FreeBSD right now support 32 CPUs on all the architectures at least.

With the arrival of 128+ cores it is necessary to handle more than that.
One of the first thing to change is the support for cpumask_t that needs
to handle more than 32 bits masking (which happens now).  Some places,
however, still assume that cpumask_t is a 32 bits mask.
Fix that situation by using always correctly cpumask_t when needed.

While here, remove the part under STOP_NMI for the Xen support as it
is broken in any case.

Additively make ipi_nmi_pending as static.

Reviewed by:	jhb, kmacy
Tested by:	Giovanni Trematerra <giovanni dot trematerra at gmail dot com>

4 files changed, 21 insertions, 78 deletions

diff --git a/sys/i386/i386/mp_machdep.c b/sys/i386/i386/mp_machdep.c
index 4495a57..c233f25 100644
--- a/sys/i386/i386/mp_machdep.c
+++ b/sys/i386/i386/mp_machdep.c
@@ -155,9 +155,9 @@ vm_offset_t smp_tlb_addr2;
 volatile int smp_tlb_wait;
 
 #ifdef STOP_NMI
-volatile cpumask_t ipi_nmi_pending;
+static volatile cpumask_t ipi_nmi_pending;
 
-static void	ipi_nmi_selected(u_int32_t cpus);
+static void	ipi_nmi_selected(cpumask_t cpus);
 #endif 
 
 #ifdef COUNT_IPIS
@@ -1146,7 +1146,7 @@ smp_tlb_shootdown(u_int vector, vm_offset_t addr1, vm_offset_t addr2)
 }
 
 static void
-smp_targeted_tlb_shootdown(u_int mask, u_int vector, vm_offset_t addr1, vm_offset_t addr2)
+smp_targeted_tlb_shootdown(cpumask_t mask, u_int vector, vm_offset_t addr1, vm_offset_t addr2)
 {
 	int ncpu, othercpus;
 
@@ -1231,7 +1231,7 @@ smp_invlpg_range(vm_offset_t addr1, vm_offset_t addr2)
 }
 
 void
-smp_masked_invltlb(u_int mask)
+smp_masked_invltlb(cpumask_t mask)
 {
 
 	if (smp_started) {
@@ -1243,7 +1243,7 @@ smp_masked_invltlb(u_int mask)
 }
 
 void
-smp_masked_invlpg(u_int mask, vm_offset_t addr)
+smp_masked_invlpg(cpumask_t mask, vm_offset_t addr)
 {
 
 	if (smp_started) {
@@ -1255,7 +1255,7 @@ smp_masked_invlpg(u_int mask, vm_offset_t addr)
 }
 
 void
-smp_masked_invlpg_range(u_int mask, vm_offset_t addr1, vm_offset_t addr2)
+smp_masked_invlpg_range(cpumask_t mask, vm_offset_t addr1, vm_offset_t addr2)
 {
 
 	if (smp_started) {
@@ -1303,7 +1303,7 @@ ipi_bitmap_handler(struct trapframe frame)
  * send an IPI to a set of cpus.
  */
 void
-ipi_selected(u_int32_t cpus, u_int ipi)
+ipi_selected(cpumask_t cpus, u_int ipi)
 {
 	int cpu;
 	u_int bitmap = 0;
@@ -1367,7 +1367,7 @@ ipi_all_but_self(u_int ipi)
 #define	BEFORE_SPIN	1000000
 
 void
-ipi_nmi_selected(u_int32_t cpus)
+ipi_nmi_selected(cpumask_t cpus)
 {
 	int cpu;
 	register_t icrlo;
@@ -1456,7 +1456,7 @@ SYSINIT(start_aps, SI_SUB_SMP, SI_ORDER_FIRST, release_aps, NULL);
 static int
 sysctl_hlt_cpus(SYSCTL_HANDLER_ARGS)
 {
-	u_int mask;
+	cpumask_t mask;
 	int error;
 
 	mask = hlt_cpus_mask;
diff --git a/sys/i386/i386/pmap.c b/sys/i386/i386/pmap.c
index 5f016d5..5d7dc97 100644
--- a/sys/i386/i386/pmap.c
+++ b/sys/i386/i386/pmap.c
@@ -1624,7 +1624,7 @@ retry:
  * Deal with a SMP shootdown of other users of the pmap that we are
  * trying to dispose of.  This can be a bit hairy.
  */
-static u_int *lazymask;
+static cpumask_t *lazymask;
 static u_int lazyptd;
 static volatile u_int lazywait;
 
@@ -1633,7 +1633,7 @@ void pmap_lazyfix_action(void);
 void
 pmap_lazyfix_action(void)
 {
-	u_int mymask = PCPU_GET(cpumask);
+	cpumask_t mymask = PCPU_GET(cpumask);
 
 #ifdef COUNT_IPIS
 	(*ipi_lazypmap_counts[PCPU_GET(cpuid)])++;
@@ -1645,7 +1645,7 @@ pmap_lazyfix_action(void)
 }
 
 static void
-pmap_lazyfix_self(u_int mymask)
+pmap_lazyfix_self(cpumask_t mymask)
 {
 
 	if (rcr3() == lazyptd)
@@ -1657,8 +1657,7 @@ pmap_lazyfix_self(u_int mymask)
 static void
 pmap_lazyfix(pmap_t pmap)
 {
-	u_int mymask;
-	u_int mask;
+	cpumask_t mymask, mask;
 	u_int spins;
 
 	while ((mask = pmap->pm_active) != 0) {
diff --git a/sys/i386/include/smp.h b/sys/i386/include/smp.h
index 33739cc..917c285 100644
--- a/sys/i386/include/smp.h
+++ b/sys/i386/include/smp.h
@@ -69,12 +69,12 @@ u_int	mp_bootaddress(u_int);
 int	mp_grab_cpu_hlt(void);
 void	smp_cache_flush(void);
 void	smp_invlpg(vm_offset_t addr);
-void	smp_masked_invlpg(u_int mask, vm_offset_t addr);
+void	smp_masked_invlpg(cpumask_t mask, vm_offset_t addr);
 void	smp_invlpg_range(vm_offset_t startva, vm_offset_t endva);
-void	smp_masked_invlpg_range(u_int mask, vm_offset_t startva,
+void	smp_masked_invlpg_range(cpumask_t mask, vm_offset_t startva,
 	    vm_offset_t endva);
 void	smp_invltlb(void);
-void	smp_masked_invltlb(u_int mask);
+void	smp_masked_invltlb(cpumask_t mask);
 
 #ifdef STOP_NMI
 int	ipi_nmi_handler(void);
diff --git a/sys/i386/xen/mp_machdep.c b/sys/i386/xen/mp_machdep.c
index 7af4437..d8e1eea 100644
--- a/sys/i386/xen/mp_machdep.c
+++ b/sys/i386/xen/mp_machdep.c
@@ -993,7 +993,7 @@ smp_tlb_shootdown(u_int vector, vm_offset_t addr1, vm_offset_t addr2)
 }
 
 static void
-smp_targeted_tlb_shootdown(u_int mask, u_int vector, vm_offset_t addr1, vm_offset_t addr2)
+smp_targeted_tlb_shootdown(cpumask_t mask, u_int vector, vm_offset_t addr1, vm_offset_t addr2)
 {
 	int ncpu, othercpus;
 	struct _call_data data;
@@ -1072,7 +1072,7 @@ smp_invlpg_range(vm_offset_t addr1, vm_offset_t addr2)
 }
 
 void
-smp_masked_invltlb(u_int mask)
+smp_masked_invltlb(cpumask_t mask)
 {
 
 	if (smp_started) {
@@ -1081,7 +1081,7 @@ smp_masked_invltlb(u_int mask)
 }
 
 void
-smp_masked_invlpg(u_int mask, vm_offset_t addr)
+smp_masked_invlpg(cpumask_t mask, vm_offset_t addr)
 {
 
 	if (smp_started) {
@@ -1090,7 +1090,7 @@ smp_masked_invlpg(u_int mask, vm_offset_t addr)
 }
 
 void
-smp_masked_invlpg_range(u_int mask, vm_offset_t addr1, vm_offset_t addr2)
+smp_masked_invlpg_range(cpumask_t mask, vm_offset_t addr1, vm_offset_t addr2)
 {
 
 	if (smp_started) {
@@ -1102,7 +1102,7 @@ smp_masked_invlpg_range(u_int mask, vm_offset_t addr1, vm_offset_t addr2)
  * send an IPI to a set of cpus.
  */
 void
-ipi_selected(uint32_t cpus, u_int ipi)
+ipi_selected(cpumask_t cpus, u_int ipi)
 {
 	int cpu;
 	u_int bitmap = 0;
@@ -1114,12 +1114,6 @@ ipi_selected(uint32_t cpus, u_int ipi)
 		ipi = IPI_BITMAP_VECTOR;
 	} 
 
-#ifdef STOP_NMI
-	if (ipi == IPI_STOP && stop_cpus_with_nmi) {
-		ipi_nmi_selected(cpus);
-		return;
-	}
-#endif
 	CTR3(KTR_SMP, "%s: cpus: %x ipi: %x", __func__, cpus, ipi);
 	while ((cpu = ffs(cpus)) != 0) {
 		cpu--;
@@ -1160,56 +1154,6 @@ ipi_all_but_self(u_int ipi)
 	ipi_selected(PCPU_GET(other_cpus), ipi);
 }
 
-#ifdef STOP_NMI
-/*
- * send NMI IPI to selected CPUs
- */
-
-#define	BEFORE_SPIN	1000000
-
-void
-ipi_nmi_selected(u_int32_t cpus)
-{
-	int cpu;
-	register_t icrlo;
-
-	icrlo = APIC_DELMODE_NMI | APIC_DESTMODE_PHY | APIC_LEVEL_ASSERT 
-		| APIC_TRIGMOD_EDGE; 
-	
-	CTR2(KTR_SMP, "%s: cpus: %x nmi", __func__, cpus);
-
-	atomic_set_int(&ipi_nmi_pending, cpus);
-
-	while ((cpu = ffs(cpus)) != 0) {
-		cpu--;
-		cpus &= ~(1 << cpu);
-
-		KASSERT(cpu_apic_ids[cpu] != -1,
-		    ("IPI NMI to non-existent CPU %d", cpu));
-		
-		/* Wait for an earlier IPI to finish. */
-		if (!lapic_ipi_wait(BEFORE_SPIN))
-			panic("ipi_nmi_selected: previous IPI has not cleared");
-
-		lapic_ipi_raw(icrlo, cpu_apic_ids[cpu]);
-	}
-}
-
-int
-ipi_nmi_handler(void)
-{
-	int cpumask = PCPU_GET(cpumask);
-
-	if (!(ipi_nmi_pending & cpumask))
-		return 1;
-
-	atomic_clear_int(&ipi_nmi_pending, cpumask);
-	cpustop_handler();
-	return 0;
-}
-
-#endif /* STOP_NMI */
-
 /*
  * Handle an IPI_STOP by saving our current context and spinning until we
  * are resumed.