The bitstream reader needs 512 bytes ready to read to examine the
headers in the first frame. If that frame is too small, prepend it
with a filler NAL.
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@s-opensource.com>

Keeping buffers filled into the bitstream on a temporary list instead of
immediately calling vb2_buffer_done on each of them immediately allows
start_streaming to correctly decide whether they should be marked as
done or requeued if an error occurs after the bitstream has been filled.
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@s-opensource.com>

The VDOA is able to transform the NV12 custom macroblock tiled format of
the CODA to YUYV format. If and only if the VDOA is available, the
driver can also provide YUYV support.

While the driver is configured to produce YUYV output, the CODA must be
configured to produce NV12 macroblock tiled frames and the VDOA must
transform the intermediate result into the final YUYV output.
Signed-off-by: NMichael Tretter <m.tretter@pengutronix.de>
Reviewed-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@s-opensource.com>

If the CODA driver is configured to produce NV12 output and the VDOA is
available, the VDOA can be used to transform the custom macroblock tiled
format to a raster-ordered format for scanout.

In this case, set the output format of the CODA to the custom macroblock
tiled format, disable the rotator, and use the VDOA to write to the v4l2
buffer. The VDOA is synchronized with the CODA to always un-tile the
frame that the CODA finished in the previous run.
Signed-off-by: NMichael Tretter <m.tretter@pengutronix.de>
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@s-opensource.com>

display_idx refers to the frame that will be returned in the next round.
The currently processed frame is ctx->display_idx and errors should be
reported for this frame.
Signed-off-by: NMichael Tretter <m.tretter@pengutronix.de>
Acked-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@s-opensource.com>

Rename dma_*_writecombine() to dma_*_wc(), so that the naming
is coherent across the various write-combining APIs. Keep the
old names for compatibility for a while, these can be removed
at a later time. A guard is left to enable backporting of the
rename, and later remove of the old mapping defines seemlessly.

Build tested successfully with allmodconfig.

The following Coccinelle SmPL patch was used for this simple
transformation:

@ rename_dma_alloc_writecombine @
expression dev, size, dma_addr, gfp;
@@

-dma_alloc_writecombine(dev, size, dma_addr, gfp)
+dma_alloc_wc(dev, size, dma_addr, gfp)

@ rename_dma_free_writecombine @
expression dev, size, cpu_addr, dma_addr;
@@

-dma_free_writecombine(dev, size, cpu_addr, dma_addr)
+dma_free_wc(dev, size, cpu_addr, dma_addr)

@ rename_dma_mmap_writecombine @
expression dev, vma, cpu_addr, dma_addr, size;
@@

-dma_mmap_writecombine(dev, vma, cpu_addr, dma_addr, size)
+dma_mmap_wc(dev, vma, cpu_addr, dma_addr, size)

We also keep the old names as compatibility helpers, and
guard against their definition to make backporting easier.

Generated-by: Coccinelle SmPL
Suggested-by: NIngo Molnar <mingo@kernel.org>
Signed-off-by: NLuis R. Rodriguez <mcgrof@suse.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: airlied@linux.ie
Cc: akpm@linux-foundation.org
Cc: benh@kernel.crashing.org
Cc: bhelgaas@google.com
Cc: bp@suse.de
Cc: dan.j.williams@intel.com
Cc: daniel.vetter@ffwll.ch
Cc: dhowells@redhat.com
Cc: julia.lawall@lip6.fr
Cc: konrad.wilk@oracle.com
Cc: linux-fbdev@vger.kernel.org
Cc: linux-pci@vger.kernel.org
Cc: luto@amacapital.net
Cc: mst@redhat.com
Cc: tomi.valkeinen@ti.com
Cc: toshi.kani@hp.com
Cc: vinod.koul@intel.com
Cc: xen-devel@lists.xensource.com
Link: http://lkml.kernel.org/r/1453516462-4844-1-git-send-email-mcgrof@do-not-panic.comSigned-off-by: NIngo Molnar <mingo@kernel.org>

During the recent vb2_buffer restructuring, the calculation of the
buffer payload reported to userspace was accidentally broken for the
first encoded frame, counting only the length of the headers.
This patch re-adds the length of the actual frame data.

Fixes: 2d700715 ("[media] media: videobuf2: Restructure vb2_buffer")
Reported-by: NMichael Olbrich <m.olbrich@pengutronix.de>
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Tested-by: NJan Luebbe <jlu@pengutronix.de>
Cc: <stable@vger.kernel.org>      # for v4.4 and up
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

coda_jpeg_check_buffer only cares about the buffer length and contents,
so change the parameter type back from v4l2_vb2_buffer to just the
vb2_buffer.
Instead of just checking the first and last bytes for the SOI and EOI
markers, relax the EOI marker check a bit and allow up to 32 trailing
bytes after the EOI marker as hardware generated JPEGs sometimes contain
some alignment overhead.
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NKamil Debski <k.debski@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

Move timestamp from struct vb2_v4l2_buffer to struct vb2_buffer
for common use, and change its type to u64 in order to handling
y2038 problem. This patch also includes all device drivers' changes related to
this restructuring.
Signed-off-by: NJunghak Sung <jh1009.sung@samsung.com>
Signed-off-by: NGeunyoung Kim <nenggun.kim@samsung.com>
Acked-by: NSeung-Woo Kim <sw0312.kim@samsung.com>
Acked-by: NInki Dae <inki.dae@samsung.com>
Signed-off-by: NHans Verkuil <hansverk@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

Remove v4l2 stuff - v4l2_buf, v4l2_plane - from struct vb2_buffer.

Add new member variables - bytesused, length, offset, userptr, fd,
data_offset - to struct vb2_plane in order to cover all information
of v4l2_plane.
struct vb2_plane {
        <snip>
        unsigned int            bytesused;
        unsigned int            length;
        union {
                unsigned int    offset;
                unsigned long   userptr;
                int             fd;
        } m;
        unsigned int            data_offset;
}

Replace v4l2_buf with new member variables - index, type, memory - which
are common fields for buffer management.
struct vb2_buffer {
        <snip>
        unsigned int            index;
        unsigned int            type;
        unsigned int            memory;
        unsigned int            num_planes;
        struct vb2_plane        planes[VIDEO_MAX_PLANES];
        <snip>
};

v4l2 specific fields - flags, field, timestamp, timecode,
sequence - are moved to vb2_v4l2_buffer in videobuf2-v4l2.c
struct vb2_v4l2_buffer {
        struct vb2_buffer       vb2_buf;

        __u32                   flags;
        __u32                   field;
        struct timeval          timestamp;
        struct v4l2_timecode    timecode;
        __u32                   sequence;
};
Signed-off-by: NJunghak Sung <jh1009.sung@samsung.com>
Signed-off-by: NGeunyoung Kim <nenggun.kim@samsung.com>
Acked-by: NSeung-Woo Kim <sw0312.kim@samsung.com>
Acked-by: NInki Dae <inki.dae@samsung.com>
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

Make videobuf2-v4l2 as a wrapper of videobuf2-core for v4l2-use.
And replace videobuf2-core.h with videobuf2-v4l2.h.
This renaming change should be accompanied by the modifications
of all device drivers that include videobuf2-core.h.
It can be done with just running this shell script.

replace()
{
str1=$1
str2=$2
dir=$3
for file in $(find $dir -name *.h -o -name *.c -o -name Makefile)
do
    echo $file
    sed "s/$str1/$str2/g" $file > $file.out
    mv $file.out $file
done
}

replace "videobuf2-core" "videobuf2-v4l2" "include/media/"
replace "videobuf2-core" "videobuf2-v4l2" "drivers/media/"
replace "videobuf2-core" "videobuf2-v4l2" "drivers/usb/gadget/"
replace "videobuf2-core" "videobuf2-v4l2" "drivers/staging/media/"
Signed-off-by: NJunghak Sung <jh1009.sung@samsung.com>
Signed-off-by: NGeunyoung Kim <nenggun.kim@samsung.com>
Acked-by: NSeung-Woo Kim <sw0312.kim@samsung.com>
Acked-by: NInki Dae <inki.dae@samsung.com>
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

The buffers with zero payload are now dumped in coda_fill_bitstream and not
passed to coda_bitstream_queue. This avoids unnecessary fifo addition and
buffer sequence counter increment.
Signed-off-by: NZahari Doychev <zahari.doychev@linux.com>
Acked-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

Storing internal frames in macroblock tiled order improves memory
access patterns by allowing increased burst sizes when transferring
the uncompressed macroblocks to or from main memory.
The translation logic only supports a single chroma base address,
so this is only supported for the chroma interleaved NV12 format.

Since the rotator used to copy the decoder output into the v4l2
capture buffers does not seem to support the tiled format correctly,
only enable it in the encoder for now.
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

The frame cache should be set up correctly to encode NV12 source frames.
This was not done before, so move the cache setup out of start_decoding
into its own function and call it from both start_encoding and
start_decoding.
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

The encoder allows to specify the VBV model reference decoder's initial
delay and buffer size. Export the corresponding V4L2 controls.
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

On i.MX6, sometimes after decoding a stream, encoding will produce macroblock
errors caused by missing 8-byte sequences in the output stream. Until the cause
for this is found, reset the hardware after sequence end, which seems to help.
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

Keep count of number of buffer meta structures in the list and use
a separate spinlock for operations on this counted list instead
of reusing the bitstream mutex in some places and none at all in
others.
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

Trace events with exactly the same parameters and trace output, such as
coda_enc_pic_run and coda_enc_pic_done, are supposed to use the
DECLARE_EVENT_CLASS and DEFINE_EVENT macros instead of duplicated
TRACE_EVENT macro calls.
This patch changes the order of parameters to coda_dec_rot_done and adds
a timestamp so it can share an event class with coda_bit_queue.
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

Allow coda_seq_end_work to be called multiple times, move the setting
of ctx->initialized from coda_start/stop_streaming() into
coda_start_encoding/decoding and coda_seq_end_work, respectively,
and skip the SEQ_END command in coda_seq_end_work if the context is
already deinitialized before.
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

In stream end mode the hardware will read the bitstream to its end,
overshooting the write pointer. Do not write additional data into
the bitstream in this mode.
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

The mvcol buffer is allocated at the end of the first internal buffer.
This patch fixes an out of bounds array access.
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

This is already done for one side of the comparison with the expectation
that the HW counter rolls over at the 16 bit boundary. This is true when
decoding a h.264 stream, but doesn't hold for at least MJPEG. As we don't
know the exact wrap-around point for this format just clamp the HW counter
to the same 16 bits. This should be enough to detect most of the errors
and saves us from doing different comparisons based on the decoded format.
Signed-off-by: NLucas Stach <l.stach@pengutronix.de>
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

Setting the last buffer flag causes the videobuf2 core to return -EPIPE from
DQBUF calls on the capture queue after the last buffer is dequeued.
This patch also fixes the EOS event to conform to the specification. It now is
sent right after the last buffer has been decoded instead of when the last
buffer is dequeued.
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NKamil Debski <k.debski@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

This patch adds tracepoints to the coda driver that can be used together
with the v4l2:v4l2_qbuf and v4l2:v4l2_dqbuf tracepoints to to follow video
frames through the mem2mem device.

For encoding with the BIT processor:
    coda:coda_enc_pic_run
    coda:coda_enc_pic_done

For decoding with the BIT processor:
    coda:coda_bit_queue
    coda:coda_dec_pic_run
    coda:coda_dec_pic_done
    coda:coda_dec_rot_done

Additionally, two low level tracepoints register whenever the BIT processor
is started and returns:
    coda:coda_bit_run
    coda:coda_bit_done
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NKamil Debski <k.debski@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

Issuing a cache flush for the whole bitstream buffer is not optimal in the first
place when only a part of it was written. But given that the buffer is mapped in
writecombine mode, it is not needed at all.
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Tested-by: NIan Molton <imolton@ad-holdings.co.uk>
Signed-off-by: NKamil Debski <k.debski@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

When queueing a buffer into the bitstream fails, it has to be requeued
in the videobuf2 queue before streaming starts, but while streaming it
should be returned to userspace with an error.
Signed-off-by: NPhilipp Zabel <philipp.zabel@gmail.com>
Signed-off-by: NKamil Debski <k.debski@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

coda_alloc_aux_buf already prints an error, no need to print duplicate
error messages all over the place.
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NKamil Debski <k.debski@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

The parameter buffer is a per-context buffer, so we can allocate and free it
together with the other context buffers during REQBUFS.
Since this was the last context buffer allocated in coda-common.c, we can now
move coda_alloc_context_buf into coda-bit.c.
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NKamil Debski <k.debski@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

Allocating the bitstream buffer only when the format is set allows to guarantee
that at least two frames fit into the bitstream buffer. For small frame sizes
a smaller bitstream buffer can be allocated. Since the bitstream buffer size now
depends on the format, replace CODA_MAX_FRAME_SIZE with ctx->bitstream.size
where appropriate and remove the now unused constant.
Since REQBUFS can be called multiple times, but the format can't be changed
unless REQBUFS 0 was called before, we can just keep the allocated context and
bitstream buffers if REQBUFS is called multiple times with a non-zero buffer
count.

[fixed a resource leak preventing repeatedly decoding]
Signed-off-by: NPeter Seiderer <ps.report@gmx.net>
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NKamil Debski <k.debski@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

Allocate the per-context buffers from REQBUFS instead in start_encoding or
start_decoding. This allows to stop and start streaming independently of
buffer (re)allocation
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NKamil Debski <k.debski@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

Make sure the buffer_mutex lock is taken in coda_bit_release
while coda_free_framebuffers and coda_free_context_buffers
are called.
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NKamil Debski <k.debski@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

On i.MX6, two subsampling ping-pong buffers are used for motion estimation and
deblocking They should not be counted as framebuffers, or they will be also used
to store reconstructed frames, causing visible artifacts in P-frames.
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NKamil Debski <k.debski@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

Compare rounded up width to fit into bytesperline.
Signed-off-by: NMarkus Pargmann <mpa@pengutronix.de>
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NKamil Debski <k.debski@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

To make sure a set RET_DEC_PIC_SUCCESS flag is not a leftover from
a previous successful run, clear it in prepare_decode.
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NKamil Debski <k.debski@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

If userspace doesn't properly separate the bitstream input into
individual frames (which may happen for example on slightly
corrupted streams) the CODA hardware may decode more frames
than we expect. We already log an error in this case, but it's
also necessary to adjust the sequence offset. Otherwise we
spam the log with a sequence number mismatch on every frame
frame after the unexpected one.
Signed-off-by: NLucas Stach <l.stach@pengutronix.de>
Signed-off-by: NKamil Debski <k.debski@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

Handle an empty buffer metadata list without crashing. This can happen
if the decoder is fed a broken stream, or multiple compressed frames in
a single queued buffer.
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NKamil Debski <k.debski@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

This patch adds a quick check for valid JPEG frames before feeding them into
the bitstream buffer: Frames that do not begin with the JPEG start of image
marker and end with the end of image marker are dropped.
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NKamil Debski <k.debski@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

For small frame sizes, allocating 1 MiB per compressed buffer is a waste of
space. On the other hand, incompressible 1080p data can produce JPEGs larger
than 1 MiB at higher quality settings. Allow userspace to set the compressed
buffer size and clamp the value to a sensible range.
Also set the initial sizeimage to a value inside the range allowed by try_fmt.
While at it, reduce the default image size to a maximum of 1920*1088 (otherwise
JPEG will default to 8k*8k and 96 MiB buffers).
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NKamil Debski <k.debski@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

With bitstream padding, it is possible to decode a single JPEG in the bitstream
immediately. This allows us to only ever queue a single JPEG into the bitstream
buffer, except to increase payload over 512 bytes or to back out of hold state.
This is a measure to decrease JPEG decoder latency.
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NKamil Debski <k.debski@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>

Before starting a PIC_RUN, pad the bitstream with 0xff until 256 bytes
past the next multiple of 256 bytes, if the buffer to be decoded is the
last buffer in the bitstream.
Signed-off-by: NPhilipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: NKamil Debski <k.debski@samsung.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@osg.samsung.com>