[PATCH] drm/amd/powerplay: custom peak clock freq for navi10

Wang, Kevin(Yang) Kevin1.Wang at amd.com
Fri Jul 19 05:36:20 UTC 2019


Comment inline

________________________________
From: Quan, Evan <Evan.Quan at amd.com>
Sent: Friday, July 19, 2019 1:03 PM
To: Wang, Kevin(Yang) <Kevin1.Wang at amd.com>; amd-gfx at lists.freedesktop.org <amd-gfx at lists.freedesktop.org>
Cc: Feng, Kenneth <Kenneth.Feng at amd.com>; Huang, Ray <Ray.Huang at amd.com>; Xu, Feifei <Feifei.Xu at amd.com>; Gui, Jack <Jack.Gui at amd.com>
Subject: RE: [PATCH] drm/amd/powerplay: custom peak clock freq for navi10

Comment inline

> -----Original Message-----
> From: Wang, Kevin(Yang) <Kevin1.Wang at amd.com>
> Sent: Friday, July 19, 2019 11:46 AM
> To: amd-gfx at lists.freedesktop.org
> Cc: Feng, Kenneth <Kenneth.Feng at amd.com>; Quan, Evan
> <Evan.Quan at amd.com>; Huang, Ray <Ray.Huang at amd.com>; Xu, Feifei
> <Feifei.Xu at amd.com>; Gui, Jack <Jack.Gui at amd.com>; Wang, Kevin(Yang)
> <Kevin1.Wang at amd.com>
> Subject: [PATCH] drm/amd/powerplay: custom peak clock freq for navi10
>
> 1.NAVI10_PEAK_SCLK_XTX    1830 Mhz
> 2.NAVI10_PEAK_SCLK_XT     1755 Mhz
> 3.NAVI10_PEAK_SCLK_XL     1625 Mhz
>
> Change-Id: I48863a9d0e261b9e7778a6c0e4a8762d7c978da6
> Signed-off-by: Kevin Wang <kevin1.wang at amd.com>
> ---
>  drivers/gpu/drm/amd/powerplay/amdgpu_smu.c    | 65 ++++++++++--------
> -
>  .../gpu/drm/amd/powerplay/inc/amdgpu_smu.h    |  4 ++
>  drivers/gpu/drm/amd/powerplay/navi10_ppt.c    | 55 ++++++++++++++++
>  drivers/gpu/drm/amd/powerplay/navi10_ppt.h    |  4 ++
>  4 files changed, 97 insertions(+), 31 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/powerplay/amdgpu_smu.c
> b/drivers/gpu/drm/amd/powerplay/amdgpu_smu.c
> index 7f51bbd2ac90..ab389dde9562 100644
> --- a/drivers/gpu/drm/amd/powerplay/amdgpu_smu.c
> +++ b/drivers/gpu/drm/amd/powerplay/amdgpu_smu.c
> @@ -1360,37 +1360,40 @@ int smu_adjust_power_state_dynamic(struct
> smu_context *smu,
>        }
>
>        if (smu_dpm_ctx->dpm_level != level) {
> -             switch (level) {
> -             case AMD_DPM_FORCED_LEVEL_HIGH:
> -                     ret = smu_force_dpm_limit_value(smu, true);
> -                     break;
> -             case AMD_DPM_FORCED_LEVEL_LOW:
> -                     ret = smu_force_dpm_limit_value(smu, false);
> -                     break;
> -
> -             case AMD_DPM_FORCED_LEVEL_AUTO:
> -             case AMD_DPM_FORCED_LEVEL_PROFILE_STANDARD:
> -                     ret = smu_unforce_dpm_levels(smu);
> -                     break;
> -
> -             case AMD_DPM_FORCED_LEVEL_PROFILE_MIN_SCLK:
> -             case AMD_DPM_FORCED_LEVEL_PROFILE_MIN_MCLK:
> -             case AMD_DPM_FORCED_LEVEL_PROFILE_PEAK:
> -                     ret = smu_get_profiling_clk_mask(smu, level,
> -                                                      &sclk_mask,
> -                                                      &mclk_mask,
> -                                                      &soc_mask);
> -                     if (ret)
> -                             return ret;
> -                     smu_force_clk_levels(smu, SMU_SCLK, 1 <<
> sclk_mask);
> -                     smu_force_clk_levels(smu, SMU_MCLK, 1 <<
> mclk_mask);
> -                     smu_force_clk_levels(smu, SMU_SOCCLK, 1 <<
> soc_mask);
> -                     break;
> -
> -             case AMD_DPM_FORCED_LEVEL_MANUAL:
> -             case AMD_DPM_FORCED_LEVEL_PROFILE_EXIT:
> -             default:
> -                     break;
> +             ret = smu_set_performance_level(smu, level);
> +             if (ret) {
[Quan, Evan] Since this actually handles AMD_DPM_FORCED_LEVEL_PROFILE_PEAK only. Please move this under "case AMD_DPM_FORCED_LEVEL_PROFILE_PEAK:".
And please give it a more meaningful name e.g. smu_set_asic_peak_profile().

[kevin]: in the future, for specifc asic, it may be custom other performance level, so we should have this capacity to support other asic with any performance level. and i think it is not a good way to handle performance level use case by case method for each asic.

> +                     switch (level) {
> +                     case AMD_DPM_FORCED_LEVEL_HIGH:
> +                             ret = smu_force_dpm_limit_value(smu, true);
> +                             break;
> +                     case AMD_DPM_FORCED_LEVEL_LOW:
> +                             ret = smu_force_dpm_limit_value(smu,
> false);
> +                             break;
> +
> +                     case AMD_DPM_FORCED_LEVEL_AUTO:
> +                     case
> AMD_DPM_FORCED_LEVEL_PROFILE_STANDARD:
> +                             ret = smu_unforce_dpm_levels(smu);
> +                             break;
> +
> +                     case
> AMD_DPM_FORCED_LEVEL_PROFILE_MIN_SCLK:
> +                     case
> AMD_DPM_FORCED_LEVEL_PROFILE_MIN_MCLK:
> +                     case AMD_DPM_FORCED_LEVEL_PROFILE_PEAK:
> +                             ret = smu_get_profiling_clk_mask(smu, level,
> +                                                              &sclk_mask,
> +                                                              &mclk_mask,
> +                                                              &soc_mask);
> +                             if (ret)
> +                                     return ret;
> +                             smu_force_clk_levels(smu, SMU_SCLK, 1 <<
> sclk_mask);
> +                             smu_force_clk_levels(smu, SMU_MCLK, 1 <<
> mclk_mask);
> +                             smu_force_clk_levels(smu, SMU_SOCCLK, 1
> << soc_mask);
> +                             break;
> +
> +                     case AMD_DPM_FORCED_LEVEL_MANUAL:
> +                     case AMD_DPM_FORCED_LEVEL_PROFILE_EXIT:
> +                     default:
> +                             break;
> +                     }
>                }
>
>                if (!ret)
> diff --git a/drivers/gpu/drm/amd/powerplay/inc/amdgpu_smu.h
> b/drivers/gpu/drm/amd/powerplay/inc/amdgpu_smu.h
> index 514d31518853..ba5ddafcbdba 100644
> --- a/drivers/gpu/drm/amd/powerplay/inc/amdgpu_smu.h
> +++ b/drivers/gpu/drm/amd/powerplay/inc/amdgpu_smu.h
> @@ -631,6 +631,7 @@ struct pptable_funcs {
>        int (*get_thermal_temperature_range)(struct smu_context *smu,
> struct smu_temperature_range *range);
>        int (*get_uclk_dpm_states)(struct smu_context *smu, uint32_t
> *clocks_in_khz, uint32_t *num_states);
>        int (*set_default_od_settings)(struct smu_context *smu, bool
> initialize);
> +     int (*set_performance_level)(struct smu_context *smu, int32_t
> level);
>  };
>
>  struct smu_funcs
> @@ -928,6 +929,9 @@ struct smu_funcs
>        ((smu)->funcs->baco_get_state? (smu)->funcs-
> >baco_get_state((smu), (state)) : 0)
>  #define smu_baco_reset(smu) \
>        ((smu)->funcs->baco_reset? (smu)->funcs->baco_reset((smu)) : 0)
> +#define smu_set_performance_level(smu, level) \
> +     ((smu)->ppt_funcs->set_performance_level? (smu)->ppt_funcs-
> >set_performance_level((smu), (level)) : -EINVAL);
> +
>
>  extern int smu_get_atom_data_table(struct smu_context *smu, uint32_t
> table,
>                                   uint16_t *size, uint8_t *frev, uint8_t *crev,
> diff --git a/drivers/gpu/drm/amd/powerplay/navi10_ppt.c
> b/drivers/gpu/drm/amd/powerplay/navi10_ppt.c
> index e44041a25e64..9024d91477f0 100644
> --- a/drivers/gpu/drm/amd/powerplay/navi10_ppt.c
> +++ b/drivers/gpu/drm/amd/powerplay/navi10_ppt.c
> @@ -1590,6 +1590,60 @@ static int navi10_set_ppfeature_status(struct
> smu_context *smu,
>        return 0;
>  }
>
> +static int navi10_set_peak_clock_by_device(struct smu_context *smu)
> +{
> +     struct amdgpu_device *adev = smu->adev;
> +     int ret = 0;
> +     uint32_t sclk_freq = 0, uclk_freq = 0;
> +     uint32_t uclk_level = 0;
> +
> +     switch (adev->rev_id) {
> +     case 0xf0: /* XTX */
> +     case 0xC0:
> +             sclk_freq = NAVI10_PEAK_SCLK_XTX;
> +             break;
> +     case 0xf1: /* XT */
> +     case 0xC1:
> +             sclk_freq = NAVI10_PEAK_SCLK_XT;
> +             break;
> +     default: /* XL */
> +             sclk_freq = NAVI10_PEAK_SCLK_XL;
> +             break;
> +     }
> +
> +     ret = smu_get_dpm_level_count(smu, SMU_UCLK, &uclk_level);
> +     if (ret)
> +             return ret;
> +     ret = smu_get_dpm_freq_by_index(smu, SMU_UCLK, uclk_level - 1,
> &uclk_freq);
> +     if (ret)
> +             return ret;
[Quan, Evan] I think with dpm level (3rd argument) set as 0xff directly,  you can  get the clock frequency for the max level.
[kevin]: not right, 0xff will return dpm clock level count, see function smu_get_dpm_level_count
> +
> +     ret = smu_set_soft_freq_range(smu, SMU_SCLK, sclk_freq,
> sclk_freq);
> +     if (ret)
> +             return ret;
> +     ret = smu_set_soft_freq_range(smu, SMU_UCLK, uclk_freq,
> uclk_freq);
> +     if (ret)
> +             return ret;
> +
> +     return ret;
> +}
> +
> +static int navi10_set_performance_level(struct smu_context *smu, int32_t
> level)
> +{
> +     int ret = 0;
> +
> +     switch (level) {
> +     case AMD_DPM_FORCED_LEVEL_PROFILE_PEAK:
> +             ret = navi10_set_peak_clock_by_device(smu);
> +             break;
> +     default:
> +             ret = -EINVAL;
> +             break;
> +     }
> +
> +     return ret;
> +}
> +
>  static const struct pptable_funcs navi10_ppt_funcs = {
>        .tables_init = navi10_tables_init,
>        .alloc_dpm_context = navi10_allocate_dpm_context,
> @@ -1625,6 +1679,7 @@ static const struct pptable_funcs navi10_ppt_funcs
> = {
>        .get_uclk_dpm_states = navi10_get_uclk_dpm_states,
>        .get_ppfeature_status = navi10_get_ppfeature_status,
>        .set_ppfeature_status = navi10_set_ppfeature_status,
> +     .set_performance_level = navi10_set_performance_level,
>  };
>
>  void navi10_set_ppt_funcs(struct smu_context *smu)
> diff --git a/drivers/gpu/drm/amd/powerplay/navi10_ppt.h
> b/drivers/gpu/drm/amd/powerplay/navi10_ppt.h
> index 957288e22f47..620ff17c2fef 100644
> --- a/drivers/gpu/drm/amd/powerplay/navi10_ppt.h
> +++ b/drivers/gpu/drm/amd/powerplay/navi10_ppt.h
> @@ -23,6 +23,10 @@
>  #ifndef __NAVI10_PPT_H__
>  #define __NAVI10_PPT_H__
>
> +#define NAVI10_PEAK_SCLK_XTX         (1830)
> +#define NAVI10_PEAK_SCLK_XT           (1755)
> +#define NAVI10_PEAK_SCLK_XL           (1625)
> +
>  extern void navi10_set_ppt_funcs(struct smu_context *smu);
>
>  #endif
> --
> 2.22.0

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/amd-gfx/attachments/20190719/d469de23/attachment-0001.html>


More information about the amd-gfx mailing list