[PATCH 2/2] RFC drm/xe: Enable Coarse Power Gating

Rodrigo Vivi rodrigo.vivi at intel.com
Wed May 15 03:11:16 UTC 2024


On Tue, May 14, 2024 at 12:13:09PM +0530, Riana Tauro wrote:
> Coarse Power Gating (CPG), where Render and Media can enter C6
> independent of the remaining GT. Enable render and media
> power gating.

Something strange with this phrase...

> 
> Also enable VD HCP/MFX sub-pipe power gating.
> HCP/MFX power gating is disabled by default, turn it on for
> the vd units available

Could be simplified to something like:
"Enable every power gating that are disabled by default,
for every unit and sub-pipe available"
(or something like that).

> 
> Signed-off-by: Riana Tauro <riana.tauro at intel.com>
> ---
>  drivers/gpu/drm/xe/regs/xe_gt_regs.h |  2 ++
>  drivers/gpu/drm/xe/xe_gt.c           | 10 +++++++
>  drivers/gpu/drm/xe/xe_gt_idle.c      | 45 ++++++++++++++++++++++++++--
>  drivers/gpu/drm/xe/xe_gt_idle.h      |  2 ++
>  4 files changed, 56 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/gpu/drm/xe/regs/xe_gt_regs.h b/drivers/gpu/drm/xe/regs/xe_gt_regs.h
> index 7c173db7d585..1cb0343ab581 100644
> --- a/drivers/gpu/drm/xe/regs/xe_gt_regs.h
> +++ b/drivers/gpu/drm/xe/regs/xe_gt_regs.h
> @@ -317,6 +317,8 @@
>  #define FORCEWAKE_GT				XE_REG(0xa188)
>  
>  #define POWERGATE_ENABLE			XE_REG(0xa210)
> +#define   RENDER_POWERGATE_ENABLE		REG_BIT(0)
> +#define   MEDIA_POWERGATE_ENABLE		REG_BIT(1)
>  #define   VDN_HCP_POWERGATE_ENABLE(n)		REG_BIT(3 + 2 * (n))
>  #define   VDN_MFXVDENC_POWERGATE_ENABLE(n)	REG_BIT(4 + 2 * (n))
>  
> diff --git a/drivers/gpu/drm/xe/xe_gt.c b/drivers/gpu/drm/xe/xe_gt.c
> index 36c7b1631fa6..8a09630f5f38 100644
> --- a/drivers/gpu/drm/xe/xe_gt.c
> +++ b/drivers/gpu/drm/xe/xe_gt.c
> @@ -567,6 +567,10 @@ int xe_gt_init(struct xe_gt *gt)
>  	if (err)
>  		return err;
>  
> +	err = xe_gt_idle_enable_pg(gt);
> +	if (err)
> +		return err;
> +
>  	return drmm_add_action_or_reset(&gt_to_xe(gt)->drm, gt_fini, gt);
>  }
>  
> @@ -624,6 +628,10 @@ static int do_gt_restart(struct xe_gt *gt)
>  	if (err)
>  		return err;
>  
> +	err = xe_gt_idle_enable_pg(gt);
> +	if (err)
> +		return err;
> +
>  	for_each_hw_engine(hwe, gt, id) {
>  		xe_reg_sr_apply_mmio(&hwe->reg_sr, gt);
>  		xe_reg_sr_apply_whitelist(hwe);
> @@ -720,6 +728,8 @@ void xe_gt_suspend_prepare(struct xe_gt *gt)
>  {
>  	XE_WARN_ON(xe_force_wake_get(gt_to_fw(gt), XE_FORCEWAKE_ALL));
>  
> +	xe_gt_idle_disable_pg(gt);
> +

they are not the symmetric equivalents... we should probably find
a more suitable symmetric place for that.

>  	xe_uc_stop_prepare(&gt->uc);
>  
>  	XE_WARN_ON(xe_force_wake_put(gt_to_fw(gt), XE_FORCEWAKE_ALL));
> diff --git a/drivers/gpu/drm/xe/xe_gt_idle.c b/drivers/gpu/drm/xe/xe_gt_idle.c
> index 4384f7e80258..f549bbb0e110 100644
> --- a/drivers/gpu/drm/xe/xe_gt_idle.c
> +++ b/drivers/gpu/drm/xe/xe_gt_idle.c
> @@ -12,6 +12,7 @@
>  #include "xe_gt_sysfs.h"
>  #include "xe_guc_pc.h"
>  #include "regs/xe_gt_regs.h"
> +#include "xe_macros.h"
>  #include "xe_mmio.h"
>  #include "xe_pm.h"
>  
> @@ -93,6 +94,42 @@ static u64 get_residency_ms(struct xe_gt_idle *gtidle, u64 cur_residency)
>  	return cur_residency;
>  }
>  
> +int xe_gt_idle_enable_pg(struct xe_gt *gt)
> +{
> +	int ret, i;
> +	u32 pg_enable;
> +
> +	xe_device_assert_mem_access(gt_to_xe(gt));
> +
> +	pg_enable = RENDER_POWERGATE_ENABLE | MEDIA_POWERGATE_ENABLE;
> +
> +	for (i = XE_HW_ENGINE_VCS0; i <= XE_HW_ENGINE_VCS7; i++) {
> +		if ((gt->info.engine_mask & BIT(i)))
> +			pg_enable |= (VDN_HCP_POWERGATE_ENABLE(i) |
> +				      VDN_MFXVDENC_POWERGATE_ENABLE(i));
> +	}
> +
> +	ret = xe_force_wake_get(gt_to_fw(gt), XE_FW_GT);
> +	if (ret)
> +		return ret;
> +
> +	xe_mmio_write32(gt, POWERGATE_ENABLE, pg_enable);
> +
> +	XE_WARN_ON(xe_force_wake_put(gt_to_fw(gt), XE_FW_GT));
> +
> +	return 0;
> +}
> +
> +void xe_gt_idle_disable_pg(struct xe_gt *gt)
> +{
> +	xe_device_assert_mem_access(gt_to_xe(gt));
> +	XE_WARN_ON(xe_force_wake_get(gt_to_fw(gt), XE_FW_GT));
> +
> +	xe_mmio_write32(gt, POWERGATE_ENABLE, 0);
> +
> +	XE_WARN_ON(xe_force_wake_put(gt_to_fw(gt), XE_FW_GT));
> +}
> +
>  static ssize_t name_show(struct device *dev,
>  			 struct device_attribute *attr, char *buff)
>  {
> @@ -145,9 +182,12 @@ static const struct attribute *gt_idle_attrs[] = {
>  	NULL,
>  };
>  
> -static void gt_idle_sysfs_fini(struct drm_device *drm, void *arg)
> +static void gt_idle_fini(struct drm_device *drm, void *arg)
>  {
>  	struct kobject *kobj = arg;
> +	struct xe_gt *gt = kobj_to_gt(kobj->parent);
> +
> +	xe_gt_idle_disable_pg(gt);
>  
>  	sysfs_remove_files(kobj, gt_idle_attrs);
>  	kobject_put(kobj);
> @@ -182,7 +222,7 @@ int xe_gt_idle_sysfs_init(struct xe_gt_idle *gtidle)

we should perhaps also change the name of the init function
to ensure symmetry?

>  		return err;
>  	}
>  
> -	return drmm_add_action_or_reset(&xe->drm, gt_idle_sysfs_fini, kobj);
> +	return drmm_add_action_or_reset(&xe->drm, gt_idle_fini, kobj);
>  }
>  
>  void xe_gt_idle_enable_c6(struct xe_gt *gt)
> @@ -202,7 +242,6 @@ void xe_gt_idle_disable_c6(struct xe_gt *gt)
>  	xe_device_assert_mem_access(gt_to_xe(gt));
>  	xe_force_wake_assert_held(gt_to_fw(gt), XE_FORCEWAKE_ALL);
>  
> -	xe_mmio_write32(gt, POWERGATE_ENABLE, 0);
>  	xe_mmio_write32(gt, RC_CONTROL, 0);
>  	xe_mmio_write32(gt, RC_STATE, 0);
>  }
> diff --git a/drivers/gpu/drm/xe/xe_gt_idle.h b/drivers/gpu/drm/xe/xe_gt_idle.h
> index 75bd99659b1b..ba7aa36348fc 100644
> --- a/drivers/gpu/drm/xe/xe_gt_idle.h
> +++ b/drivers/gpu/drm/xe/xe_gt_idle.h
> @@ -13,5 +13,7 @@ struct xe_gt;
>  int xe_gt_idle_sysfs_init(struct xe_gt_idle *gtidle);
>  void xe_gt_idle_enable_c6(struct xe_gt *gt);
>  void xe_gt_idle_disable_c6(struct xe_gt *gt);
> +void xe_gt_idle_disable_pg(struct xe_gt *gt);
> +int  xe_gt_idle_enable_pg(struct xe_gt *gt);
>  
>  #endif /* _XE_GT_IDLE_H_ */
> -- 
> 2.40.0
> 


More information about the Intel-xe mailing list