diff options
Diffstat (limited to 'drivers/gpu/drm/i915/gt/intel_sseu.c')
-rw-r--r-- | drivers/gpu/drm/i915/gt/intel_sseu.c | 591 |
1 files changed, 589 insertions, 2 deletions
diff --git a/drivers/gpu/drm/i915/gt/intel_sseu.c b/drivers/gpu/drm/i915/gt/intel_sseu.c index d173271c7397..f1c039e1b5ad 100644 --- a/drivers/gpu/drm/i915/gt/intel_sseu.c +++ b/drivers/gpu/drm/i915/gt/intel_sseu.c @@ -60,10 +60,552 @@ intel_sseu_subslices_per_slice(const struct sseu_dev_info *sseu, u8 slice) return hweight32(intel_sseu_get_subslices(sseu, slice)); } -u32 intel_sseu_make_rpcs(struct drm_i915_private *i915, +static int sseu_eu_idx(const struct sseu_dev_info *sseu, int slice, + int subslice) +{ + int slice_stride = sseu->max_subslices * sseu->eu_stride; + + return slice * slice_stride + subslice * sseu->eu_stride; +} + +static u16 sseu_get_eus(const struct sseu_dev_info *sseu, int slice, + int subslice) +{ + int i, offset = sseu_eu_idx(sseu, slice, subslice); + u16 eu_mask = 0; + + for (i = 0; i < sseu->eu_stride; i++) + eu_mask |= + ((u16)sseu->eu_mask[offset + i]) << (i * BITS_PER_BYTE); + + return eu_mask; +} + +static void sseu_set_eus(struct sseu_dev_info *sseu, int slice, int subslice, + u16 eu_mask) +{ + int i, offset = sseu_eu_idx(sseu, slice, subslice); + + for (i = 0; i < sseu->eu_stride; i++) + sseu->eu_mask[offset + i] = + (eu_mask >> (BITS_PER_BYTE * i)) & 0xff; +} + +static u16 compute_eu_total(const struct sseu_dev_info *sseu) +{ + u16 i, total = 0; + + for (i = 0; i < ARRAY_SIZE(sseu->eu_mask); i++) + total += hweight8(sseu->eu_mask[i]); + + return total; +} + +static void gen11_compute_sseu_info(struct sseu_dev_info *sseu, + u8 s_en, u32 ss_en, u16 eu_en) +{ + int s, ss; + + /* ss_en represents entire subslice mask across all slices */ + GEM_BUG_ON(sseu->max_slices * sseu->max_subslices > + sizeof(ss_en) * BITS_PER_BYTE); + + for (s = 0; s < sseu->max_slices; s++) { + if ((s_en & BIT(s)) == 0) + continue; + + sseu->slice_mask |= BIT(s); + + intel_sseu_set_subslices(sseu, s, ss_en); + + for (ss = 0; ss < sseu->max_subslices; ss++) + if (intel_sseu_has_subslice(sseu, s, ss)) + sseu_set_eus(sseu, s, ss, eu_en); + } + sseu->eu_per_subslice = hweight16(eu_en); + sseu->eu_total = compute_eu_total(sseu); +} + +static void gen12_sseu_info_init(struct intel_gt *gt) +{ + struct sseu_dev_info *sseu = >->info.sseu; + struct intel_uncore *uncore = gt->uncore; + u32 dss_en; + u16 eu_en = 0; + u8 eu_en_fuse; + u8 s_en; + int eu; + + /* + * Gen12 has Dual-Subslices, which behave similarly to 2 gen11 SS. + * Instead of splitting these, provide userspace with an array + * of DSS to more closely represent the hardware resource. + */ + intel_sseu_set_info(sseu, 1, 6, 16); + + s_en = intel_uncore_read(uncore, GEN11_GT_SLICE_ENABLE) & + GEN11_GT_S_ENA_MASK; + + dss_en = intel_uncore_read(uncore, GEN12_GT_DSS_ENABLE); + + /* one bit per pair of EUs */ + eu_en_fuse = ~(intel_uncore_read(uncore, GEN11_EU_DISABLE) & + GEN11_EU_DIS_MASK); + for (eu = 0; eu < sseu->max_eus_per_subslice / 2; eu++) + if (eu_en_fuse & BIT(eu)) + eu_en |= BIT(eu * 2) | BIT(eu * 2 + 1); + + gen11_compute_sseu_info(sseu, s_en, dss_en, eu_en); + + /* TGL only supports slice-level power gating */ + sseu->has_slice_pg = 1; +} + +static void gen11_sseu_info_init(struct intel_gt *gt) +{ + struct sseu_dev_info *sseu = >->info.sseu; + struct intel_uncore *uncore = gt->uncore; + u32 ss_en; + u8 eu_en; + u8 s_en; + + if (IS_ELKHARTLAKE(gt->i915)) + intel_sseu_set_info(sseu, 1, 4, 8); + else + intel_sseu_set_info(sseu, 1, 8, 8); + + s_en = intel_uncore_read(uncore, GEN11_GT_SLICE_ENABLE) & + GEN11_GT_S_ENA_MASK; + ss_en = ~intel_uncore_read(uncore, GEN11_GT_SUBSLICE_DISABLE); + + eu_en = ~(intel_uncore_read(uncore, GEN11_EU_DISABLE) & + GEN11_EU_DIS_MASK); + + gen11_compute_sseu_info(sseu, s_en, ss_en, eu_en); + + /* ICL has no power gating restrictions. */ + sseu->has_slice_pg = 1; + sseu->has_subslice_pg = 1; + sseu->has_eu_pg = 1; +} + +static void gen10_sseu_info_init(struct intel_gt *gt) +{ + struct intel_uncore *uncore = gt->uncore; + struct sseu_dev_info *sseu = >->info.sseu; + const u32 fuse2 = intel_uncore_read(uncore, GEN8_FUSE2); + const int eu_mask = 0xff; + u32 subslice_mask, eu_en; + int s, ss; + + intel_sseu_set_info(sseu, 6, 4, 8); + + sseu->slice_mask = (fuse2 & GEN10_F2_S_ENA_MASK) >> + GEN10_F2_S_ENA_SHIFT; + + /* Slice0 */ + eu_en = ~intel_uncore_read(uncore, GEN8_EU_DISABLE0); + for (ss = 0; ss < sseu->max_subslices; ss++) + sseu_set_eus(sseu, 0, ss, (eu_en >> (8 * ss)) & eu_mask); + /* Slice1 */ + sseu_set_eus(sseu, 1, 0, (eu_en >> 24) & eu_mask); + eu_en = ~intel_uncore_read(uncore, GEN8_EU_DISABLE1); + sseu_set_eus(sseu, 1, 1, eu_en & eu_mask); + /* Slice2 */ + sseu_set_eus(sseu, 2, 0, (eu_en >> 8) & eu_mask); + sseu_set_eus(sseu, 2, 1, (eu_en >> 16) & eu_mask); + /* Slice3 */ + sseu_set_eus(sseu, 3, 0, (eu_en >> 24) & eu_mask); + eu_en = ~intel_uncore_read(uncore, GEN8_EU_DISABLE2); + sseu_set_eus(sseu, 3, 1, eu_en & eu_mask); + /* Slice4 */ + sseu_set_eus(sseu, 4, 0, (eu_en >> 8) & eu_mask); + sseu_set_eus(sseu, 4, 1, (eu_en >> 16) & eu_mask); + /* Slice5 */ + sseu_set_eus(sseu, 5, 0, (eu_en >> 24) & eu_mask); + eu_en = ~intel_uncore_read(uncore, GEN10_EU_DISABLE3); + sseu_set_eus(sseu, 5, 1, eu_en & eu_mask); + + subslice_mask = (1 << 4) - 1; + subslice_mask &= ~((fuse2 & GEN10_F2_SS_DIS_MASK) >> + GEN10_F2_SS_DIS_SHIFT); + + for (s = 0; s < sseu->max_slices; s++) { + u32 subslice_mask_with_eus = subslice_mask; + + for (ss = 0; ss < sseu->max_subslices; ss++) { + if (sseu_get_eus(sseu, s, ss) == 0) + subslice_mask_with_eus &= ~BIT(ss); + } + + /* + * Slice0 can have up to 3 subslices, but there are only 2 in + * slice1/2. + */ + intel_sseu_set_subslices(sseu, s, s == 0 ? + subslice_mask_with_eus : + subslice_mask_with_eus & 0x3); + } + + sseu->eu_total = compute_eu_total(sseu); + + /* + * 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. + */ + sseu->eu_per_subslice = + intel_sseu_subslice_total(sseu) ? + DIV_ROUND_UP(sseu->eu_total, intel_sseu_subslice_total(sseu)) : + 0; + + /* 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 intel_gt *gt) +{ + struct sseu_dev_info *sseu = >->info.sseu; + u32 fuse; + u8 subslice_mask = 0; + + fuse = intel_uncore_read(gt->uncore, CHV_FUSE_GT); + + sseu->slice_mask = BIT(0); + intel_sseu_set_info(sseu, 1, 2, 8); + + if (!(fuse & CHV_FGT_DISABLE_SS0)) { + u8 disabled_mask = + ((fuse & CHV_FGT_EU_DIS_SS0_R0_MASK) >> + CHV_FGT_EU_DIS_SS0_R0_SHIFT) | + (((fuse & CHV_FGT_EU_DIS_SS0_R1_MASK) >> + CHV_FGT_EU_DIS_SS0_R1_SHIFT) << 4); + + subslice_mask |= BIT(0); + sseu_set_eus(sseu, 0, 0, ~disabled_mask); + } + + if (!(fuse & CHV_FGT_DISABLE_SS1)) { + u8 disabled_mask = + ((fuse & CHV_FGT_EU_DIS_SS1_R0_MASK) >> + CHV_FGT_EU_DIS_SS1_R0_SHIFT) | + (((fuse & CHV_FGT_EU_DIS_SS1_R1_MASK) >> + CHV_FGT_EU_DIS_SS1_R1_SHIFT) << 4); + + subslice_mask |= BIT(1); + sseu_set_eus(sseu, 0, 1, ~disabled_mask); + } + + intel_sseu_set_subslices(sseu, 0, subslice_mask); + + sseu->eu_total = compute_eu_total(sseu); + + /* + * CHV expected to always have a uniform distribution of EU + * across subslices. + */ + sseu->eu_per_subslice = intel_sseu_subslice_total(sseu) ? + sseu->eu_total / + intel_sseu_subslice_total(sseu) : + 0; + /* + * CHV supports subslice power gating on devices with more than + * one subslice, and supports EU power gating on devices with + * more than one EU pair per subslice. + */ + sseu->has_slice_pg = 0; + sseu->has_subslice_pg = intel_sseu_subslice_total(sseu) > 1; + sseu->has_eu_pg = (sseu->eu_per_subslice > 2); +} + +static void gen9_sseu_info_init(struct intel_gt *gt) +{ + struct drm_i915_private *i915 = gt->i915; + struct intel_device_info *info = mkwrite_device_info(i915); + struct sseu_dev_info *sseu = >->info.sseu; + struct intel_uncore *uncore = gt->uncore; + u32 fuse2, eu_disable, subslice_mask; + const u8 eu_mask = 0xff; + int s, ss; + + fuse2 = intel_uncore_read(uncore, GEN8_FUSE2); + sseu->slice_mask = (fuse2 & GEN8_F2_S_ENA_MASK) >> GEN8_F2_S_ENA_SHIFT; + + /* BXT has a single slice and at most 3 subslices. */ + intel_sseu_set_info(sseu, IS_GEN9_LP(i915) ? 1 : 3, + IS_GEN9_LP(i915) ? 3 : 4, 8); + + /* + * The subslice disable field is global, i.e. it applies + * to each of the enabled slices. + */ + subslice_mask = (1 << sseu->max_subslices) - 1; + subslice_mask &= ~((fuse2 & GEN9_F2_SS_DIS_MASK) >> + GEN9_F2_SS_DIS_SHIFT); + + /* + * Iterate through enabled slices and subslices to + * count the total enabled EU. + */ + for (s = 0; s < sseu->max_slices; s++) { + if (!(sseu->slice_mask & BIT(s))) + /* skip disabled slice */ + continue; + + intel_sseu_set_subslices(sseu, s, subslice_mask); + + eu_disable = intel_uncore_read(uncore, GEN9_EU_DISABLE(s)); + for (ss = 0; ss < sseu->max_subslices; ss++) { + int eu_per_ss; + u8 eu_disabled_mask; + + if (!intel_sseu_has_subslice(sseu, s, ss)) + /* skip disabled subslice */ + continue; + + eu_disabled_mask = (eu_disable >> (ss * 8)) & eu_mask; + + sseu_set_eus(sseu, s, ss, ~eu_disabled_mask); + + eu_per_ss = sseu->max_eus_per_subslice - + hweight8(eu_disabled_mask); + + /* + * Record which subslice(s) has(have) 7 EUs. we + * can tune the hash used to spread work among + * subslices if they are unbalanced. + */ + if (eu_per_ss == 7) + sseu->subslice_7eu[s] |= BIT(ss); + } + } + + sseu->eu_total = compute_eu_total(sseu); + + /* + * SKL 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. BXT is expected to be perfectly uniform in EU + * distribution. + */ + sseu->eu_per_subslice = + intel_sseu_subslice_total(sseu) ? + DIV_ROUND_UP(sseu->eu_total, intel_sseu_subslice_total(sseu)) : + 0; + + /* + * SKL+ supports slice power gating on devices with more than + * one slice, and supports EU power gating on devices with + * more than one EU pair per subslice. BXT+ supports subslice + * power gating on devices with more than one subslice, and + * supports EU power gating on devices with more than one EU + * pair per subslice. + */ + sseu->has_slice_pg = + !IS_GEN9_LP(i915) && hweight8(sseu->slice_mask) > 1; + sseu->has_subslice_pg = + IS_GEN9_LP(i915) && intel_sseu_subslice_total(sseu) > 1; + sseu->has_eu_pg = sseu->eu_per_subslice > 2; + + if (IS_GEN9_LP(i915)) { +#define IS_SS_DISABLED(ss) (!(sseu->subslice_mask[0] & BIT(ss))) + info->has_pooled_eu = hweight8(sseu->subslice_mask[0]) == 3; + + sseu->min_eu_in_pool = 0; + if (info->has_pooled_eu) { + if (IS_SS_DISABLED(2) || IS_SS_DISABLED(0)) + sseu->min_eu_in_pool = 3; + else if (IS_SS_DISABLED(1)) + sseu->min_eu_in_pool = 6; + else + sseu->min_eu_in_pool = 9; + } +#undef IS_SS_DISABLED + } +} + +static void bdw_sseu_info_init(struct intel_gt *gt) +{ + struct sseu_dev_info *sseu = >->info.sseu; + struct intel_uncore *uncore = gt->uncore; + int s, ss; + u32 fuse2, subslice_mask, eu_disable[3]; /* s_max */ + u32 eu_disable0, eu_disable1, eu_disable2; + + fuse2 = intel_uncore_read(uncore, GEN8_FUSE2); + sseu->slice_mask = (fuse2 & GEN8_F2_S_ENA_MASK) >> GEN8_F2_S_ENA_SHIFT; + intel_sseu_set_info(sseu, 3, 3, 8); + + /* + * The subslice disable field is global, i.e. it applies + * to each of the enabled slices. + */ + subslice_mask = GENMASK(sseu->max_subslices - 1, 0); + subslice_mask &= ~((fuse2 & GEN8_F2_SS_DIS_MASK) >> + GEN8_F2_SS_DIS_SHIFT); + eu_disable0 = intel_uncore_read(uncore, GEN8_EU_DISABLE0); + eu_disable1 = intel_uncore_read(uncore, GEN8_EU_DISABLE1); + eu_disable2 = intel_uncore_read(uncore, GEN8_EU_DISABLE2); + eu_disable[0] = eu_disable0 & GEN8_EU_DIS0_S0_MASK; + eu_disable[1] = (eu_disable0 >> GEN8_EU_DIS0_S1_SHIFT) | + ((eu_disable1 & GEN8_EU_DIS1_S1_MASK) << + (32 - GEN8_EU_DIS0_S1_SHIFT)); + eu_disable[2] = (eu_disable1 >> GEN8_EU_DIS1_S2_SHIFT) | + ((eu_disable2 & GEN8_EU_DIS2_S2_MASK) << + (32 - GEN8_EU_DIS1_S2_SHIFT)); + + /* + * Iterate through enabled slices and subslices to + * count the total enabled EU. + */ + for (s = 0; s < sseu->max_slices; s++) { + if (!(sseu->slice_mask & BIT(s))) + /* skip disabled slice */ + continue; + + intel_sseu_set_subslices(sseu, s, subslice_mask); + + for (ss = 0; ss < sseu->max_subslices; ss++) { + u8 eu_disabled_mask; + u32 n_disabled; + + if (!intel_sseu_has_subslice(sseu, s, ss)) + /* skip disabled subslice */ + continue; + + eu_disabled_mask = + eu_disable[s] >> (ss * sseu->max_eus_per_subslice); + + sseu_set_eus(sseu, s, ss, ~eu_disabled_mask); + + n_disabled = hweight8(eu_disabled_mask); + + /* + * Record which subslices have 7 EUs. + */ + if (sseu->max_eus_per_subslice - n_disabled == 7) + sseu->subslice_7eu[s] |= 1 << ss; + } + } + + sseu->eu_total = compute_eu_total(sseu); + + /* + * BDW 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. + */ + sseu->eu_per_subslice = + intel_sseu_subslice_total(sseu) ? + DIV_ROUND_UP(sseu->eu_total, intel_sseu_subslice_total(sseu)) : + 0; + + /* + * BDW supports slice power gating on devices with more than + * one slice. + */ + sseu->has_slice_pg = hweight8(sseu->slice_mask) > 1; + sseu->has_subslice_pg = 0; + sseu->has_eu_pg = 0; +} + +static void hsw_sseu_info_init(struct intel_gt *gt) +{ + struct drm_i915_private *i915 = gt->i915; + struct sseu_dev_info *sseu = >->info.sseu; + u32 fuse1; + u8 subslice_mask = 0; + int s, ss; + + /* + * There isn't a register to tell us how many slices/subslices. We + * work off the PCI-ids here. + */ + switch (INTEL_INFO(i915)->gt) { + default: + MISSING_CASE(INTEL_INFO(i915)->gt); + fallthrough; + case 1: + sseu->slice_mask = BIT(0); + subslice_mask = BIT(0); + break; + case 2: + sseu->slice_mask = BIT(0); + subslice_mask = BIT(0) | BIT(1); + break; + case 3: + sseu->slice_mask = BIT(0) | BIT(1); + subslice_mask = BIT(0) | BIT(1); + break; + } + + fuse1 = intel_uncore_read(gt->uncore, HSW_PAVP_FUSE1); + switch ((fuse1 & HSW_F1_EU_DIS_MASK) >> HSW_F1_EU_DIS_SHIFT) { + default: + MISSING_CASE((fuse1 & HSW_F1_EU_DIS_MASK) >> + HSW_F1_EU_DIS_SHIFT); + fallthrough; + case HSW_F1_EU_DIS_10EUS: + sseu->eu_per_subslice = 10; + break; + case HSW_F1_EU_DIS_8EUS: + sseu->eu_per_subslice = 8; + break; + case HSW_F1_EU_DIS_6EUS: + sseu->eu_per_subslice = 6; + break; + } + + intel_sseu_set_info(sseu, hweight8(sseu->slice_mask), + hweight8(subslice_mask), + sseu->eu_per_subslice); + + for (s = 0; s < sseu->max_slices; s++) { + intel_sseu_set_subslices(sseu, s, subslice_mask); + + for (ss = 0; ss < sseu->max_subslices; ss++) { + sseu_set_eus(sseu, s, ss, + (1UL << sseu->eu_per_subslice) - 1); + } + } + + sseu->eu_total = compute_eu_total(sseu); + + /* No powergating for you. */ + sseu->has_slice_pg = 0; + sseu->has_subslice_pg = 0; + sseu->has_eu_pg = 0; +} + +void intel_sseu_info_init(struct intel_gt *gt) +{ + struct drm_i915_private *i915 = gt->i915; + + if (IS_HASWELL(i915)) + hsw_sseu_info_init(gt); + else if (IS_CHERRYVIEW(i915)) + cherryview_sseu_info_init(gt); + else if (IS_BROADWELL(i915)) + bdw_sseu_info_init(gt); + else if (IS_GEN(i915, 9)) + gen9_sseu_info_init(gt); + else if (IS_GEN(i915, 10)) + gen10_sseu_info_init(gt); + else if (IS_GEN(i915, 11)) + gen11_sseu_info_init(gt); + else if (INTEL_GEN(i915) >= 12) + gen12_sseu_info_init(gt); +} + +u32 intel_sseu_make_rpcs(struct intel_gt *gt, const struct intel_sseu *req_sseu) { - const struct sseu_dev_info *sseu = &RUNTIME_INFO(i915)->sseu; + struct drm_i915_private *i915 = gt->i915; + const struct sseu_dev_info *sseu = >->info.sseu; bool subslice_pg = sseu->has_subslice_pg; u8 slices, subslices; u32 rpcs = 0; @@ -173,3 +715,48 @@ u32 intel_sseu_make_rpcs(struct drm_i915_private *i915, return rpcs; } + +void intel_sseu_dump(const struct sseu_dev_info *sseu, struct drm_printer *p) +{ + int s; + + drm_printf(p, "slice total: %u, mask=%04x\n", + hweight8(sseu->slice_mask), sseu->slice_mask); + drm_printf(p, "subslice total: %u\n", intel_sseu_subslice_total(sseu)); + for (s = 0; s < sseu->max_slices; s++) { + drm_printf(p, "slice%d: %u subslices, mask=%08x\n", + s, intel_sseu_subslices_per_slice(sseu, s), + intel_sseu_get_subslices(sseu, s)); + } + drm_printf(p, "EU total: %u\n", sseu->eu_total); + drm_printf(p, "EU per subslice: %u\n", sseu->eu_per_subslice); + drm_printf(p, "has slice power gating: %s\n", + yesno(sseu->has_slice_pg)); + drm_printf(p, "has subslice power gating: %s\n", + yesno(sseu->has_subslice_pg)); + drm_printf(p, "has EU power gating: %s\n", yesno(sseu->has_eu_pg)); +} + +void intel_sseu_print_topology(const struct sseu_dev_info *sseu, + struct drm_printer *p) +{ + int s, ss; + + if (sseu->max_slices == 0) { + drm_printf(p, "Unavailable\n"); + return; + } + + for (s = 0; s < sseu->max_slices; s++) { + drm_printf(p, "slice%d: %u subslice(s) (0x%08x):\n", + s, intel_sseu_subslices_per_slice(sseu, s), + intel_sseu_get_subslices(sseu, s)); + + for (ss = 0; ss < sseu->max_subslices; ss++) { + u16 enabled_eus = sseu_get_eus(sseu, s, ss); + + drm_printf(p, "\tsubslice%d: %u EUs (0x%hx)\n", + ss, hweight16(enabled_eus), enabled_eus); + } + } +} |