[PATCH 42/43] drm/i915/bdw: Pin the context backing objects to GGTT on-demand

[Thomas Daniel <thomas.daniel@intel.com>]

From: Oscar Mateo <oscar.mateo@intel.com>

Up until now, we have pinned every logical ring context backing object
during creation, and left it pinned until destruction. This made my life
easier, but it's a harmful thing to do, because we cause fragmentation
of the GGTT (and, eventually, we would run out of space).

This patch makes the pinning on-demand: the backing objects of the two
contexts that are written to the ELSP are pinned right before submission
and unpinned once the hardware is done with them. The only context that
is still pinned regardless is the global default one, so that the HWS can
still be accessed in the same way (ring->status_page).

v2: In the early version of this patch, we were pinning the context as
we put it into the ELSP: on the one hand, this is very efficient because
only a maximum two contexts are pinned at any given time, but on the other
hand, we cannot really pin in interrupt time :(

Signed-off-by: Oscar Mateo <oscar.mateo@intel.com>
---
 drivers/gpu/drm/i915/i915_debugfs.c |   11 +++++++--
 drivers/gpu/drm/i915/i915_drv.h     |    1 +
 drivers/gpu/drm/i915/i915_gem.c     |   44 ++++++++++++++++++++++++-----------
 drivers/gpu/drm/i915/intel_lrc.c    |   42 ++++++++++++++++++++++++---------
 drivers/gpu/drm/i915/intel_lrc.h    |    2 ++
 5 files changed, 73 insertions(+), 27 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
index 968c3c0..84531cc 100644
--- a/drivers/gpu/drm/i915/i915_debugfs.c
+++ b/drivers/gpu/drm/i915/i915_debugfs.c
@@ -1721,10 +1721,15 @@ static int i915_dump_lrc(struct seq_file *m, void *unused)
 				continue;
 
 			if (ctx_obj) {
-				struct page *page = i915_gem_object_get_page(ctx_obj, 1);
-				uint32_t *reg_state = kmap_atomic(page);
+				struct page *page ;
+				uint32_t *reg_state;
 				int j;
 
+				i915_gem_obj_ggtt_pin(ctx_obj, GEN8_LR_CONTEXT_ALIGN, 0);
+
+				page = i915_gem_object_get_page(ctx_obj, 1);
+				reg_state = kmap_atomic(page);
+
 				seq_printf(m, "CONTEXT: %s %u\n", ring->name,
 						intel_execlists_ctx_id(ctx_obj));
 
@@ -1736,6 +1741,8 @@ static int i915_dump_lrc(struct seq_file *m, void *unused)
 				}
 				kunmap_atomic(reg_state);
 
+				i915_gem_object_ggtt_unpin(ctx_obj);
+
 				seq_putc(m, '\n');
 			}
 		}
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 1ce51d6..70466af 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -628,6 +628,7 @@ struct intel_context {
 	struct {
 		struct drm_i915_gem_object *state;
 		struct intel_ringbuffer *ringbuf;
+		atomic_t unpin_count;
 	} engine[I915_NUM_RINGS];
 
 	struct list_head link;
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 143cff7..42faaa3 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -2491,12 +2491,23 @@ static void i915_set_reset_status(struct drm_i915_private *dev_priv,
 
 static void i915_gem_free_request(struct drm_i915_gem_request *request)
 {
+	struct intel_context *ctx = request->ctx;
+
 	list_del(&request->list);
 	i915_gem_request_remove_from_client(request);
 
-	if (request->ctx)
-		i915_gem_context_unreference(request->ctx);
+	if (ctx) {
+		struct intel_engine_cs *ring = request->ring;
+		struct drm_i915_gem_object *ctx_obj = ctx->engine[ring->id].state;
+		atomic_t *unpin_count = &ctx->engine[ring->id].unpin_count;
 
+		if (ctx_obj) {
+			if (atomic_dec_return(unpin_count) == 0 &&
+					ctx != ring->default_context)
+				i915_gem_object_ggtt_unpin(ctx_obj);
+		}
+		i915_gem_context_unreference(ctx);
+	}
 	kfree(request);
 }
 
@@ -2551,6 +2562,23 @@ static void i915_gem_reset_ring_cleanup(struct drm_i915_private *dev_priv,
 	}
 
 	/*
+	 * Clear the execlists queue up before freeing the requests, as those
+	 * are the ones that keep the context and ringbuffer backing objects
+	 * pinned in place.
+	 */
+	while (!list_empty(&ring->execlist_queue)) {
+		struct intel_ctx_submit_request *submit_req;
+
+		submit_req = list_first_entry(&ring->execlist_queue,
+				struct intel_ctx_submit_request,
+				execlist_link);
+		list_del(&submit_req->execlist_link);
+		intel_runtime_pm_put(dev_priv);
+		i915_gem_context_unreference(submit_req->ctx);
+		kfree(submit_req);
+	}
+
+	/*
 	 * We must free the requests after all the corresponding objects have
 	 * been moved off active lists. Which is the same order as the normal
 	 * retire_requests function does. This is important if object hold
@@ -2567,18 +2595,6 @@ static void i915_gem_reset_ring_cleanup(struct drm_i915_private *dev_priv,
 		i915_gem_free_request(request);
 	}
 
-	while (!list_empty(&ring->execlist_queue)) {
-		struct intel_ctx_submit_request *submit_req;
-
-		submit_req = list_first_entry(&ring->execlist_queue,
-				struct intel_ctx_submit_request,
-				execlist_link);
-		list_del(&submit_req->execlist_link);
-		intel_runtime_pm_put(dev_priv);
-		i915_gem_context_unreference(submit_req->ctx);
-		kfree(submit_req);
-	}
-
 	/* These may not have been flush before the reset, do so now */
 	kfree(ring->preallocated_lazy_request);
 	ring->preallocated_lazy_request = NULL;
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index 5faa084..9fa8e35 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -139,8 +139,6 @@
 #define GEN8_LR_CONTEXT_RENDER_SIZE (20 * PAGE_SIZE)
 #define GEN8_LR_CONTEXT_OTHER_SIZE (2 * PAGE_SIZE)
 
-#define GEN8_LR_CONTEXT_ALIGN 4096
-
 #define RING_EXECLIST_QFULL		(1 << 0x2)
 #define RING_EXECLIST1_VALID		(1 << 0x3)
 #define RING_EXECLIST0_VALID		(1 << 0x4)
@@ -767,16 +765,30 @@ void intel_logical_ring_advance_and_submit(struct intel_ringbuffer *ringbuf)
 static int logical_ring_alloc_seqno(struct intel_engine_cs *ring,
 				    struct intel_context *ctx)
 {
+	int ret;
+
 	if (ring->outstanding_lazy_seqno)
 		return 0;
 
 	if (ring->preallocated_lazy_request == NULL) {
 		struct drm_i915_gem_request *request;
+		struct drm_i915_gem_object *ctx_obj = ctx->engine[ring->id].state;
+		atomic_t *unpin_count = &ctx->engine[ring->id].unpin_count;
 
 		request = kmalloc(sizeof(*request), GFP_KERNEL);
 		if (request == NULL)
 			return -ENOMEM;
 
+		if (atomic_inc_return(unpin_count) == 1 &&
+				ctx != ring->default_context) {
+			ret = i915_gem_obj_ggtt_pin(ctx_obj,
+					GEN8_LR_CONTEXT_ALIGN, 0);
+			if (ret) {
+				kfree(request);
+				return ret;
+			}
+		}
+
 		/* Hold a reference to the context this request belongs to
 		 * (we will need it when the time comes to emit/retire the
 		 * request).
@@ -1610,12 +1622,15 @@ void intel_lr_context_free(struct intel_context *ctx)
 
 	for (i = 0; i < I915_NUM_RINGS; i++) {
 		struct drm_i915_gem_object *ctx_obj = ctx->engine[i].state;
-		struct intel_ringbuffer *ringbuf = ctx->engine[i].ringbuf;
 
 		if (ctx_obj) {
+			struct intel_ringbuffer *ringbuf = ctx->engine[i].ringbuf;
+			struct intel_engine_cs *ring = ringbuf->ring;
+
 			intel_destroy_ringbuffer_obj(ringbuf);
 			kfree(ringbuf);
-			i915_gem_object_ggtt_unpin(ctx_obj);
+			if (ctx == ring->default_context)
+				i915_gem_object_ggtt_unpin(ctx_obj);
 			drm_gem_object_unreference(&ctx_obj->base);
 		}
 	}
@@ -1658,6 +1673,7 @@ static uint32_t get_lr_context_size(struct intel_engine_cs *ring)
 int intel_lr_context_deferred_create(struct intel_context *ctx,
 				     struct intel_engine_cs *ring)
 {
+	const bool is_global_default_ctx = (ctx == ring->default_context);
 	struct drm_device *dev = ring->dev;
 	struct drm_i915_gem_object *ctx_obj;
 	uint32_t context_size;
@@ -1677,18 +1693,21 @@ int intel_lr_context_deferred_create(struct intel_context *ctx,
 		return ret;
 	}
 
-	ret = i915_gem_obj_ggtt_pin(ctx_obj, GEN8_LR_CONTEXT_ALIGN, 0);
-	if (ret) {
-		DRM_DEBUG_DRIVER("Pin LRC backing obj failed: %d\n", ret);
-		drm_gem_object_unreference(&ctx_obj->base);
-		return ret;
+	if (is_global_default_ctx) {
+		ret = i915_gem_obj_ggtt_pin(ctx_obj, GEN8_LR_CONTEXT_ALIGN, 0);
+		if (ret) {
+			DRM_DEBUG_DRIVER("Pin LRC backing obj failed: %d\n", ret);
+			drm_gem_object_unreference(&ctx_obj->base);
+			return ret;
+		}
 	}
 
 	ringbuf = kzalloc(sizeof(*ringbuf), GFP_KERNEL);
 	if (!ringbuf) {
 		DRM_DEBUG_DRIVER("Failed to allocate ringbuffer %s\n",
 				ring->name);
-		i915_gem_object_ggtt_unpin(ctx_obj);
+		if (is_global_default_ctx)
+			i915_gem_object_ggtt_unpin(ctx_obj);
 		drm_gem_object_unreference(&ctx_obj->base);
 		ret = -ENOMEM;
 		return ret;
@@ -1744,7 +1763,8 @@ int intel_lr_context_deferred_create(struct intel_context *ctx,
 
 error:
 	kfree(ringbuf);
-	i915_gem_object_ggtt_unpin(ctx_obj);
+	if (is_global_default_ctx)
+		i915_gem_object_ggtt_unpin(ctx_obj);
 	drm_gem_object_unreference(&ctx_obj->base);
 	return ret;
 }
diff --git a/drivers/gpu/drm/i915/intel_lrc.h b/drivers/gpu/drm/i915/intel_lrc.h
index ac32890..5999e05 100644
--- a/drivers/gpu/drm/i915/intel_lrc.h
+++ b/drivers/gpu/drm/i915/intel_lrc.h
@@ -24,6 +24,8 @@
 #ifndef _INTEL_LRC_H_
 #define _INTEL_LRC_H_
 
+#define GEN8_LR_CONTEXT_ALIGN 4096
+
 /* Execlists regs */
 #define RING_ELSP(ring)			((ring)->mmio_base+0x230)
 #define RING_EXECLIST_STATUS(ring)	((ring)->mmio_base+0x234)
-- 
1.7.9.5