[PATCH v2] hw/i386/amd_iommu: Allow migration

[PATCH v2] hw/i386/amd_iommu: Allow migration

[Suravee Suthikulpanit]

Add migration support for AMD IOMMU model by saving necessary AMDVIState
parameters for MMIO registers, device table, command buffer, and event
buffers.

Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
---
Changes from v1:
(https://lore.kernel.org/all/9ecffa7a-f4c6-45a5-a066-84826ccb55dd@amd.com/T/)
* Include ppr_log, pprlog_len, pprlog_head, pprlog_tail per Joao.

 hw/i386/amd_iommu.c | 43 ++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 42 insertions(+), 1 deletion(-)

diff --git a/hw/i386/amd_iommu.c b/hw/i386/amd_iommu.c
index 13af7211e1..a1940a0ab3 100644
--- a/hw/i386/amd_iommu.c
+++ b/hw/i386/amd_iommu.c
@@ -1673,7 +1673,48 @@ static Property amdvi_properties[] = {
 
 static const VMStateDescription vmstate_amdvi_sysbus = {
     .name = "amd-iommu",
-    .unmigratable = 1
+    .version_id = 1,
+    .minimum_version_id = 1,
+    .priority = MIG_PRI_IOMMU,
+    .fields = (VMStateField[]) {
+      /* Updated in  amdvi_handle_control_write() */
+      VMSTATE_BOOL(enabled, AMDVIState),
+      VMSTATE_BOOL(ga_enabled, AMDVIState),
+      VMSTATE_BOOL(ats_enabled, AMDVIState),
+      VMSTATE_BOOL(cmdbuf_enabled, AMDVIState),
+      VMSTATE_BOOL(completion_wait_intr, AMDVIState),
+      VMSTATE_BOOL(evtlog_enabled, AMDVIState),
+      VMSTATE_BOOL(evtlog_intr, AMDVIState),
+      /* Updated in amdvi_handle_devtab_write() */
+      VMSTATE_UINT64(devtab, AMDVIState),
+      VMSTATE_UINT64(devtab_len, AMDVIState),
+      /* Updated in amdvi_handle_cmdbase_write() */
+      VMSTATE_UINT64(cmdbuf, AMDVIState),
+      VMSTATE_UINT64(cmdbuf_len, AMDVIState),
+      /* Updated in amdvi_handle_cmdhead_write() */
+      VMSTATE_UINT32(cmdbuf_head, AMDVIState),
+      /* Updated in amdvi_handle_cmdtail_write() */
+      VMSTATE_UINT32(cmdbuf_tail, AMDVIState),
+      /* Updated in amdvi_handle_evtbase_write() */
+      VMSTATE_UINT64(evtlog, AMDVIState),
+      VMSTATE_UINT32(evtlog_len, AMDVIState),
+      /* Updated in amdvi_handle_evthead_write() */
+      VMSTATE_UINT32(evtlog_head, AMDVIState),
+      /* Updated in amdvi_handle_evttail_write() */
+      VMSTATE_UINT32(evtlog_tail, AMDVIState),
+      /* Updated in amdvi_handle_pprbase_write() */
+      VMSTATE_UINT64(ppr_log, AMDVIState),
+      VMSTATE_UINT32(pprlog_len, AMDVIState),
+      /* Updated in amdvi_handle_pprhead_write() */
+      VMSTATE_UINT32(pprlog_head, AMDVIState),
+      /* Updated in amdvi_handle_tailhead_write() */
+      VMSTATE_UINT32(pprlog_tail, AMDVIState),
+      /* MMIO registers */
+      VMSTATE_UINT8_ARRAY(mmior, AMDVIState, AMDVI_MMIO_SIZE),
+      VMSTATE_UINT8_ARRAY(romask, AMDVIState, AMDVI_MMIO_SIZE),
+      VMSTATE_UINT8_ARRAY(w1cmask, AMDVIState, AMDVI_MMIO_SIZE),
+      VMSTATE_END_OF_LIST()
+    }
 };
 
 static void amdvi_sysbus_instance_init(Object *klass)
-- 
2.34.1

Re: [PATCH v2] hw/i386/amd_iommu: Allow migration

[Daniel P. Berrangé]

On Thu, Feb 06, 2025 at 05:18:56AM +0000, Suravee Suthikulpanit wrote:
> Add migration support for AMD IOMMU model by saving necessary AMDVIState
> parameters for MMIO registers, device table, command buffer, and event
> buffers.
> 
> Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
> ---
> Changes from v1:
> (https://lore.kernel.org/all/9ecffa7a-f4c6-45a5-a066-84826ccb55dd@amd.com/T/)
> * Include ppr_log, pprlog_len, pprlog_head, pprlog_tail per Joao.
> 
>  hw/i386/amd_iommu.c | 43 ++++++++++++++++++++++++++++++++++++++++++-
>  1 file changed, 42 insertions(+), 1 deletion(-)
> 
> diff --git a/hw/i386/amd_iommu.c b/hw/i386/amd_iommu.c
> index 13af7211e1..a1940a0ab3 100644
> --- a/hw/i386/amd_iommu.c
> +++ b/hw/i386/amd_iommu.c
> @@ -1673,7 +1673,48 @@ static Property amdvi_properties[] = {
>  
>  static const VMStateDescription vmstate_amdvi_sysbus = {
>      .name = "amd-iommu",
> -    .unmigratable = 1
> +    .version_id = 1,

IMHO we should not remove the 'unmigratable = 1'  setting, until we
have fixed the design problem with this device, to split the creation
of the 'AMDVI-PCI' device off from the 'amd-iommu' device, so that the
former is user creatable.

As it stands now, there's no mechanism to guarantee that the internally
created 'AMDVI-PCI' device gets the same PCI address properties on every
boot. Thus it isn't safe to claim this device is migratable yet IMHO.

> +    .minimum_version_id = 1,
> +    .priority = MIG_PRI_IOMMU,
> +    .fields = (VMStateField[]) {


With regards,
Daniel
-- 
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|

Re: [PATCH v2] hw/i386/amd_iommu: Allow migration

[Suthikulpanit, Suravee]

Daniel,

On 2/6/2025 4:01 PM, Daniel P. Berrangé wrote:
> On Thu, Feb 06, 2025 at 05:18:56AM +0000, Suravee Suthikulpanit wrote:
>> Add migration support for AMD IOMMU model by saving necessary AMDVIState
>> parameters for MMIO registers, device table, command buffer, and event
>> buffers.
>>
>> Signed-off-by: Suravee Suthikulpanit<suravee.suthikulpanit@amd.com>
>> ---
>> Changes from v1:
>> (https://lore.kernel.org/all/9ecffa7a-f4c6-45a5- 
>> a066-84826ccb55dd@amd.com/T/)
>> * Include ppr_log, pprlog_len, pprlog_head, pprlog_tail per Joao.
>>
>>   hw/i386/amd_iommu.c | 43 ++++++++++++++++++++++++++++++++++++++++++-
>>   1 file changed, 42 insertions(+), 1 deletion(-)
>>
>> diff --git a/hw/i386/amd_iommu.c b/hw/i386/amd_iommu.c
>> index 13af7211e1..a1940a0ab3 100644
>> --- a/hw/i386/amd_iommu.c
>> +++ b/hw/i386/amd_iommu.c
>> @@ -1673,7 +1673,48 @@ static Property amdvi_properties[] = {
>>   
>>   static const VMStateDescription vmstate_amdvi_sysbus = {
>>       .name = "amd-iommu",
>> -    .unmigratable = 1
>> +    .version_id = 1,
> IMHO we should not remove the 'unmigratable = 1'  setting, until we
> have fixed the design problem with this device, to split the creation
> of the 'AMDVI-PCI' device off from the 'amd-iommu' device, so that the
> former is user creatable.
> 
> As it stands now, there's no mechanism to guarantee that the internally
> created 'AMDVI-PCI' device gets the same PCI address properties on every
> boot. Thus it isn't safe to claim this device is migratable yet IMHO.

I'll send out v3 including the change you suggested, where we split the 
creation of AMDVI-PCI device to allow fix enumeration of amd-iommu device.

Thanks,
Suravee