[25/41] drm/i915: Move GEM activity tracking into a common struct reservation_object

Submitted by Chris Wilson on Oct. 14, 2016, 12:18 p.m.

Details

Message ID 20161014121833.439-26-chris@chris-wilson.co.uk
State New
Headers show
Series "Series without cover letter" ( rev: 1 ) in Intel GFX

Browsing this patch as part of:
"Series without cover letter" rev 1 in Intel GFX
<< prev patch [25/41] next patch >>

Commit Message

Chris Wilson Oct. 14, 2016, 12:18 p.m.
In preparation to support many distinct timelines, we need to expand the
activity tracking on the GEM object to handle more than just a request
per engine. We already use the struct reservation_object on the dma-buf
to handle many fence contexts, so integrating that into the GEM object
itself is the preferred solution. (For example, we can now share the same
reservation_object between every consumer/producer using this buffer and
skip the manual import/export via dma-buf.)

v2: Reimplement busy-ioctl (by walking the reservation object), postpone
the ABI change for another day. Similarly use the reservation object to
find the last_write request (if active and from i915) for choosing
display CS flips.

Caveats:

 * busy-ioctl: busy-ioctl only reports on the native fences, it will not
warn of stalls (in set-domain-ioctl, pread/pwrite etc) if the object is
being rendered to by external fences. It also will not report the same
busy state as wait-ioctl (or polling on the dma-buf) in the same
circumstances. On the plus side, it does retain reporting of which
*i915* engines are engaged with this object.

 * non-blocking atomic modesets take a step backwards as the wait for
render completion blocks the ioctl. This is fixed in a subsequent
patch to use a fence instead for awaiting on the rendering, see
"drm/i915: Restore nonblocking awaits for modesetting"

 * dynamic array manipulation for shared-fences in reservation is slower
than the previous lockless static assignment (e.g. gem_exec_lut_handle
runtime on ivb goes from 42s to 66s), mainly due to atomic operations.

 * loss of object-level retirement callbacks, emulated by VMA retirement
tracking.

 * minor loss of object-level last activity information from debugfs,
could be replaced with per-vma information if desired

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_debugfs.c        |  15 +-
 drivers/gpu/drm/i915/i915_drv.h            |  62 +++----
 drivers/gpu/drm/i915/i915_gem.c            | 266 ++++++++---------------------
 drivers/gpu/drm/i915/i915_gem_batch_pool.c |  11 +-
 drivers/gpu/drm/i915/i915_gem_dmabuf.c     |  53 +-----
 drivers/gpu/drm/i915/i915_gem_dmabuf.h     |  45 -----
 drivers/gpu/drm/i915/i915_gem_execbuffer.c |  82 +++------
 drivers/gpu/drm/i915/i915_gem_gtt.c        |  32 ++++
 drivers/gpu/drm/i915/i915_gem_gtt.h        |   1 +
 drivers/gpu/drm/i915/i915_gem_request.c    |  48 +++---
 drivers/gpu/drm/i915/i915_gem_request.h    |  37 +---
 drivers/gpu/drm/i915/i915_gpu_error.c      |   6 +-
 drivers/gpu/drm/i915/intel_atomic_plane.c  |   2 -
 drivers/gpu/drm/i915/intel_display.c       | 114 +++----------
 drivers/gpu/drm/i915/intel_drv.h           |   3 -
 15 files changed, 223 insertions(+), 554 deletions(-)
 delete mode 100644 drivers/gpu/drm/i915/i915_gem_dmabuf.h

Patch hide | download patch | download mbox

diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
index 27fd5370f0cc..3fcdfff273e4 100644
--- a/drivers/gpu/drm/i915/i915_debugfs.c
+++ b/drivers/gpu/drm/i915/i915_debugfs.c
@@ -136,11 +136,10 @@  describe_obj(struct seq_file *m, struct drm_i915_gem_object *obj)
 	struct i915_vma *vma;
 	unsigned int frontbuffer_bits;
 	int pin_count = 0;
-	enum intel_engine_id id;
 
 	lockdep_assert_held(&obj->base.dev->struct_mutex);
 
-	seq_printf(m, "%pK: %c%c%c%c%c %8zdKiB %02x %02x [ ",
+	seq_printf(m, "%pK: %c%c%c%c%c %8zdKiB %02x %02x %s%s%s",
 		   &obj->base,
 		   get_active_flag(obj),
 		   get_pin_flag(obj),
@@ -149,14 +148,7 @@  describe_obj(struct seq_file *m, struct drm_i915_gem_object *obj)
 		   get_pin_mapped_flag(obj),
 		   obj->base.size / 1024,
 		   obj->base.read_domains,
-		   obj->base.write_domain);
-	for_each_engine(engine, dev_priv, id)
-		seq_printf(m, "%x ",
-			   i915_gem_active_get_seqno(&obj->last_read[id],
-						     &obj->base.dev->struct_mutex));
-	seq_printf(m, "] %x %s%s%s",
-		   i915_gem_active_get_seqno(&obj->last_write,
-					     &obj->base.dev->struct_mutex),
+		   obj->base.write_domain,
 		   i915_cache_level_str(dev_priv, obj->cache_level),
 		   obj->mm.dirty ? " dirty" : "",
 		   obj->mm.madv == I915_MADV_DONTNEED ? " purgeable" : "");
@@ -187,8 +179,7 @@  describe_obj(struct seq_file *m, struct drm_i915_gem_object *obj)
 	if (obj->stolen)
 		seq_printf(m, " (stolen: %08llx)", obj->stolen->start);
 
-	engine = i915_gem_active_get_engine(&obj->last_write,
-					    &dev_priv->drm.struct_mutex);
+	engine = i915_gem_object_last_write_engine(obj);
 	if (engine)
 		seq_printf(m, " (%s)", engine->name);
 
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index b6192aca57ec..3245280c8b6c 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -41,6 +41,7 @@ 
 #include <linux/intel-iommu.h>
 #include <linux/kref.h>
 #include <linux/pm_qos.h>
+#include <linux/reservation.h>
 #include <linux/shmem_fs.h>
 
 #include <drm/drmP.h>
@@ -2233,21 +2234,12 @@  struct drm_i915_gem_object {
 	struct list_head batch_pool_link;
 
 	unsigned long flags;
-	/**
-	 * This is set if the object is on the active lists (has pending
-	 * rendering and so a non-zero seqno), and is not set if it i s on
-	 * inactive (ready to be unbound) list.
-	 */
-#define I915_BO_ACTIVE_SHIFT 0
-#define I915_BO_ACTIVE_MASK ((1 << I915_NUM_ENGINES) - 1)
-#define __I915_BO_ACTIVE(bo) \
-	((READ_ONCE((bo)->flags) >> I915_BO_ACTIVE_SHIFT) & I915_BO_ACTIVE_MASK)
 
 	/**
 	 * Have we taken a reference for the object for incomplete GPU
 	 * activity?
 	 */
-#define I915_BO_ACTIVE_REF (I915_BO_ACTIVE_SHIFT + I915_NUM_ENGINES)
+#define I915_BO_ACTIVE_REF 0
 
 	/*
 	 * Is the object to be mapped as read-only to the GPU
@@ -2268,6 +2260,7 @@  struct drm_i915_gem_object {
 
 	/** Count of VMA actually bound by this object */
 	unsigned int bind_count;
+	unsigned int active_count;
 	unsigned int pin_display;
 
 	struct {
@@ -2307,8 +2300,7 @@  struct drm_i915_gem_object {
 	 * read request. This allows for the CPU to read from an active
 	 * buffer by only waiting for the write to complete.
 	 */
-	struct i915_gem_active last_read[I915_NUM_ENGINES];
-	struct i915_gem_active last_write;
+	struct reservation_object *resv;
 
 	/** References from framebuffers, locks out tiling changes. */
 	unsigned long framebuffer_references;
@@ -2327,6 +2319,8 @@  struct drm_i915_gem_object {
 
 	/** for phys allocated objects */
 	struct drm_dma_handle *phys_handle;
+
+	struct reservation_object __builtin_resv;
 };
 
 static inline struct drm_i915_gem_object *
@@ -2412,35 +2406,10 @@  i915_gem_object_has_struct_page(const struct drm_i915_gem_object *obj)
 	return obj->ops->flags & I915_GEM_OBJECT_HAS_STRUCT_PAGE;
 }
 
-static inline unsigned long
-i915_gem_object_get_active(const struct drm_i915_gem_object *obj)
-{
-	return (obj->flags >> I915_BO_ACTIVE_SHIFT) & I915_BO_ACTIVE_MASK;
-}
-
 static inline bool
 i915_gem_object_is_active(const struct drm_i915_gem_object *obj)
 {
-	return i915_gem_object_get_active(obj);
-}
-
-static inline void
-i915_gem_object_set_active(struct drm_i915_gem_object *obj, int engine)
-{
-	obj->flags |= BIT(engine + I915_BO_ACTIVE_SHIFT);
-}
-
-static inline void
-i915_gem_object_clear_active(struct drm_i915_gem_object *obj, int engine)
-{
-	obj->flags &= ~BIT(engine + I915_BO_ACTIVE_SHIFT);
-}
-
-static inline bool
-i915_gem_object_has_active_engine(const struct drm_i915_gem_object *obj,
-				  int engine)
-{
-	return obj->flags & BIT(engine + I915_BO_ACTIVE_SHIFT);
+	return obj->active_count;
 }
 
 static inline bool
@@ -2483,6 +2452,23 @@  i915_gem_object_get_stride(struct drm_i915_gem_object *obj)
 	return obj->tiling_and_stride & STRIDE_MASK;
 }
 
+static inline struct intel_engine_cs *
+i915_gem_object_last_write_engine(struct drm_i915_gem_object *obj)
+{
+	struct intel_engine_cs *engine = NULL;
+	struct fence *fence;
+
+	rcu_read_lock();
+	fence = reservation_object_get_excl_rcu(obj->resv);
+	rcu_read_unlock();
+
+	if (fence && fence_is_i915(fence) && !fence_is_signaled(fence))
+		engine = to_request(fence)->engine;
+	fence_put(fence);
+
+	return engine;
+}
+
 static inline struct i915_vma *i915_vma_get(struct i915_vma *vma)
 {
 	i915_gem_object_get(vma->obj);
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index a3e39f412a87..5f6f32ac2a64 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -29,7 +29,6 @@ 
 #include <drm/drm_vma_manager.h>
 #include <drm/i915_drm.h>
 #include "i915_drv.h"
-#include "i915_gem_dmabuf.h"
 #include "i915_vgpu.h"
 #include "i915_trace.h"
 #include "intel_drv.h"
@@ -447,11 +446,6 @@  i915_gem_object_wait(struct drm_i915_gem_object *obj,
 		     long timeout,
 		     struct intel_rps_client *rps)
 {
-	struct reservation_object *resv;
-	struct i915_gem_active *active;
-	unsigned long active_mask;
-	int idx;
-
 	might_sleep();
 #if IS_ENABLED(CONFIG_LOCKDEP)
 	GEM_BUG_ON(!!lockdep_is_held(&obj->base.dev->struct_mutex) !=
@@ -459,33 +453,9 @@  i915_gem_object_wait(struct drm_i915_gem_object *obj,
 #endif
 	GEM_BUG_ON(timeout < 0);
 
-	if (flags & I915_WAIT_ALL) {
-		active = obj->last_read;
-		active_mask = i915_gem_object_get_active(obj);
-	} else {
-		active_mask = 1;
-		active = &obj->last_write;
-	}
-
-	for_each_active(active_mask, idx) {
-		struct drm_i915_gem_request *request;
-
-		request = i915_gem_active_get_unlocked(&active[idx]);
-		if (request) {
-			timeout = i915_gem_object_wait_fence(&request->fence,
-							     flags, timeout,
-							     rps);
-			i915_gem_request_put(request);
-		}
-		if (timeout < 0)
-			return timeout;
-	}
-
-	resv = i915_gem_object_get_dmabuf_resv(obj);
-	if (resv)
-		timeout = i915_gem_object_wait_reservation(resv,
-							   flags, timeout,
-							   rps);
+	timeout = i915_gem_object_wait_reservation(obj->resv,
+						   flags, timeout,
+						   rps);
 	return timeout < 0 ? timeout : 0;
 }
 
@@ -2561,44 +2531,6 @@  err_unlock:
 	goto out_unlock;
 }
 
-static void
-i915_gem_object_retire__write(struct i915_gem_active *active,
-			      struct drm_i915_gem_request *request)
-{
-	struct drm_i915_gem_object *obj =
-		container_of(active, struct drm_i915_gem_object, last_write);
-
-	intel_fb_obj_flush(obj, true, ORIGIN_CS);
-}
-
-static void
-i915_gem_object_retire__read(struct i915_gem_active *active,
-			     struct drm_i915_gem_request *request)
-{
-	int idx = request->engine->id;
-	struct drm_i915_gem_object *obj =
-		container_of(active, struct drm_i915_gem_object, last_read[idx]);
-
-	GEM_BUG_ON(!i915_gem_object_has_active_engine(obj, idx));
-
-	i915_gem_object_clear_active(obj, idx);
-	if (i915_gem_object_is_active(obj))
-		return;
-
-	/* Bump our place on the bound list to keep it roughly in LRU order
-	 * so that we don't steal from recently used but inactive objects
-	 * (unless we are forced to ofc!)
-	 */
-	if (obj->bind_count)
-		list_move_tail(&obj->global_list,
-			       &request->i915->mm.bound_list);
-
-	if (i915_gem_object_has_active_reference(obj)) {
-		i915_gem_object_clear_active_reference(obj);
-		i915_gem_object_put(obj);
-	}
-}
-
 static bool i915_context_is_banned(const struct i915_gem_context *ctx)
 {
 	unsigned long elapsed;
@@ -2978,6 +2910,13 @@  int i915_vma_unbind(struct i915_vma *vma)
 		 * In order to prevent it from being recursively closed,
 		 * take a pin on the vma so that the second unbind is
 		 * aborted.
+		 *
+		 * Even more scary is that the retire callback may free
+		 * the object (last active vma). To prevent the explosion
+		 * we defer the actual object free to a worker that can
+		 * only proceed once it acquires the struct_mutex (which
+		 * we currently hold, therefore it cannot free this object
+		 * before we are finished).
 		 */
 		__i915_vma_pin(vma);
 
@@ -4020,83 +3959,42 @@  static __always_inline unsigned int __busy_write_id(unsigned int id)
 }
 
 static __always_inline unsigned int
-__busy_set_if_active(const struct i915_gem_active *active,
+__busy_set_if_active(const struct fence *fence,
 		     unsigned int (*flag)(unsigned int id))
 {
-	struct drm_i915_gem_request *request;
-
-	request = rcu_dereference(active->request);
-	if (!request || i915_gem_request_completed(request))
-		return 0;
+	struct drm_i915_gem_request *rq;
 
-	/* This is racy. See __i915_gem_active_get_rcu() for an in detail
-	 * discussion of how to handle the race correctly, but for reporting
-	 * the busy state we err on the side of potentially reporting the
-	 * wrong engine as being busy (but we guarantee that the result
-	 * is at least self-consistent).
-	 *
-	 * As we use SLAB_DESTROY_BY_RCU, the request may be reallocated
-	 * whilst we are inspecting it, even under the RCU read lock as we are.
-	 * This means that there is a small window for the engine and/or the
-	 * seqno to have been overwritten. The seqno will always be in the
-	 * future compared to the intended, and so we know that if that
-	 * seqno is idle (on whatever engine) our request is idle and the
-	 * return 0 above is correct.
-	 *
-	 * The issue is that if the engine is switched, it is just as likely
-	 * to report that it is busy (but since the switch happened, we know
-	 * the request should be idle). So there is a small chance that a busy
-	 * result is actually the wrong engine.
+	/* We have to check the current hw status of the fence as the uABI
+	 * guarantees forward progress. We could rely on the idle worker
+	 * to eventually flush us, but to minimise latency just ask the
+	 * hardware.
 	 *
-	 * So why don't we care?
-	 *
-	 * For starters, the busy ioctl is a heuristic that is by definition
-	 * racy. Even with perfect serialisation in the driver, the hardware
-	 * state is constantly advancing - the state we report to the user
-	 * is stale.
-	 *
-	 * The critical information for the busy-ioctl is whether the object
-	 * is idle as userspace relies on that to detect whether its next
-	 * access will stall, or if it has missed submitting commands to
-	 * the hardware allowing the GPU to stall. We never generate a
-	 * false-positive for idleness, thus busy-ioctl is reliable at the
-	 * most fundamental level, and we maintain the guarantee that a
-	 * busy object left to itself will eventually become idle (and stay
-	 * idle!).
-	 *
-	 * We allow ourselves the leeway of potentially misreporting the busy
-	 * state because that is an optimisation heuristic that is constantly
-	 * in flux. Being quickly able to detect the busy/idle state is much
-	 * more important than accurate logging of exactly which engines were
-	 * busy.
-	 *
-	 * For accuracy in reporting the engine, we could use
-	 *
-	 *	result = 0;
-	 *	request = __i915_gem_active_get_rcu(active);
-	 *	if (request) {
-	 *		if (!i915_gem_request_completed(request))
-	 *			result = flag(request->engine->exec_id);
-	 *		i915_gem_request_put(request);
-	 *	}
-	 *
-	 * but that still remains susceptible to both hardware and userspace
-	 * races. So we accept making the result of that race slightly worse,
-	 * given the rarity of the race and its low impact on the result.
+	 * Note we only report on the status of native fences.
 	 */
-	return flag(READ_ONCE(request->engine->exec_id));
+	if (!fence_is_i915(fence))
+		return 0;
+
+	/* opencode to_request() in order to avoid const warnings */
+	rq = container_of(fence, struct drm_i915_gem_request, fence);
+	if (i915_gem_request_completed(rq))
+		return 0;
+
+	return flag(rq->engine->exec_id);
 }
 
 static __always_inline unsigned int
-busy_check_reader(const struct i915_gem_active *active)
+busy_check_reader(const struct fence *fence)
 {
-	return __busy_set_if_active(active, __busy_read_flag);
+	return __busy_set_if_active(fence, __busy_read_flag);
 }
 
 static __always_inline unsigned int
-busy_check_writer(const struct i915_gem_active *active)
+busy_check_writer(const struct fence *fence)
 {
-	return __busy_set_if_active(active, __busy_write_id);
+	if (!fence)
+		return 0;
+
+	return __busy_set_if_active(fence, __busy_write_id);
 }
 
 int
@@ -4105,63 +4003,52 @@  i915_gem_busy_ioctl(struct drm_device *dev, void *data,
 {
 	struct drm_i915_gem_busy *args = data;
 	struct drm_i915_gem_object *obj;
-	unsigned long active;
+	struct reservation_object_list *list;
+	unsigned int seq;
 	int err;
 
+	err = -ENOENT;
 	rcu_read_lock();
 	obj = i915_gem_object_lookup_rcu(file, args->handle);
-	if (!obj) {
-		err = -ENOENT;
+	if (!obj)
 		goto out;
-	}
 
-	args->busy = 0;
-	active = __I915_BO_ACTIVE(obj);
-	if (active) {
-		int idx;
-
-		/* Yes, the lookups are intentionally racy.
-		 *
-		 * First, we cannot simply rely on __I915_BO_ACTIVE. We have
-		 * to regard the value as stale and as our ABI guarantees
-		 * forward progress, we confirm the status of each active
-		 * request with the hardware.
-		 *
-		 * Even though we guard the pointer lookup by RCU, that only
-		 * guarantees that the pointer and its contents remain
-		 * dereferencable and does *not* mean that the request we
-		 * have is the same as the one being tracked by the object.
-		 *
-		 * Consider that we lookup the request just as it is being
-		 * retired and freed. We take a local copy of the pointer,
-		 * but before we add its engine into the busy set, the other
-		 * thread reallocates it and assigns it to a task on another
-		 * engine with a fresh and incomplete seqno. Guarding against
-		 * that requires careful serialisation and reference counting,
-		 * i.e. using __i915_gem_active_get_request_rcu(). We don't,
-		 * instead we expect that if the result is busy, which engines
-		 * are busy is not completely reliable - we only guarantee
-		 * that the object was busy.
-		 */
-
-		for_each_active(active, idx)
-			args->busy |= busy_check_reader(&obj->last_read[idx]);
-
-		/* For ABI sanity, we only care that the write engine is in
-		 * the set of read engines. This should be ensured by the
-		 * ordering of setting last_read/last_write in
-		 * i915_vma_move_to_active(), and then in reverse in retire.
-		 * However, for good measure, we always report the last_write
-		 * request as a busy read as well as being a busy write.
-		 *
-		 * We don't care that the set of active read/write engines
-		 * may change during construction of the result, as it is
-		 * equally liable to change before userspace can inspect
-		 * the result.
-		 */
-		args->busy |= busy_check_writer(&obj->last_write);
+	/* A discrepancy here is that we do not report the status of
+	 * non-i915 fences, i.e. even though we may report the object as idle,
+	 * a call to set-domain may still stall waiting for foreign rendering.
+	 * This also means that wait-ioctl may report an object as busy,
+	 * where busy-ioctl considers it idle.
+	 *
+	 * We trade the ability to warn of foreign fences to report on which
+	 * i915 engines are active for the object.
+	 *
+	 * Alternatively, we can trade that extra information on read/write
+	 * activity with
+	 *	args->busy =
+	 *		!reservation_object_test_signaled_rcu(obj->resv, true);
+	 * to report the overall busyness. This is what the wait-ioctl does.
+	 *
+	 */
+retry:
+	seq = raw_read_seqcount(&obj->resv->seq);
+
+	/* Translate the exclusive fence to the READ *and* WRITE engine */
+	args->busy = busy_check_writer(rcu_dereference(obj->resv->fence_excl));
+
+	/* Translate shared fences to READ set of engines */
+	list = rcu_dereference(obj->resv->fence);
+	if (list) {
+		unsigned int shared_count = list->shared_count, i;
+		for (i = 0; i < shared_count; ++i) {
+			struct fence *fence = rcu_dereference(list->shared[i]);
+			args->busy |= busy_check_reader(fence);
+		}
 	}
 
+	if (args->busy && read_seqcount_retry(&obj->resv->seq, seq))
+		goto retry;
+
+	err = 0;
 out:
 	rcu_read_unlock();
 	return err;
@@ -4226,23 +4113,19 @@  out:
 void i915_gem_object_init(struct drm_i915_gem_object *obj,
 			  const struct drm_i915_gem_object_ops *ops)
 {
-	int i;
-
 	mutex_init(&obj->mm.lock);
 
 	INIT_LIST_HEAD(&obj->global_list);
 	INIT_LIST_HEAD(&obj->userfault_link);
-	for (i = 0; i < I915_NUM_ENGINES; i++)
-		init_request_active(&obj->last_read[i],
-				    i915_gem_object_retire__read);
-	init_request_active(&obj->last_write,
-			    i915_gem_object_retire__write);
 	INIT_LIST_HEAD(&obj->obj_exec_link);
 	INIT_LIST_HEAD(&obj->vma_list);
 	INIT_LIST_HEAD(&obj->batch_pool_link);
 
 	obj->ops = ops;
 
+	reservation_object_init(&obj->__builtin_resv);
+	obj->resv = &obj->__builtin_resv;
+
 	obj->frontbuffer_ggtt_origin = ORIGIN_GTT;
 
 	obj->mm.madv = I915_MADV_WILLNEED;
@@ -4380,6 +4263,7 @@  static void __i915_gem_free_objects(struct drm_i915_private *i915,
 		if (obj->base.import_attach)
 			drm_prime_gem_destroy(&obj->base, NULL);
 
+		reservation_object_fini(&obj->__builtin_resv);
 		drm_gem_object_release(&obj->base);
 		i915_gem_info_remove_obj(i915, obj->base.size);
 
diff --git a/drivers/gpu/drm/i915/i915_gem_batch_pool.c b/drivers/gpu/drm/i915/i915_gem_batch_pool.c
index e0f38e5c0fbb..b3bc119ec1bb 100644
--- a/drivers/gpu/drm/i915/i915_gem_batch_pool.c
+++ b/drivers/gpu/drm/i915/i915_gem_batch_pool.c
@@ -114,11 +114,18 @@  i915_gem_batch_pool_get(struct i915_gem_batch_pool *pool,
 
 	list_for_each_entry(tmp, list, batch_pool_link) {
 		/* The batches are strictly LRU ordered */
-		if (!i915_gem_active_is_idle(&tmp->last_read[pool->engine->id],
-					     &tmp->base.dev->struct_mutex))
+		if (i915_gem_object_is_active(tmp))
 			break;
 
+		GEM_BUG_ON(!reservation_object_test_signaled_rcu(tmp->resv,
+								 true));
+
 		if (tmp->base.size >= size) {
+			/* Clear the set of shared fences early */
+			ww_mutex_lock(&tmp->resv->lock, NULL);
+			reservation_object_add_excl_fence(tmp->resv, NULL);
+			ww_mutex_unlock(&tmp->resv->lock);
+
 			obj = tmp;
 			break;
 		}
diff --git a/drivers/gpu/drm/i915/i915_gem_dmabuf.c b/drivers/gpu/drm/i915/i915_gem_dmabuf.c
index 4d45f20d11ed..5e38299b5df6 100644
--- a/drivers/gpu/drm/i915/i915_gem_dmabuf.c
+++ b/drivers/gpu/drm/i915/i915_gem_dmabuf.c
@@ -211,60 +211,17 @@  static const struct dma_buf_ops i915_dmabuf_ops =  {
 	.end_cpu_access = i915_gem_end_cpu_access,
 };
 
-static void export_fences(struct drm_i915_gem_object *obj,
-			  struct dma_buf *dma_buf)
-{
-	struct reservation_object *resv = dma_buf->resv;
-	struct drm_i915_gem_request *req;
-	unsigned long active;
-	int idx;
-
-	active = __I915_BO_ACTIVE(obj);
-	if (!active)
-		return;
-
-	/* Serialise with execbuf to prevent concurrent fence-loops */
-	mutex_lock(&obj->base.dev->struct_mutex);
-
-	/* Mark the object for future fences before racily adding old fences */
-	obj->base.dma_buf = dma_buf;
-
-	ww_mutex_lock(&resv->lock, NULL);
-
-	for_each_active(active, idx) {
-		req = i915_gem_active_get(&obj->last_read[idx],
-					  &obj->base.dev->struct_mutex);
-		if (!req)
-			continue;
-
-		if (reservation_object_reserve_shared(resv) == 0)
-			reservation_object_add_shared_fence(resv, &req->fence);
-
-		i915_gem_request_put(req);
-	}
-
-	req = i915_gem_active_get(&obj->last_write,
-				  &obj->base.dev->struct_mutex);
-	if (req) {
-		reservation_object_add_excl_fence(resv, &req->fence);
-		i915_gem_request_put(req);
-	}
-
-	ww_mutex_unlock(&resv->lock);
-	mutex_unlock(&obj->base.dev->struct_mutex);
-}
-
 struct dma_buf *i915_gem_prime_export(struct drm_device *dev,
 				      struct drm_gem_object *gem_obj, int flags)
 {
 	struct drm_i915_gem_object *obj = to_intel_bo(gem_obj);
 	DEFINE_DMA_BUF_EXPORT_INFO(exp_info);
-	struct dma_buf *dma_buf;
 
 	exp_info.ops = &i915_dmabuf_ops;
 	exp_info.size = gem_obj->size;
 	exp_info.flags = flags;
 	exp_info.priv = gem_obj;
+	exp_info.resv = obj->resv;
 
 	if (obj->ops->dmabuf_export) {
 		int ret = obj->ops->dmabuf_export(obj);
@@ -272,12 +229,7 @@  struct dma_buf *i915_gem_prime_export(struct drm_device *dev,
 			return ERR_PTR(ret);
 	}
 
-	dma_buf = drm_gem_dmabuf_export(dev, &exp_info);
-	if (IS_ERR(dma_buf))
-		return dma_buf;
-
-	export_fences(obj, dma_buf);
-	return dma_buf;
+	return drm_gem_dmabuf_export(dev, &exp_info);
 }
 
 static struct sg_table *
@@ -335,6 +287,7 @@  struct drm_gem_object *i915_gem_prime_import(struct drm_device *dev,
 	drm_gem_private_object_init(dev, &obj->base, dma_buf->size);
 	i915_gem_object_init(obj, &i915_gem_object_dmabuf_ops);
 	obj->base.import_attach = attach;
+	obj->resv = dma_buf->resv;
 
 	/* We use GTT as shorthand for a coherent domain, one that is
 	 * neither in the GPU cache nor in the CPU cache, where all
diff --git a/drivers/gpu/drm/i915/i915_gem_dmabuf.h b/drivers/gpu/drm/i915/i915_gem_dmabuf.h
deleted file mode 100644
index 91315557e421..000000000000
--- a/drivers/gpu/drm/i915/i915_gem_dmabuf.h
+++ /dev/null
@@ -1,45 +0,0 @@ 
-/*
- * Copyright 2016 Intel Corporation
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice (including the next
- * paragraph) shall be included in all copies or substantial portions of the
- * Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
- * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
- * DEALINGS IN THE SOFTWARE.
- *
- */
-
-#ifndef _I915_GEM_DMABUF_H_
-#define _I915_GEM_DMABUF_H_
-
-#include <linux/dma-buf.h>
-
-static inline struct reservation_object *
-i915_gem_object_get_dmabuf_resv(struct drm_i915_gem_object *obj)
-{
-	struct dma_buf *dma_buf;
-
-	if (obj->base.dma_buf)
-		dma_buf = obj->base.dma_buf;
-	else if (obj->base.import_attach)
-		dma_buf = obj->base.import_attach->dmabuf;
-	else
-		return NULL;
-
-	return dma_buf->resv;
-}
-
-#endif
diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
index d6214b99bbf5..9b67df033ff8 100644
--- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
@@ -34,7 +34,6 @@ 
 #include <drm/i915_drm.h>
 
 #include "i915_drv.h"
-#include "i915_gem_dmabuf.h"
 #include "i915_trace.h"
 #include "intel_drv.h"
 #include "intel_frontbuffer.h"
@@ -390,6 +389,11 @@  static void *reloc_kmap(struct drm_i915_gem_object *obj,
 		unsigned int flushes;
 		int ret;
 
+		/* We can't wait for rendering with pagefaults disabled */
+		if (pagefault_disabled() &&
+		    !reservation_object_test_signaled_rcu(obj->resv, true))
+			return ERR_PTR(-EFAULT);
+
 		ret = i915_gem_obj_prepare_shmem_write(obj, &flushes);
 		if (ret)
 			return ERR_PTR(ret);
@@ -436,6 +440,11 @@  static void *reloc_iomap(struct drm_i915_gem_object *obj,
 		if (use_cpu_reloc(obj))
 			return NULL;
 
+		/* We can't wait for rendering with pagefaults disabled */
+		if (pagefault_disabled() &&
+		    !reservation_object_test_signaled_rcu(obj->resv, true))
+			return ERR_PTR(-EFAULT);
+
 		ret = i915_gem_object_set_to_gtt_domain(obj, true);
 		if (ret)
 			return ERR_PTR(ret);
@@ -551,20 +560,6 @@  repeat:
 	return 0;
 }
 
-static bool object_is_idle(struct drm_i915_gem_object *obj)
-{
-	unsigned long active = i915_gem_object_get_active(obj);
-	int idx;
-
-	for_each_active(active, idx) {
-		if (!i915_gem_active_is_idle(&obj->last_read[idx],
-					     &obj->base.dev->struct_mutex))
-			return false;
-	}
-
-	return true;
-}
-
 static int
 i915_gem_execbuffer_relocate_entry(struct drm_i915_gem_object *obj,
 				   struct eb_vmas *eb,
@@ -648,10 +643,6 @@  i915_gem_execbuffer_relocate_entry(struct drm_i915_gem_object *obj,
 		return -EINVAL;
 	}
 
-	/* We can't wait for rendering with pagefaults disabled */
-	if (pagefault_disabled() && !object_is_idle(obj))
-		return -EFAULT;
-
 	ret = relocate_entry(obj, reloc, cache, target_offset);
 	if (ret)
 		return ret;
@@ -1110,44 +1101,20 @@  err:
 	return ret;
 }
 
-static unsigned int eb_other_engines(struct drm_i915_gem_request *req)
-{
-	unsigned int mask;
-
-	mask = ~intel_engine_flag(req->engine) & I915_BO_ACTIVE_MASK;
-	mask <<= I915_BO_ACTIVE_SHIFT;
-
-	return mask;
-}
-
 static int
 i915_gem_execbuffer_move_to_gpu(struct drm_i915_gem_request *req,
 				struct list_head *vmas)
 {
-	const unsigned int other_rings = eb_other_engines(req);
 	struct i915_vma *vma;
 	int ret;
 
 	list_for_each_entry(vma, vmas, exec_list) {
 		struct drm_i915_gem_object *obj = vma->obj;
-		struct reservation_object *resv;
 
-		if (obj->flags & other_rings) {
-			ret = i915_gem_request_await_object
-				(req, obj, obj->base.pending_write_domain);
-			if (ret)
-				return ret;
-		}
-
-		resv = i915_gem_object_get_dmabuf_resv(obj);
-		if (resv) {
-			ret = i915_sw_fence_await_reservation
-				(&req->submit, resv, &i915_fence_ops,
-				 obj->base.pending_write_domain, 10*HZ,
-				 GFP_KERNEL | __GFP_NOWARN);
-			if (ret < 0)
-				return ret;
-		}
+		ret = i915_gem_request_await_object
+			(req, obj, obj->base.pending_write_domain);
+		if (ret)
+			return ret;
 
 		if (obj->base.write_domain & I915_GEM_DOMAIN_CPU)
 			i915_gem_clflush_object(obj, false);
@@ -1289,8 +1256,6 @@  void i915_vma_move_to_active(struct i915_vma *vma,
 
 	GEM_BUG_ON(!drm_mm_node_allocated(&vma->node));
 
-	obj->mm.dirty = true; /* be paranoid  */
-
 	/* Add a reference if we're newly entering the active list.
 	 * The order in which we add operations to the retirement queue is
 	 * vital here: mark_active adds to the start of the callback list,
@@ -1298,11 +1263,14 @@  void i915_vma_move_to_active(struct i915_vma *vma,
 	 * add the active reference first and queue for it to be dropped
 	 * *last*.
 	 */
-	i915_gem_object_set_active(obj, idx);
-	i915_gem_active_set(&obj->last_read[idx], req);
+	if (!i915_vma_is_active(vma))
+		obj->active_count++;
+	i915_vma_set_active(vma, idx);
+	i915_gem_active_set(&vma->last_read[idx], req);
+	list_move_tail(&vma->vm_link, &vma->vm->active_list);
 
 	if (flags & EXEC_OBJECT_WRITE) {
-		i915_gem_active_set(&obj->last_write, req);
+		i915_gem_active_set(&vma->last_write, req);
 
 		intel_fb_obj_invalidate(obj, ORIGIN_CS);
 
@@ -1312,21 +1280,13 @@  void i915_vma_move_to_active(struct i915_vma *vma,
 
 	if (flags & EXEC_OBJECT_NEEDS_FENCE)
 		i915_gem_active_set(&vma->last_fence, req);
-
-	i915_vma_set_active(vma, idx);
-	i915_gem_active_set(&vma->last_read[idx], req);
-	list_move_tail(&vma->vm_link, &vma->vm->active_list);
 }
 
 static void eb_export_fence(struct drm_i915_gem_object *obj,
 			    struct drm_i915_gem_request *req,
 			    unsigned int flags)
 {
-	struct reservation_object *resv;
-
-	resv = i915_gem_object_get_dmabuf_resv(obj);
-	if (!resv)
-		return;
+	struct reservation_object *resv = obj->resv;
 
 	/* Ignore errors from failing to allocate the new fence, we can't
 	 * handle an error right now. Worst case should be missed
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index 1008209ca797..1b1a459e2b68 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -31,6 +31,7 @@ 
 #include "i915_vgpu.h"
 #include "i915_trace.h"
 #include "intel_drv.h"
+#include "intel_frontbuffer.h"
 
 #define I915_GFP_DMA (GFP_KERNEL | __GFP_HIGHMEM)
 
@@ -3343,6 +3344,7 @@  i915_vma_retire(struct i915_gem_active *active,
 	const unsigned int idx = rq->engine->id;
 	struct i915_vma *vma =
 		container_of(active, struct i915_vma, last_read[idx]);
+	struct drm_i915_gem_object *obj = vma->obj;
 
 	GEM_BUG_ON(!i915_vma_has_active_engine(vma, idx));
 
@@ -3353,6 +3355,34 @@  i915_vma_retire(struct i915_gem_active *active,
 	list_move_tail(&vma->vm_link, &vma->vm->inactive_list);
 	if (unlikely(i915_vma_is_closed(vma) && !i915_vma_is_pinned(vma)))
 		WARN_ON(i915_vma_unbind(vma));
+
+	GEM_BUG_ON(!i915_gem_object_is_active(obj));
+	if (--obj->active_count)
+		return;
+
+	/* Bump our place on the bound list to keep it roughly in LRU order
+	 * so that we don't steal from recently used but inactive objects
+	 * (unless we are forced to ofc!)
+	 */
+	if (obj->bind_count)
+		list_move_tail(&obj->global_list, &rq->i915->mm.bound_list);
+
+	obj->mm.dirty = true; /* be paranoid  */
+
+	if (i915_gem_object_has_active_reference(obj)) {
+		i915_gem_object_clear_active_reference(obj);
+		i915_gem_object_put(obj);
+	}
+}
+
+static void
+i915_ggtt_retire__write(struct i915_gem_active *active,
+			struct drm_i915_gem_request *request)
+{
+	struct i915_vma *vma =
+		container_of(active, struct i915_vma, last_write);
+
+	intel_fb_obj_flush(vma->obj, true, ORIGIN_CS);
 }
 
 void i915_vma_destroy(struct i915_vma *vma)
@@ -3396,6 +3426,8 @@  __i915_vma_create(struct drm_i915_gem_object *obj,
 	INIT_LIST_HEAD(&vma->exec_list);
 	for (i = 0; i < ARRAY_SIZE(vma->last_read); i++)
 		init_request_active(&vma->last_read[i], i915_vma_retire);
+	init_request_active(&vma->last_write,
+			    i915_is_ggtt(vm) ? i915_ggtt_retire__write : NULL);
 	init_request_active(&vma->last_fence, NULL);
 	list_add(&vma->vm_link, &vm->unbound_list);
 	vma->vm = vm;
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.h b/drivers/gpu/drm/i915/i915_gem_gtt.h
index dbe6a6cec20d..9f0327e5176a 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.h
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.h
@@ -211,6 +211,7 @@  struct i915_vma {
 
 	unsigned int active;
 	struct i915_gem_active last_read[I915_NUM_ENGINES];
+	struct i915_gem_active last_write;
 	struct i915_gem_active last_fence;
 
 	/**
diff --git a/drivers/gpu/drm/i915/i915_gem_request.c b/drivers/gpu/drm/i915/i915_gem_request.c
index f09b8e3a8276..225c4c684ac4 100644
--- a/drivers/gpu/drm/i915/i915_gem_request.c
+++ b/drivers/gpu/drm/i915/i915_gem_request.c
@@ -196,6 +196,8 @@  static void i915_gem_request_retire(struct drm_i915_gem_request *request)
 	}
 
 	i915_gem_context_put(request->ctx);
+
+	fence_signal(&request->fence);
 	i915_gem_request_put(request);
 }
 
@@ -553,33 +555,41 @@  i915_gem_request_await_object(struct drm_i915_gem_request *to,
 			      struct drm_i915_gem_object *obj,
 			      bool write)
 {
-	struct i915_gem_active *active;
-	unsigned long active_mask;
-	int idx;
+	struct fence *excl;
+	int ret = 0;
 
 	if (write) {
-		active_mask = i915_gem_object_get_active(obj);
-		active = obj->last_read;
+		struct fence **shared;
+		unsigned int count, i;
+
+		ret = reservation_object_get_fences_rcu(obj->resv,
+							&excl, &count, &shared);
+		if (ret)
+			return ret;
+
+		for (i = 0; i < count; i++) {
+			ret = i915_gem_request_await_fence(to, shared[i]);
+			if (ret)
+				break;
+
+			fence_put(shared[i]);
+		}
+
+		for (; i < count; i++)
+			fence_put(shared[i]);
+		kfree(shared);
 	} else {
-		active_mask = 1;
-		active = &obj->last_write;
+		excl = reservation_object_get_excl_rcu(obj->resv);
 	}
 
-	for_each_active(active_mask, idx) {
-		struct drm_i915_gem_request *request;
-		int ret;
-
-		request = i915_gem_active_peek(&active[idx],
-					       &obj->base.dev->struct_mutex);
-		if (!request)
-			continue;
+	if (excl) {
+		if (ret == 0)
+			ret = i915_gem_request_await_fence(to, excl);
 
-		ret = i915_gem_request_await_request(to, request);
-		if (ret)
-			return ret;
+		fence_put(excl);
 	}
 
-	return 0;
+	return ret;
 }
 
 static void i915_gem_mark_busy(const struct intel_engine_cs *engine)
diff --git a/drivers/gpu/drm/i915/i915_gem_request.h b/drivers/gpu/drm/i915/i915_gem_request.h
index 45998eedda2c..b8feff5857a0 100644
--- a/drivers/gpu/drm/i915/i915_gem_request.h
+++ b/drivers/gpu/drm/i915/i915_gem_request.h
@@ -147,7 +147,7 @@  struct drm_i915_gem_request {
 
 extern const struct fence_ops i915_fence_ops;
 
-static inline bool fence_is_i915(struct fence *fence)
+static inline bool fence_is_i915(const struct fence *fence)
 {
 	return fence->ops == &i915_fence_ops;
 }
@@ -554,22 +554,7 @@  i915_gem_active_isset(const struct i915_gem_active *active)
 }
 
 /**
- * i915_gem_active_is_idle - report whether the active tracker is idle
- * @active - the active tracker
- *
- * i915_gem_active_is_idle() returns true if the active tracker is currently
- * unassigned or if the request is complete (but not yet retired). Requires
- * the caller to hold struct_mutex (but that can be relaxed if desired).
- */
-static inline bool
-i915_gem_active_is_idle(const struct i915_gem_active *active,
-			struct mutex *mutex)
-{
-	return !i915_gem_active_peek(active, mutex);
-}
-
-/**
- * i915_gem_active_wait- waits until the request is completed
+ * i915_gem_active_wait - waits until the request is completed
  * @active - the active request on which to wait
  * @flags - how to wait
  * @timeout - how long to wait at most
@@ -639,24 +624,6 @@  i915_gem_active_retire(struct i915_gem_active *active,
 	return 0;
 }
 
-/* Convenience functions for peeking at state inside active's request whilst
- * guarded by the struct_mutex.
- */
-
-static inline uint32_t
-i915_gem_active_get_seqno(const struct i915_gem_active *active,
-			  struct mutex *mutex)
-{
-	return i915_gem_request_get_seqno(i915_gem_active_peek(active, mutex));
-}
-
-static inline struct intel_engine_cs *
-i915_gem_active_get_engine(const struct i915_gem_active *active,
-			   struct mutex *mutex)
-{
-	return i915_gem_request_get_engine(i915_gem_active_peek(active, mutex));
-}
-
 #define for_each_active(mask, idx) \
 	for (; mask ? idx = ffs(mask) - 1, 1 : 0; mask &= ~BIT(idx))
 
diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c b/drivers/gpu/drm/i915/i915_gpu_error.c
index b81f0fa3c6e3..4f62b82fc4df 100644
--- a/drivers/gpu/drm/i915/i915_gpu_error.c
+++ b/drivers/gpu/drm/i915/i915_gpu_error.c
@@ -880,9 +880,9 @@  static void capture_bo(struct drm_i915_error_buffer *err,
 	err->name = obj->base.name;
 
 	for (i = 0; i < I915_NUM_ENGINES; i++)
-		err->rseqno[i] = __active_get_seqno(&obj->last_read[i]);
-	err->wseqno = __active_get_seqno(&obj->last_write);
-	err->engine = __active_get_engine_id(&obj->last_write);
+		err->rseqno[i] = __active_get_seqno(&vma->last_read[i]);
+	err->wseqno = __active_get_seqno(&vma->last_write);
+	err->engine = __active_get_engine_id(&vma->last_write);
 
 	err->gtt_offset = vma->node.start;
 	err->read_domains = obj->base.read_domains;
diff --git a/drivers/gpu/drm/i915/intel_atomic_plane.c b/drivers/gpu/drm/i915/intel_atomic_plane.c
index b82de3072d4f..a8927929c740 100644
--- a/drivers/gpu/drm/i915/intel_atomic_plane.c
+++ b/drivers/gpu/drm/i915/intel_atomic_plane.c
@@ -84,7 +84,6 @@  intel_plane_duplicate_state(struct drm_plane *plane)
 	state = &intel_state->base;
 
 	__drm_atomic_helper_plane_duplicate_state(plane, state);
-	intel_state->wait_req = NULL;
 
 	return state;
 }
@@ -101,7 +100,6 @@  void
 intel_plane_destroy_state(struct drm_plane *plane,
 			  struct drm_plane_state *state)
 {
-	WARN_ON(state && to_intel_plane_state(state)->wait_req);
 	drm_atomic_helper_plane_destroy_state(plane, state);
 }
 
diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c
index 26b6ad03f84a..1f1f792c5191 100644
--- a/drivers/gpu/drm/i915/intel_display.c
+++ b/drivers/gpu/drm/i915/intel_display.c
@@ -37,7 +37,6 @@ 
 #include "intel_frontbuffer.h"
 #include <drm/i915_drm.h>
 #include "i915_drv.h"
-#include "i915_gem_dmabuf.h"
 #include "intel_dsi.h"
 #include "i915_trace.h"
 #include <drm/drm_atomic.h>
@@ -11945,8 +11944,6 @@  static int intel_gen7_queue_flip(struct drm_device *dev,
 static bool use_mmio_flip(struct intel_engine_cs *engine,
 			  struct drm_i915_gem_object *obj)
 {
-	struct reservation_object *resv;
-
 	/*
 	 * This is not being used for older platforms, because
 	 * non-availability of flip done interrupt forces us to use
@@ -11968,12 +11965,7 @@  static bool use_mmio_flip(struct intel_engine_cs *engine,
 	else if (i915.enable_execlists)
 		return true;
 
-	resv = i915_gem_object_get_dmabuf_resv(obj);
-	if (resv && !reservation_object_test_signaled_rcu(resv, false))
-		return true;
-
-	return engine != i915_gem_active_get_engine(&obj->last_write,
-						    &obj->base.dev->struct_mutex);
+	return engine != i915_gem_object_last_write_engine(obj);
 }
 
 static void skl_do_mmio_flip(struct intel_crtc *intel_crtc,
@@ -12046,17 +12038,8 @@  static void intel_mmio_flip_work_func(struct work_struct *w)
 	struct intel_framebuffer *intel_fb =
 		to_intel_framebuffer(crtc->base.primary->fb);
 	struct drm_i915_gem_object *obj = intel_fb->obj;
-	struct reservation_object *resv;
 
-	if (work->flip_queued_req)
-		WARN_ON(i915_wait_request(work->flip_queued_req,
-					  0, MAX_SCHEDULE_TIMEOUT) < 0);
-
-	/* For framebuffer backed by dmabuf, wait for fence */
-	resv = i915_gem_object_get_dmabuf_resv(obj);
-	if (resv)
-		WARN_ON(reservation_object_wait_timeout_rcu(resv, false, false,
-							    MAX_SCHEDULE_TIMEOUT) < 0);
+	WARN_ON(i915_gem_object_wait(obj, 0, MAX_SCHEDULE_TIMEOUT, NULL) < 0);
 
 	intel_pipe_update_start(crtc);
 
@@ -12257,8 +12240,7 @@  static int intel_crtc_page_flip(struct drm_crtc *crtc,
 	} else if (IS_IVYBRIDGE(dev_priv) || IS_HASWELL(dev_priv)) {
 		engine = dev_priv->engine[BCS];
 	} else if (INTEL_INFO(dev)->gen >= 7) {
-		engine = i915_gem_active_get_engine(&obj->last_write,
-						    &obj->base.dev->struct_mutex);
+		engine = i915_gem_object_last_write_engine(obj);
 		if (engine == NULL || engine->id != RCS)
 			engine = dev_priv->engine[BCS];
 	} else {
@@ -12290,9 +12272,6 @@  static int intel_crtc_page_flip(struct drm_crtc *crtc,
 
 	if (mmio_flip) {
 		INIT_WORK(&work->mmio_work, intel_mmio_flip_work_func);
-
-		work->flip_queued_req = i915_gem_active_get(&obj->last_write,
-							    &obj->base.dev->struct_mutex);
 		queue_work(system_unbound_wq, &work->mmio_work);
 	} else {
 		request = i915_gem_request_alloc(engine, engine->last_context);
@@ -14067,13 +14046,10 @@  static int intel_atomic_check(struct drm_device *dev,
 }
 
 static int intel_atomic_prepare_commit(struct drm_device *dev,
-				       struct drm_atomic_state *state,
-				       bool nonblock)
+				       struct drm_atomic_state *state)
 {
 	struct drm_i915_private *dev_priv = to_i915(dev);
-	struct drm_plane_state *plane_state;
 	struct drm_crtc_state *crtc_state;
-	struct drm_plane *plane;
 	struct drm_crtc *crtc;
 	int i, ret;
 
@@ -14096,30 +14072,6 @@  static int intel_atomic_prepare_commit(struct drm_device *dev,
 	ret = drm_atomic_helper_prepare_planes(dev, state);
 	mutex_unlock(&dev->struct_mutex);
 
-	if (!ret && !nonblock) {
-		for_each_plane_in_state(state, plane, plane_state, i) {
-			struct intel_plane_state *intel_plane_state =
-				to_intel_plane_state(plane_state);
-			long timeout;
-
-			if (!intel_plane_state->wait_req)
-				continue;
-
-			timeout = i915_wait_request(intel_plane_state->wait_req,
-						    I915_WAIT_INTERRUPTIBLE,
-						    MAX_SCHEDULE_TIMEOUT);
-			if (timeout < 0) {
-				/* Any hang should be swallowed by the wait */
-				WARN_ON(timeout == -EIO);
-				mutex_lock(&dev->struct_mutex);
-				drm_atomic_helper_cleanup_planes(dev, state);
-				mutex_unlock(&dev->struct_mutex);
-				ret = timeout;
-				break;
-			}
-		}
-	}
-
 	return ret;
 }
 
@@ -14311,26 +14263,11 @@  static void intel_atomic_commit_tail(struct drm_atomic_state *state)
 	struct drm_crtc_state *old_crtc_state;
 	struct drm_crtc *crtc;
 	struct intel_crtc_state *intel_cstate;
-	struct drm_plane *plane;
-	struct drm_plane_state *plane_state;
 	bool hw_check = intel_state->modeset;
 	unsigned long put_domains[I915_MAX_PIPES] = {};
 	unsigned crtc_vblank_mask = 0;
 	int i;
 
-	for_each_plane_in_state(state, plane, plane_state, i) {
-		struct intel_plane_state *intel_plane_state =
-			to_intel_plane_state(plane_state);
-
-		if (!intel_plane_state->wait_req)
-			continue;
-
-		/* EIO should be eaten, and we can't get interrupted in the
-		 * worker, and blocking commits have waited already. */
-		WARN_ON(i915_wait_request(intel_plane_state->wait_req,
-					  0, MAX_SCHEDULE_TIMEOUT) < 0);
-	}
-
 	drm_atomic_helper_wait_for_dependencies(state);
 
 	if (intel_state->modeset) {
@@ -14537,7 +14474,7 @@  static int intel_atomic_commit(struct drm_device *dev,
 
 	INIT_WORK(&state->commit_work, intel_atomic_commit_work);
 
-	ret = intel_atomic_prepare_commit(dev, state, nonblock);
+	ret = intel_atomic_prepare_commit(dev, state);
 	if (ret) {
 		DRM_DEBUG_ATOMIC("Preparing state failed with %i\n", ret);
 		return ret;
@@ -14670,7 +14607,7 @@  intel_prepare_plane_fb(struct drm_plane *plane,
 	struct drm_framebuffer *fb = new_state->fb;
 	struct drm_i915_gem_object *obj = intel_fb_obj(fb);
 	struct drm_i915_gem_object *old_obj = intel_fb_obj(plane->state->fb);
-	struct reservation_object *resv;
+	long lret;
 	int ret = 0;
 
 	if (!obj && !old_obj)
@@ -14708,39 +14645,34 @@  intel_prepare_plane_fb(struct drm_plane *plane,
 		return 0;
 
 	/* For framebuffer backed by dmabuf, wait for fence */
-	resv = i915_gem_object_get_dmabuf_resv(obj);
-	if (resv) {
-		long lret;
+	lret = i915_gem_object_wait(obj,
+				    I915_WAIT_INTERRUPTIBLE | I915_WAIT_LOCKED,
+				    MAX_SCHEDULE_TIMEOUT,
+				    NULL);
+	if (lret == -ERESTARTSYS)
+		return lret;
 
-		lret = reservation_object_wait_timeout_rcu(resv, false, true,
-							   MAX_SCHEDULE_TIMEOUT);
-		if (lret == -ERESTARTSYS)
-			return lret;
-
-		WARN(lret < 0, "waiting returns %li\n", lret);
-	}
+	WARN(lret < 0, "waiting returns %li\n", lret);
 
 	if (plane->type == DRM_PLANE_TYPE_CURSOR &&
 	    INTEL_INFO(dev)->cursor_needs_physical) {
 		int align = IS_I830(dev_priv) ? 16 * 1024 : 256;
 		ret = i915_gem_object_attach_phys(obj, align);
-		if (ret)
+		if (ret) {
 			DRM_DEBUG_KMS("failed to attach phys object\n");
+			return ret;
+		}
 	} else {
 		struct i915_vma *vma;
 
 		vma = intel_pin_and_fence_fb_obj(fb, new_state->rotation);
-		if (IS_ERR(vma))
-			ret = PTR_ERR(vma);
-	}
-
-	if (ret == 0) {
-		to_intel_plane_state(new_state)->wait_req =
-			i915_gem_active_get(&obj->last_write,
-					    &obj->base.dev->struct_mutex);
+		if (IS_ERR(vma)) {
+			DRM_DEBUG_KMS("failed to pin object\n");
+			return PTR_ERR(vma);
+		}
 	}
 
-	return ret;
+	return 0;
 }
 
 /**
@@ -14758,7 +14690,6 @@  intel_cleanup_plane_fb(struct drm_plane *plane,
 {
 	struct drm_device *dev = plane->dev;
 	struct intel_plane_state *old_intel_state;
-	struct intel_plane_state *intel_state = to_intel_plane_state(plane->state);
 	struct drm_i915_gem_object *old_obj = intel_fb_obj(old_state->fb);
 	struct drm_i915_gem_object *obj = intel_fb_obj(plane->state->fb);
 
@@ -14770,9 +14701,6 @@  intel_cleanup_plane_fb(struct drm_plane *plane,
 	if (old_obj && (plane->type != DRM_PLANE_TYPE_CURSOR ||
 	    !INTEL_INFO(dev)->cursor_needs_physical))
 		intel_unpin_fb_obj(old_state->fb, old_state->rotation);
-
-	i915_gem_request_assign(&intel_state->wait_req, NULL);
-	i915_gem_request_assign(&old_intel_state->wait_req, NULL);
 }
 
 int
diff --git a/drivers/gpu/drm/i915/intel_drv.h b/drivers/gpu/drm/i915/intel_drv.h
index 2948615573d6..3dfe99516722 100644
--- a/drivers/gpu/drm/i915/intel_drv.h
+++ b/drivers/gpu/drm/i915/intel_drv.h
@@ -401,9 +401,6 @@  struct intel_plane_state {
 	int scaler_id;
 
 	struct drm_intel_sprite_colorkey ckey;
-
-	/* async flip related structures */
-	struct drm_i915_gem_request *wait_req;
 };
 
 struct intel_initial_plane_config {

Comments

On pe, 2016-10-14 at 13:18 +0100, Chris Wilson wrote:
> In preparation to support many distinct timelines, we need to expand the
> activity tracking on the GEM object to handle more than just a request
> per engine. We already use the struct reservation_object on the dma-buf
> to handle many fence contexts, so integrating that into the GEM object
> itself is the preferred solution. (For example, we can now share the same
> reservation_object between every consumer/producer using this buffer and
> skip the manual import/export via dma-buf.)
> 
> v2: Reimplement busy-ioctl (by walking the reservation object), postpone
> the ABI change for another day. Similarly use the reservation object to
> find the last_write request (if active and from i915) for choosing
> display CS flips.
> 
> Caveats:
> 
>  * busy-ioctl: busy-ioctl only reports on the native fences, it will not
> warn of stalls (in set-domain-ioctl, pread/pwrite etc) if the object is
> being rendered to by external fences. It also will not report the same
> busy state as wait-ioctl (or polling on the dma-buf) in the same
> circumstances. On the plus side, it does retain reporting of which
> *i915* engines are engaged with this object.
> 
>  * non-blocking atomic modesets take a step backwards as the wait for
> render completion blocks the ioctl. This is fixed in a subsequent
> patch to use a fence instead for awaiting on the rendering, see
> "drm/i915: Restore nonblocking awaits for modesetting"
> 
>  * dynamic array manipulation for shared-fences in reservation is slower
> than the previous lockless static assignment (e.g. gem_exec_lut_handle
> runtime on ivb goes from 42s to 66s), mainly due to atomic operations.
> 
>  * loss of object-level retirement callbacks, emulated by VMA retirement
> tracking.
> 
>  * minor loss of object-level last activity information from debugfs,
> could be replaced with per-vma information if desired
> 
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

This was already:

Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

Regards, Joonas