Skip to contentSkip to conversation

arm_gic: Initialize VGIC on machine init done

PATCH In review Eric Auger 2 participants 7 messages Dec 2, 2014 – Dec 3, 2014

lore
Share & tools
lore

Eric Auger

author
#0 lore raw

cover arm_gic: Initialize VGIC on machine init done

With current implementation the vgic is initialized very late, when the VM is started, on the first VCPU run execution. This series kicks the initialization off on machine init done instead.

Among other things, this earlier VGIC initialization allows to have a simpler handling of VFIO. VFIO platform signaling can be setup on reset handler, making sure the VGIC is ready, at that time, to accept injected virtual IRQs.

Depends on the following kernel patch:
- [PATCH] KVM: arm/arm64: vgic: add init entry to VGIC KVM device

Tested on Calxeda Miday with VFIO platform use case

Eric Auger (2):
linux-headers: Update KVM headers from linux-next tag ToBeFilled hw/intc/arm_gic: Initialize the vgic on machine init done

 hw/intc/arm_gic_kvm.c            | 21 +++++++++++++++++++++
 include/hw/intc/arm_gic_common.h |  1 +
 linux-headers/asm-arm/kvm.h      |  2 ++
 linux-headers/asm-arm64/kvm.h    |  2 ++
 4 files changed, 26 insertions(+)
Signature · 1 line
1.8.3.2
author
#1 lore raw

PATCH 1/2 linux-headers: Update KVM headers from linux-next tag ToBeFilled

Syncup KVM related linux headers from linux-next tree using scripts/update-linux-headers.sh.

Add a new group/attribute in VGIC KVM device enabling to force vgic init: KVM_DEV_ARM_VGIC_GRP_CTRL/KVM_DEV_ARM_VGIC_CTRL_INIT

---
 linux-headers/asm-arm/kvm.h   | 2 ++
 linux-headers/asm-arm64/kvm.h | 2 ++
 2 files changed, 4 insertions(+)
linux-headers/asm-arm/kvm.h +2 -0
index 09ee408..0db25bc 100644
@@ -175,6 +175,8 @@ struct kvm_arch_memory_slot {
 #define   KVM_DEV_ARM_VGIC_OFFSET_SHIFT	0
 #define   KVM_DEV_ARM_VGIC_OFFSET_MASK	(0xffffffffULL << KVM_DEV_ARM_VGIC_OFFSET_SHIFT)
 #define KVM_DEV_ARM_VGIC_GRP_NR_IRQS	3
+#define KVM_DEV_ARM_VGIC_GRP_CTRL       4
+#define   KVM_DEV_ARM_VGIC_CTRL_INIT    0
 
 /* KVM_IRQ_LINE irq field index values */
 #define KVM_ARM_IRQ_TYPE_SHIFT		24
linux-headers/asm-arm64/kvm.h +2 -0
index 8e38878..cbce625 100644
@@ -161,6 +161,8 @@ struct kvm_arch_memory_slot {
 #define   KVM_DEV_ARM_VGIC_OFFSET_SHIFT	0
 #define   KVM_DEV_ARM_VGIC_OFFSET_MASK	(0xffffffffULL << KVM_DEV_ARM_VGIC_OFFSET_SHIFT)
 #define KVM_DEV_ARM_VGIC_GRP_NR_IRQS	3
+#define KVM_DEV_ARM_VGIC_GRP_CTRL       4
+#define   KVM_DEV_ARM_VGIC_CTRL_INIT    0
 
 /* KVM_IRQ_LINE irq field index values */
 #define KVM_ARM_IRQ_TYPE_SHIFT		24
Signature · 1 line
1.8.3.2
#2 lore raw

Re: [Qemu-devel] [PATCH 1/2] linux-headers: Update KVM headers from linux-next tag ToBeFilled

Quoted 7 lines · Eric Auger

Syncup KVM related linux headers from linux-next tree using scripts/update-linux-headers.sh.

Add a new group/attribute in VGIC KVM device enabling to force vgic init: KVM_DEV_ARM_VGIC_GRP_CTRL/KVM_DEV_ARM_VGIC_CTRL_INIT

How can this be from linux-next? I'm pretty sure we didn't merge this patch yet.

Also, if you synced your headers from your local tree, I think that generally makes such a QEMU patches series an RFC one.

-Christoffer

author
#3 lore raw

Re: [Qemu-devel] [PATCH 1/2] linux-headers: Update KVM headers from linux-next tag ToBeFilled

Quoted 11 lines · Christoffer Dall
Quoted 7 lines · Eric Auger

Syncup KVM related linux headers from linux-next tree using scripts/update-linux-headers.sh.

Add a new group/attribute in VGIC KVM device enabling to force vgic init: KVM_DEV_ARM_VGIC_GRP_CTRL/KVM_DEV_ARM_VGIC_CTRL_INIT

How can this be from linux-next? I'm pretty sure we didn't merge this patch yet.

Hi Christoffer,

yes that's correct . I don't know how to format that kind of patch that refers to linux headers (I mentioned in the title tag ToBeFilled). I looked at other similar ones and I thought this was the wording that is generally used. I can move it to RFC but it will stay as is until kernel patches are upstreamed. In that case I should also move the QEMU VFIO patch back to RFC due to forwarding stuff;-)

Best Regards

Eric

Quoted 6 lines

Also, if you synced your headers from your local tree, I think that generally makes such a QEMU patches series an RFC one.

-Christoffer

#4 lore raw

Re: [Qemu-devel] [PATCH 1/2] linux-headers: Update KVM headers from linux-next tag ToBeFilled

Quoted 13 lines · Eric Auger
Quoted 11 lines · Christoffer Dall
Quoted 7 lines · Eric Auger

Syncup KVM related linux headers from linux-next tree using scripts/update-linux-headers.sh.

Add a new group/attribute in VGIC KVM device enabling to force vgic init: KVM_DEV_ARM_VGIC_GRP_CTRL/KVM_DEV_ARM_VGIC_CTRL_INIT

How can this be from linux-next? I'm pretty sure we didn't merge this patch yet.

Quoted 9 lines

Hi Christoffer,

yes that's correct . I don't know how to format that kind of patch that refers to linux headers (I mentioned in the title tag ToBeFilled). I looked at other similar ones and I thought this was the wording that is generally used. I can move it to RFC but it will stay as is until kernel patches are upstreamed. In that case I should also move the QEMU VFIO patch back to RFC due to forwarding stuff;-)

This patch subject should just be something like this:

Subject: [RFC PATCH 1/2] linux-headers: Update KVM headers for KVM_DEV_ARM_VGIC_GRP_CTRL

And in the commit message you point to a personal git tree that contains the patches you're syncing against.

-Christoffer

author
#5 lore raw

Re: [Qemu-devel] [PATCH 1/2] linux-headers: Update KVM headers from linux-next tag ToBeFilled

Quoted 23 lines · Christoffer Dall
Quoted 13 lines · Eric Auger
Quoted 11 lines · Christoffer Dall
Quoted 7 lines · Eric Auger

Syncup KVM related linux headers from linux-next tree using scripts/update-linux-headers.sh.

Add a new group/attribute in VGIC KVM device enabling to force vgic init: KVM_DEV_ARM_VGIC_GRP_CTRL/KVM_DEV_ARM_VGIC_CTRL_INIT

How can this be from linux-next? I'm pretty sure we didn't merge this patch yet.

Quoted 9 lines

Hi Christoffer,

yes that's correct . I don't know how to format that kind of patch that refers to linux headers (I mentioned in the title tag ToBeFilled). I looked at other similar ones and I thought this was the wording that is generally used. I can move it to RFC but it will stay as is until kernel patches are upstreamed. In that case I should also move the QEMU VFIO patch back to RFC due to forwarding stuff;-)

Quoted 6 lines

This patch subject should just be something like this:

Subject: [RFC PATCH 1/2] linux-headers: Update KVM headers for KVM_DEV_ARM_VGIC_GRP_CTRL

And in the commit message you point to a personal git tree that contains the patches you're syncing against.

OK thanks

I will update accordingly

Eric

Quoted 3 lines

-Christoffer

author
#6 lore raw

PATCH 2/2 hw/intc/arm_gic: Initialize the vgic on machine init done

In reply to Eric Auger

With current implementation the vgic is instantiated very late, when the VM is started, on the first VCPU run execution. This patch forces its initialization in a machine init done notifier. This latter is registered in the vgic realize function. The notifier uses a new group/attribute that allows such operation. In practice, vgic initialization can be done as soon as those dimensioning parameters are known:
- the number of IRQs,
- the number of vcpus
- start address of dist and CPU interface (set in a memory listener).

This earlier initialization allows, for example, to setup VFIO signaling after vgic initialization, on a reset notifier.

---
 hw/intc/arm_gic_kvm.c            | 21 +++++++++++++++++++++
 include/hw/intc/arm_gic_common.h |  1 +
 2 files changed, 22 insertions(+)
hw/intc/arm_gic_kvm.c +21 -0
index 5038885..24fdf98 100644
@@ -23,6 +23,7 @@
 #include "sysemu/kvm.h"
 #include "kvm_arm.h"
 #include "gic_internal.h"
+#include "sysemu/sysemu.h"
 
 //#define DEBUG_GIC_KVM
 
@@ -511,6 +512,17 @@ static void kvm_arm_gic_reset(DeviceState *dev)
     kvm_arm_gic_put(s);
 }
 
+static void kvm_arm_vgic_init_notify(Notifier *notifier, void *data)
+{
+    GICState *s = container_of(notifier, GICState, init_notifier);
+    struct kvm_device_attr attr = {
+         .group = KVM_DEV_ARM_VGIC_GRP_CTRL,
+         .attr = KVM_DEV_ARM_VGIC_CTRL_INIT};
+    if (s->dev_fd >= 0) {
+        kvm_device_ioctl(s->dev_fd, KVM_SET_DEVICE_ATTR, &attr);
+    }
+}
+
 static void kvm_arm_gic_realize(DeviceState *dev, Error **errp)
 {
     int i;
@@ -553,6 +565,15 @@ static void kvm_arm_gic_realize(DeviceState *dev, Error **errp)
         return;
     }
 
+    /*
+     * Register a notifier that will request the VGIC initialization.
+     * It will be executed after the notifier registered by
+     * kvm_arm_register_device whose role is to set dist and cpu interface
+     * base address. Also IRQ and VCPU number is known.
+     */
+    s->init_notifier.notify = kvm_arm_vgic_init_notify;
+    qemu_add_machine_init_done_notifier(&s->init_notifier);
+
     /* Distributor */
     memory_region_init_reservation(&s->iomem, OBJECT(s),
                                    "kvm-gic_dist", 0x1000);
include/hw/intc/arm_gic_common.h +1 -0
index f6887ed..98653c9 100644
@@ -105,6 +105,7 @@ typedef struct GICState {
     uint32_t num_irq;
     uint32_t revision;
     int dev_fd; /* kvm device fd if backed by kvm vgic support */
+    Notifier init_notifier; /* notifier used to init the vgic */
 } GICState;
 
 #define TYPE_ARM_GIC_COMMON "arm_gic_common"
Signature · 1 line
1.8.3.2