linux-kernel - [PATCH v2 09/23] drm/msm/gem: Avoid obj lock in job

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <20230320144356.803762-10-robdclark@gmail.com>
Date:   Mon, 20 Mar 2023 07:43:31 -0700
From:   Rob Clark <robdclark@...il.com>
To:     dri-devel@...ts.freedesktop.org
Cc:     Daniel Vetter <daniel@...ll.ch>, freedreno@...ts.freedesktop.org,
        linux-arm-msm@...r.kernel.org, Rob Clark <robdclark@...omium.org>,
        Rob Clark <robdclark@...il.com>,
        Abhinav Kumar <quic_abhinavk@...cinc.com>,
        Dmitry Baryshkov <dmitry.baryshkov@...aro.org>,
        Sean Paul <sean@...rly.run>, David Airlie <airlied@...il.com>,
        Sumit Semwal <sumit.semwal@...aro.org>,
        Christian König <christian.koenig@....com>,
        linux-kernel@...r.kernel.org (open list),
        linux-media@...r.kernel.org (open list:DMA BUFFER SHARING FRAMEWORK),
        linaro-mm-sig@...ts.linaro.org (moderated list:DMA BUFFER SHARING
        FRAMEWORK)
Subject: [PATCH v2 09/23] drm/msm/gem: Avoid obj lock in job_run()

From: Rob Clark <robdclark@...omium.org>

Now that everything that controls which LRU an obj lives in *except* the
backing pages is protected by the LRU lock, add a special path to unpin
in the job_run() path, we we are assured that we already have backing
pages and will not be racing against eviction (because the GEM object's
dma_resv contains the fence that will be signaled when the submit/job
completes).

Signed-off-by: Rob Clark <robdclark@...omium.org>
---
 drivers/gpu/drm/msm/msm_gem.c        | 44 +++++++++++++++++++++++-----
 drivers/gpu/drm/msm/msm_gem.h        |  1 +
 drivers/gpu/drm/msm/msm_ringbuffer.c |  4 +--
 3 files changed, 39 insertions(+), 10 deletions(-)

diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c
index d0ac3e704b66..9628e8d8dd02 100644
--- a/drivers/gpu/drm/msm/msm_gem.c
+++ b/drivers/gpu/drm/msm/msm_gem.c
@@ -61,18 +61,14 @@ static void sync_for_cpu(struct msm_gem_object *msm_obj)
 	dma_unmap_sgtable(dev, msm_obj->sgt, DMA_BIDIRECTIONAL, 0);
 }
 
-static void update_lru_locked(struct drm_gem_object *obj)
+static void update_lru_active(struct drm_gem_object *obj)
 {
 	struct msm_drm_private *priv = obj->dev->dev_private;
 	struct msm_gem_object *msm_obj = to_msm_bo(obj);
 
-	msm_gem_assert_locked(&msm_obj->base);
-
-	if (!msm_obj->pages) {
-		GEM_WARN_ON(msm_obj->pin_count);
+	GEM_WARN_ON(!msm_obj->pages);
 
-		drm_gem_lru_move_tail_locked(&priv->lru.unbacked, obj);
-	} else if (msm_obj->pin_count) {
+	if (msm_obj->pin_count) {
 		drm_gem_lru_move_tail_locked(&priv->lru.pinned, obj);
 	} else if (msm_obj->madv == MSM_MADV_WILLNEED) {
 		drm_gem_lru_move_tail_locked(&priv->lru.willneed, obj);
@@ -83,6 +79,22 @@ static void update_lru_locked(struct drm_gem_object *obj)
 	}
 }
 
+static void update_lru_locked(struct drm_gem_object *obj)
+{
+	struct msm_drm_private *priv = obj->dev->dev_private;
+	struct msm_gem_object *msm_obj = to_msm_bo(obj);
+
+	msm_gem_assert_locked(&msm_obj->base);
+
+	if (!msm_obj->pages) {
+		GEM_WARN_ON(msm_obj->pin_count);
+
+		drm_gem_lru_move_tail_locked(&priv->lru.unbacked, obj);
+	} else {
+		update_lru_active(obj);
+	}
+}
+
 static void update_lru(struct drm_gem_object *obj)
 {
 	struct msm_drm_private *priv = obj->dev->dev_private;
@@ -489,6 +501,24 @@ void msm_gem_unpin_locked(struct drm_gem_object *obj)
 	mutex_unlock(&priv->lru.lock);
 }
 
+/* Special unpin path for use in fence-signaling path, avoiding the need
+ * to hold the obj lock by only depending on things that a protected by
+ * the LRU lock.  In particular we know that that we already have backing
+ * and and that the object's dma_resv has the fence for the current
+ * submit/job which will prevent us racing against page eviction.
+ */
+void msm_gem_unpin_active(struct drm_gem_object *obj)
+{
+	struct msm_drm_private *priv = obj->dev->dev_private;
+	struct msm_gem_object *msm_obj = to_msm_bo(obj);
+
+	mutex_lock(&priv->lru.lock);
+	msm_obj->pin_count--;
+	GEM_WARN_ON(msm_obj->pin_count < 0);
+	update_lru_active(obj);
+	mutex_unlock(&priv->lru.lock);
+}
+
 struct msm_gem_vma *msm_gem_get_vma_locked(struct drm_gem_object *obj,
 					   struct msm_gem_address_space *aspace)
 {
diff --git a/drivers/gpu/drm/msm/msm_gem.h b/drivers/gpu/drm/msm/msm_gem.h
index 0057e8e8fa13..2bd6846c83a9 100644
--- a/drivers/gpu/drm/msm/msm_gem.h
+++ b/drivers/gpu/drm/msm/msm_gem.h
@@ -128,6 +128,7 @@ struct msm_gem_object {
 uint64_t msm_gem_mmap_offset(struct drm_gem_object *obj);
 int msm_gem_pin_vma_locked(struct drm_gem_object *obj, struct msm_gem_vma *vma);
 void msm_gem_unpin_locked(struct drm_gem_object *obj);
+void msm_gem_unpin_active(struct drm_gem_object *obj);
 struct msm_gem_vma *msm_gem_get_vma_locked(struct drm_gem_object *obj,
 					   struct msm_gem_address_space *aspace);
 int msm_gem_get_iova(struct drm_gem_object *obj,
diff --git a/drivers/gpu/drm/msm/msm_ringbuffer.c b/drivers/gpu/drm/msm/msm_ringbuffer.c
index 31b4fbf96c36..b60199184409 100644
--- a/drivers/gpu/drm/msm/msm_ringbuffer.c
+++ b/drivers/gpu/drm/msm/msm_ringbuffer.c
@@ -24,9 +24,7 @@ static struct dma_fence *msm_job_run(struct drm_sched_job *job)
 		struct drm_gem_object *obj = &submit->bos[i].obj->base;
 
 		msm_gem_vma_unpin_fenced(submit->bos[i].vma, fctx);
-		msm_gem_lock(obj);
-		msm_gem_unpin_locked(obj);
-		msm_gem_unlock(obj);
+		msm_gem_unpin_active(obj);
 		submit->bos[i].flags &= ~(BO_VMA_PINNED | BO_OBJ_PINNED);
 	}
 
-- 
2.39.2