There is only one device mutex to lock vb2_queue when running
multi-instance encoding, it can be set by each encoder context.
[hverkuil: fix q_mutex documentation in the header]
Signed-off-by: Irui Wang <irui.wang@mediatek.com>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>
Encoder driver got iova from IOMMU is 34-bit, for example:
Here is the sample code:
encoder input frame buffer dma address is:
frm_buf =
vb2_dma_contig_plane_dma_addr(&vb2_v4l2_buffer->vb2_buf, 0);
the value of frm_buf is 0x1_ff30_0000.
encoder driver got the frm_buf and send the iova to SCP firmware
through SCP IPI message, then write to encoder hardware in SCP.
The iova is stored in IPI message as uint32_t data type, so the
value will be truncated from *0x1_ff30_0000* to *0xff30_0000*,
and then *0xff30_0000* will be written to encoder hardware, but
IOMMU will help to add the high *0x1_* bit back, so IOMMU can
translate the iova to PA correctly, encoder hardware can access
the correct memory for encoding.
Another reason to do this is the encoder hardware can't access
the 34-bit iova, IOMMU will help to add the remaining high bits
of iova. But for mt8188, encoder hardware can access 34-bit iova
directly, and encoder driver need write all 34 bits because
IOMMU can't help driver do this if the hardware support access
34-bit iova.
For the reasons above, this patch is added to support transfer
34-bit iova between kernel and SCP encoder driver. Use uint64_t
data type to store the iova, for compatibility with old chipsets,
add some new struct definitions for 34-bit.
[hverkuil: initialize wb and wb_34 to NULL]
Signed-off-by: Irui Wang <irui.wang@mediatek.com>
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>
This partially reverts commit fd9f8050e3,
and changes things so that the capability string fields of the encoder
conform to their requirements.
The driver name field should contain the actual driver name, not some
otherwise unused string macro from the driver. To make this clear,
copy the name from the driver's name field.
The card name for the video encoder previously held a static platform
name that was fixed to match MT8173. This obviously doesn't make sense
for newer chips. Since commit fd9f8050e3 ("media: mediatek: vcodec:
Change encoder v4l2 capability value"), this field was changed to hold
the driver's name, or "mtk-vcodec-dec". This doesn't make much sense
either, since this still doesn't reflect what chip this is.
Instead, fill in the card name with "MTxxxx video encoder" with the
proper chip number.
Since commit f2d8b6917f ("media: v4l: ioctl: Set bus_info in
v4l_querycap()"), the V4L2 core provides a default value for the
bus_info field for platform and PCI devices. This value will match
the default value for media devices added by commit cef699749f
("media: mc: Set bus_info in media_device_init()"). These defaults
are stable and device-specific.
Drop the custom capability bus_info from the mtk-vcodec encoder
driver, and use the defaults.
As this patch removes the last usage of MTK_VCODEC_DRV_NAME, remove
the macro as well.
Fixes: fd9f8050e3 ("media: mediatek: vcodec: Change encoder v4l2 capability value")
Signed-off-by: Chen-Yu Tsai <wenst@chromium.org>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>
Right now the decoder maintains two separate lists for supported pixel
formats and frame sizes. Getting the supported frame sizes for the
current set format is a bit convoluted, requiring a search through
the separate frame size list. The frame sizes are used to clamp and
align requested resolutions.
Instead, the frame size structure could be embedded inside the pixel
format structure. Getting one also gets the other. And since the
the driver already keeps pointers to the current set format, getting
the frame sizes becomes straightforward.
Do just that. Move v4l2_frmsize_stepwise inside mtk_video_fmt, and get
rid of mtk_codec_framesizes.
Signed-off-by: Chen-Yu Tsai <wenst@chromium.org>
Tested-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>
This partially reverts commit b018be06f3 ("media: mediatek: vcodec:
Read max resolution from dec_capability").
After the previous patches:
- media: mediatek: vcodec: decoder: Fix 4K frame size enumeration
- media: mediatek: vcodec: decoder: Skip alignment for default resolution
- media: mediatek: vcodec: decoder: Fix resolution clamping in TRY_FMT
the max_{width,height} fields in |struct mtk_vcodec_ctx| no longer have
any real users. Remove them.
Fixes: b018be06f3 ("media: mediatek: vcodec: Read max resolution from dec_capability")
Signed-off-by: Chen-Yu Tsai <wenst@chromium.org>
Tested-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>
In order to reduce decoder latency, enable H264 inner racing mode.
Send lat trans buffer information to core when trigger lat to work,
need not to wait until lat decode done.
Signed-off-by: Yunfei Dong <yunfei.dong@mediatek.com>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>
The capture queue format type may be differ depending on platform: for
stateless decoder drivers, we need to calculate the capture buffer size
according to the capture queue format type in SCP.
As a preparation for introducing drivers for stateless decoding, save the
current capture queue type on a per vcodec context basis.
Signed-off-by: Yunfei Dong <yunfei.dong@mediatek.com>
Reviewed-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>
Tested-by: Nícolas F. R. A. Prado <nfraprado@collabora.com>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>
For lat and core decode in parallel, need to get capture buffer
when core start to decode and put capture buffer to display
list when core decode done.
Signed-off-by: Yunfei Dong <yunfei.dong@mediatek.com>
Tested-by: Nícolas F. R. A. Prado <nfraprado@collabora.com>
Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>
As the end goal is to have platform drivers split by vendor,
rename mediatek/mtk-jpeg/ to mediatek/jpeg/.
Requested-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>
