drm/msm: Shrinker fixes and opts
Subsystems (4)
cover drm/msm: Shrinker fixes and opts
From: Rob Clark <[email protected]>
The last patch is the main thing, motivated by some cases where we would spend a lot of time in msm_gem_shrinker_count(). First two are fixes I noticed along the way.
Rob Clark (3):
drm/msm: Protect obj->active_count under obj lock drm/msm/shrinker: We can vmap shrink active_list too drm/msm/shrinker: Only iterate dontneed objs
drivers/gpu/drm/msm/msm_debugfs.c | 3 +- drivers/gpu/drm/msm/msm_drv.c | 3 +- drivers/gpu/drm/msm/msm_drv.h | 8 ++-- drivers/gpu/drm/msm/msm_gem.c | 45 ++++++++++++++++------ drivers/gpu/drm/msm/msm_gem.h | 5 ++- drivers/gpu/drm/msm/msm_gem_shrinker.c | 52 +++++++++++++++++++------- drivers/gpu/drm/msm/msm_gpu.c | 10 +++-- 7 files changed, 89 insertions(+), 37 deletions(-)
Signature · 1 line
2.28.0
PATCH 1/3 drm/msm: Protect obj->active_count under obj lock
From: Rob Clark <[email protected]>
Previously we only held obj lock in the _active_get() path, and relied on atomic_dec_return() to not be racy in the _active_put() path where obj lock was not held.
But this is a false sense of security. Unlike obj lifetime refcnt, where you do not expect to *increase* the refcnt after the last put (which would mean that something has gone horribly wrong with the object liveness reference counting), the active_count can increase again from zero. Racing _active_put()s and _active_get()s could leave the obj on the wrong mm list.
But in the retire path, immediately after the _active_put(), the _unpin_iova() would acquire obj lock. So just move the locking earlier and rely on that to protect obj->active_count.
--- drivers/gpu/drm/msm/msm_gem.c | 11 ++++++----- drivers/gpu/drm/msm/msm_gem.h | 5 +++-- drivers/gpu/drm/msm/msm_gpu.c | 10 ++++++---- 3 files changed, 15 insertions(+), 11 deletions(-)
drivers/gpu/drm/msm/msm_gem.c +6 -5
@@ -770,7 +770,7 @@ void msm_gem_active_get(struct drm_gem_object *obj, struct msm_gpu *gpu)WARN_ON (!msm_gem_is_locked (obj));WARN_ON (msm_obj->madv !=MSM_MADV_WILLNEED ); -if (!atomic_fetch_inc (&msm_obj->active_count)) { +if (msm_obj->active_count++ ==0 ) {mutex_lock (&priv->mm_lock);list_del_init (&msm_obj->mm_list);list_add_tail (&msm_obj->mm_list, &gpu->active_list);
@@ -784,8 +784,9 @@ void msm_gem_active_put(struct drm_gem_object *obj)struct msm_drm_private *priv = obj->dev->dev_private;might_sleep (); +WARN_ON (!msm_gem_is_locked (obj)); -if (!atomic_dec_return (&msm_obj->active_count)) { +if (--msm_obj->active_count ==0 ) {mutex_lock (&priv->mm_lock);list_del_init (&msm_obj->mm_list);list_add_tail (&msm_obj->mm_list, &priv->inactive_list);
@@ -936,15 +937,15 @@ void msm_gem_free_object(struct drm_gem_object *obj)struct drm_device *dev = obj->dev;struct msm_drm_private *priv = dev->dev_private; -/* object should not be on active list: */ -WARN_ON (is_active (msm_obj)); -mutex_lock (&priv->mm_lock);list_del (&msm_obj->mm_list);mutex_unlock (&priv->mm_lock);msm_gem_lock (obj); +/* object should not be on active list: */ +WARN_ON (is_active (msm_obj)); +put_iova (obj);if (obj->import_attach) {
drivers/gpu/drm/msm/msm_gem.h +3 -2
@@ -87,7 +87,7 @@ struct msm_gem_object {char name[32 ];/* Identifier to print for the debugfs files */ -atomic_t active_count; +int active_count; };#define to_msm_bo (x) container_of(x, struct msm_gem_object, base)
@@ -185,7 +185,8 @@ msm_gem_is_locked(struct drm_gem_object *obj)static inline bool is_active (struct msm_gem_object *msm_obj) { -return atomic_read (&msm_obj->active_count); +WARN_ON (!msm_gem_is_locked (&msm_obj->base)); +return msm_obj->active_count; }static inline bool is_purgeable (struct msm_gem_object *msm_obj)
drivers/gpu/drm/msm/msm_gpu.c +6 -4
@@ -717,11 +717,13 @@ static void retire_submit(struct msm_gpu *gpu, struct msm_ringbuffer *ring, stats->alwayson_start, stats->alwayson_end);for (i =0 ; i < submit->nr_bos; i++) { -struct msm_gem_object *msm_obj = submit->bos[i].obj; +struct drm_gem_object *obj = &submit->bos[i].obj->base; -msm_gem_active_put (&msm_obj->base); -msm_gem_unpin_iova (&msm_obj->base, submit->aspace); -drm_gem_object_put (&msm_obj->base); +msm_gem_lock (obj); +msm_gem_active_put (obj); +msm_gem_unpin_iova_locked (obj, submit->aspace); +msm_gem_unlock (obj); +drm_gem_object_put (obj); }pm_runtime_mark_last_busy (&gpu->pdev->dev);
Signature · 6 lines
2.28.0 _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
PATCH 3/3 drm/msm/shrinker: Only iterate dontneed objs
In reply to Rob ClarkFrom: Rob Clark <[email protected]>
In situations where the GPU is mostly idle, all or nearly all buffer objects will be in the inactive list. But if the system is under memory pressure (from something other than GPU), we could still get a lot of shrinker calls. Which results in traversing a list of thousands of objs and in the end finding nothing to shrink. Which isn't so efficient.
Instead split the inactive_list into two lists, one inactive objs which are shrinkable, and a second one for those that are not. This way we can avoid traversing objs which we know are not shrinker candidates.
--- drivers/gpu/drm/msm/msm_debugfs.c | 3 ++- drivers/gpu/drm/msm/msm_drv.c | 3 ++- drivers/gpu/drm/msm/msm_drv.h | 8 +++--- drivers/gpu/drm/msm/msm_gem.c | 34 ++++++++++++++++++++------ drivers/gpu/drm/msm/msm_gem_shrinker.c | 7 +++--- 5 files changed, 40 insertions(+), 15 deletions(-)
drivers/gpu/drm/msm/msm_debugfs.c +2 -1
@@ -124,7 +124,8 @@ static int msm_gem_show(struct drm_device *dev, struct seq_file *m) }seq_printf (m,"Inactive Objects:\n" ); -msm_gem_describe_objects (&priv->inactive_list, m); +msm_gem_describe_objects (&priv->inactive_dontneed, m); +msm_gem_describe_objects (&priv->inactive_willneed, m);mutex_unlock (&priv->mm_lock);
drivers/gpu/drm/msm/msm_drv.c +2 -1
@@ -465,7 +465,8 @@ static int msm_drm_init(struct device *dev, struct drm_driver *drv) priv->wq =alloc_ordered_workqueue ("msm" ,0 ); -INIT_LIST_HEAD (&priv->inactive_list); +INIT_LIST_HEAD (&priv->inactive_willneed); +INIT_LIST_HEAD (&priv->inactive_dontneed);mutex_init (&priv->mm_lock);/* Teach lockdep about lock ordering wrt. shrinker: */
drivers/gpu/drm/msm/msm_drv.h +5 -3
@@ -175,8 +175,9 @@ struct msm_drm_private {struct msm_perf_state *perf;/* -* List of inactive GEM objects. Every bo is either in the inactive_list -* or gpu->active_list (for the gpu it is active on[1]) +* Lists of inactive GEM objects. Every bo is either in one of the +* 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
@@ -185,7 +186,8 @@ struct msm_drm_private {* [1] if someone ever added support for the old 2d cores, there could be * more than one gpu object */ -struct list_head inactive_list; +struct list_head inactive_willneed;/* inactive + !shrinkable */ +struct list_head inactive_dontneed;/* inactive + shrinkable */ struct mutex mm_lock;struct workqueue_struct *wq;
drivers/gpu/drm/msm/msm_gem.c +27 -7
@@ -17,6 +17,7 @@#include "msm_gpu.h" #include "msm_mmu.h" +static void update_inactive (struct msm_gem_object *msm_obj);static dma_addr_t physaddr (struct drm_gem_object *obj) {
@@ -678,6 +679,12 @@ int msm_gem_madvise(struct drm_gem_object *obj, unsigned madv) madv = msm_obj->madv; +/* If the obj is inactive, we might need to move it +* between inactive lists +*/ +if (msm_obj->active_count ==0 ) +update_inactive (msm_obj); +msm_gem_unlock (obj);return (madv != __MSM_MADV_PURGED);
@@ -781,19 +788,31 @@ void msm_gem_active_get(struct drm_gem_object *obj, struct msm_gpu *gpu)void msm_gem_active_put (struct drm_gem_object *obj) {struct msm_gem_object *msm_obj =to_msm_bo (obj); -struct msm_drm_private *priv = obj->dev->dev_private;might_sleep ();WARN_ON (!msm_gem_is_locked (obj));if (--msm_obj->active_count ==0 ) { -mutex_lock (&priv->mm_lock); -list_del_init (&msm_obj->mm_list); -list_add_tail (&msm_obj->mm_list, &priv->inactive_list); -mutex_unlock (&priv->mm_lock); +update_inactive (msm_obj); } } +static void update_inactive (struct msm_gem_object *msm_obj) +{ +struct msm_drm_private *priv = msm_obj->base.dev->dev_private; + +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_DONTNEED ) +list_add_tail (&msm_obj->mm_list, &priv->inactive_willneed); +else +list_add_tail (&msm_obj->mm_list, &priv->inactive_dontneed); + +mutex_unlock (&priv->mm_lock); +} +int msm_gem_cpu_prep (struct drm_gem_object *obj,uint32_t op,ktime_t *timeout) {bool write = !!(op &MSM_PREP_WRITE );
@@ -1099,7 +1118,8 @@ static struct drm_gem_object *_msm_gem_new(struct drm_device *dev, }mutex_lock (&priv->mm_lock); -list_add_tail (&msm_obj->mm_list, &priv->inactive_list); +/* Initially obj is idle, obj->madv == WILLNEED: */ +list_add_tail (&msm_obj->mm_list, &priv->inactive_willneed);mutex_unlock (&priv->mm_lock);return obj;
@@ -1169,7 +1189,7 @@ struct drm_gem_object *msm_gem_import(struct drm_device *dev,msm_gem_unlock (obj);mutex_lock (&priv->mm_lock); -list_add_tail (&msm_obj->mm_list, &priv->inactive_list); +list_add_tail (&msm_obj->mm_list, &priv->inactive_willneed);mutex_unlock (&priv->mm_lock);return obj;
drivers/gpu/drm/msm/msm_gem_shrinker.c +4 -3
@@ -19,7 +19,7 @@ msm_gem_shrinker_count(struct shrinker *shrinker, struct shrink_control *sc)mutex_lock (&priv->mm_lock); -list_for_each_entry (msm_obj, &priv->inactive_list, mm_list) { +list_for_each_entry (msm_obj, &priv->inactive_dontneed, mm_list) {if (!msm_gem_trylock (&msm_obj->base))continue ;if (is_purgeable (msm_obj))
@@ -42,7 +42,7 @@ msm_gem_shrinker_scan(struct shrinker *shrinker, struct shrink_control *sc)mutex_lock (&priv->mm_lock); -list_for_each_entry (msm_obj, &priv->inactive_list, mm_list) { +list_for_each_entry (msm_obj, &priv->inactive_dontneed, mm_list) {if (freed >= sc->nr_to_scan)break ;if (!msm_gem_trylock (&msm_obj->base))
@@ -96,7 +96,8 @@ msm_gem_shrinker_vmap(struct notifier_block *nb, unsigned long event, void *ptr)struct msm_drm_private *priv = container_of(nb, struct msm_drm_private, vmap_notifier);struct list_head *mm_lists[] = { - &priv->inactive_list, + &priv->inactive_dontneed, + &priv->inactive_willneed, priv->gpu ? &priv->gpu->active_list :NULL ,NULL , };
Signature · 6 lines
2.28.0 _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
PATCH 2/3 drm/msm/shrinker: We can vmap shrink active_list too
In reply to Rob ClarkFrom: Rob Clark <[email protected]>
Just because a obj is active, if the vmap_count is zero, we can still tear down the vmap.
--- drivers/gpu/drm/msm/msm_gem_shrinker.c | 47 +++++++++++++++++++------- 1 file changed, 35 insertions(+), 12 deletions(-)
drivers/gpu/drm/msm/msm_gem_shrinker.c +35 -12
@@ -6,6 +6,7 @@#include "msm_drv.h" #include "msm_gem.h" +#include "msm_gpu.h" #include "msm_gpu_trace.h" static unsigned long
@@ -61,17 +62,19 @@ msm_gem_shrinker_scan(struct shrinker *shrinker, struct shrink_control *sc)return freed; } -static int -msm_gem_shrinker_vmap (struct notifier_block *nb,unsigned long event,void *ptr) +/* since we don't know any better, lets bail after a few +* and if necessary the shrinker will be invoked again. +* Seems better than unmapping *everything* +*/ +static const int vmap_shrink_limit =15 ; + +static unsigned +vmap_shrink (struct list_head *mm_list) { -struct msm_drm_private *priv = - container_of(nb, struct msm_drm_private, vmap_notifier);struct msm_gem_object *msm_obj;unsigned unmapped =0 ; -mutex_lock (&priv->mm_lock); - -list_for_each_entry (msm_obj, &priv->inactive_list, mm_list) { +list_for_each_entry (msm_obj, mm_list, mm_list) {if (!msm_gem_trylock (&msm_obj->base))continue ;if (is_vunmapable (msm_obj)) {
@@ -80,11 +83,31 @@ msm_gem_shrinker_vmap(struct notifier_block *nb, unsigned long event, void *ptr) }msm_gem_unlock (&msm_obj->base); -/* since we don't know any better, lets bail after a few -* and if necessary the shrinker will be invoked again. -* Seems better than unmapping *everything* -*/ -if (++unmapped >=15 ) +if (++unmapped >= vmap_shrink_limit) +break ; + } + +return unmapped; +} + +static int +msm_gem_shrinker_vmap (struct notifier_block *nb,unsigned long event,void *ptr) +{ +struct msm_drm_private *priv = + container_of(nb, struct msm_drm_private, vmap_notifier); +struct list_head *mm_lists[] = { + &priv->inactive_list, + priv->gpu ? &priv->gpu->active_list :NULL , +NULL , + }; +unsigned idx, unmapped =0 ; + +mutex_lock (&priv->mm_lock); + +for (idx =0 ; mm_lists[idx]; idx++) { + unmapped +=vmap_shrink (mm_lists[idx]); + +if (unmapped >= vmap_shrink_limit)break ; }
Signature · 6 lines
2.28.0 _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
Re: [PATCH 3/3] drm/msm/shrinker: Only iterate dontneed objs
In reply to Rob ClarkQuoted 120 lines · Rob Clark
From: Rob Clark <[email protected]>
In situations where the GPU is mostly idle, all or nearly all buffer objects will be in the inactive list. But if the system is under memory pressure (from something other than GPU), we could still get a lot of shrinker calls. Which results in traversing a list of thousands of objs and in the end finding nothing to shrink. Which isn't so efficient.
Instead split the
inactive_listinto two lists, one inactive objs which are shrinkable, and a second one for those that are not. This way we can avoid traversing objs which we know are not shrinker candidates.Signed-off-by Rob Clark <[email protected]>--- drivers/gpu/drm/msm/msm_debugfs.c | 3 ++- drivers/gpu/drm/msm/msm_drv.c | 3 ++- drivers/gpu/drm/msm/msm_drv.h | 8 +++--- drivers/gpu/drm/msm/msm_gem.c | 34 ++++++++++++++++++++------ drivers/gpu/drm/msm/msm_gem_shrinker.c | 7 +++--- 5 files changed, 40 insertions(+), 15 deletions(-)drivers/gpu/drm/msm/msm_debugfs.c +2
-1@@ -124,7 +124,8 @@ static int msm_gem_show(struct drm_device *dev, struct seq_file *m) }seq_printf (m,"Inactive Objects:\n" ); -msm_gem_describe_objects (&priv->inactive_list, m); +msm_gem_describe_objects (&priv->inactive_dontneed, m); +msm_gem_describe_objects (&priv->inactive_willneed, m);mutex_unlock (&priv->mm_lock);drivers/gpu/drm/msm/msm_drv.c +2
-1@@ -465,7 +465,8 @@ static int msm_drm_init(struct device *dev, struct drm_driver *drv) priv->wq =alloc_ordered_workqueue ("msm" ,0 ); -INIT_LIST_HEAD (&priv->inactive_list); +INIT_LIST_HEAD (&priv->inactive_willneed); +INIT_LIST_HEAD (&priv->inactive_dontneed);mutex_init (&priv->mm_lock);/* Teach lockdep about lock ordering wrt. shrinker: */ drivers/gpu/drm/msm/msm_drv.h +5
-3@@ -175,8 +175,9 @@ struct msm_drm_private {struct msm_perf_state *perf;/* -* List of inactive GEM objects. Every bo is either in the inactive_list -* or gpu->active_list (for the gpu it is active on[1]) +* Lists of inactive GEM objects. Every bo is either in one of the +* 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 @@ -185,7 +186,8 @@ struct msm_drm_private {* [1] if someone ever added support for the old 2d cores, there could be * more than one gpu object */ -struct list_head inactive_list; +struct list_head inactive_willneed;/* inactive + !shrinkable */ +struct list_head inactive_dontneed;/* inactive + shrinkable */ struct mutex mm_lock;struct workqueue_struct *wq;drivers/gpu/drm/msm/msm_gem.c +10
-5@@ -17,6 +17,7 @@#include "msm_gpu.h" #include "msm_mmu.h" +static void update_inactive (struct msm_gem_object *msm_obj);static dma_addr_t physaddr (struct drm_gem_object *obj) {@@ -678,6 +679,12 @@ int msm_gem_madvise(struct drm_gem_object *obj, unsigned madv) madv = msm_obj->madv; +/* If the obj is inactive, we might need to move it +* between inactive lists +*/ +if (msm_obj->active_count ==0 ) +update_inactive (msm_obj); +msm_gem_unlock (obj);return (madv != __MSM_MADV_PURGED);@@ -781,19 +788,31 @@ void msm_gem_active_get(struct drm_gem_object *obj, struct msm_gpu *gpu)void msm_gem_active_put (struct drm_gem_object *obj) {struct msm_gem_object *msm_obj =to_msm_bo (obj); -struct msm_drm_private *priv = obj->dev->dev_private;might_sleep ();WARN_ON (!msm_gem_is_locked (obj));if (--msm_obj->active_count ==0 ) { -mutex_lock (&priv->mm_lock); -list_del_init (&msm_obj->mm_list); -list_add_tail (&msm_obj->mm_list, &priv->inactive_list); -mutex_unlock (&priv->mm_lock); +update_inactive (msm_obj); } } +static void update_inactive (struct msm_gem_object *msm_obj) +{
drivers/gpu/drm/msm/msm_gem.c void msm_gem_active_get(struct drm_gem_object *obj, struct msm_gpu *gpu) · line 811
Quoted 10 lines
drivers/gpu/drm/msm/msm_gem.c +10
-0+struct msm_drm_private *priv = msm_obj->base.dev->dev_private; + +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_DONTNEED ) +list_add_tail (&msm_obj->mm_list, &priv->inactive_willneed); +else +list_add_tail (&msm_obj->mm_list, &priv->inactive_dontneed);
Is the logic here inverted or is this just really confusing nomenclature? If it is correct a comment might help remind us whats happening.
Jordan
Quoted 65 lines
drivers/gpu/drm/msm/msm_gem.c +7
-2+ +mutex_unlock (&priv->mm_lock); +} +int msm_gem_cpu_prep (struct drm_gem_object *obj,uint32_t op,ktime_t *timeout) {bool write = !!(op &MSM_PREP_WRITE );@@ -1099,7 +1118,8 @@ static struct drm_gem_object *_msm_gem_new(struct drm_device *dev, }mutex_lock (&priv->mm_lock); -list_add_tail (&msm_obj->mm_list, &priv->inactive_list); +/* Initially obj is idle, obj->madv == WILLNEED: */ +list_add_tail (&msm_obj->mm_list, &priv->inactive_willneed);mutex_unlock (&priv->mm_lock);return obj;@@ -1169,7 +1189,7 @@ struct drm_gem_object *msm_gem_import(struct drm_device *dev,msm_gem_unlock (obj);mutex_lock (&priv->mm_lock); -list_add_tail (&msm_obj->mm_list, &priv->inactive_list); +list_add_tail (&msm_obj->mm_list, &priv->inactive_willneed);mutex_unlock (&priv->mm_lock);return obj;drivers/gpu/drm/msm/msm_gem_shrinker.c +4
-3@@ -19,7 +19,7 @@ msm_gem_shrinker_count(struct shrinker *shrinker, struct shrink_control *sc)mutex_lock (&priv->mm_lock); -list_for_each_entry (msm_obj, &priv->inactive_list, mm_list) { +list_for_each_entry (msm_obj, &priv->inactive_dontneed, mm_list) {if (!msm_gem_trylock (&msm_obj->base))continue ;if (is_purgeable (msm_obj))@@ -42,7 +42,7 @@ msm_gem_shrinker_scan(struct shrinker *shrinker, struct shrink_control *sc)mutex_lock (&priv->mm_lock); -list_for_each_entry (msm_obj, &priv->inactive_list, mm_list) { +list_for_each_entry (msm_obj, &priv->inactive_dontneed, mm_list) {if (freed >= sc->nr_to_scan)break ;if (!msm_gem_trylock (&msm_obj->base))@@ -96,7 +96,8 @@ msm_gem_shrinker_vmap(struct notifier_block *nb, unsigned long event, void *ptr)struct msm_drm_private *priv = container_of(nb, struct msm_drm_private, vmap_notifier);struct list_head *mm_lists[] = { - &priv->inactive_list, + &priv->inactive_dontneed, + &priv->inactive_willneed, priv->gpu ? &priv->gpu->active_list :NULL ,NULL , };Signature · 6 lines
2.28.0 _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
Signature · 6 lines
The Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
Re: [Freedreno] [PATCH 3/3] drm/msm/shrinker: Only iterate dontneed objs
Quoted 120 lines · Jordan Crouse
Quoted 118 lines · Rob Clark
From: Rob Clark <[email protected]>
In situations where the GPU is mostly idle, all or nearly all buffer objects will be in the inactive list. But if the system is under memory pressure (from something other than GPU), we could still get a lot of shrinker calls. Which results in traversing a list of thousands of objs and in the end finding nothing to shrink. Which isn't so efficient.
Instead split the
inactive_listinto two lists, one inactive objs which are shrinkable, and a second one for those that are not. This way we can avoid traversing objs which we know are not shrinker candidates.Signed-off-by Rob Clark <[email protected]>--- drivers/gpu/drm/msm/msm_debugfs.c | 3 ++- drivers/gpu/drm/msm/msm_drv.c | 3 ++- drivers/gpu/drm/msm/msm_drv.h | 8 +++--- drivers/gpu/drm/msm/msm_gem.c | 34 ++++++++++++++++++++------ drivers/gpu/drm/msm/msm_gem_shrinker.c | 7 +++--- 5 files changed, 40 insertions(+), 15 deletions(-)drivers/gpu/drm/msm/msm_debugfs.c +2
-1@@ -124,7 +124,8 @@ static int msm_gem_show(struct drm_device *dev, struct seq_file *m) }seq_printf (m,"Inactive Objects:\n" ); -msm_gem_describe_objects (&priv->inactive_list, m); +msm_gem_describe_objects (&priv->inactive_dontneed, m); +msm_gem_describe_objects (&priv->inactive_willneed, m);mutex_unlock (&priv->mm_lock);drivers/gpu/drm/msm/msm_drv.c +2
-1@@ -465,7 +465,8 @@ static int msm_drm_init(struct device *dev, struct drm_driver *drv) priv->wq =alloc_ordered_workqueue ("msm" ,0 ); -INIT_LIST_HEAD (&priv->inactive_list); +INIT_LIST_HEAD (&priv->inactive_willneed); +INIT_LIST_HEAD (&priv->inactive_dontneed);mutex_init (&priv->mm_lock);/* Teach lockdep about lock ordering wrt. shrinker: */ drivers/gpu/drm/msm/msm_drv.h +5
-3@@ -175,8 +175,9 @@ struct msm_drm_private {struct msm_perf_state *perf;/* -* List of inactive GEM objects. Every bo is either in the inactive_list -* or gpu->active_list (for the gpu it is active on[1]) +* Lists of inactive GEM objects. Every bo is either in one of the +* 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 @@ -185,7 +186,8 @@ struct msm_drm_private {* [1] if someone ever added support for the old 2d cores, there could be * more than one gpu object */ -struct list_head inactive_list; +struct list_head inactive_willneed;/* inactive + !shrinkable */ +struct list_head inactive_dontneed;/* inactive + shrinkable */ struct mutex mm_lock;struct workqueue_struct *wq;drivers/gpu/drm/msm/msm_gem.c +8
-5@@ -17,6 +17,7 @@#include "msm_gpu.h" #include "msm_mmu.h" +static void update_inactive (struct msm_gem_object *msm_obj);static dma_addr_t physaddr (struct drm_gem_object *obj) {@@ -678,6 +679,12 @@ int msm_gem_madvise(struct drm_gem_object *obj, unsigned madv) madv = msm_obj->madv; +/* If the obj is inactive, we might need to move it +* between inactive lists +*/ +if (msm_obj->active_count ==0 ) +update_inactive (msm_obj); +msm_gem_unlock (obj);return (madv != __MSM_MADV_PURGED);@@ -781,19 +788,31 @@ void msm_gem_active_get(struct drm_gem_object *obj, struct msm_gpu *gpu)void msm_gem_active_put (struct drm_gem_object *obj) {struct msm_gem_object *msm_obj =to_msm_bo (obj); -struct msm_drm_private *priv = obj->dev->dev_private;might_sleep ();WARN_ON (!msm_gem_is_locked (obj));if (--msm_obj->active_count ==0 ) { -mutex_lock (&priv->mm_lock); -list_del_init (&msm_obj->mm_list); -list_add_tail (&msm_obj->mm_list, &priv->inactive_list); -mutex_unlock (&priv->mm_lock); +update_inactive (msm_obj); } }
Quoted 15 lines
drivers/gpu/drm/msm/msm_gem.c void msm_gem_active_get(struct drm_gem_object *obj, struct msm_gpu *gpu) · line 811
Quoted 12 lines
drivers/gpu/drm/msm/msm_gem.c +12
-0+static void update_inactive (struct msm_gem_object *msm_obj) +{ +struct msm_drm_private *priv = msm_obj->base.dev->dev_private; + +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_DONTNEED ) +list_add_tail (&msm_obj->mm_list, &priv->inactive_willneed); +else +list_add_tail (&msm_obj->mm_list, &priv->inactive_dontneed);Is the logic here inverted or is this just really confusing nomenclature? If it is correct a comment might help remind us whats happening.
Oh, whoops, yeah that is inverted
BR,
-R
Quoted 63 lines
Jordan
Quoted 61 lines
drivers/gpu/drm/msm/msm_gem.c +7
-2+ +mutex_unlock (&priv->mm_lock); +} +int msm_gem_cpu_prep (struct drm_gem_object *obj,uint32_t op,ktime_t *timeout) {bool write = !!(op &MSM_PREP_WRITE );@@ -1099,7 +1118,8 @@ static struct drm_gem_object *_msm_gem_new(struct drm_device *dev, }mutex_lock (&priv->mm_lock); -list_add_tail (&msm_obj->mm_list, &priv->inactive_list); +/* Initially obj is idle, obj->madv == WILLNEED: */ +list_add_tail (&msm_obj->mm_list, &priv->inactive_willneed);mutex_unlock (&priv->mm_lock);return obj;@@ -1169,7 +1189,7 @@ struct drm_gem_object *msm_gem_import(struct drm_device *dev,msm_gem_unlock (obj);mutex_lock (&priv->mm_lock); -list_add_tail (&msm_obj->mm_list, &priv->inactive_list); +list_add_tail (&msm_obj->mm_list, &priv->inactive_willneed);mutex_unlock (&priv->mm_lock);return obj;drivers/gpu/drm/msm/msm_gem_shrinker.c +4
-4@@ -19,7 +19,7 @@ msm_gem_shrinker_count(struct shrinker *shrinker, struct shrink_control *sc)mutex_lock (&priv->mm_lock); -list_for_each_entry (msm_obj, &priv->inactive_list, mm_list) { +list_for_each_entry (msm_obj, &priv->inactive_dontneed, mm_list) {if (!msm_gem_trylock (&msm_obj->base))continue ;if (is_purgeable (msm_obj))@@ -42,7 +42,7 @@ msm_gem_shrinker_scan(struct shrinker *shrinker, struct shrink_control *sc)mutex_lock (&priv->mm_lock); -list_for_each_entry (msm_obj, &priv->inactive_list, mm_list) { +list_for_each_entry (msm_obj, &priv->inactive_dontneed, mm_list) {if (freed >= sc->nr_to_scan)break ;if (!msm_gem_trylock (&msm_obj->base))@@ -96,7 +96,8 @@ msm_gem_shrinker_vmap(struct notifier_block *nb, unsigned long event, void *ptr)struct msm_drm_private *priv = container_of(nb, struct msm_drm_private, vmap_notifier);struct list_head *mm_lists[] = { - &priv->inactive_list, + &priv->inactive_dontneed, + &priv->inactive_willneed, priv->gpu ? &priv->gpu->active_list :NULL ,NULL , }; --2.28.0
Quoted 12 lines
Quoted 4 lines
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel--
The Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project
_______________________________________________
Freedreno mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/freedreno
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel