Limit the auto-frame-rate feature to the original fps value the client
set, ignore other cases.
Change-Id: Ib8f374cc385da1cc4de464bc1d0520eb1d637b7e
Signed-off-by: Wenjie Yin <quic_wenjyin@quicinc.com>
- Moved DPB list from input to output
- change the log level of xo_reset control to high instead
of error log.
Change-Id: Ie3fd1f5fc846b08452d1c0d56bb065a11135e1e3
Signed-off-by: Rajathi S <quic_rajathis@quicinc.com>
Add comments to the error logs that are used
in RMA device screening.
Change-Id: I830e7116ceae535cf291aae8e137b38c9bc77fee
Signed-off-by: Manikanta Kanamarlapudi <quic_kmanikan@quicinc.com>
- Add ARES for Core clock.
- After controller power down, re-initiate controller/core power
on sequence and wait for QACCEPT and then power down the video
subsystem.
Change-Id: Icec985f33dc6e05adf71bfa40901120ceffea655
Signed-off-by: Manikanta Kanamarlapudi <quic_kmanikan@quicinc.com>
Update ssr type slection logic to msm_vidc_driver.c.
And add ssr type specific to target for kalama and
waipio.
Change-Id: Ifbdd47f3cc13f69cd93ecbfdbfefa20a112fca7e
Signed-off-by: Megha Byahatti <quic_mbyahatt@quicinc.com>
- Milos doesn't support video_xo_reset.
- Since its a common code added a condition to checkif vpu
version is iris2_2pipe and video_xo_reset is configured then
return with success.
Change-Id: I0baf3b1e89245065be4a4c21c55bc138077dd214
Signed-off-by: Rajathi S <quic_rajathis@quicinc.com>
- Add the basic driver support for volcano variant.
- Add platform specific files for volcano.
Change-Id: Iaa736415dc495912ed598a94fca736937015ac7b
Signed-off-by: Rajathi S <quic_rajathis@quicinc.com>
After drain command when there is inactive session,the
difference between last qbuf input and current time
was exceeding the MSM_VIDC_SESSION_INACTIVE_THRESHOLD_MS
which was leading to ddr voting 0(total_bw_ddr = 0).
When there are no video frames to process, ensure
minimum passive ddr voting for Tensilica(1000kbps).
Change-Id: I5ead2dcf95581b302fd711e4ec08962e876e09b9
Signed-off-by: Megha Byahatti <quic_mbyahatt@quicinc.com>
HFI_CMD_RESUME is not required on output port for encoder.
Add change to skip the same.
HFI Doc snippet for reference:
* There is no usage of resume command for encoder output port.
Change-Id: Ib2f7177f38de4b780c4627c7bdb03cbc61a63399
Signed-off-by: Vedang Nagar <quic_vnagar@quicinc.com>
- Allow SSR type according to the target with
defining SSR type in platform file.
Change-Id: Idca62ce34b66a885da388d798eddb0c6c00871e5
Signed-off-by: Megha Byahatti <quic_mbyahatt@quicinc.com>
[1] Acquire xo_reset mutex for gdsc access.
[2] Retry acquire_regulator for 1sec and BUG_ON if still fails.
Change-Id: Ie0ed4561a0c12b8526e2ede96a879f9d311aafab
Signed-off-by: Vedang Nagar <quic_vnagar@quicinc.com>
Problem:
First IPSC is received from firmware and driver sub state is set to
INPUT PAUSE. And driver state is INPUT_STREAMING at this point of time.
Now client calls for seek (or flush), which invokes streamoff and
stream on INPUT port. As a result, INPUT PAUSE sub state is cleared.
Now when client handles the first IPSC by calling stream on output,
internal buffers are not queued as INPUT pause is cleared with seek.
But for split mode, LINE buffer is mandatory to queue in FIRST_IPSC
streamon of the OUTPUT port
Solution:
Introduced a new sub state check of FIRST_IPSC, during stream on of
the INPUT port after FLUSH, we will check if FIRST_IPSC is pending,
then we will PAUSE the INPUT port, later in OUTPUT stream on,
we check for INPUT PAUSE and if it is set, we will queue the
internal buffers to firmware and resume the input port.
Change-Id: Ia9b566443bc9c4546247d31625384d89aff6985d
Signed-off-by: Roopangkumar Patel <quic_rooppate@quicinc.com>
AV1 Tile Info is sent by FW during EBD and these values
are being used by driver to compute clock votings.
For few(such as SEF) frames, there is no TILE info
present in the bitstream and FW will send default
(1, 1) value to driver which affects clock voting
for next frames and hence leading to performance issue.
To avoid such issue, if there is no tile infor, FW will
not send any such stale TILE info values.
But still during EBD, driver is doing memset to 0 for
inst->hfi_frame_info and hence overriding the value to 0.
Added fix to retain previous frames tile info by keeping
check for 0(since min value for tile info from FW = 1, 1)
if no value is being sent from the FW.
Change-Id: I6c23ed769f29efec2a7f318e3997d3f7532685c9
Signed-off-by: Vedang Nagar <quic_vnagar@quicinc.com>
Fix Alignment should match open parenthesis
Change-Id: I57689cff7d2251dac9a8b8e65ac126c324bf84f2
Signed-off-by: Anshul Agarwal <quic_anshagar@quicinc.com>
Fix Alignment should match open paranthesis.
Change-Id: I2b4c1a89fedd88df79ddb8c5711123f339303604
Signed-off-by: Anshul Agarwal <quic_anshagar@quicinc.com>
Fix warnings for vendor checker in vidc/inc folder.
[1] CHECK: Alignment should match open parenthesis.
[2] Lines should not end with a '('.
[3] Please use a blank line after.
Change-Id: Iac0591fc1e544e837a3c64c7a6e9b03cc9070ed2
Signed-off-by: Vedang Nagar <quic_vnagar@quicinc.com>
[1] Add the basic driver support for cliffs variant.
[2] Add platform specific files for cliffs.
Change-Id: If35f79fa1c72ce160d4ed4c0f1ba8babb7ce8a3d
Signed-off-by: Vedang Nagar <quic_vnagar@quicinc.com>
In vp9 splitmode cases, there is a possibility
fw reuses DPB buffers and hence does not return
these buffers as part of STOP command during
port reconfig. In this case, driver should not
destroy these DPB buffers. Added a fix for the
same.
Issue observed with: 1A1622D8_Dec_VP9_Argon_EOS0.
Change-Id: I81deded6a49290574ddda8d4f8efdfc0a9f733f3
Signed-off-by: Darshana Patil <quic_darshana@quicinc.com>
Signed-off-by: Maheshwar Ajja <quic_majja@quicinc.com>
Fix space and blank line required warning.
Change-Id: I2f754e52cdb36332a53b65ca248ae5d974a00ff4
Signed-off-by: Ankush Mitra <quic_ankumitr@quicinc.com>
[1] If request_fd enabled, msm_vb2_buf_queue() is not called
from v4l2_qbuf. Instead it's called as part of
msm_v4l2_request_queue(). Hence inst lock should be acquired
in common function i.e msm_vb2_buf_queue, to handle both
requests and non-request scenarios.
[2] If request_fd is disabled, inst_lock can be acquired in
v4l2_qbuf() call. Acquiring inst_lock from here will
ensure RO list insertion and updation i.e. attach/map will
happen under lock.
Currently, request_fd is disabled. Therefore, acquire inst_lock
from this function to ensure RO list insertion/updation is under
lock to avoid stability usecase.
Change-Id: I3cc1d4b8b5547fd5e34ce5eb06480380cb9200cc
Signed-off-by: Vedang Nagar <quic_vnagar@quicinc.com>
replace msm_vidc_vmem_alloc/free with direct calls to
vzalloc and vfree.
Change-Id: Iafde62dd7f54b99418207d10ff71589bf0efd5ed
Signed-off-by: Dikshita Agarwal <quic_dikshita@quicinc.com>
fix the error condition for pm_runtime_get_sync
as it can return 1 as well in success case.
Change-Id: I958da05b1b979adae059181432a3d917012640b7
Signed-off-by: Dikshita Agarwal <quic_dikshita@quicinc.com>
fix copyright markings and have all markings in one
multiline comment.
Change-Id: I481c6213bb25e8c278956ad6a51df3fc5c1aabf2
Signed-off-by: Dikshita Agarwal <quic_dikshita@quicinc.com>
