Skip to contentSkip to conversation

panfrost: Locking fixes

PATCH In review Rob Herring 3 participants 14 messages Aug 19, 2019 – Aug 22, 2019

Subsystems (3)
lore
Share & tools
lore

Rob Herring

author maintainer
#0 lore raw

cover panfrost: Locking fixes

With further testing of recent changes with lockdep and other locking checks enabled, we've found several bugs in the shrinker code and one sleep while atomic in panfrost_gem_open(). This series addresses those issues.

Rob

Rob Herring (4):
drm/shmem: Do dma_unmap_sg before purging pages
drm/shmem: Use mutex_trylock in drm_gem_shmem_purge drm/panfrost: Fix shrinker lockdep issues using drm_gem_shmem_purge() drm/panfrost: Fix sleeping while atomic in panfrost_gem_open

 drivers/gpu/drm/drm_gem_shmem_helper.c           | 13 +++++++++++--
 drivers/gpu/drm/panfrost/panfrost_gem.c          | 10 ++++++----
 drivers/gpu/drm/panfrost/panfrost_gem_shrinker.c | 15 ++-------------
 include/drm/drm_gem_shmem_helper.h               |  2 +-
 4 files changed, 20 insertions(+), 20 deletions(-)
Signature · 6 lines
2.20.1

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
author maintainer
#1 lore raw

PATCH 1/4 drm/shmem: Do dma_unmap_sg before purging pages

Calling dma_unmap_sg() in drm_gem_shmem_free_object() is too late if the backing pages have already been released by the shrinker. The result is the following abort:

Unable to handle kernel paging request at virtual address ffff8000098ed000

Mem abort info:
ESR = 0x96000147
Exception class = DABT (current EL), IL = 32 bits
SET = 0, FnV = 0
EA = 0, S1PTW = 0
Data abort info:
ISV = 0, ISS = 0x00000147
CM = 1, WnR = 1
swapper pgtable: 4k pages, 48-bit VAs, pgdp=0000000002f51000
[ffff8000098ed000] pgd=00000000401f8003, pud=00000000401f7003, pmd=00000000401b1003, pte=00e80000098ed712

Internal error: Oops: 96000147 [#1] SMP
Modules linked in: panfrost gpu_sched
CPU: 5 PID: 902 Comm: gnome-shell Not tainted 5.3.0-rc1+ #95
Hardware name: 96boards Rock960 (DT)
pstate: 40000005 (nZcv daif -PAN -UAO)
pc : __dma_inv_area+0x40/0x58
lr : arch_sync_dma_for_cpu+0x28/0x30

sp : ffff00001321ba30
x29: ffff00001321ba30 x28: ffff00001321bd08
x27: 0000000000000000 x26: 0000000000000009
x25: 0000ffffc1f86170 x24: 0000000000000000
x23: 0000000000000000 x22: 0000000000000000
x21: 0000000000021000 x20: ffff80003bb2d810
x19: 00000000098ed000 x18: 0000000000000000
x17: 0000000000000000 x16: ffff800023fd9480
x15: 0000000000000000 x14: 0000000000000000
x13: 0000000000000000 x12: 0000000000000000
x11: 00000000fffb9fff x10: 0000000000000000
x9 : 0000000000000000 x8 : ffff800023fd9c18
x7 : 0000000000000000 x6 : 00000000ffffffff
x5 : 0000000000000000 x4 : 0000000000021000
Purging 5693440 bytes
x3 : 000000000000003f x2 : 0000000000000040
x1 : ffff80000990e000 x0 : ffff8000098ed000

Call trace:
 __dma_inv_area+0x40/0x58
 dma_direct_sync_single_for_cpu+0x7c/0x80
 dma_direct_unmap_page+0x80/0x88
 dma_direct_unmap_sg+0x54/0x80
 drm_gem_shmem_free_object+0xfc/0x108
 panfrost_gem_free_object+0x118/0x128 [panfrost]
 drm_gem_object_free+0x18/0x90
 drm_gem_object_put_unlocked+0x58/0x80
 drm_gem_object_handle_put_unlocked+0x64/0xb0
 drm_gem_object_release_handle+0x70/0x98
 drm_gem_handle_delete+0x64/0xb0
 drm_gem_close_ioctl+0x28/0x38
 drm_ioctl_kernel+0xb8/0x110
 drm_ioctl+0x244/0x3f0
 do_vfs_ioctl+0xbc/0x910
 ksys_ioctl+0x78/0xa8
 __arm64_sys_ioctl+0x1c/0x28
 el0_svc_common.constprop.0+0x88/0x150
 el0_svc_handler+0x28/0x78
 el0_svc+0x8/0xc
 Code: 8a230000 54000060 d50b7e20 14000002 (d5087620)
Fixes 17acb9f35ed7 ("drm/shmem: Add madvise state and purge helpers")
Cc Maarten Lankhorst <[email protected]>
Cc Maxime Ripard <[email protected]>
Cc Sean Paul <[email protected]>
Cc David Airlie <[email protected]>
Cc Daniel Vetter <[email protected]>
---
 drivers/gpu/drm/drm_gem_shmem_helper.c | 6 ++++++
 1 file changed, 6 insertions(+)
drivers/gpu/drm/drm_gem_shmem_helper.c +6 -0
index df8f2c8adb2b..5423ec56b535 100644
@@ -390,6 +390,12 @@ void drm_gem_shmem_purge_locked(struct drm_gem_object *obj)
 
 	WARN_ON(!drm_gem_shmem_is_purgeable(shmem));
 
+	dma_unmap_sg(obj->dev->dev, shmem->sgt->sgl,
+		     shmem->sgt->nents, DMA_BIDIRECTIONAL);
+	sg_free_table(shmem->sgt);
+	kfree(shmem->sgt);
+	shmem->sgt = NULL;
+
 	drm_gem_shmem_put_pages_locked(shmem);
 
 	shmem->madv = -1;
Signature · 6 lines
2.20.1

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
maintainer Reviewed-by on v1 1/4
#2 lore raw

Re: [PATCH 1/4] drm/shmem: Do dma_unmap_sg before purging pages

Quoted 62 lines · Rob Herring

Calling dma_unmap_sg() in drm_gem_shmem_free_object() is too late if the backing pages have already been released by the shrinker. The result is the following abort:

Unable to handle kernel paging request at virtual address ffff8000098ed000

Mem abort info:
ESR = 0x96000147
Exception class = DABT (current EL), IL = 32 bits
SET = 0, FnV = 0
EA = 0, S1PTW = 0
Data abort info:
ISV = 0, ISS = 0x00000147
CM = 1, WnR = 1
swapper pgtable: 4k pages, 48-bit VAs, pgdp=0000000002f51000
[ffff8000098ed000] pgd=00000000401f8003, pud=00000000401f7003, pmd=00000000401b1003, pte=00e80000098ed712

Internal error: Oops: 96000147 [#1] SMP
Modules linked in: panfrost gpu_sched
CPU: 5 PID: 902 Comm: gnome-shell Not tainted 5.3.0-rc1+ #95
Hardware name: 96boards Rock960 (DT)
pstate: 40000005 (nZcv daif -PAN -UAO)
pc : __dma_inv_area+0x40/0x58
lr : arch_sync_dma_for_cpu+0x28/0x30

sp : ffff00001321ba30
x29: ffff00001321ba30 x28: ffff00001321bd08
x27: 0000000000000000 x26: 0000000000000009
x25: 0000ffffc1f86170 x24: 0000000000000000
x23: 0000000000000000 x22: 0000000000000000
x21: 0000000000021000 x20: ffff80003bb2d810
x19: 00000000098ed000 x18: 0000000000000000
x17: 0000000000000000 x16: ffff800023fd9480
x15: 0000000000000000 x14: 0000000000000000
x13: 0000000000000000 x12: 0000000000000000
x11: 00000000fffb9fff x10: 0000000000000000
x9 : 0000000000000000 x8 : ffff800023fd9c18
x7 : 0000000000000000 x6 : 00000000ffffffff
x5 : 0000000000000000 x4 : 0000000000021000
Purging 5693440 bytes
x3 : 000000000000003f x2 : 0000000000000040
x1 : ffff80000990e000 x0 : ffff8000098ed000

Call trace:
 __dma_inv_area+0x40/0x58
 dma_direct_sync_single_for_cpu+0x7c/0x80
 dma_direct_unmap_page+0x80/0x88
 dma_direct_unmap_sg+0x54/0x80
 drm_gem_shmem_free_object+0xfc/0x108
 panfrost_gem_free_object+0x118/0x128 [panfrost]
 drm_gem_object_free+0x18/0x90
 drm_gem_object_put_unlocked+0x58/0x80
 drm_gem_object_handle_put_unlocked+0x64/0xb0
 drm_gem_object_release_handle+0x70/0x98
 drm_gem_handle_delete+0x64/0xb0
 drm_gem_close_ioctl+0x28/0x38
 drm_ioctl_kernel+0xb8/0x110
 drm_ioctl+0x244/0x3f0
 do_vfs_ioctl+0xbc/0x910
 ksys_ioctl+0x78/0xa8
 __arm64_sys_ioctl+0x1c/0x28
 el0_svc_common.constprop.0+0x88/0x150
 el0_svc_handler+0x28/0x78
 el0_svc+0x8/0xc
 Code: 8a230000 54000060 d50b7e20 14000002 (d5087620)
Quoted 7 lines
Fixes 17acb9f35ed7 ("drm/shmem: Add madvise state and purge helpers")
Cc Maarten Lankhorst <[email protected]>
Cc Maxime Ripard <[email protected]>
Cc Sean Paul <[email protected]>
Cc David Airlie <[email protected]>
Cc Daniel Vetter <[email protected]>

Looks good to me:

Quoted 12 lines
---
 drivers/gpu/drm/drm_gem_shmem_helper.c | 6 ++++++
 1 file changed, 6 insertions(+)
drivers/gpu/drm/drm_gem_shmem_helper.c +0 -0
index df8f2c8adb2b..5423ec56b535 100644
@@ -390,6 +390,12 @@ void drm_gem_shmem_purge_locked(struct drm_gem_object *obj)
 
 	WARN_ON(!drm_gem_shmem_is_purgeable(shmem));

drivers/gpu/drm/drm_gem_shmem_helper.c void drm_gem_shmem_purge_locked(struct drm_gem_object *obj) · line 401

Quoted 10 lines
drivers/gpu/drm/drm_gem_shmem_helper.c +6 -0
 
+	dma_unmap_sg(obj->dev->dev, shmem->sgt->sgl,
+		     shmem->sgt->nents, DMA_BIDIRECTIONAL);
+	sg_free_table(shmem->sgt);
+	kfree(shmem->sgt);
+	shmem->sgt = NULL;
+
 	drm_gem_shmem_put_pages_locked(shmem);
 
 	shmem->madv = -1;

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

author maintainer
#3 lore raw

PATCH 2/4 drm/shmem: Use mutex_trylock in drm_gem_shmem_purge

In reply to Rob Herring

Lockdep reports a circular locking dependency with pages_lock taken in the shrinker callback. The deadlock can't actually happen with current users at least as a BO will never be purgeable when pages_lock is held. To be safe, let's use mutex_trylock() instead and bail if a BO is locked already.

WARNING: possible circular locking dependency detected
5.3.0-rc1+ #100 Tainted: G L
------------------------------------------------------
kswapd0/171 is trying to acquire lock:
000000009b9823fd (&shmem->pages_lock){+.+.}, at: drm_gem_shmem_purge+0x20/0x40

but task is already holding lock:
00000000f82369b6 (fs_reclaim){+.+.}, at: __fs_reclaim_acquire+0x0/0x40

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #1 (fs_reclaim){+.+.}:

       fs_reclaim_acquire.part.18+0x34/0x40
       fs_reclaim_acquire+0x20/0x28
       __kmalloc_node+0x6c/0x4c0
       kvmalloc_node+0x38/0xa8
       drm_gem_get_pages+0x80/0x1d0
       drm_gem_shmem_get_pages+0x58/0xa0
       drm_gem_shmem_get_pages_sgt+0x48/0xd0
       panfrost_mmu_map+0x38/0xf8 [panfrost]
       panfrost_gem_open+0xc0/0xe8 [panfrost]
       drm_gem_handle_create_tail+0xe8/0x198
       drm_gem_handle_create+0x3c/0x50
       panfrost_gem_create_with_handle+0x70/0xa0 [panfrost]
       panfrost_ioctl_create_bo+0x48/0x80 [panfrost]
       drm_ioctl_kernel+0xb8/0x110
       drm_ioctl+0x244/0x3f0
       do_vfs_ioctl+0xbc/0x910
       ksys_ioctl+0x78/0xa8
       __arm64_sys_ioctl+0x1c/0x28
       el0_svc_common.constprop.0+0x90/0x168
       el0_svc_handler+0x28/0x78
       el0_svc+0x8/0xc

-> #0 (&shmem->pages_lock){+.+.}:

       __lock_acquire+0xa2c/0x1d70
       lock_acquire+0xdc/0x228
       __mutex_lock+0x8c/0x800
       mutex_lock_nested+0x1c/0x28
       drm_gem_shmem_purge+0x20/0x40
       panfrost_gem_shrinker_scan+0xc0/0x180 [panfrost]
       do_shrink_slab+0x208/0x500
       shrink_slab+0x10c/0x2c0
       shrink_node+0x28c/0x4d8
       balance_pgdat+0x2c8/0x570
       kswapd+0x22c/0x638
       kthread+0x128/0x130
       ret_from_fork+0x10/0x18

other info that might help us debug this:

 Possible unsafe locking scenario:
       CPU0                    CPU1
       ----                    ----
  lock(fs_reclaim);
                               lock(&shmem->pages_lock);
                               lock(fs_reclaim);

lock(&shmem->pages_lock);

*** DEADLOCK ***

3 locks held by kswapd0/171:
#0: 00000000f82369b6 (fs_reclaim){+.+.}, at: __fs_reclaim_acquire+0x0/0x40
#1: 00000000ceb37808 (shrinker_rwsem){++++}, at: shrink_slab+0xbc/0x2c0
#2: 00000000f31efa81 (&pfdev->shrinker_lock){+.+.}, at: panfrost_gem_shrinker_scan+0x34/0x180 [panfrost]

Fixes 17acb9f35ed7 ("drm/shmem: Add madvise state and purge helpers")
Cc Maarten Lankhorst <[email protected]>
Cc Maxime Ripard <[email protected]>
Cc Sean Paul <[email protected]>
Cc David Airlie <[email protected]>
Cc Daniel Vetter <[email protected]>
---
 drivers/gpu/drm/drm_gem_shmem_helper.c | 7 +++++--
 include/drm/drm_gem_shmem_helper.h     | 2 +-
 2 files changed, 6 insertions(+), 3 deletions(-)
drivers/gpu/drm/drm_gem_shmem_helper.c +5 -2
index 5423ec56b535..f5918707672f 100644
@@ -415,13 +415,16 @@ void drm_gem_shmem_purge_locked(struct drm_gem_object *obj)
 }
 EXPORT_SYMBOL(drm_gem_shmem_purge_locked);
 
-void drm_gem_shmem_purge(struct drm_gem_object *obj)
+bool drm_gem_shmem_purge(struct drm_gem_object *obj)
 {
 	struct drm_gem_shmem_object *shmem = to_drm_gem_shmem_obj(obj);
 
-	mutex_lock(&shmem->pages_lock);
+	if (!mutex_trylock(&shmem->pages_lock))
+		return false;
 	drm_gem_shmem_purge_locked(obj);
 	mutex_unlock(&shmem->pages_lock);
+
+	return true;
 }
 EXPORT_SYMBOL(drm_gem_shmem_purge);
 
include/drm/drm_gem_shmem_helper.h +1 -1
index ce1600fdfc3e..01f514521687 100644
@@ -134,7 +134,7 @@ static inline bool drm_gem_shmem_is_purgeable(struct drm_gem_shmem_object *shmem
 }
 
 void drm_gem_shmem_purge_locked(struct drm_gem_object *obj);
-void drm_gem_shmem_purge(struct drm_gem_object *obj);
+bool drm_gem_shmem_purge(struct drm_gem_object *obj);
 
 struct drm_gem_shmem_object *
 drm_gem_shmem_create_with_handle(struct drm_file *file_priv,
Signature · 6 lines
2.20.1

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#4 lore raw

Re: [PATCH 2/4] drm/shmem: Use mutex_trylock in drm_gem_shmem_purge

Quoted 92 lines · Rob Herring

Lockdep reports a circular locking dependency with pages_lock taken in the shrinker callback. The deadlock can't actually happen with current users at least as a BO will never be purgeable when pages_lock is held. To be safe, let's use mutex_trylock() instead and bail if a BO is locked already.

WARNING: possible circular locking dependency detected
5.3.0-rc1+ #100 Tainted: G L
------------------------------------------------------
kswapd0/171 is trying to acquire lock:
000000009b9823fd (&shmem->pages_lock){+.+.}, at: drm_gem_shmem_purge+0x20/0x40

but task is already holding lock:
00000000f82369b6 (fs_reclaim){+.+.}, at: __fs_reclaim_acquire+0x0/0x40

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #1 (fs_reclaim){+.+.}:

       fs_reclaim_acquire.part.18+0x34/0x40
       fs_reclaim_acquire+0x20/0x28
       __kmalloc_node+0x6c/0x4c0
       kvmalloc_node+0x38/0xa8
       drm_gem_get_pages+0x80/0x1d0
       drm_gem_shmem_get_pages+0x58/0xa0
       drm_gem_shmem_get_pages_sgt+0x48/0xd0
       panfrost_mmu_map+0x38/0xf8 [panfrost]
       panfrost_gem_open+0xc0/0xe8 [panfrost]
       drm_gem_handle_create_tail+0xe8/0x198
       drm_gem_handle_create+0x3c/0x50
       panfrost_gem_create_with_handle+0x70/0xa0 [panfrost]
       panfrost_ioctl_create_bo+0x48/0x80 [panfrost]
       drm_ioctl_kernel+0xb8/0x110
       drm_ioctl+0x244/0x3f0
       do_vfs_ioctl+0xbc/0x910
       ksys_ioctl+0x78/0xa8
       __arm64_sys_ioctl+0x1c/0x28
       el0_svc_common.constprop.0+0x90/0x168
       el0_svc_handler+0x28/0x78
       el0_svc+0x8/0xc

-> #0 (&shmem->pages_lock){+.+.}:

       __lock_acquire+0xa2c/0x1d70
       lock_acquire+0xdc/0x228
       __mutex_lock+0x8c/0x800
       mutex_lock_nested+0x1c/0x28
       drm_gem_shmem_purge+0x20/0x40
       panfrost_gem_shrinker_scan+0xc0/0x180 [panfrost]
       do_shrink_slab+0x208/0x500
       shrink_slab+0x10c/0x2c0
       shrink_node+0x28c/0x4d8
       balance_pgdat+0x2c8/0x570
       kswapd+0x22c/0x638
       kthread+0x128/0x130
       ret_from_fork+0x10/0x18

other info that might help us debug this:

 Possible unsafe locking scenario:
       CPU0                    CPU1
       ----                    ----
  lock(fs_reclaim);
                               lock(&shmem->pages_lock);
                               lock(fs_reclaim);

lock(&shmem->pages_lock);

*** DEADLOCK ***

3 locks held by kswapd0/171:
#0: 00000000f82369b6 (fs_reclaim){+.+.}, at: __fs_reclaim_acquire+0x0/0x40
#1: 00000000ceb37808 (shrinker_rwsem){++++}, at: shrink_slab+0xbc/0x2c0
#2: 00000000f31efa81 (&pfdev->shrinker_lock){+.+.}, at: panfrost_gem_shrinker_scan+0x34/0x180 [panfrost]

Fixes 17acb9f35ed7 ("drm/shmem: Add madvise state and purge helpers")
Cc Maarten Lankhorst <[email protected]>
Cc Maxime Ripard <[email protected]>
Cc Sean Paul <[email protected]>
Cc David Airlie <[email protected]>
Cc Daniel Vetter <[email protected]>
---
 drivers/gpu/drm/drm_gem_shmem_helper.c | 7 +++++--
 include/drm/drm_gem_shmem_helper.h     | 2 +-
 2 files changed, 6 insertions(+), 3 deletions(-)
drivers/gpu/drm/drm_gem_shmem_helper.c +0 -0
index 5423ec56b535..f5918707672f 100644

drivers/gpu/drm/drm_gem_shmem_helper.c void drm_gem_shmem_purge_locked(struct drm_gem_object *obj) · line 422

Quoted 10 lines
drivers/gpu/drm/drm_gem_shmem_helper.c +2 -2
@@ -415,13 +415,16 @@ void drm_gem_shmem_purge_locked(struct drm_gem_object *obj)
 }
 EXPORT_SYMBOL(drm_gem_shmem_purge_locked);
 
-void drm_gem_shmem_purge(struct drm_gem_object *obj)
+bool drm_gem_shmem_purge(struct drm_gem_object *obj)
 {
 	struct drm_gem_shmem_object *shmem = to_drm_gem_shmem_obj(obj);
 
-	mutex_lock(&shmem->pages_lock);
+	if (!mutex_trylock(&shmem->pages_lock))

Did you see my ping about cutting all the locking over to dma_resv? Would align shmem helpers with ttm a lot more, for that bright glorious future taste. Should we capture that in some todo.rst entry?

Cheers, Daniel

Quoted 24 lines
drivers/gpu/drm/drm_gem_shmem_helper.c +3 -0
+		return false;
 	drm_gem_shmem_purge_locked(obj);
 	mutex_unlock(&shmem->pages_lock);
+
+	return true;
 }
 EXPORT_SYMBOL(drm_gem_shmem_purge);
 
include/drm/drm_gem_shmem_helper.h +1 -1
index ce1600fdfc3e..01f514521687 100644
@@ -134,7 +134,7 @@ static inline bool drm_gem_shmem_is_purgeable(struct drm_gem_shmem_object *shmem
 }
 
 void drm_gem_shmem_purge_locked(struct drm_gem_object *obj);
-void drm_gem_shmem_purge(struct drm_gem_object *obj);
+bool drm_gem_shmem_purge(struct drm_gem_object *obj);
 
 struct drm_gem_shmem_object *
 drm_gem_shmem_create_with_handle(struct drm_file *file_priv,
Signature · 1 line
2.20.1
Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
author maintainer
#5 lore raw

Re: [PATCH 2/4] drm/shmem: Use mutex_trylock in drm_gem_shmem_purge

Quoted 102 lines · Daniel Vetter
Quoted 100 lines · Rob Herring

Lockdep reports a circular locking dependency with pages_lock taken in the shrinker callback. The deadlock can't actually happen with current users at least as a BO will never be purgeable when pages_lock is held. To be safe, let's use mutex_trylock() instead and bail if a BO is locked already.

WARNING: possible circular locking dependency detected
5.3.0-rc1+ #100 Tainted: G L
------------------------------------------------------
kswapd0/171 is trying to acquire lock:
000000009b9823fd (&shmem->pages_lock){+.+.}, at: drm_gem_shmem_purge+0x20/0x40

but task is already holding lock:
00000000f82369b6 (fs_reclaim){+.+.}, at: __fs_reclaim_acquire+0x0/0x40

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #1 (fs_reclaim){+.+.}:

       fs_reclaim_acquire.part.18+0x34/0x40
       fs_reclaim_acquire+0x20/0x28
       __kmalloc_node+0x6c/0x4c0
       kvmalloc_node+0x38/0xa8
       drm_gem_get_pages+0x80/0x1d0
       drm_gem_shmem_get_pages+0x58/0xa0
       drm_gem_shmem_get_pages_sgt+0x48/0xd0
       panfrost_mmu_map+0x38/0xf8 [panfrost]
       panfrost_gem_open+0xc0/0xe8 [panfrost]
       drm_gem_handle_create_tail+0xe8/0x198
       drm_gem_handle_create+0x3c/0x50
       panfrost_gem_create_with_handle+0x70/0xa0 [panfrost]
       panfrost_ioctl_create_bo+0x48/0x80 [panfrost]
       drm_ioctl_kernel+0xb8/0x110
       drm_ioctl+0x244/0x3f0
       do_vfs_ioctl+0xbc/0x910
       ksys_ioctl+0x78/0xa8
       __arm64_sys_ioctl+0x1c/0x28
       el0_svc_common.constprop.0+0x90/0x168
       el0_svc_handler+0x28/0x78
       el0_svc+0x8/0xc

-> #0 (&shmem->pages_lock){+.+.}:

       __lock_acquire+0xa2c/0x1d70
       lock_acquire+0xdc/0x228
       __mutex_lock+0x8c/0x800
       mutex_lock_nested+0x1c/0x28
       drm_gem_shmem_purge+0x20/0x40
       panfrost_gem_shrinker_scan+0xc0/0x180 [panfrost]
       do_shrink_slab+0x208/0x500
       shrink_slab+0x10c/0x2c0
       shrink_node+0x28c/0x4d8
       balance_pgdat+0x2c8/0x570
       kswapd+0x22c/0x638
       kthread+0x128/0x130
       ret_from_fork+0x10/0x18

other info that might help us debug this:

 Possible unsafe locking scenario:
       CPU0                    CPU1
       ----                    ----
  lock(fs_reclaim);
                               lock(&shmem->pages_lock);
                               lock(fs_reclaim);

lock(&shmem->pages_lock);

*** DEADLOCK ***

3 locks held by kswapd0/171:
#0: 00000000f82369b6 (fs_reclaim){+.+.}, at: __fs_reclaim_acquire+0x0/0x40
#1: 00000000ceb37808 (shrinker_rwsem){++++}, at: shrink_slab+0xbc/0x2c0
#2: 00000000f31efa81 (&pfdev->shrinker_lock){+.+.}, at: panfrost_gem_shrinker_scan+0x34/0x180 [panfrost]

Fixes 17acb9f35ed7 ("drm/shmem: Add madvise state and purge helpers")
Cc Maarten Lankhorst <[email protected]>
Cc Maxime Ripard <[email protected]>
Cc Sean Paul <[email protected]>
Cc David Airlie <[email protected]>
Cc Daniel Vetter <[email protected]>
---
 drivers/gpu/drm/drm_gem_shmem_helper.c | 7 +++++--
 include/drm/drm_gem_shmem_helper.h     | 2 +-
 2 files changed, 6 insertions(+), 3 deletions(-)
drivers/gpu/drm/drm_gem_shmem_helper.c +1 -1
index 5423ec56b535..f5918707672f 100644
@@ -415,13 +415,16 @@ void drm_gem_shmem_purge_locked(struct drm_gem_object *obj)
 }
 EXPORT_SYMBOL(drm_gem_shmem_purge_locked);
 
-void drm_gem_shmem_purge(struct drm_gem_object *obj)
+bool drm_gem_shmem_purge(struct drm_gem_object *obj)
 {
      struct drm_gem_shmem_object *shmem = to_drm_gem_shmem_obj(obj);
 
Quoted 4 lines

drivers/gpu/drm/drm_gem_shmem_helper.c void drm_gem_shmem_purge_locked(struct drm_gem_object *obj) · line 422

Quoted 2 lines
drivers/gpu/drm/drm_gem_shmem_helper.c +1 -1
-     mutex_lock(&shmem->pages_lock);
+     if (!mutex_trylock(&shmem->pages_lock))

Did you see my ping about cutting all the locking over to dma_resv?

Yes, but you didn't reply to Rob C. about it. I guess I'll have to go figure out how reservation objects work...

Quoted 3 lines

Would
align shmem helpers with ttm a lot more, for that bright glorious future taste. Should we capture that in some todo.rst entry?

Sure.

Rob
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

#6 lore raw

Re: [PATCH 2/4] drm/shmem: Use mutex_trylock in drm_gem_shmem_purge

Quoted 103 lines · Rob Herring
Quoted 102 lines · Daniel Vetter
Quoted 100 lines · Rob Herring

Lockdep reports a circular locking dependency with pages_lock taken in the shrinker callback. The deadlock can't actually happen with current users at least as a BO will never be purgeable when pages_lock is held. To be safe, let's use mutex_trylock() instead and bail if a BO is locked already.

WARNING: possible circular locking dependency detected
5.3.0-rc1+ #100 Tainted: G L
------------------------------------------------------
kswapd0/171 is trying to acquire lock:
000000009b9823fd (&shmem->pages_lock){+.+.}, at: drm_gem_shmem_purge+0x20/0x40

but task is already holding lock:
00000000f82369b6 (fs_reclaim){+.+.}, at: __fs_reclaim_acquire+0x0/0x40

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #1 (fs_reclaim){+.+.}:

       fs_reclaim_acquire.part.18+0x34/0x40
       fs_reclaim_acquire+0x20/0x28
       __kmalloc_node+0x6c/0x4c0
       kvmalloc_node+0x38/0xa8
       drm_gem_get_pages+0x80/0x1d0
       drm_gem_shmem_get_pages+0x58/0xa0
       drm_gem_shmem_get_pages_sgt+0x48/0xd0
       panfrost_mmu_map+0x38/0xf8 [panfrost]
       panfrost_gem_open+0xc0/0xe8 [panfrost]
       drm_gem_handle_create_tail+0xe8/0x198
       drm_gem_handle_create+0x3c/0x50
       panfrost_gem_create_with_handle+0x70/0xa0 [panfrost]
       panfrost_ioctl_create_bo+0x48/0x80 [panfrost]
       drm_ioctl_kernel+0xb8/0x110
       drm_ioctl+0x244/0x3f0
       do_vfs_ioctl+0xbc/0x910
       ksys_ioctl+0x78/0xa8
       __arm64_sys_ioctl+0x1c/0x28
       el0_svc_common.constprop.0+0x90/0x168
       el0_svc_handler+0x28/0x78
       el0_svc+0x8/0xc

-> #0 (&shmem->pages_lock){+.+.}:

       __lock_acquire+0xa2c/0x1d70
       lock_acquire+0xdc/0x228
       __mutex_lock+0x8c/0x800
       mutex_lock_nested+0x1c/0x28
       drm_gem_shmem_purge+0x20/0x40
       panfrost_gem_shrinker_scan+0xc0/0x180 [panfrost]
       do_shrink_slab+0x208/0x500
       shrink_slab+0x10c/0x2c0
       shrink_node+0x28c/0x4d8
       balance_pgdat+0x2c8/0x570
       kswapd+0x22c/0x638
       kthread+0x128/0x130
       ret_from_fork+0x10/0x18

other info that might help us debug this:

 Possible unsafe locking scenario:
       CPU0                    CPU1
       ----                    ----
  lock(fs_reclaim);
                               lock(&shmem->pages_lock);
                               lock(fs_reclaim);

lock(&shmem->pages_lock);

*** DEADLOCK ***

3 locks held by kswapd0/171:
#0: 00000000f82369b6 (fs_reclaim){+.+.}, at: __fs_reclaim_acquire+0x0/0x40
#1: 00000000ceb37808 (shrinker_rwsem){++++}, at: shrink_slab+0xbc/0x2c0
#2: 00000000f31efa81 (&pfdev->shrinker_lock){+.+.}, at: panfrost_gem_shrinker_scan+0x34/0x180 [panfrost]

Fixes 17acb9f35ed7 ("drm/shmem: Add madvise state and purge helpers")
Cc Maarten Lankhorst <[email protected]>
Cc Maxime Ripard <[email protected]>
Cc Sean Paul <[email protected]>
Cc David Airlie <[email protected]>
Cc Daniel Vetter <[email protected]>
---
 drivers/gpu/drm/drm_gem_shmem_helper.c | 7 +++++--
 include/drm/drm_gem_shmem_helper.h     | 2 +-
 2 files changed, 6 insertions(+), 3 deletions(-)
drivers/gpu/drm/drm_gem_shmem_helper.c +1 -1
index 5423ec56b535..f5918707672f 100644
@@ -415,13 +415,16 @@ void drm_gem_shmem_purge_locked(struct drm_gem_object *obj)
 }
 EXPORT_SYMBOL(drm_gem_shmem_purge_locked);
 
-void drm_gem_shmem_purge(struct drm_gem_object *obj)
+bool drm_gem_shmem_purge(struct drm_gem_object *obj)
 {
      struct drm_gem_shmem_object *shmem = to_drm_gem_shmem_obj(obj);
 
Quoted 7 lines
Quoted 4 lines

drivers/gpu/drm/drm_gem_shmem_helper.c void drm_gem_shmem_purge_locked(struct drm_gem_object *obj) · line 422

Quoted 2 lines
drivers/gpu/drm/drm_gem_shmem_helper.c +1 -1
-     mutex_lock(&shmem->pages_lock);
+     if (!mutex_trylock(&shmem->pages_lock))

Did you see my ping about cutting all the locking over to dma_resv?

Yes, but you didn't reply to Rob C. about it. I guess I'll have to go figure out how reservation objects work...

msm was the last driver that still used struct_mutex. It's a long-term dead-end, and I think with all the effort recently to create helpers for rendering drivers (shmem, vram, ttm refactoring) we should make a solid attempt to get aligned. Or did you mean that Rob Clark had some reply/questions that I didn' respond to because it fell through cracks?

Quoted 5 lines
Quoted 3 lines

Would
align shmem helpers with ttm a lot more, for that bright glorious future taste. Should we capture that in some todo.rst entry?

Sure.

Cheers, Daniel

Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
author maintainer
#7 lore raw

Re: [PATCH 2/4] drm/shmem: Use mutex_trylock in drm_gem_shmem_purge

Quoted 10 lines · Daniel Vetter
Quoted 8 lines · Rob Herring
Quoted 7 lines · Daniel Vetter
Quoted 5 lines · Rob Herring

Lockdep reports a circular locking dependency with pages_lock taken in the shrinker callback. The deadlock can't actually happen with current users at least as a BO will never be purgeable when pages_lock is held. To be safe, let's use mutex_trylock() instead and bail if a BO is locked already.

[...]

Quoted 10 lines
Quoted 10 lines
Quoted 9 lines

drivers/gpu/drm/drm_gem_shmem_helper.c void drm_gem_shmem_purge_locked(struct drm_gem_object *obj) · line 422

Quoted 7 lines
drivers/gpu/drm/drm_gem_shmem_helper.c +2 -2
-void drm_gem_shmem_purge(struct drm_gem_object *obj)
+bool drm_gem_shmem_purge(struct drm_gem_object *obj)
 {
      struct drm_gem_shmem_object *shmem = to_drm_gem_shmem_obj(obj);
 
-     mutex_lock(&shmem->pages_lock);
+     if (!mutex_trylock(&shmem->pages_lock))

Did you see my ping about cutting all the locking over to dma_resv?

Quoted 8 lines
Quoted 2 lines

Yes, but you didn't reply to Rob C. about it. I guess I'll have to go figure out how reservation objects work...

msm was the last driver that still used struct_mutex. It's a long-term dead-end, and I think with all the effort recently to create helpers for rendering drivers (shmem, vram, ttm refactoring) we should make a solid attempt to get aligned. Or did you mean that Rob Clark had some reply/questions that I didn' respond to because it fell through cracks?

I'm not using struct_mutex, so I'm confused as to why you keep mentioning it. The list of BOs for the shrinker is protected with a mutex for the list. That list head, list mutex, and the shrinker instance all have to live at the driver level, so they can't be moved into shmem as you suggested. Agreed?

Then there is the pages_lock within the shmem BO. I assume that is what you are suggesting converting to dma_resv? I'm not really sure what that would look like. You're going to have to spell it out for me. In my brief look at it, it seems like added complexity and it's not clear to me what that buys. Also, I think it would mostly be an internal implementation detail of shmem helpers, though there is one spot in panfrost that takes the lock (2 before this series). So it's kind of orthogonal to this series.

Also, I think getting more drivers using shmem is more beneficial than aligning the implementations of the GEM helpers. We should at least be able to convert vgem and vkms I would think. Various KMS drivers too, but there's an issue around kernel mappings (or lack of). There really should be little reason for most KMS drivers to have a custom BO as CMA or shmem helpers should work.

Rob
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

#8 lore raw

Re: [PATCH 2/4] drm/shmem: Use mutex_trylock in drm_gem_shmem_purge

Quoted 33 lines · Rob Herring
Quoted 10 lines · Daniel Vetter
Quoted 8 lines · Rob Herring
Quoted 7 lines · Daniel Vetter
Quoted 5 lines · Rob Herring

Lockdep reports a circular locking dependency with pages_lock taken in the shrinker callback. The deadlock can't actually happen with current users at least as a BO will never be purgeable when pages_lock is held. To be safe, let's use mutex_trylock() instead and bail if a BO is locked already.

[...]

Quoted 10 lines
Quoted 10 lines
Quoted 9 lines

drivers/gpu/drm/drm_gem_shmem_helper.c void drm_gem_shmem_purge_locked(struct drm_gem_object *obj) · line 422

Quoted 7 lines
drivers/gpu/drm/drm_gem_shmem_helper.c +2 -2
-void drm_gem_shmem_purge(struct drm_gem_object *obj)
+bool drm_gem_shmem_purge(struct drm_gem_object *obj)
 {
      struct drm_gem_shmem_object *shmem = to_drm_gem_shmem_obj(obj);
 
-     mutex_lock(&shmem->pages_lock);
+     if (!mutex_trylock(&shmem->pages_lock))

Did you see my ping about cutting all the locking over to dma_resv?

Quoted 8 lines
Quoted 2 lines

Yes, but you didn't reply to Rob C. about it. I guess I'll have to go figure out how reservation objects work...

msm was the last driver that still used struct_mutex. It's a long-term dead-end, and I think with all the effort recently to create helpers for rendering drivers (shmem, vram, ttm refactoring) we should make a solid attempt to get aligned. Or did you mean that Rob Clark had some reply/questions that I didn' respond to because it fell through cracks?

Quoted 5 lines

I'm not using struct_mutex, so I'm confused as to why you keep mentioning it. The list of BOs for the shrinker is protected with a mutex for the list. That list head, list mutex, and the shrinker instance all have to live at the driver level, so they can't be moved into shmem as you suggested. Agreed?

struct_mutex is just the historical baggage.

Wrt shrinker/lru, why not? We've talked about maybe moving that to make it easier to share ...

Quoted 8 lines

Then there is the pages_lock within the shmem BO. I assume that is what you are suggesting converting to dma_resv? I'm not really sure what that would look like. You're going to have to spell it out for me. In my brief look at it, it seems like added complexity and it's not clear to me what that buys. Also, I think it would mostly be an internal implementation detail of shmem helpers, though there is one spot in panfrost that takes the lock (2 before this series). So it's kind of orthogonal to this series.

The issue roughly is that having multiple per-bo locks gets fun, once you add in multiple drivers and dynamic dma-buf sharing. Maybe that's never going to be an issue for drivers using shmem helpers, but who knows. The cross-driver per-bo lock to untangle that maze is dma_resv, and if you then also have your own per-bo locks it can get rather interesting. Best case you end up with two locks nesting, and your own per-bo lock being fully redundant. Worst case you get different nesting depending whether you import or export. So that's roughly the context.

Of course fixing locking is going to be easier the fewer users you have. Once there's lots of code and users of it out there, it's pretty much impossible.

So yeah it would be a 1:1 replacement with all the per-bo locks you have now, and seeing how badly it bites.

Quoted 6 lines

Also, I think getting more drivers using shmem is more beneficial than aligning the implementations of the GEM helpers. We should at least be able to convert vgem and vkms I would think. Various KMS drivers too, but there's an issue around kernel mappings (or lack of). There really should be little reason for most KMS drivers to have a custom BO as CMA or shmem helpers should work.

Yeah agreed on this, I just want to make sure we're not doing this multiple times ...
-Daniel

Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
maintainer Reviewed-by on v1 2/4
#9 lore raw

Re: [PATCH 2/4] drm/shmem: Use mutex_trylock in drm_gem_shmem_purge

In reply to Rob Herring
Quoted 75 lines · Rob Herring

Lockdep reports a circular locking dependency with pages_lock taken in the shrinker callback. The deadlock can't actually happen with current users at least as a BO will never be purgeable when pages_lock is held. To be safe, let's use mutex_trylock() instead and bail if a BO is locked already.

WARNING: possible circular locking dependency detected
5.3.0-rc1+ #100 Tainted: G L
------------------------------------------------------
kswapd0/171 is trying to acquire lock:
000000009b9823fd (&shmem->pages_lock){+.+.}, at: drm_gem_shmem_purge+0x20/0x40

but task is already holding lock:
00000000f82369b6 (fs_reclaim){+.+.}, at: __fs_reclaim_acquire+0x0/0x40

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #1 (fs_reclaim){+.+.}:

       fs_reclaim_acquire.part.18+0x34/0x40
       fs_reclaim_acquire+0x20/0x28
       __kmalloc_node+0x6c/0x4c0
       kvmalloc_node+0x38/0xa8
       drm_gem_get_pages+0x80/0x1d0
       drm_gem_shmem_get_pages+0x58/0xa0
       drm_gem_shmem_get_pages_sgt+0x48/0xd0
       panfrost_mmu_map+0x38/0xf8 [panfrost]
       panfrost_gem_open+0xc0/0xe8 [panfrost]
       drm_gem_handle_create_tail+0xe8/0x198
       drm_gem_handle_create+0x3c/0x50
       panfrost_gem_create_with_handle+0x70/0xa0 [panfrost]
       panfrost_ioctl_create_bo+0x48/0x80 [panfrost]
       drm_ioctl_kernel+0xb8/0x110
       drm_ioctl+0x244/0x3f0
       do_vfs_ioctl+0xbc/0x910
       ksys_ioctl+0x78/0xa8
       __arm64_sys_ioctl+0x1c/0x28
       el0_svc_common.constprop.0+0x90/0x168
       el0_svc_handler+0x28/0x78
       el0_svc+0x8/0xc

-> #0 (&shmem->pages_lock){+.+.}:

       __lock_acquire+0xa2c/0x1d70
       lock_acquire+0xdc/0x228
       __mutex_lock+0x8c/0x800
       mutex_lock_nested+0x1c/0x28
       drm_gem_shmem_purge+0x20/0x40
       panfrost_gem_shrinker_scan+0xc0/0x180 [panfrost]
       do_shrink_slab+0x208/0x500
       shrink_slab+0x10c/0x2c0
       shrink_node+0x28c/0x4d8
       balance_pgdat+0x2c8/0x570
       kswapd+0x22c/0x638
       kthread+0x128/0x130
       ret_from_fork+0x10/0x18

other info that might help us debug this:

 Possible unsafe locking scenario:
       CPU0                    CPU1
       ----                    ----
  lock(fs_reclaim);
                               lock(&shmem->pages_lock);
                               lock(fs_reclaim);

lock(&shmem->pages_lock);

*** DEADLOCK ***

3 locks held by kswapd0/171:
#0: 00000000f82369b6 (fs_reclaim){+.+.}, at: __fs_reclaim_acquire+0x0/0x40
#1: 00000000ceb37808 (shrinker_rwsem){++++}, at: shrink_slab+0xbc/0x2c0
#2: 00000000f31efa81 (&pfdev->shrinker_lock){+.+.}, at: panfrost_gem_shrinker_scan+0x34/0x180 [panfrost]

Quoted 7 lines
Fixes 17acb9f35ed7 ("drm/shmem: Add madvise state and purge helpers")
Cc Maarten Lankhorst <[email protected]>
Cc Maxime Ripard <[email protected]>
Cc Sean Paul <[email protected]>
Cc David Airlie <[email protected]>
Cc Daniel Vetter <[email protected]>

Seems reasonable, like you state I don't think this can actually happen, but keeping lockdep happy is a good idea.

Steve

Quoted 34 lines
---
 drivers/gpu/drm/drm_gem_shmem_helper.c | 7 +++++--
 include/drm/drm_gem_shmem_helper.h     | 2 +-
 2 files changed, 6 insertions(+), 3 deletions(-)
drivers/gpu/drm/drm_gem_shmem_helper.c +5 -2
index 5423ec56b535..f5918707672f 100644
@@ -415,13 +415,16 @@ void drm_gem_shmem_purge_locked(struct drm_gem_object *obj)
 }
 EXPORT_SYMBOL(drm_gem_shmem_purge_locked);
 
-void drm_gem_shmem_purge(struct drm_gem_object *obj)
+bool drm_gem_shmem_purge(struct drm_gem_object *obj)
 {
 	struct drm_gem_shmem_object *shmem = to_drm_gem_shmem_obj(obj);
 
-	mutex_lock(&shmem->pages_lock);
+	if (!mutex_trylock(&shmem->pages_lock))
+		return false;
 	drm_gem_shmem_purge_locked(obj);
 	mutex_unlock(&shmem->pages_lock);
+
+	return true;
 }
 EXPORT_SYMBOL(drm_gem_shmem_purge);
 
include/drm/drm_gem_shmem_helper.h +0 -0
index ce1600fdfc3e..01f514521687 100644

include/drm/drm_gem_shmem_helper.h static inline bool drm_gem_shmem_is_purgeable(struct drm_gem_shmem_object *shmem · line 140

Quoted 8 lines
include/drm/drm_gem_shmem_helper.h +1 -1
@@ -134,7 +134,7 @@ static inline bool drm_gem_shmem_is_purgeable(struct drm_gem_shmem_object *shmem
 }
 
 void drm_gem_shmem_purge_locked(struct drm_gem_object *obj);
-void drm_gem_shmem_purge(struct drm_gem_object *obj);
+bool drm_gem_shmem_purge(struct drm_gem_object *obj);
 
 struct drm_gem_shmem_object *
 drm_gem_shmem_create_with_handle(struct drm_file *file_priv,

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

author maintainer
#10 lore raw

PATCH 3/4 drm/panfrost: Fix shrinker lockdep issues using drm_gem_shmem_purge()

In reply to Rob Herring

This fixes 2 issues found by lockdep. First, drm_gem_shmem_purge() now uses mutex_trylock for the pages_lock to avoid a circular dependency.

Second, it drops the call to panfrost_mmu_unmap() which takes several locks due to runtime PM calls. The call is not necessary because the unmapping is also called in panfrost_gem_close() already.

Fixes 013b65101315 ("drm/panfrost: Add madvise and shrinker support")
Cc Tomeu Vizoso <[email protected]>
Cc David Airlie <[email protected]>
Cc Daniel Vetter <[email protected]>
---
 drivers/gpu/drm/panfrost/panfrost_gem_shrinker.c | 15 ++-------------
 1 file changed, 2 insertions(+), 13 deletions(-)
drivers/gpu/drm/panfrost/panfrost_gem_shrinker.c +2 -13
index d191632b6197..cc15005dc68f 100644
@@ -36,17 +36,6 @@ panfrost_gem_shrinker_count(struct shrinker *shrinker, struct shrink_control *sc
 	return count;
 }
 
-static void panfrost_gem_purge(struct drm_gem_object *obj)
-{
-	struct drm_gem_shmem_object *shmem = to_drm_gem_shmem_obj(obj);
-	mutex_lock(&shmem->pages_lock);
-
-	panfrost_mmu_unmap(to_panfrost_bo(obj));
-	drm_gem_shmem_purge_locked(obj);
-
-	mutex_unlock(&shmem->pages_lock);
-}
-
 static unsigned long
 panfrost_gem_shrinker_scan(struct shrinker *shrinker, struct shrink_control *sc)
 {
@@ -61,8 +50,8 @@ panfrost_gem_shrinker_scan(struct shrinker *shrinker, struct shrink_control *sc)
 	list_for_each_entry_safe(shmem, tmp, &pfdev->shrinker_list, madv_list) {
 		if (freed >= sc->nr_to_scan)
 			break;
-		if (drm_gem_shmem_is_purgeable(shmem)) {
-			panfrost_gem_purge(&shmem->base);
+		if (drm_gem_shmem_is_purgeable(shmem) &&
+		    drm_gem_shmem_purge(&shmem->base)) {
 			freed += shmem->base.size >> PAGE_SHIFT;
 			list_del_init(&shmem->madv_list);
 		}
Signature · 6 lines
2.20.1

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
maintainer
#11 lore raw

Re: [PATCH 3/4] drm/panfrost: Fix shrinker lockdep issues using drm_gem_shmem_purge()

Quoted 3 lines · Rob Herring

This fixes 2 issues found by lockdep. First, drm_gem_shmem_purge() now uses mutex_trylock for the pages_lock to avoid a circular dependency.

NIT: This is in the previous patch.

Quoted 3 lines

Second, it drops the call to panfrost_mmu_unmap() which takes several locks due to runtime PM calls. The call is not necessary because the unmapping is also called in panfrost_gem_close() already.

I could be completely mistaken here, but don't we need to unmap the memory from the GPU here because the backing is free? The panfrost_gem_close() call could come significantly later, by which time a malicious user space could have run some jobs on the GPU to take a look at what those mappings now point to (quite likely some other processes memory).

So this looks to me like a crafty way of observing 'random' memory in the system.

Steve

Quoted 33 lines
Fixes 013b65101315 ("drm/panfrost: Add madvise and shrinker support")
Cc Tomeu Vizoso <[email protected]>
Cc David Airlie <[email protected]>
Cc Daniel Vetter <[email protected]>
---
 drivers/gpu/drm/panfrost/panfrost_gem_shrinker.c | 15 ++-------------
 1 file changed, 2 insertions(+), 13 deletions(-)
drivers/gpu/drm/panfrost/panfrost_gem_shrinker.c +0 -11
index d191632b6197..cc15005dc68f 100644
@@ -36,17 +36,6 @@ panfrost_gem_shrinker_count(struct shrinker *shrinker, struct shrink_control *sc
 	return count;
 }
 
-static void panfrost_gem_purge(struct drm_gem_object *obj)
-{
-	struct drm_gem_shmem_object *shmem = to_drm_gem_shmem_obj(obj);
-	mutex_lock(&shmem->pages_lock);
-
-	panfrost_mmu_unmap(to_panfrost_bo(obj));
-	drm_gem_shmem_purge_locked(obj);
-
-	mutex_unlock(&shmem->pages_lock);
-}
-
 static unsigned long
 panfrost_gem_shrinker_scan(struct shrinker *shrinker, struct shrink_control *sc)
 {

drivers/gpu/drm/panfrost/panfrost_gem_shrinker.c panfrost_gem_shrinker_scan(struct shrinker *shrinker, struct shrink_control *sc) · line 57

Quoted 10 lines
drivers/gpu/drm/panfrost/panfrost_gem_shrinker.c +2 -2
@@ -61,8 +50,8 @@ panfrost_gem_shrinker_scan(struct shrinker *shrinker, struct shrink_control *sc)
 	list_for_each_entry_safe(shmem, tmp, &pfdev->shrinker_list, madv_list) {
 		if (freed >= sc->nr_to_scan)
 			break;
-		if (drm_gem_shmem_is_purgeable(shmem)) {
-			panfrost_gem_purge(&shmem->base);
+		if (drm_gem_shmem_is_purgeable(shmem) &&
+		    drm_gem_shmem_purge(&shmem->base)) {
 			freed += shmem->base.size >> PAGE_SHIFT;
 			list_del_init(&shmem->madv_list);
 		}

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

author maintainer
#12 lore raw

PATCH 4/4 drm/panfrost: Fix sleeping while atomic in panfrost_gem_open

In reply to Rob Herring

We can't hold the mm_lock spinlock as panfrost_mmu_map() can sleep:

BUG: sleeping function called from invalid context at kernel/locking/mutex.c:909
in_atomic(): 1, irqs_disabled(): 0, pid: 974, name: glmark2-es2-drm
1 lock held by glmark2-es2-drm/974:
CPU: 5 PID: 974 Comm: glmark2-es2-drm Tainted: G        W    L    5.3.0-rc1+ #94
Hardware name: 96boards Rock960 (DT)
Call trace:
 dump_backtrace+0x0/0x130
 show_stack+0x14/0x20
 dump_stack+0xc4/0x10c
 ___might_sleep+0x158/0x228
 __might_sleep+0x50/0x88
 __mutex_lock+0x58/0x800
 mutex_lock_interruptible_nested+0x1c/0x28
 drm_gem_shmem_get_pages+0x24/0xa0
 drm_gem_shmem_get_pages_sgt+0x48/0xd0
 panfrost_mmu_map+0x38/0xf8 [panfrost]
 panfrost_gem_open+0xc0/0xd8 [panfrost]
 drm_gem_handle_create_tail+0xe8/0x198
 drm_gem_handle_create+0x3c/0x50
 panfrost_gem_create_with_handle+0x70/0xa0 [panfrost]
 panfrost_ioctl_create_bo+0x48/0x80 [panfrost]
 drm_ioctl_kernel+0xb8/0x110
 drm_ioctl+0x244/0x3f0
 do_vfs_ioctl+0xbc/0x910
 ksys_ioctl+0x78/0xa8
 __arm64_sys_ioctl+0x1c/0x28
 el0_svc_common.constprop.0+0x90/0x168
 el0_svc_handler+0x28/0x78
 el0_svc+0x8/0xc
Fixes 68337d0b8644 ("drm/panfrost: Restructure the GEM object creation")
Cc Tomeu Vizoso <[email protected]>
Cc David Airlie <[email protected]>
Cc Daniel Vetter <[email protected]>
---
 drivers/gpu/drm/panfrost/panfrost_gem.c | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)
drivers/gpu/drm/panfrost/panfrost_gem.c +6 -4
index e084bc4e9083..acb07fe06580 100644
@@ -65,16 +65,18 @@ static int panfrost_gem_open(struct drm_gem_object *obj, struct drm_file *file_p
 	spin_lock(&priv->mm_lock);
 	ret = drm_mm_insert_node_generic(&priv->mm, &bo->node,
 					 size >> PAGE_SHIFT, align, color, 0);
+	spin_unlock(&priv->mm_lock);
 	if (ret)
-		goto out;
+		return ret;
 
 	if (!bo->is_heap) {
 		ret = panfrost_mmu_map(bo);
-		if (ret)
+		if (ret) {
+			spin_lock(&priv->mm_lock);
 			drm_mm_remove_node(&bo->node);
+			spin_unlock(&priv->mm_lock);
+		}
 	}
-out:
-	spin_unlock(&priv->mm_lock);
 	return ret;
 }
 
Signature · 6 lines
2.20.1

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
maintainer Reviewed-by on v1 4/4
#13 lore raw

Re: [PATCH 4/4] drm/panfrost: Fix sleeping while atomic in panfrost_gem_open

Quoted 32 lines · Rob Herring

We can't hold the mm_lock spinlock as panfrost_mmu_map() can sleep:

BUG: sleeping function called from invalid context at kernel/locking/mutex.c:909
in_atomic(): 1, irqs_disabled(): 0, pid: 974, name: glmark2-es2-drm
1 lock held by glmark2-es2-drm/974:
CPU: 5 PID: 974 Comm: glmark2-es2-drm Tainted: G        W    L    5.3.0-rc1+ #94
Hardware name: 96boards Rock960 (DT)
Call trace:
 dump_backtrace+0x0/0x130
 show_stack+0x14/0x20
 dump_stack+0xc4/0x10c
 ___might_sleep+0x158/0x228
 __might_sleep+0x50/0x88
 __mutex_lock+0x58/0x800
 mutex_lock_interruptible_nested+0x1c/0x28
 drm_gem_shmem_get_pages+0x24/0xa0
 drm_gem_shmem_get_pages_sgt+0x48/0xd0
 panfrost_mmu_map+0x38/0xf8 [panfrost]
 panfrost_gem_open+0xc0/0xd8 [panfrost]
 drm_gem_handle_create_tail+0xe8/0x198
 drm_gem_handle_create+0x3c/0x50
 panfrost_gem_create_with_handle+0x70/0xa0 [panfrost]
 panfrost_ioctl_create_bo+0x48/0x80 [panfrost]
 drm_ioctl_kernel+0xb8/0x110
 drm_ioctl+0x244/0x3f0
 do_vfs_ioctl+0xbc/0x910
 ksys_ioctl+0x78/0xa8
 __arm64_sys_ioctl+0x1c/0x28
 el0_svc_common.constprop.0+0x90/0x168
 el0_svc_handler+0x28/0x78
 el0_svc+0x8/0xc
Quoted 5 lines
Fixes 68337d0b8644 ("drm/panfrost: Restructure the GEM object creation")
Cc Tomeu Vizoso <[email protected]>
Cc David Airlie <[email protected]>
Cc Daniel Vetter <[email protected]>
Quoted 22 lines
---
 drivers/gpu/drm/panfrost/panfrost_gem.c | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)
drivers/gpu/drm/panfrost/panfrost_gem.c +3 -2
index e084bc4e9083..acb07fe06580 100644
@@ -65,16 +65,18 @@ static int panfrost_gem_open(struct drm_gem_object *obj, struct drm_file *file_p
 	spin_lock(&priv->mm_lock);
 	ret = drm_mm_insert_node_generic(&priv->mm, &bo->node,
 					 size >> PAGE_SHIFT, align, color, 0);
+	spin_unlock(&priv->mm_lock);
 	if (ret)
-		goto out;
+		return ret;
 
 	if (!bo->is_heap) {
 		ret = panfrost_mmu_map(bo);
-		if (ret)
+		if (ret) {

drivers/gpu/drm/panfrost/panfrost_gem.c static int panfrost_gem_open(struct drm_gem_object *obj, struct drm_file *file_p · line 82

Quoted 10 lines
drivers/gpu/drm/panfrost/panfrost_gem.c +3 -2
+			spin_lock(&priv->mm_lock);
 			drm_mm_remove_node(&bo->node);
+			spin_unlock(&priv->mm_lock);
+		}
 	}
-out:
-	spin_unlock(&priv->mm_lock);
 	return ret;
 }
 

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel