[PATCH 3/3] drm/amdgpu: add xgmi perfmons for arcturus

[Kim, Jonathan]

Hi Harish.  Thanks for the review.  As discussed offline, patch 2 has a problem where attr groups array is global but allocation is done per-device causing problems with mem free and pmu unregister.  I'm sending out a second series that should fix this and simplify the solution as well as hopefully address your concerns.  Sorry again for the churn.

Jon

> -----Original Message-----
> From: Kasiviswanathan, Harish <Harish.Kasiviswanathan at amd.com>
> Sent: Tuesday, September 15, 2020 10:10 PM
> To: Kim, Jonathan <Jonathan.Kim at amd.com>; amd-
> gfx at lists.freedesktop.org
> Subject: RE: [PATCH 3/3] drm/amdgpu: add xgmi perfmons for arcturus
> 
> [AMD Official Use Only - Internal Distribution Only]
> 
> Reviewed-by: Harish Kasiviswanathan <Harish.Kasiviswanathan at amd.com>
> 
> -----Original Message-----
> From: Kim, Jonathan <Jonathan.Kim at amd.com>
> Sent: Tuesday, September 15, 2020 6:00 PM
> To: amd-gfx at lists.freedesktop.org
> Cc: Kasiviswanathan, Harish <Harish.Kasiviswanathan at amd.com>; Kim,
> Jonathan <Jonathan.Kim at amd.com>; Kim, Jonathan
> <Jonathan.Kim at amd.com>
> Subject: [PATCH 3/3] drm/amdgpu: add xgmi perfmons for arcturus
> 
> Add xgmi perfmons for Arcturus.
> 
> Signed-off-by: Jonathan Kim <Jonathan.Kim at amd.com>
> ---
>  drivers/gpu/drm/amd/amdgpu/amdgpu_pmu.c | 55
> +++++++++++++++++++++++++
>  drivers/gpu/drm/amd/amdgpu/df_v3_6.c    |  3 ++
>  2 files changed, 58 insertions(+)
> 
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_pmu.c
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_pmu.c
> index f3d2ac0e88a7..ec521c72e631 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_pmu.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_pmu.c
> @@ -34,6 +34,8 @@
>  #define NUM_EVENTS_DF_LEGACY		8
>  #define NUM_EVENTS_VEGA20_XGMI		2
>  #define NUM_EVENTS_VEGA20_MAX		2
> +#define NUM_EVENTS_ARCTURUS_XGMI	6
> +#define NUM_EVENTS_ARCTURUS_MAX		6
> 
>  /* record to keep track of pmu entry per pmu type per device */  struct
> amdgpu_pmu_entry { @@ -110,6 +112,27 @@ const struct attribute_group
> *vega20_attr_groups[] = {
>  	NULL
>  };
> 
> +/* Arcturus events */
> +static const char *arcturus_events[NUM_EVENTS_ARCTURUS_MAX][2] = {
> +	{ "xgmi_link0_data_outbound",
> "event=0x7,instance=0x4b,umask=0x2" },
> +	{ "xgmi_link1_data_outbound",
> "event=0x7,instance=0x4c,umask=0x2" },
> +	{ "xgmi_link2_data_outbound",
> "event=0x7,instance=0x4d,umask=0x2" },
> +	{ "xgmi_link3_data_outbound",
> "event=0x7,instance=0x4e,umask=0x2" },
> +	{ "xgmi_link4_data_outbound",
> "event=0x7,instance=0x4f,umask=0x2" },
> +	{ "xgmi_link5_data_outbound",
> "event=0x7,instance=0x50,umask=0x2" } };
> +
> +static struct attribute_group arcturus_event_attr_group = {
> +	.name = "events",
> +	.attrs = NULL
> +};
> +
> +const struct attribute_group *arcturus_attr_groups[] = {
> +	&amdgpu_pmu_format_attr_group,
> +	&arcturus_event_attr_group,
> +	NULL
> +};
> +
>  /* All df_vega20_* items are DEPRECATED. Use vega20_ items above
> instead. */  static const char
> *df_vega20_formats[NUM_FORMATS_DF_LEGACY][2] = {
>  	{ "event", "config:0-7" },
> @@ -400,6 +423,16 @@ static int init_pmu_by_type(struct amdgpu_device
> *adev,
> 
>  		pmu_entry->pmu.attr_groups = vega20_attr_groups;
>  		break;
> +	case CHIP_ARCTURUS:
> +		amdgpu_pmu_create_attributes(evt_attr_group, evt_attr,
> +				arcturus_events, 0,
> NUM_EVENTS_ARCTURUS_XGMI,
> +				PERF_TYPE_AMDGPU_XGMI);
> +		num_events += NUM_EVENTS_ARCTURUS_XGMI;
> +
> +		/* other events can be added here */
> +
> +		pmu_entry->pmu.attr_groups = arcturus_attr_groups;
> +		break;
>  	default:
>  		return -ENODEV;
>  	};
> @@ -530,6 +563,28 @@ int amdgpu_pmu_init(struct amdgpu_device *adev)
>  			goto err_pmu;
>  		}
> 
> +		break;
> +	case CHIP_ARCTURUS:
> +		ret =
> amdgpu_pmu_alloc_pmu_attrs(&amdgpu_pmu_format_attr_group,
> +						&fmt_attr,
> +
> 	NUM_FORMATS_AMDGPU_PMU,
> +						&arcturus_event_attr_group,
> +						&evt_attr,
> +
> 	NUM_EVENTS_ARCTURUS_MAX);
> +
> +		if (ret)
> +			goto err_alloc;
> +
> +		ret = init_pmu_by_type(adev,
> +				&amdgpu_pmu_format_attr_group,
> fmt_attr,
> +				&arcturus_event_attr_group, evt_attr,
> +				"Event", "amdgpu",
> PERF_TYPE_AMDGPU_MAX);
> +
> +		if (ret) {
> +			kfree(arcturus_event_attr_group.attrs);
> +			goto err_pmu;
> +		}
> +
>  		break;
>  	default:
>  		return 0;
> diff --git a/drivers/gpu/drm/amd/amdgpu/df_v3_6.c
> b/drivers/gpu/drm/amd/amdgpu/df_v3_6.c
> index 6e57ae95f997..6b4b30a8dce5 100644
> --- a/drivers/gpu/drm/amd/amdgpu/df_v3_6.c
> +++ b/drivers/gpu/drm/amd/amdgpu/df_v3_6.c
> @@ -513,6 +513,7 @@ static int df_v3_6_pmc_start(struct amdgpu_device
> *adev, uint64_t config,
> 
>  	switch (adev->asic_type) {
>  	case CHIP_VEGA20:
> +	case CHIP_ARCTURUS:
>  		if (is_add)
>  			return df_v3_6_pmc_add_cntr(adev, config);
> 
> @@ -554,6 +555,7 @@ static int df_v3_6_pmc_stop(struct amdgpu_device
> *adev, uint64_t config,
> 
>  	switch (adev->asic_type) {
>  	case CHIP_VEGA20:
> +	case CHIP_ARCTURUS:
>  		ret = df_v3_6_pmc_get_ctrl_settings(adev,
>  			config,
>  			counter_idx,
> @@ -590,6 +592,7 @@ static void df_v3_6_pmc_get_count(struct
> amdgpu_device *adev,
> 
>  	switch (adev->asic_type) {
>  	case CHIP_VEGA20:
> +	case CHIP_ARCTURUS:
>  		df_v3_6_pmc_get_read_settings(adev, config, counter_idx,
>  						&lo_base_addr,
> &hi_base_addr);
> 
> --
> 2.17.1