[PATCH] KVM: selftest: Add family and model check for zen4 in PMU filter test

[PATCH] KVM: selftest: Add family and model check for zen4 in PMU filter test

[Manali Shukla]

PMU event filter test fails on zen4 architecture because of
unavailability of family and model check for zen4 in use_amd_pmu().
So, add family and model check for zen4 architecture in use_amd_pmu().

Signed-off-by: Manali Shukla <manali.shukla@amd.com>
---
 .../testing/selftests/kvm/x86_64/pmu_event_filter_test.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/tools/testing/selftests/kvm/x86_64/pmu_event_filter_test.c b/tools/testing/selftests/kvm/x86_64/pmu_event_filter_test.c
index 3c85d1ae9893..c212ca4ffa72 100644
--- a/tools/testing/selftests/kvm/x86_64/pmu_event_filter_test.c
+++ b/tools/testing/selftests/kvm/x86_64/pmu_event_filter_test.c
@@ -374,6 +374,12 @@ static bool is_zen3(uint32_t family, uint32_t model)
 	return family == 0x19 && model <= 0x0f;
 }
 
+static bool is_zen4(uint32_t family, uint32_t model)
+{
+	return family == 0x19 && ((model >= 0x10 && model <= 0x1f) ||
+		(model >= 0xa0 && model <= 0xaf));
+}
+
 /*
  * Determining AMD support for a PMU event requires consulting the AMD
  * PPR for the CPU or reference material derived therefrom. The AMD
@@ -390,7 +396,8 @@ static bool use_amd_pmu(void)
 	return host_cpu_is_amd &&
 		(is_zen1(family, model) ||
 		 is_zen2(family, model) ||
-		 is_zen3(family, model));
+		 is_zen3(family, model) ||
+		 is_zen4(family, model));
 }
 
 /*
-- 
2.34.1

Re: [PATCH] KVM: selftest: Add family and model check for zen4 in PMU filter test

[Sean Christopherson]

On Wed, May 01, 2024, Manali Shukla wrote:
> PMU event filter test fails on zen4 architecture because of
> unavailability of family and model check for zen4 in use_amd_pmu().
> So, add family and model check for zen4 architecture in use_amd_pmu().

Is there a less ugly way to detect that 0xc2,0 == "branch instructions retired"?
E.g. can we instead check for v2 PMU support, or are there no guarantees going
forward?  Pivoting on FMS is so painful :-(

Re: [PATCH] KVM: selftest: Add family and model check for zen4 in PMU filter test

[Manali Shukla]

On 5/1/2024 9:02 PM, Sean Christopherson wrote:
> On Wed, May 01, 2024, Manali Shukla wrote:
>> PMU event filter test fails on zen4 architecture because of
>> unavailability of family and model check for zen4 in use_amd_pmu().
>> So, add family and model check for zen4 architecture in use_amd_pmu().
> 
> Is there a less ugly way to detect that 0xc2,0 == "branch instructions retired"?
> E.g. can we instead check for v2 PMU support, or are there no guarantees going
> forward?  Pivoting on FMS is so painful :-(

We have confirmed with the hardware team that 0xc2,0 == "branch instructions retired"
is always true going forward, we intend to maintain backward compatibility for branch
instruction retired. Since event 0xc2 is supported on all currently released F17h+ 
processors as branch instructions retired, we can check for "family >= 0x17" for all
Zen and its successors instead of checking them individually in pmu_event_filter_test.c.

- Manali

Re: [PATCH] KVM: selftest: Add family and model check for zen4 in PMU filter test

[Sean Christopherson]

On Tue, May 28, 2024, Manali Shukla wrote:
> On 5/1/2024 9:02 PM, Sean Christopherson wrote:
> > On Wed, May 01, 2024, Manali Shukla wrote:
> >> PMU event filter test fails on zen4 architecture because of
> >> unavailability of family and model check for zen4 in use_amd_pmu().
> >> So, add family and model check for zen4 architecture in use_amd_pmu().
> > 
> > Is there a less ugly way to detect that 0xc2,0 == "branch instructions retired"?
> > E.g. can we instead check for v2 PMU support, or are there no guarantees going
> > forward?  Pivoting on FMS is so painful :-(
> 
> We have confirmed with the hardware team that 0xc2,0 == "branch instructions retired"
> is always true going forward, we intend to maintain backward compatibility for branch
> instruction retired. Since event 0xc2 is supported on all currently released F17h+ 
> processors as branch instructions retired, we can check for "family >= 0x17" for all
> Zen and its successors instead of checking them individually in pmu_event_filter_test.c.

Can you send a patch for this?  Please :-)

Re: [PATCH] KVM: selftest: Add family and model check for zen4 in PMU filter test

[Manali Shukla]

On 6/4/2024 3:47 AM, Sean Christopherson wrote:
> On Tue, May 28, 2024, Manali Shukla wrote:
>> On 5/1/2024 9:02 PM, Sean Christopherson wrote:
>>> On Wed, May 01, 2024, Manali Shukla wrote:
>>>> PMU event filter test fails on zen4 architecture because of
>>>> unavailability of family and model check for zen4 in use_amd_pmu().
>>>> So, add family and model check for zen4 architecture in use_amd_pmu().
>>>
>>> Is there a less ugly way to detect that 0xc2,0 == "branch instructions retired"?
>>> E.g. can we instead check for v2 PMU support, or are there no guarantees going
>>> forward?  Pivoting on FMS is so painful :-(
>>
>> We have confirmed with the hardware team that 0xc2,0 == "branch instructions retired"
>> is always true going forward, we intend to maintain backward compatibility for branch
>> instruction retired. Since event 0xc2 is supported on all currently released F17h+ 
>> processors as branch instructions retired, we can check for "family >= 0x17" for all
>> Zen and its successors instead of checking them individually in pmu_event_filter_test.c.
> 
> Can you send a patch for this?  Please :-)

Sent. 

https://lore.kernel.org/kvm/20240605050835.30491-1-manali.shukla@amd.com/T/#u

-Manali