Since we are trying to put all WA stuff together, do not forget about the BB WAs.
v2: s/intel_bb_workarounds_init/intel_engine_init_bb_workarounds (Chris)
v3: Rebased to before the WAs are stored
Signed-off-by: Oscar Mateo <oscar.mateo@xxxxxxxxx>
Cc: Mika Kuoppala <mika.kuoppala@xxxxxxxxxxxxxxx>
Reviewed-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> (v2)
---
drivers/gpu/drm/i915/intel_lrc.c | 253 +-----------------------------
drivers/gpu/drm/i915/intel_workarounds.c | 254 +++++++++++++++++++++++++++++++
drivers/gpu/drm/i915/intel_workarounds.h | 3 +
3 files changed, 259 insertions(+), 251 deletions(-)
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index f0b4d2f..7f731d0 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -1181,255 +1181,6 @@ static int execlists_request_alloc(struct drm_i915_gem_request *request)
return 0;
}
-/*
- * In this WA we need to set GEN8_L3SQCREG4[21:21] and reset it after
- * PIPE_CONTROL instruction. This is required for the flush to happen correctly
- * but there is a slight complication as this is applied in WA batch where the
- * values are only initialized once so we cannot take register value at the
- * beginning and reuse it further; hence we save its value to memory, upload a
- * constant value with bit21 set and then we restore it back with the saved value.
- * To simplify the WA, a constant value is formed by using the default value
- * of this register. This shouldn't be a problem because we are only modifying
- * it for a short period and this batch in non-premptible. We can ofcourse
- * use additional instructions that read the actual value of the register
- * at that time and set our bit of interest but it makes the WA complicated.
- *
- * This WA is also required for Gen9 so extracting as a function avoids
- * code duplication.
- */
-static u32 *
-gen8_emit_flush_coherentl3_wa(struct intel_engine_cs *engine, u32 *batch)
-{
- *batch++ = MI_STORE_REGISTER_MEM_GEN8 | MI_SRM_LRM_GLOBAL_GTT;
- *batch++ = i915_mmio_reg_offset(GEN8_L3SQCREG4);
- *batch++ = i915_ggtt_offset(engine->scratch) + 256;
- *batch++ = 0;
-
- *batch++ = MI_LOAD_REGISTER_IMM(1);
- *batch++ = i915_mmio_reg_offset(GEN8_L3SQCREG4);
- *batch++ = 0x40400000 | GEN8_LQSC_FLUSH_COHERENT_LINES;
-
- batch = gen8_emit_pipe_control(batch,
- PIPE_CONTROL_CS_STALL |
- PIPE_CONTROL_DC_FLUSH_ENABLE,
- 0);
-
- *batch++ = MI_LOAD_REGISTER_MEM_GEN8 | MI_SRM_LRM_GLOBAL_GTT;
- *batch++ = i915_mmio_reg_offset(GEN8_L3SQCREG4);
- *batch++ = i915_ggtt_offset(engine->scratch) + 256;
- *batch++ = 0;
-
- return batch;
-}
-
-/*
- * Typically we only have one indirect_ctx and per_ctx batch buffer which are
- * initialized at the beginning and shared across all contexts but this field
- * helps us to have multiple batches at different offsets and select them based
- * on a criteria. At the moment this batch always start at the beginning of the page
- * and at this point we don't have multiple wa_ctx batch buffers.
- *
- * The number of WA applied are not known at the beginning; we use this field
- * to return the no of DWORDS written.
- *
- * It is to be noted that this batch does not contain MI_BATCH_BUFFER_END
- * so it adds NOOPs as padding to make it cacheline aligned.
- * MI_BATCH_BUFFER_END will be added to perctx batch and both of them together
- * makes a complete batch buffer.
- */
-static u32 *gen8_init_indirectctx_bb(struct intel_engine_cs *engine, u32 *batch)
-{
- /* WaDisableCtxRestoreArbitration:bdw,chv */
- *batch++ = MI_ARB_ON_OFF | MI_ARB_DISABLE;
-
- /* WaFlushCoherentL3CacheLinesAtContextSwitch:bdw */
- if (IS_BROADWELL(engine->i915))
- batch = gen8_emit_flush_coherentl3_wa(engine, batch);
-
- /* WaClearSlmSpaceAtContextSwitch:bdw,chv */
- /* Actual scratch location is at 128 bytes offset */
- batch = gen8_emit_pipe_control(batch,
- PIPE_CONTROL_FLUSH_L3 |
- PIPE_CONTROL_GLOBAL_GTT_IVB |
- PIPE_CONTROL_CS_STALL |
- PIPE_CONTROL_QW_WRITE,
- i915_ggtt_offset(engine->scratch) +
- 2 * CACHELINE_BYTES);
-
- *batch++ = MI_ARB_ON_OFF | MI_ARB_ENABLE;
-
- /* Pad to end of cacheline */
- while ((unsigned long)batch % CACHELINE_BYTES)
- *batch++ = MI_NOOP;
-
- /*
- * MI_BATCH_BUFFER_END is not required in Indirect ctx BB because
- * execution depends on the length specified in terms of cache lines
- * in the register CTX_RCS_INDIRECT_CTX
- */
-
- return batch;
-}
-
-static u32 *gen9_init_indirectctx_bb(struct intel_engine_cs *engine, u32 *batch)
-{
- *batch++ = MI_ARB_ON_OFF | MI_ARB_DISABLE;
-
- /* WaFlushCoherentL3CacheLinesAtContextSwitch:skl,bxt,glk */
- batch = gen8_emit_flush_coherentl3_wa(engine, batch);
-
- /* WaDisableGatherAtSetShaderCommonSlice:skl,bxt,kbl,glk */
- *batch++ = MI_LOAD_REGISTER_IMM(1);
- *batch++ = i915_mmio_reg_offset(COMMON_SLICE_CHICKEN2);
- *batch++ = _MASKED_BIT_DISABLE(
- GEN9_DISABLE_GATHER_AT_SET_SHADER_COMMON_SLICE);
- *batch++ = MI_NOOP;
-
- /* WaClearSlmSpaceAtContextSwitch:kbl */
- /* Actual scratch location is at 128 bytes offset */
- if (IS_KBL_REVID(engine->i915, 0, KBL_REVID_A0)) {
- batch = gen8_emit_pipe_control(batch,
- PIPE_CONTROL_FLUSH_L3 |
- PIPE_CONTROL_GLOBAL_GTT_IVB |
- PIPE_CONTROL_CS_STALL |
- PIPE_CONTROL_QW_WRITE,
- i915_ggtt_offset(engine->scratch)
- + 2 * CACHELINE_BYTES);
- }
-
- /* WaMediaPoolStateCmdInWABB:bxt,glk */
- if (HAS_POOLED_EU(engine->i915)) {
- /*
- * EU pool configuration is setup along with golden context
- * during context initialization. This value depends on
- * device type (2x6 or 3x6) and needs to be updated based
- * on which subslice is disabled especially for 2x6
- * devices, however it is safe to load default
- * configuration of 3x6 device instead of masking off
- * corresponding bits because HW ignores bits of a disabled
- * subslice and drops down to appropriate config. Please
- * see render_state_setup() in i915_gem_render_state.c for
- * possible configurations, to avoid duplication they are
- * not shown here again.
- */
- *batch++ = GEN9_MEDIA_POOL_STATE;
- *batch++ = GEN9_MEDIA_POOL_ENABLE;
- *batch++ = 0x00777000;
- *batch++ = 0;
- *batch++ = 0;
- *batch++ = 0;
- }
-
- *batch++ = MI_ARB_ON_OFF | MI_ARB_ENABLE;
-
- /* Pad to end of cacheline */
- while ((unsigned long)batch % CACHELINE_BYTES)
- *batch++ = MI_NOOP;
-
- return batch;
-}
-
-#define CTX_WA_BB_OBJ_SIZE (PAGE_SIZE)
-
-static int lrc_setup_wa_ctx(struct intel_engine_cs *engine)
-{
- struct drm_i915_gem_object *obj;
- struct i915_vma *vma;
- int err;
-
- obj = i915_gem_object_create(engine->i915, CTX_WA_BB_OBJ_SIZE);
- if (IS_ERR(obj))
- return PTR_ERR(obj);
-
- vma = i915_vma_instance(obj, &engine->i915->ggtt.base, NULL);
- if (IS_ERR(vma)) {
- err = PTR_ERR(vma);
- goto err;
- }
-
- err = i915_vma_pin(vma, 0, PAGE_SIZE, PIN_GLOBAL | PIN_HIGH);
- if (err)
- goto err;
-
- engine->wa_ctx.vma = vma;
- return 0;
-
-err:
- i915_gem_object_put(obj);
- return err;
-}
-
-static void lrc_destroy_wa_ctx(struct intel_engine_cs *engine)
-{
- i915_vma_unpin_and_release(&engine->wa_ctx.vma);
-}
-
-typedef u32 *(*wa_bb_func_t)(struct intel_engine_cs *engine, u32 *batch);
-
-static int intel_init_workaround_bb(struct intel_engine_cs *engine)
-{
- struct i915_ctx_workarounds *wa_ctx = &engine->wa_ctx;
- struct i915_wa_ctx_bb *wa_bb[2] = { &wa_ctx->indirect_ctx,
- &wa_ctx->per_ctx };
- wa_bb_func_t wa_bb_fn[2];
- struct page *page;
- void *batch, *batch_ptr;
- unsigned int i;
- int ret;
-
- if (WARN_ON(engine->id != RCS || !engine->scratch))
- return -EINVAL;
-
- switch (INTEL_GEN(engine->i915)) {
- case 10:
- return 0;
- case 9:
- wa_bb_fn[0] = gen9_init_indirectctx_bb;
- wa_bb_fn[1] = NULL;
- break;
- case 8:
- wa_bb_fn[0] = gen8_init_indirectctx_bb;
- wa_bb_fn[1] = NULL;
- break;
- default:
- MISSING_CASE(INTEL_GEN(engine->i915));
- return 0;
- }
-
- ret = lrc_setup_wa_ctx(engine);
- if (ret) {
- DRM_DEBUG_DRIVER("Failed to setup context WA page: %d\n", ret);
- return ret;
- }
-
- page = i915_gem_object_get_dirty_page(wa_ctx->vma->obj, 0);
- batch = batch_ptr = kmap_atomic(page);
-
- /*
- * Emit the two workaround batch buffers, recording the offset from the
- * start of the workaround batch buffer object for each and their
- * respective sizes.
- */
- for (i = 0; i < ARRAY_SIZE(wa_bb_fn); i++) {
- wa_bb[i]->offset = batch_ptr - batch;
- if (WARN_ON(!IS_ALIGNED(wa_bb[i]->offset, CACHELINE_BYTES))) {
- ret = -EINVAL;
- break;
- }
- if (wa_bb_fn[i])
- batch_ptr = wa_bb_fn[i](engine, batch_ptr);
- wa_bb[i]->size = batch_ptr - (batch + wa_bb[i]->offset);
- }
-
- BUG_ON(batch_ptr - batch > CTX_WA_BB_OBJ_SIZE);
-
- kunmap_atomic(batch);
- if (ret)
- lrc_destroy_wa_ctx(engine);
-
- return ret;
-}
-
static u8 gtiir[] = {
[RCS] = 0,
[BCS] = 0,
@@ -1875,7 +1626,7 @@ void intel_logical_ring_cleanup(struct intel_engine_cs *engine)
intel_engine_cleanup_common(engine);
- lrc_destroy_wa_ctx(engine);
+ intel_engine_fini_bb_workarounds(engine);
engine->i915 = NULL;
dev_priv->engine[engine->id] = NULL;
kfree(engine);
@@ -1994,7 +1745,7 @@ int logical_render_ring_init(struct intel_engine_cs *engine)
if (ret)
return ret;
- ret = intel_init_workaround_bb(engine);
+ ret = intel_engine_init_bb_workarounds(engine);
if (ret) {
/*
* We continue even if we fail to initialize WA batch
diff --git a/drivers/gpu/drm/i915/intel_workarounds.c b/drivers/gpu/drm/i915/intel_workarounds.c
index 0a8f265..d7ec27c 100644
--- a/drivers/gpu/drm/i915/intel_workarounds.c
+++ b/drivers/gpu/drm/i915/intel_workarounds.c
@@ -862,3 +862,257 @@ int intel_whitelist_workarounds_apply(struct intel_engine_cs *engine)
dev_priv->workarounds.hw_whitelist_count[engine->id]);
return 0;
}
+
+/*
+ * In this WA we need to set GEN8_L3SQCREG4[21:21] and reset it after
+ * PIPE_CONTROL instruction. This is required for the flush to happen correctly
+ * but there is a slight complication as this is applied in WA batch where the
+ * values are only initialized once so we cannot take register value at the
+ * beginning and reuse it further; hence we save its value to memory, upload a
+ * constant value with bit21 set and then we restore it back with the saved value.
+ * To simplify the WA, a constant value is formed by using the default value
+ * of this register. This shouldn't be a problem because we are only modifying
+ * it for a short period and this batch in non-premptible. We can ofcourse
+ * use additional instructions that read the actual value of the register
+ * at that time and set our bit of interest but it makes the WA complicated.
+ *
+ * This WA is also required for Gen9 so extracting as a function avoids
+ * code duplication.
+ */
+static u32 *
+gen8_emit_flush_coherentl3_wa(struct intel_engine_cs *engine, u32 *batch)
+{
+ *batch++ = MI_STORE_REGISTER_MEM_GEN8 | MI_SRM_LRM_GLOBAL_GTT;
+ *batch++ = i915_mmio_reg_offset(GEN8_L3SQCREG4);
+ *batch++ = i915_ggtt_offset(engine->scratch) + 256;
+ *batch++ = 0;
+
+ *batch++ = MI_LOAD_REGISTER_IMM(1);
+ *batch++ = i915_mmio_reg_offset(GEN8_L3SQCREG4);
+ *batch++ = 0x40400000 | GEN8_LQSC_FLUSH_COHERENT_LINES;
+
+ batch = gen8_emit_pipe_control(batch,
+ PIPE_CONTROL_CS_STALL |
+ PIPE_CONTROL_DC_FLUSH_ENABLE,
+ 0);
+
+ *batch++ = MI_LOAD_REGISTER_MEM_GEN8 | MI_SRM_LRM_GLOBAL_GTT;
+ *batch++ = i915_mmio_reg_offset(GEN8_L3SQCREG4);
+ *batch++ = i915_ggtt_offset(engine->scratch) + 256;
+ *batch++ = 0;
+
+ return batch;
+}
+
+/*
+ * Typically we only have one indirect_ctx and per_ctx batch buffer which are
+ * initialized at the beginning and shared across all contexts but this field
+ * helps us to have multiple batches at different offsets and select them based
+ * on a criteria. At the moment this batch always start at the beginning of the page
+ * and at this point we don't have multiple wa_ctx batch buffers.
+ *
+ * The number of WA applied are not known at the beginning; we use this field
+ * to return the no of DWORDS written.
+ *
+ * It is to be noted that this batch does not contain MI_BATCH_BUFFER_END
+ * so it adds NOOPs as padding to make it cacheline aligned.
+ * MI_BATCH_BUFFER_END will be added to perctx batch and both of them together
+ * makes a complete batch buffer.
+ */
+static u32 *gen8_init_indirectctx_bb(struct intel_engine_cs *engine, u32 *batch)
+{
+ /* WaDisableCtxRestoreArbitration:bdw,chv */
+ *batch++ = MI_ARB_ON_OFF | MI_ARB_DISABLE;
+
+ /* WaFlushCoherentL3CacheLinesAtContextSwitch:bdw */
+ if (IS_BROADWELL(engine->i915))
+ batch = gen8_emit_flush_coherentl3_wa(engine, batch);
+
+ /* WaClearSlmSpaceAtContextSwitch:bdw,chv */
+ /* Actual scratch location is at 128 bytes offset */
+ batch = gen8_emit_pipe_control(batch,
+ PIPE_CONTROL_FLUSH_L3 |
+ PIPE_CONTROL_GLOBAL_GTT_IVB |
+ PIPE_CONTROL_CS_STALL |
+ PIPE_CONTROL_QW_WRITE,
+ i915_ggtt_offset(engine->scratch) +
+ 2 * CACHELINE_BYTES);
+
+ *batch++ = MI_ARB_ON_OFF | MI_ARB_ENABLE;
+
+ /* Pad to end of cacheline */
+ while ((unsigned long)batch % CACHELINE_BYTES)
+ *batch++ = MI_NOOP;
+
+ /*
+ * MI_BATCH_BUFFER_END is not required in Indirect ctx BB because
+ * execution depends on the length specified in terms of cache lines
+ * in the register CTX_RCS_INDIRECT_CTX
+ */
+
+ return batch;
+}
+
+static u32 *gen9_init_indirectctx_bb(struct intel_engine_cs *engine, u32 *batch)
+{
+ *batch++ = MI_ARB_ON_OFF | MI_ARB_DISABLE;
+
+ /* WaFlushCoherentL3CacheLinesAtContextSwitch:skl,bxt,glk */
+ batch = gen8_emit_flush_coherentl3_wa(engine, batch);
+
+ /* WaDisableGatherAtSetShaderCommonSlice:skl,bxt,kbl,glk */
+ *batch++ = MI_LOAD_REGISTER_IMM(1);
+ *batch++ = i915_mmio_reg_offset(COMMON_SLICE_CHICKEN2);
+ *batch++ = _MASKED_BIT_DISABLE(
+ GEN9_DISABLE_GATHER_AT_SET_SHADER_COMMON_SLICE);
+ *batch++ = MI_NOOP;
+
+ /* WaClearSlmSpaceAtContextSwitch:kbl */
+ /* Actual scratch location is at 128 bytes offset */
+ if (IS_KBL_REVID(engine->i915, 0, KBL_REVID_A0)) {
+ batch = gen8_emit_pipe_control(batch,
+ PIPE_CONTROL_FLUSH_L3 |
+ PIPE_CONTROL_GLOBAL_GTT_IVB |
+ PIPE_CONTROL_CS_STALL |
+ PIPE_CONTROL_QW_WRITE,
+ i915_ggtt_offset(engine->scratch)
+ + 2 * CACHELINE_BYTES);
+ }
+
+ /* WaMediaPoolStateCmdInWABB:bxt,glk */
+ if (HAS_POOLED_EU(engine->i915)) {
+ /*
+ * EU pool configuration is setup along with golden context
+ * during context initialization. This value depends on
+ * device type (2x6 or 3x6) and needs to be updated based
+ * on which subslice is disabled especially for 2x6
+ * devices, however it is safe to load default
+ * configuration of 3x6 device instead of masking off
+ * corresponding bits because HW ignores bits of a disabled
+ * subslice and drops down to appropriate config. Please
+ * see render_state_setup() in i915_gem_render_state.c for
+ * possible configurations, to avoid duplication they are
+ * not shown here again.
+ */
+ *batch++ = GEN9_MEDIA_POOL_STATE;
+ *batch++ = GEN9_MEDIA_POOL_ENABLE;
+ *batch++ = 0x00777000;
+ *batch++ = 0;
+ *batch++ = 0;
+ *batch++ = 0;
+ }
+
+ *batch++ = MI_ARB_ON_OFF | MI_ARB_ENABLE;
+
+ /* Pad to end of cacheline */
+ while ((unsigned long)batch % CACHELINE_BYTES)
+ *batch++ = MI_NOOP;
+
+ return batch;
+}
+
+#define CTX_WA_BB_OBJ_SIZE (PAGE_SIZE)
+
+static int lrc_setup_wa_ctx(struct intel_engine_cs *engine)
+{
+ struct drm_i915_gem_object *obj;
+ struct i915_vma *vma;
+ int err;
+
+ obj = i915_gem_object_create(engine->i915, CTX_WA_BB_OBJ_SIZE);
+ if (IS_ERR(obj))
+ return PTR_ERR(obj);
+
+ vma = i915_vma_instance(obj, &engine->i915->ggtt.base, NULL);
+ if (IS_ERR(vma)) {
+ err = PTR_ERR(vma);
+ goto err;
+ }
+
+ err = i915_vma_pin(vma, 0, PAGE_SIZE, PIN_GLOBAL | PIN_HIGH);
+ if (err)
+ goto err;
+
+ engine->wa_ctx.vma = vma;
+ return 0;
+
+err:
+ i915_gem_object_put(obj);
+ return err;
+}
+
+static void lrc_destroy_wa_ctx(struct intel_engine_cs *engine)
+{
+ i915_vma_unpin_and_release(&engine->wa_ctx.vma);
+}
+
+typedef u32 *(*wa_bb_func_t)(struct intel_engine_cs *engine, u32 *batch);
+
+int intel_engine_init_bb_workarounds(struct intel_engine_cs *engine)
+{
+ struct i915_ctx_workarounds *wa_ctx = &engine->wa_ctx;
+ struct i915_wa_ctx_bb *wa_bb[2] = { &wa_ctx->indirect_ctx,
+ &wa_ctx->per_ctx };
+ wa_bb_func_t wa_bb_fn[2];
+ struct page *page;
+ void *batch, *batch_ptr;
+ unsigned int i;
+ int ret;
+
+ if (WARN_ON(engine->id != RCS || !engine->scratch))
+ return -EINVAL;
+
+ switch (INTEL_GEN(engine->i915)) {
+ case 10:
+ return 0;
+ case 9:
+ wa_bb_fn[0] = gen9_init_indirectctx_bb;
+ wa_bb_fn[1] = NULL;
+ break;
+ case 8:
+ wa_bb_fn[0] = gen8_init_indirectctx_bb;
+ wa_bb_fn[1] = NULL;
+ break;
+ default:
+ MISSING_CASE(INTEL_GEN(engine->i915));
+ return 0;
+ }
+
+ ret = lrc_setup_wa_ctx(engine);
+ if (ret) {
+ DRM_DEBUG_DRIVER("Failed to setup context WA page: %d\n", ret);
+ return ret;
+ }
+
+ page = i915_gem_object_get_dirty_page(wa_ctx->vma->obj, 0);
+ batch = batch_ptr = kmap_atomic(page);
+
+ /*
+ * Emit the two workaround batch buffers, recording the offset from the
+ * start of the workaround batch buffer object for each and their
+ * respective sizes.
+ */
+ for (i = 0; i < ARRAY_SIZE(wa_bb_fn); i++) {
+ wa_bb[i]->offset = batch_ptr - batch;
+ if (WARN_ON(!IS_ALIGNED(wa_bb[i]->offset, CACHELINE_BYTES))) {
+ ret = -EINVAL;
+ break;
+ }
+ if (wa_bb_fn[i])
+ batch_ptr = wa_bb_fn[i](engine, batch_ptr);
+ wa_bb[i]->size = batch_ptr - (batch + wa_bb[i]->offset);
+ }
+
+ BUG_ON(batch_ptr - batch > CTX_WA_BB_OBJ_SIZE);
+
+ kunmap_atomic(batch);
+ if (ret)
+ lrc_destroy_wa_ctx(engine);
+
+ return ret;
+}
+
+void intel_engine_fini_bb_workarounds(struct intel_engine_cs *engine)
+{
+ lrc_destroy_wa_ctx(engine);
+}
diff --git a/drivers/gpu/drm/i915/intel_workarounds.h b/drivers/gpu/drm/i915/intel_workarounds.h
index bba51bb..aa9a894 100644
--- a/drivers/gpu/drm/i915/intel_workarounds.h
+++ b/drivers/gpu/drm/i915/intel_workarounds.h
@@ -32,4 +32,7 @@
int intel_whitelist_workarounds_apply(struct intel_engine_cs *engine);
+int intel_engine_init_bb_workarounds(struct intel_engine_cs *engine);
+void intel_engine_fini_bb_workarounds(struct intel_engine_cs *engine);
+
#endif
--
1.9.1
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
