#endif /* * DEBUG stuff |