All of lore.kernel.org
 help / color / mirror / Atom feed
From: Chris Wilson <chris@chris-wilson.co.uk>
To: intel-gfx@lists.freedesktop.org
Cc: mika.kuoppala@intel.com
Subject: [PATCH 10/38] drm/i915: Defer active reference until required
Date: Tue, 20 Sep 2016 09:29:44 +0100	[thread overview]
Message-ID: <20160920083012.2754-11-chris@chris-wilson.co.uk> (raw)
In-Reply-To: <20160920083012.2754-1-chris@chris-wilson.co.uk>

We only need the active reference to keep the object alive after the
handle has been deleted (so as to prevent a synchronous gem_close). Why
then pay the price of a kref on every execbuf when we can insert that
final active ref just in time for the handle deletion?

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_drv.h              | 34 ++++++++++++++++++++++++++++
 drivers/gpu/drm/i915/i915_gem.c              | 25 +++++++++++++++++++-
 drivers/gpu/drm/i915/i915_gem_batch_pool.c   |  2 +-
 drivers/gpu/drm/i915/i915_gem_context.c      |  2 +-
 drivers/gpu/drm/i915/i915_gem_execbuffer.c   |  2 --
 drivers/gpu/drm/i915/i915_gem_gtt.c          |  2 +-
 drivers/gpu/drm/i915/i915_gem_render_state.c |  2 +-
 drivers/gpu/drm/i915/intel_ringbuffer.c      |  4 ++--
 8 files changed, 64 insertions(+), 9 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index a3401418bf4d..89eab85ab11c 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -2214,6 +2214,13 @@ struct drm_i915_gem_object {
 	((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_SHIFT (I915_BO_ACTIVE_SHIFT + I915_NUM_ENGINES)
+#define I915_BO_ACTIVE_REF BIT(I915_BO_ACTIVE_REF_SHIFT)
+
+	/**
 	 * This is set if the object has been written to since last bound
 	 * to the GTT
 	 */
@@ -2383,6 +2390,28 @@ i915_gem_object_has_active_engine(const struct drm_i915_gem_object *obj,
 	return obj->flags & BIT(engine + I915_BO_ACTIVE_SHIFT);
 }
 
+static inline bool
+i915_gem_object_has_active_reference(const struct drm_i915_gem_object *obj)
+{
+	return obj->flags & I915_BO_ACTIVE_REF;
+}
+
+static inline void
+i915_gem_object_set_active_reference(struct drm_i915_gem_object *obj)
+{
+	lockdep_assert_held(&obj->base.dev->struct_mutex);
+	obj->flags |= I915_BO_ACTIVE_REF;
+}
+
+static inline void
+i915_gem_object_clear_active_reference(struct drm_i915_gem_object *obj)
+{
+	lockdep_assert_held(&obj->base.dev->struct_mutex);
+	obj->flags &= ~I915_BO_ACTIVE_REF;
+}
+
+void __i915_gem_object_release_unless_active(struct drm_i915_gem_object *obj);
+
 static inline unsigned int
 i915_gem_object_get_tiling(struct drm_i915_gem_object *obj)
 {
@@ -2413,6 +2442,11 @@ static inline void i915_vma_put(struct i915_vma *vma)
 	i915_gem_object_put(vma->obj);
 }
 
+static inline void i915_vma_put_internal(struct i915_vma *vma)
+{
+	__i915_gem_object_release_unless_active(vma->obj);
+}
+
 /*
  * Optimised SGL iterator for GEM objects
  */
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index d9214c9d31d2..e8373f264858 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -2618,7 +2618,10 @@ i915_gem_object_retire__read(struct i915_gem_active *active,
 		list_move_tail(&obj->global_list,
 			       &request->i915->mm.bound_list);
 
-	i915_gem_object_put(obj);
+	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)
@@ -2881,6 +2884,12 @@ void i915_gem_close_object(struct drm_gem_object *gem, struct drm_file *file)
 	list_for_each_entry_safe(vma, vn, &obj->vma_list, obj_link)
 		if (vma->vm->file == fpriv)
 			i915_vma_close(vma);
+
+	if (i915_gem_object_is_active(obj) &&
+	    !i915_gem_object_has_active_reference(obj)) {
+		i915_gem_object_set_active_reference(obj);
+		i915_gem_object_get(obj);
+	}
 	mutex_unlock(&obj->base.dev->struct_mutex);
 }
 
@@ -4358,6 +4367,20 @@ void i915_gem_free_object(struct drm_gem_object *gem_obj)
 	intel_runtime_pm_put(dev_priv);
 }
 
+void __i915_gem_object_release_unless_active(struct drm_i915_gem_object *obj)
+{
+	if (!obj)
+		return;
+
+	lockdep_assert_held(&obj->base.dev->struct_mutex);
+
+	GEM_BUG_ON(i915_gem_object_has_active_reference(obj));
+	if (i915_gem_object_is_active(obj))
+		i915_gem_object_set_active_reference(obj);
+	else
+		i915_gem_object_put(obj);
+}
+
 int i915_gem_suspend(struct drm_device *dev)
 {
 	struct drm_i915_private *dev_priv = to_i915(dev);
diff --git a/drivers/gpu/drm/i915/i915_gem_batch_pool.c b/drivers/gpu/drm/i915/i915_gem_batch_pool.c
index ed989596d9a3..cb25cad3318c 100644
--- a/drivers/gpu/drm/i915/i915_gem_batch_pool.c
+++ b/drivers/gpu/drm/i915/i915_gem_batch_pool.c
@@ -73,7 +73,7 @@ void i915_gem_batch_pool_fini(struct i915_gem_batch_pool *pool)
 		list_for_each_entry_safe(obj, next,
 					 &pool->cache_list[n],
 					 batch_pool_link)
-			i915_gem_object_put(obj);
+			__i915_gem_object_release_unless_active(obj);
 
 		INIT_LIST_HEAD(&pool->cache_list[n]);
 	}
diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c
index df10f4e95736..1d2ab73a8f43 100644
--- a/drivers/gpu/drm/i915/i915_gem_context.c
+++ b/drivers/gpu/drm/i915/i915_gem_context.c
@@ -155,7 +155,7 @@ void i915_gem_context_free(struct kref *ctx_ref)
 		if (ce->ring)
 			intel_ring_free(ce->ring);
 
-		i915_vma_put(ce->state);
+		__i915_gem_object_release_unless_active(ce->state->obj);
 	}
 
 	put_pid(ctx->pid);
diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
index 33c85227643d..243f5067e7aa 100644
--- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
@@ -1299,8 +1299,6 @@ void i915_vma_move_to_active(struct i915_vma *vma,
 	 * add the active reference first and queue for it to be dropped
 	 * *last*.
 	 */
-	if (!i915_gem_object_is_active(obj))
-		i915_gem_object_get(obj);
 	i915_gem_object_set_active(obj, idx);
 	i915_gem_active_set(&obj->last_read[idx], req);
 
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index 7906fc3aa488..ed3cd714dd21 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -3718,5 +3718,5 @@ void i915_vma_unpin_and_release(struct i915_vma **p_vma)
 		return;
 
 	i915_vma_unpin(vma);
-	i915_vma_put(vma);
+	i915_vma_put_internal(vma);
 }
diff --git a/drivers/gpu/drm/i915/i915_gem_render_state.c b/drivers/gpu/drm/i915/i915_gem_render_state.c
index 95b7e9afd5f8..055139343137 100644
--- a/drivers/gpu/drm/i915/i915_gem_render_state.c
+++ b/drivers/gpu/drm/i915/i915_gem_render_state.c
@@ -225,6 +225,6 @@ int i915_gem_render_state_init(struct drm_i915_gem_request *req)
 err_unpin:
 	i915_vma_unpin(so.vma);
 err_obj:
-	i915_gem_object_put(obj);
+	__i915_gem_object_release_unless_active(obj);
 	return ret;
 }
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
index 4bf6bd056b26..6530acd8be69 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.c
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
@@ -1826,7 +1826,7 @@ static void cleanup_status_page(struct intel_engine_cs *engine)
 
 	i915_vma_unpin(vma);
 	i915_gem_object_unpin_map(vma->obj);
-	i915_vma_put(vma);
+	i915_vma_put_internal(vma);
 }
 
 static int init_status_page(struct intel_engine_cs *engine)
@@ -2023,7 +2023,7 @@ intel_engine_create_ring(struct intel_engine_cs *engine, int size)
 void
 intel_ring_free(struct intel_ring *ring)
 {
-	i915_vma_put(ring->vma);
+	i915_vma_put_internal(ring->vma);
 	kfree(ring);
 }
 
-- 
2.9.3

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

  parent reply	other threads:[~2016-09-20  8:30 UTC|newest]

Thread overview: 64+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2016-09-20  8:29 Multiple timelines, take 2 Chris Wilson
2016-09-20  8:29 ` [PATCH 01/38] drm/i915: Allow disabling error capture Chris Wilson
2016-09-21  6:13   ` Joonas Lahtinen
2016-09-20  8:29 ` [PATCH 02/38] drm/i915: Stop the machine whilst capturing the GPU crash dump Chris Wilson
2016-09-26  8:58   ` Joonas Lahtinen
2016-09-20  8:29 ` [PATCH 03/38] drm/i915: Always use the GTT for error capture Chris Wilson
2016-09-21  7:24   ` Joonas Lahtinen
2016-09-20  8:29 ` [PATCH 04/38] drm/i915: Consolidate error object printing Chris Wilson
2016-09-20  8:29 ` [PATCH 05/38] drm/i915: Compress GPU objects in error state Chris Wilson
2016-09-21  7:55   ` Joonas Lahtinen
2016-09-20  8:29 ` [PATCH 06/38] drm/i915: Support asynchronous waits on struct fence from i915_gem_request Chris Wilson
2016-09-21  8:05   ` Joonas Lahtinen
2016-09-20  8:29 ` [PATCH 07/38] drm/i915: Allow i915_sw_fence_await_sw_fence() to allocate Chris Wilson
2016-09-20  8:29 ` [PATCH 08/38] drm/i915: Rearrange i915_wait_request() accounting with callers Chris Wilson
2016-09-21  8:12   ` Joonas Lahtinen
2016-09-20  8:29 ` [PATCH 09/38] drm/i915: Remove unused i915_gem_active_wait() in favour of _unlocked() Chris Wilson
2016-09-20  8:29 ` Chris Wilson [this message]
2016-09-21  8:44   ` [PATCH 10/38] drm/i915: Defer active reference until required Joonas Lahtinen
2016-09-20  8:29 ` [PATCH 11/38] drm/i915: Introduce an internal allocator for disposable private objects Chris Wilson
2016-09-21 11:50   ` Joonas Lahtinen
2016-09-27  9:10     ` Chris Wilson
2016-09-20  8:29 ` [PATCH 12/38] drm/i915: Reuse the active golden render state batch Chris Wilson
2016-09-26  7:24   ` Joonas Lahtinen
2016-09-20  8:29 ` [PATCH 13/38] drm/i915: Markup GEM API with lockdep asserts Chris Wilson
2016-09-21 11:56   ` Joonas Lahtinen
2016-09-20  8:29 ` [PATCH 14/38] drm/i915: Use a radixtree for random access to the object's backing storage Chris Wilson
2016-09-20  8:29 ` [PATCH 15/38] drm/i915: Refactor object page API Chris Wilson
2016-09-20  8:29 ` [PATCH 16/38] drm/i915: Pass around sg_table to get_pages/put_pages backend Chris Wilson
2016-09-20 11:24   ` kbuild test robot
2016-09-20  8:29 ` [PATCH 17/38] drm/i915: Move object backing storage manipulation to its own locking Chris Wilson
2016-09-20  8:29 ` [PATCH 18/38] drm/i915/dmabuf: Acquire the backing storage outside of struct_mutex Chris Wilson
2016-09-20  8:29 ` [PATCH 19/38] drm/i915: Implement pread without struct-mutex Chris Wilson
2016-09-20  8:29 ` [PATCH 20/38] drm/i915: Implement pwrite " Chris Wilson
2016-09-20 13:47   ` kbuild test robot
2016-09-20  8:29 ` [PATCH 21/38] drm/i915: Acquire the backing storage outside of struct_mutex in set-domain Chris Wilson
2016-09-20  8:29 ` [PATCH 22/38] drm/i915: Move object release to a freelist + worker Chris Wilson
2016-09-20  8:29 ` [PATCH 23/38] drm/i915: Use lockless object free Chris Wilson
2016-09-20  8:29 ` [PATCH 24/38] drm/i915: Move GEM activity tracking into a common struct reservation_object Chris Wilson
2016-09-26  7:53   ` Joonas Lahtinen
2016-09-20  8:29 ` [PATCH 25/38] drm: Add reference counting to drm_atomic_state Chris Wilson
2016-09-21  7:24   ` Sean Paul
2016-09-20  8:30 ` [PATCH 26/38] drm/i915: Restore nonblocking awaits for modesetting Chris Wilson
2016-09-26  8:11   ` Joonas Lahtinen
2016-09-20  8:30 ` [PATCH 27/38] drm/i915: Combine seqno + tracking into a global timeline struct Chris Wilson
2016-09-20  8:30 ` [PATCH 28/38] drm/i915: Queue the idling context switch after all other timelines Chris Wilson
2016-09-26  8:49   ` Joonas Lahtinen
2016-09-20  8:30 ` [PATCH 29/38] drm/i915: Wait first for submission, before waiting for request completion Chris Wilson
2016-09-20  8:30 ` [PATCH 30/38] drm/i915: Introduce a global_seqno for each request Chris Wilson
2016-09-20  8:30 ` [PATCH 31/38] drm/i915: Record space required for request emission Chris Wilson
2016-09-20  8:30 ` [PATCH 32/38] drm/i915: Defer " Chris Wilson
2016-09-26  8:53   ` Joonas Lahtinen
2016-09-26  9:04     ` Chris Wilson
2016-09-26  9:06       ` Joonas Lahtinen
2016-09-26  9:25         ` Chris Wilson
2016-09-20  8:30 ` [PATCH 33/38] drm/i915: Move the global sync optimisation to the timeline Chris Wilson
2016-09-20  8:30 ` [PATCH 34/38] drm/i915: Create a unique name for the context Chris Wilson
2016-09-20  8:30 ` [PATCH 35/38] drm/i915: Reserve space in the global seqno during request allocation Chris Wilson
2016-09-20 18:49   ` kbuild test robot
2016-09-20 18:49   ` [PATCH] drm/i915: fix semicolon.cocci warnings kbuild test robot
2016-09-20  8:30 ` [PATCH 36/38] drm/i915: Enable multiple timelines Chris Wilson
2016-09-26  8:55   ` Joonas Lahtinen
2016-09-20  8:30 ` [PATCH 37/38] drm/i915: Enable userspace to opt-out of implicit fencing Chris Wilson
2016-09-20  8:30 ` [PATCH 38/38] drm/i915: Support explicit fencing for execbuf Chris Wilson
2016-09-20  9:24 ` ✗ Fi.CI.BAT: failure for series starting with [01/38] drm/i915: Allow disabling error capture Patchwork

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20160920083012.2754-11-chris@chris-wilson.co.uk \
    --to=chris@chris-wilson.co.uk \
    --cc=intel-gfx@lists.freedesktop.org \
    --cc=mika.kuoppala@intel.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.