[Qemu-devel] [RFC v4 0/3] vITS save/restore

[Qemu-devel] [RFC v4 0/3] vITS save/restore

[Eric Auger]

This series allows ITS save/restore and migration use cases.
It relies on not upstreamed kernel series [1].

ITS tables are flushed into guest RAM on VM stop while registers
are save on pre_save() callback. Tables and registers are restored
on ITS post_load().

That work was tested on Cavium ThunderX using virsh save/restore and
virt-manager live migration.

Best Regards

Eric

Host Kernel dependencies:
- [1] [PATCH v4 00/22] vITS save/restore

History:
v3 -> v4:
- oversight in v3, missed a last minute correction related to
  reg useless declaration in kvm_arm_its_pre_save

v2 -> v3:
- GITS_IIDR is now saved and restored to check ABI revision.
- get/put functions renamed into pre_save/post_load
- unmigratable = false removed
- changed the migration blocker message
- remove the extract64 round s->ctlr
- reword some comments

v1 -> v2:
- rebase on 2.9 soft release code
- handle case where migrate_add_blocker fails
- add comments along with ITS and GICv3 migration priorities


Eric Auger (3):
  linux-headers: Partial header update for vITS save/restore
  hw/intc/arm_gicv3_its: Implement state save/restore
  hw/intc/arm_gicv3_its: Allow save/restore

 hw/intc/arm_gicv3_common.c             |   1 +
 hw/intc/arm_gicv3_its_common.c         |  11 ++-
 hw/intc/arm_gicv3_its_kvm.c            | 119 +++++++++++++++++++++++++++++----
 include/hw/intc/arm_gicv3_its_common.h |   8 +++
 include/migration/vmstate.h            |   2 +
 linux-headers/asm-arm/kvm.h            |   2 +
 linux-headers/asm-arm64/kvm.h          |   2 +
 7 files changed, 132 insertions(+), 13 deletions(-)

-- 
2.5.5

[Qemu-devel] [RFC v4 1/3] linux-headers: Partial header update for vITS save/restore

[Eric Auger]

This is a linux header update against 4.11-rc3 plus the non
upstreamed ITS migration series.

https://github.com/eauger/linux/tree/v4.11-rc3-its-mig-v4

It aims at enhancing the KVM user API with vITS save/restore
capability. This consists in two new groups for the
ARM_VGIC_ITS KVM device, named: KVM_DEV_ARM_VGIC_GRP_ITS_REGS,
KVM_DEV_ARM_VGIC_GRP_ITS_TABLES.

Signed-off-by: Eric Auger <eric.auger@redhat.com>
---
 linux-headers/asm-arm/kvm.h   | 2 ++
 linux-headers/asm-arm64/kvm.h | 2 ++
 2 files changed, 4 insertions(+)

diff --git a/linux-headers/asm-arm/kvm.h b/linux-headers/asm-arm/kvm.h
index 1101d55..c8b9a68 100644
--- a/linux-headers/asm-arm/kvm.h
+++ b/linux-headers/asm-arm/kvm.h
@@ -192,6 +192,8 @@ struct kvm_arch_memory_slot {
 #define KVM_DEV_ARM_VGIC_GRP_REDIST_REGS 5
 #define KVM_DEV_ARM_VGIC_GRP_CPU_SYSREGS 6
 #define KVM_DEV_ARM_VGIC_GRP_LEVEL_INFO  7
+#define KVM_DEV_ARM_VGIC_GRP_ITS_REGS	8
+#define KVM_DEV_ARM_VGIC_GRP_ITS_TABLES	9
 #define KVM_DEV_ARM_VGIC_LINE_LEVEL_INFO_SHIFT	10
 #define KVM_DEV_ARM_VGIC_LINE_LEVEL_INFO_MASK \
 			(0x3fffffULL << KVM_DEV_ARM_VGIC_LINE_LEVEL_INFO_SHIFT)
diff --git a/linux-headers/asm-arm64/kvm.h b/linux-headers/asm-arm64/kvm.h
index 651ec30..a8d7d67 100644
--- a/linux-headers/asm-arm64/kvm.h
+++ b/linux-headers/asm-arm64/kvm.h
@@ -212,6 +212,8 @@ struct kvm_arch_memory_slot {
 #define KVM_DEV_ARM_VGIC_GRP_REDIST_REGS 5
 #define KVM_DEV_ARM_VGIC_GRP_CPU_SYSREGS 6
 #define KVM_DEV_ARM_VGIC_GRP_LEVEL_INFO  7
+#define KVM_DEV_ARM_VGIC_GRP_ITS_REGS 8
+#define KVM_DEV_ARM_VGIC_GRP_ITS_TABLES 9
 #define KVM_DEV_ARM_VGIC_LINE_LEVEL_INFO_SHIFT	10
 #define KVM_DEV_ARM_VGIC_LINE_LEVEL_INFO_MASK \
 			(0x3fffffULL << KVM_DEV_ARM_VGIC_LINE_LEVEL_INFO_SHIFT)
-- 
2.5.5

[Qemu-devel] [RFC v4 2/3] hw/intc/arm_gicv3_its: Implement state save/restore

[Eric Auger]

We need to handle both registers and ITS tables. While
register handling is standard, ITS table handling is more
challenging since the kernel API is devised so that the
tables are flushed into guest RAM and not in vmstate buffers.

Flushing the ITS tables on device pre_save() is too late
since the guest RAM is already saved at this point.

Table flushing needs to happen when we are sure the vcpus
are stopped and before the last dirty page saving. The
right point is RUN_STATE_FINISH_MIGRATE but sometimes the
VM gets stopped before migration launch so let's simply
flush the tables each time the VM gets stopped.

For regular ITS registers we just can use vmstate pre_save()
and post_load() callbacks.

Signed-off-by: Eric Auger <eric.auger@redhat.com>

---
v3 -> v4:
- remove useless reg declaration in pre_save()

v2 -> v3:
- take into account Peter's comments:
  - add iidr save/restore
  - reword comments
  - remove s->ctlr = extract64(reg, 0, 32);
  - rename get()/put() function into pre_save()/post_load()
- do not execute put() if iidr == 0

Signed-off-by: Eric Auger <eric.auger@redhat.com>
---
 hw/intc/arm_gicv3_its_common.c         |  9 ++++
 hw/intc/arm_gicv3_its_kvm.c            | 95 ++++++++++++++++++++++++++++++++++
 include/hw/intc/arm_gicv3_its_common.h |  8 +++
 3 files changed, 112 insertions(+)

diff --git a/hw/intc/arm_gicv3_its_common.c b/hw/intc/arm_gicv3_its_common.c
index 9d67c5c..efab8c7 100644
--- a/hw/intc/arm_gicv3_its_common.c
+++ b/hw/intc/arm_gicv3_its_common.c
@@ -49,6 +49,15 @@ static const VMStateDescription vmstate_its = {
     .pre_save = gicv3_its_pre_save,
     .post_load = gicv3_its_post_load,
     .unmigratable = true,
+    .fields = (VMStateField[]) {
+        VMSTATE_UINT32(ctlr, GICv3ITSState),
+        VMSTATE_UINT32(iidr, GICv3ITSState),
+        VMSTATE_UINT64(cbaser, GICv3ITSState),
+        VMSTATE_UINT64(cwriter, GICv3ITSState),
+        VMSTATE_UINT64(creadr, GICv3ITSState),
+        VMSTATE_UINT64_ARRAY(baser, GICv3ITSState, 8),
+        VMSTATE_END_OF_LIST()
+    },
 };
 
 static MemTxResult gicv3_its_trans_read(void *opaque, hwaddr offset,
diff --git a/hw/intc/arm_gicv3_its_kvm.c b/hw/intc/arm_gicv3_its_kvm.c
index bd4f3aa..bc4722b 100644
--- a/hw/intc/arm_gicv3_its_kvm.c
+++ b/hw/intc/arm_gicv3_its_kvm.c
@@ -53,6 +53,24 @@ static int kvm_its_send_msi(GICv3ITSState *s, uint32_t value, uint16_t devid)
     return kvm_vm_ioctl(kvm_state, KVM_SIGNAL_MSI, &msi);
 }
 
+/**
+ * vm_change_state_handler - VM change state callback aiming at flushing
+ * ITS tables into guest RAM
+ *
+ * The tables get flushed to guest RAM whenever the VM gets stopped.
+ */
+static void vm_change_state_handler(void *opaque, int running,
+                                    RunState state)
+{
+    GICv3ITSState *s = (GICv3ITSState *)opaque;
+
+    if (running) {
+        return;
+    }
+    kvm_device_access(s->dev_fd, KVM_DEV_ARM_VGIC_GRP_ITS_TABLES,
+                      0, NULL, false);
+}
+
 static void kvm_arm_its_realize(DeviceState *dev, Error **errp)
 {
     GICv3ITSState *s = ARM_GICV3_ITS_COMMON(dev);
@@ -89,6 +107,8 @@ static void kvm_arm_its_realize(DeviceState *dev, Error **errp)
     kvm_msi_use_devid = true;
     kvm_gsi_direct_mapping = false;
     kvm_msi_via_irqfd_allowed = kvm_irqfds_enabled();
+
+    qemu_add_vm_change_state_handler(vm_change_state_handler, s);
 }
 
 static void kvm_arm_its_init(Object *obj)
@@ -102,6 +122,79 @@ static void kvm_arm_its_init(Object *obj)
                              &error_abort);
 }
 
+/**
+ * kvm_arm_its_pre_save - handles the saving of ITS registers.
+ * ITS tables are flushed into guest RAM separately and earlier,
+ * through the VM change state handler, since at the moment pre_save()
+ * is called, the guest RAM has already been saved.
+ */
+static void kvm_arm_its_pre_save(GICv3ITSState *s)
+{
+    int i;
+
+    for (i = 0; i < 8; i++) {
+        kvm_device_access(s->dev_fd, KVM_DEV_ARM_VGIC_GRP_ITS_REGS,
+                          GITS_BASER + i * 8, &s->baser[i], false);
+    }
+
+    kvm_device_access(s->dev_fd, KVM_DEV_ARM_VGIC_GRP_ITS_REGS,
+                      GITS_CTLR, &s->ctlr, false);
+
+    kvm_device_access(s->dev_fd, KVM_DEV_ARM_VGIC_GRP_ITS_REGS,
+                      GITS_CBASER, &s->cbaser, false);
+
+    kvm_device_access(s->dev_fd, KVM_DEV_ARM_VGIC_GRP_ITS_REGS,
+                      GITS_CREADR, &s->creadr, false);
+
+    kvm_device_access(s->dev_fd, KVM_DEV_ARM_VGIC_GRP_ITS_REGS,
+                      GITS_CWRITER, &s->cwriter, false);
+
+    kvm_device_access(s->dev_fd, KVM_DEV_ARM_VGIC_GRP_ITS_REGS,
+                      GITS_IIDR, &s->iidr, false);
+}
+
+/**
+ * kvm_arm_its_post_load - Restore both the ITS registers and tables
+ */
+static void kvm_arm_its_post_load(GICv3ITSState *s)
+{
+    uint64_t reg;
+    int i;
+
+    if (!s->iidr) {
+        return;
+    }
+
+    kvm_device_access(s->dev_fd, KVM_DEV_ARM_VGIC_GRP_ITS_REGS,
+                      GITS_IIDR, &s->iidr, true);
+
+    /*
+     * must be written before GITS_CREADR since GITS_CBASER write
+     * access resets GITS_CREADR.
+     */
+    kvm_device_access(s->dev_fd, KVM_DEV_ARM_VGIC_GRP_ITS_REGS,
+                      GITS_CBASER, &s->cbaser, true);
+
+    kvm_device_access(s->dev_fd, KVM_DEV_ARM_VGIC_GRP_ITS_REGS,
+                      GITS_CREADR, &s->creadr, true);
+
+    kvm_device_access(s->dev_fd, KVM_DEV_ARM_VGIC_GRP_ITS_REGS,
+                      GITS_CWRITER, &s->cwriter, true);
+
+
+    for (i = 0; i < 8; i++) {
+        kvm_device_access(s->dev_fd, KVM_DEV_ARM_VGIC_GRP_ITS_REGS,
+                          GITS_BASER + i * 8, &s->baser[i], true);
+    }
+
+    kvm_device_access(s->dev_fd, KVM_DEV_ARM_VGIC_GRP_ITS_TABLES,
+                      0, NULL, true);
+
+    reg = s->ctlr;
+    kvm_device_access(s->dev_fd, KVM_DEV_ARM_VGIC_GRP_ITS_REGS,
+                      GITS_CTLR, &reg, true);
+}
+
 static void kvm_arm_its_class_init(ObjectClass *klass, void *data)
 {
     DeviceClass *dc = DEVICE_CLASS(klass);
@@ -109,6 +202,8 @@ static void kvm_arm_its_class_init(ObjectClass *klass, void *data)
 
     dc->realize = kvm_arm_its_realize;
     icc->send_msi = kvm_its_send_msi;
+    icc->pre_save = kvm_arm_its_pre_save;
+    icc->post_load = kvm_arm_its_post_load;
 }
 
 static const TypeInfo kvm_arm_its_info = {
diff --git a/include/hw/intc/arm_gicv3_its_common.h b/include/hw/intc/arm_gicv3_its_common.h
index 1ba1894..fd1fe64 100644
--- a/include/hw/intc/arm_gicv3_its_common.h
+++ b/include/hw/intc/arm_gicv3_its_common.h
@@ -28,6 +28,13 @@
 #define ITS_TRANS_SIZE   0x10000
 #define ITS_SIZE         (ITS_CONTROL_SIZE + ITS_TRANS_SIZE)
 
+#define GITS_CTLR        0x0
+#define GITS_IIDR        0x4
+#define GITS_CBASER      0x80
+#define GITS_CWRITER     0x88
+#define GITS_CREADR      0x90
+#define GITS_BASER       0x100
+
 struct GICv3ITSState {
     SysBusDevice parent_obj;
 
@@ -43,6 +50,7 @@ struct GICv3ITSState {
 
     /* Registers */
     uint32_t ctlr;
+    uint32_t iidr;
     uint64_t cbaser;
     uint64_t cwriter;
     uint64_t creadr;
-- 
2.5.5

[Qemu-devel] [RFC v4 3/3] hw/intc/arm_gicv3_its: Allow save/restore

[Eric Auger]

We change the restoration priority of both the GICv3 and ITS. The
GICv3 must be restored before the ITS and the ITS needs to be restored
before PCIe devices since it translates their MSI transactions.

Signed-off-by: Eric Auger <eric.auger@redhat.com>

---
v2 -> v3:
- reword migration blocker message
- remove unmigratable setting to false

v1 -> v2:
- handle case where migrate_add_blocker fails
- add comments along with ITS and GICv3 migration priorities
---
 hw/intc/arm_gicv3_common.c     |  1 +
 hw/intc/arm_gicv3_its_common.c |  2 +-
 hw/intc/arm_gicv3_its_kvm.c    | 24 ++++++++++++------------
 include/migration/vmstate.h    |  2 ++
 4 files changed, 16 insertions(+), 13 deletions(-)

diff --git a/hw/intc/arm_gicv3_common.c b/hw/intc/arm_gicv3_common.c
index c6493d6..4228b7c 100644
--- a/hw/intc/arm_gicv3_common.c
+++ b/hw/intc/arm_gicv3_common.c
@@ -145,6 +145,7 @@ static const VMStateDescription vmstate_gicv3 = {
     .minimum_version_id = 1,
     .pre_save = gicv3_pre_save,
     .post_load = gicv3_post_load,
+    .priority = MIG_PRI_GICV3,
     .fields = (VMStateField[]) {
         VMSTATE_UINT32(gicd_ctlr, GICv3State),
         VMSTATE_UINT32_ARRAY(gicd_statusr, GICv3State, 2),
diff --git a/hw/intc/arm_gicv3_its_common.c b/hw/intc/arm_gicv3_its_common.c
index efab8c7..22ce4c4 100644
--- a/hw/intc/arm_gicv3_its_common.c
+++ b/hw/intc/arm_gicv3_its_common.c
@@ -48,7 +48,7 @@ static const VMStateDescription vmstate_its = {
     .name = "arm_gicv3_its",
     .pre_save = gicv3_its_pre_save,
     .post_load = gicv3_its_post_load,
-    .unmigratable = true,
+    .priority = MIG_PRI_GICV3_ITS,
     .fields = (VMStateField[]) {
         VMSTATE_UINT32(ctlr, GICv3ITSState),
         VMSTATE_UINT32(iidr, GICv3ITSState),
diff --git a/hw/intc/arm_gicv3_its_kvm.c b/hw/intc/arm_gicv3_its_kvm.c
index bc4722b..9787c4c 100644
--- a/hw/intc/arm_gicv3_its_kvm.c
+++ b/hw/intc/arm_gicv3_its_kvm.c
@@ -76,18 +76,6 @@ static void kvm_arm_its_realize(DeviceState *dev, Error **errp)
     GICv3ITSState *s = ARM_GICV3_ITS_COMMON(dev);
     Error *local_err = NULL;
 
-    /*
-     * Block migration of a KVM GICv3 ITS device: the API for saving and
-     * restoring the state in the kernel is not yet available
-     */
-    error_setg(&s->migration_blocker, "vITS migration is not implemented");
-    migrate_add_blocker(s->migration_blocker, &local_err);
-    if (local_err) {
-        error_propagate(errp, local_err);
-        error_free(s->migration_blocker);
-        return;
-    }
-
     s->dev_fd = kvm_create_device(kvm_state, KVM_DEV_TYPE_ARM_VGIC_ITS, false);
     if (s->dev_fd < 0) {
         error_setg_errno(errp, -s->dev_fd, "error creating in-kernel ITS");
@@ -104,6 +92,18 @@ static void kvm_arm_its_realize(DeviceState *dev, Error **errp)
 
     gicv3_its_init_mmio(s, NULL);
 
+    if (!kvm_device_check_attr(s->dev_fd, KVM_DEV_ARM_VGIC_GRP_ITS_REGS,
+        GITS_CTLR)) {
+        error_setg(&s->migration_blocker, "This operating system kernel "
+                   "does not support vGICv3 migration");
+        migrate_add_blocker(s->migration_blocker, &local_err);
+        if (local_err) {
+            error_propagate(errp, local_err);
+            error_free(s->migration_blocker);
+            return;
+        }
+    }
+
     kvm_msi_use_devid = true;
     kvm_gsi_direct_mapping = false;
     kvm_msi_via_irqfd_allowed = kvm_irqfds_enabled();
diff --git a/include/migration/vmstate.h b/include/migration/vmstate.h
index f2dbf84..8dab9c7 100644
--- a/include/migration/vmstate.h
+++ b/include/migration/vmstate.h
@@ -198,6 +198,8 @@ enum VMStateFlags {
 typedef enum {
     MIG_PRI_DEFAULT = 0,
     MIG_PRI_IOMMU,              /* Must happen before PCI devices */
+    MIG_PRI_GICV3_ITS,          /* Must happen before PCI devices */
+    MIG_PRI_GICV3,              /* Must happen before the ITS */
     MIG_PRI_MAX,
 } MigrationPriority;
 
-- 
2.5.5