From: Rob Clark robdclark@chromium.org
I've been spending some time looking into how things behave under high memory pressure. The first patch is a random cleanup I noticed along the way. The second improves the situation significantly when we are getting shrinker called from many threads in parallel. And the last two are $debugfs/gem fixes I needed so I could monitor the state of GEM objects (ie. how many are active/purgable/purged) while triggering high memory pressure.
We could probably go a bit further with dropping the mm_lock in the shrinker->scan() loop, but this is already a pretty big improvement. The next step is probably actually to add support to unpin/evict inactive objects. (We are part way there since we have already de- coupled the iova lifetime from the pages lifetime, but there are a few sharp corners to work through.)
Rob Clark (4): drm/msm: Remove unused freed llist node drm/msm: Avoid mutex in shrinker_count() drm/msm: Fix debugfs deadlock drm/msm: Improved debugfs gem stats
drivers/gpu/drm/msm/msm_debugfs.c | 14 ++--- drivers/gpu/drm/msm/msm_drv.c | 4 ++ drivers/gpu/drm/msm/msm_drv.h | 15 ++++-- drivers/gpu/drm/msm/msm_fb.c | 3 +- drivers/gpu/drm/msm/msm_gem.c | 65 ++++++++++++++++++----- drivers/gpu/drm/msm/msm_gem.h | 72 +++++++++++++++++++++++--- drivers/gpu/drm/msm/msm_gem_shrinker.c | 28 ++++------ 7 files changed, 150 insertions(+), 51 deletions(-)
From: Rob Clark robdclark@chromium.org
When the system is under heavy memory pressure, we can end up with lots of concurrent calls into the shrinker. Keeping a running tab on what we can shrink avoids grabbing a lock in shrinker->count(), and avoids shrinker->scan() getting called when not profitable.
Also, we can keep purged objects in their own list to avoid re-traversing them to help cut down time in the critical section further.
Signed-off-by: Rob Clark robdclark@chromium.org Tested-by: Douglas Anderson dianders@chromium.org --- drivers/gpu/drm/msm/msm_drv.c | 1 + drivers/gpu/drm/msm/msm_drv.h | 6 ++- drivers/gpu/drm/msm/msm_gem.c | 20 ++++++++-- drivers/gpu/drm/msm/msm_gem.h | 53 ++++++++++++++++++++++++-- drivers/gpu/drm/msm/msm_gem_shrinker.c | 28 ++++++-------- 5 files changed, 81 insertions(+), 27 deletions(-)
diff --git a/drivers/gpu/drm/msm/msm_drv.c b/drivers/gpu/drm/msm/msm_drv.c index 4f9fa0189a07..3462b0ea14c6 100644 --- a/drivers/gpu/drm/msm/msm_drv.c +++ b/drivers/gpu/drm/msm/msm_drv.c @@ -476,6 +476,7 @@ static int msm_drm_init(struct device *dev, const struct drm_driver *drv)
INIT_LIST_HEAD(&priv->inactive_willneed); INIT_LIST_HEAD(&priv->inactive_dontneed); + INIT_LIST_HEAD(&priv->inactive_purged); mutex_init(&priv->mm_lock);
/* Teach lockdep about lock ordering wrt. shrinker: */ diff --git a/drivers/gpu/drm/msm/msm_drv.h b/drivers/gpu/drm/msm/msm_drv.h index a1264cfcac5e..503168817e24 100644 --- a/drivers/gpu/drm/msm/msm_drv.h +++ b/drivers/gpu/drm/msm/msm_drv.h @@ -179,8 +179,8 @@ struct msm_drm_private { * inactive lists (depending on whether or not it is shrinkable) or * gpu->active_list (for the gpu it is active on[1]) * - * These lists are protected by mm_lock. If struct_mutex is involved, it - * should be aquired prior to mm_lock. One should *not* hold mm_lock in + * These lists are protected by mm_lock (which should be acquired + * before per GEM object lock). One should *not* hold mm_lock in * get_pages()/vmap()/etc paths, as they can trigger the shrinker. * * [1] if someone ever added support for the old 2d cores, there could be @@ -188,6 +188,8 @@ struct msm_drm_private { */ struct list_head inactive_willneed; /* inactive + !shrinkable */ struct list_head inactive_dontneed; /* inactive + shrinkable */ + struct list_head inactive_purged; /* inactive + purged */ + long shrinkable_count; /* write access under mm_lock */ struct mutex mm_lock;
struct workqueue_struct *wq; diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c index 9d10739c4eb2..bec01bb48fce 100644 --- a/drivers/gpu/drm/msm/msm_gem.c +++ b/drivers/gpu/drm/msm/msm_gem.c @@ -719,6 +719,7 @@ void msm_gem_purge(struct drm_gem_object *obj) put_iova_vmas(obj);
msm_obj->madv = __MSM_MADV_PURGED; + mark_unpurgable(msm_obj);
drm_vma_node_unmap(&obj->vma_node, dev->anon_inode->i_mapping); drm_gem_free_mmap_offset(obj); @@ -790,10 +791,11 @@ void msm_gem_active_get(struct drm_gem_object *obj, struct msm_gpu *gpu) might_sleep(); WARN_ON(!msm_gem_is_locked(obj)); WARN_ON(msm_obj->madv != MSM_MADV_WILLNEED); + WARN_ON(msm_obj->dontneed);
if (msm_obj->active_count++ == 0) { mutex_lock(&priv->mm_lock); - list_del_init(&msm_obj->mm_list); + list_del(&msm_obj->mm_list); list_add_tail(&msm_obj->mm_list, &gpu->active_list); mutex_unlock(&priv->mm_lock); } @@ -818,11 +820,19 @@ static void update_inactive(struct msm_gem_object *msm_obj) mutex_lock(&priv->mm_lock); WARN_ON(msm_obj->active_count != 0);
- list_del_init(&msm_obj->mm_list); - if (msm_obj->madv == MSM_MADV_WILLNEED) + if (msm_obj->dontneed) + mark_unpurgable(msm_obj); + + list_del(&msm_obj->mm_list); + if (msm_obj->madv == MSM_MADV_WILLNEED) { list_add_tail(&msm_obj->mm_list, &priv->inactive_willneed); - else + } else if (msm_obj->madv == MSM_MADV_DONTNEED) { list_add_tail(&msm_obj->mm_list, &priv->inactive_dontneed); + mark_purgable(msm_obj); + } else { + WARN_ON(msm_obj->madv != __MSM_MADV_PURGED); + list_add_tail(&msm_obj->mm_list, &priv->inactive_purged); + }
mutex_unlock(&priv->mm_lock); } @@ -971,6 +981,8 @@ void msm_gem_free_object(struct drm_gem_object *obj) struct msm_drm_private *priv = dev->dev_private;
mutex_lock(&priv->mm_lock); + if (msm_obj->dontneed) + mark_unpurgable(msm_obj); list_del(&msm_obj->mm_list); mutex_unlock(&priv->mm_lock);
diff --git a/drivers/gpu/drm/msm/msm_gem.h b/drivers/gpu/drm/msm/msm_gem.h index 7a9107cf1818..13aabfe92dac 100644 --- a/drivers/gpu/drm/msm/msm_gem.h +++ b/drivers/gpu/drm/msm/msm_gem.h @@ -50,18 +50,24 @@ struct msm_gem_object { */ uint8_t madv;
+ /** + * Is object on inactive_dontneed list (ie. counted in priv->shrinkable_count)? + */ + bool dontneed : 1; + /** * count of active vmap'ing */ uint8_t vmap_count;
- /* And object is either: - * inactive - on priv->inactive_list + /** + * An object is either: + * inactive - on priv->inactive_dontneed/willneed/purged depending + * on status * active - on one one of the gpu's active_list.. well, at * least for now we don't have (I don't think) hw sync between * 2d and 3d one devices which have both, meaning we need to * block on submit if a bo is already on other ring - * */ struct list_head mm_list;
@@ -186,10 +192,16 @@ static inline bool is_active(struct msm_gem_object *msm_obj) return msm_obj->active_count; }
+/* imported/exported objects are not purgable: */ +static inline bool is_unpurgable(struct msm_gem_object *msm_obj) +{ + return msm_obj->base.dma_buf && msm_obj->base.import_attach; +} + static inline bool is_purgeable(struct msm_gem_object *msm_obj) { return (msm_obj->madv == MSM_MADV_DONTNEED) && msm_obj->sgt && - !msm_obj->base.dma_buf && !msm_obj->base.import_attach; + !is_unpurgable(msm_obj); }
static inline bool is_vunmapable(struct msm_gem_object *msm_obj) @@ -198,6 +210,39 @@ static inline bool is_vunmapable(struct msm_gem_object *msm_obj) return (msm_obj->vmap_count == 0) && msm_obj->vaddr; }
+static inline void mark_purgable(struct msm_gem_object *msm_obj) +{ + struct msm_drm_private *priv = msm_obj->base.dev->dev_private; + + WARN_ON(!mutex_is_locked(&priv->mm_lock)); + + if (is_unpurgable(msm_obj)) + return; + + if (WARN_ON(msm_obj->dontneed)) + return; + + priv->shrinkable_count += msm_obj->base.size >> PAGE_SHIFT; + msm_obj->dontneed = true; +} + +static inline void mark_unpurgable(struct msm_gem_object *msm_obj) +{ + struct msm_drm_private *priv = msm_obj->base.dev->dev_private; + + WARN_ON(!mutex_is_locked(&priv->mm_lock)); + + if (is_unpurgable(msm_obj)) + return; + + if (WARN_ON(!msm_obj->dontneed)) + return; + + priv->shrinkable_count -= msm_obj->base.size >> PAGE_SHIFT; + WARN_ON(priv->shrinkable_count < 0); + msm_obj->dontneed = false; +} + void msm_gem_purge(struct drm_gem_object *obj); void msm_gem_vunmap(struct drm_gem_object *obj);
diff --git a/drivers/gpu/drm/msm/msm_gem_shrinker.c b/drivers/gpu/drm/msm/msm_gem_shrinker.c index 9d5248be746f..f3e948af01c5 100644 --- a/drivers/gpu/drm/msm/msm_gem_shrinker.c +++ b/drivers/gpu/drm/msm/msm_gem_shrinker.c @@ -14,22 +14,7 @@ msm_gem_shrinker_count(struct shrinker *shrinker, struct shrink_control *sc) { struct msm_drm_private *priv = container_of(shrinker, struct msm_drm_private, shrinker); - struct msm_gem_object *msm_obj; - unsigned long count = 0; - - mutex_lock(&priv->mm_lock); - - list_for_each_entry(msm_obj, &priv->inactive_dontneed, mm_list) { - if (!msm_gem_trylock(&msm_obj->base)) - continue; - if (is_purgeable(msm_obj)) - count += msm_obj->base.size >> PAGE_SHIFT; - msm_gem_unlock(&msm_obj->base); - } - - mutex_unlock(&priv->mm_lock); - - return count; + return priv->shrinkable_count; }
static unsigned long @@ -45,6 +30,9 @@ msm_gem_shrinker_scan(struct shrinker *shrinker, struct shrink_control *sc) list_for_each_entry(msm_obj, &priv->inactive_dontneed, mm_list) { if (freed >= sc->nr_to_scan) break; + /* Use trylock, because we cannot block on a obj that + * might be trying to acquire mm_lock + */ if (!msm_gem_trylock(&msm_obj->base)) continue; if (is_purgeable(msm_obj)) { @@ -56,8 +44,11 @@ msm_gem_shrinker_scan(struct shrinker *shrinker, struct shrink_control *sc)
mutex_unlock(&priv->mm_lock);
- if (freed > 0) + if (freed > 0) { trace_msm_gem_purge(freed << PAGE_SHIFT); + } else { + return SHRINK_STOP; + }
return freed; } @@ -75,6 +66,9 @@ vmap_shrink(struct list_head *mm_list) unsigned unmapped = 0;
list_for_each_entry(msm_obj, mm_list, mm_list) { + /* Use trylock, because we cannot block on a obj that + * might be trying to acquire mm_lock + */ if (!msm_gem_trylock(&msm_obj->base)) continue; if (is_vunmapable(msm_obj)) {
linaro-mm-sig@lists.linaro.org