Linaro-mm-sig February 2017

linaro-mm-sig@lists.linaro.org

6 participants
6 discussions

Re: [Linaro-mm-sig] [RFCv3][PATCH 3/5] arm64: Implement ARCH_HAS_FORCE_CACHE

by Laura Abbott

On 02/22/2017 05:01 PM, Chen Feng wrote: > > > On 2017/2/22 3:29, Laura Abbott wrote: >> On 02/20/2017 10:05 PM, Chen Feng wrote: >>> Hi Laura, >>> >>> When we enable kernel v4.4 or newer version on our platform, we meet the issue >>> of flushing cache without reference device. It seems that this patch set is >>> a solution. I'm curious the progress of the discussion. Do you have any plan >>> to fix it in v4.4 and newer kernel verison? >>> >> >> No, I've abandoned this approach based on feedback. The APIs had too much >> potential for incorrect usage. I'm ripping out the implicit caching in Ion >> and switching it to a model where there should always be a device available. >> >> What's your use case where you don't have a device structure? >> > Userspace use ioctl to flush cache for device. > > ion_sync_for_device > dma_sync_sg_for_device(NULL, buffer->sg_table->sgl, > buffer->sg_table->nents, DMA_BIDIRECTIONAL); > > And sys-heap when allocate a zero buffer flush zero data to ddr. > alloc_buffer_page > ion_pages_sync_for_device(NULL, page, PAGE_SIZE << order, > DMA_BIDIRECTIONAL); > > Yes, those calls are being removed. This is what I proposed back in December https://marc.info/?l=linux-kernel&m=148176054902921&w=2 I never heard any feedback on it so I assume everyone was okay with the general direction. I plan on pushing a revised and expanded version of that series once the merge window closes. Thanks, Laura > >> Thanks, >> Laura >> >>> On 2016/9/14 2:41, Laura Abbott wrote: >>>> On 09/13/2016 08:14 AM, Will Deacon wrote: >>>>> On Tue, Sep 13, 2016 at 08:02:20AM -0700, Laura Abbott wrote: >>>>>> On 09/13/2016 02:19 AM, Will Deacon wrote: >>>>>>> On Mon, Sep 12, 2016 at 02:32:56PM -0700, Laura Abbott wrote: >>>>>>>> >>>>>>>> arm64 may need to guarantee the caches are synced. Implement versions of >>>>>>>> the kernel_force_cache API to allow this. >>>>>>>> >>>>>>>> Signed-off-by: Laura Abbott <labbott(a)redhat.com> >>>>>>>> --- >>>>>>>> v3: Switch to calling cache operations directly instead of relying on >>>>>>>> DMA mapping. >>>>>>>> --- >>>>>>>> arch/arm64/include/asm/cacheflush.h | 8 ++++++++ >>>>>>>> arch/arm64/mm/cache.S | 24 ++++++++++++++++++++---- >>>>>>>> arch/arm64/mm/flush.c | 11 +++++++++++ >>>>>>>> 3 files changed, 39 insertions(+), 4 deletions(-) >>>>>>> >>>>>>> I'm really hesitant to expose these cache routines as an API solely to >>>>>>> support a driver sitting in staging/. I appreciate that there's a chicken >>>>>>> and egg problem here, but we *really* don't want people using these routines >>>>>>> in preference to the DMA API, and I fear that we'll simply grow a bunch >>>>>>> more users of these things if we promote it as an API like you're proposing. >>>>>>> >>>>>>> Can the code not be contained under staging/, as part of ion? >>>>>>> >>>>>> >>>>>> I proposed that in V1 and it was suggested I make it a proper API >>>>>> >>>>>> http://www.mail-archive.com/driverdev-devel@linuxdriverproject.org/msg47654… >>>>>> http://www.mail-archive.com/driverdev-devel@linuxdriverproject.org/msg47672… >>>>> >>>>> :/ then I guess we're in disagreement. If ion really needs this stuff >>>>> (which I don't fully grok), perhaps we should be exposing something at >>>>> a higher level from the architecture, so it really can't be used for >>>>> anything other than ion. >>>> >>>> I talked/complained about this at a past plumbers. The gist is that Ion >>>> ends up acting as a fake DMA layer for clients. It doesn't match nicely >>>> because clients can allocate both coherent and non-coherent memory. >>>> Trying to use dma_map doesn't work because a) a device for coherency isn't >>>> known at allocation time b) it kills performance. Part of the motivation >>>> for taking this approach is to avoid the need to rework the existing >>>> Android userspace and keep the existing behavior, as terrible as it >>>> is. Having Ion out of staging and not actually usable isn't helpful. >>>> >>>> I'll give this all some more thought and hopefully have one or two more >>>> proposals before Connect/Plumbers. >>>> >>>>> >>>>> Will >>>>> >>>> >>>> Thanks, >>>> Laura >>>> _______________________________________________ >>>> Linaro-mm-sig mailing list >>>> Linaro-mm-sig(a)lists.linaro.org >>>> https://lists.linaro.org/mailman/listinfo/linaro-mm-sig >>> >>> >>> _______________________________________________ >>> linux-arm-kernel mailing list >>> linux-arm-kernel(a)lists.infradead.org >>> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel >>> >> >> >> . >> > > > _______________________________________________ > linux-arm-kernel mailing list > linux-arm-kernel(a)lists.infradead.org > http://lists.infradead.org/mailman/listinfo/linux-arm-kernel >

7 years, 2 months

Re: [Linaro-mm-sig] [PATCH] dma-buf: add support for compat ioctl

by Daniel Vetter

On Tue, Feb 21, 2017 at 4:08 PM, Christian König <deathsimple(a)vodafone.de> wrote: > Am 21.02.2017 um 15:55 schrieb Marek Szyprowski: >> >> Dear All, >> >> On 2017-02-21 15:37, Marek Szyprowski wrote: >>> >>> Hi Christian, >>> >>> On 2017-02-21 14:59, Christian König wrote: >>>> >>>> Am 21.02.2017 um 14:21 schrieb Marek Szyprowski: >>>>> >>>>> Add compat ioctl support to dma-buf. This lets one to use >>>>> DMA_BUF_IOCTL_SYNC >>>>> ioctl from 32bit application on 64bit kernel. Data structures for both >>>>> 32 >>>>> and 64bit modes are same, so there is no need for additional >>>>> translation >>>>> layer. >>>> >>>> >>>> Well I might be wrong, but IIRC compat_ioctl was just optional and if >>>> not specified unlocked_ioctl was called instead. >>>> >>>> If that is true your patch wouldn't have any effect at all. >>> >>> >>> Well, then why I got -ENOTTY in the 32bit test app for this ioctl on >>> 64bit ARM64 kernel without this patch? >>> >> >> I've checked in fs/compat_ioctl.c, I see no fallback in >> COMPAT_SYSCALL_DEFINE3, >> so one has to provide compat_ioctl callback to have ioctl working with >> 32bit >> apps. > > > Then my memory cheated on me. > > In this case the patch is Reviewed-by: Christian König > <christian.koenig(a)amd.com>. Since you have commit rights for drm-misc, care to push this to drm-misc-next-fixes pls? Also I think this warrants a cc: stable, clearly an obvious screw-up in creating this api on our side :( So feel free to smash my ack on the patch. Thanks, Daniel -- Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch

7 years, 2 months

Re: [Linaro-mm-sig] [RFCv3][PATCH 3/5] arm64: Implement ARCH_HAS_FORCE_CACHE

by Laura Abbott

On 02/20/2017 10:05 PM, Chen Feng wrote: > Hi Laura, > > When we enable kernel v4.4 or newer version on our platform, we meet the issue > of flushing cache without reference device. It seems that this patch set is > a solution. I'm curious the progress of the discussion. Do you have any plan > to fix it in v4.4 and newer kernel verison? > No, I've abandoned this approach based on feedback. The APIs had too much potential for incorrect usage. I'm ripping out the implicit caching in Ion and switching it to a model where there should always be a device available. What's your use case where you don't have a device structure? Thanks, Laura > On 2016/9/14 2:41, Laura Abbott wrote: >> On 09/13/2016 08:14 AM, Will Deacon wrote: >>> On Tue, Sep 13, 2016 at 08:02:20AM -0700, Laura Abbott wrote: >>>> On 09/13/2016 02:19 AM, Will Deacon wrote: >>>>> On Mon, Sep 12, 2016 at 02:32:56PM -0700, Laura Abbott wrote: >>>>>> >>>>>> arm64 may need to guarantee the caches are synced. Implement versions of >>>>>> the kernel_force_cache API to allow this. >>>>>> >>>>>> Signed-off-by: Laura Abbott <labbott(a)redhat.com> >>>>>> --- >>>>>> v3: Switch to calling cache operations directly instead of relying on >>>>>> DMA mapping. >>>>>> --- >>>>>> arch/arm64/include/asm/cacheflush.h | 8 ++++++++ >>>>>> arch/arm64/mm/cache.S | 24 ++++++++++++++++++++---- >>>>>> arch/arm64/mm/flush.c | 11 +++++++++++ >>>>>> 3 files changed, 39 insertions(+), 4 deletions(-) >>>>> >>>>> I'm really hesitant to expose these cache routines as an API solely to >>>>> support a driver sitting in staging/. I appreciate that there's a chicken >>>>> and egg problem here, but we *really* don't want people using these routines >>>>> in preference to the DMA API, and I fear that we'll simply grow a bunch >>>>> more users of these things if we promote it as an API like you're proposing. >>>>> >>>>> Can the code not be contained under staging/, as part of ion? >>>>> >>>> >>>> I proposed that in V1 and it was suggested I make it a proper API >>>> >>>> http://www.mail-archive.com/driverdev-devel@linuxdriverproject.org/msg47654… >>>> http://www.mail-archive.com/driverdev-devel@linuxdriverproject.org/msg47672… >>> >>> :/ then I guess we're in disagreement. If ion really needs this stuff >>> (which I don't fully grok), perhaps we should be exposing something at >>> a higher level from the architecture, so it really can't be used for >>> anything other than ion. >> >> I talked/complained about this at a past plumbers. The gist is that Ion >> ends up acting as a fake DMA layer for clients. It doesn't match nicely >> because clients can allocate both coherent and non-coherent memory. >> Trying to use dma_map doesn't work because a) a device for coherency isn't >> known at allocation time b) it kills performance. Part of the motivation >> for taking this approach is to avoid the need to rework the existing >> Android userspace and keep the existing behavior, as terrible as it >> is. Having Ion out of staging and not actually usable isn't helpful. >> >> I'll give this all some more thought and hopefully have one or two more >> proposals before Connect/Plumbers. >> >>> >>> Will >>> >> >> Thanks, >> Laura >> _______________________________________________ >> Linaro-mm-sig mailing list >> Linaro-mm-sig(a)lists.linaro.org >> https://lists.linaro.org/mailman/listinfo/linaro-mm-sig > > > _______________________________________________ > linux-arm-kernel mailing list > linux-arm-kernel(a)lists.infradead.org > http://lists.infradead.org/mailman/listinfo/linux-arm-kernel >

7 years, 2 months

Re: [Linaro-mm-sig] [PATCH] dma-buf: add support for compat ioctl

by Marek Szyprowski

Hi Christian, On 2017-02-21 14:59, Christian König wrote: > Am 21.02.2017 um 14:21 schrieb Marek Szyprowski: >> Add compat ioctl support to dma-buf. This lets one to use >> DMA_BUF_IOCTL_SYNC >> ioctl from 32bit application on 64bit kernel. Data structures for >> both 32 >> and 64bit modes are same, so there is no need for additional translation >> layer. > > Well I might be wrong, but IIRC compat_ioctl was just optional and if > not specified unlocked_ioctl was called instead. > > If that is true your patch wouldn't have any effect at all. Well, then why I got -ENOTTY in the 32bit test app for this ioctl on 64bit ARM64 kernel without this patch? Best regards -- Marek Szyprowski, PhD Samsung R&D Institute Poland

7 years, 2 months

[PATCH] dma-buf: add support for compat ioctl

by Marek Szyprowski

Add compat ioctl support to dma-buf. This lets one to use DMA_BUF_IOCTL_SYNC ioctl from 32bit application on 64bit kernel. Data structures for both 32 and 64bit modes are same, so there is no need for additional translation layer. Signed-off-by: Marek Szyprowski <m.szyprowski(a)samsung.com> --- drivers/dma-buf/dma-buf.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/drivers/dma-buf/dma-buf.c b/drivers/dma-buf/dma-buf.c index 718f832a5c71..0007b792827b 100644 --- a/drivers/dma-buf/dma-buf.c +++ b/drivers/dma-buf/dma-buf.c @@ -325,6 +325,9 @@ static long dma_buf_ioctl(struct file *file, .llseek = dma_buf_llseek, .poll = dma_buf_poll, .unlocked_ioctl = dma_buf_ioctl, +#ifdef CONFIG_COMPAT + .compat_ioctl = dma_buf_ioctl, +#endif }; /* -- 1.9.1

7 years, 2 months

[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, 2 months

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

Linaro-mm-sig February 2017