[Intel-gfx] [PATCH 3/5] drm/i915/guc: Add GuC ADS - scheduler policies

yu.dai at intel.com yu.dai at intel.com
Wed Dec 16 13:40:53 PST 2015


From: Alex Dai <yu.dai at intel.com>

GuC supports different scheduling policies for its four internal
queues. Currently these have been set to the same default values
as KMD_NORMAL queue.

Particularly POLICY_MAX_NUM_WI is set to 15 to match GuC internal
maximum submit queue numbers to avoid an out-of-space problem.
This value indicates max number of work items allowed to be queued
for one DPC process. A smaller value will let GuC schedule more
frequently while a larger number may increase chances to optimize
cmds (such as collapse cmds from same lrc) with risks that keeps
CS idle.

Signed-off-by: Alex Dai <yu.dai at intel.com>
---
 drivers/gpu/drm/i915/i915_guc_submission.c | 31 +++++++++++++++++++-
 drivers/gpu/drm/i915/intel_guc_fwif.h      | 45 ++++++++++++++++++++++++++++++
 2 files changed, 75 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/i915/i915_guc_submission.c b/drivers/gpu/drm/i915/i915_guc_submission.c
index 66d85c3..a5c555c 100644
--- a/drivers/gpu/drm/i915/i915_guc_submission.c
+++ b/drivers/gpu/drm/i915/i915_guc_submission.c
@@ -842,17 +842,39 @@ static void guc_create_log(struct intel_guc *guc)
 	guc->log_flags = (offset << GUC_LOG_BUF_ADDR_SHIFT) | flags;
 }
 
+static void init_guc_policies(struct guc_policies *policies)
+{
+	struct guc_policy *policy;
+	u32 p, i;
+
+	policies->dpc_promote_time = 500000;
+	policies->max_num_work_items = POLICY_MAX_NUM_WI;
+
+	for (p = 0; p < GUC_CTX_PRIORITY_NUM; p++)
+	for (i = 0; i < I915_NUM_RINGS; i++) {
+		policy = &policies->policy[p][i];
+
+		policy->execution_quantum = 1000000;
+		policy->preemption_time = 500000;
+		policy->fault_time = 250000;
+		policy->policy_flags = 0;
+	}
+
+	policies->is_valid = 1;
+}
+
 static void guc_create_ads(struct intel_guc *guc)
 {
 	struct drm_i915_private *dev_priv = guc_to_i915(guc);
 	struct drm_i915_gem_object *obj;
 	struct guc_ads *ads;
+	struct guc_policies *policies;
 	struct intel_engine_cs *ring;
 	struct page *page;
 	u32 size, i;
 
 	/* The ads obj includes the struct itself and buffers passed to GuC */
-	size = sizeof(struct guc_ads);
+	size = sizeof(struct guc_ads) + sizeof(struct guc_policies);
 
 	obj = guc->ads_obj;
 	if (!obj) {
@@ -884,6 +906,13 @@ static void guc_create_ads(struct intel_guc *guc)
 	for_each_ring(ring, dev_priv, i)
 		ads->eng_state_size[i] = intel_lr_context_size(ring);
 
+	/* GuC scheduling policies */
+	policies = (void *)ads + sizeof(struct guc_ads);
+	init_guc_policies(policies);
+
+	ads->scheduler_policies = i915_gem_obj_ggtt_offset(obj) +
+			sizeof(struct guc_ads);
+
 	kunmap_atomic(ads);
 }
 
diff --git a/drivers/gpu/drm/i915/intel_guc_fwif.h b/drivers/gpu/drm/i915/intel_guc_fwif.h
index 76ecc85..0cc17c7 100644
--- a/drivers/gpu/drm/i915/intel_guc_fwif.h
+++ b/drivers/gpu/drm/i915/intel_guc_fwif.h
@@ -39,6 +39,7 @@
 #define GUC_CTX_PRIORITY_HIGH		1
 #define GUC_CTX_PRIORITY_KMD_NORMAL	2
 #define GUC_CTX_PRIORITY_NORMAL		3
+#define GUC_CTX_PRIORITY_NUM		4
 
 #define GUC_MAX_GPU_CONTEXTS		1024
 #define	GUC_INVALID_CTX_ID		GUC_MAX_GPU_CONTEXTS
@@ -316,6 +317,50 @@ struct guc_context_desc {
 #define GUC_POWER_D2		3
 #define GUC_POWER_D3		4
 
+/* Scheduling policy settings */
+
+/* Reset engine upon preempt failure */
+#define POLICY_RESET_ENGINE		(1<<0)
+/* Preempt to idle on quantum expiry */
+#define POLICY_PREEMPT_TO_IDLE		(1<<1)
+
+#define POLICY_MAX_NUM_WI		15
+
+struct guc_policy {
+	/* Time for one workload to execute. (in micro seconds) */
+	u32 execution_quantum;
+	u32 reserved1;
+
+	/* Time to wait for a preemption request to completed before issuing a
+	 * reset. (in micro seconds). */
+	u32 preemption_time;
+
+	/* How much time to allow to run after the first fault is observed.
+	 * Then preempt afterwards. (in micro seconds) */
+	u32 fault_time;
+
+	u32 policy_flags;
+	u32 reserved[2];
+} __packed;
+
+struct guc_policies {
+	struct guc_policy policy[GUC_CTX_PRIORITY_NUM][I915_NUM_RINGS];
+
+	/* In micro seconds. How much time to allow before DPC processing is
+	 * called back via interrupt (to prevent DPC queue drain starving).
+	 * Typically 1000s of micro seconds (example only, not granularity). */
+	u32 dpc_promote_time;
+
+	/* Must be set to take these new values. */
+	u32 is_valid;
+
+	/* Max number of WIs to process per call. A large value may keep CS
+	 * idle. */
+	u32 max_num_work_items;
+
+	u32 reserved[19];
+} __packed;
+
 /* GuC Additional Data Struct */
 
 struct guc_ads {
-- 
2.5.0



More information about the Intel-gfx mailing list