Workaround for GPU hang if boost turns GPU on before it is
initialized.
Bug 1435870
Change-Id: I07d0617049612344ca7c494da8cb8d75789984e5
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: http://git-master/r/453375
"isr_enable_lock" was used to protect pmu's isr_enabled flag
and pmu enable/disable calls
Instead of this extra lock, we can reuse "isr_mutex" for this
purpose
Bug 200014542
Bug 200014887
Change-Id: Ifbb7d6108effc132266a20517820e470d52a7110
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: http://git-master/r/453348
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Tested-by: Terje Bergstrom <tbergstrom@nvidia.com>
Add below ioctls for channels
1. NVHOST_IOCTL_CHANNEL_ENABLE
To enable the channel
2. NVHOST_IOCTL_CHANNEL_DISABLE
To disable the channel
3. NVHOST_IOCTL_CHANNEL_PREEMPT
To preempt the channel
(Not supported for a channel in TSG)
Bug 1514064
Change-Id: Ie9315f9742bb27efb22f993799c51a1ecda91756
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: http://git-master/r/449229
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Tested-by: Terje Bergstrom <tbergstrom@nvidia.com>
This patch fixes a refcounting issue in semaphore handling.
Change-Id: I03327c60ed6923a90663f0b845566e81af4b94d4
Signed-off-by: Arto Merilainen <amerilainen@nvidia.com>
Reviewed-on: http://git-master/r/453056
Reviewed-by: Automatic_Commit_Validation_User
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
In runlist we first write channel count in TSG entry and then
follow those many channel entries
If no. of channel entries does not match to count then it is
considered as error
To detect this, add a counter while adding channel entries and
give warning if channel count does not match with this counter
bug 1470692
Change-Id: I4bbfd9b696fbfafa25dffb27979373f057a7f35a
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: http://git-master/r/449228
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Tested-by: Terje Bergstrom <tbergstrom@nvidia.com>
Currently we clear the runlist and re-create it in
scheduled work during fifo recovery process
But we can post-pone this runlist re-generation for
later time i.e. when channel is closed
Hence, remove runlist locks and re-generation from
handle_mmu_fault() methods. Instead of that, disable
gr fifo access at start of recovery and re-enable
it at end of recovery process.
Also, delete scheduled work to re-create runlist.
Re-enable EPLG and fifo access in
finish_mmu_fault_handling() itself.
bug 1470692
Change-Id: I705a6a5236734c7207a01d9a9fa9eca22bdbe7eb
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: http://git-master/r/449225
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Tested-by: Terje Bergstrom <tbergstrom@nvidia.com>
Modify TSG's channel list as "ch_list" for all channels
instead of "ch_runnable_list" for only runnable list
We can traverse this list and check runnable status of
channel in active_channels to get runnable channels
Remove below APIs as they are no longer required :
gk20a_bind_runnable_channel_to_tsg()
gk20a_unbind_channel_from_tsg()
While closing the channel, call gk20a_tsg_unbind_channel()
to unbind the channel from TSG
bug 1470692
Change-Id: I0178fa74b3e8bb4e5c0b3e3b2b2f031491761ba7
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: http://git-master/r/449227
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Tested-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reading the load value may increase CPU power consumption
temprorarily. In most cases we are ok with a value that
was read a moment earlier.
This patch introduces a software shadow for gpu load. The shadow
is updated before starting scaling and all scaling code paths use
the sw shadow.
Change-Id: I53d2ccb8e7f83147f411a14d3104d890dd9af9a3
Signed-off-by: Arto Merilainen <amerilainen@nvidia.com>
Reviewed-on: http://git-master/r/453347
Reviewed-by: Automatic_Commit_Validation_User
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
1. Register tegra-throttle cooling device as a
platform driver.
2. Obtain all the platform data (throtlle table
info) for all instances of blanced-throtlled cdev
from device tree and register them.
Change-Id: Ie92685eea3eb5cb18068b195adc9ab5f83762399
Signed-off-by: Arun Kumar Swain <arswain@nvidia.com>
Reviewed-on: http://git-master/r/449104
Reviewed-by: Diwakar Tundlam <dtundlam@nvidia.com>
Tested-by: Diwakar Tundlam <dtundlam@nvidia.com>
return error from pmu_mutex_acquire() and release() if
pmu->initialized is not set
Bug 1533644
Change-Id: I341a5831bc5beeccb4587668f61c954ce7576226
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Currently if pmu_mutex_acquire() fails, we disable ELPG
and move ahead. But it is not clear why it is required
to disable ELPG in case where we fail to acquire mutex.
Hence skip disabling ELPG if mutex_acquire() fails
Bug 1533644
Change-Id: I7e8e99a701d0ba071eb31ac17582b04072ee55eb
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: http://git-master/r/448131
Reviewed-by: Automatic_Commit_Validation_User
GVS: Gerrit_Virtual_Submit
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Compression bit base was calculated incorrectly in cases where
number of LTCs was not 1. This patch fixes the code.
Change-Id: I25e3fa7446b238202d93ce8a72ed919d11fb6e30
Signed-off-by: Arto Merilainen <amerilainen@nvidia.com>
Reviewed-on: http://git-master/r/449281
Reviewed-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-by: Automatic_Commit_Validation_User
Tested-by: Jussi Rasanen <jrasanen@nvidia.com>
GVS: Gerrit_Virtual_Submit
If code coverage is enabled on GCC 4.7, the kernel build fails in
gk20a_init_kind_attr() since GCC decides to inline almost everything in
this file into it, leading to a massive stack frame with over kilobyte's
worth of temporary variables generated by gcov, leading to this error:
kind_gk20a.c: In function 'gk20a_init_kind_attr':
kind_gk20a.c:424:1: error: the frame size of 1232 bytes is larger than
1024 bytes [-Werror=frame-larger-than=]
(Just removing the inline keyword doesn't work, as GCC still decides to
inline it, so noinline_for_stack is actually required.)
Change-Id: I819fd2a5b20581f0ac60e1ee490899c977379151
Signed-off-by: Tuomas Tynkkynen <ttynkkynen@nvidia.com>
Reviewed-on: http://git-master/r/448914
Reviewed-by: Juha Tukkinen <jtukkinen@nvidia.com>
Tested-by: Juha Tukkinen <jtukkinen@nvidia.com>
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
This patch adds support for executing a precompiled GPU program to
allow exporting GPU buffers to other graphics units that have color
decompression engine (CDE) support.
Bug 1409151
Change-Id: Id0c930923f2449b85a6555de71d7ec93eed238ae
Signed-off-by: Arto Merilainen <amerilainen@nvidia.com>
Reviewed-on: http://git-master/r/360418
Reviewed-by: Lauri Peltonen <lpeltonen@nvidia.com>
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
When moving compression state tracking and compbit management ops to
kernel, we need to attach a fence to dma-buf metadata, along with the
compbit state.
To make in-kernel fence management easier, introduce a new gk20a_fence
abstraction. A gk20a_fence may be backed by a semaphore or a syncpoint
(id, value) pair. If the kernel is configured with CONFIG_SYNC, it will
also contain a sync_fence. The gk20a_fence can easily be converted back
to a syncpoint (id, value) parir or sync FD when we need to return it to
user space.
Change gk20a_submit_channel_gpfifo to return a gk20a_fence instead of
nvhost_fence. This is to facilitate work submission initiated from
kernel.
Bug 1509620
Change-Id: I6154764a279dba83f5e91ba9e0cb5e227ca08e1b
Signed-off-by: Lauri Peltonen <lpeltonen@nvidia.com>
Reviewed-on: http://git-master/r/439846
Reviewed-by: Automatic_Commit_Validation_User
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Tested-by: Terje Bergstrom <tbergstrom@nvidia.com>
Remove functions that are not used in gk20a allocator.
Bug 1523403
Change-Id: I36b2b236258d61602cb3283b59c43b40f237d514
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: http://git-master/r/432174
In channel_finish(), we submit WFI for all the channels
including channels with KEPLER_C class.
Since there is no need to submit WFI for channels with
KEPLER_C class, we can optimize to skip submitting
WFI and directly wait on last submit fence
Bug 1534272
Change-Id: I3838416cf22122728e7f1008e01d77b14a35deba
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Currently gk20a gets reference of host1x via phandle in
Device Tree. But runtime PM does not seem to be handling power
dependencies too well in this case and hence some times host1x
is off when we need it.
To fix this, exlicitly power on host1x while powering gpu up.
Do this via "busy" and "idle" callbacks from gk20a_platform
Bug 1534272
Bug 200022536
Change-Id: Ia562ee19722cfc8edc5626a5a058ab8edfe3d206
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
For LS PMU new ucode needs to be used.
Ucode has interface header file changes too.
This patch also has fixes for pmu dmem copy failure
Bug 1509680
Change-Id: I8c7018f889a82104dea590751e650e53e5524a54
Signed-off-by: Supriya <ssharatkumar@nvidia.com>
Reviewed-on: http://git-master/r/441734
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Tested-by: Terje Bergstrom <tbergstrom@nvidia.com>
Used GPU device name in clock get operation (instead of fixed name),
to make operation is common for GK20A and GM20B. Updated clock ids
in tegra clock framework accordingly.
Bug 1450787
Change-Id: Ifd5b9c3a6fd8db5b06e6dcd989285e8410794803
Signed-off-by: Alex Frid <afrid@nvidia.com>
Reviewed-on: http://git-master/r/441711
Reviewed-by: Bo Yan <byan@nvidia.com>
Tested-by: Bo Yan <byan@nvidia.com>
Made GK20A and GM20B clock operations static, since they are invoked
only via HAL interfaces.
Bug 1450787
Change-Id: Ia30218ad4244bd8790b5ef96d1963678d0ba39e1
Signed-off-by: Alex Frid <afrid@nvidia.com>
Reviewed-on: http://git-master/r/441710
Reviewed-by: Bo Yan <byan@nvidia.com>
Tested-by: Bo Yan <byan@nvidia.com>
This reverts commit 50497d4031103df1067f14ce4c1e14b15713efb9.
Simply returning error from mutex_acquire() causes the code
to call disable_elpg() which decreases elpg refcount
But we already have a race condition between pmu initialization
where we initialize elpg and runlist update where we call
this mutex_acquire and decrease the refcount
As a result of this race and returned error we might mess up
with the elpg refcount and cause abnormal behaviour
Hence revert this change for now until we have clean fix
considering this race as well
Bug 200024116
Change-Id: Ie64ca36f70aba6b15c2acc235a5d36d13c9025aa
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: http://git-master/r/441793
Reviewed-by: Arto Merilainen <amerilainen@nvidia.com>
- gk20a_enable is reading the clock after unlocking the spinlock
to flush any previous write
- This could lead to a race if any write afterwards assume
the write has been completed already
- Read the clock before unlocking to ensure all previous writes
have been completed before letting any other thread use gk20a
Bug 200007520
Change-Id: I737fbbe825c68b25ca256c4a8ee2b99aa8baf0f5
Signed-off-by: Sang-Hun Lee <sanlee@nvidia.com>
Reviewed-on: http://git-master/r/418485
(cherry picked from commit 2aed542a719caa69620766bf2dceefe50626c189)
Reviewed-on: http://git-master/r/437842
Reviewed-by: Mitch Luban <mluban@nvidia.com>
Tested-by: Mitch Luban <mluban@nvidia.com>
gm20b/gm20x requires incrementing syncpoints twice to ensure that
the data has reached memory in all cases. This patch modifies
increment push buffer to account this requirement.
Bug 1491360
Change-Id: I5c2899b26ce0e1cdf9408bb9aaa576fc3054480f
Signed-off-by: Arto Merilainen <amerilainen@nvidia.com>
Reviewed-on: http://git-master/r/437675
Reviewed-by: Automatic_Commit_Validation_User
This patch adds mm helpers to access compression backing store
from in-kernel shader.
Bug 1409151
Change-Id: Icb4f6dc0b5a35fdb97bc4221ab3657866f775fae
Signed-off-by: Arto Merilainen <amerilainen@nvidia.com>
Reviewed-on: http://git-master/r/440263
Reviewed-by: Automatic_Commit_Validation_User
Reviewed-by: Lauri Peltonen <lpeltonen@nvidia.com>
GVS: Gerrit_Virtual_Submit
In cases where a kernel channel dies, we can reload the context by
just reloading the golden context buffer. This patch makes necessary
infrastructural changes to support this behaviour.
Bug 1409151
Change-Id: Ibe6a88bf7acea2d3aced2b86a7a687279075c386
Signed-off-by: Arto Merilainen <amerilainen@nvidia.com>
Reviewed-on: http://git-master/r/440262
Reviewed-by: Automatic_Commit_Validation_User
Reviewed-by: Lauri Peltonen <lpeltonen@nvidia.com>
GVS: Gerrit_Virtual_Submit
This patch modifies channel interfaces to allow allocating the
channel for kernel use. This is needed if we want to run a shader
from kernel space.
Bug 1409151
Change-Id: I3544186bb1541120f85e01a19de106ef011c1b11
Signed-off-by: Arto Merilainen <amerilainen@nvidia.com>
Reviewed-on: http://git-master/r/440261
Reviewed-by: Automatic_Commit_Validation_User
Reviewed-by: Lauri Peltonen <lpeltonen@nvidia.com>
GVS: Gerrit_Virtual_Submit
In pmu_mutex_acquire(), we return zero (success) if
pmu->initialized is not set
Since mutex_acquire() was successful, we then call
pmu_mutex_release()
If now pmu->initialized is set in some other thread
then we proceed to validate the mutex owner and
end up causing below warning :
pmu_mutex_release: requester 0x00000000 NOT match owner 0x00000008
Hence to fix this return error from mutex_acquire()
and mutex_release() if pmu->initialized is not yet set
and in that case we proceed to call elpg enable/disable
Bug 1533644
Change-Id: Ifbb9e6a8e13f6478a13e3f9d98ced11792cc881f
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: http://git-master/r/439333
GVS: Gerrit_Virtual_Submit
Reviewed-by: Naveen Kumar S <nkumars@nvidia.com>
Tested-by: Naveen Kumar S <nkumars@nvidia.com>
Reviewed-by: Arto Merilainen <amerilainen@nvidia.com>
Add a way to force idle and reset the GPU in case where GPU
rail gating is not supported
(i.e. platform->can_railgate = false)
In this case, we follow below sequence :
- once GPU is idle, get runtime reference which enables the clocks
- call prepare_poweroff() to save the state explicitly
- perform explicit reset assert/deassert
- call finalize_poweron() to restore the state
- drop the runtime reference taken earlier
Bug 1525284
Change-Id: Id5f3ec152093acd585631dfbf785d8e0561f9048
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: http://git-master/r/435620
GVS: Gerrit_Virtual_Submit
Reviewed-by: Arto Merilainen <amerilainen@nvidia.com>
Tested-by: Arto Merilainen <amerilainen@nvidia.com>
Increase the wait delays in do_idle() to 2000 mS and make use
of msleep instead of mdelays
Also, to check if GPU is rail gated or not, add a do-while()
loop which will keep checking the status and bail out as soon
as GPU is rail gated
This increase in delays is required to allow GPU sufficient
time to complete its work and get rail gated
These delays are specially needed during stress testing where
it is possible that a large amount of GPU work is blocked
during do_idle() and then it might take more time to complete
it while next do_idle() is waiting for it
Also, remove waiting on API gk20a_wait_channel_idle() for each
channels since it is sufficient to wait for refcount to be 1
bug 1529160
Change-Id: Ie541485fbdda76d79ae4a75dda928da240fc5d8f
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: http://git-master/r/434192
(cherry picked from commit 5a621bf2aaf3355e1330a662dc98e943d68ef86d)
Reviewed-on: http://git-master/r/435133
Reviewed-by: Arto Merilainen <amerilainen@nvidia.com>
Reviewed-by: Sachin Nikam <snikam@nvidia.com>
gk20a_busy() call in channel_syncpt_incr() and corresponding
gk20a_idle() call in channel_update() are redundant since they
are already encapsulated inside another pair of busy/idle calls
This busy/idle pair will be called only from submit_gpfifo()
and submit_gpfifo() already has its own busy/idle which it
preserves for whole path and hence this redundant pair can be
removed
Also, this prevents a dead lock scenario while do_idle() is in
progress as follows :
- in submit_gpfifo() we call first gk20a_busy() which acquires
busy read semaphore
- in do_idle() we acquire busy write semaphore and wait for
current jobs to finish
- now submit_gpfifo() encounters second gk20a_busy() and requests
busy read semaphore again
- this results in dead lock where do_idle() is waiting for
submit_gpfifo() to complete and submit_gpfifo() is waiting for
busy lock held by do_idle() and hence it cannot complete
bug 1529160
Change-Id: I96e4368352f693e93524f0f61689b4447e5331ea
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: http://git-master/r/434191
(cherry picked from commit c4315c6caa42bab72ba6017c7ded25f4e9363dec)
Reviewed-on: http://git-master/r/435132
Reviewed-by: Sachin Nikam <snikam@nvidia.com>
Tested-by: Sachin Nikam <snikam@nvidia.com>
While we are executing do_idle() API, it is possible that
unrailgate() gets invoked in midst of idling the GPU and
this can result in failure of do_idle()
To prevent simultaneous execution of these methods,
add a mutex railgate_lock and acquire it during
do_idle() and unrailgate() APIs
Also, keep this lock held if do_idle() is successful.
In success, lock will be released in do_unidle(),
otherwise release this lock before returning
Note that this lock should not be held in railgate() API
since we do not want it to be blocked during do_idle()
bug 1529160
Change-Id: I87114b5367eaa217376455a2699c0d21c451c889
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: http://git-master/r/434190
(cherry picked from commit 561dc8e0933ff2d72573292968b893a52f5f783a)
Reviewed-on: http://git-master/r/435131
Reviewed-by: Arto Merilainen <amerilainen@nvidia.com>
Reviewed-by: Sachin Nikam <snikam@nvidia.com>
Channel and gpfifo allocations are entirely separated from each
other, however, the code here assumes that active channel means
that the channel also has a gpfifo.
This reverts commit a24602f094380539788696d1b1567a4f4d914b17 which
added gpfifo dump. Changing debug dumping to be safe requires
refactoring the channel release code to use proper locking.
Bug 1530226
Change-Id: I2fb02542a17dd56a0a9ce732b327e34b85ade8b9
Signed-off-by: Arto Merilainen <amerilainen@nvidia.com>
Reviewed-on: http://git-master/r/434038
Reviewed-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-by: Shridhar Rasal <srasal@nvidia.com>
Reviewed-by: Bharat Nihalani <bnihalani@nvidia.com>
This patch adds support to check if the host1x link exists and is
supported using the gpu characteristics ioctl.
Bug 1459653
Change-Id: I832eea217ed7f007e341dfde5769887e0882d6bb
Signed-off-by: Arto Merilainen <amerilainen@nvidia.com>
Reviewed-on: http://git-master/r/433058
