From a7706e0c39a01c1786b2599503cf86ac08e1b79f Mon Sep 17 00:00:00 2001
From: Zhao Yakui <yakui.zhao@intel.com>
Date: Tue, 27 Aug 2019 16:25:39 +0800
Subject: [PATCH] HV: Use the mwait instead of pause for cpu_idle

Now it will use the pause when cpu is in idle state. It will consume
more power. If the mwait is supported, it will use the monitor/mwait to
enter the deep CPU C-state. It helps to save power.

Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
---
 hypervisor/arch/x86/cpu.c            | 39 ++++++++++++++++++----------
 hypervisor/common/hv_main.c          |  2 --
 hypervisor/common/schedule.c         | 19 ++++++++++++++
 hypervisor/include/common/schedule.h |  6 +++--
 4 files changed, 49 insertions(+), 17 deletions(-)

diff --git a/hypervisor/arch/x86/cpu.c b/hypervisor/arch/x86/cpu.c
index 7e1d5c1dc..19d0234cb 100644
--- a/hypervisor/arch/x86/cpu.c
+++ b/hypervisor/arch/x86/cpu.c
@@ -367,9 +367,34 @@ void stop_pcpus(void)
 	wait_pcpus_offline(mask);
 }
 
+static
+inline void asm_monitor(const uint64_t *addr, uint64_t ecx, uint64_t edx)
+{
+	asm volatile("monitor\n" : : "a" (addr), "c" (ecx), "d" (edx));
+}
+
+static
+inline void asm_mwait(uint64_t eax, uint64_t ecx)
+{
+	asm volatile("mwait\n" : : "a" (eax), "c" (ecx));
+}
+
 void cpu_do_idle(void)
 {
-	asm_pause();
+	uint16_t pcpu_id = get_pcpu_id();
+	struct sched_context *ctx = &per_cpu(sched_ctx, pcpu_id);
+
+	if (ctx->mwait_flags) {
+		CPU_IRQ_DISABLE();
+		asm_monitor(&ctx->flags, 0UL, 0UL);
+		if (!bitmap_test(NEED_RESCHEDULE, &ctx->flags))
+			asm_mwait(0x60UL, 1UL);
+		CPU_IRQ_ENABLE();
+	} else {
+		CPU_IRQ_ENABLE();
+		asm_pause();
+		CPU_IRQ_DISABLE();
+	}
 }
 
 /**
@@ -415,18 +440,6 @@ static void print_hv_banner(void)
 	printf(boot_msg);
 }
 
-static
-inline void asm_monitor(const uint64_t *addr, uint64_t ecx, uint64_t edx)
-{
-	asm volatile("monitor\n" : : "a" (addr), "c" (ecx), "d" (edx));
-}
-
-static
-inline void asm_mwait(uint64_t eax, uint64_t ecx)
-{
-	asm volatile("mwait\n" : : "a" (eax), "c" (ecx));
-}
-
 /* wait until *sync == wake_sync */
 void wait_sync_change(uint64_t *sync, uint64_t wake_sync)
 {
diff --git a/hypervisor/common/hv_main.c b/hypervisor/common/hv_main.c
index 67e8c0cc7..70926751e 100644
--- a/hypervisor/common/hv_main.c
+++ b/hypervisor/common/hv_main.c
@@ -92,9 +92,7 @@ void default_idle(__unused struct sched_object *obj)
 		} else if (need_shutdown_vm(pcpu_id)) {
 			shutdown_vm_from_idle(pcpu_id);
 		} else {
-			CPU_IRQ_ENABLE();
 			cpu_do_idle();
-			CPU_IRQ_DISABLE();
 		}
 	}
 }
diff --git a/hypervisor/common/schedule.c b/hypervisor/common/schedule.c
index c54bbf97e..5b6c7b327 100644
--- a/hypervisor/common/schedule.c
+++ b/hypervisor/common/schedule.c
@@ -12,6 +12,12 @@
 #include <lapic.h>
 #include <schedule.h>
 #include <sprintf.h>
+#include <cpuid.h>
+#include <cpu_caps.h>
+
+#define CPUID_MWAIT_LEAF                5
+#define CPUID5_ECX_EXTENSIONS_SUPPORTED 0x1
+#define CPUID5_ECX_INTERRUPT_BREAK      0x2
 
 static uint64_t pcpu_used_bitmap;
 
@@ -20,7 +26,19 @@ void init_scheduler(void)
 	struct sched_context *ctx;
 	uint32_t i;
 	uint16_t pcpu_nums = get_pcpu_nums();
+	uint32_t mwait_flag = 0;
 
+	if (has_monitor_cap()) {
+		uint32_t eax, ebx, ecx, edx;
+
+		mwait_flag = 1;
+		cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &edx);
+		if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
+			!(ecx & CPUID5_ECX_INTERRUPT_BREAK) ||
+			!edx) {
+			mwait_flag = 0;
+		}
+	}
 	for (i = 0U; i < pcpu_nums; i++) {
 		ctx = &per_cpu(sched_ctx, i);
 
@@ -29,6 +47,7 @@ void init_scheduler(void)
 		INIT_LIST_HEAD(&ctx->runqueue);
 		ctx->flags = 0UL;
 		ctx->curr_obj = NULL;
+		ctx->mwait_flags = mwait_flag;
 	}
 }
 
diff --git a/hypervisor/include/common/schedule.h b/hypervisor/include/common/schedule.h
index 3cd45a915..29bd3e8c1 100644
--- a/hypervisor/include/common/schedule.h
+++ b/hypervisor/include/common/schedule.h
@@ -28,12 +28,14 @@ struct sched_object {
 };
 
 struct sched_context {
+	uint64_t flags;
+	uint64_t mwait_flags;
+	uint64_t rserved[6];
 	spinlock_t runqueue_lock;
 	struct list_head runqueue;
-	uint64_t flags;
 	struct sched_object *curr_obj;
 	spinlock_t scheduler_lock;
-};
+} __aligned(64);
 
 void init_scheduler(void);
 void switch_to_idle(run_thread_t idle_thread);