Serialize core_info_read with lock so that multiple concurrent
threads do not cause the write to overflow. Also have the bound
check to avoid overflow in write_str function.
CRs-Fixed: 2013361
Change-Id: Ia18a4b94cafd69af1d367861f2499fc202f18e9f
Signed-off-by: Abdulla Anam <abdullahanam@codeaurora.org>
Signed-off-by: Sanjay Singh <sisanj@codeaurora.org>
Hide kernel pointers from unprivileged ussers by using %pK format-
specifier instead of %p. This respects the kptr_restrict sysctl
setting which is by default on. So by default %pK will print zeroes
as address. echo 1 to kptr_restrict to print proper kernel addresses.
CRs-Fixed: 987018
Change-Id: I4772257a557c6730ecc0624cbc8e5614e893e9fd
Signed-off-by: Sanjay Singh <sisanj@codeaurora.org>
qos_request variable should be instance variable rather than global as
we want to raise the request for each instance and remove it on closing
of instance. Also modify the latency value to 332 from 1000 ms as per
system team recommendation.
Change-Id: I28138d0abcc3a724d7c116a12443fee6898ecb01
Signed-off-by: Deepak Kushwah <dkushwah@codeaurora.org>
CPU L2 cache is being power collapsed and resumed during
video sessions which is consuming more power unnecessarily.
So disable CPU L2 cache power collapse during video sessions
using qos requests to save power.
Change-Id: Iaf47231bed1304510ea221b9d50866d3b4adfed0
Signed-off-by: Maheshwar Ajja <majja@codeaurora.org>
Video hardware on msm8956/76 version v1 chipset supports
VP9 decoder with a hardware limitation which requires
more voltage for a given video clock rate. So read
hardware version from efuse register and do necessary
voltage scaling by defining new VP9 decoder clock vs
voltage table in dtsi and use it when VP9 decoder
session is running.
Change-Id: I2824d64aa1abd7f2d3c1853d6f23f2d172bf3929
Signed-off-by: Maheshwar Ajja <majja@codeaurora.org>
.compat_ioctl32 is used to support custom ioctls for 64 bit kernel
and 32 bit userspace. Since vidc driver doesn't have any custom
controls, remove the definition.
When userspace calls some fuzzy ioctls, V4L2 framework calls
compat_ioctl32 to corresponding driver from v4l2_compat_ioctl32.
If vidc driver defines compat_ioctl32 to be v4l2_compat_ioctl32
then v4l2_compat_ioctl32 function gets calls infinitely.
Hence remove the definition so that V4L2 framework returns failure
to ioctl caller.
Change-Id: I8d0e7afb14c4fd16e877145c603283cb936b82e2
Signed-off-by: Praneeth Paladugu <ppaladug@codeaurora.org>
Free the registered buffers when closing instead of stopping
session to avoid NULL pointer access error if client tries to
dequeue buffer while stopping session.
Change-Id: Iecce727fe7f5afc0287e38904beb5855a6e11e02
Signed-off-by: Maheshwar Ajja <majja@codeaurora.org>
Currently video driver will power down the video hardware
if video hardware is in IDLE state for 2 seconds. This is
causing CTS testcase failure if power down sequence executed
within 5 seconds from the start of the video playback.
GDSC is expected to be turned off when power down sequence
executed from video driver but it is not because PIL driver
has 5 seconds timeout value to remove GDSC vote from PIL side.
Hence GDSC is not being turned off within 5 seconds from the
video start even though video driver executed power down sequence.
Increase the timeout value to 10 seconds from 2 seconds in
video driver to power down the video hardware which ensures the
PIL driver has removed its GDSC vote and resolve the CTS failure.
Change-Id: I5b79c793595ce7e19666a88cd39f0a08675e9382
Signed-off-by: Maheshwar Ajja <majja@codeaurora.org>
Load firmware early during probe, only if early-fw-load entry
is set to true in video dtsi. Also fix a race condition issue
related to ocmem allocation during early firmware download.
Video driver calls suspend immediatly after sys init done response
is received from firmware. The suspend call leads to triggering
power collapse immediately. During power collapse, venus hfi
detects there is no ocmem to unset. The ocmem gets allocated only
after the first power collapse is triggered. So the first
power collapse fails. In mean time ocmem gets allocated and
driver goes in to power collapse after two seconds. By removing
suspend call during early firmware download, we prevent the race
condition.
Change-Id: I33b0155faf14c11477a8c205c714ae429dc54550
Signed-off-by: Arun Menon <avmenon@codeaurora.org>
This change will load the firmware during v4l2 video driver
probe. If there are no active video sessions, then video
hardware will be power collapsed. On targets which can afford
to keep firmware always loaded in memory, this change will help
overcome the delay associated with loading firmware when a new
video session is instantiated.
Change-Id: I4f9981d8db4ff62d652628681bb58fa79ee9b7d4
Signed-off-by: Arun Menon <avmenon@codeaurora.org>
Add checks for IOMMU presence around specific operations.
Set VBIF registers that are required when IOMMU is not
present. Use existing resources from vidc driver and avoid
duplication.
Change-Id: Id23a7a086ffe3ef8aa702560531c45b287b0be08
Signed-off-by: Sachin Shah <sachins@codeaurora.org>
Thermal agent from userspace reports thermal level
value through sysfs node. Based on thermal levels,
video driver takes below action to avoid any further
overheating.
1. Normal thermal level(T0) is ignored.
2. Low(T1) or high(T2) thermal level is ignored for
usecase running in nominal/svs clock whereas all
video sessions are terminated for turbo usecase.
3. During critical(T3) thermal level, all video
sessions are terminated.
Change-Id: I516761fc51e824b35957a85009de7f08b15d887c
Signed-off-by: Vikash Garodia <vgarodia@codeaurora.org>
Signed-off-by: Maheshwar Ajja <majja@codeaurora.org>
This change flushes any queued work related to power collapse,
once the device is power suspended.
There are applications which do not destroy video instance
on suspending the device. For such applications, if the device
is power suspended, video driver does not prepare venus for
power collapse.
Considerable power saving is observed by power collapsing
venus before the device goes in suspend state.
Change-Id: I11252e45b10d0d3d3eefb38994acd083c847bbb8
Signed-off-by: Vikash Garodia <vgarodia@codeaurora.org>
Video driver loads and unloads firmware at the start
and end of each session. Firmware download takes up
to 100 msec. When there are many b2b video sessions
e.g. thumbnail generation, this is an unnecessary
overhead. To overcome this problem, video driver will
unload the firmware 10 secs after the session is closed.
If a new session starts during this period, it will not
be required to download the firmware again.
Change-Id: Ia515c3ed89c75ea9e43852958d1cbb2d77fe4c51
Signed-off-by: Arun Menon <avmenon@codeaurora.org>
Add trace events for ETBs/FTBs, buffer allocation and mappings,
fw load times etc.
Change-Id: Ie57db1ddbad886e78f8d0a6e5eaac06206a6c53d
Signed-off-by: Shalaj Jain <shalajj@codeaurora.org>
The core->instances list structure is guarded by
mutex core->lock, whenever a new video instance is added
or deleted from the list. However this list was accessed
from certain functions without acquiring the lock, which
resulted in race condition, causing the device to crash.
Also clean up the core locking mechanism.
Change-Id: I6f431bc381d54948611783e19fac853c83a87203
Signed-off-by: Arun Menon <avmenon@codeaurora.org>
In cases where IS_ERR_OR_NULL() was true, the driver was forming the
return via PTR_ERR. However if the OR_NULL part happens to be true,
PTR_ERR(NULL) would be 0, which would generally mean success. So, if
PTR_ERR returns 0, assign a valid error code to rc.
Change-Id: I363c69963040074e7b97d85914ea168b8eba3ab3
Signed-off-by: Deva Ramasubramanian <dramasub@codeaurora.org>
Remove using and parsing properties from board files.
Change-Id: Ife51e04ea5bb8857cce1220eb03ffd5237ec44a9
Signed-off-by: Shalaj Jain <shalajj@codeaurora.org>
Header files are no longer allowed in the directory
arch/arm/mach-msm/include/mach/ .
Move the iommu related header files to a more suitable place.
Change-Id: Ib7bbce1485d6185f669935b507040cac75368985
Signed-off-by: Larry Bassel <lbassel@codeaurora.org>
Compat IOCTLs are used when Userspace operates in 32 bit
and kernel in 64 bit. The IOCTLs in V4L2 framework needs to
map to compat functions. Add support in driver to call compat
V4L2 framework IOCTLs.
Change-Id: I3f92e40d314e7717535193806259cef4eef99827
Signed-off-by: Praneeth Paladugu <ppaladug@codeaurora.org>
Without NULL termination, of_dev_lookup() can run
off the end of the array and looking at invalid data.
Change-Id: Ib7d8b3afdc76d3f196d620d47eb77d4e0bc74d53
CRs-fixed: 582111
Signed-off-by: Sachin Shah <sachins@codeaurora.org>
This change will ensure that vidc clients specify the
core id as venus or q6 to open a driver session.
Without this change, sometimes due to driver probe deferral,
core_id_0 would be taken by q6 and vidc clients intending to open a
venus decode session ended up opening a q6 decode session.
Change-Id: I4f9c862b8057999e4db7ca75be493138737a58a8
Signed-off-by: Arun Menon <avmenon@codeaurora.org>
This reverts commit a63cf09355. The commit
was a temporary hack to work around power collapse issues that have
since been fixed.
Change-Id: I74be8a1b163e814592056ca5556ace78a05b55f2
Signed-off-by: Deva Ramasubramanian <dramasub@codeaurora.org>
v4l2 video driver probe needs to defer till
iommu domains probe has been called. Without this
change, v4l2 video driver probe will fail if
iommu domains probe is not invoked prior to it.
Change-Id: Idbc1a4c90e558952f41f51ec96fdd3e7974490cd
Signed-off-by: Arun Menon <avmenon@codeaurora.org>
The iommu_domain code isn't really MSM specific and is better suited
to live in the iommu directory. Move it accordingly.
Change-Id: Ie88b4aba6901722166bb180275d352b745821772
Signed-off-by: Laura Abbott <lauraa@codeaurora.org>
Temporarily increase power collapse delay to
a high number, in effect disabling it, to work around
stability issues.
Change-Id: I65d71c96081fa5eddffc5e91db2c65167dec59c7
Signed-off-by: Arun Menon <avmenon@codeaurora.org>
When parsing the device tree the driver emits false error messages when
optional properties are not found. This commit gracefully handles the
absence of optional properties.
Change-Id: I34c58513e8fa62e6862b8c630814930554ff18fb
Signed-off-by: Deva Ramasubramanian <dramasub@codeaurora.org>
On chipsets where there can be multiple video cores, when setting up the
debugfs entries for the secondary cores, several debugfs_create_files
often fail because the driver attempts to create new entries with the same
file names.
This commit moves out a lot of driver specific entries to a new function
(msm_vidc_debugfs_init_drv) that's created during driver init and during
driver probe which might be called multiple times.
Change-Id: I9fee95acbe07244c558d6ede8539dfb67e2597dc
Signed-off-by: Deva Ramasubramanian <dramasub@codeaurora.org>
Increase power collase timeout to avoid venus power
collapse before the video video/encoding decoding is
actually started. This is needed to avoid iommu map
failures at the beginning of the playback.
Change-Id: I05cade4f6464122d7f3c3b42b4976b9b7a33827f
CRs-fixed: 563706
Signed-off-by: Vinay Kalia <vkalia@codeaurora.org>
Dynamic buffer mode support will require buffer
handling support in vidc layer, rather than v4l2
layer.
Change-Id: I4732a09afc6b9dc279d7dbf28e362bf400e09cc9
Signed-off-by: Arun Menon <avmenon@codeaurora.org>
For Hybrid hevc codec, the output buffer contains
intermediate data. In case of cached buffers,
these intermediate buffers need to be invalidated before
they are sent to the dsp.
Change-Id: I08916b73ee7cbad8fc4de7831842056434d0e251
Signed-off-by: Arun Menon <avmenon@codeaurora.org>
Provide debugfs node for HW response timeout and a sysfs device
attribute for power collapse delay.
Change-Id: I586f10f3e90df5384c8e465fbfe8e7ebec0915eb
Signed-off-by: Sachin Shah <sachins@codeaurora.org>
Migrate to kernel 3.10 which has contains minor changes in the
V4L2 interface:
- Driver needs to specify how it provides timestamps
- Driver needs to declare itself as a mem2mem driver so that
certain ioctls are not masked out before reaching the driver.
- Vendor specific buffer flags have been changed as to not
conflict with standard flags
- Controls that are of boolean type need to explicitly set the
step to 1.
Change-Id: I1a38471d9ee0c74abe39ae7ce1a0f2912195db30
Signed-off-by: Deva Ramasubramanian <dramasub@codeaurora.org>
Unregister input buffers when reqbufs is called for input
port with zero buffer count.
Change-Id: I676fa0857d675ae45f4328b26c2f82d7397315b1
Signed-off-by: Surajit Podder <spodder@codeaurora.org>
Signed-off-by: Praveen Chavan <pchavan@codeaurora.org>
Due to how HAL buffer types are tied into the secure hints, switch up
the HAL buffer types if the current session is of type encoder.
Change-Id: Idb74340ffad25278dd926c869745e235b4821ae3
Signed-off-by: Deva Ramasubramanian <dramasub@codeaurora.org>
Keep v4l2 code cleaner by moving parsing logic to separate file.
Change-Id: I65999400ad4391caa396a2bdcc8494db1727bed9
Signed-off-by: Eric Ho <yinlokh@codeaurora.org>
Add support for different node values for simultaneous
venus and q6 usecase. This is needed to support HEVC
decode on msm8974, where venus and q6 both will be
present. Also create "link_name" device attribute
to provide a meaningful name to the video device
instead of video32, video33 etc. Userspace reads
this "link_name" sysfs node and creates appropiate
dev node symbolic link.
Change-Id: Iab14cad2c43b78eb7553ac78521611727597a42f
Signed-off-by: Sachin Shah <sachins@codeaurora.org>
Add necessary check to confirm absence of IOMMU and fallback
to physical memory address when IOMMU is absent.
Change-Id: I6957ce93392f67161f48371d5fc5ac74773886d8
Signed-off-by: Sachin Shah <sachins@codeaurora.org>
Read the "qcom,max-load" parameter from the device tree to determine the
max support load for each target. This allows us to do away with
the hardcoded max load with in the driver.
Change-Id: I9e874624b9209b53c7bd6d0524dd2c1314b485cb
Signed-off-by: Deva Ramasubramanian <dramasub@codeaurora.org>
Read the video hardware capabilities and add
check for unsupported video sessions.
Change-Id: If3ea9400f92bccbf2e69de61b5aad3ff2673b905
Signed-off-by: Arun Menon <menon@codeaurora.org>
Abdulla Anam
Sanjay Singh
Deepak Kushwah
Manikanta Sivapala
Linux Build Service Account
Maheshwar Ajja
Praneeth Paladugu
Arun Menon
Sachin Shah
Vikash Garodia
Shalaj Jain
Deva Ramasubramanian
Larry Bassel
Ashray Kulkarni
Laura Abbott
Vinay Kalia
Stephen Boyd
Surajit Podder
Eric Ho
Arun Menon