Re: [Intel-gfx] [PATCH v8 4/4] drm/i915: Correctly map DBUF slices to pipes

["Ville Syrjälä" <ville.syrjala@linux.intel.com>]

On Fri, Dec 13, 2019 at 03:02:28PM +0200, Stanislav Lisovskiy wrote:
> Added proper DBuf slice mapping to correspondent
> pipes, depending on pipe configuration as stated
> in BSpec.
> 
> v2:
>     - Remove unneeded braces
>     - Stop using macro for DBuf assignments as
>       it seems to reduce readability.
> 
> v3: Start using enabled slices mask in dev_priv
> 
> v4: Renamed "enabled_slices" used in dev_priv
>     to "enabled_dbuf_slices_mask"(Matt Roper)
> 
> Signed-off-by: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com>
> ---
>  drivers/gpu/drm/i915/intel_pm.c | 226 ++++++++++++++++++++++++++++++--
>  1 file changed, 216 insertions(+), 10 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/intel_pm.c b/drivers/gpu/drm/i915/intel_pm.c
> index 111bcafd6e4c..a13052b2c2ef 100644
> --- a/drivers/gpu/drm/i915/intel_pm.c
> +++ b/drivers/gpu/drm/i915/intel_pm.c
> @@ -3832,13 +3832,30 @@ bool intel_can_enable_sagv(struct intel_atomic_state *state)
>  	return true;
>  }
>  
> +/*
> + * Calculate initial DBuf slice offset, based on slice size
> + * and mask(i.e if slice size is 1024 and second slice is enabled
> + * offset would be 1024)
> + */
> +static u32 icl_get_first_dbuf_slice_offset(u32 dbuf_slice_mask,
> +					   u32 slice_size, u32 ddb_size)

s/u32/unsigned int/ or something. No point in using sized types when
we're not dealing with some thing that is actually that size.

> +{
> +	u32 offset = 0;
> +
> +	if (!dbuf_slice_mask)
> +		return 0;
> +
> +	offset = (ffs(dbuf_slice_mask) - 1) * slice_size;
> +
> +	WARN_ON(offset >= ddb_size);
> +	return offset;
> +}
> +
>  static u16 intel_get_ddb_size(struct drm_i915_private *dev_priv,
>  			      const struct intel_crtc_state *crtc_state,
>  			      const u64 total_data_rate,
>  			      const int num_active)
>  {
> -	struct drm_atomic_state *state = crtc_state->uapi.state;
> -	struct intel_atomic_state *intel_state = to_intel_atomic_state(state);
>  	u16 ddb_size = INTEL_INFO(dev_priv)->ddb_size;
>  
>  	WARN_ON(ddb_size == 0);
> @@ -3846,12 +3863,13 @@ static u16 intel_get_ddb_size(struct drm_i915_private *dev_priv,
>  	if (INTEL_GEN(dev_priv) < 11)
>  		return ddb_size - 4; /* 4 blocks for bypass path allocation */
>  
> -	intel_state->enabled_dbuf_slices_mask = DBUF_S1_BIT;
> -	ddb_size /= 2;
> -
>  	return ddb_size;
>  }
>  
> +u32 i915_possible_dbuf_slices(struct drm_i915_private *dev_priv,
> +			      int pipe, u32 active_pipes,
> +			      const struct intel_crtc_state *crtc_state);
> +
>  static void
>  skl_ddb_get_pipe_allocation_limits(struct drm_i915_private *dev_priv,
>  				   const struct intel_crtc_state *crtc_state,
> @@ -3866,7 +3884,14 @@ skl_ddb_get_pipe_allocation_limits(struct drm_i915_private *dev_priv,
>  	u32 pipe_width = 0, total_width = 0, width_before_pipe = 0;
>  	enum pipe for_pipe = to_intel_crtc(for_crtc)->pipe;
>  	u16 ddb_size;
> +	u32 ddb_range_size;
>  	u32 i;
> +	u32 dbuf_slice_mask;
> +	u32 active_pipes;
> +	u32 offset;
> +	u32 slice_size;
> +	u32 total_slice_mask;
> +	u32 start, end;
>  
>  	if (WARN_ON(!state) || !crtc_state->hw.active) {
>  		alloc->start = 0;
> @@ -3875,14 +3900,19 @@ skl_ddb_get_pipe_allocation_limits(struct drm_i915_private *dev_priv,
>  		return;
>  	}
>  
> -	if (intel_state->active_pipe_changes)
> +	if (intel_state->active_pipe_changes) {
>  		*num_active = hweight8(intel_state->active_pipes);
> -	else
> +		active_pipes = intel_state->active_pipes;
> +	} else {
>  		*num_active = hweight8(dev_priv->active_pipes);
> +		active_pipes = dev_priv->active_pipes;
> +	}
>  
>  	ddb_size = intel_get_ddb_size(dev_priv, crtc_state, total_data_rate,
>  				      *num_active);
>  
> +	slice_size = ddb_size / INTEL_INFO(dev_priv)->num_supported_dbuf_slices;
> +
>  	/*
>  	 * If the state doesn't change the active CRTC's or there is no
>  	 * modeset request, then there's no need to recalculate;
> @@ -3900,18 +3930,68 @@ skl_ddb_get_pipe_allocation_limits(struct drm_i915_private *dev_priv,
>  		return;
>  	}
>  
> +	/*
> +	 * Get allowed DBuf slices for correspondent pipe and platform.
> +	 */
> +	dbuf_slice_mask = i915_possible_dbuf_slices(dev_priv, for_pipe,
> +						    active_pipes, crtc_state);
> +
> +	DRM_DEBUG_KMS("DBuf slice mask %x pipe %d active pipes %x\n",
> +		      dbuf_slice_mask,
> +		      for_pipe, active_pipes);
> +
> +	/*
> +	 * Figure out at which DBuf slice we start, i.e if we start at Dbuf S2
> +	 * and slice size is 1024, the offset would be 1024
> +	 */
> +	offset = icl_get_first_dbuf_slice_offset(dbuf_slice_mask,
> +						 slice_size, ddb_size);
> +
> +	/*
> +	 * Figure out total size of allowed DBuf slices, which is basically
> +	 * a number of allowed slices for that pipe multiplied by slice size.
> +	 * Inside of this
> +	 * range ddb entries are still allocated in proportion to display width.
> +	 */
> +	ddb_range_size = hweight8(dbuf_slice_mask) * slice_size;
> +
>  	/*
>  	 * Watermark/ddb requirement highly depends upon width of the
>  	 * framebuffer, So instead of allocating DDB equally among pipes
>  	 * distribute DDB based on resolution/width of the display.
>  	 */
> +	total_slice_mask = dbuf_slice_mask;
>  	for_each_new_intel_crtc_in_state(intel_state, crtc, crtc_state, i) {
>  		const struct drm_display_mode *adjusted_mode =
>  			&crtc_state->hw.adjusted_mode;
>  		enum pipe pipe = crtc->pipe;
>  		int hdisplay, vdisplay;
> +		u32 pipe_dbuf_slice_mask =
> +					i915_possible_dbuf_slices(dev_priv,
> +								  pipe,
> +								  active_pipes,
> +								  crtc_state);
> +
> +		if (!crtc_state->hw.active)
> +			continue;
> +
> +		/*
> +		 * According to BSpec pipe can share one dbuf slice with another
> +		 * pipes or pipe can use multiple dbufs, in both cases we
> +		 * account for other pipes only if they have exactly same mask.
> +		 * However we need to account how many slices we should enable
> +		 * in total.
> +		 */
> +		total_slice_mask |= pipe_dbuf_slice_mask;
>  
> -		if (!crtc_state->hw.enable)
> +		/*
> +		 * Do not account pipes using other slice sets
> +		 * luckily as of current BSpec slice sets do not partially
> +		 * intersect(pipes share either same one slice or same slice set
> +		 * i.e no partial intersection), so it is enough to check for
> +		 * equality for now.
> +		 */
> +		if (dbuf_slice_mask != pipe_dbuf_slice_mask)
>  			continue;
>  
>  		drm_mode_get_hv_timing(adjusted_mode, &hdisplay, &vdisplay);
> @@ -3923,8 +4003,19 @@ skl_ddb_get_pipe_allocation_limits(struct drm_i915_private *dev_priv,
>  			pipe_width = hdisplay;
>  	}
>  
> -	alloc->start = ddb_size * width_before_pipe / total_width;
> -	alloc->end = ddb_size * (width_before_pipe + pipe_width) / total_width;
> +	intel_state->enabled_dbuf_slices_mask = total_slice_mask;
> +
> +	start = ddb_range_size * width_before_pipe / total_width;
> +	end = ddb_range_size * (width_before_pipe + pipe_width) / total_width;
> +
> +	alloc->start = offset + start;
> +	alloc->end = offset + end;
> +
> +	DRM_DEBUG_KMS("Pipe %d ddb %d-%d\n", for_pipe,
> +		      alloc->start, alloc->end);
> +	DRM_DEBUG_KMS("Enabled ddb slices mask %x num supported %d\n",
> +		      intel_state->enabled_dbuf_slices_mask,
> +		      INTEL_INFO(dev_priv)->num_supported_dbuf_slices);
>  }

That function really is quite the monster. Probably needs to be split up
into sensible chunks... later.

>  
>  static int skl_compute_wm_params(const struct intel_crtc_state *crtc_state,
> @@ -4094,6 +4185,121 @@ skl_plane_downscale_amount(const struct intel_crtc_state *crtc_state,
>  	return mul_fixed16(downscale_w, downscale_h);
>  }
>  
> +struct dbuf_slice_conf_entry {
> +	u32 active_pipes;
> +	u32 dbuf_mask[I915_MAX_PIPES];

u8 will do for everything.

> +};
> +
> +/*
> + * Table taken from Bspec 12716
> + * Pipes do have some preferred DBuf slice affinity,
> + * plus there are some hardcoded requirements on how
> + * those should be distributed for multipipe scenarios.
> + * For more DBuf slices algorithm can get even more messy
> + * and less readable, so decided to use a table almost
> + * as is from BSpec itself - that way it is at least easier
> + * to compare, change and check.
> + */
> +static struct dbuf_slice_conf_entry icl_allowed_dbufs[] = {
> +	{ BIT(PIPE_A), { DBUF_S1_BIT, 0, 0, 0 } },

Still don't like that DBUF_S1_BIT. I still think it should be an
enum dbuf_slice + explcit BIT(DBUF_S1) where needed.

And please use named initializers instead of sprinkling zeroes.
IMO this should just look something like:

	{ .active_pipes = BIT(PIPE_A) | BIT(PIPE_B),
	  .dbuf_slices = { [PIPE_A] = BIT(DBUF_S1),
	  		   [PIPE_B] = BIT(DBUF_S2),
			 },
	},

> +	{ BIT(PIPE_B), { 0, DBUF_S1_BIT, 0, 0 } },
> +	{ BIT(PIPE_C), { 0, 0, DBUF_S2_BIT, 0 } },
> +	{ BIT(PIPE_A) | BIT(PIPE_B), { DBUF_S1_BIT, DBUF_S2_BIT, 0, 0 } },
> +	{ BIT(PIPE_A) | BIT(PIPE_C), { DBUF_S1_BIT, 0, DBUF_S2_BIT, 0 } },
> +	{ BIT(PIPE_B) | BIT(PIPE_C), { 0, DBUF_S1_BIT, DBUF_S2_BIT, 0 } },
> +	{ BIT(PIPE_A) | BIT(PIPE_B) | BIT(PIPE_C),
> +		{ DBUF_S1_BIT, DBUF_S1_BIT, DBUF_S2_BIT, 0 } }
> +};
> +
> +/*
> + * Table taken from Bspec 49255
> + * Pipes do have some preferred DBuf slice affinity,
> + * plus there are some hardcoded requirements on how
> + * those should be distributed for multipipe scenarios.
> + * For more DBuf slices algorithm can get even more messy
> + * and less readable, so decided to use a table almost
> + * as is from BSpec itself - that way it is at least easier
> + * to compare, change and check.
> + */
> +static struct dbuf_slice_conf_entry tgl_allowed_dbufs[] = {
> +	{ BIT(PIPE_A), { DBUF_S1_BIT | DBUF_S2_BIT, 0, 0, 0 } },
> +	{ BIT(PIPE_B), { 0, DBUF_S1_BIT | DBUF_S2_BIT, 0, 0 } },
> +	{ BIT(PIPE_C), { 0, 0, DBUF_S1_BIT | DBUF_S2_BIT, 0 } },
> +	{ BIT(PIPE_D), { 0, 0, 0, DBUF_S1_BIT | DBUF_S2_BIT } },
> +	{ BIT(PIPE_A) | BIT(PIPE_B), { DBUF_S2_BIT, DBUF_S1_BIT, 0, 0 } },
> +	{ BIT(PIPE_A) | BIT(PIPE_C), { DBUF_S1_BIT, 0, DBUF_S2_BIT, 0 } },
> +	{ BIT(PIPE_A) | BIT(PIPE_D), { DBUF_S1_BIT, 0, 0, DBUF_S2_BIT } },
> +	{ BIT(PIPE_B) | BIT(PIPE_C), { 0, DBUF_S1_BIT, DBUF_S2_BIT, 0 } },
> +	{ BIT(PIPE_B) | BIT(PIPE_D), { 0, DBUF_S1_BIT, 0, DBUF_S2_BIT } },
> +	{ BIT(PIPE_C) | BIT(PIPE_D), { 0, 0, DBUF_S2_BIT, DBUF_S2_BIT } },

I believe pipe C should use S1 here.

> +	{ BIT(PIPE_A) | BIT(PIPE_B) | BIT(PIPE_C),
> +		{ DBUF_S1_BIT, DBUF_S1_BIT, DBUF_S2_BIT, 0 } },
> +	{ BIT(PIPE_A) | BIT(PIPE_B) | BIT(PIPE_D),
> +		{ DBUF_S1_BIT, DBUF_S1_BIT, 0, DBUF_S2_BIT } },
> +	{ BIT(PIPE_A) | BIT(PIPE_C) | BIT(PIPE_D),
> +		{ DBUF_S1_BIT, 0, DBUF_S2_BIT, DBUF_S2_BIT } },
> +	{ BIT(PIPE_B) | BIT(PIPE_C) | BIT(PIPE_D),
> +		{ 0, DBUF_S1_BIT, DBUF_S2_BIT, DBUF_S2_BIT } },
> +	{ BIT(PIPE_A) | BIT(PIPE_B) | BIT(PIPE_C) | BIT(PIPE_D),
> +		{ DBUF_S1_BIT, DBUF_S1_BIT, DBUF_S2_BIT, DBUF_S2_BIT } },
> +};
> +
> +static u32 i915_find_pipe_conf(int pipe,

enum pipe pipe

> +			       u32 active_pipes,
> +			       const struct dbuf_slice_conf_entry *dbuf_slices,
> +			       int size)

Could just add a sentinel to the end of the arrays and avoid this
'size' stuff.

The function name is still very confusing. icl_compute_dbuf_slices()
or something perhaps. Or even without a prefix since it's static
anyway.

> +{
> +	int i;
> +
> +	for (i = 0; i < size; i++) {
> +		if (dbuf_slices[i].active_pipes == active_pipes)
> +			return dbuf_slices[i].dbuf_mask[pipe];
> +	}
> +	return 0;
> +}
> +
> +/*
> + * This function finds an entry with same enabled pipe configuration and
> + * returns correspondent DBuf slice mask as stated in BSpec for particular
> + * platform.
> + */
> +static u32 icl_possible_dbuf_slices(int pipe,
> +				    u32 active_pipes,
> +				    const struct intel_crtc_state *crtc_state)
> +{
> +	return i915_find_pipe_conf(pipe, active_pipes,
> +				   icl_allowed_dbufs,
> +				   ARRAY_SIZE(icl_allowed_dbufs));
> +}
> +
> +static u32 tgl_possible_dbuf_slices(int pipe,
> +				    u32 active_pipes,
> +				    const struct intel_crtc_state *crtc_state)
> +{
> +	return i915_find_pipe_conf(pipe, active_pipes,
> +				   tgl_allowed_dbufs,
> +				   ARRAY_SIZE(tgl_allowed_dbufs));
> +}

These wrappers aren't really doing anything for us.

> +
> +u32 i915_possible_dbuf_slices(struct drm_i915_private *dev_priv,

dev_priv and pipe are redundant parameters. Both
can be extracted from crtc_state. I would put crtc_state as the first
parameter. Matches common practice more closely.

s/i915/skl/ in the function name.

> +			      int pipe, u32 active_pipes,
> +			      const struct intel_crtc_state *crtc_state)
> +{
> +	if (IS_GEN(dev_priv, 11))
> +		return icl_possible_dbuf_slices(pipe,
> +						active_pipes,
> +						crtc_state);
> +	else if (IS_GEN(dev_priv, 12))
> +		return tgl_possible_dbuf_slices(pipe,
> +						active_pipes,
> +						crtc_state);

The established convention is to go from new platforms
to old platforms, if possible.

> +	/*
> +	 * For anything else just return one slice yet.
> +	 * Should be extended for other platforms.
> +	 */
> +	return DBUF_S1_BIT;
> +}
> +
>  static u64
>  skl_plane_relative_data_rate(const struct intel_crtc_state *crtc_state,
>  			     const struct intel_plane_state *plane_state,
> -- 
> 2.17.1

-- 
Ville Syrjälä
Intel
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx