Merge branch 'sparc64-Use-low-latency-path-to-resume-idle-cpu'

Vijay Kumar says:

====================
sparc64: Use low latency path to resume idle cpu

CPU_POKE is a low latency path to resume the target cpu if suspended
using CPU_YIELD. Use CPU_POKE to resume cpu if supported by hypervisor.

	     Hackbench results (lower is better):
Number of
Process:		w/o fix		with fix
1  			0.012		 0.010
10			0.021		 0.019
100			0.151		 0.148

Changelog:
v2:
  - Fixed comments and spacing (2/2)
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

arch/sparc/include/asm/hypervisor.h

@@ -298,6 +298,24 @@ unsigned long sun4v_cpu_stop(unsigned long cpuid);
 unsigned long sun4v_cpu_yield(void);
 #endif
 
+/* cpu_poke()
+ * TRAP:	HV_FAST_TRAP
+ * FUNCTION:	HV_FAST_CPU_POKE
+ * RET0:	status
+ * ERRORS:	ENOCPU		cpuid refers to a CPU that does not exist
+ *		EINVAL		cpuid is current CPU
+ *
+ * Poke CPU cpuid. If the target CPU is currently suspended having
+ * invoked the cpu-yield service, that vCPU will be resumed.
+ * Poke interrupts may only be sent to valid, non-local CPUs.
+ * It is not legal to poke the current vCPU.
+ */
+#define HV_FAST_CPU_POKE                0x13
+
+#ifndef __ASSEMBLY__
+unsigned long sun4v_cpu_poke(unsigned long cpuid);
+#endif
+
 /* cpu_qconf()
  * TRAP:	HV_FAST_TRAP
  * FUNCTION:	HV_FAST_CPU_QCONF

arch/sparc/include/asm/smp_64.h

@@ -33,6 +33,9 @@
 DECLARE_PER_CPU(cpumask_t, cpu_sibling_map);
 extern cpumask_t cpu_core_map[NR_CPUS];
 
+void smp_init_cpu_poke(void);
+void scheduler_poke(void);
+
 void arch_send_call_function_single_ipi(int cpu);
 void arch_send_call_function_ipi_mask(const struct cpumask *mask);
 
@@ -74,6 +77,8 @@ void __cpu_die(unsigned int cpu);
 #define smp_fetch_global_regs() do { } while (0)
 #define smp_fetch_global_pmu() do { } while (0)
 #define smp_fill_in_cpu_possible_map() do { } while (0)
+#define smp_init_cpu_poke() do { } while (0)
+#define scheduler_poke() do { } while (0)
 
 #endif /* !(CONFIG_SMP) */
 

arch/sparc/kernel/hvapi.c

@@ -189,7 +189,7 @@ void __init sun4v_hvapi_init(void)
 
 	group = HV_GRP_CORE;
 	major = 1;
-	minor = 1;
+	minor = 6;
 	if (sun4v_hvapi_register(group, major, &minor))
 		goto bad;
 

arch/sparc/kernel/hvcalls.S

@@ -106,6 +106,17 @@ ENTRY(sun4v_cpu_yield)
 	 nop
 ENDPROC(sun4v_cpu_yield)
 
+	/* %o0: cpuid
+	 *
+	 * returns %o0:	status
+	 */
+ENTRY(sun4v_cpu_poke)
+	mov     HV_FAST_CPU_POKE, %o5
+	ta      HV_FAST_TRAP
+	retl
+	 nop
+ENDPROC(sun4v_cpu_poke)
+
 	/* %o0:	type
 	 * %o1:	queue paddr
 	 * %o2:	num queue entries

arch/sparc/kernel/process_64.c

@@ -77,8 +77,13 @@ void arch_cpu_idle(void)
 			: "=&r" (pstate)
 			: "i" (PSTATE_IE));
 
-		if (!need_resched() && !cpu_is_offline(smp_processor_id()))
+		if (!need_resched() && !cpu_is_offline(smp_processor_id())) {
 			sun4v_cpu_yield();
+			/* If resumed by cpu_poke then we need to explicitly
+			 * call scheduler_ipi().
+			 */
+			scheduler_poke();
+		}
 
 		/* Re-enable interrupts. */
 		__asm__ __volatile__(

arch/sparc/kernel/setup_64.c

@@ -356,6 +356,7 @@ void __init start_early_boot(void)
 	check_if_starfire();
 	per_cpu_patch();
 	sun4v_patch();
+	smp_init_cpu_poke();
 
 	cpu = hard_smp_processor_id();
 	if (cpu >= NR_CPUS) {

arch/sparc/kernel/smp_64.c

@@ -74,6 +74,9 @@ EXPORT_SYMBOL(cpu_core_sib_cache_map);
 
 static cpumask_t smp_commenced_mask;
 
+static DEFINE_PER_CPU(bool, poke);
+static bool cpu_poke;
+
 void smp_info(struct seq_file *m)
 {
 	int i;
@@ -1439,15 +1442,86 @@ void __init smp_cpus_done(unsigned int max_cpus)
 {
 }
 
+static void send_cpu_ipi(int cpu)
+{
+	xcall_deliver((u64) &xcall_receive_signal,
+			0, 0, cpumask_of(cpu));
+}
+
+void scheduler_poke(void)
+{
+	if (!cpu_poke)
+		return;
+
+	if (!__this_cpu_read(poke))
+		return;
+
+	__this_cpu_write(poke, false);
+	set_softint(1 << PIL_SMP_RECEIVE_SIGNAL);
+}
+
+static unsigned long send_cpu_poke(int cpu)
+{
+	unsigned long hv_err;
+
+	per_cpu(poke, cpu) = true;
+	hv_err = sun4v_cpu_poke(cpu);
+	if (hv_err != HV_EOK) {
+		per_cpu(poke, cpu) = false;
+		pr_err_ratelimited("%s: sun4v_cpu_poke() fails err=%lu\n",
+				    __func__, hv_err);
+	}
+
+	return hv_err;
+}
+
 void smp_send_reschedule(int cpu)
 {
 	if (cpu == smp_processor_id()) {
 		WARN_ON_ONCE(preemptible());
 		set_softint(1 << PIL_SMP_RECEIVE_SIGNAL);
-	} else {
-		xcall_deliver((u64) &xcall_receive_signal,
-			      0, 0, cpumask_of(cpu));
+		return;
 	}
+
+	/* Use cpu poke to resume idle cpu if supported. */
+	if (cpu_poke && idle_cpu(cpu)) {
+		unsigned long ret;
+
+		ret = send_cpu_poke(cpu);
+		if (ret == HV_EOK)
+			return;
+	}
+
+	/* Use IPI in following cases:
+	 * - cpu poke not supported
+	 * - cpu not idle
+	 * - send_cpu_poke() returns with error
+	 */
+	send_cpu_ipi(cpu);
+}
+
+void smp_init_cpu_poke(void)
+{
+	unsigned long major;
+	unsigned long minor;
+	int ret;
+
+	if (tlb_type != hypervisor)
+		return;
+
+	ret = sun4v_hvapi_get(HV_GRP_CORE, &major, &minor);
+	if (ret) {
+		pr_debug("HV_GRP_CORE is not registered\n");
+		return;
+	}
+
+	if (major == 1 && minor >= 6) {
+		/* CPU POKE is registered. */
+		cpu_poke = true;
+		return;
+	}
+
+	pr_debug("CPU_POKE not supported\n");
 }
 
 void __irq_entry smp_receive_signal_client(int irq, struct pt_regs *regs)