hv: add priority based scheduler

This patch adds a new priority based scheduler to support
vCPU scheduling based on their pre-configured priorities.
A vCPU can be running only if there is no higher priority
vCPU running on the same pCPU.

Tracked-On: #6571
Signed-off-by: Jie Deng <jie.deng@intel.com>

hypervisor/Makefile

@@ -244,6 +244,9 @@ endif
 ifeq ($(CONFIG_SCHED_BVT),y)
 HW_C_SRCS += common/sched_bvt.c
 endif
+ifeq ($(CONFIG_SCHED_PRIO),y)
+HW_C_SRCS += common/sched_prio.c
+endif
 HW_C_SRCS += hw/pci.c
 HW_C_SRCS += arch/x86/configs/vm_config.c
 HW_C_SRCS += boot/acpi_base.c

hypervisor/arch/x86/guest/vcpu.c

@@ -963,6 +963,7 @@ int32_t prepare_vcpu(struct acrn_vm *vm, uint16_t pcpu_id)
 		vcpu->thread_obj.host_sp = build_stack_frame(vcpu);
 		vcpu->thread_obj.switch_out = context_switch_out;
 		vcpu->thread_obj.switch_in = context_switch_in;
+		vcpu->thread_obj.priority = get_vm_config(vm->vm_id)->vm_prio;
 		init_thread_data(&vcpu->thread_obj);
 		for (i = 0; i < VCPU_EVENT_NUM; i++) {
 			init_event(&vcpu->events[i]);

hypervisor/common/hv_main.c

@@ -105,6 +105,7 @@ void run_idle_thread(void)
 	idle->thread_entry = default_idle;
 	idle->switch_out = NULL;
 	idle->switch_in = NULL;
+	idle->priority = PRIO_IDLE;
 
 	run_thread(idle);
 

hypervisor/common/sched_prio.c

@@ -0,0 +1,98 @@
+/*
+ * Copyright (C) 2021 Intel Corporation.
+ *
+ * SPDX-License-Identifier: BSD-3-Clause
+ */
+
+#include <list.h>
+#include <asm/per_cpu.h>
+#include <schedule.h>
+
+struct sched_prio_data {
+	/* keep list as the first item */
+	struct list_head list;
+};
+
+static int sched_prio_init(struct sched_control *ctl)
+{
+	struct sched_prio_control *prio_ctl = &per_cpu(sched_prio_ctl, ctl->pcpu_id);
+
+	ASSERT(ctl->pcpu_id == get_pcpu_id(), "Init scheduler on wrong CPU!");
+
+	ctl->priv = prio_ctl;
+	INIT_LIST_HEAD(&prio_ctl->prio_queue);
+
+	return 0;
+}
+
+static void sched_prio_init_data(struct thread_object *obj)
+{
+	struct sched_prio_data *data;
+
+	data = (struct sched_prio_data *)obj->data;
+	INIT_LIST_HEAD(&data->list);
+}
+
+static struct thread_object *sched_prio_pick_next(struct sched_control *ctl)
+{
+	struct sched_prio_control *prio_ctl = (struct sched_prio_control *)ctl->priv;
+	struct thread_object *next = NULL;
+
+	if (!list_empty(&prio_ctl->prio_queue)) {
+		next = get_first_item(&prio_ctl->prio_queue, struct thread_object, data);
+	} else {
+		next = &get_cpu_var(idle);
+	}
+
+	return next;
+}
+
+static void prio_queue_add(struct thread_object *obj)
+{
+	struct sched_prio_control *prio_ctl =
+		(struct sched_prio_control *)obj->sched_ctl->priv;
+	struct sched_prio_data *data = (struct sched_prio_data *)obj->data;
+	struct thread_object *iter_obj;
+	struct list_head *pos;
+
+	if (list_empty(&prio_ctl->prio_queue)) {
+		list_add(&data->list, &prio_ctl->prio_queue);
+	} else {
+		list_for_each(pos, &prio_ctl->prio_queue) {
+			iter_obj = container_of(pos, struct thread_object, data);
+			if (iter_obj->priority < obj->priority) {
+				list_add_node(&data->list, pos->prev, pos);
+				break;
+			}
+		}
+		if (list_empty(&data->list)) {
+			list_add_tail(&data->list, &prio_ctl->prio_queue);
+		}
+	}
+}
+
+static void prio_queue_remove(struct thread_object *obj)
+{
+	struct sched_prio_data *data = (struct sched_prio_data *)obj->data;
+
+	list_del_init(&data->list);
+}
+
+static void sched_prio_sleep(struct thread_object *obj)
+{
+	prio_queue_remove(obj);
+}
+
+static void sched_prio_wake(struct thread_object *obj)
+{
+	prio_queue_add(obj);
+}
+
+struct acrn_scheduler sched_prio = {
+	.name		= "sched_prio",
+	.init		= sched_prio_init,
+	.init_data	= sched_prio_init_data,
+	.pick_next	= sched_prio_pick_next,
+	.sleep		= sched_prio_sleep,
+	.wake		= sched_prio_wake,
+};

hypervisor/common/schedule.c

@@ -77,6 +77,9 @@ void init_sched(uint16_t pcpu_id)
 #endif
 #ifdef CONFIG_SCHED_BVT
 	ctl->scheduler = &sched_bvt;
+#endif
+#ifdef CONFIG_SCHED_PRIO
+	ctl->scheduler = &sched_prio;
 #endif
 	if (ctl->scheduler->init != NULL) {
 		ctl->scheduler->init(ctl);

hypervisor/include/arch/x86/asm/per_cpu.h

@@ -41,6 +41,7 @@ struct per_cpu_region {
 	struct sched_noop_control sched_noop_ctl;
 	struct sched_iorr_control sched_iorr_ctl;
 	struct sched_bvt_control sched_bvt_ctl;
+	struct sched_prio_control sched_prio_ctl;
 	struct thread_object idle;
 	struct host_gdt gdt;
 	struct tss_64 tss;

hypervisor/include/arch/x86/asm/vm_config.h

@@ -169,6 +169,7 @@ struct acrn_vm_config {
 							 *	GUEST_FLAG_LAPIC_PASSTHROUGH
 							 * We could add more guest flags in future;
 							 */
+	uint32_t vm_prio;				/* The priority for VM vCPU scheduling */
 	struct acrn_vm_mem_config memory;		/* memory configuration of VM */
 	struct epc_section epc;				/* EPC memory configuration of VM */
 	uint16_t pci_dev_num;				/* indicate how many PCI devices in VM */

hypervisor/include/common/schedule.h

@@ -28,6 +28,14 @@ enum sched_notify_mode {
 	SCHED_NOTIFY_IPI
 };
 
+/* Tools can configure a VM to use PRIO_LOW or PRIO_HIGH */
+enum thread_priority {
+	PRIO_IDLE = 0,
+	PRIO_LOW,
+	PRIO_HIGH,
+	PRIO_MAX
+};
+
 struct thread_object;
 typedef void (*thread_entry_t)(struct thread_object *obj);
 typedef void (*switch_t)(struct thread_object *obj);
@@ -44,6 +52,8 @@ struct thread_object {
 	switch_t switch_out;
 	switch_t switch_in;
 
+	int priority;
+
 	uint8_t data[THREAD_DATA_SIZE];
 };
 
@@ -97,6 +107,11 @@ struct sched_bvt_control {
 	struct hv_timer tick_timer;
 };
 
+extern struct acrn_scheduler sched_prio;
+struct sched_prio_control {
+	struct list_head prio_queue;
+};
+
 bool is_idle_thread(const struct thread_object *obj);
 uint16_t sched_get_pcpuid(const struct thread_object *obj);
 struct thread_object *sched_get_current(uint16_t pcpu_id);