[PATCH 7/9] drm/i915: expose eu topology to userspace

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



With the introduction of asymetric slices in CNL, we cannot rely on the
previous SUBSLICE_MASK getparam. Here we introduce a more detailed way of
querying the Gen's GPU topology that doesn't aggregate numbers.

This is essential for monitoring parts of the GPU with the OA unit, because
counters need to be normalized to the number of EUs/subslices/slices. The
current aggregated numbers like EU_TOTAL do not gives us sufficient
information.

As a bonus we can draw representations of the GPU :

    https://imgur.com/a/vuqpa

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@xxxxxxxxx>
---
 drivers/gpu/drm/i915/i915_debugfs.c      |  24 +++--
 drivers/gpu/drm/i915/i915_drv.c          |  51 ++++++++-
 drivers/gpu/drm/i915/i915_drv.h          |  23 ++++-
 drivers/gpu/drm/i915/intel_device_info.c | 171 ++++++++++++++++++++++---------
 drivers/gpu/drm/i915/intel_lrc.c         |   2 +-
 drivers/gpu/drm/i915/intel_ringbuffer.h  |   2 +-
 include/uapi/drm/i915_drm.h              |  59 +++++++++++
 7 files changed, 270 insertions(+), 62 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
index 5890a49e64c7..c5bb943f1c16 100644
--- a/drivers/gpu/drm/i915/i915_debugfs.c
+++ b/drivers/gpu/drm/i915/i915_debugfs.c
@@ -4441,7 +4441,7 @@ static void cherryview_sseu_device_status(struct drm_i915_private *dev_priv,
 			continue;
 
 		sseu->slice_mask = BIT(0);
-		sseu->subslice_mask |= BIT(ss);
+		sseu->subslices_mask[0] |= BIT(ss);
 		eu_cnt = ((sig1[ss] & CHV_EU08_PG_ENABLE) ? 0 : 2) +
 			 ((sig1[ss] & CHV_EU19_PG_ENABLE) ? 0 : 2) +
 			 ((sig1[ss] & CHV_EU210_PG_ENABLE) ? 0 : 2) +
@@ -4488,7 +4488,7 @@ static void gen10_sseu_device_status(struct drm_i915_private *dev_priv,
 			continue;
 
 		sseu->slice_mask |= BIT(s);
-		sseu->subslice_mask = info->sseu.subslice_mask;
+		sseu->subslices_mask[s] = info->sseu.subslices_mask[s];
 
 		for (ss = 0; ss < ss_max; ss++) {
 			unsigned int eu_cnt;
@@ -4543,8 +4543,8 @@ static void gen9_sseu_device_status(struct drm_i915_private *dev_priv,
 		sseu->slice_mask |= BIT(s);
 
 		if (IS_GEN9_BC(dev_priv))
-			sseu->subslice_mask =
-				INTEL_INFO(dev_priv)->sseu.subslice_mask;
+			sseu->subslices_mask[s] =
+				INTEL_INFO(dev_priv)->sseu.subslices_mask[s];
 
 		for (ss = 0; ss < ss_max; ss++) {
 			unsigned int eu_cnt;
@@ -4554,7 +4554,7 @@ static void gen9_sseu_device_status(struct drm_i915_private *dev_priv,
 					/* skip disabled subslice */
 					continue;
 
-				sseu->subslice_mask |= BIT(ss);
+				sseu->subslices_mask[s] |= BIT(ss);
 			}
 
 			eu_cnt = 2 * hweight32(eu_reg[2*s + ss/2] &
@@ -4576,9 +4576,12 @@ static void broadwell_sseu_device_status(struct drm_i915_private *dev_priv,
 	sseu->slice_mask = slice_info & GEN8_LSLICESTAT_MASK;
 
 	if (sseu->slice_mask) {
-		sseu->subslice_mask = INTEL_INFO(dev_priv)->sseu.subslice_mask;
 		sseu->eu_per_subslice =
 				INTEL_INFO(dev_priv)->sseu.eu_per_subslice;
+		for (s = 0; s < fls(sseu->slice_mask); s++) {
+			sseu->subslices_mask[s] =
+				INTEL_INFO(dev_priv)->sseu.subslices_mask[s];
+		}
 		sseu->eu_total = sseu->eu_per_subslice *
 				 sseu_subslice_total(sseu);
 
@@ -4597,6 +4600,7 @@ static void i915_print_sseu_info(struct seq_file *m, bool is_available_info,
 {
 	struct drm_i915_private *dev_priv = node_to_i915(m->private);
 	const char *type = is_available_info ? "Available" : "Enabled";
+	int s;
 
 	seq_printf(m, "  %s Slice Mask: %04x\n", type,
 		   sseu->slice_mask);
@@ -4604,10 +4608,10 @@ static void i915_print_sseu_info(struct seq_file *m, bool is_available_info,
 		   hweight8(sseu->slice_mask));
 	seq_printf(m, "  %s Subslice Total: %u\n", type,
 		   sseu_subslice_total(sseu));
-	seq_printf(m, "  %s Subslice Mask: %04x\n", type,
-		   sseu->subslice_mask);
-	seq_printf(m, "  %s Subslice Per Slice: %u\n", type,
-		   hweight8(sseu->subslice_mask));
+	for (s = 0; s < fls(sseu->slice_mask); s++) {
+		seq_printf(m, "  %s Slice%i Subslice Mask: %04x\n", type,
+			   s, sseu->subslices_mask[s]);
+	}
 	seq_printf(m, "  %s EU Total: %u\n", type,
 		   sseu->eu_total);
 	seq_printf(m, "  %s EU Per Subslice: %u\n", type,
diff --git a/drivers/gpu/drm/i915/i915_drv.c b/drivers/gpu/drm/i915/i915_drv.c
index 32ad49c81b79..dfa6af28289f 100644
--- a/drivers/gpu/drm/i915/i915_drv.c
+++ b/drivers/gpu/drm/i915/i915_drv.c
@@ -412,13 +412,62 @@ static int i915_getparam(struct drm_device *dev, void *data,
 			return -ENODEV;
 		break;
 	case I915_PARAM_SUBSLICE_MASK:
-		value = INTEL_INFO(dev_priv)->sseu.subslice_mask;
+		value = INTEL_INFO(dev_priv)->sseu.subslices_mask[0];
 		if (!value)
 			return -ENODEV;
 		break;
 	case I915_PARAM_CS_TIMESTAMP_FREQUENCY:
 		value = INTEL_INFO(dev_priv)->cs_timestamp_frequency;
 		break;
+	case I915_PARAM_EU_TOPOLOGY: {
+		const struct sseu_dev_info *sseu = &INTEL_INFO(dev_priv)->sseu;
+		drm_i915_eu_topology_t __user *user_topology =
+			(drm_i915_eu_topology_t __user *) param->value;
+		drm_i915_eu_topology_t topology;
+		int ret, subslice_size, eu_size;
+
+		if (sseu->max_slices == 0)
+			return -ENODEV;
+
+		ret = copy_from_user(&topology, user_topology, sizeof(topology));
+		if (ret)
+			return ret;
+
+		if (topology.n_slices == 0 &&
+		    topology.n_subslices_per_slice == 0 &&
+		    topology.n_eus_per_subslice == 0) {
+			topology.n_slices = sseu->max_slices;
+			topology.n_subslices_per_slice = sseu->max_subslices;
+			topology.n_eus_per_subslice = sseu->max_eus_per_subslice;
+
+			return copy_to_user(user_topology, &topology,
+					    sizeof(topology));
+		}
+
+		if (topology.n_slices != sseu->max_slices ||
+		    topology.n_subslices_per_slice != sseu->max_subslices ||
+		    topology.n_eus_per_subslice != sseu->max_eus_per_subslice)
+			return -EINVAL;
+
+		ret = copy_to_user(user_topology->data, &sseu->slice_mask,
+				   sizeof(sseu->slice_mask));
+		if (ret)
+			return ret;
+
+		subslice_size =
+			sseu->max_slices * (ALIGN(sseu->max_subslices, 8) / 8);
+		ret = copy_to_user(user_topology->data + sizeof(sseu->slice_mask),
+				   sseu->subslices_mask,
+				   subslice_size);
+		if (ret)
+			return ret;
+
+		eu_size = sseu->max_slices * sseu->max_subslices *
+			(ALIGN(sseu->max_eus_per_subslice, 8) / 8);
+		return copy_to_user((user_topology->data +
+				     sizeof(sseu->slice_mask) + subslice_size),
+				    sseu->eu_mask, eu_size);
+	}
 	default:
 		DRM_DEBUG("Unknown parameter %d\n", param->param);
 		return -EINVAL;
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index fd1a6dc0c70b..457901024ed0 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -799,9 +799,12 @@ struct intel_csr {
 	func(supports_tv); \
 	func(has_ipc);
 
+#define GEN_MAX_SLICES		(6) /* CNL upper bound */
+#define GEN_MAX_SUBSLICES	(7)
+
 struct sseu_dev_info {
 	u8 slice_mask;
-	u8 subslice_mask;
+	u8 subslices_mask[GEN_MAX_SLICES];
 	u8 eu_total;
 	u8 eu_per_subslice;
 	u8 min_eu_in_pool;
@@ -810,11 +813,27 @@ struct sseu_dev_info {
 	u8 has_slice_pg:1;
 	u8 has_subslice_pg:1;
 	u8 has_eu_pg:1;
+
+	/* Topology fields */
+	u8 max_slices;
+	u8 max_subslices;
+	u8 max_eus_per_subslice;
+
+	/* We don't have more than 8 eus per subslice at the moment and as we
+	 * store eus enabled using bits, no need to multiply by eus per
+	 * subslice.
+	 */
+	u8 eu_mask[GEN_MAX_SLICES * GEN_MAX_SUBSLICES];
 };
 
 static inline unsigned int sseu_subslice_total(const struct sseu_dev_info *sseu)
 {
-	return hweight8(sseu->slice_mask) * hweight8(sseu->subslice_mask);
+	unsigned s, total = 0;
+
+	for (s = 0; s < ARRAY_SIZE(sseu->subslices_mask); s++)
+		total += hweight8(sseu->subslices_mask[s]);
+
+	return total;
 }
 
 /* Keep in gen based order, and chronological order within a gen */
diff --git a/drivers/gpu/drm/i915/intel_device_info.c b/drivers/gpu/drm/i915/intel_device_info.c
index c6b0e4ec84b4..7e395e7e2892 100644
--- a/drivers/gpu/drm/i915/intel_device_info.c
+++ b/drivers/gpu/drm/i915/intel_device_info.c
@@ -82,22 +82,74 @@ void intel_device_info_dump(struct drm_i915_private *dev_priv)
 #undef PRINT_FLAG
 }
 
+static u8 compute_eu_total(const struct sseu_dev_info *sseu)
+{
+	u8 i, total = 0;
+
+	for (i = 0; i < ARRAY_SIZE(sseu->eu_mask); i++)
+		total += hweight8(sseu->eu_mask[i]);
+
+	return total;
+}
+
 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);
+	int s, ss, eu_mask = 0xff;
+	u32 subslice_mask, eu_en;
 
 	sseu->slice_mask = (fuse2 & GEN10_F2_S_ENA_MASK) >>
 			    GEN10_F2_S_ENA_SHIFT;
-	sseu->subslice_mask = (1 << 4) - 1;
-	sseu->subslice_mask &= ~((fuse2 & GEN10_F2_SS_DIS_MASK) >>
-				 GEN10_F2_SS_DIS_SHIFT);
+	sseu->max_slices = 6;
+	sseu->max_subslices = 4;
+	sseu->max_eus_per_subslice = 8;
+
+	subslice_mask = (1 << 4) - 1;
+	subslice_mask &= ~((fuse2 & GEN10_F2_SS_DIS_MASK) >>
+			   GEN10_F2_SS_DIS_SHIFT);
+
+	/* Slice0 can have up to 3 subslices, but there are only 2 in
+	 * slice1/2.
+	 */
+	sseu->subslices_mask[0] = subslice_mask;
+	for (s = 1; s < sseu->max_slices; s++)
+		sseu->subslices_mask[s] = subslice_mask & 0x3;
+
+	/* Slice0 */
+	eu_en = ~I915_READ(GEN8_EU_DISABLE0);
+	for (ss = 0; ss < sseu->max_subslices; ss++)
+		sseu->eu_mask[ss]     = (eu_en >> (8 * ss)) & eu_mask;
+	/* Slice1 */
+	sseu->eu_mask[sseu->max_subslices]         = (eu_en >> 24) & eu_mask;
+	eu_en = ~I915_READ(GEN8_EU_DISABLE1);
+	sseu->eu_mask[sseu->max_subslices + 1]     = eu_en & eu_mask;
+	/* Slice2 */
+	sseu->eu_mask[2 * sseu->max_subslices]     = (eu_en >> 8) & eu_mask;
+	sseu->eu_mask[2 * sseu->max_subslices + 1] = (eu_en >> 16) & eu_mask;
+	/* Slice3 */
+	sseu->eu_mask[3 * sseu->max_subslices]     = (eu_en >> 24) & eu_mask;
+	eu_en = ~I915_READ(GEN8_EU_DISABLE2);
+	sseu->eu_mask[3 * sseu->max_subslices + 1] = eu_en & eu_mask;
+	/* Slice4 */
+	sseu->eu_mask[4 * sseu->max_subslices]     = (eu_en >> 8) & eu_mask;
+	sseu->eu_mask[4 * sseu->max_subslices + 1] = (eu_en >> 16) & eu_mask;
+	/* Slice5 */
+	sseu->eu_mask[5 * sseu->max_subslices]     = (eu_en >> 24) & eu_mask;
+	eu_en = ~I915_READ(GEN10_EU_DISABLE3);
+	sseu->eu_mask[5 * sseu->max_subslices + 1] = eu_en & eu_mask;
+
+	/* Do a second pass where we marked the subslices disabled if all
+	 * their eus are off.
+	 */
+	for (s = 0; s < sseu->max_slices; s++) {
+		for (ss = 0; ss < sseu->max_subslices; ss++) {
+			if (sseu->eu_mask[s * sseu->max_subslices + ss] == 0)
+				sseu->subslices_mask[s] &= ~BIT(ss);
+		}
+	}
 
-	sseu->eu_total = hweight32(~I915_READ(GEN8_EU_DISABLE0));
-	sseu->eu_total += hweight32(~I915_READ(GEN8_EU_DISABLE1));
-	sseu->eu_total += hweight32(~I915_READ(GEN8_EU_DISABLE2));
-	sseu->eu_total += hweight8(~(I915_READ(GEN10_EU_DISABLE3) &
-				     GEN10_EU_DIS_SS_MASK));
+	sseu->eu_total = compute_eu_total(sseu);
 
 	/*
 	 * CNL is expected to always have a uniform distribution
@@ -118,26 +170,30 @@ static void gen10_sseu_info_init(struct drm_i915_private *dev_priv)
 static void cherryview_sseu_info_init(struct drm_i915_private *dev_priv)
 {
 	struct sseu_dev_info *sseu = &mkwrite_device_info(dev_priv)->sseu;
-	u32 fuse, eu_dis;
+	u32 fuse;
 
 	fuse = I915_READ(CHV_FUSE_GT);
 
 	sseu->slice_mask = BIT(0);
+	sseu->max_slices = 1;
+	sseu->max_subslices = 2;
+	sseu->max_eus_per_subslice = 8;
 
 	if (!(fuse & CHV_FGT_DISABLE_SS0)) {
-		sseu->subslice_mask |= BIT(0);
-		eu_dis = fuse & (CHV_FGT_EU_DIS_SS0_R0_MASK |
-				 CHV_FGT_EU_DIS_SS0_R1_MASK);
-		sseu->eu_total += 8 - hweight32(eu_dis);
+		sseu->subslices_mask[0] |= BIT(0);
+		sseu->eu_mask[0] = (fuse & CHV_FGT_EU_DIS_SS0_R0_MASK) >> CHV_FGT_EU_DIS_SS0_R0_SHIFT;
+		sseu->eu_mask[0] |= ((fuse & CHV_FGT_EU_DIS_SS0_R1_MASK) >> CHV_FGT_EU_DIS_SS0_R1_SHIFT) << 4;
+		sseu->subslices_mask[0] = 1;
 	}
 
 	if (!(fuse & CHV_FGT_DISABLE_SS1)) {
-		sseu->subslice_mask |= BIT(1);
-		eu_dis = fuse & (CHV_FGT_EU_DIS_SS1_R0_MASK |
-				 CHV_FGT_EU_DIS_SS1_R1_MASK);
-		sseu->eu_total += 8 - hweight32(eu_dis);
+		sseu->subslices_mask[0] |= BIT(1);
+		sseu->eu_mask[1] = (fuse & CHV_FGT_EU_DIS_SS1_R0_MASK) >> CHV_FGT_EU_DIS_SS0_R0_SHIFT;
+		sseu->eu_mask[2] |= ((fuse & CHV_FGT_EU_DIS_SS1_R1_MASK) >> CHV_FGT_EU_DIS_SS0_R1_SHIFT) << 4;
 	}
 
+	sseu->eu_total = compute_eu_total(sseu);
+
 	/*
 	 * CHV expected to always have a uniform distribution of EU
 	 * across subslices.
@@ -159,41 +215,50 @@ static void gen9_sseu_info_init(struct drm_i915_private *dev_priv)
 {
 	struct intel_device_info *info = mkwrite_device_info(dev_priv);
 	struct sseu_dev_info *sseu = &info->sseu;
-	int s_max = 3, ss_max = 4, eu_max = 8;
 	int s, ss;
-	u32 fuse2, eu_disable;
+	u32 fuse2, eu_disable, subslice_mask;
 	u8 eu_mask = 0xff;
 
 	fuse2 = I915_READ(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. */
+	sseu->max_slices = IS_GEN9_LP(dev_priv) ? 1 : 3;
+	sseu->max_subslices = IS_GEN9_LP(dev_priv) ? 3 : 4;
+	sseu->max_eus_per_subslice = 8;
+
 	/*
 	 * The subslice disable field is global, i.e. it applies
 	 * to each of the enabled slices.
 	*/
-	sseu->subslice_mask = (1 << ss_max) - 1;
-	sseu->subslice_mask &= ~((fuse2 & GEN9_F2_SS_DIS_MASK) >>
-				 GEN9_F2_SS_DIS_SHIFT);
+	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 < s_max; s++) {
+	for (s = 0; s < sseu->max_slices; s++) {
 		if (!(sseu->slice_mask & BIT(s)))
 			/* skip disabled slice */
 			continue;
 
+		sseu->subslices_mask[s] = subslice_mask;
+
 		eu_disable = I915_READ(GEN9_EU_DISABLE(s));
-		for (ss = 0; ss < ss_max; ss++) {
+		for (ss = 0; ss < sseu->max_subslices; ss++) {
 			int eu_per_ss;
 
-			if (!(sseu->subslice_mask & BIT(ss)))
+			if (!(sseu->subslices_mask[s] & BIT(ss)))
 				/* skip disabled subslice */
 				continue;
 
-			eu_per_ss = eu_max - hweight8((eu_disable >> (ss*8)) &
-						      eu_mask);
+			sseu->eu_mask[ss + s * sseu->max_subslices] =
+				~((eu_disable >> (ss*8)) & eu_mask);
+
+			eu_per_ss = sseu->max_eus_per_subslice -
+				hweight8((eu_disable >> (ss*8)) & eu_mask);
 
 			/*
 			 * Record which subslice(s) has(have) 7 EUs. we
@@ -202,11 +267,11 @@ static void gen9_sseu_info_init(struct drm_i915_private *dev_priv)
 			 */
 			if (eu_per_ss == 7)
 				sseu->subslice_7eu[s] |= BIT(ss);
-
-			sseu->eu_total += eu_per_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
@@ -232,8 +297,8 @@ static void gen9_sseu_info_init(struct drm_i915_private *dev_priv)
 	sseu->has_eu_pg = sseu->eu_per_subslice > 2;
 
 	if (IS_GEN9_LP(dev_priv)) {
-#define IS_SS_DISABLED(ss)	(!(sseu->subslice_mask & BIT(ss)))
-		info->has_pooled_eu = hweight8(sseu->subslice_mask) == 3;
+#define IS_SS_DISABLED(ss)	(!(sseu->subslices_mask[0] & BIT(ss)))
+		info->has_pooled_eu = hweight8(sseu->subslices_mask[0]) == 3;
 
 		/*
 		 * There is a HW issue in 2x6 fused down parts that requires
@@ -242,7 +307,7 @@ static void gen9_sseu_info_init(struct drm_i915_private *dev_priv)
 		 * doesn't affect if the device has all 3 subslices enabled.
 		 */
 		/* WaEnablePooledEuFor2x6:bxt */
-		info->has_pooled_eu |= (hweight8(sseu->subslice_mask) == 2 &&
+		info->has_pooled_eu |= (hweight8(sseu->subslices_mask[0]) == 2 &&
 					IS_BXT_REVID(dev_priv, 0, BXT_REVID_B_LAST));
 
 		sseu->min_eu_in_pool = 0;
@@ -261,19 +326,22 @@ static void gen9_sseu_info_init(struct drm_i915_private *dev_priv)
 static void broadwell_sseu_info_init(struct drm_i915_private *dev_priv)
 {
 	struct sseu_dev_info *sseu = &mkwrite_device_info(dev_priv)->sseu;
-	const int s_max = 3, ss_max = 3, eu_max = 8;
 	int s, ss;
-	u32 fuse2, eu_disable[3]; /* s_max */
+	u32 fuse2, subslice_mask, eu_disable[3]; /* s_max */
 
 	fuse2 = I915_READ(GEN8_FUSE2);
 	sseu->slice_mask = (fuse2 & GEN8_F2_S_ENA_MASK) >> GEN8_F2_S_ENA_SHIFT;
+	sseu->max_slices = 3;
+	sseu->max_subslices = 3;
+	sseu->max_eus_per_subslice = 8;
+
 	/*
 	 * The subslice disable field is global, i.e. it applies
 	 * to each of the enabled slices.
 	 */
-	sseu->subslice_mask = GENMASK(ss_max - 1, 0);
-	sseu->subslice_mask &= ~((fuse2 & GEN8_F2_SS_DIS_MASK) >>
-				 GEN8_F2_SS_DIS_SHIFT);
+	subslice_mask = GENMASK(sseu->max_subslices - 1, 0);
+	subslice_mask &= ~((fuse2 & GEN8_F2_SS_DIS_MASK) >>
+			   GEN8_F2_SS_DIS_SHIFT);
 
 	eu_disable[0] = I915_READ(GEN8_EU_DISABLE0) & GEN8_EU_DIS0_S0_MASK;
 	eu_disable[1] = (I915_READ(GEN8_EU_DISABLE0) >> GEN8_EU_DIS0_S1_SHIFT) |
@@ -287,30 +355,36 @@ static void broadwell_sseu_info_init(struct drm_i915_private *dev_priv)
 	 * Iterate through enabled slices and subslices to
 	 * count the total enabled EU.
 	 */
-	for (s = 0; s < s_max; s++) {
+	for (s = 0; s < sseu->max_slices; s++) {
 		if (!(sseu->slice_mask & BIT(s)))
 			/* skip disabled slice */
 			continue;
 
-		for (ss = 0; ss < ss_max; ss++) {
+		sseu->subslices_mask[s] = subslice_mask;
+
+		for (ss = 0; ss < sseu->max_subslices; ss++) {
 			u32 n_disabled;
 
-			if (!(sseu->subslice_mask & BIT(ss)))
+			if (!(sseu->subslices_mask[ss] & BIT(ss)))
 				/* skip disabled subslice */
 				continue;
 
-			n_disabled = hweight8(eu_disable[s] >> (ss * eu_max));
+			sseu->eu_mask[ss + s * sseu->max_subslices] =
+				~(eu_disable[s] >>
+				  (ss * sseu->max_eus_per_subslice));
+			n_disabled = hweight8(eu_disable[s] >>
+					      (ss * sseu->max_eus_per_subslice));
 
 			/*
 			 * Record which subslices have 7 EUs.
 			 */
-			if (eu_max - n_disabled == 7)
+			if (sseu->max_eus_per_subslice - n_disabled == 7)
 				sseu->subslice_7eu[s] |= 1 << ss;
-
-			sseu->eu_total += eu_max - n_disabled;
 		}
 	}
 
+	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
@@ -440,6 +514,7 @@ void intel_device_info_runtime_init(struct drm_i915_private *dev_priv)
 {
 	struct intel_device_info *info = mkwrite_device_info(dev_priv);
 	enum pipe pipe;
+	int s;
 
 	if (INTEL_GEN(dev_priv) >= 9) {
 		info->num_scalers[PIPE_A] = 2;
@@ -548,9 +623,11 @@ void intel_device_info_runtime_init(struct drm_i915_private *dev_priv)
 	DRM_DEBUG_DRIVER("slice total: %u\n", hweight8(info->sseu.slice_mask));
 	DRM_DEBUG_DRIVER("subslice total: %u\n",
 			 sseu_subslice_total(&info->sseu));
-	DRM_DEBUG_DRIVER("subslice mask %04x\n", info->sseu.subslice_mask);
-	DRM_DEBUG_DRIVER("subslice per slice: %u\n",
-			 hweight8(info->sseu.subslice_mask));
+	for (s = 0; s < ARRAY_SIZE(info->sseu.subslices_mask); s++) {
+		DRM_DEBUG_DRIVER("subslice mask %04x\n", info->sseu.subslices_mask[s]);
+		DRM_DEBUG_DRIVER("subslice per slice: %u\n",
+				 hweight8(info->sseu.subslices_mask[s]));
+	}
 	DRM_DEBUG_DRIVER("EU total: %u\n", info->sseu.eu_total);
 	DRM_DEBUG_DRIVER("EU per subslice: %u\n", info->sseu.eu_per_subslice);
 	DRM_DEBUG_DRIVER("has slice power gating: %s\n",
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index 6840ec8db037..98b65cb6199a 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -2037,7 +2037,7 @@ make_rpcs(struct drm_i915_private *dev_priv)
 
 	if (INTEL_INFO(dev_priv)->sseu.has_subslice_pg) {
 		rpcs |= GEN8_RPCS_SS_CNT_ENABLE;
-		rpcs |= hweight8(INTEL_INFO(dev_priv)->sseu.subslice_mask) <<
+		rpcs |= hweight8(INTEL_INFO(dev_priv)->sseu.subslices_mask[0]) <<
 			GEN8_RPCS_SS_CNT_SHIFT;
 		rpcs |= GEN8_RPCS_ENABLE;
 	}
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
index 69ad875fd011..f9f73dfc8eae 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.h
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
@@ -98,7 +98,7 @@ hangcheck_action_to_str(const enum intel_engine_hangcheck_action a)
 
 #define instdone_subslice_mask(dev_priv__) \
 	(INTEL_GEN(dev_priv__) == 7 ? \
-	 1 : INTEL_INFO(dev_priv__)->sseu.subslice_mask)
+	 1 : INTEL_INFO(dev_priv__)->sseu.subslices_mask[0])
 
 #define for_each_instdone_slice_subslice(dev_priv__, slice__, subslice__) \
 	for ((slice__) = 0, (subslice__) = 0; \
diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h
index c3ff0d4947af..bb66fea826a2 100644
--- a/include/uapi/drm/i915_drm.h
+++ b/include/uapi/drm/i915_drm.h
@@ -456,6 +456,65 @@ typedef struct drm_i915_irq_wait {
  */
 #define I915_PARAM_CS_TIMESTAMP_FREQUENCY   50
 
+/* List of active slices, subslices & execution units. Starting with CNL we
+ * can have asynmetric slices (for example 3 subslices in slice0 and 2
+ * subslices in slice1+). This means we cannot rely on PARAM_SUBSLICE_MASK
+ * anymore.
+ *
+ * When using this parameter, getparam value should point to a structure of
+ * type drm_i915_eus_enabled_t. Call this once with n_slices,
+ * n_subslices_per_slice, & n_eus_per_subslice set to 0 and the kernel will
+ * fill out only those fields. You can then allocate the
+ * drm_i915_eu_topology_t properly to have enough space to store the full
+ * topology.
+ */
+#define I915_PARAM_EU_TOPOLOGY           51
+typedef struct drm_i915_eu_topology {
+	__u32 n_slices;
+	__u32 n_subslices_per_slice;
+	__u32 n_eus_per_subslice;
+
+	/* We define the following :
+	 *
+	 *      slice_mask_size        = ALIGN(n_slices, 8) / 8
+	 *      subslices_mask_offset  = slice_mask_size
+	 *      sublisces_mask_stride  = ALIGN(n_subslices_per_slice, 8) / 8
+	 *      subslices_mask_size    = n_slices * sublisces_mask_stride
+	 *      eu_mask_offset         = slice_mask_size + subslices_mask_size
+	 *      eu_subslice_stride     = ALIGN(n_eus_per_subslice, 8) / 8
+	 *      eu_slice_stride        = n_subslices_per_slice * eu_subslice_stride
+	 *      eu_mask_size           = n_slices * eu_slice_stride
+	 *      data_size              = slice_mask_size + subslices_mask_size + eu_mask_size
+	 *
+	 * When called with n_slices, n_subslices_per_slice &
+	 * n_eus_per_subslice filled out with the kernel values, the length of
+	 * data should be data_size.
+	 *
+	 * The kernel will write the following into data :
+	 *
+	 *  data[0 .. slice_mask_size - 1] :
+	 *
+	 *      Where each bit represent slices available.
+	 *
+	 *  data[slice_mask_offset .. slice_mask_offset + subslices_mask_size - 1] :
+	 *
+	 *      Where each bit represent the subslices available for each
+	 *      slice. Formula to tell if sliceX-subsliceY is enabled :
+	 *
+	 *           ((data[slice_mask_offset + X * sublisces_mask_stride + Y / 8] >> (Y % 8)) & 1)
+	 *
+	 *  data[eu_mask_offset .. eu_mask_offset + eu_mask_size - 1] :
+	 *
+	 *      Where each bit reports the status of one execution unit. User
+	 *      should use eu_slice_stride & eu_subslice_stride to through the
+	 *      different subslices. Formula to tell if sliceX-subsliceY-euZ
+	 *      is enabled :
+	 *
+	 *           ((data[eu_mask_offset + X * eu_slice_stride + Y * eu_subslice_stride + Z / 8] >> (Z % 8)) & 1)
+	 */
+	__u8 data[];
+} drm_i915_eu_topology_t;
+
 typedef struct drm_i915_getparam {
 	__s32 param;
 	/*
-- 
2.15.0.rc2

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




[Index of Archives]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]
  Powered by Linux