- Unlike cache invalidation and TLB demapping IPIs, reading registers from

other CPUs doesn't require locking so get rid of it. As the latter is used
  for the timecounter on certain machine models, using a spin lock in this
  case can lead to a deadlock with the upcoming callout(9) rework.
- Merge r134227/r167250 from x86:
  Avoid cross-IPI SMP deadlock by using the smp_ipi_mtx spin lock not only
  for smp_rendezvous_cpus() but also for the MD cache invalidation and TLB
  demapping IPIs.
- Mark some unused function arguments as such.

MFC after:	1 week

sys/kern/subr_witness.c

@@ -669,9 +669,6 @@ static struct witness_order_list_entry order_lists[] = {
 	 */
 	{ "intrcnt", &lock_class_mtx_spin },
 	{ "icu", &lock_class_mtx_spin },
-#if defined(SMP) && defined(__sparc64__)
-	{ "ipi", &lock_class_mtx_spin },
-#endif
 #ifdef __i386__
 	{ "allpmaps", &lock_class_mtx_spin },
 	{ "descriptor tables", &lock_class_mtx_spin },

sys/sparc64/include/smp.h

@@ -109,7 +109,6 @@ extern	cpu_ipi_single_t *cpu_ipi_single;
 
 void	mp_init(u_int cpu_impl);
 
-extern	struct mtx ipi_mtx;
 extern	struct ipi_cache_args ipi_cache_args;
 extern	struct ipi_rd_args ipi_rd_args;
 extern	struct ipi_tlb_args ipi_tlb_args;
@@ -169,7 +168,7 @@ ipi_dcache_page_inval(void *func, vm_paddr_t pa)
 		return (NULL);
 	sched_pin();
 	ica = &ipi_cache_args;
-	mtx_lock_spin(&ipi_mtx);
+	mtx_lock_spin(&smp_ipi_mtx);
 	ica->ica_mask = all_cpus;
 	CPU_CLR(PCPU_GET(cpuid), &ica->ica_mask);
 	ica->ica_pa = pa;
@@ -186,7 +185,7 @@ ipi_icache_page_inval(void *func, vm_paddr_t pa)
 		return (NULL);
 	sched_pin();
 	ica = &ipi_cache_args;
-	mtx_lock_spin(&ipi_mtx);
+	mtx_lock_spin(&smp_ipi_mtx);
 	ica->ica_mask = all_cpus;
 	CPU_CLR(PCPU_GET(cpuid), &ica->ica_mask);
 	ica->ica_pa = pa;
@@ -203,7 +202,6 @@ ipi_rd(u_int cpu, void *func, u_long *val)
 		return (NULL);
 	sched_pin();
 	ira = &ipi_rd_args;
-	mtx_lock_spin(&ipi_mtx);
 	CPU_SETOF(cpu, &ira->ira_mask);
 	ira->ira_val = val;
 	cpu_ipi_single(cpu, 0, (u_long)func, (u_long)ira);
@@ -227,7 +225,7 @@ ipi_tlb_context_demap(struct pmap *pm)
 		return (NULL);
 	}
 	ita = &ipi_tlb_args;
-	mtx_lock_spin(&ipi_mtx);
+	mtx_lock_spin(&smp_ipi_mtx);
 	ita->ita_mask = cpus;
 	ita->ita_pmap = pm;
 	cpu_ipi_selected(cpus, 0, (u_long)tl_ipi_tlb_context_demap,
@@ -252,7 +250,7 @@ ipi_tlb_page_demap(struct pmap *pm, vm_offset_t va)
 		return (NULL);
 	}
 	ita = &ipi_tlb_args;
-	mtx_lock_spin(&ipi_mtx);
+	mtx_lock_spin(&smp_ipi_mtx);
 	ita->ita_mask = cpus;
 	ita->ita_pmap = pm;
 	ita->ita_va = va;
@@ -277,7 +275,7 @@ ipi_tlb_range_demap(struct pmap *pm, vm_offset_t start, vm_offset_t end)
 		return (NULL);
 	}
 	ita = &ipi_tlb_args;
-	mtx_lock_spin(&ipi_mtx);
+	mtx_lock_spin(&smp_ipi_mtx);
 	ita->ita_mask = cpus;
 	ita->ita_pmap = pm;
 	ita->ita_start = start;
@@ -295,7 +293,19 @@ ipi_wait(void *cookie)
 	if ((mask = cookie) != NULL) {
 		while (!CPU_EMPTY(mask))
 			;
-		mtx_unlock_spin(&ipi_mtx);
+		mtx_unlock_spin(&smp_ipi_mtx);
+		sched_unpin();
+	}
+}
+
+static __inline void
+ipi_wait_unlocked(void *cookie)
+{
+	volatile cpuset_t *mask;
+
+	if ((mask = cookie) != NULL) {
+		while (!CPU_EMPTY(mask))
+			;
 		sched_unpin();
 	}
 }
@@ -352,7 +362,13 @@ ipi_tlb_range_demap(struct pmap *pm __unused, vm_offset_t start __unused,
 }
 
 static __inline void
-ipi_wait(void *cookie)
+ipi_wait(void *cookie __unused)
+{
+
+}
+
+static __inline void
+ipi_wait_unlocked(void *cookie __unused)
 {
 
 }

sys/sparc64/sparc64/mp_machdep.c

@@ -113,8 +113,6 @@ struct	ipi_rd_args ipi_rd_args;
 struct	ipi_tlb_args ipi_tlb_args;
 struct	pcb stoppcbs[MAXCPU];
 
-struct	mtx ipi_mtx;
-
 cpu_ipi_selected_t *cpu_ipi_selected;
 cpu_ipi_single_t *cpu_ipi_single;
 
@@ -280,8 +278,6 @@ void
 cpu_mp_start(void)
 {
 
-	mtx_init(&ipi_mtx, "ipi", NULL, MTX_SPIN);
-
 	intr_setup(PIL_AST, cpu_ipi_ast, -1, NULL, NULL);
 	intr_setup(PIL_RENDEZVOUS, (ih_func_t *)smp_rendezvous_action,
 	    -1, NULL, NULL);
@@ -503,13 +499,13 @@ cpu_mp_shutdown(void)
 }
 
 static void
-cpu_ipi_ast(struct trapframe *tf)
+cpu_ipi_ast(struct trapframe *tf __unused)
 {
 
 }
 
 static void
-cpu_ipi_stop(struct trapframe *tf)
+cpu_ipi_stop(struct trapframe *tf __unused)
 {
 	u_int cpuid;
 

sys/sparc64/sparc64/tick.c

@@ -332,7 +332,7 @@ stick_get_timecount_mp(struct timecounter *tc)
 	if (curcpu == 0)
 		stick = rdstick();
 	else
-		ipi_wait(ipi_rd(0, tl_ipi_stick_rd, &stick));
+		ipi_wait_unlocked(ipi_rd(0, tl_ipi_stick_rd, &stick));
 	sched_unpin();
 	return (stick);
 }
@@ -346,7 +346,7 @@ tick_get_timecount_mp(struct timecounter *tc)
 	if (curcpu == 0)
 		tick = rd(tick);
 	else
-		ipi_wait(ipi_rd(0, tl_ipi_tick_rd, &tick));
+		ipi_wait_unlocked(ipi_rd(0, tl_ipi_tick_rd, &tick));
 	sched_unpin();
 	return (tick);
 }