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);

misc/config_tools/library/hv_cfg_lib.py

@@ -12,7 +12,7 @@ import board_cfg_lib
 
 ERR_LIST = {}
 N_Y = ['n', 'y']
-SCHEDULER_TYPE = ['SCHED_NOOP', 'SCHED_IORR', 'SCHED_BVT']
+SCHEDULER_TYPE = ['SCHED_NOOP', 'SCHED_IORR', 'SCHED_BVT', 'SCHED_PRIO']
 
 RANGE_DB = {
     'LOG_LEVEL':{'min':0,'max':6},

misc/config_tools/schema/config.xsd

@@ -400,6 +400,11 @@ Refer SDM 17.19.2 for details, and use with caution.</xs:documentation>
         <xs:documentation>Specify memory information for Service and User VMs.</xs:documentation>
       </xs:annotation>
     </xs:element>
+    <xs:element name="priority" type="PriorityType"  default="PRIO_LOW">
+      <xs:annotation>
+        <xs:documentation>Specify the VM vCPU priority for scheduling.</xs:documentation>
+      </xs:annotation>
+    </xs:element>
     <xs:element name="os_config" type="OSConfigurations" minOccurs="0">
       <xs:annotation>
         <xs:documentation>General information for host kernel, boot

misc/config_tools/schema/types.xsd

@@ -101,7 +101,7 @@ higher value (lower severity) are discarded.</xs:documentation>
 
 <xs:simpleType name="SchedulerType">
   <xs:annotation>
-    <xs:documentation>Three scheduler options are supported:
+    <xs:documentation>Four scheduler options are supported:
 
 - ``SCHED_NOOP``: The NOOP (No-Operation) scheduler means there is a
   strict 1 to 1 mapping between vCPUs and pCPUs.
@@ -113,6 +113,8 @@ higher value (lower severity) are discarded.</xs:documentation>
   earliest effective virtual time. *TODO: BVT scheduler will be built on
   top of a prioritized scheduling mechanism, i.e. higher priority threads
   get scheduled first, and same priority tasks are scheduled per BVT.*
+- ``SCHED_PRIO``: The priority based scheduler. vCPU scheduling will be based on
+  their pre-configured priorities.
 
 Read more about the available scheduling options in :ref:`cpu_sharing`.</xs:documentation>
   </xs:annotation>
@@ -120,6 +122,21 @@ Read more about the available scheduling options in :ref:`cpu_sharing`.</xs:docu
     <xs:enumeration value="SCHED_NOOP" />
     <xs:enumeration value="SCHED_IORR" />
     <xs:enumeration value="SCHED_BVT" />
+    <xs:enumeration value="SCHED_PRIO" />
+  </xs:restriction>
+</xs:simpleType>
+
+<xs:simpleType name="PriorityType">
+  <xs:annotation>
+    <xs:documentation>Two priorities are supported for priority based scheduler:
+
+- ``PRIO_LOW``: low priority for vCPU scheduling.
+- ``PRIO_HIGH``: high priority for vCPU scheduling.
+    </xs:documentation>
+  </xs:annotation>
+  <xs:restriction base="xs:string">
+    <xs:enumeration value="PRIO_LOW" />
+    <xs:enumeration value="PRIO_HIGH" />
   </xs:restriction>
 </xs:simpleType>
 

misc/config_tools/xforms/vm_configurations.c.xsl

@@ -18,6 +18,7 @@
     <xsl:value-of select="acrn:include('vuart.h')" />
     <xsl:value-of select="acrn:include('asm/pci_dev.h')" />
     <xsl:value-of select="acrn:include('asm/pgtable.h')" />
+    <xsl:value-of select="acrn:include('schedule.h')" />
 
     <xsl:apply-templates select="config-data/acrn-config" />
   </xsl:template>
@@ -67,6 +68,7 @@
       <xsl:value-of select="acrn:comment('Allow Service VM to reboot the system since it is the highest priority VM.')" />
       <xsl:value-of select="$newline" />
     </xsl:if>
+    <xsl:value-of select="acrn:initializer('vm_prio', priority)" />
     <xsl:apply-templates select="guest_flags" />
     <xsl:apply-templates select="clos" />
     <xsl:call-template name="cpu_affinity" />