- Move all of the PS_ flags into either p_flag or td_flags.

- p_sflag was mostly protected by PROC_LOCK rather than the PROC_SLOCK or previously the sched_lock. These bugs have existed for some time. - Allow swapout to try each thread in a process individually and then swapin the whole process if any of these fail. This allows us to move most scheduler related swap flags into td_flags. - Keep ki_sflag for backwards compat but change all in source tools to use the new and more correct location of P_INMEM. Reported by: pho Reviewed by: attilio, kib Approved by: re (kensmith)
svn path=/head/; revision=172207
2024-09-06 17:18:32 +00:00 · 2007-09-17 05:31:39 +00:00 · 2007-09-17 05:31:39 +00:00 · b61ce5b0e6 · 2020-12-20 02:59:44 +00:00
parent 0e8c5adf21
commit b61ce5b0e6
29 changed files with 168 additions and 172 deletions
--- a/bin/ps/print.c
+++ b/bin/ps/print.c
@ -197,14 +197,13 @@ logname(KINFO *k, VARENT *ve)
 void
 state(KINFO *k, VARENT *ve)
 {
-	int flag, sflag, tdflags;
+	int flag, tdflags;
 	char *cp;
 	VAR *v;
 	char buf[16];

 	v = ve->var;
 	flag = k->ki_p->ki_flag;
-	sflag = k->ki_p->ki_sflag;
 	tdflags = k->ki_p->ki_tdflags;	/* XXXKSE */
 	cp = buf;

@ -242,7 +241,7 @@ state(KINFO *k, VARENT *ve)
 		*cp = '?';
 	}
 	cp++;
-	if (!(sflag & PS_INMEM))
+	if (!(flag & P_INMEM))
 		*cp++ = 'W';
 	if (k->ki_p->ki_nice < NZERO)
 		*cp++ = '<';
@ -591,7 +590,7 @@ getpcpu(const KINFO *k)
 #define	fxtofl(fixpt)	((double)(fixpt) / fscale)

 	/* XXX - I don't like this */
-	if (k->ki_p->ki_swtime == 0 || (k->ki_p->ki_sflag & PS_INMEM) == 0)
+	if (k->ki_p->ki_swtime == 0 || (k->ki_p->ki_flag & P_INMEM) == 0)
 		return (0.0);
 	if (rawcpu)
 		return (100.0 * fxtofl(k->ki_p->ki_pctcpu));
@ -619,7 +618,7 @@ getpmem(KINFO *k)
 	if (failure)
 		return (0.0);

-	if ((k->ki_p->ki_sflag & PS_INMEM) == 0)
+	if ((k->ki_p->ki_flag & P_INMEM) == 0)
 		return (0.0);
 	/* XXX want pmap ptpages, segtab, etc. (per architecture) */
 	/* XXX don't have info about shared */
--- a/bin/ps/ps.c
+++ b/bin/ps/ps.c
@ -994,13 +994,13 @@ fmt(char **(*fn)(kvm_t *, const struct kinfo_proc *, int), KINFO *ki,
 	return (s);
 }

-#define UREADOK(ki)	(forceuread || (ki->ki_p->ki_sflag & PS_INMEM))
+#define UREADOK(ki)	(forceuread || (ki->ki_p->ki_flag & P_INMEM))

 static void
 saveuser(KINFO *ki)
 {

-	if (ki->ki_p->ki_sflag & PS_INMEM) {
+	if (ki->ki_p->ki_flag & P_INMEM) {
 		/*
 		 * The u-area might be swapped out, and we can't get
 		 * at it because we have a crashdump and no swap.
--- a/lib/libkvm/kvm_proc.c
+++ b/lib/libkvm/kvm_proc.c
@ -209,7 +209,7 @@ kvm_proclist(kd, what, arg, p, bp, maxcnt)
 			kp->ki_sigcatch = sigacts.ps_sigcatch;
 		}
 #if 0
-		if ((proc.p_sflag & PS_INMEM) && proc.p_stats != NULL) {
+		if ((proc.p_flag & P_INMEM) && proc.p_stats != NULL) {
 			if (KREAD(kd, (u_long)proc.p_stats, &pstats)) {
 				_kvm_err(kd, kd->program,
 				    "can't read stats at %x", proc.p_stats);
@ -370,7 +370,7 @@ kvm_proclist(kd, what, arg, p, bp, maxcnt)
 		if (proc.p_state != PRS_ZOMBIE) {
 			kp->ki_swtime = proc.p_swtime;
 			kp->ki_flag = proc.p_flag;
-			kp->ki_sflag = proc.p_sflag;
+			kp->ki_sflag = 0;
 			kp->ki_nice = proc.p_nice;
 			kp->ki_traceflag = proc.p_traceflag;
 			if (proc.p_state == PRS_NORMAL) { 
--- a/sys/amd64/amd64/genassym.c
+++ b/sys/amd64/amd64/genassym.c
@ -74,7 +74,6 @@ __FBSDID("$FreeBSD$");
 ASSYM(P_VMSPACE, offsetof(struct proc, p_vmspace));
 ASSYM(VM_PMAP, offsetof(struct vmspace, vm_pmap));
 ASSYM(PM_ACTIVE, offsetof(struct pmap, pm_active));
-ASSYM(P_SFLAG, offsetof(struct proc, p_sflag));

 ASSYM(TD_LOCK, offsetof(struct thread, td_lock));
 ASSYM(TD_FLAGS, offsetof(struct thread, td_flags));
--- a/sys/ddb/db_ps.c
+++ b/sys/ddb/db_ps.c
@ -162,7 +162,7 @@ db_ps(db_expr_t addr, boolean_t hasaddr, db_expr_t count, char *modif)
 		state[1] = '\0';

 		/* Additional process state flags. */
-		if (!p->p_sflag & PS_INMEM)
+		if (!p->p_flag & P_INMEM)
 			strlcat(state, "W", sizeof(state));
 		if (p->p_flag & P_TRACED)
 			strlcat(state, "X", sizeof(state));
--- a/sys/fs/procfs/procfs_status.c
+++ b/sys/fs/procfs/procfs_status.c
@ -112,7 +112,6 @@ procfs_doprocstatus(PFS_FILL_ARGS)
 		sbuf_printf(sb, "noflags");
 	}

-	PROC_SLOCK(p);
 #ifdef KSE
 	if (p->p_flag & P_SA)
 		wmesg = "-kse- ";
@ -128,9 +127,10 @@ procfs_doprocstatus(PFS_FILL_ARGS)
 			wmesg = "nochan";
 	}

-	if (p->p_sflag & PS_INMEM) {
+	if (p->p_flag & P_INMEM) {
 		struct timeval start, ut, st;

+		PROC_SLOCK(p);
 		calcru(p, &ut, &st);
 		PROC_SUNLOCK(p);
 		start = p->p_stats->p_start;
@ -139,10 +139,8 @@ procfs_doprocstatus(PFS_FILL_ARGS)
 		    (intmax_t)start.tv_sec, start.tv_usec,
 		    (intmax_t)ut.tv_sec, ut.tv_usec,
 		    (intmax_t)st.tv_sec, st.tv_usec);
-	} else {
-		PROC_SUNLOCK(p);
+	} else
 		sbuf_printf(sb, " -1,-1 -1,-1 -1,-1");
-	}

 	sbuf_printf(sb, " %s", wmesg);

--- a/sys/i386/i386/genassym.c
+++ b/sys/i386/i386/genassym.c
@ -78,7 +78,6 @@ __FBSDID("$FreeBSD$");
 ASSYM(P_VMSPACE, offsetof(struct proc, p_vmspace));
 ASSYM(VM_PMAP, offsetof(struct vmspace, vm_pmap));
 ASSYM(PM_ACTIVE, offsetof(struct pmap, pm_active));
-ASSYM(P_SFLAG, offsetof(struct proc, p_sflag));

 ASSYM(TD_FLAGS, offsetof(struct thread, td_flags));
 ASSYM(TD_LOCK, offsetof(struct thread, td_lock));
--- a/sys/i386/linux/linux_ptrace.c
+++ b/sys/i386/linux/linux_ptrace.c
@ -222,7 +222,7 @@ linux_proc_read_fpxregs(struct thread *td, struct linux_pt_fpxreg *fpxregs)
 {

 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
-	if (cpu_fxsr == 0 || (td->td_proc->p_sflag & PS_INMEM) == 0)
+	if (cpu_fxsr == 0 || (td->td_proc->p_flag & P_INMEM) == 0)
 		return (EIO);
 	bcopy(&td->td_pcb->pcb_save.sv_xmm, fpxregs, sizeof(*fpxregs));
 	return (0);
@ -233,7 +233,7 @@ linux_proc_write_fpxregs(struct thread *td, struct linux_pt_fpxreg *fpxregs)
 {

 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
-	if (cpu_fxsr == 0 || (td->td_proc->p_sflag & PS_INMEM) == 0)
+	if (cpu_fxsr == 0 || (td->td_proc->p_flag & P_INMEM) == 0)
 		return (EIO);
 	bcopy(fpxregs, &td->td_pcb->pcb_save.sv_xmm, sizeof(*fpxregs));
 	return (0);
--- a/sys/kern/init_main.c
+++ b/sys/kern/init_main.c
@ -415,8 +415,7 @@ proc0_init(void *dummy __unused)
 	session0.s_leader = p;

 	p->p_sysent = &null_sysvec;
-	p->p_flag = P_SYSTEM;
-	p->p_sflag = PS_INMEM;
+	p->p_flag = P_SYSTEM | P_INMEM;
 	p->p_state = PRS_NORMAL;
 	knlist_init(&p->p_klist, &p->p_mtx, NULL, NULL, NULL);
 	STAILQ_INIT(&p->p_ktr);
@ -428,6 +427,7 @@ proc0_init(void *dummy __unused)
 	td->td_priority = PVM;
 	td->td_base_pri = PUSER;
 	td->td_oncpu = 0;
+	td->td_flags = TDF_INMEM;
 	p->p_peers = 0;
 	p->p_leader = p;

@ -710,7 +710,7 @@ create_init(const void *udata __unused)
 	/* divorce init's credentials from the kernel's */
 	newcred = crget();
 	PROC_LOCK(initproc);
-	initproc->p_flag |= P_SYSTEM;
+	initproc->p_flag |= P_SYSTEM | P_INMEM;
 	oldcred = initproc->p_ucred;
 	crcopy(newcred, oldcred);
 #ifdef MAC
@ -723,9 +723,6 @@ create_init(const void *udata __unused)
 	PROC_UNLOCK(initproc);
 	crfree(oldcred);
 	cred_update_thread(FIRST_THREAD_IN_PROC(initproc));
-	PROC_SLOCK(initproc);
-	initproc->p_sflag |= PS_INMEM;
-	PROC_SUNLOCK(initproc);
 	cpu_set_fork_handler(FIRST_THREAD_IN_PROC(initproc), start_init, NULL);
 }
 SYSINIT(init, SI_SUB_CREATE_INIT, SI_ORDER_FIRST, create_init, NULL)
--- a/sys/kern/kern_clock.c
+++ b/sys/kern/kern_clock.c
@ -201,34 +201,29 @@ hardclock_cpu(int usermode)
 	struct pstats *pstats;
 	struct thread *td = curthread;
 	struct proc *p = td->td_proc;
-	int ast;
+	int flags;

 	/*
 	 * Run current process's virtual and profile time, as needed.
 	 */
 	pstats = p->p_stats;
-	ast = 0;
+	flags = 0;
 	if (usermode &&
 	    timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value)) {
 		PROC_SLOCK(p);
-		if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) {
-			p->p_sflag |= PS_ALRMPEND;
-			ast = 1;
-		}
+		if (itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0)
+			flags |= TDF_ALRMPEND | TDF_ASTPENDING;
 		PROC_SUNLOCK(p);
 	}
 	if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value)) {
 		PROC_SLOCK(p);
-		if (itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) {
-			p->p_sflag |= PS_PROFPEND;
-			ast = 1;
-		}
+		if (itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0)
+			flags |= TDF_PROFPEND | TDF_ASTPENDING;
 		PROC_SUNLOCK(p);
 	}
 	thread_lock(td);
 	sched_tick();
-	if (ast)
-		td->td_flags |= TDF_ASTPENDING;
+	td->td_flags |= flags;
 	thread_unlock(td);

 #ifdef	HWPMC_HOOKS
--- a/sys/kern/kern_fork.c
+++ b/sys/kern/kern_fork.c
@ -493,17 +493,15 @@ fork1(td, flags, pages, procp)

 	td2->td_sigstk = td->td_sigstk;
 	td2->td_sigmask = td->td_sigmask;
+	td2->td_flags = TDF_INMEM;

 	/*
 	 * Duplicate sub-structures as needed.
 	 * Increase reference counts on shared objects.
 	 */
-	p2->p_flag = 0;
+	p2->p_flag = P_INMEM;
 	if (p1->p_flag & P_PROFIL)
 		startprofclock(p2);
-	PROC_SLOCK(p2);
-	p2->p_sflag = PS_INMEM;
-	PROC_SUNLOCK(p2);
 	td2->td_ucred = crhold(p2->p_ucred);
 	pargs_hold(p2->p_args);

--- a/sys/kern/kern_kse.c
+++ b/sys/kern/kern_kse.c
@ -1002,6 +1002,7 @@ thread_alloc_spare(struct thread *td)
 	    __rangeof(struct thread, td_startzero, td_endzero));
 	spare->td_proc = td->td_proc;
 	spare->td_ucred = crhold(td->td_ucred);
+	spare->td_flags = TDF_INMEM;
 }

 /*
@ -1042,7 +1043,6 @@ thread_schedule_upcall(struct thread *td, struct kse_upcall *ku)
 	/* Let the new thread become owner of the upcall */
 	ku->ku_owner   = td2;
 	td2->td_upcall = ku;
-	td2->td_flags  = 0;
 	td2->td_pflags = TDP_SA|TDP_UPCALLING;
 	td2->td_state  = TDS_CAN_RUN;
 	td2->td_inhibitors = 0;
--- a/sys/kern/kern_proc.c
+++ b/sys/kern/kern_proc.c
@ -690,14 +690,17 @@ fill_kinfo_proc_only(struct proc *p, struct kinfo_proc *kp)
 		kp->ki_ssize = vm->vm_ssize;
 	} else if (p->p_state == PRS_ZOMBIE)
 		kp->ki_stat = SZOMB;
-	kp->ki_sflag = p->p_sflag;
+	if (kp->ki_flag & P_INMEM)
+		kp->ki_sflag = PS_INMEM;
+	else
+		kp->ki_sflag = 0;
 	kp->ki_swtime = p->p_swtime;
 	kp->ki_pid = p->p_pid;
 	kp->ki_nice = p->p_nice;
 	rufetch(p, &kp->ki_rusage);
 	kp->ki_runtime = cputick2usec(p->p_rux.rux_runtime);
 	PROC_SUNLOCK(p);
-	if ((p->p_sflag & PS_INMEM) && p->p_stats != NULL) {
+	if ((p->p_flag & P_INMEM) && p->p_stats != NULL) {
 		kp->ki_start = p->p_stats->p_start;
 		timevaladd(&kp->ki_start, &boottime);
 		PROC_SLOCK(p);
--- a/sys/kern/kern_switch.c
+++ b/sys/kern/kern_switch.c
@ -558,8 +558,8 @@ runq_remove_idx(struct runq *rq, struct td_sched *ts, u_char *idx)
 	struct rqhead *rqh;
 	u_char pri;

-	KASSERT(ts->ts_thread->td_proc->p_sflag & PS_INMEM,
-		("runq_remove_idx: process swapped out"));
+	KASSERT(ts->ts_thread->td_flags & TDF_INMEM,
+		("runq_remove_idx: thread swapped out"));
 	pri = ts->ts_rqindex;
 	KASSERT(pri < RQ_NQS, ("runq_remove_idx: Invalid index %d\n", pri));
 	rqh = &rq->rq_queues[pri];
--- a/sys/kern/kern_synch.c
+++ b/sys/kern/kern_synch.c
@ -463,16 +463,10 @@ mi_switch(int flags, struct thread *newtd)
 void
 setrunnable(struct thread *td)
 {
-	struct proc *p;

-	p = td->td_proc;
 	THREAD_LOCK_ASSERT(td, MA_OWNED);
-	switch (p->p_state) {
-	case PRS_ZOMBIE:
-		panic("setrunnable(1)");
-	default:
-		break;
-	}
+	KASSERT(td->td_proc->p_state != PRS_ZOMBIE,
+	    ("setrunnable: pid %d is a zombie", td->td_proc->p_pid));
 	switch (td->td_state) {
 	case TDS_RUNNING:
 	case TDS_RUNQ:
@ -491,9 +485,9 @@ setrunnable(struct thread *td)
 		printf("state is 0x%x", td->td_state);
 		panic("setrunnable(2)");
 	}
-	if ((p->p_sflag & PS_INMEM) == 0) {
-		if ((p->p_sflag & PS_SWAPPINGIN) == 0) {
-			p->p_sflag |= PS_SWAPINREQ;
+	if ((td->td_flags & TDF_INMEM) == 0) {
+		if ((td->td_flags & TDF_SWAPINREQ) == 0) {
+			td->td_flags |= TDF_SWAPINREQ;
 			/*
 			 * due to a LOR between the thread lock and
 			 * the sleepqueue chain locks, use
--- a/sys/kern/kern_thread.c
+++ b/sys/kern/kern_thread.c
@ -531,7 +531,7 @@ thread_link(struct thread *td, struct proc *p)
 	 */
 	td->td_state    = TDS_INACTIVE;
 	td->td_proc     = p;
-	td->td_flags    = 0;
+	td->td_flags    = TDF_INMEM;

 	LIST_INIT(&td->td_contested);
 	sigqueue_init(&td->td_sigqueue, p);
--- a/sys/kern/sched_4bsd.c
+++ b/sys/kern/sched_4bsd.c
@ -1086,8 +1086,8 @@ sched_add(struct thread *td, int flags)
 	    ("sched_add: trying to run inhibited thread"));
 	KASSERT((TD_CAN_RUN(td) || TD_IS_RUNNING(td)),
 	    ("sched_add: bad thread state"));
-	KASSERT(td->td_proc->p_sflag & PS_INMEM,
-	    ("sched_add: process swapped out"));
+	KASSERT(td->td_flags & TDF_INMEM,
+	    ("sched_add: thread swapped out"));
 	CTR5(KTR_SCHED, "sched_add: %p(%s) prio %d by %p(%s)",
 	    td, td->td_proc->p_comm, td->td_priority, curthread,
 	    curthread->td_proc->p_comm);
@ -1155,8 +1155,8 @@ sched_add(struct thread *td, int flags)
 	    ("sched_add: trying to run inhibited thread"));
 	KASSERT((TD_CAN_RUN(td) || TD_IS_RUNNING(td)),
 	    ("sched_add: bad thread state"));
-	KASSERT(td->td_proc->p_sflag & PS_INMEM,
-	    ("sched_add: process swapped out"));
+	KASSERT(td->td_flags & TDF_INMEM,
+	    ("sched_add: thread swapped out"));
 	CTR5(KTR_SCHED, "sched_add: %p(%s) prio %d by %p(%s)",
 	    td, td->td_proc->p_comm, td->td_priority, curthread,
 	    curthread->td_proc->p_comm);
@ -1200,8 +1200,8 @@ sched_rem(struct thread *td)
 	struct td_sched *ts;

 	ts = td->td_sched;
-	KASSERT(td->td_proc->p_sflag & PS_INMEM,
-	    ("sched_rem: process swapped out"));
+	KASSERT(td->td_flags & TDF_INMEM,
+	    ("sched_rem: thread swapped out"));
 	KASSERT(TD_ON_RUNQ(td),
 	    ("sched_rem: thread not on run queue"));
 	mtx_assert(&sched_lock, MA_OWNED);
@ -1253,8 +1253,8 @@ sched_choose(void)
 		runq_remove(rq, ts);
 		ts->ts_flags |= TSF_DIDRUN;

-		KASSERT(ts->ts_thread->td_proc->p_sflag & PS_INMEM,
-		    ("sched_choose: process swapped out"));
+		KASSERT(ts->ts_thread->td_flags & TDF_INMEM,
+		    ("sched_choose: thread swapped out"));
 		return (ts->ts_thread);
 	} 
 	return (PCPU_GET(idlethread));
--- a/sys/kern/sched_ule.c
+++ b/sys/kern/sched_ule.c
@ -2287,8 +2287,8 @@ tdq_add(struct tdq *tdq, struct thread *td, int flags)
 	    ("sched_add: trying to run inhibited thread"));
 	KASSERT((TD_CAN_RUN(td) || TD_IS_RUNNING(td)),
 	    ("sched_add: bad thread state"));
-	KASSERT(td->td_proc->p_sflag & PS_INMEM,
-	    ("sched_add: process swapped out"));
+	KASSERT(td->td_flags & TDF_INMEM,
+	    ("sched_add: thread swapped out"));

 	ts = td->td_sched;
 	class = PRI_BASE(td->td_pri_class);
--- a/sys/kern/subr_kdb.c
+++ b/sys/kern/subr_kdb.c
@ -400,7 +400,7 @@ kdb_thr_first(void)

 	p = LIST_FIRST(&allproc);
 	while (p != NULL) {
-		if (p->p_sflag & PS_INMEM) {
+		if (p->p_flag & P_INMEM) {
 			thr = FIRST_THREAD_IN_PROC(p);
 			if (thr != NULL)
 				return (thr);
@ -417,7 +417,7 @@ kdb_thr_from_pid(pid_t pid)

 	p = LIST_FIRST(&allproc);
 	while (p != NULL) {
-		if (p->p_sflag & PS_INMEM && p->p_pid == pid)
+		if (p->p_flag & P_INMEM && p->p_pid == pid)
 			return (FIRST_THREAD_IN_PROC(p));
 		p = LIST_NEXT(p, p_list);
 	}
@ -446,7 +446,7 @@ kdb_thr_next(struct thread *thr)
 		if (thr != NULL)
 			return (thr);
 		p = LIST_NEXT(p, p_list);
-		if (p != NULL && (p->p_sflag & PS_INMEM))
+		if (p != NULL && (p->p_flag & P_INMEM))
 			thr = FIRST_THREAD_IN_PROC(p);
 	} while (p != NULL);
 	return (NULL);
--- a/sys/kern/subr_trap.c
+++ b/sys/kern/subr_trap.c
@ -148,7 +148,6 @@ ast(struct trapframe *framep)
 {
 	struct thread *td;
 	struct proc *p;
-	int sflag;
 	int flags;
 	int sig;
 #if defined(DEV_NPX) && !defined(SMP)
@ -174,25 +173,17 @@ ast(struct trapframe *framep)
 #endif

 	/*
-	 * This updates the p_sflag's for the checks below in one
+	 * This updates the td_flag's for the checks below in one
 	 * "atomic" operation with turning off the astpending flag.
 	 * If another AST is triggered while we are handling the
-	 * AST's saved in sflag, the astpending flag will be set and
+	 * AST's saved in flags, the astpending flag will be set and
 	 * ast() will be called again.
 	 */
-	PROC_SLOCK(p);
-	sflag = p->p_sflag;
-	if (p->p_sflag & (PS_ALRMPEND | PS_PROFPEND))
-		p->p_sflag &= ~(PS_ALRMPEND | PS_PROFPEND);
-#ifdef MAC
-	if (p->p_sflag & PS_MACPEND)
-		p->p_sflag &= ~PS_MACPEND;
-#endif
 	thread_lock(td);
-	PROC_SUNLOCK(p);
 	flags = td->td_flags;
 	td->td_flags &= ~(TDF_ASTPENDING | TDF_NEEDSIGCHK |
-	    TDF_NEEDRESCHED | TDF_INTERRUPT);
+	    TDF_NEEDRESCHED | TDF_INTERRUPT | TDF_ALRMPEND | TDF_PROFPEND |
+	    TDF_MACPEND);
 	thread_unlock(td);
 	PCPU_INC(cnt.v_trap);

@ -210,7 +201,7 @@ ast(struct trapframe *framep)
 		td->td_profil_ticks = 0;
 		td->td_pflags &= ~TDP_OWEUPC;
 	}
-	if (sflag & PS_ALRMPEND) {
+	if (flags & TDF_ALRMPEND) {
 		PROC_LOCK(p);
 		psignal(p, SIGVTALRM);
 		PROC_UNLOCK(p);
@ -228,13 +219,13 @@ ast(struct trapframe *framep)
 		}
 	}
 #endif
-	if (sflag & PS_PROFPEND) {
+	if (flags & TDF_PROFPEND) {
 		PROC_LOCK(p);
 		psignal(p, SIGPROF);
 		PROC_UNLOCK(p);
 	}
 #ifdef MAC
-	if (sflag & PS_MACPEND)
+	if (flags & TDF_MACPEND)
 		mac_thread_userret(td);
 #endif
 	if (flags & TDF_NEEDRESCHED) {
--- a/sys/kern/sys_process.c
+++ b/sys/kern/sys_process.c
@ -104,7 +104,7 @@ struct ptrace_io_desc32 {
 	int error;							\
 									\
 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);			\
-	if ((td->td_proc->p_sflag & PS_INMEM) == 0)			\
+	if ((td->td_proc->p_flag & P_INMEM) == 0)			\
 		error = EIO;						\
 	else								\
 		error = (action);					\
--- a/sys/security/mac_lomac/mac_lomac.c
+++ b/sys/security/mac_lomac/mac_lomac.c
@ -537,8 +537,7 @@ maybe_demote(struct mac_lomac *subjlabel, struct mac_lomac *objlabel,
 	subj->mac_lomac.ml_rangehigh = objlabel->ml_single;
 	subj->mac_lomac.ml_flags |= MAC_LOMAC_FLAG_UPDATE;
 	thread_lock(curthread);
-	curthread->td_flags |= TDF_ASTPENDING;
-	curthread->td_proc->p_sflag |= PS_MACPEND;
+	curthread->td_flags |= TDF_ASTPENDING | TDF_MACPEND;
 	thread_unlock(curthread);

 	/*
--- a/sys/sparc64/sparc64/genassym.c
+++ b/sys/sparc64/sparc64/genassym.c
@ -219,7 +219,6 @@ ASSYM(MD_UTRAP, offsetof(struct mdproc, md_utrap));
 ASSYM(P_COMM, offsetof(struct proc, p_comm));
 ASSYM(P_MD, offsetof(struct proc, p_md));
 ASSYM(P_PID, offsetof(struct proc, p_pid));
-ASSYM(P_SFLAG, offsetof(struct proc, p_sflag));
 ASSYM(P_VMSPACE, offsetof(struct proc, p_vmspace));

 ASSYM(TD_FLAGS, offsetof(struct thread, td_flags));
--- a/sys/sys/proc.h
+++ b/sys/sys/proc.h
@ -316,6 +316,7 @@ do {									\
 */
 #define	TDF_BORROWING	0x00000001 /* Thread is borrowing pri from another. */
 #define	TDF_INPANIC	0x00000002 /* Caused a panic, let it drive crashdump. */
+#define	TDF_INMEM	0x00000004 /* Thread's stack is in memory. */
 #define	TDF_SINTR	0x00000008 /* Sleep is interruptible. */
 #define	TDF_TIMEOUT	0x00000010 /* Timing out during sleep. */
 #define	TDF_IDLETD	0x00000020 /* This is a per-CPU idle thread. */
@ -335,12 +336,15 @@ do {									\
 #define	TDF_UNUSED19	0x00080000 /* Thread is sleeping on a umtx. */
 #define	TDF_THRWAKEUP	0x00100000 /* Libthr thread must not suspend itself. */
 #define	TDF_DBSUSPEND	0x00200000 /* Thread is suspended by debugger */
-#define	TDF_UNUSED22	0x00400000 /* --available-- */
+#define	TDF_SWAPINREQ	0x00400000 /* Swapin request due to wakeup. */
 #define	TDF_UNUSED23	0x00800000 /* --available-- */
 #define	TDF_SCHED0	0x01000000 /* Reserved for scheduler private use */
 #define	TDF_SCHED1	0x02000000 /* Reserved for scheduler private use */
 #define	TDF_SCHED2	0x04000000 /* Reserved for scheduler private use */
 #define	TDF_SCHED3	0x08000000 /* Reserved for scheduler private use */
+#define	TDF_ALRMPEND	0x10000000 /* Pending SIGVTALRM needs to be posted. */
+#define	TDF_PROFPEND	0x20000000 /* Pending SIGPROF needs to be posted. */
+#define	TDF_MACPEND	0x40000000 /* AST-based MAC event pending. */

 /*
 * "Private" flags kept in td_pflags:
@ -496,7 +500,6 @@ struct proc {
 	 * See the td_ or ke_ versions of the same flags.
 	 */
 	int		p_flag;		/* (c) P_* flags. */
-	int		p_sflag;	/* (j) PS_* flags. */
 	enum {
 		PRS_NEW = 0,		/* In creation */
 		PRS_NORMAL,		/* threads can be run. */
@ -618,19 +621,13 @@ struct proc {
 #define	P_JAILED	0x1000000 /* Process is in jail. */
 #define	P_INEXEC	0x4000000 /* Process is in execve(). */
 #define	P_STATCHILD	0x8000000 /* Child process stopped or exited. */
+#define	P_INMEM		0x10000000 /* Loaded into memory. */
+#define	P_SWAPPINGOUT	0x20000000 /* Process is being swapped out. */
+#define	P_SWAPPINGIN	0x40000000 /* Process is being swapped in. */

 #define	P_STOPPED	(P_STOPPED_SIG|P_STOPPED_SINGLE|P_STOPPED_TRACE)
 #define	P_SHOULDSTOP(p)	((p)->p_flag & P_STOPPED)

-/* These flags are kept in p_sflag and are protected with proc slock. */
-#define	PS_INMEM	0x00001	/* Loaded into memory. */
-#define	PS_ALRMPEND	0x00020	/* Pending SIGVTALRM needs to be posted. */
-#define	PS_PROFPEND	0x00040	/* Pending SIGPROF needs to be posted. */
-#define	PS_SWAPINREQ	0x00100	/* Swapin request due to wakeup. */
-#define	PS_SWAPPINGOUT	0x00200	/* Process is being swapped out. */
-#define	PS_SWAPPINGIN	0x04000	/* Process is being swapped in. */
-#define	PS_MACPEND	0x08000	/* AST-based MAC event pending. */
-
 /*
 * These were process status values (p_stat), now they are only used in
 * legacy conversion code.
@ -743,7 +740,7 @@ MALLOC_DECLARE(M_ZOMBIE);
 	KASSERT(!((p)->p_flag & P_WEXIT) || (p) == curproc,		\
 	    ("PHOLD of exiting process"));				\
 	(p)->p_lock++;							\
-	if (((p)->p_sflag & PS_INMEM) == 0)				\
+	if (((p)->p_flag & P_INMEM) == 0)				\
 		faultin((p));						\
 } while (0)
 #define PROC_ASSERT_HELD(p) do {					\
--- a/sys/sys/user.h
+++ b/sys/sys/user.h
@ -207,6 +207,12 @@ void fill_kinfo_proc(struct proc *, struct kinfo_proc *);
 #define	ki_childstime	ki_rusage_ch.ru_stime
 #define	ki_childutime	ki_rusage_ch.ru_utime

+/*
+ *  Legacy PS_ flag.  This moved to p_flag but is maintained for
+ *  compatibility.
+ */
+#define	PS_INMEM	0x00001		/* Loaded into memory. */
+
 /* ki_sessflag values */
 #define	KI_CTTY		0x00000001	/* controlling tty vnode active */
 #define	KI_SLEADER	0x00000002	/* session leader */
--- a/sys/vm/vm_glue.c
+++ b/sys/vm/vm_glue.c
@ -112,7 +112,8 @@ static void scheduler(void *);
 SYSINIT(scheduler, SI_SUB_RUN_SCHEDULER, SI_ORDER_ANY, scheduler, NULL)

 #ifndef NO_SWAPPING
-static void swapout(struct proc *);
+static int swapout(struct proc *);
+static void swapclear(struct proc *);
 #endif


@ -601,7 +602,7 @@ faultin(p)
 #ifdef NO_SWAPPING

 	PROC_LOCK_ASSERT(p, MA_OWNED);
-	if ((p->p_sflag & PS_INMEM) == 0)
+	if ((p->p_flag & P_INMEM) == 0)
 		panic("faultin: proc swapped out with NO_SWAPPING!");
 #else /* !NO_SWAPPING */
 	struct thread *td;
@ -611,36 +612,34 @@ faultin(p)
 	 * If another process is swapping in this process,
 	 * just wait until it finishes.
 	 */
-	if (p->p_sflag & PS_SWAPPINGIN)
-		msleep(&p->p_sflag, &p->p_mtx, PVM, "faultin", 0);
-	else if ((p->p_sflag & PS_INMEM) == 0) {
+	if (p->p_flag & P_SWAPPINGIN) {
+		while (p->p_flag & P_SWAPPINGIN)
+			msleep(&p->p_flag, &p->p_mtx, PVM, "faultin", 0);
+		return;
+	}
+	if ((p->p_flag & P_INMEM) == 0) {
 		/*
 		 * Don't let another thread swap process p out while we are
 		 * busy swapping it in.
 		 */
 		++p->p_lock;
-		PROC_SLOCK(p);
-		p->p_sflag |= PS_SWAPPINGIN;
-		PROC_SUNLOCK(p);
+		p->p_flag |= P_SWAPPINGIN;
 		PROC_UNLOCK(p);

+		/*
+		 * We hold no lock here because the list of threads
+		 * can not change while all threads in the process are
+		 * swapped out.
+		 */
 		FOREACH_THREAD_IN_PROC(p, td)
 			vm_thread_swapin(td);
-
 		PROC_LOCK(p);
 		PROC_SLOCK(p);
-		p->p_sflag &= ~PS_SWAPPINGIN;
-		p->p_sflag |= PS_INMEM;
-		FOREACH_THREAD_IN_PROC(p, td) {
-			thread_lock(td);
-			TD_CLR_SWAPPED(td);
-			if (TD_CAN_RUN(td))
-				setrunnable(td);
-			thread_unlock(td);
-		}
+		swapclear(p);
+		p->p_swtime = 0;
 		PROC_SUNLOCK(p);

-		wakeup(&p->p_sflag);
+		wakeup(&p->p_flag);

 		/* Allow other threads to swap p out now. */
 		--p->p_lock;
@ -684,7 +683,9 @@ scheduler(dummy)
 	ppri = INT_MIN;
 	sx_slock(&allproc_lock);
 	FOREACH_PROC_IN_SYSTEM(p) {
-		if (p->p_sflag & (PS_INMEM | PS_SWAPPINGOUT | PS_SWAPPINGIN)) {
+		PROC_LOCK(p);
+		if (p->p_flag & (P_SWAPPINGOUT | P_SWAPPINGIN | P_INMEM)) {
+			PROC_UNLOCK(p);
 			continue;
 		}
 		PROC_SLOCK(p);
@ -697,10 +698,8 @@ scheduler(dummy)
 			thread_lock(td);
 			if (td->td_inhibitors == TDI_SWAPPED) {
 				pri = p->p_swtime + td->td_slptime;
-				if ((p->p_sflag & PS_SWAPINREQ) == 0) {
+				if ((td->td_flags & TDF_SWAPINREQ) == 0)
 					pri -= p->p_nice * 8;
-				}
-
 				/*
 				 * if this thread is higher priority
 				 * and there is enough space, then select
@ -715,6 +714,7 @@ scheduler(dummy)
 			thread_unlock(td);
 		}
 		PROC_SUNLOCK(p);
+		PROC_UNLOCK(p);
 	}
 	sx_sunlock(&allproc_lock);

@ -738,7 +738,7 @@ scheduler(dummy)
 	 * brought this process in while we traverse all threads.
 	 * Or, this process may even be being swapped out again.
 	 */
-	if (p->p_sflag & (PS_INMEM | PS_SWAPPINGOUT | PS_SWAPPINGIN)) {
+	if (p->p_flag & (P_INMEM | P_SWAPPINGOUT | P_SWAPPINGIN)) {
 		PROC_UNLOCK(p);
 		thread_lock(&thread0);
 		proc0_rescan = 0;
@ -746,19 +746,12 @@ scheduler(dummy)
 		goto loop;
 	}

-	PROC_SLOCK(p);
-	p->p_sflag &= ~PS_SWAPINREQ;
-	PROC_SUNLOCK(p);
-
 	/*
 	 * We would like to bring someone in. (only if there is space).
 	 * [What checks the space? ]
 	 */
 	faultin(p);
 	PROC_UNLOCK(p);
-	PROC_SLOCK(p);
-	p->p_swtime = 0;
-	PROC_SUNLOCK(p);
 	thread_lock(&thread0);
 	proc0_rescan = 0;
 	thread_unlock(&thread0);
@ -804,7 +797,7 @@ SYSCTL_INT(_vm, OID_AUTO, swap_idle_threshold2, CTLFLAG_RW,

 /*
 * 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
+ * procs and swap out their stacks.  We try to always "swap" at least one
 * process in case we need the room for a swapin.
 * If any procs have been sleeping/stopped for at least maxslp seconds,
 * they are swapped.  Else, we swap the longest-sleeping or stopped process,
@ -829,13 +822,8 @@ int action;
 		 * creation.  It may have no
 		 * address space or lock yet.
 		 */
-		PROC_SLOCK(p);
-		if (p->p_state == PRS_NEW) {
-			PROC_SUNLOCK(p);
+		if (p->p_state == PRS_NEW)
 			continue;
-		}
-		PROC_SUNLOCK(p);
-
 		/*
 		 * An aio daemon switches its
 		 * address space while running.
@ -844,7 +832,6 @@ int action;
 		 */
 		if ((p->p_flag & P_SYSTEM) != 0)
 			continue;
-
 		/*
 		 * Do not swapout a process that
 		 * is waiting for VM data
@ -874,7 +861,7 @@ int action;
 		 * skipped because of the if statement above checking 
 		 * for P_SYSTEM
 		 */
-		if ((p->p_sflag & (PS_INMEM|PS_SWAPPINGOUT|PS_SWAPPINGIN)) != PS_INMEM)
+		if ((p->p_flag & (P_INMEM|P_SWAPPINGOUT|P_SWAPPINGIN)) != P_INMEM)
 			goto nextproc2;

 		switch (p->p_state) {
@ -890,15 +877,20 @@ int action;
 			 * Check all the thread groups..
 			 */
 			FOREACH_THREAD_IN_PROC(p, td) {
-				if (PRI_IS_REALTIME(td->td_pri_class))
+				thread_lock(td);
+				if (PRI_IS_REALTIME(td->td_pri_class)) {
+					thread_unlock(td);
 					goto nextproc;
+				}

 				/*
 				 * Guarantee swap_idle_threshold1
 				 * time in memory.
 				 */
-				if (td->td_slptime < swap_idle_threshold1)
+				if (td->td_slptime < swap_idle_threshold1) {
+					thread_unlock(td);
 					goto nextproc;
+				}

 				/*
 				 * Do not swapout a process if it is
@ -910,8 +902,10 @@ int action;
 				 * swapping out a thread.
 				 */
 				if ((td->td_priority) < PSOCK ||
-				    !thread_safetoswapout(td))
+				    !thread_safetoswapout(td)) {
+					thread_unlock(td);
 					goto nextproc;
+				}
 				/*
 				 * If the system is under memory stress,
 				 * or if we are swapping
@ -920,11 +914,14 @@ int action;
 				 */
 				if (((action & VM_SWAP_NORMAL) == 0) &&
 				    (((action & VM_SWAP_IDLE) == 0) ||
-				    (td->td_slptime < swap_idle_threshold2)))
+				    (td->td_slptime < swap_idle_threshold2))) {
+					thread_unlock(td);
 					goto nextproc;
+				}

 				if (minslptime > td->td_slptime)
 					minslptime = td->td_slptime;
+				thread_unlock(td);
 			}

 			/*
@ -935,8 +932,8 @@ int action;
 			if ((action & VM_SWAP_NORMAL) ||
 				((action & VM_SWAP_IDLE) &&
 				 (minslptime > swap_idle_threshold2))) {
-				swapout(p);
-				didswap++;
+				if (swapout(p) == 0)
+					didswap++;
 				PROC_SUNLOCK(p);
 				PROC_UNLOCK(p);
 				vm_map_unlock(&vm->vm_map);
@ -964,13 +961,35 @@ int action;
 }

 static void
+swapclear(p)
+	struct proc *p;
+{
+	struct thread *td;
+
+	PROC_LOCK_ASSERT(p, MA_OWNED);
+	PROC_SLOCK_ASSERT(p, MA_OWNED);
+
+	FOREACH_THREAD_IN_PROC(p, td) {
+		thread_lock(td);
+		td->td_flags |= TDF_INMEM;
+		td->td_flags &= ~TDF_SWAPINREQ;
+		TD_CLR_SWAPPED(td);
+		if (TD_CAN_RUN(td))
+			setrunnable(td);
+		thread_unlock(td);
+	}
+	p->p_flag &= ~(P_SWAPPINGIN|P_SWAPPINGOUT);
+	p->p_flag |= P_INMEM;
+}
+
+static int
 swapout(p)
 	struct proc *p;
 {
 	struct thread *td;

 	PROC_LOCK_ASSERT(p, MA_OWNED);
-	mtx_assert(&p->p_slock, MA_OWNED | MA_NOTRECURSED);
+	PROC_SLOCK_ASSERT(p, MA_OWNED | MA_NOTRECURSED);
 #if defined(SWAP_DEBUG)
 	printf("swapping out %d\n", p->p_pid);
 #endif
@ -980,43 +999,46 @@ swapout(p)
 	 * by now.  Assuming that there is only one pageout daemon thread,
 	 * this process should still be in memory.
 	 */
-	KASSERT((p->p_sflag & (PS_INMEM|PS_SWAPPINGOUT|PS_SWAPPINGIN)) == PS_INMEM,
+	KASSERT((p->p_flag & (P_INMEM|P_SWAPPINGOUT|P_SWAPPINGIN)) == P_INMEM,
 		("swapout: lost a swapout race?"));

-#if defined(INVARIANTS)
-	/*
-	 * Make sure that all threads are safe to be swapped out.
-	 *
-	 * Alternatively, we could swap out only safe threads.
-	 */
-	FOREACH_THREAD_IN_PROC(p, td) {
-		KASSERT(thread_safetoswapout(td),
-			("swapout: there is a thread not safe for swapout"));
-	}
-#endif /* INVARIANTS */
-	td = FIRST_THREAD_IN_PROC(p);
-	++td->td_ru.ru_nswap;
 	/*
 	 * remember the process resident count
 	 */
 	p->p_vmspace->vm_swrss = vmspace_resident_count(p->p_vmspace);
-
-	p->p_sflag &= ~PS_INMEM;
-	p->p_sflag |= PS_SWAPPINGOUT;
-	PROC_UNLOCK(p);
+	/*
+	 * Check and mark all threads before we proceed.
+	 */
+	p->p_flag &= ~P_INMEM;
+	p->p_flag |= P_SWAPPINGOUT;
 	FOREACH_THREAD_IN_PROC(p, td) {
 		thread_lock(td);
+		if (!thread_safetoswapout(td)) {
+			thread_unlock(td);
+			swapclear(p);
+			return (EBUSY);
+		}
+		td->td_flags &= ~TDF_INMEM;
 		TD_SET_SWAPPED(td);
 		thread_unlock(td);
 	}
+	td = FIRST_THREAD_IN_PROC(p);
+	++td->td_ru.ru_nswap;
 	PROC_SUNLOCK(p);
+	PROC_UNLOCK(p);

+	/*
+	 * This list is stable because all threads are now prevented from
+	 * running.  The list is only modified in the context of a running
+	 * thread in this process.
+	 */
 	FOREACH_THREAD_IN_PROC(p, td)
 		vm_thread_swapout(td);

 	PROC_LOCK(p);
+	p->p_flag &= ~P_SWAPPINGOUT;
 	PROC_SLOCK(p);
-	p->p_sflag &= ~PS_SWAPPINGOUT;
 	p->p_swtime = 0;
+	return (0);
 }
 #endif /* !NO_SWAPPING */
--- a/sys/vm/vm_pageout.c
+++ b/sys/vm/vm_pageout.c
@ -1620,7 +1620,7 @@ vm_daemon()
 			 * swapped out set the limit to nothing (will force a
 			 * swap-out.)
 			 */
-			if ((p->p_sflag & PS_INMEM) == 0)
+			if ((p->p_flag & P_INMEM) == 0)
 				limit = 0;	/* XXX */
 			PROC_UNLOCK(p);

--- a/usr.bin/systat/pigs.c
+++ b/usr.bin/systat/pigs.c
@ -174,7 +174,7 @@ fetchpigs()
 		pt[i].pt_kp = &kpp[i];
 		pctp = &pt[i].pt_pctcpu;
 		ftime = kpp[i].ki_swtime;
-		if (ftime == 0 || (kpp[i].ki_sflag & PS_INMEM) == 0)
+		if (ftime == 0 || (kpp[i].ki_flag & P_INMEM) == 0)
 			*pctp = 0;
 		else
 			*pctp = ((double) kpp[i].ki_pctcpu /
--- a/usr.bin/top/machine.c
+++ b/usr.bin/top/machine.c
@ -675,7 +675,7 @@ format_next_process(caddr_t handle, char *(*get_userid)(int), int flags)
 	hp->remaining--;

 	/* get the process's command name */
-	if ((pp->ki_sflag & PS_INMEM) == 0) {
+	if ((pp->ki_flag & P_INMEM) == 0) {
 		/*
 		 * Print swapped processes as <pname>
 		 */