linux/include/media/videobuf2-core.h

1335 lines
52 KiB

media: vb2: keep a reference to the request until dqbuf When vb2_buffer_done is called the buffer is unbound from the request and put. The media_request_object_put also 'put's the request reference. If the application has already closed the request fd, then that means that the request reference at that point goes to 0 and the whole request is released. This means that the control handler associated with the request is also freed and that causes this kernel oops: [174705.995401] BUG: sleeping function called from invalid context at kernel/locking/mutex.c:908 [174705.995411] in_atomic(): 1, irqs_disabled(): 1, pid: 28071, name: vivid-000-vid-o [174705.995416] 2 locks held by vivid-000-vid-o/28071: [174705.995420] #0: 000000001ea3a232 (&dev->mutex#3){....}, at: vivid_thread_vid_out+0x3f5/0x550 [vivid] [174705.995447] #1: 00000000e30a0d1e (&(&q->done_lock)->rlock){....}, at: vb2_buffer_done+0x92/0x1d0 [videobuf2_common] [174705.995460] Preemption disabled at: [174705.995461] [<0000000000000000>] (null) [174705.995472] CPU: 11 PID: 28071 Comm: vivid-000-vid-o Tainted: G W 4.20.0-rc1-test-no #88 [174705.995476] Hardware name: VMware, Inc. VMware Virtual Platform/440BX Desktop Reference Platform, BIOS 6.00 05/19/2017 [174705.995481] Call Trace: [174705.995500] dump_stack+0x46/0x60 [174705.995512] ___might_sleep.cold.79+0xe1/0xf1 [174705.995523] __mutex_lock+0x50/0x8f0 [174705.995531] ? find_held_lock+0x2d/0x90 [174705.995536] ? find_held_lock+0x2d/0x90 [174705.995542] ? find_held_lock+0x2d/0x90 [174705.995564] ? v4l2_ctrl_handler_free.part.13+0x44/0x1d0 [videodev] [174705.995576] v4l2_ctrl_handler_free.part.13+0x44/0x1d0 [videodev] [174705.995590] v4l2_ctrl_request_release+0x1c/0x30 [videodev] [174705.995600] media_request_clean+0x64/0xe0 [media] [174705.995609] media_request_release+0x19/0x40 [media] [174705.995617] vb2_buffer_done+0xef/0x1d0 [videobuf2_common] [174705.995630] vivid_thread_vid_out+0x2c1/0x550 [vivid] [174705.995645] ? vivid_stop_generating_vid_cap+0x1c0/0x1c0 [vivid] [174705.995653] kthread+0x113/0x130 [174705.995659] ? kthread_park+0x80/0x80 [174705.995667] ret_from_fork+0x35/0x40 The vb2_buffer_done function can be called from interrupt context, so anything that sleeps is not allowed. The solution is to increment the request refcount when the buffer is queued and decrement it when the buffer is dequeued. Releasing the request is fine if that happens from VIDIOC_DQBUF. Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl> Acked-by: Sakari Ailus <sakari.ailus@linux.intel.com> Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl> Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
2018-11-28 03:37:45 -05:00
media: vb2: keep a reference to the request until dqbuf When vb2_buffer_done is called the buffer is unbound from the request and put. The media_request_object_put also 'put's the request reference. If the application has already closed the request fd, then that means that the request reference at that point goes to 0 and the whole request is released. This means that the control handler associated with the request is also freed and that causes this kernel oops: [174705.995401] BUG: sleeping function called from invalid context at kernel/locking/mutex.c:908 [174705.995411] in_atomic(): 1, irqs_disabled(): 1, pid: 28071, name: vivid-000-vid-o [174705.995416] 2 locks held by vivid-000-vid-o/28071: [174705.995420] #0: 000000001ea3a232 (&dev->mutex#3){....}, at: vivid_thread_vid_out+0x3f5/0x550 [vivid] [174705.995447] #1: 00000000e30a0d1e (&(&q->done_lock)->rlock){....}, at: vb2_buffer_done+0x92/0x1d0 [videobuf2_common] [174705.995460] Preemption disabled at: [174705.995461] [<0000000000000000>] (null) [174705.995472] CPU: 11 PID: 28071 Comm: vivid-000-vid-o Tainted: G W 4.20.0-rc1-test-no #88 [174705.995476] Hardware name: VMware, Inc. VMware Virtual Platform/440BX Desktop Reference Platform, BIOS 6.00 05/19/2017 [174705.995481] Call Trace: [174705.995500] dump_stack+0x46/0x60 [174705.995512] ___might_sleep.cold.79+0xe1/0xf1 [174705.995523] __mutex_lock+0x50/0x8f0 [174705.995531] ? find_held_lock+0x2d/0x90 [174705.995536] ? find_held_lock+0x2d/0x90 [174705.995542] ? find_held_lock+0x2d/0x90 [174705.995564] ? v4l2_ctrl_handler_free.part.13+0x44/0x1d0 [videodev] [174705.995576] v4l2_ctrl_handler_free.part.13+0x44/0x1d0 [videodev] [174705.995590] v4l2_ctrl_request_release+0x1c/0x30 [videodev] [174705.995600] media_request_clean+0x64/0xe0 [media] [174705.995609] media_request_release+0x19/0x40 [media] [174705.995617] vb2_buffer_done+0xef/0x1d0 [videobuf2_common] [174705.995630] vivid_thread_vid_out+0x2c1/0x550 [vivid] [174705.995645] ? vivid_stop_generating_vid_cap+0x1c0/0x1c0 [vivid] [174705.995653] kthread+0x113/0x130 [174705.995659] ? kthread_park+0x80/0x80 [174705.995667] ret_from_fork+0x35/0x40 The vb2_buffer_done function can be called from interrupt context, so anything that sleeps is not allowed. The solution is to increment the request refcount when the buffer is queued and decrement it when the buffer is dequeued. Releasing the request is fine if that happens from VIDIOC_DQBUF. Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl> Acked-by: Sakari Ailus <sakari.ailus@linux.intel.com> Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl> Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@kernel.org>
2018-11-28 03:37:45 -05:00
dma-mapping: use unsigned long for dma_attrs The dma-mapping core and the implementations do not change the DMA attributes passed by pointer. Thus the pointer can point to const data. However the attributes do not have to be a bitfield. Instead unsigned long will do fine: 1. This is just simpler. Both in terms of reading the code and setting attributes. Instead of initializing local attributes on the stack and passing pointer to it to dma_set_attr(), just set the bits. 2. It brings safeness and checking for const correctness because the attributes are passed by value. Semantic patches for this change (at least most of them): virtual patch virtual context @r@ identifier f, attrs; @@ f(..., - struct dma_attrs *attrs + unsigned long attrs , ...) { ... } @@ identifier r.f; @@ f(..., - NULL + 0 ) and // Options: --all-includes virtual patch virtual context @r@ identifier f, attrs; type t; @@ t f(..., struct dma_attrs *attrs); @@ identifier r.f; @@ f(..., - NULL + 0 ) Link: http://lkml.kernel.org/r/1468399300-5399-2-git-send-email-k.kozlowski@samsung.com Signed-off-by: Krzysztof Kozlowski <k.kozlowski@samsung.com> Acked-by: Vineet Gupta <vgupta@synopsys.com> Acked-by: Robin Murphy <robin.murphy@arm.com> Acked-by: Hans-Christian Noren Egtvedt <egtvedt@samfundet.no> Acked-by: Mark Salter <msalter@redhat.com> [c6x] Acked-by: Jesper Nilsson <jesper.nilsson@axis.com> [cris] Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch> [drm] Reviewed-by: Bart Van Assche <bart.vanassche@sandisk.com> Acked-by: Joerg Roedel <jroedel@suse.de> [iommu] Acked-by: Fabien Dessenne <fabien.dessenne@st.com> [bdisp] Reviewed-by: Marek Szyprowski <m.szyprowski@samsung.com> [vb2-core] Acked-by: David Vrabel <david.vrabel@citrix.com> [xen] Acked-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> [xen swiotlb] Acked-by: Joerg Roedel <jroedel@suse.de> [iommu] Acked-by: Richard Kuo <rkuo@codeaurora.org> [hexagon] Acked-by: Geert Uytterhoeven <geert@linux-m68k.org> [m68k] Acked-by: Gerald Schaefer <gerald.schaefer@de.ibm.com> [s390] Acked-by: Bjorn Andersson <bjorn.andersson@linaro.org> Acked-by: Hans-Christian Noren Egtvedt <egtvedt@samfundet.no> [avr32] Acked-by: Vineet Gupta <vgupta@synopsys.com> [arc] Acked-by: Robin Murphy <robin.murphy@arm.com> [arm64 and dma-iommu] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-08-03 13:46:00 -07:00
[media] vb2: only call start_streaming if sufficient buffers are queued In commit 02f142ecd24aaf891324ffba8527284c1731b561 support was added to start_streaming to return -ENOBUFS if insufficient buffers were queued for the DMA engine to start. The vb2 core would attempt calling start_streaming again if another buffer would be queued up. Later analysis uncovered problems with the queue management if start_streaming would return an error: the buffers are enqueued to the driver before the start_streaming op is called, so after an error they are never returned to the vb2 core. The solution for this is to let the driver return them to the vb2 core in case of an error while starting the DMA engine. However, in the case of -ENOBUFS that would be weird: it is not a real error, it just says that more buffers are needed. Requiring start_streaming to give them back only to have them requeued again the next time the application calls QBUF is inefficient. This patch changes this mechanism: it adds a 'min_buffers_needed' field to vb2_queue that drivers can set with the minimum number of buffers required to start the DMA engine. The start_streaming op is only called if enough buffers are queued. The -ENOBUFS handling has been dropped in favor of this new method. Drivers are expected to return buffers back to vb2 core with state QUEUED if start_streaming would return an error. The vb2 core checks for this and produces a warning if that didn't happen and it will forcefully reclaim such buffers to ensure that the internal vb2 core state remains consistent and all buffer-related resources have been correctly freed and all op calls have been balanced. __reqbufs() has been updated to check that at least min_buffers_needed buffers could be allocated. If fewer buffers were allocated then __reqbufs will free what was allocated and return -ENOMEM. Based on a suggestion from Pawel Osciak. __create_bufs() doesn't do that check, since the use of __create_bufs assumes some advance scenario where the user might want more control. Instead streamon will check if enough buffers were allocated to prevent streaming with fewer than the minimum required number of buffers. Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com> Signed-off-by: Mauro Carvalho Chehab <m.chehab@samsung.com>
2014-02-24 13:51:03 -03:00
[media] videobuf2: fix lockdep warning The following lockdep warning has been there ever since commit a517cca6b24fc54ac209e44118ec8962051662e3 one year ago: [ 403.117947] ====================================================== [ 403.117949] [ INFO: possible circular locking dependency detected ] [ 403.117953] 3.16.0-rc6-test-media #961 Not tainted [ 403.117954] ------------------------------------------------------- [ 403.117956] v4l2-ctl/15377 is trying to acquire lock: [ 403.117959] (&dev->mutex#3){+.+.+.}, at: [<ffffffffa005a6c3>] vb2_fop_mmap+0x33/0x90 [videobuf2_core] [ 403.117974] [ 403.117974] but task is already holding lock: [ 403.117976] (&mm->mmap_sem){++++++}, at: [<ffffffff8118291f>] vm_mmap_pgoff+0x6f/0xc0 [ 403.117987] [ 403.117987] which lock already depends on the new lock. [ 403.117987] [ 403.117990] [ 403.117990] the existing dependency chain (in reverse order) is: [ 403.117992] [ 403.117992] -> #1 (&mm->mmap_sem){++++++}: [ 403.117997] [<ffffffff810d733c>] validate_chain.isra.39+0x5fc/0x9a0 [ 403.118006] [<ffffffff810d8bc3>] __lock_acquire+0x4d3/0xd30 [ 403.118010] [<ffffffff810d9da7>] lock_acquire+0xa7/0x160 [ 403.118014] [<ffffffff8118c9ec>] might_fault+0x7c/0xb0 [ 403.118018] [<ffffffffa0028a25>] video_usercopy+0x425/0x610 [videodev] [ 403.118028] [<ffffffffa0028c25>] video_ioctl2+0x15/0x20 [videodev] [ 403.118034] [<ffffffffa0022764>] v4l2_ioctl+0x184/0x1a0 [videodev] [ 403.118040] [<ffffffff811d77d0>] do_vfs_ioctl+0x2f0/0x4f0 [ 403.118307] [<ffffffff811d7a51>] SyS_ioctl+0x81/0xa0 [ 403.118311] [<ffffffff8199dc69>] system_call_fastpath+0x16/0x1b [ 403.118319] [ 403.118319] -> #0 (&dev->mutex#3){+.+.+.}: [ 403.118324] [<ffffffff810d6a96>] check_prevs_add+0x746/0x9f0 [ 403.118329] [<ffffffff810d733c>] validate_chain.isra.39+0x5fc/0x9a0 [ 403.118333] [<ffffffff810d8bc3>] __lock_acquire+0x4d3/0xd30 [ 403.118336] [<ffffffff810d9da7>] lock_acquire+0xa7/0x160 [ 403.118340] [<ffffffff81999664>] mutex_lock_interruptible_nested+0x64/0x640 [ 403.118344] [<ffffffffa005a6c3>] vb2_fop_mmap+0x33/0x90 [videobuf2_core] [ 403.118349] [<ffffffffa0022122>] v4l2_mmap+0x62/0xa0 [videodev] [ 403.118354] [<ffffffff81197270>] mmap_region+0x3d0/0x5d0 [ 403.118359] [<ffffffff8119778d>] do_mmap_pgoff+0x31d/0x400 [ 403.118363] [<ffffffff81182940>] vm_mmap_pgoff+0x90/0xc0 [ 403.118366] [<ffffffff81195cef>] SyS_mmap_pgoff+0x1df/0x2a0 [ 403.118369] [<ffffffff810085c2>] SyS_mmap+0x22/0x30 [ 403.118376] [<ffffffff8199dc69>] system_call_fastpath+0x16/0x1b [ 403.118381] [ 403.118381] other info that might help us debug this: [ 403.118381] [ 403.118383] Possible unsafe locking scenario: [ 403.118383] [ 403.118385] CPU0 CPU1 [ 403.118387] ---- ---- [ 403.118388] lock(&mm->mmap_sem); [ 403.118391] lock(&dev->mutex#3); [ 403.118394] lock(&mm->mmap_sem); [ 403.118397] lock(&dev->mutex#3); [ 403.118400] [ 403.118400] *** DEADLOCK *** [ 403.118400] [ 403.118403] 1 lock held by v4l2-ctl/15377: [ 403.118405] #0: (&mm->mmap_sem){++++++}, at: [<ffffffff8118291f>] vm_mmap_pgoff+0x6f/0xc0 [ 403.118411] [ 403.118411] stack backtrace: [ 403.118415] CPU: 0 PID: 15377 Comm: v4l2-ctl Not tainted 3.16.0-rc6-test-media #961 [ 403.118418] Hardware name: VMware, Inc. VMware Virtual Platform/440BX Desktop Reference Platform, BIOS 6.00 07/31/2013 [ 403.118420] ffffffff82a6c9d0 ffff8800af37fb00 ffffffff819916a2 ffffffff82a6c9d0 [ 403.118425] ffff8800af37fb40 ffffffff810d5715 ffff8802308e4200 0000000000000000 [ 403.118429] ffff8802308e4a48 ffff8802308e4a48 ffff8802308e4200 0000000000000001 [ 403.118433] Call Trace: [ 403.118441] [<ffffffff819916a2>] dump_stack+0x4e/0x7a [ 403.118445] [<ffffffff810d5715>] print_circular_bug+0x1d5/0x2a0 [ 403.118449] [<ffffffff810d6a96>] check_prevs_add+0x746/0x9f0 [ 403.118455] [<ffffffff8119c172>] ? find_vmap_area+0x42/0x70 [ 403.118459] [<ffffffff810d733c>] validate_chain.isra.39+0x5fc/0x9a0 [ 403.118463] [<ffffffff810d8bc3>] __lock_acquire+0x4d3/0xd30 [ 403.118468] [<ffffffff810d9da7>] lock_acquire+0xa7/0x160 [ 403.118472] [<ffffffffa005a6c3>] ? vb2_fop_mmap+0x33/0x90 [videobuf2_core] [ 403.118476] [<ffffffffa005a6c3>] ? vb2_fop_mmap+0x33/0x90 [videobuf2_core] [ 403.118480] [<ffffffff81999664>] mutex_lock_interruptible_nested+0x64/0x640 [ 403.118484] [<ffffffffa005a6c3>] ? vb2_fop_mmap+0x33/0x90 [videobuf2_core] [ 403.118488] [<ffffffffa005a6c3>] ? vb2_fop_mmap+0x33/0x90 [videobuf2_core] [ 403.118493] [<ffffffff810d8055>] ? mark_held_locks+0x75/0xa0 [ 403.118497] [<ffffffffa005a6c3>] vb2_fop_mmap+0x33/0x90 [videobuf2_core] [ 403.118502] [<ffffffffa0022122>] v4l2_mmap+0x62/0xa0 [videodev] [ 403.118506] [<ffffffff81197270>] mmap_region+0x3d0/0x5d0 [ 403.118510] [<ffffffff8119778d>] do_mmap_pgoff+0x31d/0x400 [ 403.118513] [<ffffffff81182940>] vm_mmap_pgoff+0x90/0xc0 [ 403.118517] [<ffffffff81195cef>] SyS_mmap_pgoff+0x1df/0x2a0 [ 403.118521] [<ffffffff810085c2>] SyS_mmap+0x22/0x30 [ 403.118525] [<ffffffff8199dc69>] system_call_fastpath+0x16/0x1b The reason is that vb2_fop_mmap and vb2_fop_get_unmapped_area take the core lock while they are called with the mmap_sem semaphore held. But elsewhere in the code the core lock is taken first but calls to copy_to/from_user() can take the mmap_sem semaphore as well, potentially causing a classical A-B/B-A deadlock. However, the mmap/get_unmapped_area calls really shouldn't take the core lock at all. So what would happen if they don't take the core lock anymore? There are two situations that need to be taken into account: calling mmap while new buffers are being added and calling mmap while buffers are being deleted. The first case works almost fine without a lock: in all cases mmap relies on correctly filled-in q->num_buffers/q->num_planes values and those are only updated by reqbufs and create_buffers *after* any new buffers have been initialized completely. Except in one case: if an error occurred while allocating the buffers it will increase num_buffers and rely on __vb2_queue_free to decrease it again. So there is a short period where the buffer information may be wrong. The second case definitely does pose a problem: buffers may be in the process of being deleted, without the internal structure being updated. In order to fix this a new mutex is added to vb2_queue that is taken when buffers are allocated or deleted, and in vb2_mmap. That way vb2_mmap won't get stale buffer data. Note that this is a problem only for MEMORY_MMAP, so even though __qbuf_userptr and __qbuf_dmabuf also mess around with buffers (mem_priv in particular), this doesn't clash with vb2_mmap or vb2_get_unmapped_area since those are MMAP specific. As an additional bonus the hack in __buf_prepare, the USERPTR case, can be removed as well since mmap() no longer takes the core lock. All in all a much cleaner solution. Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com> Acked-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com> Acked-by: Marek Szyprowski <m.szyprowski@samsung.com> Signed-off-by: Mauro Carvalho Chehab <m.chehab@samsung.com>
2014-08-07 03:47:14 -03:00
[media] vb2: only call start_streaming if sufficient buffers are queued In commit 02f142ecd24aaf891324ffba8527284c1731b561 support was added to start_streaming to return -ENOBUFS if insufficient buffers were queued for the DMA engine to start. The vb2 core would attempt calling start_streaming again if another buffer would be queued up. Later analysis uncovered problems with the queue management if start_streaming would return an error: the buffers are enqueued to the driver before the start_streaming op is called, so after an error they are never returned to the vb2 core. The solution for this is to let the driver return them to the vb2 core in case of an error while starting the DMA engine. However, in the case of -ENOBUFS that would be weird: it is not a real error, it just says that more buffers are needed. Requiring start_streaming to give them back only to have them requeued again the next time the application calls QBUF is inefficient. This patch changes this mechanism: it adds a 'min_buffers_needed' field to vb2_queue that drivers can set with the minimum number of buffers required to start the DMA engine. The start_streaming op is only called if enough buffers are queued. The -ENOBUFS handling has been dropped in favor of this new method. Drivers are expected to return buffers back to vb2 core with state QUEUED if start_streaming would return an error. The vb2 core checks for this and produces a warning if that didn't happen and it will forcefully reclaim such buffers to ensure that the internal vb2 core state remains consistent and all buffer-related resources have been correctly freed and all op calls have been balanced. __reqbufs() has been updated to check that at least min_buffers_needed buffers could be allocated. If fewer buffers were allocated then __reqbufs will free what was allocated and return -ENOMEM. Based on a suggestion from Pawel Osciak. __create_bufs() doesn't do that check, since the use of __create_bufs assumes some advance scenario where the user might want more control. Instead streamon will check if enough buffers were allocated to prevent streaming with fewer than the minimum required number of buffers. Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com> Signed-off-by: Mauro Carvalho Chehab <m.chehab@samsung.com>
2014-02-24 13:51:03 -03:00
[media] vb2: only call start_streaming if sufficient buffers are queued In commit 02f142ecd24aaf891324ffba8527284c1731b561 support was added to start_streaming to return -ENOBUFS if insufficient buffers were queued for the DMA engine to start. The vb2 core would attempt calling start_streaming again if another buffer would be queued up. Later analysis uncovered problems with the queue management if start_streaming would return an error: the buffers are enqueued to the driver before the start_streaming op is called, so after an error they are never returned to the vb2 core. The solution for this is to let the driver return them to the vb2 core in case of an error while starting the DMA engine. However, in the case of -ENOBUFS that would be weird: it is not a real error, it just says that more buffers are needed. Requiring start_streaming to give them back only to have them requeued again the next time the application calls QBUF is inefficient. This patch changes this mechanism: it adds a 'min_buffers_needed' field to vb2_queue that drivers can set with the minimum number of buffers required to start the DMA engine. The start_streaming op is only called if enough buffers are queued. The -ENOBUFS handling has been dropped in favor of this new method. Drivers are expected to return buffers back to vb2 core with state QUEUED if start_streaming would return an error. The vb2 core checks for this and produces a warning if that didn't happen and it will forcefully reclaim such buffers to ensure that the internal vb2 core state remains consistent and all buffer-related resources have been correctly freed and all op calls have been balanced. __reqbufs() has been updated to check that at least min_buffers_needed buffers could be allocated. If fewer buffers were allocated then __reqbufs will free what was allocated and return -ENOMEM. Based on a suggestion from Pawel Osciak. __create_bufs() doesn't do that check, since the use of __create_bufs assumes some advance scenario where the user might want more control. Instead streamon will check if enough buffers were allocated to prevent streaming with fewer than the minimum required number of buffers. Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com> Signed-off-by: Mauro Carvalho Chehab <m.chehab@samsung.com>
2014-02-24 13:51:03 -03:00
dma-mapping: use unsigned long for dma_attrs The dma-mapping core and the implementations do not change the DMA attributes passed by pointer. Thus the pointer can point to const data. However the attributes do not have to be a bitfield. Instead unsigned long will do fine: 1. This is just simpler. Both in terms of reading the code and setting attributes. Instead of initializing local attributes on the stack and passing pointer to it to dma_set_attr(), just set the bits. 2. It brings safeness and checking for const correctness because the attributes are passed by value. Semantic patches for this change (at least most of them): virtual patch virtual context @r@ identifier f, attrs; @@ f(..., - struct dma_attrs *attrs + unsigned long attrs , ...) { ... } @@ identifier r.f; @@ f(..., - NULL + 0 ) and // Options: --all-includes virtual patch virtual context @r@ identifier f, attrs; type t; @@ t f(..., struct dma_attrs *attrs); @@ identifier r.f; @@ f(..., - NULL + 0 ) Link: http://lkml.kernel.org/r/1468399300-5399-2-git-send-email-k.kozlowski@samsung.com Signed-off-by: Krzysztof Kozlowski <k.kozlowski@samsung.com> Acked-by: Vineet Gupta <vgupta@synopsys.com> Acked-by: Robin Murphy <robin.murphy@arm.com> Acked-by: Hans-Christian Noren Egtvedt <egtvedt@samfundet.no> Acked-by: Mark Salter <msalter@redhat.com> [c6x] Acked-by: Jesper Nilsson <jesper.nilsson@axis.com> [cris] Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch> [drm] Reviewed-by: Bart Van Assche <bart.vanassche@sandisk.com> Acked-by: Joerg Roedel <jroedel@suse.de> [iommu] Acked-by: Fabien Dessenne <fabien.dessenne@st.com> [bdisp] Reviewed-by: Marek Szyprowski <m.szyprowski@samsung.com> [vb2-core] Acked-by: David Vrabel <david.vrabel@citrix.com> [xen] Acked-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> [xen swiotlb] Acked-by: Joerg Roedel <jroedel@suse.de> [iommu] Acked-by: Richard Kuo <rkuo@codeaurora.org> [hexagon] Acked-by: Geert Uytterhoeven <geert@linux-m68k.org> [m68k] Acked-by: Gerald Schaefer <gerald.schaefer@de.ibm.com> [s390] Acked-by: Bjorn Andersson <bjorn.andersson@linaro.org> Acked-by: Hans-Christian Noren Egtvedt <egtvedt@samfundet.no> [avr32] Acked-by: Vineet Gupta <vgupta@synopsys.com> [arc] Acked-by: Robin Murphy <robin.murphy@arm.com> [arm64 and dma-iommu] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-08-03 13:46:00 -07:00
[media] videobuf2: fix lockdep warning The following lockdep warning has been there ever since commit a517cca6b24fc54ac209e44118ec8962051662e3 one year ago: [ 403.117947] ====================================================== [ 403.117949] [ INFO: possible circular locking dependency detected ] [ 403.117953] 3.16.0-rc6-test-media #961 Not tainted [ 403.117954] ------------------------------------------------------- [ 403.117956] v4l2-ctl/15377 is trying to acquire lock: [ 403.117959] (&dev->mutex#3){+.+.+.}, at: [<ffffffffa005a6c3>] vb2_fop_mmap+0x33/0x90 [videobuf2_core] [ 403.117974] [ 403.117974] but task is already holding lock: [ 403.117976] (&mm->mmap_sem){++++++}, at: [<ffffffff8118291f>] vm_mmap_pgoff+0x6f/0xc0 [ 403.117987] [ 403.117987] which lock already depends on the new lock. [ 403.117987] [ 403.117990] [ 403.117990] the existing dependency chain (in reverse order) is: [ 403.117992] [ 403.117992] -> #1 (&mm->mmap_sem){++++++}: [ 403.117997] [<ffffffff810d733c>] validate_chain.isra.39+0x5fc/0x9a0 [ 403.118006] [<ffffffff810d8bc3>] __lock_acquire+0x4d3/0xd30 [ 403.118010] [<ffffffff810d9da7>] lock_acquire+0xa7/0x160 [ 403.118014] [<ffffffff8118c9ec>] might_fault+0x7c/0xb0 [ 403.118018] [<ffffffffa0028a25>] video_usercopy+0x425/0x610 [videodev] [ 403.118028] [<ffffffffa0028c25>] video_ioctl2+0x15/0x20 [videodev] [ 403.118034] [<ffffffffa0022764>] v4l2_ioctl+0x184/0x1a0 [videodev] [ 403.118040] [<ffffffff811d77d0>] do_vfs_ioctl+0x2f0/0x4f0 [ 403.118307] [<ffffffff811d7a51>] SyS_ioctl+0x81/0xa0 [ 403.118311] [<ffffffff8199dc69>] system_call_fastpath+0x16/0x1b [ 403.118319] [ 403.118319] -> #0 (&dev->mutex#3){+.+.+.}: [ 403.118324] [<ffffffff810d6a96>] check_prevs_add+0x746/0x9f0 [ 403.118329] [<ffffffff810d733c>] validate_chain.isra.39+0x5fc/0x9a0 [ 403.118333] [<ffffffff810d8bc3>] __lock_acquire+0x4d3/0xd30 [ 403.118336] [<ffffffff810d9da7>] lock_acquire+0xa7/0x160 [ 403.118340] [<ffffffff81999664>] mutex_lock_interruptible_nested+0x64/0x640 [ 403.118344] [<ffffffffa005a6c3>] vb2_fop_mmap+0x33/0x90 [videobuf2_core] [ 403.118349] [<ffffffffa0022122>] v4l2_mmap+0x62/0xa0 [videodev] [ 403.118354] [<ffffffff81197270>] mmap_region+0x3d0/0x5d0 [ 403.118359] [<ffffffff8119778d>] do_mmap_pgoff+0x31d/0x400 [ 403.118363] [<ffffffff81182940>] vm_mmap_pgoff+0x90/0xc0 [ 403.118366] [<ffffffff81195cef>] SyS_mmap_pgoff+0x1df/0x2a0 [ 403.118369] [<ffffffff810085c2>] SyS_mmap+0x22/0x30 [ 403.118376] [<ffffffff8199dc69>] system_call_fastpath+0x16/0x1b [ 403.118381] [ 403.118381] other info that might help us debug this: [ 403.118381] [ 403.118383] Possible unsafe locking scenario: [ 403.118383] [ 403.118385] CPU0 CPU1 [ 403.118387] ---- ---- [ 403.118388] lock(&mm->mmap_sem); [ 403.118391] lock(&dev->mutex#3); [ 403.118394] lock(&mm->mmap_sem); [ 403.118397] lock(&dev->mutex#3); [ 403.118400] [ 403.118400] *** DEADLOCK *** [ 403.118400] [ 403.118403] 1 lock held by v4l2-ctl/15377: [ 403.118405] #0: (&mm->mmap_sem){++++++}, at: [<ffffffff8118291f>] vm_mmap_pgoff+0x6f/0xc0 [ 403.118411] [ 403.118411] stack backtrace: [ 403.118415] CPU: 0 PID: 15377 Comm: v4l2-ctl Not tainted 3.16.0-rc6-test-media #961 [ 403.118418] Hardware name: VMware, Inc. VMware Virtual Platform/440BX Desktop Reference Platform, BIOS 6.00 07/31/2013 [ 403.118420] ffffffff82a6c9d0 ffff8800af37fb00 ffffffff819916a2 ffffffff82a6c9d0 [ 403.118425] ffff8800af37fb40 ffffffff810d5715 ffff8802308e4200 0000000000000000 [ 403.118429] ffff8802308e4a48 ffff8802308e4a48 ffff8802308e4200 0000000000000001 [ 403.118433] Call Trace: [ 403.118441] [<ffffffff819916a2>] dump_stack+0x4e/0x7a [ 403.118445] [<ffffffff810d5715>] print_circular_bug+0x1d5/0x2a0 [ 403.118449] [<ffffffff810d6a96>] check_prevs_add+0x746/0x9f0 [ 403.118455] [<ffffffff8119c172>] ? find_vmap_area+0x42/0x70 [ 403.118459] [<ffffffff810d733c>] validate_chain.isra.39+0x5fc/0x9a0 [ 403.118463] [<ffffffff810d8bc3>] __lock_acquire+0x4d3/0xd30 [ 403.118468] [<ffffffff810d9da7>] lock_acquire+0xa7/0x160 [ 403.118472] [<ffffffffa005a6c3>] ? vb2_fop_mmap+0x33/0x90 [videobuf2_core] [ 403.118476] [<ffffffffa005a6c3>] ? vb2_fop_mmap+0x33/0x90 [videobuf2_core] [ 403.118480] [<ffffffff81999664>] mutex_lock_interruptible_nested+0x64/0x640 [ 403.118484] [<ffffffffa005a6c3>] ? vb2_fop_mmap+0x33/0x90 [videobuf2_core] [ 403.118488] [<ffffffffa005a6c3>] ? vb2_fop_mmap+0x33/0x90 [videobuf2_core] [ 403.118493] [<ffffffff810d8055>] ? mark_held_locks+0x75/0xa0 [ 403.118497] [<ffffffffa005a6c3>] vb2_fop_mmap+0x33/0x90 [videobuf2_core] [ 403.118502] [<ffffffffa0022122>] v4l2_mmap+0x62/0xa0 [videodev] [ 403.118506] [<ffffffff81197270>] mmap_region+0x3d0/0x5d0 [ 403.118510] [<ffffffff8119778d>] do_mmap_pgoff+0x31d/0x400 [ 403.118513] [<ffffffff81182940>] vm_mmap_pgoff+0x90/0xc0 [ 403.118517] [<ffffffff81195cef>] SyS_mmap_pgoff+0x1df/0x2a0 [ 403.118521] [<ffffffff810085c2>] SyS_mmap+0x22/0x30 [ 403.118525] [<ffffffff8199dc69>] system_call_fastpath+0x16/0x1b The reason is that vb2_fop_mmap and vb2_fop_get_unmapped_area take the core lock while they are called with the mmap_sem semaphore held. But elsewhere in the code the core lock is taken first but calls to copy_to/from_user() can take the mmap_sem semaphore as well, potentially causing a classical A-B/B-A deadlock. However, the mmap/get_unmapped_area calls really shouldn't take the core lock at all. So what would happen if they don't take the core lock anymore? There are two situations that need to be taken into account: calling mmap while new buffers are being added and calling mmap while buffers are being deleted. The first case works almost fine without a lock: in all cases mmap relies on correctly filled-in q->num_buffers/q->num_planes values and those are only updated by reqbufs and create_buffers *after* any new buffers have been initialized completely. Except in one case: if an error occurred while allocating the buffers it will increase num_buffers and rely on __vb2_queue_free to decrease it again. So there is a short period where the buffer information may be wrong. The second case definitely does pose a problem: buffers may be in the process of being deleted, without the internal structure being updated. In order to fix this a new mutex is added to vb2_queue that is taken when buffers are allocated or deleted, and in vb2_mmap. That way vb2_mmap won't get stale buffer data. Note that this is a problem only for MEMORY_MMAP, so even though __qbuf_userptr and __qbuf_dmabuf also mess around with buffers (mem_priv in particular), this doesn't clash with vb2_mmap or vb2_get_unmapped_area since those are MMAP specific. As an additional bonus the hack in __buf_prepare, the USERPTR case, can be removed as well since mmap() no longer takes the core lock. All in all a much cleaner solution. Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com> Acked-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com> Acked-by: Marek Szyprowski <m.szyprowski@samsung.com> Signed-off-by: Mauro Carvalho Chehab <m.chehab@samsung.com>
2014-08-07 03:47:14 -03:00
[media] vb2: only call start_streaming if sufficient buffers are queued In commit 02f142ecd24aaf891324ffba8527284c1731b561 support was added to start_streaming to return -ENOBUFS if insufficient buffers were queued for the DMA engine to start. The vb2 core would attempt calling start_streaming again if another buffer would be queued up. Later analysis uncovered problems with the queue management if start_streaming would return an error: the buffers are enqueued to the driver before the start_streaming op is called, so after an error they are never returned to the vb2 core. The solution for this is to let the driver return them to the vb2 core in case of an error while starting the DMA engine. However, in the case of -ENOBUFS that would be weird: it is not a real error, it just says that more buffers are needed. Requiring start_streaming to give them back only to have them requeued again the next time the application calls QBUF is inefficient. This patch changes this mechanism: it adds a 'min_buffers_needed' field to vb2_queue that drivers can set with the minimum number of buffers required to start the DMA engine. The start_streaming op is only called if enough buffers are queued. The -ENOBUFS handling has been dropped in favor of this new method. Drivers are expected to return buffers back to vb2 core with state QUEUED if start_streaming would return an error. The vb2 core checks for this and produces a warning if that didn't happen and it will forcefully reclaim such buffers to ensure that the internal vb2 core state remains consistent and all buffer-related resources have been correctly freed and all op calls have been balanced. __reqbufs() has been updated to check that at least min_buffers_needed buffers could be allocated. If fewer buffers were allocated then __reqbufs will free what was allocated and return -ENOMEM. Based on a suggestion from Pawel Osciak. __create_bufs() doesn't do that check, since the use of __create_bufs assumes some advance scenario where the user might want more control. Instead streamon will check if enough buffers were allocated to prevent streaming with fewer than the minimum required number of buffers. Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com> Signed-off-by: Mauro Carvalho Chehab <m.chehab@samsung.com>
2014-02-24 13:51:03 -03:00
[media] vb2: only call start_streaming if sufficient buffers are queued In commit 02f142ecd24aaf891324ffba8527284c1731b561 support was added to start_streaming to return -ENOBUFS if insufficient buffers were queued for the DMA engine to start. The vb2 core would attempt calling start_streaming again if another buffer would be queued up. Later analysis uncovered problems with the queue management if start_streaming would return an error: the buffers are enqueued to the driver before the start_streaming op is called, so after an error they are never returned to the vb2 core. The solution for this is to let the driver return them to the vb2 core in case of an error while starting the DMA engine. However, in the case of -ENOBUFS that would be weird: it is not a real error, it just says that more buffers are needed. Requiring start_streaming to give them back only to have them requeued again the next time the application calls QBUF is inefficient. This patch changes this mechanism: it adds a 'min_buffers_needed' field to vb2_queue that drivers can set with the minimum number of buffers required to start the DMA engine. The start_streaming op is only called if enough buffers are queued. The -ENOBUFS handling has been dropped in favor of this new method. Drivers are expected to return buffers back to vb2 core with state QUEUED if start_streaming would return an error. The vb2 core checks for this and produces a warning if that didn't happen and it will forcefully reclaim such buffers to ensure that the internal vb2 core state remains consistent and all buffer-related resources have been correctly freed and all op calls have been balanced. __reqbufs() has been updated to check that at least min_buffers_needed buffers could be allocated. If fewer buffers were allocated then __reqbufs will free what was allocated and return -ENOMEM. Based on a suggestion from Pawel Osciak. __create_bufs() doesn't do that check, since the use of __create_bufs assumes some advance scenario where the user might want more control. Instead streamon will check if enough buffers were allocated to prevent streaming with fewer than the minimum required number of buffers. Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com> Signed-off-by: Mauro Carvalho Chehab <m.chehab@samsung.com>
2014-02-24 13:51:03 -03:00
[media] vb2: fix nasty vb2_thread regression The vb2_thread implementation was made generic and was moved from videobuf2-v4l2.c to videobuf2-core.c in commit af3bac1a. Unfortunately that clearly was never tested since it broke read() causing NULL address references. The root cause was confused handling of vb2_buffer vs v4l2_buffer (the pb pointer in various core functions). The v4l2_buffer no longer exists after moving the code into the core and it is no longer needed. However, the vb2_thread code passed a pointer to a vb2_buffer to the core functions were a v4l2_buffer pointer was expected and vb2_thread expected that the vb2_buffer fields would be filled in correctly. This is obviously wrong since v4l2_buffer != vb2_buffer. Note that the pb pointer is a void pointer, so no type-checking took place. This patch fixes this problem: 1) allow pb to be NULL for vb2_core_(d)qbuf. The vb2_thread code will use a NULL pointer here since they don't care about v4l2_buffer anyway. 2) let vb2_core_dqbuf pass back the index of the received buffer. This is all vb2_thread needs: this index is the index into the q->bufs array and vb2_thread just gets the vb2_buffer from there. 3) the fileio->b pointer (that originally contained a v4l2_buffer) is removed altogether since it is no longer needed. Tested with vivid and the cobalt driver. Cc: stable@vger.kernel.org # Kernel >= 4.3 Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com> Reported-by: Matthias Schwarzott <zzam@gentoo.org> Signed-off-by: Mauro Carvalho Chehab <mchehab@osg.samsung.com>
2016-01-27 10:08:42 -02:00