Linaro-mm-sig September 2016

linaro-mm-sig@lists.linaro.org

13 participants
15 discussions

[PATCH] allow buffers allocation with DMABUF memory type

by Benjamin Gaignard

Until now the only way to make the driver allocate buffers and share them using dma_buf was to use V4L2_MEMORY_MMAP memory type. Use of MMAP memory type is a problem because vb2 never call dma_buf_map_attachment() to attach itself while queuing the buffer so dma_buf importer will not know that another device use this buffer. This patch allow to allocate buffer even for DMABUF memory type and correctly manage dma_buf buffer attachment. vb2_mem_ops attach_dmabuf() prototype has been changed to be able to distinguish if the attachment is done on a already existing buffer or on imported one. Signed-off-by: Benjamin Gaignard <benjamin.gaignard(a)linaro.org> --- drivers/media/v4l2-core/videobuf2-core.c | 76 ++++++++++++++++++++++++-- drivers/media/v4l2-core/videobuf2-dma-contig.c | 10 ++-- drivers/media/v4l2-core/videobuf2-dma-sg.c | 8 ++- drivers/media/v4l2-core/videobuf2-vmalloc.c | 8 ++- include/media/videobuf2-core.h | 5 +- 5 files changed, 91 insertions(+), 16 deletions(-) diff --git a/drivers/media/v4l2-core/videobuf2-core.c b/drivers/media/v4l2-core/videobuf2-core.c index 1329dcc..c5968aa 100644 --- a/drivers/media/v4l2-core/videobuf2-core.c +++ b/drivers/media/v4l2-core/videobuf2-core.c @@ -337,6 +337,67 @@ static void __setup_offsets(struct vb2_queue *q, unsigned int n) } /** + * __setup_dmabufs() - setup dmabuf fd for every plane in + * every buffer on the queue + */ +static void __setup_dmabufs(struct vb2_queue *q, unsigned int n) +{ + unsigned int buffer, plane; + struct vb2_buffer *vb; + struct vb2_plane *vb_plane; + struct dma_buf *dbuf; + void *mem_priv; + int fd; + int write = !V4L2_TYPE_IS_OUTPUT(q->type); + int flags = write ? O_WRONLY : O_RDONLY; + + for (buffer = q->num_buffers; buffer < q->num_buffers + n; ++buffer) { + vb = q->bufs[buffer]; + if (!vb) + continue; + + for (plane = 0; plane < vb->num_planes; ++plane) { + vb_plane = &vb->planes[plane]; + + dbuf = call_ptr_memop(vb, get_dmabuf, + vb_plane->mem_priv, + flags & O_ACCMODE); + if (IS_ERR_OR_NULL(dbuf)) { + dprintk(1, "Failed to export buffer %d, " \ + "plane %d\n", buffer, plane); + continue; + } + + fd = dma_buf_fd(dbuf, flags & ~O_ACCMODE); + if (fd < 0) { + dprintk(3, "buffer %d, plane %d failed " \ + "to export (%d)\n", buffer, plane, fd); + dma_buf_put(dbuf); + continue; + } + + /* Acquire each plane's memory */ + mem_priv = call_ptr_memop(vb, attach_dmabuf, + q->alloc_ctx[plane], dbuf, + vb->v4l2_planes[plane].length, + write, vb_plane->mem_priv); + if (IS_ERR(mem_priv)) { + dprintk(1, "Buffer %d plane %d failed "\ + "to attach dmabuf\n", buffer, plane); + dma_buf_put(dbuf); + continue; + } + + vb_plane->dbuf = dbuf; + vb->v4l2_planes[plane].m.fd = fd; + + dprintk(3, "Buffer %d, plane %d fd %x\n", + buffer, plane, fd); + } + } +} + +/** * __vb2_queue_alloc() - allocate videobuf buffer structures and (for MMAP type) * video buffer memory for all buffers/planes on the queue and initializes the * queue @@ -369,8 +430,9 @@ static int __vb2_queue_alloc(struct vb2_queue *q, enum v4l2_memory memory, vb->v4l2_buf.type = q->type; vb->v4l2_buf.memory = memory; - /* Allocate video buffer memory for the MMAP type */ - if (memory == V4L2_MEMORY_MMAP) { + /* Allocate video buffer memory for the MMAP and DMABUF types */ + if (memory == V4L2_MEMORY_MMAP || + memory == V4L2_MEMORY_DMABUF) { ret = __vb2_buf_mem_alloc(vb); if (ret) { dprintk(1, "failed allocating memory for " @@ -400,6 +462,9 @@ static int __vb2_queue_alloc(struct vb2_queue *q, enum v4l2_memory memory, if (memory == V4L2_MEMORY_MMAP) __setup_offsets(q, buffer); + if (memory == V4L2_MEMORY_DMABUF) + __setup_dmabufs(q, buffer); + dprintk(1, "allocated %d buffers, %d plane(s) each\n", buffer, num_planes); @@ -859,7 +924,7 @@ static int __verify_memory_type(struct vb2_queue *q, * to be used during streaming, * 4) allocates internal buffer structures (struct vb2_buffer), according to * the agreed parameters, - * 5) for MMAP memory type, allocates actual video memory, using the + * 5) for MMAP and DMABUF memory types, allocates actual video memory, using the * memory handling/allocation routines provided during queue initialization * * If req->count is 0, all the memory will be freed instead. @@ -885,7 +950,8 @@ static int __reqbufs(struct vb2_queue *q, struct v4l2_requestbuffers *req) * are not in use and can be freed. */ mutex_lock(&q->mmap_lock); - if (q->memory == V4L2_MEMORY_MMAP && __buffers_in_use(q)) { + if ((q->memory == V4L2_MEMORY_MMAP || + q->memory == V4L2_MEMORY_DMABUF) && __buffers_in_use(q)) { mutex_unlock(&q->mmap_lock); dprintk(1, "memory in use, cannot free\n"); return -EBUSY; @@ -1540,7 +1606,7 @@ static int __qbuf_dmabuf(struct vb2_buffer *vb, const struct v4l2_buffer *b) /* Acquire each plane's memory */ mem_priv = call_ptr_memop(vb, attach_dmabuf, q->alloc_ctx[plane], - dbuf, planes[plane].length, dma_dir); + dbuf, planes[plane].length, dma_dir, NULL); if (IS_ERR(mem_priv)) { dprintk(1, "failed to attach dmabuf\n"); ret = PTR_ERR(mem_priv); diff --git a/drivers/media/v4l2-core/videobuf2-dma-contig.c b/drivers/media/v4l2-core/videobuf2-dma-contig.c index b481d20..8dc945e 100644 --- a/drivers/media/v4l2-core/videobuf2-dma-contig.c +++ b/drivers/media/v4l2-core/videobuf2-dma-contig.c @@ -793,7 +793,7 @@ static void vb2_dc_detach_dmabuf(void *mem_priv) struct vb2_dc_buf *buf = mem_priv; /* if vb2 works correctly you should never detach mapped buffer */ - if (WARN_ON(buf->dma_addr)) + if (WARN_ON(buf->dma_sgt)) vb2_dc_unmap_dmabuf(buf); /* detach this attachment */ @@ -802,16 +802,18 @@ static void vb2_dc_detach_dmabuf(void *mem_priv) } static void *vb2_dc_attach_dmabuf(void *alloc_ctx, struct dma_buf *dbuf, - unsigned long size, enum dma_data_direction dma_dir) + unsigned long size, enum dma_data_direction dma_dir, void *buf_priv) { struct vb2_dc_conf *conf = alloc_ctx; - struct vb2_dc_buf *buf; + struct vb2_dc_buf *buf = buf_priv; struct dma_buf_attachment *dba; if (dbuf->size < size) return ERR_PTR(-EFAULT); - buf = kzalloc(sizeof(*buf), GFP_KERNEL); + if (!buf) + buf = kzalloc(sizeof(*buf), GFP_KERNEL); + if (!buf) return ERR_PTR(-ENOMEM); diff --git a/drivers/media/v4l2-core/videobuf2-dma-sg.c b/drivers/media/v4l2-core/videobuf2-dma-sg.c index b1838ab..b9855fd 100644 --- a/drivers/media/v4l2-core/videobuf2-dma-sg.c +++ b/drivers/media/v4l2-core/videobuf2-dma-sg.c @@ -667,16 +667,18 @@ static void vb2_dma_sg_detach_dmabuf(void *mem_priv) } static void *vb2_dma_sg_attach_dmabuf(void *alloc_ctx, struct dma_buf *dbuf, - unsigned long size, enum dma_data_direction dma_dir) + unsigned long size, enum dma_data_direction dma_dir, void *buf_priv) { struct vb2_dma_sg_conf *conf = alloc_ctx; - struct vb2_dma_sg_buf *buf; + struct vb2_dma_sg_buf *buf = buf_priv; struct dma_buf_attachment *dba; if (dbuf->size < size) return ERR_PTR(-EFAULT); - buf = kzalloc(sizeof(*buf), GFP_KERNEL); + if (!buf) + buf = kzalloc(sizeof(*buf), GFP_KERNEL); + if (!buf) return ERR_PTR(-ENOMEM); diff --git a/drivers/media/v4l2-core/videobuf2-vmalloc.c b/drivers/media/v4l2-core/videobuf2-vmalloc.c index bcde885..373294c 100644 --- a/drivers/media/v4l2-core/videobuf2-vmalloc.c +++ b/drivers/media/v4l2-core/videobuf2-vmalloc.c @@ -416,14 +416,16 @@ static void vb2_vmalloc_detach_dmabuf(void *mem_priv) } static void *vb2_vmalloc_attach_dmabuf(void *alloc_ctx, struct dma_buf *dbuf, - unsigned long size, enum dma_data_direction dma_dir) + unsigned long size, enum dma_data_direction dma_dir, void *buf_priv) { - struct vb2_vmalloc_buf *buf; + struct vb2_vmalloc_buf *buf = buf_priv; if (dbuf->size < size) return ERR_PTR(-EFAULT); - buf = kzalloc(sizeof(*buf), GFP_KERNEL); + if (!buf) + buf = kzalloc(sizeof(*buf), GFP_KERNEL); + if (!buf) return ERR_PTR(-ENOMEM); diff --git a/include/media/videobuf2-core.h b/include/media/videobuf2-core.h index a5790fd..374b26b 100644 --- a/include/media/videobuf2-core.h +++ b/include/media/videobuf2-core.h @@ -49,6 +49,8 @@ struct vb2_threadio_data; * used for DMABUF memory types; alloc_ctx is the alloc context * dbuf is the shared dma_buf; returns NULL on failure; * allocator private per-buffer structure on success; + * if private per-buffer structure is provided reuse it + * instead of allocating a new one. * this needs to be used for further accesses to the buffer. * @detach_dmabuf: inform the exporter of the buffer that the current DMABUF * buffer is no longer used; the buf_priv argument is the @@ -98,7 +100,8 @@ struct vb2_mem_ops { void *(*attach_dmabuf)(void *alloc_ctx, struct dma_buf *dbuf, unsigned long size, - enum dma_data_direction dma_dir); + enum dma_data_direction dma_dir, + void *buf_priv); void (*detach_dmabuf)(void *buf_priv); int (*map_dmabuf)(void *buf_priv); void (*unmap_dmabuf)(void *buf_priv); -- 1.9.1

7 years, 10 months

[PATCH] dma-buf: Wait on the reservation object when sync'ing before CPU access

by Chris Wilson

Rendering operations to the dma-buf are tracked implicitly via the reservation_object (dmabuf->resv). This is used to allow poll() to wait upon outstanding rendering (or just query the current status of rendering). The dma-buf sync ioctl allows userspace to prepare the dma-buf for CPU access, which should include waiting upon rendering. (Some drivers may need to do more work to ensure that the dma-buf mmap is coherent as well as complete.) Signed-off-by: Chris Wilson <chris(a)chris-wilson.co.uk> Cc: Sumit Semwal <sumit.semwal(a)linaro.org> Cc: Daniel Vetter <daniel.vetter(a)ffwll.ch> Cc: linux-media(a)vger.kernel.org Cc: dri-devel(a)lists.freedesktop.org Cc: linaro-mm-sig(a)lists.linaro.org Cc: linux-kernel(a)vger.kernel.org --- I'm wondering whether it makes sense just to always do the wait first. It is one of the first operations every driver has to make. A driver that wants to implement it differently (e.g. they can special case native waits) will still require a wait on the reservation object to finish external rendering. -Chris --- drivers/dma-buf/dma-buf.c | 18 ++++++++++++++++++ 1 file changed, 18 insertions(+) diff --git a/drivers/dma-buf/dma-buf.c b/drivers/dma-buf/dma-buf.c index ddaee60ae52a..123f14b8e882 100644 --- a/drivers/dma-buf/dma-buf.c +++ b/drivers/dma-buf/dma-buf.c @@ -586,6 +586,22 @@ void dma_buf_unmap_attachment(struct dma_buf_attachment *attach, } EXPORT_SYMBOL_GPL(dma_buf_unmap_attachment); +static int __dma_buf_begin_cpu_access(struct dma_buf *dmabuf, + enum dma_data_direction direction) +{ + bool write = (direction == DMA_BIDIRECTIONAL || + direction == DMA_TO_DEVICE); + struct reservation_object *resv = dma_buf->resv; + long ret; + + /* Wait on any implicit rendering fences */ + ret = reservation_object_wait_timeout_rcu(resv, write, true, + MAX_SCHEDULE_TIMEOUT); + if (ret < 0) + return ret; + + return 0; +} /** * dma_buf_begin_cpu_access - Must be called before accessing a dma_buf from the @@ -607,6 +623,8 @@ int dma_buf_begin_cpu_access(struct dma_buf *dmabuf, if (dmabuf->ops->begin_cpu_access) ret = dmabuf->ops->begin_cpu_access(dmabuf, direction); + else + ret = __dma_buf_begin_cpu_access(dmabuf, direction); return ret; } -- 2.8.1

8 years

Re: [Linaro-mm-sig] [Intel-gfx] [PATCH 11/11] dma-buf: Do a fast lockless check for poll with timeout=0

by Daniel Vetter

On Fri, Sep 23, 2016 at 07:59:44PM +0200, Christian König wrote: > Am 23.09.2016 um 17:20 schrieb Chris Wilson: > > On Fri, Sep 23, 2016 at 03:50:44PM +0200, Daniel Vetter wrote: > > > On Mon, Aug 29, 2016 at 08:08:34AM +0100, Chris Wilson wrote: > > > > Currently we install a callback for performing poll on a dma-buf, > > > > irrespective of the timeout. This involves taking a spinlock, as well as > > > > unnecessary work, and greatly reduces scaling of poll(.timeout=0) across > > > > multiple threads. > > > > > > > > We can query whether the poll will block prior to installing the > > > > callback to make the busy-query fast. > > > > > > > > Single thread: 60% faster > > > > 8 threads on 4 (+4 HT) cores: 600% faster > > > > > > > > Still not quite the perfect scaling we get with a native busy ioctl, but > > > > poll(dmabuf) is faster due to the quicker lookup of the object and > > > > avoiding drm_ioctl(). > > > > > > > > Signed-off-by: Chris Wilson <chris(a)chris-wilson.co.uk> > > > > Cc: Sumit Semwal <sumit.semwal(a)linaro.org> > > > > Cc: linux-media(a)vger.kernel.org > > > > Cc: dri-devel(a)lists.freedesktop.org > > > > Cc: linaro-mm-sig(a)lists.linaro.org > > > > Reviewed-by: Daniel Vetter <daniel.vetter(a)ffwll.ch> > > > Need to strike the r-b here, since Christian König pointed out that > > > objects won't magically switch signalling on. > > Oh, it also means that > > > > commit fb8b7d2b9d80e1e71f379e57355936bd2b024be9 > > Author: Jammy Zhou <Jammy.Zhou(a)amd.com> > > Date: Wed Jan 21 18:35:47 2015 +0800 > > > > reservation: wait only with non-zero timeout specified (v3) > > When the timeout value passed to reservation_object_wait_timeout_rcu > > is zero, no wait should be done if the fences are not signaled. > > Return '1' for idle and '0' for busy if the specified timeout is '0' > > to keep consistent with the case of non-zero timeout. > > v2: call fence_put if not signaled in the case of timeout==0 > > v3: switch to reservation_object_test_signaled_rcu > > Signed-off-by: Jammy Zhou <Jammy.Zhou(a)amd.com> > > Reviewed-by: Christian König <christian.koenig(a)amd.com> > > Reviewed-by: Alex Deucher <alexander.deucher(a)amd.com> > > Reviewed-By: Maarten Lankhorst <maarten.lankhorst(a)canonical.com> > > Signed-off-by: Sumit Semwal <sumit.semwal(a)linaro.org> > > > > is wrong. And reservation_object_test_signaled_rcu() is unreliable. > > Ups indeed, that patch is wrong as well. > > I suggest that we just enable the signaling in this case as well. Will you/Zhou take care of this corner case? Just so I can't forget about it ;-) Thanks, Daniel -- Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch

8 years, 3 months

[PATCH 10/11] dma-buf: Use seqlock to close RCU race in test_signaled_single

by Chris Wilson

With the seqlock now extended to cover the lookup of the fence and its testing, we can perform that testing solely under the seqlock guard and avoid the effective locking and serialisation of acquiring a reference to the request. As the fence is RCU protected we know it cannot disappear as we test it, the same guarantee that made it safe to acquire the reference previously. The seqlock tests whether the fence was replaced as we are testing it telling us whether or not we can trust the result (if not, we just repeat the test until stable). Signed-off-by: Chris Wilson <chris(a)chris-wilson.co.uk> Cc: Sumit Semwal <sumit.semwal(a)linaro.org> Cc: linux-media(a)vger.kernel.org Cc: dri-devel(a)lists.freedesktop.org Cc: linaro-mm-sig(a)lists.linaro.org --- drivers/dma-buf/reservation.c | 32 ++++---------------------------- 1 file changed, 4 insertions(+), 28 deletions(-) diff --git a/drivers/dma-buf/reservation.c b/drivers/dma-buf/reservation.c index e74493e7332b..1ddffa5adb5a 100644 --- a/drivers/dma-buf/reservation.c +++ b/drivers/dma-buf/reservation.c @@ -442,24 +442,6 @@ unlock_retry: } EXPORT_SYMBOL_GPL(reservation_object_wait_timeout_rcu); - -static inline int -reservation_object_test_signaled_single(struct fence *passed_fence) -{ - struct fence *fence, *lfence = passed_fence; - int ret = 1; - - if (!test_bit(FENCE_FLAG_SIGNALED_BIT, &lfence->flags)) { - fence = fence_get_rcu(lfence); - if (!fence) - return -1; - - ret = !!fence_is_signaled(fence); - fence_put(fence); - } - return ret; -} - /** * reservation_object_test_signaled_rcu - Test if a reservation object's * fences have been signaled. @@ -474,7 +456,7 @@ bool reservation_object_test_signaled_rcu(struct reservation_object *obj, bool test_all) { unsigned seq, shared_count; - int ret; + bool ret; rcu_read_lock(); retry: @@ -494,10 +476,8 @@ retry: for (i = 0; i < shared_count; ++i) { struct fence *fence = rcu_dereference(fobj->shared[i]); - ret = reservation_object_test_signaled_single(fence); - if (ret < 0) - goto retry; - else if (!ret) + ret = fence_is_signaled(fence); + if (!ret) break; } @@ -509,11 +489,7 @@ retry: struct fence *fence_excl = rcu_dereference(obj->fence_excl); if (fence_excl) { - ret = reservation_object_test_signaled_single( - fence_excl); - if (ret < 0) - goto retry; - + ret = fence_is_signaled(fence_excl); if (read_seqcount_retry(&obj->seq, seq)) goto retry; } -- 2.9.3

8 years, 3 months

[PATCH] dma-buf/sw_sync: mark sync_timeline_create() static

by Baoyou Xie

We get 1 warning when building kernel with W=1: drivers/dma-buf/sw_sync.c:87:23: warning: no previous prototype for 'sync_timeline_create' [-Wmissing-prototypes] In fact, this function is only used in the file in which it is declared and don't need a declaration, but can be made static. So this patch marks it 'static'. Signed-off-by: Baoyou Xie <baoyou.xie(a)linaro.org> --- drivers/dma-buf/sw_sync.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/dma-buf/sw_sync.c b/drivers/dma-buf/sw_sync.c index 62e8e6d..6f16c85 100644 --- a/drivers/dma-buf/sw_sync.c +++ b/drivers/dma-buf/sw_sync.c @@ -84,7 +84,7 @@ static inline struct sync_pt *fence_to_sync_pt(struct fence *fence) * Creates a new sync_timeline. Returns the sync_timeline object or NULL in * case of error. */ -struct sync_timeline *sync_timeline_create(const char *name) +static struct sync_timeline *sync_timeline_create(const char *name) { struct sync_timeline *obj; -- 2.7.4

8 years, 3 months

[PATCH] dma-buf/sw_sync: mark sync_timeline_create() static

by Baoyou Xie

8 years, 3 months

[PATCH 11/11] dma-buf: Do a fast lockless check for poll with timeout=0

by Chris Wilson

Currently we install a callback for performing poll on a dma-buf, irrespective of the timeout. This involves taking a spinlock, as well as unnecessary work, and greatly reduces scaling of poll(.timeout=0) across multiple threads. We can query whether the poll will block prior to installing the callback to make the busy-query fast. Single thread: 60% faster 8 threads on 4 (+4 HT) cores: 600% faster Still not quite the perfect scaling we get with a native busy ioctl, but poll(dmabuf) is faster due to the quicker lookup of the object and avoiding drm_ioctl(). Signed-off-by: Chris Wilson <chris(a)chris-wilson.co.uk> Cc: Sumit Semwal <sumit.semwal(a)linaro.org> Cc: linux-media(a)vger.kernel.org Cc: dri-devel(a)lists.freedesktop.org Cc: linaro-mm-sig(a)lists.linaro.org Reviewed-by: Daniel Vetter <daniel.vetter(a)ffwll.ch> --- drivers/dma-buf/dma-buf.c | 12 ++++++++++++ 1 file changed, 12 insertions(+) diff --git a/drivers/dma-buf/dma-buf.c b/drivers/dma-buf/dma-buf.c index cf04d249a6a4..c7a7bc579941 100644 --- a/drivers/dma-buf/dma-buf.c +++ b/drivers/dma-buf/dma-buf.c @@ -156,6 +156,18 @@ static unsigned int dma_buf_poll(struct file *file, poll_table *poll) if (!events) return 0; + if (poll_does_not_wait(poll)) { + if (events & POLLOUT && + !reservation_object_test_signaled_rcu(resv, true)) + events &= ~(POLLOUT | POLLIN); + + if (events & POLLIN && + !reservation_object_test_signaled_rcu(resv, false)) + events &= ~POLLIN; + + return events; + } + retry: seq = read_seqcount_begin(&resv->seq); rcu_read_lock(); -- 2.9.3

8 years, 3 months

[PATCH 09/11] dma-buf: Restart reservation_object_test_signaled_rcu() after writes

by Chris Wilson

In order to be completely generic, we have to double check the read seqlock after acquiring a reference to the fence. If the driver is allocating fences from a SLAB_DESTROY_BY_RCU, or similar freelist, then within an RCU grace period a fence may be freed and reallocated. The RCU read side critical section does not prevent this reallocation, instead we have to inspect the reservation's seqlock to double check if the fences have been reassigned as we were acquiring our reference. Signed-off-by: Chris Wilson <chris(a)chris-wilson.co.uk> Cc: Daniel Vetter <daniel.vetter(a)ffwll.ch> Cc: Maarten Lankhorst <maarten.lankhorst(a)linux.intel.com> Cc: Christian König <christian.koenig(a)amd.com> Cc: Alex Deucher <alexander.deucher(a)amd.com> Cc: Sumit Semwal <sumit.semwal(a)linaro.org> Cc: linux-media(a)vger.kernel.org Cc: dri-devel(a)lists.freedesktop.org Cc: linaro-mm-sig(a)lists.linaro.org --- drivers/dma-buf/reservation.c | 30 ++++++++++-------------------- 1 file changed, 10 insertions(+), 20 deletions(-) diff --git a/drivers/dma-buf/reservation.c b/drivers/dma-buf/reservation.c index 3369e4668e96..e74493e7332b 100644 --- a/drivers/dma-buf/reservation.c +++ b/drivers/dma-buf/reservation.c @@ -474,12 +474,13 @@ bool reservation_object_test_signaled_rcu(struct reservation_object *obj, bool test_all) { unsigned seq, shared_count; - int ret = true; + int ret; + rcu_read_lock(); retry: + ret = true; shared_count = 0; seq = read_seqcount_begin(&obj->seq); - rcu_read_lock(); if (test_all) { unsigned i; @@ -490,46 +491,35 @@ retry: if (fobj) shared_count = fobj->shared_count; - if (read_seqcount_retry(&obj->seq, seq)) - goto unlock_retry; - for (i = 0; i < shared_count; ++i) { struct fence *fence = rcu_dereference(fobj->shared[i]); ret = reservation_object_test_signaled_single(fence); if (ret < 0) - goto unlock_retry; + goto retry; else if (!ret) break; } - /* - * There could be a read_seqcount_retry here, but nothing cares - * about whether it's the old or newer fence pointers that are - * signaled. That race could still have happened after checking - * read_seqcount_retry. If you care, use ww_mutex_lock. - */ + if (read_seqcount_retry(&obj->seq, seq)) + goto retry; } if (!shared_count) { struct fence *fence_excl = rcu_dereference(obj->fence_excl); - if (read_seqcount_retry(&obj->seq, seq)) - goto unlock_retry; - if (fence_excl) { ret = reservation_object_test_signaled_single( fence_excl); if (ret < 0) - goto unlock_retry; + goto retry; + + if (read_seqcount_retry(&obj->seq, seq)) + goto retry; } } rcu_read_unlock(); return ret; - -unlock_retry: - rcu_read_unlock(); - goto retry; } EXPORT_SYMBOL_GPL(reservation_object_test_signaled_rcu); -- 2.9.3

8 years, 3 months

[PATCH 06/11] dma-buf: Introduce fence_get_rcu_safe()

by Chris Wilson

This variant of fence_get_rcu() takes an RCU protected pointer to a fence and carefully returns a reference to the fence ensuring that it is not reallocated as it does. This is required when mixing fences and SLAB_DESTROY_BY_RCU - although it serves a more pedagogical function atm Signed-off-by: Chris Wilson <chris(a)chris-wilson.co.uk> Cc: Daniel Vetter <daniel.vetter(a)ffwll.ch> Cc: Sumit Semwal <sumit.semwal(a)linaro.org> Cc: linux-media(a)vger.kernel.org Cc: dri-devel(a)lists.freedesktop.org Cc: linaro-mm-sig(a)lists.linaro.org --- include/linux/fence.h | 56 ++++++++++++++++++++++++++++++++++++++++++++++----- 1 file changed, 51 insertions(+), 5 deletions(-) diff --git a/include/linux/fence.h b/include/linux/fence.h index 0d763053f97a..c9c5ba98c302 100644 --- a/include/linux/fence.h +++ b/include/linux/fence.h @@ -183,6 +183,16 @@ void fence_release(struct kref *kref); void fence_free(struct fence *fence); /** + * fence_put - decreases refcount of the fence + * @fence: [in] fence to reduce refcount of + */ +static inline void fence_put(struct fence *fence) +{ + if (fence) + kref_put(&fence->refcount, fence_release); +} + +/** * fence_get - increases refcount of the fence * @fence: [in] fence to increase refcount of * @@ -210,13 +220,49 @@ static inline struct fence *fence_get_rcu(struct fence *fence) } /** - * fence_put - decreases refcount of the fence - * @fence: [in] fence to reduce refcount of + * fence_get_rcu_safe - acquire a reference to an RCU tracked fence + * @fence: [in] pointer to fence to increase refcount of + * + * Function returns NULL if no refcount could be obtained, or the fence. + * This function handles acquiring a reference to a fence that may be + * reallocated within the RCU grace period (such as with SLAB_DESTROY_BY_RCU), + * so long as the caller is using RCU on the pointer to the fence. + * + * An alternative mechanism is to employ a seqlock to protect a bunch of + * fences, such as used by struct reservation_object. When using a seqlock, + * the seqlock must be taken before and checked after a reference to the + * fence is acquired (as shown here). + * + * The caller is required to hold the RCU read lock. */ -static inline void fence_put(struct fence *fence) +static inline struct fence *fence_get_rcu_safe(struct fence * __rcu *fencep) { - if (fence) - kref_put(&fence->refcount, fence_release); + do { + struct fence *fence; + + fence = rcu_dereference(*fencep); + if (!fence || !fence_get_rcu(fence)) + return NULL; + + /* The atomic_inc_not_zero() inside fence_get_rcu() + * provides a full memory barrier upon success (such as now). + * This is paired with the write barrier from assigning + * to the __rcu protected fence pointer so that if that + * pointer still matches the current fence, we know we + * have successfully acquire a reference to it. If it no + * longer matches, we are holding a reference to some other + * reallocated pointer. This is possible if the allocator + * is using a freelist like SLAB_DESTROY_BY_RCU where the + * fence remains valid for the RCU grace period, but it + * may be reallocated. When using such allocators, we are + * responsible for ensuring the reference we get is to + * the right fence, as below. + */ + if (fence == rcu_access_pointer(*fencep)) + return rcu_pointer_handoff(fence); + + fence_put(fence); + } while (1); } int fence_signal(struct fence *fence); -- 2.9.3

8 years, 3 months

[PATCH 08/11] dma-buf: Restart reservation_object_wait_timeout_rcu() after writes

by Chris Wilson

In order to be completely generic, we have to double check the read seqlock after acquiring a reference to the fence. If the driver is allocating fences from a SLAB_DESTROY_BY_RCU, or similar freelist, then within an RCU grace period a fence may be freed and reallocated. The RCU read side critical section does not prevent this reallocation, instead we have to inspect the reservation's seqlock to double check if the fences have been reassigned as we were acquiring our reference. Signed-off-by: Chris Wilson <chris(a)chris-wilson.co.uk> Cc: Daniel Vetter <daniel.vetter(a)ffwll.ch> Cc: Maarten Lankhorst <maarten.lankhorst(a)linux.intel.com> Cc: Christian König <christian.koenig(a)amd.com> Cc: Alex Deucher <alexander.deucher(a)amd.com> Cc: Sumit Semwal <sumit.semwal(a)linaro.org> Cc: linux-media(a)vger.kernel.org Cc: dri-devel(a)lists.freedesktop.org Cc: linaro-mm-sig(a)lists.linaro.org --- drivers/dma-buf/reservation.c | 11 +++++------ 1 file changed, 5 insertions(+), 6 deletions(-) diff --git a/drivers/dma-buf/reservation.c b/drivers/dma-buf/reservation.c index 10fd441dd4ed..3369e4668e96 100644 --- a/drivers/dma-buf/reservation.c +++ b/drivers/dma-buf/reservation.c @@ -388,9 +388,6 @@ retry: if (fobj) shared_count = fobj->shared_count; - if (read_seqcount_retry(&obj->seq, seq)) - goto unlock_retry; - for (i = 0; i < shared_count; ++i) { struct fence *lfence = rcu_dereference(fobj->shared[i]); @@ -413,9 +410,6 @@ retry: if (!shared_count) { struct fence *fence_excl = rcu_dereference(obj->fence_excl); - if (read_seqcount_retry(&obj->seq, seq)) - goto unlock_retry; - if (fence_excl && !test_bit(FENCE_FLAG_SIGNALED_BIT, &fence_excl->flags)) { if (!fence_get_rcu(fence_excl)) @@ -430,6 +424,11 @@ retry: rcu_read_unlock(); if (fence) { + if (read_seqcount_retry(&obj->seq, seq)) { + fence_put(fence); + goto retry; + } + ret = fence_wait_timeout(fence, intr, ret); fence_put(fence); if (ret > 0 && wait_all && (i + 1 < shared_count)) -- 2.9.3

8 years, 3 months

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

Linaro-mm-sig September 2016