[Intel-gfx] [PATCH] drm/i915/cnl: Add support slice/subslice/eu configs
Oscar Mateo
oscar.mateo at intel.com
Tue Sep 5 22:14:30 UTC 2017
On 08/31/2017 04:59 PM, Rodrigo Vivi wrote:
> From: Ben Widawsky <ben at bwidawsk.net>
>
> Cannonlake Slice and Subslice information has changed.
> This Patch provided by Ben adds the proper sseu
> initialization.
>
> v2: This v2 done by Rodrigo includes:
> - Fix on Total slices count by avoiding [1][2] and [2][2].
> - Inclusion of EU Per Subslice.
> - Commit message.
>
> Cc: Oscar Mateo <oscar.mateo at intel.com>
> Signed-off-by: Ben Widawsky <ben at bwidawsk.net>
> Signed-off-by: Rodrigo Vivi <rodrigo.vivi at intel.com>
> ---
> drivers/gpu/drm/i915/i915_drv.h | 5 +++-
> drivers/gpu/drm/i915/i915_reg.h | 20 +++++++++++++
> drivers/gpu/drm/i915/intel_device_info.c | 50 +++++++++++++++++++++++++++++++-
> 3 files changed, 73 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
> index 0383e879a315..2fdd59e85189 100644
> --- a/drivers/gpu/drm/i915/i915_drv.h
> +++ b/drivers/gpu/drm/i915/i915_drv.h
> @@ -786,7 +786,10 @@ struct sseu_dev_info {
> u8 slice_mask;
> u8 subslice_mask;
> u8 eu_total;
> - u8 eu_per_subslice;
> + union {
> + u8 per_subslice_eu_disable_mask[3][3];
> + u8 eu_per_subslice;
> + };
> u8 min_eu_in_pool;
> /* For each slice, which subslice(s) has(have) 7 EUs (bitfield)? */
> u8 subslice_7eu[3];
> diff --git a/drivers/gpu/drm/i915/i915_reg.h b/drivers/gpu/drm/i915/i915_reg.h
> index 5651081ff789..7f71007baa94 100644
> --- a/drivers/gpu/drm/i915/i915_reg.h
> +++ b/drivers/gpu/drm/i915/i915_reg.h
> @@ -2730,6 +2730,11 @@ enum i915_power_well_id {
> #define GEN9_F2_SS_DIS_SHIFT 20
> #define GEN9_F2_SS_DIS_MASK (0xf << GEN9_F2_SS_DIS_SHIFT)
>
> +#define GEN10_F2_S_ENA_SHIFT 22
> +#define GEN10_F2_S_ENA_MASK (0x7 << GEN10_F2_S_ENA_SHIFT)
> +#define GEN10_F2_SS_DIS_SHIFT 18
> +#define GEN10_F2_SS_DIS_MASK (0x7 << GEN10_F2_SS_DIS_SHIFT)
> +
Technically, the BSpec allows 6 bits for slice_enable (of which only 4
are used) and 4 bits for subslice_disable, instead of 3 and 3, right?
Even if some variants are not productized, isn't it better to cover our
back and read the fuses completely?
> #define GEN8_EU_DISABLE0 _MMIO(0x9134)
> #define GEN8_EU_DIS0_S0_MASK 0xffffff
> #define GEN8_EU_DIS0_S1_SHIFT 24
> @@ -2745,6 +2750,21 @@ enum i915_power_well_id {
>
> #define GEN9_EU_DISABLE(slice) _MMIO(0x9134 + (slice)*0x4)
>
> +#define GEN10_EU_DIS0_S0_SHIFT 0
> +#define GEN10_EU_DIS0_S0_MASK (0xff << GEN10_EU_DIS0_S0_SHIFT)
> +#define GEN10_EU_DIS0_S1_SHIFT 8
> +#define GEN10_EU_DIS0_S1_MASK (0xff << GEN10_EU_DIS0_S1_SHIFT)
> +#define GEN10_EU_DIS0_S2_SHIFT 16
> +#define GEN10_EU_DIS0_S2_MASK (0xff << GEN10_EU_DIS0_S2_SHIFT)
> +#define GEN10_EU_DIS1_S0_SHIFT 24
> +#define GEN10_EU_DIS1_S0_MASK (0xff << GEN10_EU_DIS1_S0_SHIFT)
> +#define GEN10_EU_DIS1_S1_SHIFT 0
> +#define GEN10_EU_DIS1_S1_MASK (0xff << GEN10_EU_DIS1_S1_SHIFT)
> +#define GEN10_EU_DIS2_S0_SHIFT 8
> +#define GEN10_EU_DIS2_S0_MASK (0xff << GEN10_EU_DIS2_S0_SHIFT)
> +#define GEN10_EU_DIS2_S1_SHIFT 16
> +#define GEN10_EU_DIS2_S1_MASK (0xff << GEN10_EU_DIS2_S1_SHIFT)
> +
I suggest a naming like:
GEN10_EU_DIS_Sm_SSn_SHIFT
where n is the slice number and m the subslice number
Also, related to my previous comment, there should be a:
#define GEN10_EU_DIS3_S0_SHIFT 24
#define GEN10_EU_DIS3_S0_MASK (0xff << GEN10_EU_DIS3_S0_SHIFT)
> #define GEN6_BSD_SLEEP_PSMI_CONTROL _MMIO(0x12050)
> #define GEN6_BSD_SLEEP_MSG_DISABLE (1 << 0)
> #define GEN6_BSD_SLEEP_FLUSH_DISABLE (1 << 2)
> diff --git a/drivers/gpu/drm/i915/intel_device_info.c b/drivers/gpu/drm/i915/intel_device_info.c
> index 5f91ddc78c7a..da13becf97f1 100644
> --- a/drivers/gpu/drm/i915/intel_device_info.c
> +++ b/drivers/gpu/drm/i915/intel_device_info.c
> @@ -82,6 +82,52 @@ void intel_device_info_dump(struct drm_i915_private *dev_priv)
> #undef PRINT_FLAG
> }
>
> +static void gen10_sseu_info_init(struct drm_i915_private *dev_priv)
> +{
> + struct sseu_dev_info *sseu = &mkwrite_device_info(dev_priv)->sseu;
> + const u32 fuse2 = I915_READ(GEN8_FUSE2);
> + u32 temp, i, j;
> +
> + sseu->slice_mask = (fuse2 & GEN10_F2_S_ENA_MASK) >> GEN10_F2_S_ENA_SHIFT;
> + sseu->subslice_mask = (1 << 3) - 1;
> + sseu->subslice_mask &= ~((fuse2 & GEN10_F2_SS_DIS_MASK) >>
> + GEN10_F2_SS_DIS_SHIFT);
See my comments above about slice_enable and subslice_disable masks.
> + temp = I915_READ(GEN8_EU_DISABLE0);
> + sseu->per_subslice_eu_disable_mask[0][0] =
> + (temp & GEN10_EU_DIS0_S0_MASK) >> GEN10_EU_DIS0_S0_SHIFT;
> + sseu->per_subslice_eu_disable_mask[0][1] =
> + (temp & GEN10_EU_DIS0_S1_MASK) >> GEN10_EU_DIS0_S1_SHIFT;
> + sseu->per_subslice_eu_disable_mask[0][2] =
> + (temp & GEN10_EU_DIS0_S2_MASK) >> GEN10_EU_DIS0_S2_SHIFT;
> + sseu->per_subslice_eu_disable_mask[1][0] =
> + (temp & GEN10_EU_DIS1_S0_MASK) >> GEN10_EU_DIS1_S0_SHIFT;
> +
> + temp = I915_READ(GEN8_EU_DISABLE1);
> + sseu->per_subslice_eu_disable_mask[1][1] =
> + (temp & GEN10_EU_DIS1_S1_MASK) >> GEN10_EU_DIS1_S1_SHIFT;
> + sseu->per_subslice_eu_disable_mask[2][0] =
> + (temp & GEN10_EU_DIS2_S0_MASK) >> GEN10_EU_DIS2_S0_SHIFT;
> + sseu->per_subslice_eu_disable_mask[2][1] =
> + (temp & GEN10_EU_DIS2_S1_MASK) >> GEN10_EU_DIS2_S1_SHIFT;
> +
> + for (i = 0; i < 3; i++)
> + for (j = 0; j < (i == 0 ? 3 : 2); j++)
> + sseu->eu_total +=
> + hweight8(~sseu->per_subslice_eu_disable_mask[i][j]);
Same thing as before.
Also: don't we want to read all of this back in debugfs
(i915_sseu_status)? (although maybe there is a separate patch for this
that I have missed)
> + /* Let's assume uniform distribution */
> + sseu->eu_per_subslice = sseu_subslice_total(sseu) ?
> + DIV_ROUND_UP(sseu->eu_total,
> + sseu_subslice_total(sseu)) : 0;
Can we add a comment similar to the one for Gen9?:
/*
* CNL is expected to always have a uniform distribution
* of EU across subslices with the exception that any one
* EU in any one subslice may be fused off for die
* recovery.
*/
> + /* No restrictions on Power Gating */
> + sseu->has_slice_pg = 1;
> + sseu->has_subslice_pg = 1;
> + sseu->has_eu_pg = 1;
> +}
> +
> +
> static void cherryview_sseu_info_init(struct drm_i915_private *dev_priv)
> {
> struct sseu_dev_info *sseu = &mkwrite_device_info(dev_priv)->sseu;
> @@ -409,8 +455,10 @@ void intel_device_info_runtime_init(struct drm_i915_private *dev_priv)
> cherryview_sseu_info_init(dev_priv);
> else if (IS_BROADWELL(dev_priv))
> broadwell_sseu_info_init(dev_priv);
> - else if (INTEL_INFO(dev_priv)->gen >= 9)
> + else if (INTEL_INFO(dev_priv)->gen == 9)
> gen9_sseu_info_init(dev_priv);
> + else if (INTEL_INFO(dev_priv)->gen >= 10)
> + gen10_sseu_info_init(dev_priv);
Maybe leverage this change to move to INTEL_GEN(dev_priv)?
>
> info->has_snoop = !info->has_llc;
>
More information about the Intel-gfx
mailing list