Re: [PATCH 1/4] KVM: arm64: Allow saving vgic3 LPI pending status in no running vcpu context

[Gavin Shan <gshan@redhat.com>]

Hi Oliver,

On 1/18/23 7:51 AM, Oliver Upton wrote:
> On Mon, Jan 16, 2023 at 12:04:02PM +0800, Gavin Shan wrote:
>> When dirty ring is enabled, the dirty page information is pushed to
>> the dirty ring if there is a running VCPU context. Otherwise, the
>> dirty page information is still tracked by the backup dirty bitmap.
>> In order to detect if there is a running VCPU context when a guest
>> page becomes dirty, kvm_arch_allow_write_without_running_vcpu() was
>> introduced to warn when no running VCPU context exists on unknown
>> cases.
>>
>> Other than the site of saving ITS tables, it's possible to save vgic3
>> LPI pending status in no running vcpu context because it can happen when
>> ITS ITE is restored through the command KVM_DEV_ARM_ITS_RESTORE_TABLES
>> on 'kvm-arm-vgic-its' device.
>>
>> Fix it by allowing to save vgic3 LPI pending status in no running
>> vcpu context.
>>
>> Signed-off-by: Gavin Shan <gshan@redhat.com>
>> ---
>>   Documentation/virt/kvm/api.rst | 5 +++--
>>   arch/arm64/kvm/vgic/vgic-its.c | 3 ++-
>>   arch/arm64/kvm/vgic/vgic-v3.c  | 3 +++
>>   include/kvm/arm_vgic.h         | 1 +
>>   4 files changed, 9 insertions(+), 3 deletions(-)
>>
>> diff --git a/Documentation/virt/kvm/api.rst b/Documentation/virt/kvm/api.rst
>> index 9807b05a1b57..18b245a0ba02 100644
>> --- a/Documentation/virt/kvm/api.rst
>> +++ b/Documentation/virt/kvm/api.rst
>> @@ -8071,8 +8071,9 @@ state is final and avoid missing dirty pages from another ioctl ordered
>>   after the bitmap collection.
>>   
>>   NOTE: One example of using the backup bitmap is saving arm64 vgic/its
>> -tables through KVM_DEV_ARM_{VGIC_GRP_CTRL, ITS_SAVE_TABLES} command on
>> -KVM device "kvm-arm-vgic-its" when dirty ring is enabled.
>> +tables and vgic3 LPI pending status through KVM_DEV_ARM_{VGIC_GRP_CTRL,
>> +ITS_SAVE_TABLES} and KVM_DEV_ARM_{VGIC_GRP_CTRL, ITS_RESTORE_TABLES}
>> +command on KVM device "kvm-arm-vgic-its" when dirty ring is enabled.
>>   
>>   8.30 KVM_CAP_XEN_HVM
>>   --------------------
>> diff --git a/arch/arm64/kvm/vgic/vgic-its.c b/arch/arm64/kvm/vgic/vgic-its.c
>> index 94a666dd1443..119a9c7a0a52 100644
>> --- a/arch/arm64/kvm/vgic/vgic-its.c
>> +++ b/arch/arm64/kvm/vgic/vgic-its.c
>> @@ -2792,7 +2792,8 @@ bool kvm_arch_allow_write_without_running_vcpu(struct kvm *kvm)
>>   {
>>   	struct vgic_dist *dist = &kvm->arch.vgic;
>>   
>> -	return dist->save_its_tables_in_progress;
>> +	return dist->save_vgic_v3_tables_in_progress ||
>> +	       dist->save_its_tables_in_progress;
> 
> I'd much prefer using a single bool to keep track of this, i.e:
> 

Yes, it's clean to have 'dist->save_tables_in_progress' for all
3 cases. One more concern like below.

> 	return dist->save_tables_in_progress;
> 
>>   }
>>   
>>   static int vgic_its_set_attr(struct kvm_device *dev,
>> diff --git a/arch/arm64/kvm/vgic/vgic-v3.c b/arch/arm64/kvm/vgic/vgic-v3.c
>> index 2074521d4a8c..32998c8587a8 100644
>> --- a/arch/arm64/kvm/vgic/vgic-v3.c
>> +++ b/arch/arm64/kvm/vgic/vgic-v3.c
>> @@ -304,6 +304,7 @@ void vgic_v3_enable(struct kvm_vcpu *vcpu)
>>   int vgic_v3_lpi_sync_pending_status(struct kvm *kvm, struct vgic_irq *irq)
>>   {
>>   	struct kvm_vcpu *vcpu;
>> +	struct vgic_dist *dist = &kvm->arch.vgic;
>>   	int byte_offset, bit_nr;
>>   	gpa_t pendbase, ptr;
>>   	bool status;
>> @@ -339,7 +340,9 @@ int vgic_v3_lpi_sync_pending_status(struct kvm *kvm, struct vgic_irq *irq)
>>   	if (status) {
>>   		/* clear consumed data */
>>   		val &= ~(1 << bit_nr);
>> +		dist->save_vgic_v3_tables_in_progress = true;
>>   		ret = kvm_write_guest_lock(kvm, ptr, &val, 1);
>> +		dist->save_vgic_v3_tables_in_progress = false;
> 
> With the above suggestion of using a bool, this should become a helper
> used at all the affected callsites:
> 
>    static int vgic_write_guest_lock(struct kvm *kvm, gpa_t gpa,
>    				   const void *data, unsigned long len)
>    {
>    	struct vgic_dist *dist = &kvm->arch.vgic;
> 	int ret;
> 
> 	dist->save_tables_in_progress = true;
> 	ret = kvm_write_guest_lock(kvm, gpa, data, len);
> 	dist->save_tables_in_progress = false;
> 
> 	return ret;
>    }
> 

I will have vgic_write_guest_lock() in v2. Note that those 3 paths can't be
running in parallel since one switch is shared by them. Alternatively, we
extend struct vgic_dist::save_tables_in_progress from 'bool' to 'unsigned long'.
Several bit is defined for each site as below. In this way, the 3 paths can be
running in parallel:

   unsigned long struct vgic_dist::save_tables_in_progress

   #define VGIC_DIST_SAVE_ITS_ITE		0	/* ITS Translation Entry */
   #define VGIC_DIST_SAVE_ITS_DTE		1	/* ITS Device Table Entry */
   #define VGIC_DIST_SAVE_ITS_CTE		2	/* ITS Collection Table Entry */
   #define VGIC_DIST_SAVE_ITS_CT			3	/* ITS Collection Table */
   #define VGIC_DIST_SAVE_VGIC3_LPI		4	/* VGIC3 LPI Pending Status */
   #define VGIC_DIST_SAVE_VGIC3_PENDING_TABLE	5	/* VGIC3 Pending Table */

The drawback is the calls are limited to 64. If those 3 paths can't be running
in parallel, we needn't the extension at all.

Thanks,
Gavin