Message ID | 20241003-rp1-cfe-v6-0-d6762edd98a8@ideasonboard.com (mailing list archive) |
---|---|
Headers | show |
Series | media: raspberrypi: Support RPi5's CFE | expand |
Hi Tomi, I know this driver is already merged, but while checking for drivers that use q->max_num_buffers I stumbled on this cfe code: <snip> > +/* > + * vb2 ops > + */ > + > +static int cfe_queue_setup(struct vb2_queue *vq, unsigned int *nbuffers, > + unsigned int *nplanes, unsigned int sizes[], > + struct device *alloc_devs[]) > +{ > + struct cfe_node *node = vb2_get_drv_priv(vq); > + struct cfe_device *cfe = node->cfe; > + unsigned int size = is_image_node(node) ? > + node->vid_fmt.fmt.pix.sizeimage : > + node->meta_fmt.fmt.meta.buffersize; > + > + cfe_dbg(cfe, "%s: [%s] type:%u\n", __func__, node_desc[node->id].name, > + node->buffer_queue.type); > + > + if (vq->max_num_buffers + *nbuffers < 3) > + *nbuffers = 3 - vq->max_num_buffers; This makes no sense: max_num_buffers is 32, unless explicitly set when vb2_queue_init is called. So 32 + *nbuffers is never < 3. If the idea is that at least 3 buffers should be allocated by REQBUFS, then set q->min_reqbufs_allocation = 3; before calling vb2_queue_init and vb2 will handle this for you. Drivers shouldn't modify *nbuffers, except in very rare circumstances, especially since the code is almost always wrong. Regards, Hans > + > + if (*nplanes) { > + if (sizes[0] < size) { > + cfe_err(cfe, "sizes[0] %i < size %u\n", sizes[0], size); > + return -EINVAL; > + } > + size = sizes[0]; > + } > + > + *nplanes = 1; > + sizes[0] = size; > + > + return 0; > +}
Hi Hans, On 24/10/2024 11:20, Hans Verkuil wrote: > Hi Tomi, > > I know this driver is already merged, but while checking for drivers that use > q->max_num_buffers I stumbled on this cfe code: > > <snip> > >> +/* >> + * vb2 ops >> + */ >> + >> +static int cfe_queue_setup(struct vb2_queue *vq, unsigned int *nbuffers, >> + unsigned int *nplanes, unsigned int sizes[], >> + struct device *alloc_devs[]) >> +{ >> + struct cfe_node *node = vb2_get_drv_priv(vq); >> + struct cfe_device *cfe = node->cfe; >> + unsigned int size = is_image_node(node) ? >> + node->vid_fmt.fmt.pix.sizeimage : >> + node->meta_fmt.fmt.meta.buffersize; >> + >> + cfe_dbg(cfe, "%s: [%s] type:%u\n", __func__, node_desc[node->id].name, >> + node->buffer_queue.type); >> + >> + if (vq->max_num_buffers + *nbuffers < 3) >> + *nbuffers = 3 - vq->max_num_buffers; > > This makes no sense: max_num_buffers is 32, unless explicitly set when vb2_queue_init > is called. So 32 + *nbuffers is never < 3. > > If the idea is that at least 3 buffers should be allocated by REQBUFS, then set > q->min_reqbufs_allocation = 3; before calling vb2_queue_init and vb2 will handle this > for you. > > Drivers shouldn't modify *nbuffers, except in very rare circumstances, especially > since the code is almost always wrong. Indeed, the code doesn't make sense. I have to say I don't know what was the intent here, but I think "at least 3 buffers should be allocated by REQBUFS" is the likely explanation. I think the hardware should work with even just a single buffer, so is it then fine to not set either q->min_queued_buffers nor q->min_reqbufs_allocation before calling vb2_queue_init()? This seems to result in REQBUFS giving at least two buffers. Tomi > > Regards, > > Hans > >> + >> + if (*nplanes) { >> + if (sizes[0] < size) { >> + cfe_err(cfe, "sizes[0] %i < size %u\n", sizes[0], size); >> + return -EINVAL; >> + } >> + size = sizes[0]; >> + } >> + >> + *nplanes = 1; >> + sizes[0] = size; >> + >> + return 0; >> +} >
On 24/10/2024 13:08, Tomi Valkeinen wrote: > Hi Hans, > > On 24/10/2024 11:20, Hans Verkuil wrote: >> Hi Tomi, >> >> I know this driver is already merged, but while checking for drivers that use >> q->max_num_buffers I stumbled on this cfe code: >> >> <snip> >> >>> +/* >>> + * vb2 ops >>> + */ >>> + >>> +static int cfe_queue_setup(struct vb2_queue *vq, unsigned int *nbuffers, >>> + unsigned int *nplanes, unsigned int sizes[], >>> + struct device *alloc_devs[]) >>> +{ >>> + struct cfe_node *node = vb2_get_drv_priv(vq); >>> + struct cfe_device *cfe = node->cfe; >>> + unsigned int size = is_image_node(node) ? >>> + node->vid_fmt.fmt.pix.sizeimage : >>> + node->meta_fmt.fmt.meta.buffersize; >>> + >>> + cfe_dbg(cfe, "%s: [%s] type:%u\n", __func__, node_desc[node->id].name, >>> + node->buffer_queue.type); >>> + >>> + if (vq->max_num_buffers + *nbuffers < 3) >>> + *nbuffers = 3 - vq->max_num_buffers; >> >> This makes no sense: max_num_buffers is 32, unless explicitly set when vb2_queue_init >> is called. So 32 + *nbuffers is never < 3. >> >> If the idea is that at least 3 buffers should be allocated by REQBUFS, then set >> q->min_reqbufs_allocation = 3; before calling vb2_queue_init and vb2 will handle this >> for you. >> >> Drivers shouldn't modify *nbuffers, except in very rare circumstances, especially >> since the code is almost always wrong. > > Indeed, the code doesn't make sense. I have to say I don't know what was the intent here, but I think "at least 3 buffers should be allocated by REQBUFS" is the likely explanation. > > I think the hardware should work with even just a single buffer, so is it then fine to not set either q->min_queued_buffers nor q->min_reqbufs_allocation before calling vb2_queue_init()? This seems to > result in REQBUFS giving at least two buffers. min_queued_buffers is really HW dependent. If not set, then start_streaming can be called even if there are no buffers queued. If your hardware can handle that, then it's fine to not set it. Regards, Hans > > Tomi > >> >> Regards, >> >> Hans >> >>> + >>> + if (*nplanes) { >>> + if (sizes[0] < size) { >>> + cfe_err(cfe, "sizes[0] %i < size %u\n", sizes[0], size); >>> + return -EINVAL; >>> + } >>> + size = sizes[0]; >>> + } >>> + >>> + *nplanes = 1; >>> + sizes[0] = size; >>> + >>> + return 0; >>> +} >> >
On Thu, Oct 24, 2024 at 01:21:43PM +0200, Hans Verkuil wrote: > On 24/10/2024 13:08, Tomi Valkeinen wrote: > > On 24/10/2024 11:20, Hans Verkuil wrote: > >> Hi Tomi, > >> > >> I know this driver is already merged, but while checking for drivers that use > >> q->max_num_buffers I stumbled on this cfe code: > >> > >> <snip> > >> > >>> +/* > >>> + * vb2 ops > >>> + */ > >>> + > >>> +static int cfe_queue_setup(struct vb2_queue *vq, unsigned int *nbuffers, > >>> + unsigned int *nplanes, unsigned int sizes[], > >>> + struct device *alloc_devs[]) > >>> +{ > >>> + struct cfe_node *node = vb2_get_drv_priv(vq); > >>> + struct cfe_device *cfe = node->cfe; > >>> + unsigned int size = is_image_node(node) ? > >>> + node->vid_fmt.fmt.pix.sizeimage : > >>> + node->meta_fmt.fmt.meta.buffersize; > >>> + > >>> + cfe_dbg(cfe, "%s: [%s] type:%u\n", __func__, node_desc[node->id].name, > >>> + node->buffer_queue.type); > >>> + > >>> + if (vq->max_num_buffers + *nbuffers < 3) > >>> + *nbuffers = 3 - vq->max_num_buffers; > >> > >> This makes no sense: max_num_buffers is 32, unless explicitly set when vb2_queue_init > >> is called. So 32 + *nbuffers is never < 3. > >> > >> If the idea is that at least 3 buffers should be allocated by REQBUFS, then set > >> q->min_reqbufs_allocation = 3; before calling vb2_queue_init and vb2 will handle this > >> for you. > >> > >> Drivers shouldn't modify *nbuffers, except in very rare circumstances, especially > >> since the code is almost always wrong. > > > > Indeed, the code doesn't make sense. I have to say I don't know what > > was the intent here, but I think "at least 3 buffers should be > > allocated by REQBUFS" is the likely explanation. > > > > I think the hardware should work with even just a single buffer, so > > is it then fine to not set either q->min_queued_buffers nor > > q->min_reqbufs_allocation before calling vb2_queue_init()? This > > seems to result in REQBUFS giving at least two buffers. > > min_queued_buffers is really HW dependent. If not set, then > start_streaming can be called even if there are no buffers queued. Having min_queued_buffers > 1 is bad for userspace, and it's much nicer to have it set to 0. The main issue with a value of 1 is that validation of the pipeline ends up being deferred to the first QBUF if it occurs after STREAMON, and error handling is then complicated. It's not just a property of the hardware, kernel drivers can decide to work with scratch buffers if needed. In many cases, a scratch buffer allocated by the kernel could be very small, either relying on the same physical page being mapped through the IOMMU to a larger DMA space, or using a 0 stride value to write all lines to the same location. For drivers supported by libcamera, we will require min_queued_buffers <= 1 and may tighten that to == 0. Tomi, if you submit a patch, please try to target 0, and if that's too much work for now, set it to 1 at most. > If your hardware can handle that, then it's fine to not set it. > > >>> + > >>> + if (*nplanes) { > >>> + if (sizes[0] < size) { > >>> + cfe_err(cfe, "sizes[0] %i < size %u\n", sizes[0], size); > >>> + return -EINVAL; > >>> + } > >>> + size = sizes[0]; > >>> + } > >>> + > >>> + *nplanes = 1; > >>> + sizes[0] = size; > >>> + > >>> + return 0; > >>> +}
On 24/10/2024 14:05, Laurent Pinchart wrote: > On Thu, Oct 24, 2024 at 01:21:43PM +0200, Hans Verkuil wrote: >> On 24/10/2024 13:08, Tomi Valkeinen wrote: >>> On 24/10/2024 11:20, Hans Verkuil wrote: >>>> Hi Tomi, >>>> >>>> I know this driver is already merged, but while checking for drivers that use >>>> q->max_num_buffers I stumbled on this cfe code: >>>> >>>> <snip> >>>> >>>>> +/* >>>>> + * vb2 ops >>>>> + */ >>>>> + >>>>> +static int cfe_queue_setup(struct vb2_queue *vq, unsigned int *nbuffers, >>>>> + unsigned int *nplanes, unsigned int sizes[], >>>>> + struct device *alloc_devs[]) >>>>> +{ >>>>> + struct cfe_node *node = vb2_get_drv_priv(vq); >>>>> + struct cfe_device *cfe = node->cfe; >>>>> + unsigned int size = is_image_node(node) ? >>>>> + node->vid_fmt.fmt.pix.sizeimage : >>>>> + node->meta_fmt.fmt.meta.buffersize; >>>>> + >>>>> + cfe_dbg(cfe, "%s: [%s] type:%u\n", __func__, node_desc[node->id].name, >>>>> + node->buffer_queue.type); >>>>> + >>>>> + if (vq->max_num_buffers + *nbuffers < 3) >>>>> + *nbuffers = 3 - vq->max_num_buffers; >>>> >>>> This makes no sense: max_num_buffers is 32, unless explicitly set when vb2_queue_init >>>> is called. So 32 + *nbuffers is never < 3. >>>> >>>> If the idea is that at least 3 buffers should be allocated by REQBUFS, then set >>>> q->min_reqbufs_allocation = 3; before calling vb2_queue_init and vb2 will handle this >>>> for you. >>>> >>>> Drivers shouldn't modify *nbuffers, except in very rare circumstances, especially >>>> since the code is almost always wrong. >>> >>> Indeed, the code doesn't make sense. I have to say I don't know what >>> was the intent here, but I think "at least 3 buffers should be >>> allocated by REQBUFS" is the likely explanation. >>> >>> I think the hardware should work with even just a single buffer, so >>> is it then fine to not set either q->min_queued_buffers nor >>> q->min_reqbufs_allocation before calling vb2_queue_init()? This >>> seems to result in REQBUFS giving at least two buffers. >> >> min_queued_buffers is really HW dependent. If not set, then >> start_streaming can be called even if there are no buffers queued. > > Having min_queued_buffers > 1 is bad for userspace, and it's much nicer > to have it set to 0. The main issue with a value of 1 is that validation > of the pipeline ends up being deferred to the first QBUF if it occurs > after STREAMON, and error handling is then complicated. The validation can be done in the prepare_streaming callback instead of in start_streaming. prepare_streaming is called at STREAMON time. > > It's not just a property of the hardware, kernel drivers can decide to > work with scratch buffers if needed. In many cases, a scratch buffer > allocated by the kernel could be very small, either relying on the same > physical page being mapped through the IOMMU to a larger DMA space, or > using a 0 stride value to write all lines to the same location. None of which is possible for some of the older drivers (e.g. TI davinci) that do not have an IOMMU and require two contiguous buffers before they can start streaming. But for modern devices you can solve it through a scratch buffer, that's true. > > For drivers supported by libcamera, we will require min_queued_buffers > <= 1 and may tighten that to == 0. Tomi, if you submit a patch, please > try to target 0, and if that's too much work for now, set it to 1 at > most. Regards, Hans > >> If your hardware can handle that, then it's fine to not set it. >> >>>>> + >>>>> + if (*nplanes) { >>>>> + if (sizes[0] < size) { >>>>> + cfe_err(cfe, "sizes[0] %i < size %u\n", sizes[0], size); >>>>> + return -EINVAL; >>>>> + } >>>>> + size = sizes[0]; >>>>> + } >>>>> + >>>>> + *nplanes = 1; >>>>> + sizes[0] = size; >>>>> + >>>>> + return 0; >>>>> +} >
On Thu, Oct 24, 2024 at 02:13:07PM +0200, Hans Verkuil wrote: > On 24/10/2024 14:05, Laurent Pinchart wrote: > > On Thu, Oct 24, 2024 at 01:21:43PM +0200, Hans Verkuil wrote: > >> On 24/10/2024 13:08, Tomi Valkeinen wrote: > >>> On 24/10/2024 11:20, Hans Verkuil wrote: > >>>> Hi Tomi, > >>>> > >>>> I know this driver is already merged, but while checking for drivers that use > >>>> q->max_num_buffers I stumbled on this cfe code: > >>>> > >>>> <snip> > >>>> > >>>>> +/* > >>>>> + * vb2 ops > >>>>> + */ > >>>>> + > >>>>> +static int cfe_queue_setup(struct vb2_queue *vq, unsigned int *nbuffers, > >>>>> + unsigned int *nplanes, unsigned int sizes[], > >>>>> + struct device *alloc_devs[]) > >>>>> +{ > >>>>> + struct cfe_node *node = vb2_get_drv_priv(vq); > >>>>> + struct cfe_device *cfe = node->cfe; > >>>>> + unsigned int size = is_image_node(node) ? > >>>>> + node->vid_fmt.fmt.pix.sizeimage : > >>>>> + node->meta_fmt.fmt.meta.buffersize; > >>>>> + > >>>>> + cfe_dbg(cfe, "%s: [%s] type:%u\n", __func__, node_desc[node->id].name, > >>>>> + node->buffer_queue.type); > >>>>> + > >>>>> + if (vq->max_num_buffers + *nbuffers < 3) > >>>>> + *nbuffers = 3 - vq->max_num_buffers; > >>>> > >>>> This makes no sense: max_num_buffers is 32, unless explicitly set when vb2_queue_init > >>>> is called. So 32 + *nbuffers is never < 3. > >>>> > >>>> If the idea is that at least 3 buffers should be allocated by REQBUFS, then set > >>>> q->min_reqbufs_allocation = 3; before calling vb2_queue_init and vb2 will handle this > >>>> for you. > >>>> > >>>> Drivers shouldn't modify *nbuffers, except in very rare circumstances, especially > >>>> since the code is almost always wrong. > >>> > >>> Indeed, the code doesn't make sense. I have to say I don't know what > >>> was the intent here, but I think "at least 3 buffers should be > >>> allocated by REQBUFS" is the likely explanation. > >>> > >>> I think the hardware should work with even just a single buffer, so > >>> is it then fine to not set either q->min_queued_buffers nor > >>> q->min_reqbufs_allocation before calling vb2_queue_init()? This > >>> seems to result in REQBUFS giving at least two buffers. > >> > >> min_queued_buffers is really HW dependent. If not set, then > >> start_streaming can be called even if there are no buffers queued. > > > > Having min_queued_buffers > 1 is bad for userspace, and it's much nicer > > to have it set to 0. The main issue with a value of 1 is that validation > > of the pipeline ends up being deferred to the first QBUF if it occurs > > after STREAMON, and error handling is then complicated. > > The validation can be done in the prepare_streaming callback instead of > in start_streaming. prepare_streaming is called at STREAMON time. True, we have that now, I keep forgetting about it. I've noticed the annoying behaviour with drivers that predate .prepare_streaming(). It should be a requirement to validate the pipeline in .prepare_streaming() (or possibly earlier in some cases, not sure). > > It's not just a property of the hardware, kernel drivers can decide to > > work with scratch buffers if needed. In many cases, a scratch buffer > > allocated by the kernel could be very small, either relying on the same > > physical page being mapped through the IOMMU to a larger DMA space, or > > using a 0 stride value to write all lines to the same location. > > None of which is possible for some of the older drivers (e.g. TI davinci) > that do not have an IOMMU and require two contiguous buffers before they can > start streaming. But for modern devices you can solve it through a scratch > buffer, that's true. In the worst case kernel drivers could allocate a full scratch buffer. That's not nice from a memory consumption point of view of course. The stride hack may have worked with TI davinci, I haven't checked. It's something that driver authors don't usually think about even if the hardware can do it. Even if the device supports a 0 stride, it also needs to support programming the stride and the buffer addresses in a race-free way. > > For drivers supported by libcamera, we will require min_queued_buffers > > <= 1 and may tighten that to == 0. Tomi, if you submit a patch, please > > try to target 0, and if that's too much work for now, set it to 1 at > > most. > > Regards, > > Hans > > >> If your hardware can handle that, then it's fine to not set it. > >> > >>>>> + > >>>>> + if (*nplanes) { > >>>>> + if (sizes[0] < size) { > >>>>> + cfe_err(cfe, "sizes[0] %i < size %u\n", sizes[0], size); > >>>>> + return -EINVAL; > >>>>> + } > >>>>> + size = sizes[0]; > >>>>> + } > >>>>> + > >>>>> + *nplanes = 1; > >>>>> + sizes[0] = size; > >>>>> + > >>>>> + return 0; > >>>>> +}
Hi, On 24/10/2024 11:20, Hans Verkuil wrote: > Hi Tomi, > > I know this driver is already merged, but while checking for drivers that use > q->max_num_buffers I stumbled on this cfe code: > > <snip> > >> +/* >> + * vb2 ops >> + */ >> + >> +static int cfe_queue_setup(struct vb2_queue *vq, unsigned int *nbuffers, >> + unsigned int *nplanes, unsigned int sizes[], >> + struct device *alloc_devs[]) >> +{ >> + struct cfe_node *node = vb2_get_drv_priv(vq); >> + struct cfe_device *cfe = node->cfe; >> + unsigned int size = is_image_node(node) ? >> + node->vid_fmt.fmt.pix.sizeimage : >> + node->meta_fmt.fmt.meta.buffersize; >> + >> + cfe_dbg(cfe, "%s: [%s] type:%u\n", __func__, node_desc[node->id].name, >> + node->buffer_queue.type); >> + >> + if (vq->max_num_buffers + *nbuffers < 3) >> + *nbuffers = 3 - vq->max_num_buffers; > > This makes no sense: max_num_buffers is 32, unless explicitly set when vb2_queue_init > is called. So 32 + *nbuffers is never < 3. > > If the idea is that at least 3 buffers should be allocated by REQBUFS, then set > q->min_reqbufs_allocation = 3; before calling vb2_queue_init and vb2 will handle this > for you. > > Drivers shouldn't modify *nbuffers, except in very rare circumstances, especially > since the code is almost always wrong. Looking at this, the original code in the old BSP tree was, which somehow, along the long way, got turned into the above: if (vq->num_buffers + *nbuffers < 3) *nbuffers = 3 - vq->num_buffers; So... I think that is the same as "q->min_reqbufs_allocation = 3"? The distinction between min_queued_buffers and min_reqbufs_allocation, or rather the need for the latter, still escapes me. If the HW/SW requires N buffers to be queued, why would we require allocating more than N buffers? Tomi
On 28/10/2024 10:21, Tomi Valkeinen wrote: > Hi, > > On 24/10/2024 11:20, Hans Verkuil wrote: >> Hi Tomi, >> >> I know this driver is already merged, but while checking for drivers that use >> q->max_num_buffers I stumbled on this cfe code: >> >> <snip> >> >>> +/* >>> + * vb2 ops >>> + */ >>> + >>> +static int cfe_queue_setup(struct vb2_queue *vq, unsigned int *nbuffers, >>> + unsigned int *nplanes, unsigned int sizes[], >>> + struct device *alloc_devs[]) >>> +{ >>> + struct cfe_node *node = vb2_get_drv_priv(vq); >>> + struct cfe_device *cfe = node->cfe; >>> + unsigned int size = is_image_node(node) ? >>> + node->vid_fmt.fmt.pix.sizeimage : >>> + node->meta_fmt.fmt.meta.buffersize; >>> + >>> + cfe_dbg(cfe, "%s: [%s] type:%u\n", __func__, node_desc[node->id].name, >>> + node->buffer_queue.type); >>> + >>> + if (vq->max_num_buffers + *nbuffers < 3) >>> + *nbuffers = 3 - vq->max_num_buffers; >> >> This makes no sense: max_num_buffers is 32, unless explicitly set when vb2_queue_init >> is called. So 32 + *nbuffers is never < 3. >> >> If the idea is that at least 3 buffers should be allocated by REQBUFS, then set >> q->min_reqbufs_allocation = 3; before calling vb2_queue_init and vb2 will handle this >> for you. >> >> Drivers shouldn't modify *nbuffers, except in very rare circumstances, especially >> since the code is almost always wrong. > > Looking at this, the original code in the old BSP tree was, which somehow, along the long way, got turned into the above: > > if (vq->num_buffers + *nbuffers < 3) > *nbuffers = 3 - vq->num_buffers; > > So... I think that is the same as "q->min_reqbufs_allocation = 3"? > > The distinction between min_queued_buffers and min_reqbufs_allocation, or rather the need for the latter, still escapes me. If the HW/SW requires N buffers to be queued, why would we require > allocating more than N buffers? min_queued_buffers is easiest to explain: that represents the requirements of the DMA engine, i.e. how many buffers much be queued before the DMA engine can be started. Typically it is 0, 1 or 2. min_reqbufs_allocation is the minimum number of buffers that will be allocated when calling VIDIOC_REQBUFS in order for userspace to be able to stream without blocking or dropping frames. Typically this is 3 for video capture: one buffer is being DMAed, another is queued up and the third is being processed by userspace. But sometimes drivers have other requirements. The reason is that some applications will just call VIDIOC_REQBUFS with count=1 and expect it to be rounded up to whatever makes sense. See the VIDIOC_REQBUFS doc in https://hverkuil.home.xs4all.nl/spec/userspace-api/v4l/vidioc-reqbufs.html "It can be smaller than the number requested, even zero, when the driver runs out of free memory. A larger number is also possible when the driver requires more buffers to function correctly." How drivers implement this is a mess, and usually the code in the driver is wrong as well. In particular they often did not take VIDIOC_CREATE_BUFS into account, i.e. instead of 'if (vq->num_buffers + *nbuffers < 3)' they would do 'if (*nbuffers < 3)'. When we worked on the support for more than 32 buffers we added min_reqbufs_allocation to let the core take care of this. In addition, this only applies to VIDIOC_REQBUFS, if you want full control over the number of allocated buffers, then use VIDIOC_CREATE_BUFS, with this ioctl the number of buffers will never be more than requested, although it may be less if you run out of memory. I really should go through all existing drivers and fix them up if they try to handle this in the queue_setup function, I suspect a lot of them are quite messy. One thing that is missing in the V4L2 uAPI is a way to report the minimum number of buffers that need to be allocated, i.e. min_queued_buffers + 1. Since if you want to use CREATE_BUFS you need that information so you know that you have to create at least that number of buffers. We have the V4L2_CID_MIN_BUFFERS_FOR_CAPTURE control, but it is effectively codec specific. This probably should be clarified. I wonder if it wouldn't be better to add a min_num_buffers field to struct v4l2_create_buffers and set it to min_queued_buffers + 1. Regards, Hans > > Tomi >
Hi Hans, On 28/10/2024 12:11, Hans Verkuil wrote: > On 28/10/2024 10:21, Tomi Valkeinen wrote: >> Hi, >> >> On 24/10/2024 11:20, Hans Verkuil wrote: >>> Hi Tomi, >>> >>> I know this driver is already merged, but while checking for drivers that use >>> q->max_num_buffers I stumbled on this cfe code: >>> >>> <snip> >>> >>>> +/* >>>> + * vb2 ops >>>> + */ >>>> + >>>> +static int cfe_queue_setup(struct vb2_queue *vq, unsigned int *nbuffers, >>>> + unsigned int *nplanes, unsigned int sizes[], >>>> + struct device *alloc_devs[]) >>>> +{ >>>> + struct cfe_node *node = vb2_get_drv_priv(vq); >>>> + struct cfe_device *cfe = node->cfe; >>>> + unsigned int size = is_image_node(node) ? >>>> + node->vid_fmt.fmt.pix.sizeimage : >>>> + node->meta_fmt.fmt.meta.buffersize; >>>> + >>>> + cfe_dbg(cfe, "%s: [%s] type:%u\n", __func__, node_desc[node->id].name, >>>> + node->buffer_queue.type); >>>> + >>>> + if (vq->max_num_buffers + *nbuffers < 3) >>>> + *nbuffers = 3 - vq->max_num_buffers; >>> >>> This makes no sense: max_num_buffers is 32, unless explicitly set when vb2_queue_init >>> is called. So 32 + *nbuffers is never < 3. >>> >>> If the idea is that at least 3 buffers should be allocated by REQBUFS, then set >>> q->min_reqbufs_allocation = 3; before calling vb2_queue_init and vb2 will handle this >>> for you. >>> >>> Drivers shouldn't modify *nbuffers, except in very rare circumstances, especially >>> since the code is almost always wrong. >> >> Looking at this, the original code in the old BSP tree was, which somehow, along the long way, got turned into the above: >> >> if (vq->num_buffers + *nbuffers < 3) >> *nbuffers = 3 - vq->num_buffers; >> >> So... I think that is the same as "q->min_reqbufs_allocation = 3"? >> >> The distinction between min_queued_buffers and min_reqbufs_allocation, or rather the need for the latter, still escapes me. If the HW/SW requires N buffers to be queued, why would we require >> allocating more than N buffers? > > min_queued_buffers is easiest to explain: that represents the requirements of the DMA > engine, i.e. how many buffers much be queued before the DMA engine can be started. > Typically it is 0, 1 or 2. > > min_reqbufs_allocation is the minimum number of buffers that will be allocated when > calling VIDIOC_REQBUFS in order for userspace to be able to stream without blocking > or dropping frames. > > Typically this is 3 for video capture: one buffer is being DMAed, another is queued up > and the third is being processed by userspace. But sometimes drivers have other > requirements. > > The reason is that some applications will just call VIDIOC_REQBUFS with count=1 and > expect it to be rounded up to whatever makes sense. See the VIDIOC_REQBUFS doc in > https://hverkuil.home.xs4all.nl/spec/userspace-api/v4l/vidioc-reqbufs.html > > "It can be smaller than the number requested, even zero, when the driver runs out of > free memory. A larger number is also possible when the driver requires more buffers > to function correctly." > > How drivers implement this is a mess, and usually the code in the driver is wrong as > well. In particular they often did not take VIDIOC_CREATE_BUFS into account, i.e. > instead of 'if (vq->num_buffers + *nbuffers < 3)' they would do 'if (*nbuffers < 3)'. Thanks, this was educational! So. If I have a driver that has min_queued_buffers = 1, I can use VIDIOC_CREATE_BUFS to allocate a single buffer, and then capture just one buffer, right? Whereas VIDIOC_REQBUFS would give me (probably) three (or two, if the driver does not set min_reqbufs_allocation). Three buffers makes sense for full streaming, of course. > When we worked on the support for more than 32 buffers we added min_reqbufs_allocation > to let the core take care of this. In addition, this only applies to VIDIOC_REQBUFS, > if you want full control over the number of allocated buffers, then use VIDIOC_CREATE_BUFS, > with this ioctl the number of buffers will never be more than requested, although it > may be less if you run out of memory. > > I really should go through all existing drivers and fix them up if they try to > handle this in the queue_setup function, I suspect a lot of them are quite messy. > > One thing that is missing in the V4L2 uAPI is a way to report the minimum number of > buffers that need to be allocated, i.e. min_queued_buffers + 1. Since if you want Hmm, so what I wrote above is not correct? One needs min_queued_buffers + 1? Why is that? > to use CREATE_BUFS you need that information so you know that you have to create > at least that number of buffers. We have the V4L2_CID_MIN_BUFFERS_FOR_CAPTURE control, > but it is effectively codec specific. This probably should be clarified. > > I wonder if it wouldn't be better to add a min_num_buffers field to > struct v4l2_create_buffers and set it to min_queued_buffers + 1. I think this makes sense (although I still don't get the +1). However, based on the experiences from adding the streams features to various ioctls, let's be very careful =). The new 'min_num_buffers' can be filled with garbage by the userspace. If we define the 'min_num_buffers' field to be always filled by the kernel, and any value provided from the userspace to be ignored, I think it should work. Tomi
On 28/10/2024 12:05, Tomi Valkeinen wrote: > Hi Hans, > > On 28/10/2024 12:11, Hans Verkuil wrote: >> On 28/10/2024 10:21, Tomi Valkeinen wrote: >>> Hi, >>> >>> On 24/10/2024 11:20, Hans Verkuil wrote: >>>> Hi Tomi, >>>> >>>> I know this driver is already merged, but while checking for drivers that use >>>> q->max_num_buffers I stumbled on this cfe code: >>>> >>>> <snip> >>>> >>>>> +/* >>>>> + * vb2 ops >>>>> + */ >>>>> + >>>>> +static int cfe_queue_setup(struct vb2_queue *vq, unsigned int *nbuffers, >>>>> + unsigned int *nplanes, unsigned int sizes[], >>>>> + struct device *alloc_devs[]) >>>>> +{ >>>>> + struct cfe_node *node = vb2_get_drv_priv(vq); >>>>> + struct cfe_device *cfe = node->cfe; >>>>> + unsigned int size = is_image_node(node) ? >>>>> + node->vid_fmt.fmt.pix.sizeimage : >>>>> + node->meta_fmt.fmt.meta.buffersize; >>>>> + >>>>> + cfe_dbg(cfe, "%s: [%s] type:%u\n", __func__, node_desc[node->id].name, >>>>> + node->buffer_queue.type); >>>>> + >>>>> + if (vq->max_num_buffers + *nbuffers < 3) >>>>> + *nbuffers = 3 - vq->max_num_buffers; >>>> >>>> This makes no sense: max_num_buffers is 32, unless explicitly set when vb2_queue_init >>>> is called. So 32 + *nbuffers is never < 3. >>>> >>>> If the idea is that at least 3 buffers should be allocated by REQBUFS, then set >>>> q->min_reqbufs_allocation = 3; before calling vb2_queue_init and vb2 will handle this >>>> for you. >>>> >>>> Drivers shouldn't modify *nbuffers, except in very rare circumstances, especially >>>> since the code is almost always wrong. >>> >>> Looking at this, the original code in the old BSP tree was, which somehow, along the long way, got turned into the above: >>> >>> if (vq->num_buffers + *nbuffers < 3) >>> *nbuffers = 3 - vq->num_buffers; >>> >>> So... I think that is the same as "q->min_reqbufs_allocation = 3"? >>> >>> The distinction between min_queued_buffers and min_reqbufs_allocation, or rather the need for the latter, still escapes me. If the HW/SW requires N buffers to be queued, why would we require >>> allocating more than N buffers? >> >> min_queued_buffers is easiest to explain: that represents the requirements of the DMA >> engine, i.e. how many buffers much be queued before the DMA engine can be started. >> Typically it is 0, 1 or 2. >> >> min_reqbufs_allocation is the minimum number of buffers that will be allocated when >> calling VIDIOC_REQBUFS in order for userspace to be able to stream without blocking >> or dropping frames. >> >> Typically this is 3 for video capture: one buffer is being DMAed, another is queued up >> and the third is being processed by userspace. But sometimes drivers have other >> requirements. >> >> The reason is that some applications will just call VIDIOC_REQBUFS with count=1 and >> expect it to be rounded up to whatever makes sense. See the VIDIOC_REQBUFS doc in >> https://hverkuil.home.xs4all.nl/spec/userspace-api/v4l/vidioc-reqbufs.html >> >> "It can be smaller than the number requested, even zero, when the driver runs out of >> free memory. A larger number is also possible when the driver requires more buffers >> to function correctly." >> >> How drivers implement this is a mess, and usually the code in the driver is wrong as >> well. In particular they often did not take VIDIOC_CREATE_BUFS into account, i.e. >> instead of 'if (vq->num_buffers + *nbuffers < 3)' they would do 'if (*nbuffers < 3)'. > > Thanks, this was educational! > > So. If I have a driver that has min_queued_buffers = 1, I can use VIDIOC_CREATE_BUFS to allocate a single buffer, and then capture just one buffer, right? Whereas VIDIOC_REQBUFS would give me > (probably) three (or two, if the driver does not set min_reqbufs_allocation). Three buffers makes sense for full streaming, of course. > >> When we worked on the support for more than 32 buffers we added min_reqbufs_allocation >> to let the core take care of this. In addition, this only applies to VIDIOC_REQBUFS, >> if you want full control over the number of allocated buffers, then use VIDIOC_CREATE_BUFS, >> with this ioctl the number of buffers will never be more than requested, although it >> may be less if you run out of memory. >> >> I really should go through all existing drivers and fix them up if they try to >> handle this in the queue_setup function, I suspect a lot of them are quite messy. >> >> One thing that is missing in the V4L2 uAPI is a way to report the minimum number of >> buffers that need to be allocated, i.e. min_queued_buffers + 1. Since if you want > > Hmm, so what I wrote above is not correct? One needs min_queued_buffers + 1? Why is that? The DMA engine always uses min_queued_buffers, so if there are only that many buffers, then it can never return a buffer to userspace! So you need one more. That's the absolute minimum. For smooth capture you need two more to allow time for userspace to process the buffer. > >> to use CREATE_BUFS you need that information so you know that you have to create >> at least that number of buffers. We have the V4L2_CID_MIN_BUFFERS_FOR_CAPTURE control, >> but it is effectively codec specific. This probably should be clarified. >> >> I wonder if it wouldn't be better to add a min_num_buffers field to >> struct v4l2_create_buffers and set it to min_queued_buffers + 1. > > I think this makes sense (although I still don't get the +1). > > However, based on the experiences from adding the streams features to various ioctls, let's be very careful =). The new 'min_num_buffers' can be filled with garbage by the userspace. If we define the > 'min_num_buffers' field to be always filled by the kernel, and any value provided from the userspace to be ignored, I think it should work. I've posted an RFC for this. Regards, Hans > > Tomi >
Hi, On 28/10/2024 13:13, Hans Verkuil wrote: > On 28/10/2024 12:05, Tomi Valkeinen wrote: >> Hi Hans, >> >> On 28/10/2024 12:11, Hans Verkuil wrote: >>> On 28/10/2024 10:21, Tomi Valkeinen wrote: >>>> Hi, >>>> >>>> On 24/10/2024 11:20, Hans Verkuil wrote: >>>>> Hi Tomi, >>>>> >>>>> I know this driver is already merged, but while checking for drivers that use >>>>> q->max_num_buffers I stumbled on this cfe code: >>>>> >>>>> <snip> >>>>> >>>>>> +/* >>>>>> + * vb2 ops >>>>>> + */ >>>>>> + >>>>>> +static int cfe_queue_setup(struct vb2_queue *vq, unsigned int *nbuffers, >>>>>> + unsigned int *nplanes, unsigned int sizes[], >>>>>> + struct device *alloc_devs[]) >>>>>> +{ >>>>>> + struct cfe_node *node = vb2_get_drv_priv(vq); >>>>>> + struct cfe_device *cfe = node->cfe; >>>>>> + unsigned int size = is_image_node(node) ? >>>>>> + node->vid_fmt.fmt.pix.sizeimage : >>>>>> + node->meta_fmt.fmt.meta.buffersize; >>>>>> + >>>>>> + cfe_dbg(cfe, "%s: [%s] type:%u\n", __func__, node_desc[node->id].name, >>>>>> + node->buffer_queue.type); >>>>>> + >>>>>> + if (vq->max_num_buffers + *nbuffers < 3) >>>>>> + *nbuffers = 3 - vq->max_num_buffers; >>>>> >>>>> This makes no sense: max_num_buffers is 32, unless explicitly set when vb2_queue_init >>>>> is called. So 32 + *nbuffers is never < 3. >>>>> >>>>> If the idea is that at least 3 buffers should be allocated by REQBUFS, then set >>>>> q->min_reqbufs_allocation = 3; before calling vb2_queue_init and vb2 will handle this >>>>> for you. >>>>> >>>>> Drivers shouldn't modify *nbuffers, except in very rare circumstances, especially >>>>> since the code is almost always wrong. >>>> >>>> Looking at this, the original code in the old BSP tree was, which somehow, along the long way, got turned into the above: >>>> >>>> if (vq->num_buffers + *nbuffers < 3) >>>> *nbuffers = 3 - vq->num_buffers; >>>> >>>> So... I think that is the same as "q->min_reqbufs_allocation = 3"? >>>> >>>> The distinction between min_queued_buffers and min_reqbufs_allocation, or rather the need for the latter, still escapes me. If the HW/SW requires N buffers to be queued, why would we require >>>> allocating more than N buffers? >>> >>> min_queued_buffers is easiest to explain: that represents the requirements of the DMA >>> engine, i.e. how many buffers much be queued before the DMA engine can be started. >>> Typically it is 0, 1 or 2. >>> >>> min_reqbufs_allocation is the minimum number of buffers that will be allocated when >>> calling VIDIOC_REQBUFS in order for userspace to be able to stream without blocking >>> or dropping frames. >>> >>> Typically this is 3 for video capture: one buffer is being DMAed, another is queued up >>> and the third is being processed by userspace. But sometimes drivers have other >>> requirements. >>> >>> The reason is that some applications will just call VIDIOC_REQBUFS with count=1 and >>> expect it to be rounded up to whatever makes sense. See the VIDIOC_REQBUFS doc in >>> https://hverkuil.home.xs4all.nl/spec/userspace-api/v4l/vidioc-reqbufs.html >>> >>> "It can be smaller than the number requested, even zero, when the driver runs out of >>> free memory. A larger number is also possible when the driver requires more buffers >>> to function correctly." >>> >>> How drivers implement this is a mess, and usually the code in the driver is wrong as >>> well. In particular they often did not take VIDIOC_CREATE_BUFS into account, i.e. >>> instead of 'if (vq->num_buffers + *nbuffers < 3)' they would do 'if (*nbuffers < 3)'. >> >> Thanks, this was educational! >> >> So. If I have a driver that has min_queued_buffers = 1, I can use VIDIOC_CREATE_BUFS to allocate a single buffer, and then capture just one buffer, right? Whereas VIDIOC_REQBUFS would give me >> (probably) three (or two, if the driver does not set min_reqbufs_allocation). Three buffers makes sense for full streaming, of course. >> >>> When we worked on the support for more than 32 buffers we added min_reqbufs_allocation >>> to let the core take care of this. In addition, this only applies to VIDIOC_REQBUFS, >>> if you want full control over the number of allocated buffers, then use VIDIOC_CREATE_BUFS, >>> with this ioctl the number of buffers will never be more than requested, although it >>> may be less if you run out of memory. >>> >>> I really should go through all existing drivers and fix them up if they try to >>> handle this in the queue_setup function, I suspect a lot of them are quite messy. >>> >>> One thing that is missing in the V4L2 uAPI is a way to report the minimum number of >>> buffers that need to be allocated, i.e. min_queued_buffers + 1. Since if you want >> >> Hmm, so what I wrote above is not correct? One needs min_queued_buffers + 1? Why is that? > > The DMA engine always uses min_queued_buffers, so if there are only that many buffers, > then it can never return a buffer to userspace! So you need one more. That's the absolute > minimum. For smooth capture you need two more to allow time for userspace to process the > buffer. Hmm, ok, I see. Well, I guess my "I want to capture just a single frame" is not a very common case. Can I queue one buffer, start streaming, stop streaming, and get the filled buffer? But then I guess I don't when the buffer has been filled, i.e. when to call stop streaming. So, never mind, I don't actually have any use case for this, just wondering. >> >>> to use CREATE_BUFS you need that information so you know that you have to create >>> at least that number of buffers. We have the V4L2_CID_MIN_BUFFERS_FOR_CAPTURE control, >>> but it is effectively codec specific. This probably should be clarified. >>> >>> I wonder if it wouldn't be better to add a min_num_buffers field to >>> struct v4l2_create_buffers and set it to min_queued_buffers + 1. >> >> I think this makes sense (although I still don't get the +1). >> >> However, based on the experiences from adding the streams features to various ioctls, let's be very careful =). The new 'min_num_buffers' can be filled with garbage by the userspace. If we define the >> 'min_num_buffers' field to be always filled by the kernel, and any value provided from the userspace to be ignored, I think it should work. > > I've posted an RFC for this. Thanks, I'll check it out. For the original issue in this thread, I think the correct fix is to remove the lines from cfe_queue_setup(), and add "q->min_reqbufs_allocation = 3". I'll send a patch for that. Tomi
On 28/10/2024 12:25, Tomi Valkeinen wrote: > Hi, > > On 28/10/2024 13:13, Hans Verkuil wrote: >> On 28/10/2024 12:05, Tomi Valkeinen wrote: >>> Hi Hans, >>> >>> On 28/10/2024 12:11, Hans Verkuil wrote: >>>> On 28/10/2024 10:21, Tomi Valkeinen wrote: >>>>> Hi, >>>>> >>>>> On 24/10/2024 11:20, Hans Verkuil wrote: >>>>>> Hi Tomi, >>>>>> >>>>>> I know this driver is already merged, but while checking for drivers that use >>>>>> q->max_num_buffers I stumbled on this cfe code: >>>>>> >>>>>> <snip> >>>>>> >>>>>>> +/* >>>>>>> + * vb2 ops >>>>>>> + */ >>>>>>> + >>>>>>> +static int cfe_queue_setup(struct vb2_queue *vq, unsigned int *nbuffers, >>>>>>> + unsigned int *nplanes, unsigned int sizes[], >>>>>>> + struct device *alloc_devs[]) >>>>>>> +{ >>>>>>> + struct cfe_node *node = vb2_get_drv_priv(vq); >>>>>>> + struct cfe_device *cfe = node->cfe; >>>>>>> + unsigned int size = is_image_node(node) ? >>>>>>> + node->vid_fmt.fmt.pix.sizeimage : >>>>>>> + node->meta_fmt.fmt.meta.buffersize; >>>>>>> + >>>>>>> + cfe_dbg(cfe, "%s: [%s] type:%u\n", __func__, node_desc[node->id].name, >>>>>>> + node->buffer_queue.type); >>>>>>> + >>>>>>> + if (vq->max_num_buffers + *nbuffers < 3) >>>>>>> + *nbuffers = 3 - vq->max_num_buffers; >>>>>> >>>>>> This makes no sense: max_num_buffers is 32, unless explicitly set when vb2_queue_init >>>>>> is called. So 32 + *nbuffers is never < 3. >>>>>> >>>>>> If the idea is that at least 3 buffers should be allocated by REQBUFS, then set >>>>>> q->min_reqbufs_allocation = 3; before calling vb2_queue_init and vb2 will handle this >>>>>> for you. >>>>>> >>>>>> Drivers shouldn't modify *nbuffers, except in very rare circumstances, especially >>>>>> since the code is almost always wrong. >>>>> >>>>> Looking at this, the original code in the old BSP tree was, which somehow, along the long way, got turned into the above: >>>>> >>>>> if (vq->num_buffers + *nbuffers < 3) >>>>> *nbuffers = 3 - vq->num_buffers; >>>>> >>>>> So... I think that is the same as "q->min_reqbufs_allocation = 3"? >>>>> >>>>> The distinction between min_queued_buffers and min_reqbufs_allocation, or rather the need for the latter, still escapes me. If the HW/SW requires N buffers to be queued, why would we require >>>>> allocating more than N buffers? >>>> >>>> min_queued_buffers is easiest to explain: that represents the requirements of the DMA >>>> engine, i.e. how many buffers much be queued before the DMA engine can be started. >>>> Typically it is 0, 1 or 2. >>>> >>>> min_reqbufs_allocation is the minimum number of buffers that will be allocated when >>>> calling VIDIOC_REQBUFS in order for userspace to be able to stream without blocking >>>> or dropping frames. >>>> >>>> Typically this is 3 for video capture: one buffer is being DMAed, another is queued up >>>> and the third is being processed by userspace. But sometimes drivers have other >>>> requirements. >>>> >>>> The reason is that some applications will just call VIDIOC_REQBUFS with count=1 and >>>> expect it to be rounded up to whatever makes sense. See the VIDIOC_REQBUFS doc in >>>> https://hverkuil.home.xs4all.nl/spec/userspace-api/v4l/vidioc-reqbufs.html >>>> >>>> "It can be smaller than the number requested, even zero, when the driver runs out of >>>> free memory. A larger number is also possible when the driver requires more buffers >>>> to function correctly." >>>> >>>> How drivers implement this is a mess, and usually the code in the driver is wrong as >>>> well. In particular they often did not take VIDIOC_CREATE_BUFS into account, i.e. >>>> instead of 'if (vq->num_buffers + *nbuffers < 3)' they would do 'if (*nbuffers < 3)'. >>> >>> Thanks, this was educational! >>> >>> So. If I have a driver that has min_queued_buffers = 1, I can use VIDIOC_CREATE_BUFS to allocate a single buffer, and then capture just one buffer, right? Whereas VIDIOC_REQBUFS would give me >>> (probably) three (or two, if the driver does not set min_reqbufs_allocation). Three buffers makes sense for full streaming, of course. >>> >>>> When we worked on the support for more than 32 buffers we added min_reqbufs_allocation >>>> to let the core take care of this. In addition, this only applies to VIDIOC_REQBUFS, >>>> if you want full control over the number of allocated buffers, then use VIDIOC_CREATE_BUFS, >>>> with this ioctl the number of buffers will never be more than requested, although it >>>> may be less if you run out of memory. >>>> >>>> I really should go through all existing drivers and fix them up if they try to >>>> handle this in the queue_setup function, I suspect a lot of them are quite messy. >>>> >>>> One thing that is missing in the V4L2 uAPI is a way to report the minimum number of >>>> buffers that need to be allocated, i.e. min_queued_buffers + 1. Since if you want >>> >>> Hmm, so what I wrote above is not correct? One needs min_queued_buffers + 1? Why is that? >> >> The DMA engine always uses min_queued_buffers, so if there are only that many buffers, >> then it can never return a buffer to userspace! So you need one more. That's the absolute >> minimum. For smooth capture you need two more to allow time for userspace to process the >> buffer. > > Hmm, ok, I see. Well, I guess my "I want to capture just a single frame" is not a very common case. > > Can I queue one buffer, start streaming, stop streaming, and get the filled buffer? But then I guess I don't when the buffer has been filled, i.e. when to call stop streaming. Exactly. If you really want that, then the driver has to be adapted in the way that Laurent suggested, i.e. with one or more scratch buffers. But that is not always possible, esp. with older hardware without an IOMMU. Regards, Hans > > So, never mind, I don't actually have any use case for this, just wondering. > >>> >>>> to use CREATE_BUFS you need that information so you know that you have to create >>>> at least that number of buffers. We have the V4L2_CID_MIN_BUFFERS_FOR_CAPTURE control, >>>> but it is effectively codec specific. This probably should be clarified. >>>> >>>> I wonder if it wouldn't be better to add a min_num_buffers field to >>>> struct v4l2_create_buffers and set it to min_queued_buffers + 1. >>> >>> I think this makes sense (although I still don't get the +1). >>> >>> However, based on the experiences from adding the streams features to various ioctls, let's be very careful =). The new 'min_num_buffers' can be filled with garbage by the userspace. If we define the >>> 'min_num_buffers' field to be always filled by the kernel, and any value provided from the userspace to be ignored, I think it should work. >> >> I've posted an RFC for this. > > Thanks, I'll check it out. > > For the original issue in this thread, I think the correct fix is to remove the lines from cfe_queue_setup(), and add "q->min_reqbufs_allocation = 3". > > I'll send a patch for that. > > Tomi >
On Mon, Oct 28, 2024 at 12:30:45PM +0100, Hans Verkuil wrote: > On 28/10/2024 12:25, Tomi Valkeinen wrote: > > On 28/10/2024 13:13, Hans Verkuil wrote: > >> On 28/10/2024 12:05, Tomi Valkeinen wrote: > >>> On 28/10/2024 12:11, Hans Verkuil wrote: > >>>> On 28/10/2024 10:21, Tomi Valkeinen wrote: > >>>>> On 24/10/2024 11:20, Hans Verkuil wrote: > >>>>>> Hi Tomi, > >>>>>> > >>>>>> I know this driver is already merged, but while checking for drivers that use > >>>>>> q->max_num_buffers I stumbled on this cfe code: > >>>>>> > >>>>>> <snip> > >>>>>> > >>>>>>> +/* > >>>>>>> + * vb2 ops > >>>>>>> + */ > >>>>>>> + > >>>>>>> +static int cfe_queue_setup(struct vb2_queue *vq, unsigned int *nbuffers, > >>>>>>> + unsigned int *nplanes, unsigned int sizes[], > >>>>>>> + struct device *alloc_devs[]) > >>>>>>> +{ > >>>>>>> + struct cfe_node *node = vb2_get_drv_priv(vq); > >>>>>>> + struct cfe_device *cfe = node->cfe; > >>>>>>> + unsigned int size = is_image_node(node) ? > >>>>>>> + node->vid_fmt.fmt.pix.sizeimage : > >>>>>>> + node->meta_fmt.fmt.meta.buffersize; > >>>>>>> + > >>>>>>> + cfe_dbg(cfe, "%s: [%s] type:%u\n", __func__, node_desc[node->id].name, > >>>>>>> + node->buffer_queue.type); > >>>>>>> + > >>>>>>> + if (vq->max_num_buffers + *nbuffers < 3) > >>>>>>> + *nbuffers = 3 - vq->max_num_buffers; > >>>>>> > >>>>>> This makes no sense: max_num_buffers is 32, unless explicitly set when vb2_queue_init > >>>>>> is called. So 32 + *nbuffers is never < 3. > >>>>>> > >>>>>> If the idea is that at least 3 buffers should be allocated by REQBUFS, then set > >>>>>> q->min_reqbufs_allocation = 3; before calling vb2_queue_init and vb2 will handle this > >>>>>> for you. > >>>>>> > >>>>>> Drivers shouldn't modify *nbuffers, except in very rare circumstances, especially > >>>>>> since the code is almost always wrong. > >>>>> > >>>>> Looking at this, the original code in the old BSP tree was, which somehow, along the long way, got turned into the above: > >>>>> > >>>>> if (vq->num_buffers + *nbuffers < 3) > >>>>> *nbuffers = 3 - vq->num_buffers; > >>>>> > >>>>> So... I think that is the same as "q->min_reqbufs_allocation = 3"? > >>>>> > >>>>> The distinction between min_queued_buffers and > >>>>> min_reqbufs_allocation, or rather the need for the latter, still > >>>>> escapes me. If the HW/SW requires N buffers to be queued, why > >>>>> would we require allocating more than N buffers? > >>>> > >>>> min_queued_buffers is easiest to explain: that represents the requirements of the DMA > >>>> engine, i.e. how many buffers much be queued before the DMA engine can be started. > >>>> Typically it is 0, 1 or 2. That's partly true only. Even if the hardware requires 2 buffers, a driver can allocate scratch buffers to lower the requirement for userspace. Setting min_queued_buffers to 1 is usually fine, as there are few use cases for userspace to start the hardware before a buffer is available to capture a frame to. A value of 2 is much more problematic, as it prevents operating with a single buffer. I know using a single buffer results in frame drops, but there are resource-constrained systems where application don't always need all the frames (such as the Raspberry Pi Zero for instance). I very strongly encourage drivers to never set a min_queued_buffers value higher than 1. > >>>> > >>>> min_reqbufs_allocation is the minimum number of buffers that will be allocated when > >>>> calling VIDIOC_REQBUFS in order for userspace to be able to stream without blocking > >>>> or dropping frames. > >>>> > >>>> Typically this is 3 for video capture: one buffer is being DMAed, another is queued up > >>>> and the third is being processed by userspace. But sometimes drivers have other > >>>> requirements. This is exactly why I dislike min_reqbufs_allocation when set based on this logic, it encodes assumption on userspace use cases that a capture driver really shouldn't make. > >>>> > >>>> The reason is that some applications will just call VIDIOC_REQBUFS with count=1 and > >>>> expect it to be rounded up to whatever makes sense. See the VIDIOC_REQBUFS doc in > >>>> https://hverkuil.home.xs4all.nl/spec/userspace-api/v4l/vidioc-reqbufs.html > >>>> > >>>> "It can be smaller than the number requested, even zero, when the driver runs out of > >>>> free memory. A larger number is also possible when the driver requires more buffers > >>>> to function correctly." > >>>> > >>>> How drivers implement this is a mess, and usually the code in the driver is wrong as > >>>> well. In particular they often did not take VIDIOC_CREATE_BUFS into account, i.e. > >>>> instead of 'if (vq->num_buffers + *nbuffers < 3)' they would do 'if (*nbuffers < 3)'. > >>> > >>> Thanks, this was educational! > >>> > >>> So. If I have a driver that has min_queued_buffers = 1, I can use > >>> VIDIOC_CREATE_BUFS to allocate a single buffer, and then capture > >>> just one buffer, right? Whereas VIDIOC_REQBUFS would give me > >>> (probably) three (or two, if the driver does not set > >>> min_reqbufs_allocation). Three buffers makes sense for full > >>> streaming, of course. > >>> > >>>> When we worked on the support for more than 32 buffers we added min_reqbufs_allocation > >>>> to let the core take care of this. In addition, this only applies to VIDIOC_REQBUFS, I agree it's better to handle it in the core than in drivers, even if I dislike the feature in the first place. > >>>> if you want full control over the number of allocated buffers, then use VIDIOC_CREATE_BUFS, > >>>> with this ioctl the number of buffers will never be more than requested, although it > >>>> may be less if you run out of memory. On a side note, we should transition libcamera to use VIDIOC_CREATE_BUFS unconditionally. > >>>> > >>>> I really should go through all existing drivers and fix them up if they try to > >>>> handle this in the queue_setup function, I suspect a lot of them are quite messy. > >>>> > >>>> One thing that is missing in the V4L2 uAPI is a way to report the minimum number of > >>>> buffers that need to be allocated, i.e. min_queued_buffers + 1. Since if you want > >>> > >>> Hmm, so what I wrote above is not correct? One needs min_queued_buffers + 1? Why is that? > >> > >> The DMA engine always uses min_queued_buffers, so if there are only that many buffers, > >> then it can never return a buffer to userspace! So you need one more. That's the absolute > >> minimum. For smooth capture you need two more to allow time for userspace to process the > >> buffer. > > > > Hmm, ok, I see. Well, I guess my "I want to capture just a single frame" is not a very common case. It's not that uncommon, see above. > > > > Can I queue one buffer, start streaming, stop streaming, and get the > > filled buffer? But then I guess I don't when the buffer has been > > filled, i.e. when to call stop streaming. > > Exactly. If you really want that, then the driver has to be adapted in the way that Laurent > suggested, i.e. with one or more scratch buffers. But that is not always possible, esp. with > older hardware without an IOMMU. Drivers can always allocate a full-frame scratch buffer in the worst case. That can waste memory though, which is less than ideal. > > So, never mind, I don't actually have any use case for this, just wondering. > > > >>> > >>>> to use CREATE_BUFS you need that information so you know that you have to create > >>>> at least that number of buffers. We have the V4L2_CID_MIN_BUFFERS_FOR_CAPTURE control, > >>>> but it is effectively codec specific. This probably should be clarified. > >>>> > >>>> I wonder if it wouldn't be better to add a min_num_buffers field to > >>>> struct v4l2_create_buffers and set it to min_queued_buffers + 1. Don't add the +1. We should give userspace the information it needs to make informed decisions, not make decisions on its behalf. > >>> > >>> I think this makes sense (although I still don't get the +1). > >>> > >>> However, based on the experiences from adding the streams features > >>> to various ioctls, let's be very careful =). The new > >>> 'min_num_buffers' can be filled with garbage by the userspace. If > >>> we define the 'min_num_buffers' field to be always filled by the > >>> kernel, and any value provided from the userspace to be ignored, I > >>> think it should work. > >> > >> I've posted an RFC for this. > > > > Thanks, I'll check it out. > > > > For the original issue in this thread, I think the correct fix is to > > remove the lines from cfe_queue_setup(), and add > > "q->min_reqbufs_allocation = 3". Or just don't set min_reqbufs_allocation ? This is a new driver, and it requires a device-specific userspace to operate the ISP. I don't think we need to care about applications blindly calling VIDIOC_REQBUFS(1) and expecting to get more buffers. > > > > I'll send a patch for that.
Hi Laurent, On 28/10/2024 17:17, Laurent Pinchart wrote: > On Mon, Oct 28, 2024 at 12:30:45PM +0100, Hans Verkuil wrote: >> On 28/10/2024 12:25, Tomi Valkeinen wrote: >>> On 28/10/2024 13:13, Hans Verkuil wrote: >>>> On 28/10/2024 12:05, Tomi Valkeinen wrote: >>>>> On 28/10/2024 12:11, Hans Verkuil wrote: >>>>>> On 28/10/2024 10:21, Tomi Valkeinen wrote: >>>>>>> On 24/10/2024 11:20, Hans Verkuil wrote: >>>>>>>> Hi Tomi, >>>>>>>> >>>>>>>> I know this driver is already merged, but while checking for drivers that use >>>>>>>> q->max_num_buffers I stumbled on this cfe code: >>>>>>>> >>>>>>>> <snip> >>>>>>>> >>>>>>>>> +/* >>>>>>>>> + * vb2 ops >>>>>>>>> + */ >>>>>>>>> + >>>>>>>>> +static int cfe_queue_setup(struct vb2_queue *vq, unsigned int *nbuffers, >>>>>>>>> + unsigned int *nplanes, unsigned int sizes[], >>>>>>>>> + struct device *alloc_devs[]) >>>>>>>>> +{ >>>>>>>>> + struct cfe_node *node = vb2_get_drv_priv(vq); >>>>>>>>> + struct cfe_device *cfe = node->cfe; >>>>>>>>> + unsigned int size = is_image_node(node) ? >>>>>>>>> + node->vid_fmt.fmt.pix.sizeimage : >>>>>>>>> + node->meta_fmt.fmt.meta.buffersize; >>>>>>>>> + >>>>>>>>> + cfe_dbg(cfe, "%s: [%s] type:%u\n", __func__, node_desc[node->id].name, >>>>>>>>> + node->buffer_queue.type); >>>>>>>>> + >>>>>>>>> + if (vq->max_num_buffers + *nbuffers < 3) >>>>>>>>> + *nbuffers = 3 - vq->max_num_buffers; >>>>>>>> >>>>>>>> This makes no sense: max_num_buffers is 32, unless explicitly set when vb2_queue_init >>>>>>>> is called. So 32 + *nbuffers is never < 3. >>>>>>>> >>>>>>>> If the idea is that at least 3 buffers should be allocated by REQBUFS, then set >>>>>>>> q->min_reqbufs_allocation = 3; before calling vb2_queue_init and vb2 will handle this >>>>>>>> for you. >>>>>>>> >>>>>>>> Drivers shouldn't modify *nbuffers, except in very rare circumstances, especially >>>>>>>> since the code is almost always wrong. >>>>>>> >>>>>>> Looking at this, the original code in the old BSP tree was, which somehow, along the long way, got turned into the above: >>>>>>> >>>>>>> if (vq->num_buffers + *nbuffers < 3) >>>>>>> *nbuffers = 3 - vq->num_buffers; >>>>>>> >>>>>>> So... I think that is the same as "q->min_reqbufs_allocation = 3"? >>>>>>> >>>>>>> The distinction between min_queued_buffers and >>>>>>> min_reqbufs_allocation, or rather the need for the latter, still >>>>>>> escapes me. If the HW/SW requires N buffers to be queued, why >>>>>>> would we require allocating more than N buffers? >>>>>> >>>>>> min_queued_buffers is easiest to explain: that represents the requirements of the DMA >>>>>> engine, i.e. how many buffers much be queued before the DMA engine can be started. >>>>>> Typically it is 0, 1 or 2. > > That's partly true only. Even if the hardware requires 2 buffers, a > driver can allocate scratch buffers to lower the requirement for > userspace. Setting min_queued_buffers to 1 is usually fine, as there are > few use cases for userspace to start the hardware before a buffer is > available to capture a frame to. A value of 2 is much more problematic, > as it prevents operating with a single buffer. I know using a single > buffer results in frame drops, but there are resource-constrained > systems where application don't always need all the frames (such as the > Raspberry Pi Zero for instance). I very strongly encourage drivers to > never set a min_queued_buffers value higher than 1. > >>>>>> >>>>>> min_reqbufs_allocation is the minimum number of buffers that will be allocated when >>>>>> calling VIDIOC_REQBUFS in order for userspace to be able to stream without blocking >>>>>> or dropping frames. >>>>>> >>>>>> Typically this is 3 for video capture: one buffer is being DMAed, another is queued up >>>>>> and the third is being processed by userspace. But sometimes drivers have other >>>>>> requirements. > > This is exactly why I dislike min_reqbufs_allocation when set based on > this logic, it encodes assumption on userspace use cases that a capture > driver really shouldn't make. > >>>>>> >>>>>> The reason is that some applications will just call VIDIOC_REQBUFS with count=1 and >>>>>> expect it to be rounded up to whatever makes sense. See the VIDIOC_REQBUFS doc in >>>>>> https://hverkuil.home.xs4all.nl/spec/userspace-api/v4l/vidioc-reqbufs.html >>>>>> >>>>>> "It can be smaller than the number requested, even zero, when the driver runs out of >>>>>> free memory. A larger number is also possible when the driver requires more buffers >>>>>> to function correctly." >>>>>> >>>>>> How drivers implement this is a mess, and usually the code in the driver is wrong as >>>>>> well. In particular they often did not take VIDIOC_CREATE_BUFS into account, i.e. >>>>>> instead of 'if (vq->num_buffers + *nbuffers < 3)' they would do 'if (*nbuffers < 3)'. >>>>> >>>>> Thanks, this was educational! >>>>> >>>>> So. If I have a driver that has min_queued_buffers = 1, I can use >>>>> VIDIOC_CREATE_BUFS to allocate a single buffer, and then capture >>>>> just one buffer, right? Whereas VIDIOC_REQBUFS would give me >>>>> (probably) three (or two, if the driver does not set >>>>> min_reqbufs_allocation). Three buffers makes sense for full >>>>> streaming, of course. >>>>> >>>>>> When we worked on the support for more than 32 buffers we added min_reqbufs_allocation >>>>>> to let the core take care of this. In addition, this only applies to VIDIOC_REQBUFS, > > I agree it's better to handle it in the core than in drivers, even if I > dislike the feature in the first place. > >>>>>> if you want full control over the number of allocated buffers, then use VIDIOC_CREATE_BUFS, >>>>>> with this ioctl the number of buffers will never be more than requested, although it >>>>>> may be less if you run out of memory. > > On a side note, we should transition libcamera to use VIDIOC_CREATE_BUFS > unconditionally. > >>>>>> >>>>>> I really should go through all existing drivers and fix them up if they try to >>>>>> handle this in the queue_setup function, I suspect a lot of them are quite messy. >>>>>> >>>>>> One thing that is missing in the V4L2 uAPI is a way to report the minimum number of >>>>>> buffers that need to be allocated, i.e. min_queued_buffers + 1. Since if you want >>>>> >>>>> Hmm, so what I wrote above is not correct? One needs min_queued_buffers + 1? Why is that? >>>> >>>> The DMA engine always uses min_queued_buffers, so if there are only that many buffers, >>>> then it can never return a buffer to userspace! So you need one more. That's the absolute >>>> minimum. For smooth capture you need two more to allow time for userspace to process the >>>> buffer. >>> >>> Hmm, ok, I see. Well, I guess my "I want to capture just a single frame" is not a very common case. > > It's not that uncommon, see above. > >>> >>> Can I queue one buffer, start streaming, stop streaming, and get the >>> filled buffer? But then I guess I don't when the buffer has been >>> filled, i.e. when to call stop streaming. >> >> Exactly. If you really want that, then the driver has to be adapted in the way that Laurent >> suggested, i.e. with one or more scratch buffers. But that is not always possible, esp. with >> older hardware without an IOMMU. > > Drivers can always allocate a full-frame scratch buffer in the worst > case. That can waste memory though, which is less than ideal. > >>> So, never mind, I don't actually have any use case for this, just wondering. >>> >>>>> >>>>>> to use CREATE_BUFS you need that information so you know that you have to create >>>>>> at least that number of buffers. We have the V4L2_CID_MIN_BUFFERS_FOR_CAPTURE control, >>>>>> but it is effectively codec specific. This probably should be clarified. >>>>>> >>>>>> I wonder if it wouldn't be better to add a min_num_buffers field to >>>>>> struct v4l2_create_buffers and set it to min_queued_buffers + 1. > > Don't add the +1. We should give userspace the information it needs to > make informed decisions, not make decisions on its behalf. > >>>>> >>>>> I think this makes sense (although I still don't get the +1). >>>>> >>>>> However, based on the experiences from adding the streams features >>>>> to various ioctls, let's be very careful =). The new >>>>> 'min_num_buffers' can be filled with garbage by the userspace. If >>>>> we define the 'min_num_buffers' field to be always filled by the >>>>> kernel, and any value provided from the userspace to be ignored, I >>>>> think it should work. >>>> >>>> I've posted an RFC for this. >>> >>> Thanks, I'll check it out. >>> >>> For the original issue in this thread, I think the correct fix is to >>> remove the lines from cfe_queue_setup(), and add >>> "q->min_reqbufs_allocation = 3". > > Or just don't set min_reqbufs_allocation ? This is a new driver, and it > requires a device-specific userspace to operate the ISP. I don't think > we need to care about applications blindly calling VIDIOC_REQBUFS(1) and > expecting to get more buffers. It doesn't require a device-specific userspace for plain CSI-2 capture. If I understood right, the expected behavior for VIDIOC_REQBUFS is to return enough buffers for "smooth streaming". So even if device-specific userspace would be required, doesn't it still make sense to have min_reqbufs_allocation = 3? Or is your point that even a device-specific userspace, which knows exactly what it's doing, would use VIDIOC_REQBUFS, instead of VIDIOC_CREATE_BUFS? Also, if I don't set min_reqbufs_allocation, VIDIOC_REQBUFS(1) would still allocate two buffers, not one. Tomi
On Mon, Oct 28, 2024 at 05:32:27PM +0200, Tomi Valkeinen wrote: > On 28/10/2024 17:17, Laurent Pinchart wrote: > > On Mon, Oct 28, 2024 at 12:30:45PM +0100, Hans Verkuil wrote: > >> On 28/10/2024 12:25, Tomi Valkeinen wrote: > >>> On 28/10/2024 13:13, Hans Verkuil wrote: > >>>> On 28/10/2024 12:05, Tomi Valkeinen wrote: > >>>>> On 28/10/2024 12:11, Hans Verkuil wrote: > >>>>>> On 28/10/2024 10:21, Tomi Valkeinen wrote: > >>>>>>> On 24/10/2024 11:20, Hans Verkuil wrote: > >>>>>>>> Hi Tomi, > >>>>>>>> > >>>>>>>> I know this driver is already merged, but while checking for drivers that use > >>>>>>>> q->max_num_buffers I stumbled on this cfe code: > >>>>>>>> > >>>>>>>> <snip> > >>>>>>>> > >>>>>>>>> +/* > >>>>>>>>> + * vb2 ops > >>>>>>>>> + */ > >>>>>>>>> + > >>>>>>>>> +static int cfe_queue_setup(struct vb2_queue *vq, unsigned int *nbuffers, > >>>>>>>>> + unsigned int *nplanes, unsigned int sizes[], > >>>>>>>>> + struct device *alloc_devs[]) > >>>>>>>>> +{ > >>>>>>>>> + struct cfe_node *node = vb2_get_drv_priv(vq); > >>>>>>>>> + struct cfe_device *cfe = node->cfe; > >>>>>>>>> + unsigned int size = is_image_node(node) ? > >>>>>>>>> + node->vid_fmt.fmt.pix.sizeimage : > >>>>>>>>> + node->meta_fmt.fmt.meta.buffersize; > >>>>>>>>> + > >>>>>>>>> + cfe_dbg(cfe, "%s: [%s] type:%u\n", __func__, node_desc[node->id].name, > >>>>>>>>> + node->buffer_queue.type); > >>>>>>>>> + > >>>>>>>>> + if (vq->max_num_buffers + *nbuffers < 3) > >>>>>>>>> + *nbuffers = 3 - vq->max_num_buffers; > >>>>>>>> > >>>>>>>> This makes no sense: max_num_buffers is 32, unless explicitly set when vb2_queue_init > >>>>>>>> is called. So 32 + *nbuffers is never < 3. > >>>>>>>> > >>>>>>>> If the idea is that at least 3 buffers should be allocated by REQBUFS, then set > >>>>>>>> q->min_reqbufs_allocation = 3; before calling vb2_queue_init and vb2 will handle this > >>>>>>>> for you. > >>>>>>>> > >>>>>>>> Drivers shouldn't modify *nbuffers, except in very rare circumstances, especially > >>>>>>>> since the code is almost always wrong. > >>>>>>> > >>>>>>> Looking at this, the original code in the old BSP tree was, which somehow, along the long way, got turned into the above: > >>>>>>> > >>>>>>> if (vq->num_buffers + *nbuffers < 3) > >>>>>>> *nbuffers = 3 - vq->num_buffers; > >>>>>>> > >>>>>>> So... I think that is the same as "q->min_reqbufs_allocation = 3"? > >>>>>>> > >>>>>>> The distinction between min_queued_buffers and > >>>>>>> min_reqbufs_allocation, or rather the need for the latter, still > >>>>>>> escapes me. If the HW/SW requires N buffers to be queued, why > >>>>>>> would we require allocating more than N buffers? > >>>>>> > >>>>>> min_queued_buffers is easiest to explain: that represents the requirements of the DMA > >>>>>> engine, i.e. how many buffers much be queued before the DMA engine can be started. > >>>>>> Typically it is 0, 1 or 2. > > > > That's partly true only. Even if the hardware requires 2 buffers, a > > driver can allocate scratch buffers to lower the requirement for > > userspace. Setting min_queued_buffers to 1 is usually fine, as there are > > few use cases for userspace to start the hardware before a buffer is > > available to capture a frame to. A value of 2 is much more problematic, > > as it prevents operating with a single buffer. I know using a single > > buffer results in frame drops, but there are resource-constrained > > systems where application don't always need all the frames (such as the > > Raspberry Pi Zero for instance). I very strongly encourage drivers to > > never set a min_queued_buffers value higher than 1. > > > >>>>>> > >>>>>> min_reqbufs_allocation is the minimum number of buffers that will be allocated when > >>>>>> calling VIDIOC_REQBUFS in order for userspace to be able to stream without blocking > >>>>>> or dropping frames. > >>>>>> > >>>>>> Typically this is 3 for video capture: one buffer is being DMAed, another is queued up > >>>>>> and the third is being processed by userspace. But sometimes drivers have other > >>>>>> requirements. > > > > This is exactly why I dislike min_reqbufs_allocation when set based on > > this logic, it encodes assumption on userspace use cases that a capture > > driver really shouldn't make. > > > >>>>>> > >>>>>> The reason is that some applications will just call VIDIOC_REQBUFS with count=1 and > >>>>>> expect it to be rounded up to whatever makes sense. See the VIDIOC_REQBUFS doc in > >>>>>> https://hverkuil.home.xs4all.nl/spec/userspace-api/v4l/vidioc-reqbufs.html > >>>>>> > >>>>>> "It can be smaller than the number requested, even zero, when the driver runs out of > >>>>>> free memory. A larger number is also possible when the driver requires more buffers > >>>>>> to function correctly." > >>>>>> > >>>>>> How drivers implement this is a mess, and usually the code in the driver is wrong as > >>>>>> well. In particular they often did not take VIDIOC_CREATE_BUFS into account, i.e. > >>>>>> instead of 'if (vq->num_buffers + *nbuffers < 3)' they would do 'if (*nbuffers < 3)'. > >>>>> > >>>>> Thanks, this was educational! > >>>>> > >>>>> So. If I have a driver that has min_queued_buffers = 1, I can use > >>>>> VIDIOC_CREATE_BUFS to allocate a single buffer, and then capture > >>>>> just one buffer, right? Whereas VIDIOC_REQBUFS would give me > >>>>> (probably) three (or two, if the driver does not set > >>>>> min_reqbufs_allocation). Three buffers makes sense for full > >>>>> streaming, of course. > >>>>> > >>>>>> When we worked on the support for more than 32 buffers we added min_reqbufs_allocation > >>>>>> to let the core take care of this. In addition, this only applies to VIDIOC_REQBUFS, > > > > I agree it's better to handle it in the core than in drivers, even if I > > dislike the feature in the first place. > > > >>>>>> if you want full control over the number of allocated buffers, then use VIDIOC_CREATE_BUFS, > >>>>>> with this ioctl the number of buffers will never be more than requested, although it > >>>>>> may be less if you run out of memory. > > > > On a side note, we should transition libcamera to use VIDIOC_CREATE_BUFS > > unconditionally. > > > >>>>>> > >>>>>> I really should go through all existing drivers and fix them up if they try to > >>>>>> handle this in the queue_setup function, I suspect a lot of them are quite messy. > >>>>>> > >>>>>> One thing that is missing in the V4L2 uAPI is a way to report the minimum number of > >>>>>> buffers that need to be allocated, i.e. min_queued_buffers + 1. Since if you want > >>>>> > >>>>> Hmm, so what I wrote above is not correct? One needs min_queued_buffers + 1? Why is that? > >>>> > >>>> The DMA engine always uses min_queued_buffers, so if there are only that many buffers, > >>>> then it can never return a buffer to userspace! So you need one more. That's the absolute > >>>> minimum. For smooth capture you need two more to allow time for userspace to process the > >>>> buffer. > >>> > >>> Hmm, ok, I see. Well, I guess my "I want to capture just a single frame" is not a very common case. > > > > It's not that uncommon, see above. > > > >>> > >>> Can I queue one buffer, start streaming, stop streaming, and get the > >>> filled buffer? But then I guess I don't when the buffer has been > >>> filled, i.e. when to call stop streaming. > >> > >> Exactly. If you really want that, then the driver has to be adapted in the way that Laurent > >> suggested, i.e. with one or more scratch buffers. But that is not always possible, esp. with > >> older hardware without an IOMMU. > > > > Drivers can always allocate a full-frame scratch buffer in the worst > > case. That can waste memory though, which is less than ideal. > > > >>> So, never mind, I don't actually have any use case for this, just wondering. > >>> > >>>>> > >>>>>> to use CREATE_BUFS you need that information so you know that you have to create > >>>>>> at least that number of buffers. We have the V4L2_CID_MIN_BUFFERS_FOR_CAPTURE control, > >>>>>> but it is effectively codec specific. This probably should be clarified. > >>>>>> > >>>>>> I wonder if it wouldn't be better to add a min_num_buffers field to > >>>>>> struct v4l2_create_buffers and set it to min_queued_buffers + 1. > > > > Don't add the +1. We should give userspace the information it needs to > > make informed decisions, not make decisions on its behalf. > > > >>>>> > >>>>> I think this makes sense (although I still don't get the +1). > >>>>> > >>>>> However, based on the experiences from adding the streams features > >>>>> to various ioctls, let's be very careful =). The new > >>>>> 'min_num_buffers' can be filled with garbage by the userspace. If > >>>>> we define the 'min_num_buffers' field to be always filled by the > >>>>> kernel, and any value provided from the userspace to be ignored, I > >>>>> think it should work. > >>>> > >>>> I've posted an RFC for this. > >>> > >>> Thanks, I'll check it out. > >>> > >>> For the original issue in this thread, I think the correct fix is to > >>> remove the lines from cfe_queue_setup(), and add > >>> "q->min_reqbufs_allocation = 3". > > > > Or just don't set min_reqbufs_allocation ? This is a new driver, and it > > requires a device-specific userspace to operate the ISP. I don't think > > we need to care about applications blindly calling VIDIOC_REQBUFS(1) and > > expecting to get more buffers. > > It doesn't require a device-specific userspace for plain CSI-2 capture. > > If I understood right, the expected behavior for VIDIOC_REQBUFS is to > return enough buffers for "smooth streaming". So even if device-specific > userspace would be required, doesn't it still make sense to have > min_reqbufs_allocation = 3? "Smooth streaming" is use case-dependent, you will need different number of buffers for different use cases. That's why I don't like hardcoding this in a video capture driver. I'd rather expose information about the driver behaviour (in particular, how many buffers it will hold on without returning anything to userspace until a new buffer gets queued) and let applications make a decision. I don't expect applications relying on VIDIOC_REQBUFS(1) to work out-of-the-box on Pi 5 anyway, as the media graph needs to be configured. > Or is your point that even a device-specific userspace, which knows > exactly what it's doing, would use VIDIOC_REQBUFS, instead of > VIDIOC_CREATE_BUFS? I expect a device-specific userspace not to require drivers to make policy decisions on its behalf. > Also, if I don't set min_reqbufs_allocation, VIDIOC_REQBUFS(1) would > still allocate two buffers, not one.
On 28/10/2024 17:32, Laurent Pinchart wrote: > On Mon, Oct 28, 2024 at 05:32:27PM +0200, Tomi Valkeinen wrote: >> On 28/10/2024 17:17, Laurent Pinchart wrote: >>> On Mon, Oct 28, 2024 at 12:30:45PM +0100, Hans Verkuil wrote: >>>> On 28/10/2024 12:25, Tomi Valkeinen wrote: >>>>> On 28/10/2024 13:13, Hans Verkuil wrote: >>>>>> On 28/10/2024 12:05, Tomi Valkeinen wrote: >>>>>>> On 28/10/2024 12:11, Hans Verkuil wrote: >>>>>>>> On 28/10/2024 10:21, Tomi Valkeinen wrote: >>>>>>>>> On 24/10/2024 11:20, Hans Verkuil wrote: >>>>>>>>>> Hi Tomi, >>>>>>>>>> >>>>>>>>>> I know this driver is already merged, but while checking for drivers that use >>>>>>>>>> q->max_num_buffers I stumbled on this cfe code: >>>>>>>>>> >>>>>>>>>> <snip> >>>>>>>>>> >>>>>>>>>>> +/* >>>>>>>>>>> + * vb2 ops >>>>>>>>>>> + */ >>>>>>>>>>> + >>>>>>>>>>> +static int cfe_queue_setup(struct vb2_queue *vq, unsigned int *nbuffers, >>>>>>>>>>> + unsigned int *nplanes, unsigned int sizes[], >>>>>>>>>>> + struct device *alloc_devs[]) >>>>>>>>>>> +{ >>>>>>>>>>> + struct cfe_node *node = vb2_get_drv_priv(vq); >>>>>>>>>>> + struct cfe_device *cfe = node->cfe; >>>>>>>>>>> + unsigned int size = is_image_node(node) ? >>>>>>>>>>> + node->vid_fmt.fmt.pix.sizeimage : >>>>>>>>>>> + node->meta_fmt.fmt.meta.buffersize; >>>>>>>>>>> + >>>>>>>>>>> + cfe_dbg(cfe, "%s: [%s] type:%u\n", __func__, node_desc[node->id].name, >>>>>>>>>>> + node->buffer_queue.type); >>>>>>>>>>> + >>>>>>>>>>> + if (vq->max_num_buffers + *nbuffers < 3) >>>>>>>>>>> + *nbuffers = 3 - vq->max_num_buffers; >>>>>>>>>> >>>>>>>>>> This makes no sense: max_num_buffers is 32, unless explicitly set when vb2_queue_init >>>>>>>>>> is called. So 32 + *nbuffers is never < 3. >>>>>>>>>> >>>>>>>>>> If the idea is that at least 3 buffers should be allocated by REQBUFS, then set >>>>>>>>>> q->min_reqbufs_allocation = 3; before calling vb2_queue_init and vb2 will handle this >>>>>>>>>> for you. >>>>>>>>>> >>>>>>>>>> Drivers shouldn't modify *nbuffers, except in very rare circumstances, especially >>>>>>>>>> since the code is almost always wrong. >>>>>>>>> >>>>>>>>> Looking at this, the original code in the old BSP tree was, which somehow, along the long way, got turned into the above: >>>>>>>>> >>>>>>>>> if (vq->num_buffers + *nbuffers < 3) >>>>>>>>> *nbuffers = 3 - vq->num_buffers; >>>>>>>>> >>>>>>>>> So... I think that is the same as "q->min_reqbufs_allocation = 3"? >>>>>>>>> >>>>>>>>> The distinction between min_queued_buffers and >>>>>>>>> min_reqbufs_allocation, or rather the need for the latter, still >>>>>>>>> escapes me. If the HW/SW requires N buffers to be queued, why >>>>>>>>> would we require allocating more than N buffers? >>>>>>>> >>>>>>>> min_queued_buffers is easiest to explain: that represents the requirements of the DMA >>>>>>>> engine, i.e. how many buffers much be queued before the DMA engine can be started. >>>>>>>> Typically it is 0, 1 or 2. >>> >>> That's partly true only. Even if the hardware requires 2 buffers, a >>> driver can allocate scratch buffers to lower the requirement for >>> userspace. Setting min_queued_buffers to 1 is usually fine, as there are >>> few use cases for userspace to start the hardware before a buffer is >>> available to capture a frame to. A value of 2 is much more problematic, >>> as it prevents operating with a single buffer. I know using a single >>> buffer results in frame drops, but there are resource-constrained >>> systems where application don't always need all the frames (such as the >>> Raspberry Pi Zero for instance). I very strongly encourage drivers to >>> never set a min_queued_buffers value higher than 1. >>> >>>>>>>> >>>>>>>> min_reqbufs_allocation is the minimum number of buffers that will be allocated when >>>>>>>> calling VIDIOC_REQBUFS in order for userspace to be able to stream without blocking >>>>>>>> or dropping frames. >>>>>>>> >>>>>>>> Typically this is 3 for video capture: one buffer is being DMAed, another is queued up >>>>>>>> and the third is being processed by userspace. But sometimes drivers have other >>>>>>>> requirements. >>> >>> This is exactly why I dislike min_reqbufs_allocation when set based on >>> this logic, it encodes assumption on userspace use cases that a capture >>> driver really shouldn't make. >>> >>>>>>>> >>>>>>>> The reason is that some applications will just call VIDIOC_REQBUFS with count=1 and >>>>>>>> expect it to be rounded up to whatever makes sense. See the VIDIOC_REQBUFS doc in >>>>>>>> https://hverkuil.home.xs4all.nl/spec/userspace-api/v4l/vidioc-reqbufs.html >>>>>>>> >>>>>>>> "It can be smaller than the number requested, even zero, when the driver runs out of >>>>>>>> free memory. A larger number is also possible when the driver requires more buffers >>>>>>>> to function correctly." >>>>>>>> >>>>>>>> How drivers implement this is a mess, and usually the code in the driver is wrong as >>>>>>>> well. In particular they often did not take VIDIOC_CREATE_BUFS into account, i.e. >>>>>>>> instead of 'if (vq->num_buffers + *nbuffers < 3)' they would do 'if (*nbuffers < 3)'. >>>>>>> >>>>>>> Thanks, this was educational! >>>>>>> >>>>>>> So. If I have a driver that has min_queued_buffers = 1, I can use >>>>>>> VIDIOC_CREATE_BUFS to allocate a single buffer, and then capture >>>>>>> just one buffer, right? Whereas VIDIOC_REQBUFS would give me >>>>>>> (probably) three (or two, if the driver does not set >>>>>>> min_reqbufs_allocation). Three buffers makes sense for full >>>>>>> streaming, of course. >>>>>>> >>>>>>>> When we worked on the support for more than 32 buffers we added min_reqbufs_allocation >>>>>>>> to let the core take care of this. In addition, this only applies to VIDIOC_REQBUFS, >>> >>> I agree it's better to handle it in the core than in drivers, even if I >>> dislike the feature in the first place. >>> >>>>>>>> if you want full control over the number of allocated buffers, then use VIDIOC_CREATE_BUFS, >>>>>>>> with this ioctl the number of buffers will never be more than requested, although it >>>>>>>> may be less if you run out of memory. >>> >>> On a side note, we should transition libcamera to use VIDIOC_CREATE_BUFS >>> unconditionally. >>> >>>>>>>> >>>>>>>> I really should go through all existing drivers and fix them up if they try to >>>>>>>> handle this in the queue_setup function, I suspect a lot of them are quite messy. >>>>>>>> >>>>>>>> One thing that is missing in the V4L2 uAPI is a way to report the minimum number of >>>>>>>> buffers that need to be allocated, i.e. min_queued_buffers + 1. Since if you want >>>>>>> >>>>>>> Hmm, so what I wrote above is not correct? One needs min_queued_buffers + 1? Why is that? >>>>>> >>>>>> The DMA engine always uses min_queued_buffers, so if there are only that many buffers, >>>>>> then it can never return a buffer to userspace! So you need one more. That's the absolute >>>>>> minimum. For smooth capture you need two more to allow time for userspace to process the >>>>>> buffer. >>>>> >>>>> Hmm, ok, I see. Well, I guess my "I want to capture just a single frame" is not a very common case. >>> >>> It's not that uncommon, see above. >>> >>>>> >>>>> Can I queue one buffer, start streaming, stop streaming, and get the >>>>> filled buffer? But then I guess I don't when the buffer has been >>>>> filled, i.e. when to call stop streaming. >>>> >>>> Exactly. If you really want that, then the driver has to be adapted in the way that Laurent >>>> suggested, i.e. with one or more scratch buffers. But that is not always possible, esp. with >>>> older hardware without an IOMMU. >>> >>> Drivers can always allocate a full-frame scratch buffer in the worst >>> case. That can waste memory though, which is less than ideal. >>> >>>>> So, never mind, I don't actually have any use case for this, just wondering. >>>>> >>>>>>> >>>>>>>> to use CREATE_BUFS you need that information so you know that you have to create >>>>>>>> at least that number of buffers. We have the V4L2_CID_MIN_BUFFERS_FOR_CAPTURE control, >>>>>>>> but it is effectively codec specific. This probably should be clarified. >>>>>>>> >>>>>>>> I wonder if it wouldn't be better to add a min_num_buffers field to >>>>>>>> struct v4l2_create_buffers and set it to min_queued_buffers + 1. >>> >>> Don't add the +1. We should give userspace the information it needs to >>> make informed decisions, not make decisions on its behalf. >>> >>>>>>> >>>>>>> I think this makes sense (although I still don't get the +1). >>>>>>> >>>>>>> However, based on the experiences from adding the streams features >>>>>>> to various ioctls, let's be very careful =). The new >>>>>>> 'min_num_buffers' can be filled with garbage by the userspace. If >>>>>>> we define the 'min_num_buffers' field to be always filled by the >>>>>>> kernel, and any value provided from the userspace to be ignored, I >>>>>>> think it should work. >>>>>> >>>>>> I've posted an RFC for this. >>>>> >>>>> Thanks, I'll check it out. >>>>> >>>>> For the original issue in this thread, I think the correct fix is to >>>>> remove the lines from cfe_queue_setup(), and add >>>>> "q->min_reqbufs_allocation = 3". >>> >>> Or just don't set min_reqbufs_allocation ? This is a new driver, and it >>> requires a device-specific userspace to operate the ISP. I don't think >>> we need to care about applications blindly calling VIDIOC_REQBUFS(1) and >>> expecting to get more buffers. >> >> It doesn't require a device-specific userspace for plain CSI-2 capture. >> >> If I understood right, the expected behavior for VIDIOC_REQBUFS is to >> return enough buffers for "smooth streaming". So even if device-specific >> userspace would be required, doesn't it still make sense to have >> min_reqbufs_allocation = 3? > > "Smooth streaming" is use case-dependent, you will need different number > of buffers for different use cases. That's why I don't like hardcoding > this in a video capture driver. I'd rather expose information about the > driver behaviour (in particular, how many buffers it will hold on > without returning anything to userspace until a new buffer gets queued) > and let applications make a decision. I don't expect applications > relying on VIDIOC_REQBUFS(1) to work out-of-the-box on Pi 5 anyway, as > the media graph needs to be configured. > >> Or is your point that even a device-specific userspace, which knows >> exactly what it's doing, would use VIDIOC_REQBUFS, instead of >> VIDIOC_CREATE_BUFS? > > I expect a device-specific userspace not to require drivers to make > policy decisions on its behalf. Remember that libcamera is a specialized library that indeed wants to make policy decisions itself. But many other drivers for much simpler pipelines (typically for video receivers) don't need this and can use the standard V4L2 API. My goal is to have the standard V4L2 API behave in a well-defined manner, while giving enough information to specialized userspace code like libcamera to do their own thing. Regards, Hans > >> Also, if I don't set min_reqbufs_allocation, VIDIOC_REQBUFS(1) would >> still allocate two buffers, not one. >
On Tue, Oct 29, 2024 at 09:23:35AM +0100, Hans Verkuil wrote: > On 28/10/2024 17:32, Laurent Pinchart wrote: > > On Mon, Oct 28, 2024 at 05:32:27PM +0200, Tomi Valkeinen wrote: > >> On 28/10/2024 17:17, Laurent Pinchart wrote: > >>> On Mon, Oct 28, 2024 at 12:30:45PM +0100, Hans Verkuil wrote: > >>>> On 28/10/2024 12:25, Tomi Valkeinen wrote: > >>>>> On 28/10/2024 13:13, Hans Verkuil wrote: > >>>>>> On 28/10/2024 12:05, Tomi Valkeinen wrote: > >>>>>>> On 28/10/2024 12:11, Hans Verkuil wrote: > >>>>>>>> On 28/10/2024 10:21, Tomi Valkeinen wrote: > >>>>>>>>> On 24/10/2024 11:20, Hans Verkuil wrote: > >>>>>>>>>> Hi Tomi, > >>>>>>>>>> > >>>>>>>>>> I know this driver is already merged, but while checking for drivers that use > >>>>>>>>>> q->max_num_buffers I stumbled on this cfe code: > >>>>>>>>>> > >>>>>>>>>> <snip> > >>>>>>>>>> > >>>>>>>>>>> +/* > >>>>>>>>>>> + * vb2 ops > >>>>>>>>>>> + */ > >>>>>>>>>>> + > >>>>>>>>>>> +static int cfe_queue_setup(struct vb2_queue *vq, unsigned int *nbuffers, > >>>>>>>>>>> + unsigned int *nplanes, unsigned int sizes[], > >>>>>>>>>>> + struct device *alloc_devs[]) > >>>>>>>>>>> +{ > >>>>>>>>>>> + struct cfe_node *node = vb2_get_drv_priv(vq); > >>>>>>>>>>> + struct cfe_device *cfe = node->cfe; > >>>>>>>>>>> + unsigned int size = is_image_node(node) ? > >>>>>>>>>>> + node->vid_fmt.fmt.pix.sizeimage : > >>>>>>>>>>> + node->meta_fmt.fmt.meta.buffersize; > >>>>>>>>>>> + > >>>>>>>>>>> + cfe_dbg(cfe, "%s: [%s] type:%u\n", __func__, node_desc[node->id].name, > >>>>>>>>>>> + node->buffer_queue.type); > >>>>>>>>>>> + > >>>>>>>>>>> + if (vq->max_num_buffers + *nbuffers < 3) > >>>>>>>>>>> + *nbuffers = 3 - vq->max_num_buffers; > >>>>>>>>>> > >>>>>>>>>> This makes no sense: max_num_buffers is 32, unless explicitly set when vb2_queue_init > >>>>>>>>>> is called. So 32 + *nbuffers is never < 3. > >>>>>>>>>> > >>>>>>>>>> If the idea is that at least 3 buffers should be allocated by REQBUFS, then set > >>>>>>>>>> q->min_reqbufs_allocation = 3; before calling vb2_queue_init and vb2 will handle this > >>>>>>>>>> for you. > >>>>>>>>>> > >>>>>>>>>> Drivers shouldn't modify *nbuffers, except in very rare circumstances, especially > >>>>>>>>>> since the code is almost always wrong. > >>>>>>>>> > >>>>>>>>> Looking at this, the original code in the old BSP tree was, which somehow, along the long way, got turned into the above: > >>>>>>>>> > >>>>>>>>> if (vq->num_buffers + *nbuffers < 3) > >>>>>>>>> *nbuffers = 3 - vq->num_buffers; > >>>>>>>>> > >>>>>>>>> So... I think that is the same as "q->min_reqbufs_allocation = 3"? > >>>>>>>>> > >>>>>>>>> The distinction between min_queued_buffers and > >>>>>>>>> min_reqbufs_allocation, or rather the need for the latter, still > >>>>>>>>> escapes me. If the HW/SW requires N buffers to be queued, why > >>>>>>>>> would we require allocating more than N buffers? > >>>>>>>> > >>>>>>>> min_queued_buffers is easiest to explain: that represents the requirements of the DMA > >>>>>>>> engine, i.e. how many buffers much be queued before the DMA engine can be started. > >>>>>>>> Typically it is 0, 1 or 2. > >>> > >>> That's partly true only. Even if the hardware requires 2 buffers, a > >>> driver can allocate scratch buffers to lower the requirement for > >>> userspace. Setting min_queued_buffers to 1 is usually fine, as there are > >>> few use cases for userspace to start the hardware before a buffer is > >>> available to capture a frame to. A value of 2 is much more problematic, > >>> as it prevents operating with a single buffer. I know using a single > >>> buffer results in frame drops, but there are resource-constrained > >>> systems where application don't always need all the frames (such as the > >>> Raspberry Pi Zero for instance). I very strongly encourage drivers to > >>> never set a min_queued_buffers value higher than 1. > >>> > >>>>>>>> > >>>>>>>> min_reqbufs_allocation is the minimum number of buffers that will be allocated when > >>>>>>>> calling VIDIOC_REQBUFS in order for userspace to be able to stream without blocking > >>>>>>>> or dropping frames. > >>>>>>>> > >>>>>>>> Typically this is 3 for video capture: one buffer is being DMAed, another is queued up > >>>>>>>> and the third is being processed by userspace. But sometimes drivers have other > >>>>>>>> requirements. > >>> > >>> This is exactly why I dislike min_reqbufs_allocation when set based on > >>> this logic, it encodes assumption on userspace use cases that a capture > >>> driver really shouldn't make. > >>> > >>>>>>>> > >>>>>>>> The reason is that some applications will just call VIDIOC_REQBUFS with count=1 and > >>>>>>>> expect it to be rounded up to whatever makes sense. See the VIDIOC_REQBUFS doc in > >>>>>>>> https://hverkuil.home.xs4all.nl/spec/userspace-api/v4l/vidioc-reqbufs.html > >>>>>>>> > >>>>>>>> "It can be smaller than the number requested, even zero, when the driver runs out of > >>>>>>>> free memory. A larger number is also possible when the driver requires more buffers > >>>>>>>> to function correctly." > >>>>>>>> > >>>>>>>> How drivers implement this is a mess, and usually the code in the driver is wrong as > >>>>>>>> well. In particular they often did not take VIDIOC_CREATE_BUFS into account, i.e. > >>>>>>>> instead of 'if (vq->num_buffers + *nbuffers < 3)' they would do 'if (*nbuffers < 3)'. > >>>>>>> > >>>>>>> Thanks, this was educational! > >>>>>>> > >>>>>>> So. If I have a driver that has min_queued_buffers = 1, I can use > >>>>>>> VIDIOC_CREATE_BUFS to allocate a single buffer, and then capture > >>>>>>> just one buffer, right? Whereas VIDIOC_REQBUFS would give me > >>>>>>> (probably) three (or two, if the driver does not set > >>>>>>> min_reqbufs_allocation). Three buffers makes sense for full > >>>>>>> streaming, of course. > >>>>>>> > >>>>>>>> When we worked on the support for more than 32 buffers we added min_reqbufs_allocation > >>>>>>>> to let the core take care of this. In addition, this only applies to VIDIOC_REQBUFS, > >>> > >>> I agree it's better to handle it in the core than in drivers, even if I > >>> dislike the feature in the first place. > >>> > >>>>>>>> if you want full control over the number of allocated buffers, then use VIDIOC_CREATE_BUFS, > >>>>>>>> with this ioctl the number of buffers will never be more than requested, although it > >>>>>>>> may be less if you run out of memory. > >>> > >>> On a side note, we should transition libcamera to use VIDIOC_CREATE_BUFS > >>> unconditionally. > >>> > >>>>>>>> > >>>>>>>> I really should go through all existing drivers and fix them up if they try to > >>>>>>>> handle this in the queue_setup function, I suspect a lot of them are quite messy. > >>>>>>>> > >>>>>>>> One thing that is missing in the V4L2 uAPI is a way to report the minimum number of > >>>>>>>> buffers that need to be allocated, i.e. min_queued_buffers + 1. Since if you want > >>>>>>> > >>>>>>> Hmm, so what I wrote above is not correct? One needs min_queued_buffers + 1? Why is that? > >>>>>> > >>>>>> The DMA engine always uses min_queued_buffers, so if there are only that many buffers, > >>>>>> then it can never return a buffer to userspace! So you need one more. That's the absolute > >>>>>> minimum. For smooth capture you need two more to allow time for userspace to process the > >>>>>> buffer. > >>>>> > >>>>> Hmm, ok, I see. Well, I guess my "I want to capture just a single frame" is not a very common case. > >>> > >>> It's not that uncommon, see above. > >>> > >>>>> > >>>>> Can I queue one buffer, start streaming, stop streaming, and get the > >>>>> filled buffer? But then I guess I don't when the buffer has been > >>>>> filled, i.e. when to call stop streaming. > >>>> > >>>> Exactly. If you really want that, then the driver has to be adapted in the way that Laurent > >>>> suggested, i.e. with one or more scratch buffers. But that is not always possible, esp. with > >>>> older hardware without an IOMMU. > >>> > >>> Drivers can always allocate a full-frame scratch buffer in the worst > >>> case. That can waste memory though, which is less than ideal. > >>> > >>>>> So, never mind, I don't actually have any use case for this, just wondering. > >>>>> > >>>>>>> > >>>>>>>> to use CREATE_BUFS you need that information so you know that you have to create > >>>>>>>> at least that number of buffers. We have the V4L2_CID_MIN_BUFFERS_FOR_CAPTURE control, > >>>>>>>> but it is effectively codec specific. This probably should be clarified. > >>>>>>>> > >>>>>>>> I wonder if it wouldn't be better to add a min_num_buffers field to > >>>>>>>> struct v4l2_create_buffers and set it to min_queued_buffers + 1. > >>> > >>> Don't add the +1. We should give userspace the information it needs to > >>> make informed decisions, not make decisions on its behalf. > >>> > >>>>>>> > >>>>>>> I think this makes sense (although I still don't get the +1). > >>>>>>> > >>>>>>> However, based on the experiences from adding the streams features > >>>>>>> to various ioctls, let's be very careful =). The new > >>>>>>> 'min_num_buffers' can be filled with garbage by the userspace. If > >>>>>>> we define the 'min_num_buffers' field to be always filled by the > >>>>>>> kernel, and any value provided from the userspace to be ignored, I > >>>>>>> think it should work. > >>>>>> > >>>>>> I've posted an RFC for this. > >>>>> > >>>>> Thanks, I'll check it out. > >>>>> > >>>>> For the original issue in this thread, I think the correct fix is to > >>>>> remove the lines from cfe_queue_setup(), and add > >>>>> "q->min_reqbufs_allocation = 3". > >>> > >>> Or just don't set min_reqbufs_allocation ? This is a new driver, and it > >>> requires a device-specific userspace to operate the ISP. I don't think > >>> we need to care about applications blindly calling VIDIOC_REQBUFS(1) and > >>> expecting to get more buffers. > >> > >> It doesn't require a device-specific userspace for plain CSI-2 capture. > >> > >> If I understood right, the expected behavior for VIDIOC_REQBUFS is to > >> return enough buffers for "smooth streaming". So even if device-specific > >> userspace would be required, doesn't it still make sense to have > >> min_reqbufs_allocation = 3? > > > > "Smooth streaming" is use case-dependent, you will need different number > > of buffers for different use cases. That's why I don't like hardcoding > > this in a video capture driver. I'd rather expose information about the > > driver behaviour (in particular, how many buffers it will hold on > > without returning anything to userspace until a new buffer gets queued) > > and let applications make a decision. I don't expect applications > > relying on VIDIOC_REQBUFS(1) to work out-of-the-box on Pi 5 anyway, as > > the media graph needs to be configured. > > > >> Or is your point that even a device-specific userspace, which knows > >> exactly what it's doing, would use VIDIOC_REQBUFS, instead of > >> VIDIOC_CREATE_BUFS? > > > > I expect a device-specific userspace not to require drivers to make > > policy decisions on its behalf. > > Remember that libcamera is a specialized library that indeed wants to > make policy decisions itself. But many other drivers for much simpler > pipelines (typically for video receivers) don't need this and can use > the standard V4L2 API. > > My goal is to have the standard V4L2 API behave in a well-defined manner, > while giving enough information to specialized userspace code like libcamera > to do their own thing. I think we all agree on well-defined :-) Where we may not agree is that, regardless of whether an application goes through libcamera, interfaces with V4L2 through another framework, or directly, I don't think the kernel should make policy decisions on behalf of userspace. I'm fine keeping VIDIOC_REQBUFS(1) operating as it was meant to because we have to ensure backward compatibility, but I don't think it's the kind of API we should design today. Let's start by making sure we expose the information userspace needs, and then we can discuss the next step. > >> Also, if I don't set min_reqbufs_allocation, VIDIOC_REQBUFS(1) would > >> still allocate two buffers, not one.
This series adds support to the CFE hardware block on RaspberryPi 5. The CFE (Camera Front End) contains a CSI-2 receiver and Front End, a small ISP. To run this, you need the basic RPi5 kernel support plus relevant dts changes to enable the cfe and camera. My work branch with everything needed to run CFE can be found from: git://git.kernel.org/pub/scm/linux/kernel/git/tomba/linux.git rp1-cfe A few notes about the patches: - The original work was done by RaspberryPi, mostly by Naushir Patuck. - The second video node only sets V4L2_CAP_META_CAPTURE instead of both V4L2_CAP_META_CAPTURE and V4L2_CAP_META_CAPTURE like the other nodes. This is a temporary workaround for userspace (libcamera), and hopefully can be removed soon. I have tested this with: - A single IMX219 sensor connected to the RPi5's CSI-2 port - Arducam's UB960 FPD-Link board with four imx219 sensors connected Tomi Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com> --- Changes in v6: - Rename "isr_handler" functions to just "isr" - Fix checkpatch warnings, mainly lines over 80 chars - A diff of the changes be be found from: https://paste.debian.net/1331173/ (note: the tab there seems to be 4 chars) - Link to v5: https://lore.kernel.org/r/20240910-rp1-cfe-v5-0-9ab4c4c8eace@ideasonboard.com Changes in v5: - Add "depends on PM". The platforms that use CFE will always have PM in practice, and it's not worth supporting both the PM and !PM cases as it adds complexity to the driver. - Link to v4: https://lore.kernel.org/r/20240904-rp1-cfe-v4-0-f1b5b3d69c81@ideasonboard.com Changes in v4: - Drop unnecessary clock-lanes from the DT bindings - Drop unnecessary linux-media from MAINTAINERS entry - Drop unnecessary conversion to bool with !! - Don't set cap->bus_info in cfe_querycap() - Make debugfs files not readable by the world - Check the return value of v4l2_fwnode_endpoint_parse() - Remove the code dealing with remote_ep_fwnode. Instead use v4l2_create_fwnode_links_to_pad() and media_pad_remote_pad_unique() to create the link and get the pad index. - Add cfe/csi2/fe/dphy argument to the respective dbg/info/err print macros. - Drop some debug prints and add a few, clarifying the prints for enabling and disabling the streams. - Some cosmetic changes (linefeed, drop unnecessary assignment, move a define) - Link to v3: https://lore.kernel.org/r/20240815-rp1-cfe-v3-0-e15a979db327@ideasonboard.com Changes in v3: - Based on v6.11-rc3. The PiSP BE series is now in upstream so no extra dependencies are needed. - Fixed cfe_remove() return value, as the .remove hook has changed - Added Krzysztof's Rb. - Link to v2: https://lore.kernel.org/r/20240620-rp1-cfe-v2-0-b8b48fdba3b3@ideasonboard.com Changes in v2: - Change the compatible string back to raspberrypi,rp1-cfe from raspberrypi,rpi5-rp1-cfe - Drop the references to rp1 headers in the DT binding example. This allows compiling the example without the rp1 support. - Fix missing remap lines for mono formats - Fix csi2_pad_set_fmt() so that the format can be changed back to the sink's format from 16-bit or compressed format. - Link to v1: https://lore.kernel.org/r/20240318-rp1-cfe-v1-0-ac6d960ff22d@ideasonboard.com --- Tomi Valkeinen (4): media: uapi: Add meta formats for PiSP FE config and stats dt-bindings: media: Add bindings for raspberrypi,rp1-cfe media: raspberrypi: Add support for RP1-CFE media: admin-guide: Document the Raspberry Pi CFE (rp1-cfe) .../admin-guide/media/raspberrypi-rp1-cfe.dot | 27 + .../admin-guide/media/raspberrypi-rp1-cfe.rst | 78 + Documentation/admin-guide/media/v4l-drivers.rst | 1 + .../bindings/media/raspberrypi,rp1-cfe.yaml | 93 + .../userspace-api/media/v4l/meta-formats.rst | 1 + .../userspace-api/media/v4l/metafmt-pisp-fe.rst | 39 + MAINTAINERS | 7 + drivers/media/platform/raspberrypi/Kconfig | 1 + drivers/media/platform/raspberrypi/Makefile | 1 + drivers/media/platform/raspberrypi/rp1-cfe/Kconfig | 15 + .../media/platform/raspberrypi/rp1-cfe/Makefile | 6 + .../media/platform/raspberrypi/rp1-cfe/cfe-fmts.h | 332 +++ .../media/platform/raspberrypi/rp1-cfe/cfe-trace.h | 202 ++ drivers/media/platform/raspberrypi/rp1-cfe/cfe.c | 2504 ++++++++++++++++++++ drivers/media/platform/raspberrypi/rp1-cfe/cfe.h | 43 + drivers/media/platform/raspberrypi/rp1-cfe/csi2.c | 586 +++++ drivers/media/platform/raspberrypi/rp1-cfe/csi2.h | 89 + drivers/media/platform/raspberrypi/rp1-cfe/dphy.c | 181 ++ drivers/media/platform/raspberrypi/rp1-cfe/dphy.h | 27 + .../media/platform/raspberrypi/rp1-cfe/pisp-fe.c | 605 +++++ .../media/platform/raspberrypi/rp1-cfe/pisp-fe.h | 53 + drivers/media/v4l2-core/v4l2-ioctl.c | 2 + .../uapi/linux/media/raspberrypi/pisp_fe_config.h | 273 +++ .../linux/media/raspberrypi/pisp_fe_statistics.h | 64 + include/uapi/linux/videodev2.h | 2 + 25 files changed, 5232 insertions(+) --- base-commit: 98f7e32f20d28ec452afb208f9cffc08448a2652 change-id: 20240314-rp1-cfe-142b628b7214 Best regards,