add memory ops support so that different
implementations can exist for memory_alloc,
memory_free, memory_map and memory_unmap,
dma buf attach, detach, map, unmap and
get buffer region in upstream and downstream
drivers.
Change-Id: Ifabc34e7a8b0284579c1bc4a8f477fe558d068f4
Signed-off-by: Darshana Patil <quic_darshana@quicinc.com>
Fix compilation issues due to wrong format specifiers being
used for printing device address.
Change-Id: Ic8ee8e9cdb563ea3efea9fa2e713befad74447cb
Signed-off-by: Dikshita Agarwal <quic_dikshita@quicinc.com>
When client queued a buffer, v4l2 framework
compares the client Dmabuf with the existing
dmabuf in the bufs array at the same index.
If the dmabuf is different then v4l2 framework
would call unmap_dmabuf, detach_dmabuf callbacks
for the old buffer and then calls attach_dmabuf,
map_dmabuf callbacks for the new buffer before
qbuf callback is made to driver. If the dmabuf
is same then qbuf callback will be made directly.
V4l2 framework expects that clients recycles the
same buffers in the same indices so that when
v4l2 framework compares dmabuf it won't unmap
or map buffers.
If client recycle the buffers in different indices
still usecase works but there will be unnecessary
unmap, map overhead will be incurred.
If client does not recycle the buffers at all,
meaning every time client queues new buffers
then special handling is required for decoder
output buffers which are used for reference by
video hardware.If any buffers are used by video
hardware as reference buffers then firmware returns
those buffers with READ_ONLY flag and they are kept
in read_only list and when read_only grows beyond
some limit then driver would ask firmware to release
those READ_ONLY buffers and when firmware responds
driver would delete these READ_ONLY buffers.
Change-Id: I8a0b11c986dd0b9464895498efd8c1831a754198
Signed-off-by: Darshana Patil <quic_darshana@quicinc.com>
Added change to print etb, ebd, ftb, fbd,
(ebd-etb), (fbd - etb), (etb-ftb) wallclock
time in ms.
It will help to debug perf issues.
Change-Id: Ideaecb2d51763711d6628729aa0f9056d05f0d51
Signed-off-by: Govindaraj Rajagopal <quic_grajagop@quicinc.com>
Move mapping of v4l2 to vidc macros for codec, color format,
color primaries etc to target specific file to restirct the
usage of private v4l2 defines to target specific file.
Change-Id: Ie83968cb944c756e55b6514c496c0b23c5d16413
Signed-off-by: Dikshita Agarwal <quic_dikshita@quicinc.com>
- hierarchical state model which comprises of
top level states and sub states
- top level states include OPEN, CLOSE, ERROR,
INPUT_STREAMING, OUTPUT_STREAMING, STREAMING.
- sub states include DRAIN, DRC, DRAIN_LAST_BUFFER,
DRC_LAST_BUFFER, INPUT_PAUSE, OUTPUT_PAUSE.
- follows deterministic approach for key sequences
like DRC, DRAIN and last flag handling i.e none
of these events are ignored or discarded.
- removal of auto start of firmware input port as
part of ipsc handling.
- introduction of hfi commands for host controlled
PAUSE and RESUME of firmware ports.
- introduction of information last flag packets
for DRC and DRAIN when FW has insufficient
output buffers.
Change-Id: Ie8c8f44af464d06f5a7bb76822f749c9874f869a
Signed-off-by: Darshana Patil <quic_darshana@quicinc.com>
Sometimes uninitialized local pointer variables were passed to
msm_vidc_vmem_alloc() call for allocating memory. Uninitialized
variables might garbage value, so msm_vidc_vmem_alloc() is
treating that double alloc request and returning error. So i.e
leading to undefined behaviour.
For e.x, msm_vidc_update_input_cr() call will never add any
entries into &inst->enc_input_crs(due to above mentioned issue),
So i.e leading to populate invalid input compression ratio in
encoder usecase.
Change-Id: I4507b343bee8eec7252cf946ad8d3120efd7bacb
Signed-off-by: Govindaraj Rajagopal <quic_grajagop@quicinc.com>
kzalloc might fail to allocate physically contiguous
memory which may not be available always and hence use
vzalloc to allocate virtually contiguous memory
Change-Id: I8ad89ef107da301217f3d1ec307b9b27b4edd2de
Signed-off-by: Deepa Guthyappa Madivalara <quic_dmadival@quicinc.com>
Acquire client lock for every entry point
in driver. This is needed especially for
streamoff and close calls where inst lock
is released while driver is waiting for
stop_done and close_done responses from fw.
When inst lock is released, client can send
a new ioctl call and driver state machine
can get updated leading to unexpected behaviour.
Change-Id: Ibb62c3431c88e2f1d9088d54fe7c9e53b1c284fd
Signed-off-by: Darshana Patil <quic_darshana@quicinc.com>
v4l2 compliance fixes for test formats
fixes:
test VIDIOC_ENUM_FMT/FRAMESIZES/FRAMEINTERVALS: OK
test VIDIOC_G_FMT: OK
test VIDIOC_TRY_FMT: OK
test VIDIOC_S_FMT: OK
Change-Id: I23c193ab5785bb10d070fa3085678d9669f53a92
Signed-off-by: Darshana Patil <quic_darshana@quicinc.com>
Use client input buffer queuing rate for power
calculations for non-realtime sessions
Change-Id: I498422ba1ae9ae96b782a2096dc0cd0aa7b05798
Signed-off-by: Deepa Guthyappa Madivalara <quic_dmadival@quicinc.com>
Video-banner contains compilation timestamp, which will
be helpful in stability debugging to find mismatch b/w
ramdump and symbols.
Change-Id: Ibc446fc4398a184b840867c568ed9e8940e490ae
Signed-off-by: Govindaraj Rajagopal <quic_grajagop@quicinc.com>
Add support for AV1D Intra-Block Copy (IBC) feature. This requires a
dedicated internal buffer (HFI_BUFFER_IBC_AV1D), which was previously
part of LINE buffer. A new internal buffer type called
HFI_BUFFER_PARTIAL_DATA is introduced, and this buffer type holds
HFI_BUFFER_IBC_AV1D for AV1D. HFI_BUFFER_PARTIAL_DATA is mapped to
NON_SECURE_PIXEL context bank for non-secure and to SECURE_PIXEL
context bank for secure use case.
Change-Id: If7ae22495f9981f275d54acc342c25ccdfc0c7b9
Signed-off-by: Mihir Ganu <quic_mganu@quicinc.com>
If client queues input buffers before streamon of
input port, these buffers are not actually queued
to driver but instead held in v4l2 framework. They
are queued to driver when input port is streamed on.
In this scenario when input port is streamed on, inst
lock is acquired and as part of this call, vb2 startes
queueing buffers to driver. During this enqueuing,
buf_queue also tries to acquire inst lock which leads
to deadlock. Hence fixed this issue by moving inst lock
acquiring/releasing to start_streaming call for streamon
and stop_streamping call for streamoff.
Change-Id: I67ed28b4f270ea899c4ace88a368148848b31072
Signed-off-by: Darshana Patil <quic_darshana@quicinc.com>
- add support for prepare_buf
- check if client has configured invalid fd for metadata
buffer fd control.
Change-Id: I576d02be05e4692a5ab45286895c34284f08651c
Signed-off-by: Darshana Patil <darshana@quicinc.com>
When client queues buffer via requests using
ioctl MEDIA_REQUEST_IOC_QUEUE, vidioc_qbuf
callback is not called to driver but instead
buf_queue callback is called directly.
Video driver is acquiring lock in vidioc_qbuf
but not in buf_queue and hence acquiring lock
was missing for requests which is leading to
errors. Moved lock from vidioc_qbuf to buf_queue
callback to resolve the issue.
Change-Id: I76e3fcf07b7b24218717075f2c7d4a688ad18e86
Signed-off-by: Darshana Patil <quic_darshana@quicinc.com>
Decouple fence structure from video buffer. Also add some
misc fence flow setup support.
Change-Id: Ia4e223a3a288f0dae3238850685f3a73d4b9400a
Signed-off-by: Akshata Sahukar <quic_asahukar@quicinc.com>
Use v4l2_m2m_ioctl_streamon helper function for input and output ports
Change-Id: I20375259321cc5925ccaf7f8eab95a94f55d2821
Signed-off-by: Deepa Guthyappa Madivalara <quic_dmadival@quicinc.com>
packetize dynamic controls set via request api along
with corresponding input buffer under the same hfi header
and queue to FW.
Change-Id: I8bf82f7ccb1ab16c42ce02189a208f5f7cb8dc70
Signed-off-by: Darshana Patil <quic_darshana@quicinc.com>
Add fence support for decoder output buffers for
the consumer to use it to improve latency.
Change-Id: I7384b4a9793248988a2d2d546a535f26636d5bb3
Signed-off-by: Maheshwar Ajja <quic_majja@quicinc.com>
[1] Remove CAP_FLAG_ROOT flag. Root willbe identified
with has_parents() api.
[2] Replace field(struct msm_vidc_inst_cap) cap -> cap_id.
[3] Rename below pool apis.
msm_memory_alloc -> msm_memory_pool_alloc
msm_memory_free -> msm_memory_pool_free
[4] Remove struct msm_vidc_inst_cap_entry usage in inst.
inst->firmware.list -> inst->firmware_list
inst->children.list -> inst->children_list
[5] Organize msm_vidc_inst_capability_type enums in such
a way that all parents placed first then children. It
will save CPU cycles while preparing caps_list.
Change-Id: I376f4e6ddec37ce326c393fba133e19ba70bab09
Signed-off-by: Govindaraj Rajagopal <quic_grajagop@quicinc.com>
Convert multi-parent, multi-child based dependency tree
into flattened list. It is organized in such a way that
all parents willbe present before self or any child node.
For some caps, adjust sequence is called without preparing
parent nodes. So that is leading to undefined behaviour.
Added change to address these problems.
Change-Id: Ib28cae8a4ed7e019c80ecfc06519ef78ffa4dc08
Signed-off-by: Govindaraj Rajagopal <quic_grajagop@quicinc.com>
Add handling for VIDIOC_CREATE_BUFS ioctl in driver.
This ioctl is used to create buffers for memory mapped
or DMA buffer I/O.
It can be used as an alternative or in addition to the
ioctl VIDIOC_REQBUFS ioctl, when a tighter control over
buffers is required.
This ioctl can be called multiple times to create buffers
of different sizes.
Partially Fixes: v4l2-compliance:
test VIDIOC_REQBUFS/CREATE_BUFS/QUERYBUF.
Change-Id: I9dd6c3f48e48e297b3e56fd4c8062869c8b55bee
Signed-off-by: Dikshita Agarwal <quic_dikshita@quicinc.com>
Add handling for VIDIOC_QUERYBUF ioctl in driver.
This can be used to query the status of a buffer
at any time after buffers have been allocated with
the VIDIOC_REQBUFS ioctl.
Partially Fixes: v4l2-compliance:
test VIDIOC_REQBUFS/CREATE_BUFS/QUERYBUF.
Change-Id: Ifd014bd8a8b02ce9fa15c62d7583bc8bcfb6756a
Signed-off-by: Dikshita Agarwal <quic_dikshita@quicinc.com>
Invoke correct APIs to handle different types of events.
Fixes: v4l2-compliance:
test VIDIOC_(UN)SUBSCRIBE_EVENT/DQEVENT.
Change-Id: I442fae3111d65a9612c73f2d16f9ed8031c77ec0
Signed-off-by: Dikshita Agarwal <quic_dikshita@quicinc.com>
- Request api allows to synchronize controls along
with buffers.
- Client allocates a request fd. This request fd is
associated with control and set using s_ext_ctrls.
Such controls are stored by the framework and not
applied immediately.
- Request fd is also associated with input qbuf and
queued to driver. Following this actual request is
queued. Buffers do not enter the queue untill the
request is queued.
- As part of qbuf, driver call request_setup() which
inturn sets the control to driver.
Change-Id: I316c4dc2a0091b21059230be5ee5fd6379259456
Signed-off-by: Darshana Patil <quic_darshana@quicinc.com>
Added changes to call do msm_vidc_core_init() inside
msm_vidc_probe_video_device() itself. It will uniform
core init and deinit as part of insmod and rmmod.
follow core_init and core_init_wait for all new session
open sequence.
Change-Id: I081d5ff3632296b5d2086992e6f3388daedfd1d7
Signed-off-by: Govindaraj Rajagopal <quic_grajagop@quicinc.com>
Added support for timestamp reorder. It is default
disabled.
Change-Id: I51a6fc785860081b651fc67f443132cb2675b9fd
Signed-off-by: Govindaraj Rajagopal <quic_grajagop@quicinc.com>
Both &inst->stats_work and &inst->response_work both were using
&inst->response_workq for scheduling works. From reverse thread
handle_session_response_work_handler incremented refcount via
get_inst_ref(), at the same time driverfd got closed by HAL.
So strongref is held by handle_session_response_work_handler and
as part of put_inst(refcount is zero), will attempt to destry
&inst->response_workq inside msm_vidc_close_helper.
destroy_workqueue() call will wait until all already fired events
to complete, which will never happen in this case i.e leading to
deadlock situation. It is an attempt to self destruction, which is
not allowed.
Added changes to avoid above mentioned situation.
Change-Id: Ieb26a4b02f87f2bb738027cfd16e4c261b756dc3
Signed-off-by: Govindaraj Rajagopal <quic_grajagop@quicinc.com>
-Removed unwanted functions
-Removed unwanted part of code from function definitions
-Added proper format specifier according to the data type.
Change-Id: I700202f1cf588506b329d202ce3e2729c027dcae
Signed-off-by: Gaviraju Doddabettahalli Bettegowda <quic_gdoddabe@quicinc.com>
Signed-off-by: Govindaraj Rajagopal <quic_grajagop@quicinc.com>
qbuf call for actual bitstream and yuv buffer 'struct v4l2_buffer'
length field indicates numplanes and for metadata buffers, length field
indicates capacity. So always check length field before accessing/
dereferencing b->m.planes or b->m.fd.
Change-Id: I409b28e0a66bd5c031b2c98c6d7614da99164a6c
Signed-off-by: Govindaraj Rajagopal <quic_grajagop@quicinc.com>
Decoder needs to support VIDIOC_TRY_DECODER_CMD and
Encoder needs to be support VIDIOC_TRY_ENCODER_CMD.
Partially Fixes: V4l2-complience:
testDecoder(VIDIOC_(TRY)_DECODER_CMD) for decoder.
testEncoder(VIDIOC_(TRY)_ENCODER_CMD) for encoder.
Change-Id: Ia68587412ed316f0c871397da83d6c56665cfbb5
Signed-off-by: Dikshita Agarwal <quic_dikshita@quicinc.com>
- Do not clear V4L2_FL_USES_V4L2_FH flag because this
flag indicates that file->private_data points to
&struct v4l2_fh.
- Initialize v4l2 ctrl handler with driver ctrl handler.
- Add ctrl ops for g_volatile_ctrl to return updated
value of volatile ctrls from driver.
Fixes: v4l2-compliance:
testPrio (VIDIOC_G/S_PRIORITY).
testQueryControls (VIDIOC_QUERYCTRL).
Change-Id: Ibe4b6fa67d2c02a5afeb627f2060e4761c5ef717
Signed-off-by: Dikshita Agarwal <quic_dikshita@quicinc.com>
Use maximum framerate from published
instance caps to make it platform
agnostic.
Change-Id: I319768e722774969287abb9e782ddbece7593ff9
Signed-off-by: Priyanka Gujjula <quic_pgujjula@quicinc.com>
There are 2 possible race issues with msm_vidc_stats_handler.
[1] msm_vidc_close acquired inst->lock and called cancel_delayed_work,
by that time stats_handler already fired and incremented inst->kref
via get_inst_ref. Close sequence releases the lock and called
put_inst().
So now inst strong ref is held by stats_handler, which will acquire
inst->lock and schedules new stats_work and does put_inst. inst->kref
count reaches zero and it will free inst struct(using close_helper).
So that will lead to use-after-free crash at core kernel side.
[2] msm_vidc_close acquired inst->lock and called cancel_delayed_work,
by that time stats_handler is scheduled. So process_one_work() from
workqueue is ready to call stats_handler api. But before it invokes
stats handler(context switch), msm_vidc_close sequence continued to
run and completed the close sequence and called put_inst. So inst
struct got freed up(because inst->kref count reached to zero). Now if
core kernel(workqueue) attempts to invoke stats_handler by calling
worker->current_func(work), will lead to again use-after-free crash.
Added changes to avoid above mentioned issues.
Change-Id: I55bc33a753f4dbae4a8cdc6373b5465d183da3bc
Signed-off-by: Govindaraj Rajagopal <grajagop@codeaurora.org>
