Linaro-mm-sig December 2016

linaro-mm-sig@lists.linaro.org

3 participants
3 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

[RFC PATCH 0/4] Ion caching (yet again) proof of concept

by Laura Abbott

Hi, I've been (once again) looking at alternate caching models for Ion. Part of this work is also to make Ion fit better in to the dma_buf model. Ion is a bit unusual for dma_buf. Most drivers that support dma_buf have two parts: exporting buffers that a driver allocates and importing buffers allocated elsewhere for use by the driver. Ion is basically designed to export only and not import buffers from other drivers (the need for import is also on my TODO list) Even more unusual, there is no actual 'driver' to map into. Ion currently does nothing except pass back the same sg_table each time without calling dma_map. The description of the .map_dma_buf function in dma_buf_ops * @map_dma_buf: returns list of scatter pages allocated, increases usecount * of the buffer. Requires atleast one attach to be called * before. Returned sg list should already be mapped into * _device_ address space. This call may sleep. May also return * -EINTR. Should return -EINVAL if attach hasn't been called yet. So Ion is definitely not doing this correctly. This ties back into correcting the caching model. If we call dma_map_sg/dma_unmap_sg with begin_cpu_access, this should be enough to allow the caches to always be properly synchronized and means we can drop the various dma_sync calls floating around. This is going to violate one of the big fat comments in ion_buffer_create /* * this will set up dma addresses for the sglist -- it is not * technically correct as per the dma api -- a specific * device isn't really taking ownership here. However, in practice on * our systems the only dma_address space is physical addresses. * Additionally, we can't afford the overhead of invalidating every * allocation via dma_map_sg. The implicit contract here is that * memory coming from the heaps is ready for dma, ie if it has a * cached mapping that mapping has been invalidated */ for_each_sg(buffer->sg_table->sgl, sg, buffer->sg_table->nents, i) { sg_dma_address(sg) = sg_phys(sg); sg_dma_len(sg) = sg->length; } The overhead of invalidating is a valid concern. I'm hoping that the architecture has either evolved such that this won't be a problem or we can figure out some clever use of DMA_ATTR_SKIP_CPU_SYNC. As part of this, I'm considering dropping the fault synchronization. If we have explicit use begin_cpu_access and use of the dma_buf sync ioctls, I don't think it should be necessary. I have a 'pre-RFC' tree at https://pagure.io/kernel-ion/branch/ion_cache_proof_dec14 Yes, the patches are not bisectable and there is more to be done. These have been compile tested only and haven't been hooked up to anything to actually run (another actually big TODO). I'm mostly looking for feedback if this looks like the right direction and if there are going to be major problems with this approach. I don't actually anticipate this getting merged into drivers/staging/android/ion but this is the easiest way to continue discussion. Thanks, Laura Laura Abbott (4): staging: android: ion: Some cleanup staging: android: ion: Duplicate sg_table staging: android: ion: Remove page faulting support staging: android: ion: Call dma_map_sg for syncing and mapping drivers/staging/android/ion/ion-ioctl.c | 6 - drivers/staging/android/ion/ion.c | 251 ++++++++---------------- drivers/staging/android/ion/ion.h | 5 +- drivers/staging/android/ion/ion_carveout_heap.c | 16 +- drivers/staging/android/ion/ion_chunk_heap.c | 15 +- drivers/staging/android/ion/ion_cma_heap.c | 5 +- drivers/staging/android/ion/ion_page_pool.c | 3 - drivers/staging/android/ion/ion_priv.h | 4 +- drivers/staging/android/ion/ion_system_heap.c | 14 +- 9 files changed, 90 insertions(+), 229 deletions(-) -- 2.7.4

8 years

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

Linaro-mm-sig December 2016