Fix the false IPIs on smp when using LAZY_SWITCH caused by pmap_activate()

not releasing the pm_active bit in the old pmap.

4 files changed, 18 insertions, 9 deletions

diff --git a/sys/i386/i386/genassym.c b/sys/i386/i386/genassym.c
index 675f80a..54e2ec2 100644
--- a/sys/i386/i386/genassym.c
+++ b/sys/i386/i386/genassym.c
@@ -201,6 +201,7 @@ ASSYM(PC_COMMON_TSSD, offsetof(struct pcpu, pc_common_tssd));
 ASSYM(PC_TSS_GDT, offsetof(struct pcpu, pc_tss_gdt));
 ASSYM(PC_CURRENTLDT, offsetof(struct pcpu, pc_currentldt));
 ASSYM(PC_CPUID, offsetof(struct pcpu, pc_cpuid));
+ASSYM(PC_CURPMAP, offsetof(struct pcpu, pc_curpmap));
 
 #ifdef SMP
 ASSYM(LA_VER, offsetof(struct LAPIC, version));
diff --git a/sys/i386/i386/pmap.c b/sys/i386/i386/pmap.c
index a2401c1..1e45a97 100644
--- a/sys/i386/i386/pmap.c
+++ b/sys/i386/i386/pmap.c
@@ -1073,6 +1073,7 @@ pmap_pinit0(pmap)
 	pmap->pm_pdpt = (pdpt_entry_t *)(KERNBASE + (vm_offset_t)IdlePDPT);
 #endif
 	pmap->pm_active = 0;
+	PCPU_SET(curpmap, pmap);
 	TAILQ_INIT(&pmap->pm_pvlist);
 	bzero(&pmap->pm_stats, sizeof pmap->pm_stats);
 	mtx_lock_spin(&allpmaps_lock);
@@ -3240,14 +3241,17 @@ void
 pmap_activate(struct thread *td)
 {
 	struct proc *p = td->td_proc;
-	pmap_t	pmap;
+	pmap_t	pmap, oldpmap;
 	u_int32_t  cr3;
 
 	critical_enter();
 	pmap = vmspace_pmap(td->td_proc->p_vmspace);
+	oldpmap = PCPU_GET(curpmap);
 #if defined(SMP)
+	atomic_clear_int(&oldpmap->pm_active, PCPU_GET(cpumask));
 	atomic_set_int(&pmap->pm_active, PCPU_GET(cpumask));
 #else
+	oldpmap->pm_active &= ~1;
 	pmap->pm_active |= 1;
 #endif
 #ifdef PAE
@@ -3271,6 +3275,7 @@ pmap_activate(struct thread *td)
 #ifdef SWTCH_OPTIM_STATS
 	tlb_flush_count++;
 #endif
+	PCPU_SET(curpmap, pmap);
 	critical_exit();
 }
 
diff --git a/sys/i386/i386/swtch.s b/sys/i386/i386/swtch.s
index a19baa7..810d82e 100644
--- a/sys/i386/i386/swtch.s
+++ b/sys/i386/i386/swtch.s
@@ -67,12 +67,11 @@ ENTRY(cpu_throw)
 	testl	%ecx,%ecx			/* no thread? */
 	jz	1f
 	/* release bit from old pm_active */
-	movl	TD_PROC(%ecx), %eax		/* thread->td_proc */
-	movl	P_VMSPACE(%eax), %ebx		/* proc->p_vmspace */
+	movl	PCPU(CURPMAP), %ebx
 #ifdef SMP
 	lock
 #endif
-	btrl	%esi, VM_PMAP+PM_ACTIVE(%ebx)	/* clear old */
+	btrl	%esi, PM_ACTIVE(%ebx)		/* clear old */
 1:
 	movl	8(%esp),%ecx			/* New thread */
 	movl	TD_PCB(%ecx),%edx
@@ -84,10 +83,12 @@ ENTRY(cpu_throw)
 	/* set bit in new pm_active */
 	movl	TD_PROC(%ecx),%eax
 	movl	P_VMSPACE(%eax), %ebx
+	addl	$VM_PMAP, %ebx
+	movl	%ebx, PCPU(CURPMAP)
 #ifdef SMP
 	lock
 #endif
-	btsl	%esi, VM_PMAP+PM_ACTIVE(%ebx)	/* set new */
+	btsl	%esi, PM_ACTIVE(%ebx)		/* set new */
 	jmp	sw1
 
 /*
@@ -188,20 +189,21 @@ ENTRY(cpu_switch)
 	movl	%eax,%cr3			/* new address space */
 
 	/* Release bit from old pmap->pm_active */
-	movl	TD_PROC(%edi), %eax		/* oldproc */
-	movl	P_VMSPACE(%eax), %ebx
+	movl	PCPU(CURPMAP), %ebx
 #ifdef SMP
 	lock
 #endif
-	btrl	%esi, VM_PMAP+PM_ACTIVE(%ebx)	/* clear old */
+	btrl	%esi, PM_ACTIVE(%ebx)		/* clear old */
 
 	/* Set bit in new pmap->pm_active */
 	movl	TD_PROC(%ecx),%eax		/* newproc */
 	movl	P_VMSPACE(%eax), %ebx
+	addl	$VM_PMAP, %ebx
+	movl	%ebx, PCPU(CURPMAP)
 #ifdef SMP
 	lock
 #endif
-	btsl	%esi, VM_PMAP+PM_ACTIVE(%ebx)	/* set new */
+	btsl	%esi, PM_ACTIVE(%ebx)		/* set new */
 
 #ifdef LAZY_SWITCH
 #ifdef SWTCH_OPTIM_STATS
diff --git a/sys/i386/include/pcpu.h b/sys/i386/include/pcpu.h
index a680645..799eefb 100644
--- a/sys/i386/include/pcpu.h
+++ b/sys/i386/include/pcpu.h
@@ -43,6 +43,7 @@
  */
 #define	PCPU_MD_FIELDS							\
 	struct	pcpu *pc_prvspace;		/* Self-reference */	\
+	struct	pmap *pc_curpmap;					\
 	struct	i386tss pc_common_tss;					\
 	struct	segment_descriptor pc_common_tssd;			\
 	struct	segment_descriptor *pc_tss_gdt;				\