Following changes are added to fix the issue.
1) Threads having higher priority e.g. RT may preempt
threads with sched-normal priority. As a consequence, higher priority
threads might not still see initialization of data in another thread
resulting in failures such as accessing a condition value before initialization.
Any initialization in the parent thread must be accompanied by a barrier
to make it visible in other thread. Added appropriate barriers to prevent
reordering of the initialization in the thread construction path.
2) There is a race condition between nvgpu_cond_signal() and
nvgpu_cond_destroy() in the asynchronous submit code and corresponding
worker thread's process_item callback for NVS. This may lead to
data corruption and resulting in the above errors as well. Fixed
that by adding a refcount based mechanism for ownership sharing
of the struct nvgpu_nvs_worker_item between the two threads.
Bug 3778235
Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Change-Id: Ie9b9ba57bc1dcbb8780801be79863adc39690f72
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2771535
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: svc-mobile-cert <svc-mobile-cert@nvidia.com>
Reviewed-by: Prateek Sethi <prsethi@nvidia.com>
Reviewed-by: Ketan Patil <ketanp@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
GR context associated with channel is updated in various driver paths.
Sequence to do the same is disable the TSG, preempt the TSG, update
the GR context or instance block and then enable the TSG.
These operations and runlist updates for channel have to be done under
TSG specific ctx_init_lock to avoid the race.
suspend_contexts and resume_contexts needs special handling which is
not covered in this patch.
Bug 3677982
Change-Id: I837257fe9d9ef3eb6f69f5d7e0707e0bb6d4ea72
Signed-off-by: Sagar Kamble <skamble@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2720222
Reviewed-by: Scott Long <scottl@nvidia.com>
Reviewed-by: Ankur Kishore <ankkishore@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
With subcontexts support added, nvgpu has to allocate VEID0 channel
itself to initialize the golden context image. Allocate the channel
and init the golden context image at the beginning of alloc_obj_ctx
call for first user channel.
It can't be initialized at the end of probe as tpc pg settings need
to be updated before golden context image is initialized.
Bug 3677982
Change-Id: Ia82f6ad6e088c2bc1578a6bd32b7c7a707a17224
Signed-off-by: Sagar Kamble <skamble@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2756289
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: svc-mobile-cert <svc-mobile-cert@nvidia.com>
Reviewed-by: Vaibhav Kachore <vkachore@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
- Change enables CONFIG_NVS_PRESENT for safety build.
- Fixes misra vioations.
- Renames sched.h to nvs_sched.h to avoid the conflict with QNX system
sched.h file for the safety support.
- Disable test_channel_close, test_tsg_unbind_channel,
test_channel_enable_disable_tsg, test_gv11b_fifo_preempt_tsg,
test_tsg_unbind_channel_check_hw_state and test_rc_deinit unit tests.
Jira NVGPU-8619
Change-Id: I7c983de2f4910fcb23687ec23368a060ce89c918
Signed-off-by: prsethi <prsethi@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2763579
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
- With ELPG + RG enabled, gpu_module_reload test fails.
- This happens because the test tries to unload nvgpu.ko
module and then reload it. This all happens with RG enabled.
- During rmmod of nvgpu.ko module the code path taken is:
nvgpu_remove() -> nvgpu_quiesce() -> gk20a_pm_prepare_poweroff
-> nvgpu_prepare_poweroff -> pmu_destroy
- In this code path, NVGPU_DRIVER_IS_DYING flag is not set.
- Thus, in pmu_pg_task thread (which keeps on running in parallel),
commands are sent to the PMU and the driver keeps waiting for the
ACK in nvgpu_pmu_wait_fw_ack_status().
- Add nvgpu_start_gpu_idle() in nvgpu_remove() path, before calling
nvgpu_quiesce().
- This will set NVGPU_DRIVER_IS_DYING flag to true.
- nvgpu_can_busy() will return 0 when the driver is shutting down or
getting removed.
Bug 3676200
Change-Id: Ic24f58c210e4b477e5d560b053b70c16308e16f1
Signed-off-by: Divya <dsinghatwari@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2762310
Reviewed-by: Debarshi Dutta <ddutta@nvidia.com>
Reviewed-by: Tejal Kudav <tkudav@nvidia.com>
Reviewed-by: Mahantesh Kumbar <mkumbar@nvidia.com>
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
(cherry picked from commit 8f1792565e71b822a6e9cc50af4b43c1b48518e0)
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2765300
Tested-by: Mahantesh Kumbar <mkumbar@nvidia.com>
Following changes are added here to simplify the overall
sequence.
1) Remove deferred update for runlists. NVS worker thread
shall submit the updated runlist.
2) Moved Runlist mem swap inside update itself. Protect
the swap() and hw_submit() path with a spinlock. This
is temporary till GSP.
3) Enable Control-Fifo mode from nvgpu driver.
Jira NVGPU-8609
Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Change-Id: Icc52e5d8ccec9d3653c9bc1cf40400fc01a08fde
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2757406
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
In linux threaded interrupts run with a Realtime priority
of 50. This bumps up the priority of bottom-half handlers
over regular kernel/User threads even during process
context.
In the current implementation scheduler thread still
runs in normal kernel thread priority. In order to
allow a seamless scheduling experience, the worker
thread is now created with a Realtime priority of 1.
This allows for the Worker thread to work at a priority
lower than interrupt handlers but higher than the regular
kernel threads.
Linux kernel allows setting priority with the help of
sched_set_fifo() API. Only two modes are supported
i.e. sched_set_fifo() and sched_set_fifo_low().
For more reference, refer to this article
https://lwn.net/Articles/818388/.
Added an implementation of nvgpu_thread_create_priority()
for linux thread using the above two APIs.
Jira NVGPU-860
Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Change-Id: I0a5a611bf0e0a5b9bb51354c6ff0a99e42e76e2f
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2751736
Reviewed-by: Prateek Sethi <prsethi@nvidia.com>
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
In the current setting, clock arbiter skips setting
the clock if its already set previously. The value
set by the arbiter is stored in
"struct nvgpu_clk_arb->actual" whenever the clock is
updated via the arbiter. However, DVFS might also
update the clock and the updates are not synchronized
with the arbiter. Hence, ensure that any clock
requests are always updated i.e. the requested rate is
set even if the previous rate remains the same.
In the devfreq scale() part, scale emc when clk_arb
is active and skip setting of clocks.
Bug 3666615
Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Change-Id: I32bf4dbf81b19fdd6fa0bdec3a6c9a9312b78eca
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2727787
Reviewed-by: svcacv <svcacv@nvidia.com>
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
This is adding a mutex lock to synchronise between
stall isr threads.
Orin(t234) has three interrupt lines and three ISR
threads to handle bottom half of the ISR. The threads
sharing same data between them without proper synchronization.
When multiple interrupts trigger simeltaneously, causing the
threads running in parallell like below traces
#0 nvgpu_cic_mon_intr_stall_isr (g=g@entry=0x5ed62a9318)
at /home/dt/automotive-dev-main-20220802T015100095/kernel/nvgpu/drivers/gpu/nvgpu/common/cic/mon/mon_intr.c:158
#1 0x00000013758cae30 in nvgpu_intr_stall (arg=0x5ed62a9120)
at /home/dt/automotive-dev-main-20220802T015100095/qnx/src/resmgrs/nvrm/nvgpu_rmos/os/intr.c:140
#2 0x00000013758ec090 in nvgpu_posix_thread_wrapper (data=<optimized out>)
at /home/dt/automotive-dev-main-20220802T015100095/kernel/nvgpu/drivers/gpu/nvgpu/os/posix/thread.c:77
#3 0x0000001375b01000 in pthread_attr_setdetachstate ()
from /home/dt/automotive-dev-main-20220802T015100095/out/embedded-qnx-t186ref-debug-none/target_rootfs/lib/libc.so.5
Backtrace stopped: previous frame identical to this frame (corrupt stack?)
This is causing some race in shared data access and causing
multiple issues.
Bug 3647988
Change-Id: If40e581635b52cce288d8f4b00af6a040f7f9a6e
Signed-off-by: Dinesh T <dt@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2755874
Reviewed-by: Tejal Kudav <tkudav@nvidia.com>
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: svc-mobile-cert <svc-mobile-cert@nvidia.com>
Reviewed-by: svc-mobile-misra <svc-mobile-misra@nvidia.com>
Reviewed-by: Ankur Kishore <ankkishore@nvidia.com>
GVS: Gerrit_Virtual_Submit
The function tegra_bpmp_get() returns an error pointer on failure and
so if the call to tegra_bpmp_get() fails, because the device-tree
property is missing, then this is not detected and leads to a crash when
trying to dereference the pointer to the bpmp handle. Fix this by
correctly checking the return value from tegra_bpmp_get().
Bug 3752030
Change-Id: I944063ab7e116fc81769c9dbbfefe6b6dc4bf0f4
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2759251
Reviewed-by: svcacv <svcacv@nvidia.com>
Reviewed-by: Sagar Kamble <skamble@nvidia.com>
Reviewed-by: Alex Waterman <alexw@nvidia.com>
GVS: Gerrit_Virtual_Submit
Add support for accessing Event Queue for non-exclusive
users. Allows, non-exclusive users to open Event Queues
before exclusive users. Non-Exclusive users can only
use the Event Queue in a read-only mode.
Add VM_SHARED for Event Queues across all users instead of just
Read-Only users. Event queues are shared with multiple processes
and as such require VM_SHARED across all users(exclusive and
observers).
Jira NVGPU-8608
Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Change-Id: Id9733c2511ded6f06dd9feea880005bdc92e51a0
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2745083
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
GVS: Gerrit_Virtual_Submit
This patch fixes a bug where GPU mappings with the L3 hint
would fail. The failure happens because the L3 map flag does
not get consumed if L3 allocations are disabled. The fix
is to consume the flag.
Bug 3717951
Bug 3486025
Change-Id: Ib10ee58cc060318c810f86013de7311f73c25729
Signed-off-by: Martin Radev <mradev@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2750419
Reviewed-by: Sagar Kamble <skamble@nvidia.com>
Reviewed-by: Vaibhav Kachore <vkachore@nvidia.com>
GVS: Gerrit_Virtual_Submit
- Add AP_INIT RPC to initialize the AELPG feature.
- Add AP_CTRL_INIT_AND_ENABLE RPC to program some
APCTRL values for Adaptive ELPG.
- Add AP_CTRL_ENABLE and AP_CTRL_DISABLE RPCs to
send AELPG enable/disable request to PMU via sysfs
node.
- Re-structure the rpc handler based on PG_LOADING
and PG unit id. This is needed to handle different
types of new RPCs from PMU.
JIRA NVGPU-7182
Change-Id: If00b00730507f17ff1883a67094f7e16da5b81ea
Signed-off-by: Divya <dsinghatwari@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2728286
(cherry picked from commit fffb58703bd718600e8c983dcd1c81d9abe83802)
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2603161
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
GVS: Gerrit_Virtual_Submit
In order to maintain separate mappings of GR TSG and global context
buffers for different subcontexts, we need to separate the memory
struct and the mapping struct for the buffers. This patch moves
the mappings of all GR ctx buffers to new structure
nvgpu_gr_ctx_mappings.
This will be instantiated per subcontext in the upcoming patches.
Summary of changes:
1. Various context buffers were allocated and mapped separately.
All TSG context buffers are now stored in gr_ctx->mem[] array
since allocation and mapping is unified for them.
2. Mapping/unmapping and querying the GPU VA of the context
buffers is now handled in ctx_mappings unit. Structure
nvgpu_gr_ctx_mappings in nvgpu_gr_ctx holds the maps.
On ALLOC_OBJ_CTX this struct is instantiated and deleted
on free_gr_ctx.
3. Introduce mapping flags for TSG and global context buffers.
This is to map different buffers with different caching
attribute. Map all buffers as cacheable except
PRIV_ACCESS_MAP, RTV_CIRCULAR_BUFFER, FECS_TRACE, GR CTX
and PATCH ctx buffers. Map all buffers as privileged.
4. Wherever VM or GPU VA is passed in the obj_ctx allocation
functions, they are now replaced by nvgpu_gr_ctx_mappings.
5. free_gr_ctx API need not accept the VM as mappings struct
will hold the VM. mappings struct will be kept in gr_ctx.
6. Move preemption buffers allocation logic out of
nvgpu_gr_obj_ctx_set_graphics_preemption_mode.
7. set_preemption_mode and gr_gk20a_update_hwpm_ctxsw_mode
functions need update to ensure buffers are allocated
and mapped.
8. Keep the unit tests and documentation updated.
With these changes there is clear seggregation of allocation and
mapping of GR context buffers. This will simplify further change
to add multiple address spaces support. With multiple address
spaces in a TSG, subcontexts created after first subcontext
just need to map the buffers.
Bug 3677982
Change-Id: I3cd5f1311dd85aad1cf547da8fa45293fb7a7cb3
Signed-off-by: Sagar Kamble <skamble@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2712222
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Added implementation for following IOCTLs
NVGPU_NVS_CTRL_FIFO_CREATE_QUEUE
NVGPU_NVS_CTRL_FIFO_RELEASE_QUEUE
The above ioctls are supported only for users with
R/W permissions.
1) NVGPU_NVS_CTRL_FIFO_CREATE_QUEUE constructs a memory region
via the nvgpu_dma_alloc_sys() API and creates the corresponding
GPU and kernel mappings. Upon successful creation, KMD exports
this buffer to the userspace via a dmabuf fd that the UMD
can use to mmap it into its process address space.
2) Added plumbing to store VMA's corresponding to different users
for event queue in future.
3) Added necessary validation checks for the IOCTLs
4) NVGPU_NVS_CTRL_FIFO_RELEASE_QUEUE is used to clear the queues.
5) Using a global queue lock to protect access to the queues. This
could be modified to be more fine-grained in future when there
is more clarity on GSP's implementation and access of queues.
6) Added plumbing to enable user subscription to queues.
NVGPU_NVS_CTRL_FIFO_RELEASE_QUEUE is used to unsubscribe
the user from the queue. Once, the last user is deleted,
all the queues will be cleared. User must ensure that
any mappings are removed before calling release queue.
7) Set the default queue_size for event queues to
PAGE_SIZE. This can be modified later. For event
queues, UMD shall fetch the queue_size.
Jira NVGPU-8129
Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Change-Id: I31633174e960ec6feb77caede9d143b3b3c145d7
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2723198
Reviewed-by: svc-mobile-cert <svc-mobile-cert@nvidia.com>
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
GVS: Gerrit_Virtual_Submit
Add a device node for management of nvs control fifo buffers for
scheduling domains. The current design consists of a master structure
struct nvgpu_nvs_domain_sched_ctrl for management of users as well
as control queues. Initially all users are added as non-exclusive users.
Subsequent changes will add support for IOCTLS to manage opening of
Send/Receive and Event buffers, querying characteristics etc.
In subsequent changes, a user that tries to open a Send/Receive queue
will first try to reserve itself as an exclusive user and only if that
succeeds can proceed with creation of both Send/Receive queues.
Exclusive users will be reset to non-exclusive users just before they
close their device node handle.
Jira NVGPU-8128
Change-Id: I15a83f70cd49c685510a9fd5ea4476ebb3544378
Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2691404
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
GVS: Gerrit_Virtual_Submit
On host copy engine PBDMA interrupt, channel is aborted as part of the
recovery and its syncpt value is set to the max threshold.
Syncpoint may then get incremented by PBDMA (incr cmd gets processed)
after this interrupt is handled leading to syncpoint value becoming
greater than the max threshold.
Again while unbinding the channel, syncpoint value is incremented until
it reaches max threshold. Since syncpoint value is already greater than
max threshold, host1x version of nvgpu_nvhost_syncpt_set_minval will
loop for entire u32 range until it reaches max threshold and this
will hang the channel unbind.
nvgpu_nvhost_syncpt_set_minval can ensure the syncpoint value is greater
than or equal to max threshold. Hence update the check for syncpoint
value from not equal to less than.
Bug 3681100
Change-Id: I96e7a1f53d4037e9ed858a2e90dd5a8d17ed6bb0
Signed-off-by: Sagar Kamble <skamble@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2742604
(cherry picked from commit f246facd01)
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2742603
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
GVS: Gerrit_Virtual_Submit
bpmp will floorsweep GPCs as per parameters to tpc_pg_mask sysfs.
While doing that corresponding GPC clocks are also disabled.
nvgpu should re-initialize the clocks every time the
GPC/TPC pg_masks are passed to bpmp mrq.
Also print error when clk_prepare_enable fails.
Introduce platform->clks_lock to protect access to platform->clks
and platform->num_clks done from unrailgate/railgate and bpmp
mrq set calls from sysfs.
Acquire static_pg_lock in railgate path to synchronize railgate
with sysfs.
Bug 3688506
Change-Id: I3203d78b87289e7a847d78b3117e2d3119be3425
Signed-off-by: Sagar Kamble <skamble@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2738920
(cherry picked from commit 28ddb0996f)
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2741029
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
GVS: Gerrit_Virtual_Submit
The HSI error injection utility is an on-bench debug and test utility
which can be used by customers and SQA to test end-to-end error
detection and reporting path.
Inplement callback function to integrate with this utility and allow
injecting GPU HSI related errors.
As part of callback function hsierrrpt_inj(), invoke the driver's
error-reporting logic which uses the EPD MISC_EC APIs. In future,
we can enhance the callback function to trigger driver's error
handling logic incrementally for different errors.
Bug 3413214
Change-Id: I2d050b6c850d6151b40095f243a6733b4ba74f47
Signed-off-by: Tejal Kudav <tkudav@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2727198
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
gr ctx buffer in non-cacheable hence there is no need to do L2 cache
flush when updating the buffer. Remove the flushes.
pm ctx buffer is cacheable hence add l2 flush in the function
nvgpu_profiler_quiesce_hwpm_streamout_non_resident since it
updates the buffer.
Bug 3677982
Change-Id: I0c15ec7a7f8fa250af1d25891122acc24443a872
Signed-off-by: Sagar Kamble <skamble@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2713916
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Enable DEVFREQ for OOT module unconditionally as the podgov governor
module.
linux/pm_qos is only used for downstream supported modifications
which is currently determined by CONFIG_GK20A_PM_QOS.
struct devfreq_dev_status doesn't have any field 'busy' in the upstream
driver hence enable it only for when downstream driver is in use
activated by CONFIG_GK20A_PM_QOS.
governor.h is only needed for android platforms which depend on 4.9
version of the kernel in downstream builds. Hence, added an compile
time flag to remove it for kernels versions greater than 4.9.
Jira LS-418
Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Change-Id: Id242bd28e66ed187208f0d7975ee0bc508730a88
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2705766
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: svc-mobile-cert <svc-mobile-cert@nvidia.com>
Reviewed-by: Jonathan Hunter <jonathanh@nvidia.com>
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
GVS: Gerrit_Virtual_Submit
vidmem buffers are using fd as buffer handles and we need to allocate more
than 1024 fds. tegra_alloc_fd was exported by TEGRA_MC driver that allowed
allocating more than 1024 fds, however that function is to be removed from
that driver.
Hence use now kernel exported function __alloc_fd directly from nvgpu.
This is currently to be used only for dgpu on downstream kernel 5.10.
Bug 3535321
Change-Id: I10cfc41a6439f07309cda9eb2f22746f3fbac996
Signed-off-by: Dinesh T <dt@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2702794
Reviewed-by: Alex Waterman <alexw@nvidia.com>
Reviewed-by: Sagar Kamble <skamble@nvidia.com>
Reviewed-by: Sachin Nikam <snikam@nvidia.com>
Tested-by: Sagar Kamble <skamble@nvidia.com>
GVS: Gerrit_Virtual_Submit
This patch creates a sysfs node which is easier to parse than
the existing mig_mode_config_list node. This new node outputs
in the following format:
active: -1
num_configs: 2
num_instances: 2
id:000000000001 gr:000000000000 gpc:0003
id:000000000002 gr:000000000004 gpc:0003
num_instances: 3
id:000000000001 gr:000000000000 gpc:0003
id:000000000005 gr:000000000004 gpc:0002
id:000000000013 gr:000000000006 gpc:0001
Bug 200740852
Change-Id: I8a3d4425ccb88dd4e58bbe1908e0f7cc577ff191
Signed-off-by: Martin Radev <mradev@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2704349
Reviewed-by: Alex Waterman <alexw@nvidia.com>
GVS: Gerrit_Virtual_Submit
To be able to access the full physical memory range, gpu's dma_mask
needs to be set to the max value of H/W compatible range.
For example. In order to support from 2GB to 66 GB, GV11B's dma_mask
needs to be atleast 37 bits. Set GV11B's dma_mask to 38 bit
and T23X's dma_mask to 39 bit. These values are supported by H/W
Bug 3656729
Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Change-Id: Icfff3c36a8c9cf074a254fa773c42e18020ae5de
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2723640
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: svc-mobile-cert <svc-mobile-cert@nvidia.com>
Reviewed-by: Brad Griffis <bgriffis@nvidia.com>
Reviewed-by: Alex Waterman <alexw@nvidia.com>
GVS: Gerrit_Virtual_Submit
Tested-by: Brad Griffis <bgriffis@nvidia.com>
Background: In case of a deferred suspend implemented by gk20a_idle,
the device waits for a delay before suspending and invoking
power gating callbacks. This helps minimize resume latency for any
resume calls(gk20a_busy) that occur before the delay.
Now, some APIs spread across the driver requires that if the device
is powered on, then they can proceed with register writes, but if its
powered off, then it must return. Examples of such APIs include
l2_flush, fb_flush and even nvs_thread. We have relied on
some hacks to ensure the device is kept powered on to prevent any such
delayed suspension to proceed. However, this still raced for some calls
like ioctl l2_flush, so gk20a_busy() was added (Refer to commit Id
dd341e7ecbaf65843cb8059f9d57a8be58952f63)
Upstream linux kernel has introduced the API pm_runtime_get_if_active
specifically to handle the corner case for locking the state during the
event of a deferred suspend.
According to the Linux kernel docs, invoking the API with
ign_usage_count parameter set to true, prevents an incoming suspend
if it has not already suspended.
With this, there is no longer a need to check whether
nvgpu_is_powered_off(). Changed the behavior of gk20a_busy_noresume()
to return bool. It returns true, iff it managed to prevent
an imminent suspend, else returns false. For cases where
PM runtime is disabled, the code follows the existing implementation.
Added missing gk20a_busy_noresume() calls to tlb_invalidate.
Also, moved gk20a_pm_deinit to after nvgpu_quiesce() in
the module removal path. This is done to prevent regs access
after registers are locked out at the end of nvgpu_quiesce. This
can happen as some free function calls post quiesce might still
have l2_flush, fb_flush deep inside their stack, hence invoke
gk20a_pm_deinit to disable pm_runtime immediately after quiesce.
Kept the legacy implementation same for VGPU and
older kernels
Jira NVGPU-8487
Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Change-Id: I972f9afe577b670c44fc09e3177a5ce8a44ca338
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2715654
Reviewed-by: Sagar Kamble <skamble@nvidia.com>
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
GVS: Gerrit_Virtual_Submit
Below print is misleading and seems like an error.
[INFO] Missing support-gpu-tools property, ret =-22
'support-gpu-tools' property was added to allow disabling debugger
features on prod boards. The debugger/profiler support will be
enabled by default, even if the property is missing.
Make the INFO print conditional, more informational and less
dramatic.
Bug 3539518
Change-Id: I5fc50df30be23e1fd1ecc06282a0d50f3ca7ac64
Signed-off-by: Tejal Kudav <tkudav@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2668464
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
