[PATCH 10/10] move-switch-ctx
Chris Wilson
chris at chris-wilson.co.uk
Sat Nov 18 00:07:28 UTC 2017
---
drivers/gpu/drm/i915/i915_gem_context.c | 197 --------------------------------
drivers/gpu/drm/i915/intel_ringbuffer.c | 185 +++++++++++++++++++++++++++++-
2 files changed, 184 insertions(+), 198 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c
index f63bec08cc85..aee0f6d72d33 100644
--- a/drivers/gpu/drm/i915/i915_gem_context.c
+++ b/drivers/gpu/drm/i915/i915_gem_context.c
@@ -567,203 +567,6 @@ void i915_gem_context_close(struct drm_file *file)
idr_destroy(&file_priv->context_idr);
}
-static inline int
-mi_set_context(struct drm_i915_gem_request *req, u32 flags)
-{
- struct drm_i915_private *dev_priv = req->i915;
- struct intel_engine_cs *engine = req->engine;
- enum intel_engine_id id;
- const int num_rings =
- /* Use an extended w/a on gen7 if signalling from other rings */
- (HAS_LEGACY_SEMAPHORES(dev_priv) && IS_GEN7(dev_priv)) ?
- INTEL_INFO(dev_priv)->num_rings - 1 :
- 0;
- int len;
- u32 *cs;
-
- flags |= MI_MM_SPACE_GTT;
- if (IS_HASWELL(dev_priv))
- /* These flags are for resource streamer on HSW+ */
- flags |= HSW_MI_RS_SAVE_STATE_EN | HSW_MI_RS_RESTORE_STATE_EN;
- else
- flags |= MI_SAVE_EXT_STATE_EN | MI_RESTORE_EXT_STATE_EN;
-
- len = 4;
- if (IS_GEN7(dev_priv))
- len += 2 + (num_rings ? 4*num_rings + 6 : 0);
-
- cs = intel_ring_begin(req, len);
- if (IS_ERR(cs))
- return PTR_ERR(cs);
-
- /* WaProgramMiArbOnOffAroundMiSetContext:ivb,vlv,hsw,bdw,chv */
- if (IS_GEN7(dev_priv)) {
- *cs++ = MI_ARB_ON_OFF | MI_ARB_DISABLE;
- if (num_rings) {
- struct intel_engine_cs *signaller;
-
- *cs++ = MI_LOAD_REGISTER_IMM(num_rings);
- for_each_engine(signaller, dev_priv, id) {
- if (signaller == engine)
- continue;
-
- *cs++ = i915_mmio_reg_offset(
- RING_PSMI_CTL(signaller->mmio_base));
- *cs++ = _MASKED_BIT_ENABLE(
- GEN6_PSMI_SLEEP_MSG_DISABLE);
- }
- }
- }
-
- *cs++ = MI_NOOP;
- *cs++ = MI_SET_CONTEXT;
- *cs++ = i915_ggtt_offset(req->ctx->engine[RCS].state) | flags;
- /*
- * w/a: MI_SET_CONTEXT must always be followed by MI_NOOP
- * WaMiSetContext_Hang:snb,ivb,vlv
- */
- *cs++ = MI_NOOP;
-
- if (IS_GEN7(dev_priv)) {
- if (num_rings) {
- struct intel_engine_cs *signaller;
- i915_reg_t last_reg = {}; /* keep gcc quiet */
-
- *cs++ = MI_LOAD_REGISTER_IMM(num_rings);
- for_each_engine(signaller, dev_priv, id) {
- if (signaller == engine)
- continue;
-
- last_reg = RING_PSMI_CTL(signaller->mmio_base);
- *cs++ = i915_mmio_reg_offset(last_reg);
- *cs++ = _MASKED_BIT_DISABLE(
- GEN6_PSMI_SLEEP_MSG_DISABLE);
- }
-
- /* Insert a delay before the next switch! */
- *cs++ = MI_STORE_REGISTER_MEM | MI_SRM_LRM_GLOBAL_GTT;
- *cs++ = i915_mmio_reg_offset(last_reg);
- *cs++ = i915_ggtt_offset(engine->scratch);
- *cs++ = MI_NOOP;
- }
- *cs++ = MI_ARB_ON_OFF | MI_ARB_ENABLE;
- }
-
- intel_ring_advance(req, cs);
-
- return 0;
-}
-
-static int remap_l3(struct drm_i915_gem_request *req, int slice)
-{
- u32 *cs, *remap_info = req->i915->l3_parity.remap_info[slice];
- int i;
-
- if (!remap_info)
- return 0;
-
- cs = intel_ring_begin(req, GEN7_L3LOG_SIZE/4 * 2 + 2);
- if (IS_ERR(cs))
- return PTR_ERR(cs);
-
- /*
- * Note: We do not worry about the concurrent register cacheline hang
- * here because no other code should access these registers other than
- * at initialization time.
- */
- *cs++ = MI_LOAD_REGISTER_IMM(GEN7_L3LOG_SIZE/4);
- for (i = 0; i < GEN7_L3LOG_SIZE/4; i++) {
- *cs++ = i915_mmio_reg_offset(GEN7_L3LOG(slice, i));
- *cs++ = remap_info[i];
- }
- *cs++ = MI_NOOP;
- intel_ring_advance(req, cs);
-
- return 0;
-}
-
-/**
- * i915_switch_context() - perform a GPU context switch.
- * @rq: request for which we'll execute the context switch
- *
- * The context life cycle is simple. The context refcount is incremented and
- * decremented by 1 and create and destroy. If the context is in use by the GPU,
- * it will have a refcount > 1. This allows us to destroy the context abstract
- * object while letting the normal object tracking destroy the backing BO.
- *
- * This function should not be used in execlists mode. Instead the context is
- * switched by writing to the ELSP and requests keep a reference to their
- * context.
- */
-int i915_switch_context(struct drm_i915_gem_request *rq)
-{
- struct intel_engine_cs *engine = rq->engine;
- struct i915_gem_context *to = rq->ctx;
- struct i915_hw_ppgtt *ppgtt = to->ppgtt ?: rq->i915->mm.aliasing_ppgtt;
- struct i915_gem_context *saved_ctx = engine->legacy_active_context;
- struct i915_hw_ppgtt *saved_mm = engine->legacy_active_ppgtt;
- u32 hw_flags = 0;
- int ret, i;
-
- lockdep_assert_held(&rq->i915->drm.struct_mutex);
- GEM_BUG_ON(HAS_EXECLISTS(rq->i915));
-
- if (ppgtt != saved_mm ||
- (ppgtt && intel_engine_flag(engine) & ppgtt->pd_dirty_rings)) {
- trace_switch_mm(engine, to);
- ret = ppgtt->switch_mm(ppgtt, rq);
- if (ret)
- goto err;
-
- ppgtt->pd_dirty_rings &= ~intel_engine_flag(engine);
- engine->legacy_active_ppgtt = ppgtt;
- hw_flags = MI_FORCE_RESTORE;
- }
-
- if (to->engine[engine->id].state &&
- (to != saved_ctx || hw_flags & MI_FORCE_RESTORE)) {
- GEM_BUG_ON(engine->id != RCS);
-
- /*
- * The kernel context(s) is treated as pure scratch and is not
- * expected to retain any state (as we sacrifice it during
- * suspend and on resume it may be corrupted). This is ok,
- * as nothing actually executes using the kernel context; it
- * is purely used for flushing user contexts.
- */
- if (i915_gem_context_is_kernel(to))
- hw_flags = MI_RESTORE_INHIBIT;
-
- ret = mi_set_context(rq, hw_flags);
- if (ret)
- goto err_mm;
-
- engine->legacy_active_context = to;
- }
-
- if (to->remap_slice) {
- for (i = 0; i < MAX_L3_SLICES; i++) {
- if (!(to->remap_slice & BIT(i)))
- continue;
-
- ret = remap_l3(rq, i);
- if (ret)
- goto err_ctx;
- }
-
- to->remap_slice = 0;
- }
-
- return 0;
-
-err_ctx:
- engine->legacy_active_context = saved_ctx;
-err_mm:
- engine->legacy_active_ppgtt = saved_mm;
-err:
- return ret;
-}
-
static bool engine_has_idle_kernel_context(struct intel_engine_cs *engine)
{
struct i915_gem_timeline *timeline;
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
index a904b0353bec..11f3fbc12e65 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.c
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
@@ -1385,6 +1385,189 @@ void intel_legacy_submission_resume(struct drm_i915_private *dev_priv)
intel_ring_reset(engine->buffer, 0);
}
+static inline int mi_set_context(struct drm_i915_gem_request *rq, u32 flags)
+{
+ struct drm_i915_private *dev_priv = rq->i915;
+ struct intel_engine_cs *engine = rq->engine;
+ enum intel_engine_id id;
+ const int num_rings =
+ /* Use an extended w/a on gen7 if signalling from other rings */
+ (HAS_LEGACY_SEMAPHORES(dev_priv) && IS_GEN7(dev_priv)) ?
+ INTEL_INFO(dev_priv)->num_rings - 1 :
+ 0;
+ int len;
+ u32 *cs;
+
+ flags |= MI_MM_SPACE_GTT;
+ if (IS_HASWELL(dev_priv))
+ /* These flags are for resource streamer on HSW+ */
+ flags |= HSW_MI_RS_SAVE_STATE_EN | HSW_MI_RS_RESTORE_STATE_EN;
+ else
+ flags |= MI_SAVE_EXT_STATE_EN | MI_RESTORE_EXT_STATE_EN;
+
+ len = 4;
+ if (IS_GEN7(dev_priv))
+ len += 2 + (num_rings ? 4*num_rings + 6 : 0);
+
+ cs = intel_ring_begin(rq, len);
+ if (IS_ERR(cs))
+ return PTR_ERR(cs);
+
+ /* WaProgramMiArbOnOffAroundMiSetContext:ivb,vlv,hsw,bdw,chv */
+ if (IS_GEN7(dev_priv)) {
+ *cs++ = MI_ARB_ON_OFF | MI_ARB_DISABLE;
+ if (num_rings) {
+ struct intel_engine_cs *signaller;
+
+ *cs++ = MI_LOAD_REGISTER_IMM(num_rings);
+ for_each_engine(signaller, dev_priv, id) {
+ if (signaller == engine)
+ continue;
+
+ *cs++ = i915_mmio_reg_offset(
+ RING_PSMI_CTL(signaller->mmio_base));
+ *cs++ = _MASKED_BIT_ENABLE(
+ GEN6_PSMI_SLEEP_MSG_DISABLE);
+ }
+ }
+ }
+
+ *cs++ = MI_NOOP;
+ *cs++ = MI_SET_CONTEXT;
+ *cs++ = i915_ggtt_offset(rq->ctx->engine[RCS].state) | flags;
+ /*
+ * w/a: MI_SET_CONTEXT must always be followed by MI_NOOP
+ * WaMiSetContext_Hang:snb,ivb,vlv
+ */
+ *cs++ = MI_NOOP;
+
+ if (IS_GEN7(dev_priv)) {
+ if (num_rings) {
+ struct intel_engine_cs *signaller;
+ i915_reg_t last_reg = {}; /* keep gcc quiet */
+
+ *cs++ = MI_LOAD_REGISTER_IMM(num_rings);
+ for_each_engine(signaller, dev_priv, id) {
+ if (signaller == engine)
+ continue;
+
+ last_reg = RING_PSMI_CTL(signaller->mmio_base);
+ *cs++ = i915_mmio_reg_offset(last_reg);
+ *cs++ = _MASKED_BIT_DISABLE(
+ GEN6_PSMI_SLEEP_MSG_DISABLE);
+ }
+
+ /* Insert a delay before the next switch! */
+ *cs++ = MI_STORE_REGISTER_MEM | MI_SRM_LRM_GLOBAL_GTT;
+ *cs++ = i915_mmio_reg_offset(last_reg);
+ *cs++ = i915_ggtt_offset(engine->scratch);
+ *cs++ = MI_NOOP;
+ }
+ *cs++ = MI_ARB_ON_OFF | MI_ARB_ENABLE;
+ }
+
+ intel_ring_advance(rq, cs);
+
+ return 0;
+}
+
+static int remap_l3(struct drm_i915_gem_request *rq, int slice)
+{
+ u32 *cs, *remap_info = rq->i915->l3_parity.remap_info[slice];
+ int i;
+
+ if (!remap_info)
+ return 0;
+
+ cs = intel_ring_begin(rq, GEN7_L3LOG_SIZE/4 * 2 + 2);
+ if (IS_ERR(cs))
+ return PTR_ERR(cs);
+
+ /*
+ * Note: We do not worry about the concurrent register cacheline hang
+ * here because no other code should access these registers other than
+ * at initialization time.
+ */
+ *cs++ = MI_LOAD_REGISTER_IMM(GEN7_L3LOG_SIZE/4);
+ for (i = 0; i < GEN7_L3LOG_SIZE/4; i++) {
+ *cs++ = i915_mmio_reg_offset(GEN7_L3LOG(slice, i));
+ *cs++ = remap_info[i];
+ }
+ *cs++ = MI_NOOP;
+ intel_ring_advance(rq, cs);
+
+ return 0;
+}
+
+static int switch_context(struct drm_i915_gem_request *rq)
+{
+ struct intel_engine_cs *engine = rq->engine;
+ struct i915_gem_context *to = rq->ctx;
+ struct i915_hw_ppgtt *ppgtt = to->ppgtt ?: rq->i915->mm.aliasing_ppgtt;
+ struct i915_gem_context *saved_ctx = engine->legacy_active_context;
+ struct i915_hw_ppgtt *saved_mm = engine->legacy_active_ppgtt;
+ u32 hw_flags = 0;
+ int ret, i;
+
+ lockdep_assert_held(&rq->i915->drm.struct_mutex);
+ GEM_BUG_ON(HAS_EXECLISTS(rq->i915));
+
+ if (ppgtt != saved_mm ||
+ (ppgtt && intel_engine_flag(engine) & ppgtt->pd_dirty_rings)) {
+ trace_switch_mm(engine, to);
+ ret = ppgtt->switch_mm(ppgtt, rq);
+ if (ret)
+ goto err;
+
+ ppgtt->pd_dirty_rings &= ~intel_engine_flag(engine);
+ engine->legacy_active_ppgtt = ppgtt;
+ hw_flags = MI_FORCE_RESTORE;
+ }
+
+ if (to->engine[engine->id].state &&
+ (to != saved_ctx || hw_flags & MI_FORCE_RESTORE)) {
+ GEM_BUG_ON(engine->id != RCS);
+
+ /*
+ * The kernel context(s) is treated as pure scratch and is not
+ * expected to retain any state (as we sacrifice it during
+ * suspend and on resume it may be corrupted). This is ok,
+ * as nothing actually executes using the kernel context; it
+ * is purely used for flushing user contexts.
+ */
+ if (i915_gem_context_is_kernel(to))
+ hw_flags = MI_RESTORE_INHIBIT;
+
+ ret = mi_set_context(rq, hw_flags);
+ if (ret)
+ goto err_mm;
+
+ engine->legacy_active_context = to;
+ }
+
+ if (to->remap_slice) {
+ for (i = 0; i < MAX_L3_SLICES; i++) {
+ if (!(to->remap_slice & BIT(i)))
+ continue;
+
+ ret = remap_l3(rq, i);
+ if (ret)
+ goto err_ctx;
+ }
+
+ to->remap_slice = 0;
+ }
+
+ return 0;
+
+err_ctx:
+ engine->legacy_active_context = saved_ctx;
+err_mm:
+ engine->legacy_active_ppgtt = saved_mm;
+err:
+ return ret;
+}
+
static int ring_request_alloc(struct drm_i915_gem_request *request)
{
int ret;
@@ -1401,7 +1584,7 @@ static int ring_request_alloc(struct drm_i915_gem_request *request)
if (ret)
return ret;
- ret = i915_switch_context(request);
+ ret = switch_context(request);
if (ret)
return ret;
--
2.15.0
More information about the Intel-gfx-trybot
mailing list