With retirement of cpumask_t and usage of cpuset_t for representing a

mask of CPUs, pc_other_cpus and pc_cpumask become highly inefficient.

Remove them and replace their usage with custom pc_cpuid magic (as,
atm, pc_cpumask can be easilly represented by (1 << pc_cpuid) and
pc_other_cpus by (all_cpus & ~(1 << pc_cpuid))).

This change is not targeted for MFC because of struct pcpu members
removal and dependency by cpumask_t retirement.

MD review by:	marcel, marius, alc
Tested by:	pluknet
MD testing by:	marcel, marius, gonzo, andreast

5 files changed, 41 insertions, 43 deletions

diff --git a/sys/kern/kern_rmlock.c b/sys/kern/kern_rmlock.c
index 3214e1b..1c7337d 100644
--- a/sys/kern/kern_rmlock.c
+++ b/sys/kern/kern_rmlock.c
@@ -263,7 +263,7 @@ _rm_rlock_hard(struct rmlock *rm, struct rm_priotracker *tracker, int trylock)
 	pc = pcpu_find(curcpu);
 
 	/* Check if we just need to do a proper critical_exit. */
-	if (!CPU_OVERLAP(&pc->pc_cpumask, &rm->rm_writecpus)) {
+	if (!CPU_ISSET(pc->pc_cpuid, &rm->rm_writecpus)) {
 		critical_exit();
 		return (1);
 	}
@@ -325,7 +325,7 @@ _rm_rlock_hard(struct rmlock *rm, struct rm_priotracker *tracker, int trylock)
 
 	critical_enter();
 	pc = pcpu_find(curcpu);
-	CPU_NAND(&rm->rm_writecpus, &pc->pc_cpumask);
+	CPU_CLR(pc->pc_cpuid, &rm->rm_writecpus);
 	rm_tracker_add(pc, tracker);
 	sched_pin();
 	critical_exit();
@@ -367,7 +367,7 @@ _rm_rlock(struct rmlock *rm, struct rm_priotracker *tracker, int trylock)
 	 * conditional jump.
 	 */
 	if (0 == (td->td_owepreempt |
-	    CPU_OVERLAP(&rm->rm_writecpus,  &pc->pc_cpumask)))
+	    CPU_ISSET(pc->pc_cpuid, &rm->rm_writecpus)))
 		return (1);
 
 	/* We do not have a read token and need to acquire one. */
diff --git a/sys/kern/sched_4bsd.c b/sys/kern/sched_4bsd.c
index 592bb80..574755f0 100644
--- a/sys/kern/sched_4bsd.c
+++ b/sys/kern/sched_4bsd.c
@@ -951,8 +951,7 @@ sched_switch(struct thread *td, struct thread *newtd, int flags)
 	if (td->td_flags & TDF_IDLETD) {
 		TD_SET_CAN_RUN(td);
 #ifdef SMP
-		/* Spinlock held here, assume no migration. */
-		CPU_NAND(&idle_cpus_mask, PCPU_PTR(cpumask));
+		CPU_CLR(PCPU_GET(cpuid), &idle_cpus_mask);
 #endif
 	} else {
 		if (TD_IS_RUNNING(td)) {
@@ -1026,7 +1025,7 @@ sched_switch(struct thread *td, struct thread *newtd, int flags)
 
 #ifdef SMP
 	if (td->td_flags & TDF_IDLETD)
-		CPU_OR(&idle_cpus_mask, PCPU_PTR(cpumask));
+		CPU_SET(PCPU_GET(cpuid), &idle_cpus_mask);
 #endif
 	sched_lock.mtx_lock = (uintptr_t)td;
 	td->td_oncpu = PCPU_GET(cpuid);
@@ -1055,7 +1054,8 @@ static int
 forward_wakeup(int cpunum)
 {
 	struct pcpu *pc;
-	cpuset_t dontuse, id, map, map2, me;
+	cpuset_t dontuse, map, map2;
+	u_int id, me;
 	int iscpuset;
 
 	mtx_assert(&sched_lock, MA_OWNED);
@@ -1073,27 +1073,24 @@ forward_wakeup(int cpunum)
 	/*
 	 * Check the idle mask we received against what we calculated
 	 * before in the old version.
-	 *
-	 * Also note that sched_lock is held now, thus no migration is
-	 * expected.
 	 */
-	me = PCPU_GET(cpumask);
+	me = PCPU_GET(cpuid);
 
 	/* Don't bother if we should be doing it ourself. */
-	if (CPU_OVERLAP(&me, &idle_cpus_mask) &&
-	    (cpunum == NOCPU || CPU_ISSET(cpunum, &me)))
+	if (CPU_ISSET(me, &idle_cpus_mask) &&
+	    (cpunum == NOCPU || me == cpunum))
 		return (0);
 
-	dontuse = me;
+	CPU_SETOF(me, &dontuse);
 	CPU_OR(&dontuse, &stopped_cpus);
 	CPU_OR(&dontuse, &hlt_cpus_mask);
 	CPU_ZERO(&map2);
 	if (forward_wakeup_use_loop) {
 		STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
-			id = pc->pc_cpumask;
-			if (!CPU_OVERLAP(&id, &dontuse) &&
+			id = pc->pc_cpuid;
+			if (!CPU_ISSET(id, &dontuse) &&
 			    pc->pc_curthread == pc->pc_idlethread) {
-				CPU_OR(&map2, &id);
+				CPU_SET(id, &map2);
 			}
 		}
 	}
@@ -1125,11 +1122,11 @@ forward_wakeup(int cpunum)
 	if (!CPU_EMPTY(&map)) {
 		forward_wakeups_delivered++;
 		STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
-			id = pc->pc_cpumask;
-			if (!CPU_OVERLAP(&map, &id))
+			id = pc->pc_cpuid;
+			if (!CPU_ISSET(id, &map))
 				continue;
 			if (cpu_idle_wakeup(pc->pc_cpuid))
-				CPU_NAND(&map, &id);
+				CPU_CLR(id, &map);
 		}
 		if (!CPU_EMPTY(&map))
 			ipi_selected(map, IPI_AST);
@@ -1147,7 +1144,7 @@ kick_other_cpu(int pri, int cpuid)
 	int cpri;
 
 	pcpu = pcpu_find(cpuid);
-	if (CPU_OVERLAP(&idle_cpus_mask, &pcpu->pc_cpumask)) {
+	if (CPU_ISSET(cpuid, &idle_cpus_mask)) {
 		forward_wakeups_delivered++;
 		if (!cpu_idle_wakeup(cpuid))
 			ipi_cpu(cpuid, IPI_AST);
@@ -1205,10 +1202,10 @@ void
 sched_add(struct thread *td, int flags)
 #ifdef SMP
 {
-	cpuset_t idle, me, tidlemsk;
+	cpuset_t tidlemsk;
 	struct td_sched *ts;
+	u_int cpu, cpuid;
 	int forwarded = 0;
-	int cpu;
 	int single_cpu = 0;
 
 	ts = td->td_sched;
@@ -1271,23 +1268,17 @@ sched_add(struct thread *td, int flags)
 		ts->ts_runq = &runq;
 	}
 
-	if (single_cpu && (cpu != PCPU_GET(cpuid))) {
+	cpuid = PCPU_GET(cpuid);
+	if (single_cpu && cpu != cpuid) {
 	        kick_other_cpu(td->td_priority, cpu);
 	} else {
 		if (!single_cpu) {
+			tidlemsk = idle_cpus_mask;
+			CPU_NAND(&tidlemsk, &hlt_cpus_mask);
+			CPU_CLR(cpuid, &tidlemsk);
 
-			/*
-			 * Thread spinlock is held here, assume no
-			 * migration is possible.
-			 */
-			me = PCPU_GET(cpumask);
-			idle = idle_cpus_mask;
-			tidlemsk = idle;
-			CPU_AND(&idle, &me);
-			CPU_OR(&me, &hlt_cpus_mask);
-			CPU_NAND(&tidlemsk, &me);
-
-			if (CPU_EMPTY(&idle) && ((flags & SRQ_INTR) == 0) &&
+			if (!CPU_ISSET(cpuid, &idle_cpus_mask) &&
+			    ((flags & SRQ_INTR) == 0) &&
 			    !CPU_EMPTY(&tidlemsk))
 				forwarded = forward_wakeup(cpu);
 		}
diff --git a/sys/kern/subr_kdb.c b/sys/kern/subr_kdb.c
index 2bc5ab2..f5cb31e 100644
--- a/sys/kern/subr_kdb.c
+++ b/sys/kern/subr_kdb.c
@@ -197,9 +197,12 @@ kdb_sysctl_trap_code(SYSCTL_HANDLER_ARGS)
 void
 kdb_panic(const char *msg)
 {
-	
 #ifdef SMP
-	stop_cpus_hard(PCPU_GET(other_cpus));
+	cpuset_t other_cpus;
+
+	other_cpus = all_cpus;
+	CPU_CLR(PCPU_GET(cpuid), &other_cpus);
+	stop_cpus_hard(other_cpus);
 #endif
 	printf("KDB: panic\n");
 	panic("%s", msg);
@@ -415,7 +418,7 @@ kdb_thr_ctx(struct thread *thr)
 #if defined(SMP) && defined(KDB_STOPPEDPCB)
 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu)  {
 		if (pc->pc_curthread == thr &&
-		    CPU_OVERLAP(&stopped_cpus, &pc->pc_cpumask))
+		    CPU_ISSET(pc->pc_cpuid, &stopped_cpus))
 			return (KDB_STOPPEDPCB(pc));
 	}
 #endif
@@ -499,6 +502,9 @@ kdb_thr_select(struct thread *thr)
 int
 kdb_trap(int type, int code, struct trapframe *tf)
 {
+#ifdef SMP
+	cpuset_t other_cpus;
+#endif
 	struct kdb_dbbe *be;
 	register_t intr;
 	int handled;
@@ -514,7 +520,9 @@ kdb_trap(int type, int code, struct trapframe *tf)
 	intr = intr_disable();
 
 #ifdef SMP
-	stop_cpus_hard(PCPU_GET(other_cpus));
+	other_cpus = all_cpus;
+	CPU_CLR(PCPU_GET(cpuid), &other_cpus);
+	stop_cpus_hard(other_cpus);
 #endif
 
 	kdb_active++;
diff --git a/sys/kern/subr_pcpu.c b/sys/kern/subr_pcpu.c
index a6b3ae0..ec6b590 100644
--- a/sys/kern/subr_pcpu.c
+++ b/sys/kern/subr_pcpu.c
@@ -87,7 +87,6 @@ pcpu_init(struct pcpu *pcpu, int cpuid, size_t size)
 	KASSERT(cpuid >= 0 && cpuid < MAXCPU,
 	    ("pcpu_init: invalid cpuid %d", cpuid));
 	pcpu->pc_cpuid = cpuid;
-	CPU_SETOF(cpuid, &pcpu->pc_cpumask);
 	cpuid_to_pcpu[cpuid] = pcpu;
 	STAILQ_INSERT_TAIL(&cpuhead, pcpu, pc_allcpu);
 	cpu_pcpu_init(pcpu, cpuid, size);
diff --git a/sys/kern/subr_smp.c b/sys/kern/subr_smp.c
index 0929ab9..caec965 100644
--- a/sys/kern/subr_smp.c
+++ b/sys/kern/subr_smp.c
@@ -142,7 +142,7 @@ mp_start(void *dummy)
 	/* Probe for MP hardware. */
 	if (smp_disabled != 0 || cpu_mp_probe() == 0) {
 		mp_ncpus = 1;
-		all_cpus = PCPU_GET(cpumask);
+		CPU_SETOF(PCPU_GET(cpuid), &all_cpus);
 		return;
 	}
 
@@ -706,7 +706,7 @@ mp_setvariables_for_up(void *dummy)
 {
 	mp_ncpus = 1;
 	mp_maxid = PCPU_GET(cpuid);
-	all_cpus = PCPU_GET(cpumask);
+	CPU_SETOF(mp_maxid, &all_cpus);
 	KASSERT(PCPU_GET(cpuid) == 0, ("UP must have a CPU ID of zero"));
 }
 SYSINIT(cpu_mp_setvariables, SI_SUB_TUNABLES, SI_ORDER_FIRST,