Skip to contentSkip to conversation

drm/ttm: remove some bo->mutex remains

PATCH NAK Maarten Maathuis 3 participants 6 messages Mar 1, 2010 – Mar 12, 2010

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/2 drm/ttm: remove some bo->mutex remains

- A few comments existed here and there that referred to a bo->mutex.

---
 drivers/gpu/drm/ttm/ttm_bo.c    |    6 +-----
 drivers/gpu/drm/ttm/ttm_bo_vm.c |    2 +-
 include/drm/ttm/ttm_bo_api.h    |    3 +--
 include/drm/ttm/ttm_bo_driver.h |    1 +
 4 files changed, 4 insertions(+), 8 deletions(-)
drivers/gpu/drm/ttm/ttm_bo.c +1 -5
index 2920f9a..f5333d9 100644
@@ -46,7 +46,6 @@
 #include <linux/file.h>
 #include <linux/module.h>
 
-#define TTM_ASSERT_LOCKED(param)
 #define TTM_DEBUG(fmt, arg...)
 #define TTM_BO_HASH_ORDER 13
 
@@ -306,9 +305,6 @@ void ttm_bo_unreserve(struct ttm_buffer_object *bo)
 }
 EXPORT_SYMBOL(ttm_bo_unreserve);
 
-/*
- * Call bo->mutex locked.
- */
 static int ttm_bo_add_ttm(struct ttm_buffer_object *bo, bool zero_alloc)
 {
 	struct ttm_bo_device *bdev = bo->bdev;
@@ -316,7 +312,7 @@ static int ttm_bo_add_ttm(struct ttm_buffer_object *bo, bool zero_alloc)
 	int ret = 0;
 	uint32_t page_flags = 0;
 
-	TTM_ASSERT_LOCKED(&bo->mutex);
+	BUG_ON(!atomic_read(&bo->reserved));
 	bo->ttm = NULL;
 
 	if (bdev->need_dma32)
drivers/gpu/drm/ttm/ttm_bo_vm.c +1 -1
index 668dbe8..41b0c1e 100644
@@ -146,7 +146,7 @@ static int ttm_bo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 	 * since the mmap_sem is only held in read mode. However, we
 	 * modify only the caching bits of vma->vm_page_prot and
 	 * consider those bits protected by
-	 * the bo->mutex, as we should be the only writers.
+	 * bo->reserved, as we should be the only writers.
 	 * There shouldn't really be any readers of these bits except
 	 * within vm_insert_mixed()? fork?
 	 *
include/drm/ttm/ttm_bo_api.h +1 -2
index 81eb9f4..c1093ae 100644
@@ -35,7 +35,6 @@
 #include <linux/kref.h>
 #include <linux/list.h>
 #include <linux/wait.h>
-#include <linux/mutex.h>
 #include <linux/mm.h>
 #include <linux/rbtree.h>
 #include <linux/bitmap.h>
@@ -298,7 +297,7 @@ ttm_bo_reference(struct ttm_buffer_object *bo)
  * @interruptible:  Use interruptible wait.
  * @no_wait:  Return immediately if buffer is busy.
  *
- * This function must be called with the bo::mutex held, and makes
+ * This function must be called with bo->reserved held, and makes
  * sure any previous rendering to the buffer is completed.
  * Note: It might be necessary to block validations before the
  * wait by reserving the buffer.
include/drm/ttm/ttm_bo_driver.h +1 -0
index ff7664e..d3fc5f8 100644
@@ -37,6 +37,7 @@
 #include "linux/workqueue.h"
 #include "linux/fs.h"
 #include "linux/spinlock.h"
+#include <linux/mutex.h>
 
 struct ttm_backend;
 
Signature · 10 lines
1.7.0


------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
--
#1 lore raw

PATCH 2/2 drm/ttm: don't write to bo->reserved without holding glob->lru_lock

- The headerfile says you can't write to it without holding the lock.

---
 drivers/gpu/drm/ttm/ttm_bo.c |    8 +++++++-
 1 files changed, 7 insertions(+), 1 deletions(-)
drivers/gpu/drm/ttm/ttm_bo.c +7 -1
index f5333d9..2104885 100644
@@ -476,9 +476,9 @@ static int ttm_bo_cleanup_refs(struct ttm_buffer_object *bo, bool remove_all)
 			drm_mm_put_block(bo->mem.mm_node);
 			bo->mem.mm_node = NULL;
 		}
-		spin_unlock(&glob->lru_lock);
 
 		atomic_set(&bo->reserved, 0);
+		spin_unlock(&glob->lru_lock);
 
 		while (put_count--)
 			kref_put(&bo->list_kref, ttm_bo_ref_bug);
@@ -1707,8 +1707,12 @@ EXPORT_SYMBOL(ttm_bo_wait);
 
 void ttm_bo_unblock_reservation(struct ttm_buffer_object *bo)
 {
+	struct ttm_bo_global *glob = bo->glob;
+
+	spin_lock(&glob->lru_lock);
 	atomic_set(&bo->reserved, 0);
 	wake_up_all(&bo->event_queue);
+	spin_unlock(&glob->lru_lock);
 }
 
 int ttm_bo_block_reservation(struct ttm_buffer_object *bo, bool interruptible,
@@ -1849,8 +1853,10 @@ out:
 	 * already swapped buffer.
 	 */
 
+	spin_lock(&glob->lru_lock);
 	atomic_set(&bo->reserved, 0);
 	wake_up_all(&bo->event_queue);
+	spin_unlock(&glob->lru_lock);
 	kref_put(&bo->list_kref, ttm_bo_release_list);
 	return ret;
 }
Signature · 10 lines
1.7.0


------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
--
#2 lore raw

Re: [PATCH 2/2] drm/ttm: don't write to bo->reserved without holding glob->lru_lock

Quoted 3 lines · Maarten Maathuis

- The headerfile says you can't write to it without holding the lock.

NAK, from my POV as we always use atomic_* on reserved it's useless to protect it with spinlock.

Cheers,
Jerome

Quoted 59 lines
---
 drivers/gpu/drm/ttm/ttm_bo.c |    8 +++++++-
 1 files changed, 7 insertions(+), 1 deletions(-)
drivers/gpu/drm/ttm/ttm_bo.c +7 -1
index f5333d9..2104885 100644
@@ -476,9 +476,9 @@ static int ttm_bo_cleanup_refs(struct ttm_buffer_object *bo, bool remove_all)
 			drm_mm_put_block(bo->mem.mm_node);
 			bo->mem.mm_node = NULL;
 		}
-		spin_unlock(&glob->lru_lock);
 
 		atomic_set(&bo->reserved, 0);
+		spin_unlock(&glob->lru_lock);
 
 		while (put_count--)
 			kref_put(&bo->list_kref, ttm_bo_ref_bug);
@@ -1707,8 +1707,12 @@ EXPORT_SYMBOL(ttm_bo_wait);
 
 void ttm_bo_unblock_reservation(struct ttm_buffer_object *bo)
 {
+	struct ttm_bo_global *glob = bo->glob;
+
+	spin_lock(&glob->lru_lock);
 	atomic_set(&bo->reserved, 0);
 	wake_up_all(&bo->event_queue);
+	spin_unlock(&glob->lru_lock);
 }
 
 int ttm_bo_block_reservation(struct ttm_buffer_object *bo, bool interruptible,
@@ -1849,8 +1853,10 @@ out:
 	 * already swapped buffer.
 	 */
 
+	spin_lock(&glob->lru_lock);
 	atomic_set(&bo->reserved, 0);
 	wake_up_all(&bo->event_queue);
+	spin_unlock(&glob->lru_lock);
 	kref_put(&bo->list_kref, ttm_bo_release_list);
 	return ret;
 }
Signature · 14 lines
1.7.0


------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
--
_______________________________________________
Dri-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/dri-devel

------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs proactively, and fine-tune applications for parallel performance. See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
--

#3 lore raw

Re: [PATCH 2/2] drm/ttm: don't write to bo->reserved without holding glob->lru_lock

In reply to Maarten Maathuis
Quoted 2 lines · Maarten Maathuis

- The headerfile says you can't write to it without holding the lock.

NAK.

The header-file is incorrect. It should say that the lru_lock needs to be held only when we transition from 0 to 1.

That guarantees that if the lru lock is held, and we read 0, we can safely assume that we're the only ones trying to lock.

/Thomas

Quoted 35 lines
---
 drivers/gpu/drm/ttm/ttm_bo.c |    8 +++++++-
 1 files changed, 7 insertions(+), 1 deletions(-)
drivers/gpu/drm/ttm/ttm_bo.c +5 -1
index f5333d9..2104885 100644
@@ -476,9 +476,9 @@ static int ttm_bo_cleanup_refs(struct ttm_buffer_object *bo, bool remove_all)
 			drm_mm_put_block(bo->mem.mm_node);
 			bo->mem.mm_node = NULL;
 		}
-		spin_unlock(&glob->lru_lock);
 
 		atomic_set(&bo->reserved, 0);
+		spin_unlock(&glob->lru_lock);
 
 		while (put_count--)
 			kref_put(&bo->list_kref, ttm_bo_ref_bug);
@@ -1707,8 +1707,12 @@ EXPORT_SYMBOL(ttm_bo_wait);
 
 void ttm_bo_unblock_reservation(struct ttm_buffer_object *bo)
 {
+	struct ttm_bo_global *glob = bo->glob;
+
+	spin_lock(&glob->lru_lock);
 	atomic_set(&bo->reserved, 0);
 	wake_up_all(&bo->event_queue);
+	spin_unlock(&glob->lru_lock);
 }
 
 int ttm_bo_block_reservation(struct ttm_buffer_object *bo, bool interruptible,

drivers/gpu/drm/ttm/ttm_bo.c out: · line 1862

Quoted 10 lines
drivers/gpu/drm/ttm/ttm_bo.c +2 -0
@@ -1849,8 +1853,10 @@ out:
 	 * already swapped buffer.
 	 */
 
+	spin_lock(&glob->lru_lock);
 	atomic_set(&bo->reserved, 0);
 	wake_up_all(&bo->event_queue);
+	spin_unlock(&glob->lru_lock);
 	kref_put(&bo->list_kref, ttm_bo_release_list);
 	return ret;
 }

------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs proactively, and fine-tune applications for parallel performance. See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
--

Reviewed-by on v1 1/2
#4 lore raw
In reply to Maarten Maathuis
Quoted 3 lines · Maarten Maathuis

- A few comments existed here and there that referred to a bo->mutex.

Quoted 100 lines
---
 drivers/gpu/drm/ttm/ttm_bo.c    |    6 +-----
 drivers/gpu/drm/ttm/ttm_bo_vm.c |    2 +-
 include/drm/ttm/ttm_bo_api.h    |    3 +--
 include/drm/ttm/ttm_bo_driver.h |    1 +
 4 files changed, 4 insertions(+), 8 deletions(-)
drivers/gpu/drm/ttm/ttm_bo.c +1 -5
index 2920f9a..f5333d9 100644
@@ -46,7 +46,6 @@
 #include <linux/file.h>
 #include <linux/module.h>
 
-#define TTM_ASSERT_LOCKED(param)
 #define TTM_DEBUG(fmt, arg...)
 #define TTM_BO_HASH_ORDER 13
 
@@ -306,9 +305,6 @@ void ttm_bo_unreserve(struct ttm_buffer_object *bo)
 }
 EXPORT_SYMBOL(ttm_bo_unreserve);
 
-/*
- * Call bo->mutex locked.
- */
 static int ttm_bo_add_ttm(struct ttm_buffer_object *bo, bool zero_alloc)
 {
 	struct ttm_bo_device *bdev = bo->bdev;
@@ -316,7 +312,7 @@ static int ttm_bo_add_ttm(struct ttm_buffer_object *bo, bool zero_alloc)
 	int ret = 0;
 	uint32_t page_flags = 0;
 
-	TTM_ASSERT_LOCKED(&bo->mutex);
+	BUG_ON(!atomic_read(&bo->reserved));
 	bo->ttm = NULL;
 
 	if (bdev->need_dma32)
drivers/gpu/drm/ttm/ttm_bo_vm.c +1 -1
index 668dbe8..41b0c1e 100644
@@ -146,7 +146,7 @@ static int ttm_bo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 	 * since the mmap_sem is only held in read mode. However, we
 	 * modify only the caching bits of vma->vm_page_prot and
 	 * consider those bits protected by
-	 * the bo->mutex, as we should be the only writers.
+	 * bo->reserved, as we should be the only writers.
 	 * There shouldn't really be any readers of these bits except
 	 * within vm_insert_mixed()? fork?
 	 *
include/drm/ttm/ttm_bo_api.h +1 -2
index 81eb9f4..c1093ae 100644
@@ -35,7 +35,6 @@
 #include <linux/kref.h>
 #include <linux/list.h>
 #include <linux/wait.h>
-#include <linux/mutex.h>
 #include <linux/mm.h>
 #include <linux/rbtree.h>
 #include <linux/bitmap.h>
@@ -298,7 +297,7 @@ ttm_bo_reference(struct ttm_buffer_object *bo)
  * @interruptible:  Use interruptible wait.
  * @no_wait:  Return immediately if buffer is busy.
  *
- * This function must be called with the bo::mutex held, and makes
+ * This function must be called with bo->reserved held, and makes
  * sure any previous rendering to the buffer is completed.
  * Note: It might be necessary to block validations before the
  * wait by reserving the buffer.
include/drm/ttm/ttm_bo_driver.h +1 -0
index ff7664e..d3fc5f8 100644
@@ -37,6 +37,7 @@
 #include "linux/workqueue.h"
 #include "linux/fs.h"
 #include "linux/spinlock.h"
+#include <linux/mutex.h>
 
 struct ttm_backend;
 
Signature · 14 lines
1.7.0


------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
--
_______________________________________________
Dri-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/dri-devel

------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs proactively, and fine-tune applications for parallel performance. See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
--

#5 lore raw

Ping?

Quoted 95 lines · Jerome Glisse
Quoted 3 lines · Maarten Maathuis

- A few comments existed here and there that referred to a bo->mutex.

Quoted 88 lines

---
 drivers/gpu/drm/ttm/ttm_bo.c    |    6 +-----
 drivers/gpu/drm/ttm/ttm_bo_vm.c |    2 +-
 include/drm/ttm/ttm_bo_api.h    |    3 +--
 include/drm/ttm/ttm_bo_driver.h |    1 +

 4 files changed, 4 insertions(+), 8 deletions(-)
drivers/gpu/drm/ttm/ttm_bo.c +1 -5
index 2920f9a..f5333d9 100644
@@ -46,7 +46,6 @@
  #include <linux/file.h>
  #include <linux/module.h>
 
-#define TTM_ASSERT_LOCKED(param)
  #define TTM_DEBUG(fmt, arg...)
  #define TTM_BO_HASH_ORDER 13
 
@@ -306,9 +305,6 @@ void ttm_bo_unreserve(struct ttm_buffer_object *bo)
  }
  EXPORT_SYMBOL(ttm_bo_unreserve);
 
-/*
- * Call bo->mutex locked.
- */
  static int ttm_bo_add_ttm(struct ttm_buffer_object *bo, bool zero_alloc)
  {
       struct ttm_bo_device *bdev = bo->bdev;
@@ -316,7 +312,7 @@ static int ttm_bo_add_ttm(struct ttm_buffer_object *bo, bool zero_alloc)
       int ret = 0;
       uint32_t page_flags = 0;
 
-     TTM_ASSERT_LOCKED(&bo->mutex);
+     BUG_ON(!atomic_read(&bo->reserved));
       bo->ttm = NULL;
 
       if (bdev->need_dma32)
drivers/gpu/drm/ttm/ttm_bo_vm.c +1 -1
index 668dbe8..41b0c1e 100644
@@ -146,7 +146,7 @@ static int ttm_bo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
        * since the mmap_sem is only held in read mode. However, we
        * modify only the caching bits of vma->vm_page_prot and
        * consider those bits protected by
-      * the bo->mutex, as we should be the only writers.
+      * bo->reserved, as we should be the only writers.
        * There shouldn't really be any readers of these bits except
        * within vm_insert_mixed()? fork?
        *
include/drm/ttm/ttm_bo_api.h +1 -2
index 81eb9f4..c1093ae 100644
@@ -35,7 +35,6 @@
  #include <linux/kref.h>
  #include <linux/list.h>
  #include <linux/wait.h>
-#include <linux/mutex.h>
  #include <linux/mm.h>
  #include <linux/rbtree.h>
  #include <linux/bitmap.h>
@@ -298,7 +297,7 @@ ttm_bo_reference(struct ttm_buffer_object *bo)
   * @interruptible:  Use interruptible wait.
   * @no_wait:  Return immediately if buffer is busy.
   *
- * This function must be called with the bo::mutex held, and makes
+ * This function must be called with bo->reserved held, and makes
   * sure any previous rendering to the buffer is completed.
   * Note: It might be necessary to block validations before the
   * wait by reserving the buffer.
include/drm/ttm/ttm_bo_driver.h +1 -1
index ff7664e..d3fc5f8 100644
@@ -37,6 +37,7 @@
  #include "linux/workqueue.h"
  #include "linux/fs.h"
  #include "linux/spinlock.h"
+#include <linux/mutex.h>
 
  struct ttm_backend;
 
--

1.7.0

Quoted 13 lines
Quoted 12 lines

------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs proactively, and fine-tune applications for parallel performance. See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
--
_______________________________________________
Dri-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/dri-devel

------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs proactively, and fine-tune applications for parallel performance. See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
--