Skip to contentSkip to conversation

drm/i915/mtl: Add OAG 32 bit format support for MTL

PATCH Superseded Umesh Nerlige Ramappa 1 participant 5 messages Dec 1, 2022

Subsystems (2)

v2 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

v2 cover drm/i915/mtl: Add OAG 32 bit format support for MTL

Enable OA for MTL by adding 32-bit OA format support and relevant fixes.

Umesh Nerlige Ramappa (4):
drm/i915/mtl: Resize noa_wait BO size to save restore GPR regs drm/i915/mtl: Add Wa_14015846243 to fix OA vs CS timestamp mismatch drm/i915/mtl: Update OA mux whitelist for MTL drm/i915/mtl: Add OA support by enabling 32 bit OAG formats for MTL

 drivers/gpu/drm/i915/gt/intel_gt_types.h |  6 ---
 drivers/gpu/drm/i915/i915_perf.c         | 49 ++++++++++++++++++------
 2 files changed, 38 insertions(+), 17 deletions(-)
Signature · 1 line
2.36.1
#1 lore raw

PATCH v2 1/4 drm/i915/mtl: Resize noa_wait BO size to save restore GPR regs

On MTL, gt->scratch was using stolen lmem. An MI_SRM to stolen lmem caused a hang that was attributed to saving and restoring the GPR registers used for noa_wait.

Add an additional page in noa_wait BO to save/restore GPR registers for the noa_wait logic.

---
 drivers/gpu/drm/i915/gt/intel_gt_types.h |  6 ------
 drivers/gpu/drm/i915/i915_perf.c         | 25 ++++++++++++++++--------
 2 files changed, 17 insertions(+), 14 deletions(-)
drivers/gpu/drm/i915/gt/intel_gt_types.h +0 -6
index c1d9cd255e06..13dffe0a3d20 100644
@@ -296,12 +296,6 @@ enum intel_gt_scratch_field {
 
 	/* 8 bytes */
 	INTEL_GT_SCRATCH_FIELD_COHERENTL3_WA = 256,
-
-	/* 6 * 8 bytes */
-	INTEL_GT_SCRATCH_FIELD_PERF_CS_GPR = 2048,
-
-	/* 4 bytes */
-	INTEL_GT_SCRATCH_FIELD_PERF_PREDICATE_RESULT_1 = 2096,
 };
 
 #endif /* __INTEL_GT_TYPES_H__ */
drivers/gpu/drm/i915/i915_perf.c +17 -8
index 00e09bb18b13..7790a88f10d8 100644
@@ -1842,8 +1842,7 @@ static u32 *save_restore_register(struct i915_perf_stream *stream, u32 *cs,
 	for (d = 0; d < dword_count; d++) {
 		*cs++ = cmd;
 		*cs++ = i915_mmio_reg_offset(reg) + 4 * d;
-		*cs++ = intel_gt_scratch_offset(stream->engine->gt,
-						offset) + 4 * d;
+		*cs++ = i915_ggtt_offset(stream->noa_wait) + offset + 4 * d;
 		*cs++ = 0;
 	}
 
@@ -1876,7 +1875,13 @@ static int alloc_noa_wait(struct i915_perf_stream *stream)
 					  MI_PREDICATE_RESULT_2_ENGINE(base) :
 					  MI_PREDICATE_RESULT_1(RENDER_RING_BASE);
 
-	bo = i915_gem_object_create_internal(i915, 4096);
+	/*
+	 * gt->scratch was being used to save/restore the GPR registers, but on
+	 * MTL the scratch uses stolen lmem. An MI_SRM to this memory region
+	 * causes an engine hang. Instead allocate an additional page here to
+	 * save/restore GPR registers
+	 */
+	bo = i915_gem_object_create_internal(i915, 8192);
 	if (IS_ERR(bo)) {
 		drm_err(&i915->drm,
 			"Failed to allocate NOA wait batchbuffer\n");
@@ -1910,14 +1915,19 @@ static int alloc_noa_wait(struct i915_perf_stream *stream)
 		goto err_unpin;
 	}
 
+	stream->noa_wait = vma;
+
+#define GPR_SAVE_OFFSET 4096
+#define PREDICATE_SAVE_OFFSET 4160
+
 	/* Save registers. */
 	for (i = 0; i < N_CS_GPR; i++)
 		cs = save_restore_register(
 			stream, cs, true /* save */, CS_GPR(i),
-			INTEL_GT_SCRATCH_FIELD_PERF_CS_GPR + 8 * i, 2);
+			GPR_SAVE_OFFSET + 8 * i, 2);
 	cs = save_restore_register(
 		stream, cs, true /* save */, mi_predicate_result,
-		INTEL_GT_SCRATCH_FIELD_PERF_PREDICATE_RESULT_1, 1);
+		PREDICATE_SAVE_OFFSET, 1);
 
 	/* First timestamp snapshot location. */
 	ts0 = cs;
@@ -2033,10 +2043,10 @@ static int alloc_noa_wait(struct i915_perf_stream *stream)
 	for (i = 0; i < N_CS_GPR; i++)
 		cs = save_restore_register(
 			stream, cs, false /* restore */, CS_GPR(i),
-			INTEL_GT_SCRATCH_FIELD_PERF_CS_GPR + 8 * i, 2);
+			GPR_SAVE_OFFSET + 8 * i, 2);
 	cs = save_restore_register(
 		stream, cs, false /* restore */, mi_predicate_result,
-		INTEL_GT_SCRATCH_FIELD_PERF_PREDICATE_RESULT_1, 1);
+		PREDICATE_SAVE_OFFSET, 1);
 
 	/* And return to the ring. */
 	*cs++ = MI_BATCH_BUFFER_END;
@@ -2046,7 +2056,6 @@ static int alloc_noa_wait(struct i915_perf_stream *stream)
 	i915_gem_object_flush_map(bo);
 	__i915_gem_object_release_map(bo);
 
-	stream->noa_wait = vma;
 	goto out_ww;
 
 err_unpin:
Signature · 1 line
2.36.1
#2 lore raw

PATCH v2 2/4 drm/i915/mtl: Add Wa_14015846243 to fix OA vs CS timestamp mismatch

In reply to Umesh Nerlige Ramappa

Similar to ACM, OA timestamp that is part of the OA report is shifted when compared to the CS timestamp. Add MTL to the WA.

---
 drivers/gpu/drm/i915/i915_perf.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
drivers/gpu/drm/i915/i915_perf.c +5 -2
index 7790a88f10d8..8ed9af571de9 100644
@@ -3136,8 +3136,11 @@ get_sseu_config(struct intel_sseu *out_sseu,
  */
 u32 i915_perf_oa_timestamp_frequency(struct drm_i915_private *i915)
 {
-	/* Wa_18013179988:dg2 */
-	if (IS_DG2(i915)) {
+	/*
+	 * Wa_18013179988:dg2
+	 * Wa_14015846243:mtl
+	 */
+	if (IS_DG2(i915) || IS_METEORLAKE(i915)) {
 		intel_wakeref_t wakeref;
 		u32 reg, shift;
 
Signature · 1 line
2.36.1
#3 lore raw

PATCH v2 3/4 drm/i915/mtl: Update OA mux whitelist for MTL

In reply to Umesh Nerlige Ramappa

0x20cc (WAIT_FOR_RC6_EXIT on other platforms) is repurposed on MTL. Use a separate mux table to verify oa configs passed by user.

---
 drivers/gpu/drm/i915/i915_perf.c | 16 +++++++++++++++-
 1 file changed, 15 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/i915_perf.c +15 -1
index 8ed9af571de9..8369ae4b850d 100644
@@ -4318,6 +4318,17 @@ static const struct i915_range gen12_oa_mux_regs[] = {
 	{}
 };
 
+/*
+ * Ref: 14010536224:
+ * 0x20cc is repurposed on MTL, so use a separate array for MTL.
+ */
+static const struct i915_range mtl_oa_mux_regs[] = {
+	{ .start = 0x0d00, .end = 0x0d04 },	/* RPM_CONFIG[0-1] */
+	{ .start = 0x0d0c, .end = 0x0d2c },	/* NOA_CONFIG[0-8] */
+	{ .start = 0x9840, .end = 0x9840 },	/* GDT_CHICKEN_BITS */
+	{ .start = 0x9884, .end = 0x9888 },	/* NOA_WRITE */
+};
+
 static bool gen7_is_valid_b_counter_addr(struct i915_perf *perf, u32 addr)
 {
 	return reg_in_range_table(addr, gen7_oa_b_counters);
@@ -4361,7 +4372,10 @@ static bool xehp_is_valid_b_counter_addr(struct i915_perf *perf, u32 addr)
 
 static bool gen12_is_valid_mux_addr(struct i915_perf *perf, u32 addr)
 {
-	return reg_in_range_table(addr, gen12_oa_mux_regs);
+	if (IS_METEORLAKE(perf->i915))
+		return reg_in_range_table(addr, mtl_oa_mux_regs);
+	else
+		return reg_in_range_table(addr, gen12_oa_mux_regs);
 }
 
 static u32 mask_reg_value(u32 reg, u32 val)
Signature · 1 line
2.36.1
#4 lore raw

PATCH v2 4/4 drm/i915/mtl: Add OA support by enabling 32 bit OAG formats for MTL

In reply to Umesh Nerlige Ramappa

Without an entry in oa_init_supported_formats, OA will not be functional in MTL. Enable OA support by enabling 32 bit OAG formats for MTL.

---
 drivers/gpu/drm/i915/i915_perf.c | 1 +
 1 file changed, 1 insertion(+)
drivers/gpu/drm/i915/i915_perf.c +1 -0
index 8369ae4b850d..a735b9540113 100644
@@ -4772,6 +4772,7 @@ static void oa_init_supported_formats(struct i915_perf *perf)
 		break;
 
 	case INTEL_DG2:
+	case INTEL_METEORLAKE:
 		oa_format_add(perf, I915_OAR_FORMAT_A32u40_A4u32_B8_C8);
 		oa_format_add(perf, I915_OA_FORMAT_A24u40_A14u32_B8_C8);
 		break;
Signature · 1 line
2.36.1