[PATCH 1/3] drm/xe/pmu: Enable PMU interface

[Lucas De Marchi]

On Tue, Oct 01, 2024 at 03:08:03PM GMT, Belgaumkar, Vinay wrote:
>
>On 10/1/2024 2:32 PM, Lucas De Marchi wrote:
>>On Thu, Sep 26, 2024 at 05:23:42PM GMT, Vinay Belgaumkar wrote:
>>>diff --git a/include/uapi/drm/xe_drm.h b/include/uapi/drm/xe_drm.h
>>>index b6fbe4988f2e..2c5f258eee3a 100644
>>>--- a/include/uapi/drm/xe_drm.h
>>>+++ b/include/uapi/drm/xe_drm.h
>>>@@ -1389,6 +1389,14 @@ struct drm_xe_wait_user_fence {
>>>    __u64 reserved[2];
>>>};
>>>
>>>+/*
>>>+ * Top bits of every counter are GT id.
>>>+ */
>>>+#define __XE_PMU_GT_SHIFT (56)
>>>+
>>>+#define ___XE_PMU_OTHER(gt, x) \
>>>+    (((__u64)(x)) | ((__u64)(gt) << __XE_PMU_GT_SHIFT))
>>>+
>>
>>why are you touching this?
>>
>>I thought we agreed it's not needed at all on previous version of this
>>patch: https://lore.kernel.org/intel-xe/602bffcd-d66f-4b49-b3b4-abb934b00f3a@intel.com/
>>
>>what changed?
>
>IGT tests like perf_pmu use this interface to open specific PMU files, 
>for example-
>
>                fd[pmus] = perf_i915_open_group(gem_fd,
>__I915_PMU_RC6_RESIDENCY(gt_),
>                                                fd[0]);
>If we define these inside non-header files, these kinds of tests and 
>any other user programs that may be doing this, will need to look at 
>kernel code to determine the PMU config? I assumed we needed these 
>bare minimum header definitions.

no, they should look in sysfs and that should be sufficient. This is how
the `perf` tool works and you shouldn't need to change userspace tools
to support additional events


Lucas De Marchi

>
>Thanks,
>
>Vinay.
>
>>
>>
>>Lucas De Marchi