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.
Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221212220902.1819159-2-umesh.nerlige.ramappa@intel.com

drivers/gpu/drm/i915/gt/intel_gt_types.h

@@ -304,12 +304,6 @@ 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

@@ -1845,8 +1845,7 @@ 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;
 	}
 
@@ -1879,7 +1878,13 @@ 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");
@@ -1913,14 +1918,19 @@ 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;
@@ -2036,10 +2046,10 @@ 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;
@@ -2049,7 +2059,6 @@ 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: