Re: [RFC]: shmem fd for non-DMA buffer sharing cross drivers
Hi, On Fri, 25 Aug 2023 at 08:56, Hsia-Jun Li wrote: > On 8/25/23 15:40, Pekka Paalanen wrote: > > if userspace cannot access things like an image's HDR metadata, then it > > will be impossible for userspace to program KMS to have the correct > > color pipeline, or to send intended HDR metadata to a video sink. > > > > You cannot leave userspace out of HDR metadata handling, because quite > > probably the V4L2 buffer is not the only thing on screen. That means > > there must composition of multiple sources with different image > > properties and metadata, which means it is no longer obvious what HDR > > metadata should be sent to the video sink. > > > > Even if it is a TV-like application rather than a windowed desktop, you > > will still have other contents to composite: OSD (volume indicators, > > channels indicators, program guide, ...), sub-titles, channel logos, > > notifications... These components ideally should not change their > > appearance arbitrarily with the main program content and metadata > > changes. Either the metadata sent to the video sink is kept static and > > the main program adapted on the fly, or main program metadata is sent > > to the video sink and the additional content is adapted on the fly. > > > > There is only one set of HDR metadata and one composited image that can > > be sent to a video sink, so both must be chosen and produced correctly > > at the source side. This cannot be done automatically inside KMS kernel > > drivers. > > There may be some misunderstanding. > Let suppose this HDR data is in a vendor specific format. > Both upstream(decoder) and downstream(DRM) hardware devices are coming > from the same vendor. > Then we just need to delivery the reference to this metadata buffer from > the upstream to downstream, both of drivers know how to handle it. > > Despite the userspace, we just need to extend a wayland protocol that > making wayland compositor know how to receive the reference to the > metadata and set it to the DRM plane. > > If you want a common HDR formats for all HDR variants(HDR10+, DV), I am > not against it. But it won't make the userspace be able to fill the HDR > metadata even the HDR data comes from the bitstream(likes SEI). We must > consider the case of Secure Video Path(Digital Right), the bitstream is > not accessible from (REE) userspace nor linux kernel, the downstream > must take what the upstream feed. To summarise from IRC, so it's properly documented: the community will not accept this. The mechanism (shmem, dmabuf, copy_from_user, whatever) is _not_ the problem. The problem is the concept. There have been incredibly extensive discussions on this list about colour management and HDR, summarised in documentation in the DRM repository, as well as three talks at the last XDC. This design is the result of discussion between many community participants - including hardware vendors - who have all come up with a design which prioritises transparency and explicit operation. What you are suggesting is exactly the opposite of this. A design in which opaque magic blobs are passed around and the kernel does unknown things based on the contents of those blobs, contradicts this design. (This is different to compression, where even if the format is proprietary, the effect is well-understood - for the current compression mechanisms, it is a lossless transform.) The old Android Display Framework (ADF) was based around the same design with blobs of opaque driver-specific data, where generic code - either in the kernel or in userspace - could not understand the effect of these blobs. This design was rejected, and we made a clear choice to follow the DRM design principles instead. Upstream will not accept any design which hides magic away. 'GKI is hard', 'other vendors won't let us', etc, are not good enough reasons to change our mind on this fundamental principle. Cheers, Daniel
Re: [RFC]: shmem fd for non-DMA buffer sharing cross drivers
On Fri, 25 Aug 2023 15:56:18 +0800 Hsia-Jun Li wrote: > On 8/25/23 15:40, Pekka Paalanen wrote: > > Subject: > > Re: [RFC]: shmem fd for non-DMA buffer sharing cross drivers > > From: > > Pekka Paalanen > > Date: > > 8/25/23, 15:40 > > > > To: > > Hsia-Jun Li > > CC: > > Tomasz Figa , linux...@kvack.org, > > dri-devel@lists.freedesktop.org, Linux Media Mailing List > > , hu...@google.com, > > a...@linux-foundation.org, Simon Ser , Hans Verkuil > > , dani...@collabora.com, ayaka > > , linux-ker...@vger.kernel.org, Nicolas Dufresne > > > > > > > > On Wed, 23 Aug 2023 15:11:23 +0800 > > Hsia-Jun Li wrote: > > > >> On 8/23/23 12:46, Tomasz Figa wrote: > >>> CAUTION: Email originated externally, do not click links or open > >>> attachments unless you recognize the sender and know the content is safe. > >>> > >>> > >>> Hi Hsia-Jun, > >>> > >>> On Tue, Aug 22, 2023 at 8:14 PM Hsia-Jun Li > >>> wrote: > >>>> Hello > >>>> > >>>> I would like to introduce a usage of SHMEM slimier to DMA-buf, the major > >>>> purpose of that is sharing metadata or just a pure container for cross > >>>> drivers. > >>>> > >>>> We need to exchange some sort of metadata between drivers, likes dynamic > >>>> HDR data between video4linux2 and DRM. > >>> If the metadata isn't too big, would it be enough to just have the > >>> kernel copy_from_user() to a kernel buffer in the ioctl code? > >>> > >>>> Or the graphics frame buffer is > >>>> too complex to be described with plain plane's DMA-buf fd. > >>>> An issue between DRM and V4L2 is that DRM could only support 4 planes > >>>> while it is 8 for V4L2. It would be pretty hard for DRM to expend its > >>>> interface to support that 4 more planes which would lead to revision of > >>>> many standard likes Vulkan, EGL. > >>> Could you explain how a shmem buffer could be used to support frame > >>> buffers with more than 4 planes? > >>> If you are asking why we need this: > >> 1. metadata likes dynamic HDR tone data > >> 2. DRM also challenges with this problem, let me quote what sima said: > >> "another trick that we iirc used for afbc is that sometimes the planes > >> have a fixed layout > >> like nv12 > >> and so logically it's multiple planes, but you only need one plane slot > >> to describe the buffer > >> since I think afbc had the "we need more than 4 planes" issue too" > >> > >> Unfortunately, there are vendor pixel formats are not fixed layout. > >> > >> 3. Secure(REE, trusted video piepline) info. > >> > >> For how to assign such metadata data. > >> In case with a drm fb_id, it is simple, we just add a drm plane property > >> for it. The V4L2 interface is not flexible, we could only leave into > >> CAPTURE request_fd as a control. > >>>> Also, there is no reason to consume a device's memory for the content > >>>> that device can't read it, or wasting an entry of IOMMU for such data. > >>> That's right, but DMA-buf doesn't really imply any of those. DMA-buf > >>> is just a kernel object with some backing memory. It's up to the > >>> allocator to decide how the backing memory is allocated and up to the > >>> importer on whether it would be mapped into an IOMMU. > >>> > >> I just want to say it can't be allocated at the same place which was for > >> those DMA bufs(graphics or compressed bitstream). > >> This also could be answer for your first question, if we place this kind > >> of buffer in a plane for DMABUF(importing) in V4L2, V4L2 core would try > >> to prepare it, which could map it into IOMMU. > >> > >>>> Usually, such a metadata would be the value should be written to a > >>>> hardware's registers, a 4KiB page would be 1024 items of 32 bits > >>>> registers. > >>>> > >>>> Still, I have some problems with SHMEM: > >>>> 1. I don't want the userspace modify the context of the SHMEM allocated > >>>> by the kernel, is there a way to do so? > >>> This is generally impossible without doing any of the two: > >>> 1) copying the contents to an internal buffer not accessible to the >
Re: [RFC]: shmem fd for non-DMA buffer sharing cross drivers
On 8/25/23 15:40, Pekka Paalanen wrote: Subject: Re: [RFC]: shmem fd for non-DMA buffer sharing cross drivers From: Pekka Paalanen Date: 8/25/23, 15:40 To: Hsia-Jun Li CC: Tomasz Figa , linux...@kvack.org, dri-devel@lists.freedesktop.org, Linux Media Mailing List , hu...@google.com, a...@linux-foundation.org, Simon Ser , Hans Verkuil , dani...@collabora.com, ayaka , linux-ker...@vger.kernel.org, Nicolas Dufresne On Wed, 23 Aug 2023 15:11:23 +0800 Hsia-Jun Li wrote: On 8/23/23 12:46, Tomasz Figa wrote: CAUTION: Email originated externally, do not click links or open attachments unless you recognize the sender and know the content is safe. Hi Hsia-Jun, On Tue, Aug 22, 2023 at 8:14 PM Hsia-Jun Li wrote: Hello I would like to introduce a usage of SHMEM slimier to DMA-buf, the major purpose of that is sharing metadata or just a pure container for cross drivers. We need to exchange some sort of metadata between drivers, likes dynamic HDR data between video4linux2 and DRM. If the metadata isn't too big, would it be enough to just have the kernel copy_from_user() to a kernel buffer in the ioctl code? Or the graphics frame buffer is too complex to be described with plain plane's DMA-buf fd. An issue between DRM and V4L2 is that DRM could only support 4 planes while it is 8 for V4L2. It would be pretty hard for DRM to expend its interface to support that 4 more planes which would lead to revision of many standard likes Vulkan, EGL. Could you explain how a shmem buffer could be used to support frame buffers with more than 4 planes? If you are asking why we need this: 1. metadata likes dynamic HDR tone data 2. DRM also challenges with this problem, let me quote what sima said: "another trick that we iirc used for afbc is that sometimes the planes have a fixed layout like nv12 and so logically it's multiple planes, but you only need one plane slot to describe the buffer since I think afbc had the "we need more than 4 planes" issue too" Unfortunately, there are vendor pixel formats are not fixed layout. 3. Secure(REE, trusted video piepline) info. For how to assign such metadata data. In case with a drm fb_id, it is simple, we just add a drm plane property for it. The V4L2 interface is not flexible, we could only leave into CAPTURE request_fd as a control. Also, there is no reason to consume a device's memory for the content that device can't read it, or wasting an entry of IOMMU for such data. That's right, but DMA-buf doesn't really imply any of those. DMA-buf is just a kernel object with some backing memory. It's up to the allocator to decide how the backing memory is allocated and up to the importer on whether it would be mapped into an IOMMU. I just want to say it can't be allocated at the same place which was for those DMA bufs(graphics or compressed bitstream). This also could be answer for your first question, if we place this kind of buffer in a plane for DMABUF(importing) in V4L2, V4L2 core would try to prepare it, which could map it into IOMMU. Usually, such a metadata would be the value should be written to a hardware's registers, a 4KiB page would be 1024 items of 32 bits registers. Still, I have some problems with SHMEM: 1. I don't want the userspace modify the context of the SHMEM allocated by the kernel, is there a way to do so? This is generally impossible without doing any of the two: 1) copying the contents to an internal buffer not accessible to the userspace, OR 2) modifying any of the buffer mappings to read-only 2) can actually be more costly than 1) (depending on the architecture, data size, etc.), so we shouldn't just discard the option of a simple copy_from_user() in the ioctl. I don't want the userspace access it at all. So that won't be a problem. Hi, if userspace cannot access things like an image's HDR metadata, then it will be impossible for userspace to program KMS to have the correct color pipeline, or to send intended HDR metadata to a video sink. You cannot leave userspace out of HDR metadata handling, because quite probably the V4L2 buffer is not the only thing on screen. That means there must composition of multiple sources with different image properties and metadata, which means it is no longer obvious what HDR metadata should be sent to the video sink. Even if it is a TV-like application rather than a windowed desktop, you will still have other contents to composite: OSD (volume indicators, channels indicators, program guide, ...), sub-titles, channel logos, notifications... These components ideally should not change their appearance arbitrarily with the main program content and metadata changes. Either the metadata sent to the video sink is kept static and the main program adapted on the fly, or main program metadata is sent to the video sink and the additional content is adapted on the fly. There is only one set of HDR metadata and one composited image that can be sent to a video sink, so
Re: [RFC]: shmem fd for non-DMA buffer sharing cross drivers
On Wed, 23 Aug 2023 15:11:23 +0800 Hsia-Jun Li wrote: > On 8/23/23 12:46, Tomasz Figa wrote: > > CAUTION: Email originated externally, do not click links or open > > attachments unless you recognize the sender and know the content is safe. > > > > > > Hi Hsia-Jun, > > > > On Tue, Aug 22, 2023 at 8:14 PM Hsia-Jun Li wrote: > > > >> > >> Hello > >> > >> I would like to introduce a usage of SHMEM slimier to DMA-buf, the major > >> purpose of that is sharing metadata or just a pure container for cross > >> drivers. > >> > >> We need to exchange some sort of metadata between drivers, likes dynamic > >> HDR data between video4linux2 and DRM. > > > > If the metadata isn't too big, would it be enough to just have the > > kernel copy_from_user() to a kernel buffer in the ioctl code? > > > >> Or the graphics frame buffer is > >> too complex to be described with plain plane's DMA-buf fd. > >> An issue between DRM and V4L2 is that DRM could only support 4 planes > >> while it is 8 for V4L2. It would be pretty hard for DRM to expend its > >> interface to support that 4 more planes which would lead to revision of > >> many standard likes Vulkan, EGL. > > > > Could you explain how a shmem buffer could be used to support frame > > buffers with more than 4 planes? > > If you are asking why we need this: > 1. metadata likes dynamic HDR tone data > 2. DRM also challenges with this problem, let me quote what sima said: > "another trick that we iirc used for afbc is that sometimes the planes > have a fixed layout > like nv12 > and so logically it's multiple planes, but you only need one plane slot > to describe the buffer > since I think afbc had the "we need more than 4 planes" issue too" > > Unfortunately, there are vendor pixel formats are not fixed layout. > > 3. Secure(REE, trusted video piepline) info. > > For how to assign such metadata data. > In case with a drm fb_id, it is simple, we just add a drm plane property > for it. The V4L2 interface is not flexible, we could only leave into > CAPTURE request_fd as a control. > >> > >> Also, there is no reason to consume a device's memory for the content > >> that device can't read it, or wasting an entry of IOMMU for such data. > > > > That's right, but DMA-buf doesn't really imply any of those. DMA-buf > > is just a kernel object with some backing memory. It's up to the > > allocator to decide how the backing memory is allocated and up to the > > importer on whether it would be mapped into an IOMMU. > > > I just want to say it can't be allocated at the same place which was for > those DMA bufs(graphics or compressed bitstream). > This also could be answer for your first question, if we place this kind > of buffer in a plane for DMABUF(importing) in V4L2, V4L2 core would try > to prepare it, which could map it into IOMMU. > > >> Usually, such a metadata would be the value should be written to a > >> hardware's registers, a 4KiB page would be 1024 items of 32 bits registers. > >> > >> Still, I have some problems with SHMEM: > >> 1. I don't want the userspace modify the context of the SHMEM allocated > >> by the kernel, is there a way to do so? > > > > This is generally impossible without doing any of the two: > > 1) copying the contents to an internal buffer not accessible to the > > userspace, OR > > 2) modifying any of the buffer mappings to read-only > > > > 2) can actually be more costly than 1) (depending on the architecture, > > data size, etc.), so we shouldn't just discard the option of a simple > > copy_from_user() in the ioctl. > > > I don't want the userspace access it at all. So that won't be a problem. Hi, if userspace cannot access things like an image's HDR metadata, then it will be impossible for userspace to program KMS to have the correct color pipeline, or to send intended HDR metadata to a video sink. You cannot leave userspace out of HDR metadata handling, because quite probably the V4L2 buffer is not the only thing on screen. That means there must composition of multiple sources with different image properties and metadata, which means it is no longer obvious what HDR metadata should be sent to the video sink. Even if it is a TV-like application rather than a windowed desktop, you will still have other contents to composite: OSD (volume indicators, channels indicators, program guide, ...), sub-titles, channel logos, notifications... These components ideally should not change their appearance arbitrarily with the main program content and metadata changes. Either the metadata sent to the video sink is kept static and the main program adapted on the fly, or main program metadata is sent to the video sink and the additional content is adapted on the fly. There is only one set of HDR metadata and one composited image that can be sent to a video sink, so both must be chosen and produced correctly at the source side. This cannot be done automatically inside KMS kernel drivers. Thanks, pq > >> 2. Should I
Re: [RFC]: shmem fd for non-DMA buffer sharing cross drivers
On 8/23/23 21:15, Tomasz Figa wrote: CAUTION: Email originated externally, do not click links or open attachments unless you recognize the sender and know the content is safe. On Wed, Aug 23, 2023 at 4:11 PM Hsia-Jun Li wrote: On 8/23/23 12:46, Tomasz Figa wrote: CAUTION: Email originated externally, do not click links or open attachments unless you recognize the sender and know the content is safe. Hi Hsia-Jun, On Tue, Aug 22, 2023 at 8:14 PM Hsia-Jun Li wrote: Hello I would like to introduce a usage of SHMEM slimier to DMA-buf, the major purpose of that is sharing metadata or just a pure container for cross drivers. We need to exchange some sort of metadata between drivers, likes dynamic HDR data between video4linux2 and DRM. If the metadata isn't too big, would it be enough to just have the kernel copy_from_user() to a kernel buffer in the ioctl code? Or the graphics frame buffer is too complex to be described with plain plane's DMA-buf fd. An issue between DRM and V4L2 is that DRM could only support 4 planes while it is 8 for V4L2. It would be pretty hard for DRM to expend its interface to support that 4 more planes which would lead to revision of many standard likes Vulkan, EGL. Could you explain how a shmem buffer could be used to support frame buffers with more than 4 planes? If you are asking why we need this: I'm asking how your proposal to use shmem FD solves the problem for those cases. The shmem fd is the reference to a metadata container(A C struct in kernel). Then drivers(V4L2 and DRM) could read this metadata when it processes the major buffer(SHMEM buf is the buffer assigned with a major buffer like the graphics buffer). 1. metadata likes dynamic HDR tone data 2. DRM also challenges with this problem, let me quote what sima said: "another trick that we iirc used for afbc is that sometimes the planes have a fixed layout like nv12 and so logically it's multiple planes, but you only need one plane slot to describe the buffer since I think afbc had the "we need more than 4 planes" issue too" Unfortunately, there are vendor pixel formats are not fixed layout. 3. Secure(REE, trusted video piepline) info. For how to assign such metadata data. In case with a drm fb_id, it is simple, we just add a drm plane property for it. The V4L2 interface is not flexible, we could only leave into CAPTURE request_fd as a control. Also, there is no reason to consume a device's memory for the content that device can't read it, or wasting an entry of IOMMU for such data. That's right, but DMA-buf doesn't really imply any of those. DMA-buf is just a kernel object with some backing memory. It's up to the allocator to decide how the backing memory is allocated and up to the importer on whether it would be mapped into an IOMMU. I just want to say it can't be allocated at the same place which was for those DMA bufs(graphics or compressed bitstream). This also could be answer for your first question, if we place this kind of buffer in a plane for DMABUF(importing) in V4L2, V4L2 core would try to prepare it, which could map it into IOMMU. V4L2 core will prepare it according to the struct device that is given to it. For the planes that don't have to go to the hardware a struct device could be given that doesn't require any DMA mapping. Also you can check how the uvcvideo driver handles it. It doesn't use the vb2 Because it uses vb2_vmalloc_memops? That vb2_vmalloc_attach_dmabuf() won't work anything. buffers directly, but always writes to them using CPU (due to how the Yes I noticed it would copy UBR buffer to vb2 buffer. UVC protocol is designed). I don't know what stops that, because we can't assume xHCI or EHCI have the IOMMU? I think that is not I want. If you were not talking about META_CAPTURE, which would be a ioslated dev node. For example, we have a NV15(2 planes) buffer with its HDR data. We need its NV15 planes be accessed by DMA directly or it would be a performance issue(so UVC memcpy is not acceptable), while its HDR data, we just read it from the devices' registers or somewhere, should be ship with the exactly buffer. Even we could expand the vb2_mem_ops interfaces, making it know which plane(ex. plane 0, 1 are graphics plane 2 is the metadata). The purpose here it not invoke the metadata buffer with any DMA buffer procedure. Usually, such a metadata would be the value should be written to a hardware's registers, a 4KiB page would be 1024 items of 32 bits registers. Still, I have some problems with SHMEM: 1. I don't want the userspace modify the context of the SHMEM allocated by the kernel, is there a way to do so? This is generally impossible without doing any of the two: 1) copying the contents to an internal buffer not accessible to the userspace, OR 2) modifying any of the buffer mappings to read-only 2) can actually be more costly than 1) (depending on the architecture, data size, etc.), so we shouldn't just discard the option of a
Re: [RFC]: shmem fd for non-DMA buffer sharing cross drivers
On Wed, Aug 23, 2023 at 4:11 PM Hsia-Jun Li wrote: > > > > On 8/23/23 12:46, Tomasz Figa wrote: > > CAUTION: Email originated externally, do not click links or open > > attachments unless you recognize the sender and know the content is safe. > > > > > > Hi Hsia-Jun, > > > > On Tue, Aug 22, 2023 at 8:14 PM Hsia-Jun Li wrote: > >> > >> Hello > >> > >> I would like to introduce a usage of SHMEM slimier to DMA-buf, the major > >> purpose of that is sharing metadata or just a pure container for cross > >> drivers. > >> > >> We need to exchange some sort of metadata between drivers, likes dynamic > >> HDR data between video4linux2 and DRM. > > > > If the metadata isn't too big, would it be enough to just have the > > kernel copy_from_user() to a kernel buffer in the ioctl code? > > > >> Or the graphics frame buffer is > >> too complex to be described with plain plane's DMA-buf fd. > >> An issue between DRM and V4L2 is that DRM could only support 4 planes > >> while it is 8 for V4L2. It would be pretty hard for DRM to expend its > >> interface to support that 4 more planes which would lead to revision of > >> many standard likes Vulkan, EGL. > > > > Could you explain how a shmem buffer could be used to support frame > > buffers with more than 4 planes? > > If you are asking why we need this: I'm asking how your proposal to use shmem FD solves the problem for those cases. > 1. metadata likes dynamic HDR tone data > 2. DRM also challenges with this problem, let me quote what sima said: > "another trick that we iirc used for afbc is that sometimes the planes > have a fixed layout > like nv12 > and so logically it's multiple planes, but you only need one plane slot > to describe the buffer > since I think afbc had the "we need more than 4 planes" issue too" > > Unfortunately, there are vendor pixel formats are not fixed layout. > > 3. Secure(REE, trusted video piepline) info. > > For how to assign such metadata data. > In case with a drm fb_id, it is simple, we just add a drm plane property > for it. The V4L2 interface is not flexible, we could only leave into > CAPTURE request_fd as a control. > >> > >> Also, there is no reason to consume a device's memory for the content > >> that device can't read it, or wasting an entry of IOMMU for such data. > > > > That's right, but DMA-buf doesn't really imply any of those. DMA-buf > > is just a kernel object with some backing memory. It's up to the > > allocator to decide how the backing memory is allocated and up to the > > importer on whether it would be mapped into an IOMMU. > > > I just want to say it can't be allocated at the same place which was for > those DMA bufs(graphics or compressed bitstream). > This also could be answer for your first question, if we place this kind > of buffer in a plane for DMABUF(importing) in V4L2, V4L2 core would try > to prepare it, which could map it into IOMMU. > V4L2 core will prepare it according to the struct device that is given to it. For the planes that don't have to go to the hardware a struct device could be given that doesn't require any DMA mapping. Also you can check how the uvcvideo driver handles it. It doesn't use the vb2 buffers directly, but always writes to them using CPU (due to how the UVC protocol is designed). > >> Usually, such a metadata would be the value should be written to a > >> hardware's registers, a 4KiB page would be 1024 items of 32 bits registers. > >> > >> Still, I have some problems with SHMEM: > >> 1. I don't want the userspace modify the context of the SHMEM allocated > >> by the kernel, is there a way to do so? > > > > This is generally impossible without doing any of the two: > > 1) copying the contents to an internal buffer not accessible to the > > userspace, OR > > 2) modifying any of the buffer mappings to read-only > > > > 2) can actually be more costly than 1) (depending on the architecture, > > data size, etc.), so we shouldn't just discard the option of a simple > > copy_from_user() in the ioctl. > > > I don't want the userspace access it at all. So that won't be a problem. In this case, wouldn't it be enough to have a DMA-buf exporter that doesn't provide the mmap op? > >> 2. Should I create a helper function for installing the SHMEM file as a fd? > > > > We already have the udmabuf device [1] to turn a memfd into a DMA-buf, > > so maybe that would be enough? > > > > [1] > > https://elixir.bootlin.com/linux/v6.5-rc7/source/drivers/dma-buf/udmabuf.c > > > It is the kernel driver that allocate this buffer. For example, v4l2 > CAPTURE allocate a buffer for metadata when VIDIOC_REQBUFS. > Or GBM give you a fd which is assigned with a surface. > > So we need a kernel interface. Sorry, I'm confused. If we're talking about buffers allocated by the specific allocators like V4L2 or GBM, why do we need SHMEM at all? Best, Tomasz > > Best, > > Tomasz > > > >> > >> -- > >> Hsia-Jun(Randy) Li > > -- > Hsia-Jun(Randy) Li
Re: [RFC]: shmem fd for non-DMA buffer sharing cross drivers
On 8/23/23 12:46, Tomasz Figa wrote: CAUTION: Email originated externally, do not click links or open attachments unless you recognize the sender and know the content is safe. Hi Hsia-Jun, On Tue, Aug 22, 2023 at 8:14 PM Hsia-Jun Li wrote: Hello I would like to introduce a usage of SHMEM slimier to DMA-buf, the major purpose of that is sharing metadata or just a pure container for cross drivers. We need to exchange some sort of metadata between drivers, likes dynamic HDR data between video4linux2 and DRM. If the metadata isn't too big, would it be enough to just have the kernel copy_from_user() to a kernel buffer in the ioctl code? Or the graphics frame buffer is too complex to be described with plain plane's DMA-buf fd. An issue between DRM and V4L2 is that DRM could only support 4 planes while it is 8 for V4L2. It would be pretty hard for DRM to expend its interface to support that 4 more planes which would lead to revision of many standard likes Vulkan, EGL. Could you explain how a shmem buffer could be used to support frame buffers with more than 4 planes? If you are asking why we need this: 1. metadata likes dynamic HDR tone data 2. DRM also challenges with this problem, let me quote what sima said: "another trick that we iirc used for afbc is that sometimes the planes have a fixed layout like nv12 and so logically it's multiple planes, but you only need one plane slot to describe the buffer since I think afbc had the "we need more than 4 planes" issue too" Unfortunately, there are vendor pixel formats are not fixed layout. 3. Secure(REE, trusted video piepline) info. For how to assign such metadata data. In case with a drm fb_id, it is simple, we just add a drm plane property for it. The V4L2 interface is not flexible, we could only leave into CAPTURE request_fd as a control. Also, there is no reason to consume a device's memory for the content that device can't read it, or wasting an entry of IOMMU for such data. That's right, but DMA-buf doesn't really imply any of those. DMA-buf is just a kernel object with some backing memory. It's up to the allocator to decide how the backing memory is allocated and up to the importer on whether it would be mapped into an IOMMU. I just want to say it can't be allocated at the same place which was for those DMA bufs(graphics or compressed bitstream). This also could be answer for your first question, if we place this kind of buffer in a plane for DMABUF(importing) in V4L2, V4L2 core would try to prepare it, which could map it into IOMMU. Usually, such a metadata would be the value should be written to a hardware's registers, a 4KiB page would be 1024 items of 32 bits registers. Still, I have some problems with SHMEM: 1. I don't want the userspace modify the context of the SHMEM allocated by the kernel, is there a way to do so? This is generally impossible without doing any of the two: 1) copying the contents to an internal buffer not accessible to the userspace, OR 2) modifying any of the buffer mappings to read-only 2) can actually be more costly than 1) (depending on the architecture, data size, etc.), so we shouldn't just discard the option of a simple copy_from_user() in the ioctl. I don't want the userspace access it at all. So that won't be a problem. 2. Should I create a helper function for installing the SHMEM file as a fd? We already have the udmabuf device [1] to turn a memfd into a DMA-buf, so maybe that would be enough? [1] https://elixir.bootlin.com/linux/v6.5-rc7/source/drivers/dma-buf/udmabuf.c It is the kernel driver that allocate this buffer. For example, v4l2 CAPTURE allocate a buffer for metadata when VIDIOC_REQBUFS. Or GBM give you a fd which is assigned with a surface. So we need a kernel interface. Best, Tomasz -- Hsia-Jun(Randy) Li -- Hsia-Jun(Randy) Li
Re: [RFC]: shmem fd for non-DMA buffer sharing cross drivers
Hi Hsia-Jun, On Tue, Aug 22, 2023 at 8:14 PM Hsia-Jun Li wrote: > > Hello > > I would like to introduce a usage of SHMEM slimier to DMA-buf, the major > purpose of that is sharing metadata or just a pure container for cross > drivers. > > We need to exchange some sort of metadata between drivers, likes dynamic > HDR data between video4linux2 and DRM. If the metadata isn't too big, would it be enough to just have the kernel copy_from_user() to a kernel buffer in the ioctl code? > Or the graphics frame buffer is > too complex to be described with plain plane's DMA-buf fd. > An issue between DRM and V4L2 is that DRM could only support 4 planes > while it is 8 for V4L2. It would be pretty hard for DRM to expend its > interface to support that 4 more planes which would lead to revision of > many standard likes Vulkan, EGL. Could you explain how a shmem buffer could be used to support frame buffers with more than 4 planes? > > Also, there is no reason to consume a device's memory for the content > that device can't read it, or wasting an entry of IOMMU for such data. That's right, but DMA-buf doesn't really imply any of those. DMA-buf is just a kernel object with some backing memory. It's up to the allocator to decide how the backing memory is allocated and up to the importer on whether it would be mapped into an IOMMU. > Usually, such a metadata would be the value should be written to a > hardware's registers, a 4KiB page would be 1024 items of 32 bits registers. > > Still, I have some problems with SHMEM: > 1. I don't want thhe userspace modify the context of the SHMEM allocated > by the kernel, is there a way to do so? This is generally impossible without doing any of the two: 1) copying the contents to an internal buffer not accessible to the userspace, OR 2) modifying any of the buffer mappings to read-only 2) can actually be more costly than 1) (depending on the architecture, data size, etc.), so we shouldn't just discard the option of a simple copy_from_user() in the ioctl. > 2. Should I create a helper function for installing the SHMEM file as a fd? We already have the udmabuf device [1] to turn a memfd into a DMA-buf, so maybe that would be enough? [1] https://elixir.bootlin.com/linux/v6.5-rc7/source/drivers/dma-buf/udmabuf.c Best, Tomasz > > -- > Hsia-Jun(Randy) Li
Re: [RFC]: shmem fd for non-DMA buffer sharing cross drivers
On 8/23/23 03:55, Nicolas Dufresne wrote:
CAUTION: Email originated externally, do not click links or open attachments
unless you recognize the sender and know the content is safe.
Hi,
Le mardi 22 août 2023 à 19:14 +0800, Hsia-Jun Li a écrit :
Hello
I would like to introduce a usage of SHMEM slimier to DMA-buf, the major
purpose of that is sharing metadata or just a pure container for cross
drivers.
We need to exchange some sort of metadata between drivers, likes dynamic
HDR data between video4linux2 and DRM. Or the graphics frame buffer is
too complex to be described with plain plane's DMA-buf fd.
An issue between DRM and V4L2 is that DRM could only support 4 planes
while it is 8 for V4L2. It would be pretty hard for DRM to expend its
interface to support that 4 more planes which would lead to revision of
many standard likes Vulkan, EGL.
Also, there is no reason to consume a device's memory for the content
that device can't read it, or wasting an entry of IOMMU for such data.
Usually, such a metadata would be the value should be written to a
hardware's registers, a 4KiB page would be 1024 items of 32 bits registers.
Still, I have some problems with SHMEM:
1. I don't want thhe userspace modify the context of the SHMEM allocated
by the kernel, is there a way to do so?
2. Should I create a helper function for installing the SHMEM file as a fd?
Please have a look at memfd and the seal feature, it does cover the reason why
That is the implement I need, it would affact the userspace not the
kernel space. Should I expand a kAPI for memfd or just take the
implement for the SHMEM?
This interfaces need to offer three things:
1. a fd for userspace to exchange between drivers
2. a kernel virtual address for accessing
3. userspace SEAL
Meanwhile, I am thinking whether we should offer a generic context
header for such usage. Or we need another fields in a driver to describe it.
struct shmem_generic_container {
u64 format; /* use DRM modifier vendor bits but */
u32 size; /* size of the payload */
u8 payload[];
};
/* format linear for nesting dolls context */
struct shmem_nesting_container {
u32 num;
u64 formats[num];
u32 sizes[num];
u32 offsets[num]; /* offset from the payload below */
u8 payload[];
};
unsealed shared memory require full trust. For controls, the SEAL_WRITE is even
needed, as with appropriate timing, a malicous process can modify the data in-
between validation and allocation, causing possible memory overflow.
https://man7.org/linux/man-pages/man2/memfd_create.2.html
File sealing
In the absence of file sealing, processes that communicate via
shared memory must either trust each other, or take measures to
deal with the possibility that an untrusted peer may manipulate
the shared memory region in problematic ways. For example, an
untrusted peer might modify the contents of the shared memory at
any time, or shrink the shared memory region. The former
possibility leaves the local process vulnerable to time-of-check-
to-time-of-use race conditions (typically dealt with by copying
data from the shared memory region before checking and using it).
The latter possibility leaves the local process vulnerable to
SIGBUS signals when an attempt is made to access a now-
nonexistent location in the shared memory region. (Dealing with
this possibility necessitates the use of a handler for the SIGBUS
signal.)
Dealing with untrusted peers imposes extra complexity on code
that employs shared memory. Memory sealing enables that extra
complexity to be eliminated, by allowing a process to operate
secure in the knowledge that its peer can't modify the shared
memory in an undesired fashion.
[...]
regards,
Nicolas
--
Hsia-Jun(Randy) Li
Re: [RFC]: shmem fd for non-DMA buffer sharing cross drivers
Hi, Le mardi 22 août 2023 à 19:14 +0800, Hsia-Jun Li a écrit : > Hello > > I would like to introduce a usage of SHMEM slimier to DMA-buf, the major > purpose of that is sharing metadata or just a pure container for cross > drivers. > > We need to exchange some sort of metadata between drivers, likes dynamic > HDR data between video4linux2 and DRM. Or the graphics frame buffer is > too complex to be described with plain plane's DMA-buf fd. > An issue between DRM and V4L2 is that DRM could only support 4 planes > while it is 8 for V4L2. It would be pretty hard for DRM to expend its > interface to support that 4 more planes which would lead to revision of > many standard likes Vulkan, EGL. > > Also, there is no reason to consume a device's memory for the content > that device can't read it, or wasting an entry of IOMMU for such data. > Usually, such a metadata would be the value should be written to a > hardware's registers, a 4KiB page would be 1024 items of 32 bits registers. > > Still, I have some problems with SHMEM: > 1. I don't want thhe userspace modify the context of the SHMEM allocated > by the kernel, is there a way to do so? > 2. Should I create a helper function for installing the SHMEM file as a fd? Please have a look at memfd and the seal feature, it does cover the reason why unsealed shared memory require full trust. For controls, the SEAL_WRITE is even needed, as with appropriate timing, a malicous process can modify the data in- between validation and allocation, causing possible memory overflow. https://man7.org/linux/man-pages/man2/memfd_create.2.html File sealing In the absence of file sealing, processes that communicate via shared memory must either trust each other, or take measures to deal with the possibility that an untrusted peer may manipulate the shared memory region in problematic ways. For example, an untrusted peer might modify the contents of the shared memory at any time, or shrink the shared memory region. The former possibility leaves the local process vulnerable to time-of-check- to-time-of-use race conditions (typically dealt with by copying data from the shared memory region before checking and using it). The latter possibility leaves the local process vulnerable to SIGBUS signals when an attempt is made to access a now- nonexistent location in the shared memory region. (Dealing with this possibility necessitates the use of a handler for the SIGBUS signal.) Dealing with untrusted peers imposes extra complexity on code that employs shared memory. Memory sealing enables that extra complexity to be eliminated, by allowing a process to operate secure in the knowledge that its peer can't modify the shared memory in an undesired fashion. [...] regards, Nicolas
[RFC]: shmem fd for non-DMA buffer sharing cross drivers
Hello I would like to introduce a usage of SHMEM slimier to DMA-buf, the major purpose of that is sharing metadata or just a pure container for cross drivers. We need to exchange some sort of metadata between drivers, likes dynamic HDR data between video4linux2 and DRM. Or the graphics frame buffer is too complex to be described with plain plane's DMA-buf fd. An issue between DRM and V4L2 is that DRM could only support 4 planes while it is 8 for V4L2. It would be pretty hard for DRM to expend its interface to support that 4 more planes which would lead to revision of many standard likes Vulkan, EGL. Also, there is no reason to consume a device's memory for the content that device can't read it, or wasting an entry of IOMMU for such data. Usually, such a metadata would be the value should be written to a hardware's registers, a 4KiB page would be 1024 items of 32 bits registers. Still, I have some problems with SHMEM: 1. I don't want thhe userspace modify the context of the SHMEM allocated by the kernel, is there a way to do so? 2. Should I create a helper function for installing the SHMEM file as a fd? -- Hsia-Jun(Randy) Li
