mm_gv11b.c has several direct calls to fb_gv11b.h. Redirect them to
go via a HAL. Also make sure the HALs are using parameter with
correct signedness and prefix the parameter constants with
NVGPU_FB_MMU_.
MMU buffer table indices were also defined in fb_gv11b.h, even though
the tables themselves are defined in include/nvgpu/mm.h. Move the
indices to include/nvgpu/mm.h and prefix them with NVGPU_MM_MMU_.
JIRA NVGPU-714
Change-Id: Ieeae7c5664b8f53f8313cfad0a771d14637caa08
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1776131
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Runtime PM is enabled only for iGPU and not for dGPU. For dGPU,
the .probe() of driver pm_runtime_disable()s, if rail-gating is
not enabled. With nvgpu kernel module load/unload, .probe() is
called multiple times for same struct device *. This results
in an overflow of disable_depth (3 bit refcount) and enables
runtime PM on 8th iteration and calls RTPM routines even if it's
disabled.
To effectively manage pm_runtime_disable(), move it from common
nvgpu_remove() to iGPU/dGPU specific routines.
Also, add restore pm_runtime state of device on driver .remove().
Bug 1987855
Change-Id: I781278da546ef9c9ef7d7da7dbea0757df32716f
Signed-off-by: Nitin Kumbhar <nkumbhar@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1770804
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
On nvgpu module unload, platform_driver_unregister() detaches
driver from device (driver_detach()). As part of this,
__device_release_driver() results a race between driver's
.runtime_resume(), .remove() and .runtime_suspend().
As nvgpu's .remove() is handling all steps of cleaning up
driver state and shutting down gpu, .runtime_suspend()
shall have no work. So skip .runtime_suspend() is gk20a *g
has already been processed.
Bug 1987855
Change-Id: I024ac63d321689ea04c64b1ffc125da943d482f9
Signed-off-by: Nitin Kumbhar <nkumbhar@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1770803
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Seshendra Gadagottu <sgadagottu@nvidia.com>
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
This reverts commit 74d786dd13.
With the recent powergating changes that got merged in k4.14 and by
disabling CONFIG_PROVE_LOCKING, we no longer see any issues in T186.
The reboots observer earlier have stopped and the status of nvgpu_submit
and nvrm_gpu_tests remain the same without causing any more issues
Bug 200414723
Change-Id: Iea872612a233cb3d79df772a78fd6349c69323f2
Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1775199
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
GK20A_DEVFREQ scaling depends on Nvidia downstream kernel
modifications to PM QoS framework. To break that dependency and to
allow devfreq based scaling to work without those changes, introduce a
Kconfig option to conditionally enable the PM Qos based constraints.
Bug 200414723
Bug 200414600
Change-Id: If39a144dfb322176c2d0e6c17d57d0cd6d885c41
Signed-off-by: Timo Alho <talho@nvidia.com>
Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1762987
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: Deepak Nibade <dnibade@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
- Write new pm mode to context buffer header. Ucode use
this mode to enable mode-e context switch. This is Mode-B
context switch of PMs with Mode-E streamout on one context.
If this mode is set, Ucode makes sure that Mode-E pipe
(perfmons, routers, pma) is idle before it context switches PMs.
- This allows us to collect counters in a secure way
(i.e. on context basis) with stream out.
Bug 2106999
Change-Id: I5a7435f09d1bf053ca428e538b0a57f3a175ac37
Signed-off-by: Vaibhav Kachore <vkachore@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1760366
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
-Created common falcon function nvgpu_flcn_bl_bootstrap() to
bootstrap falcon bootloader
-Created HAL gk20a_falcon_bl_bootstrap() which does actual
bootloader bootstrap by fetching parameters and loading
code/parameters as needed.
-Created HAL ops bl_bootstrap under nvgpu_falcon_ops.
-Created struct nvgpu_falcon_bl_info to hold info required
for bootloader to pass to common function
-Removed falcons bootstrap code in multiple file & made
changes to fill struct nvgpu_falcon_bl_info & call
nvgpu_flcn_bl_bootstrap().
Change-Id: Iee275233915ff11f9afb5207ac0c3338ca9dacc1
Signed-off-by: Mahantesh Kumbar <mkumbar@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1756104
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: svc-misra-checker <svc-misra-checker@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Move queries of FB MMU configuration to FB HAL. Also use g->ltc_count
instead of reading the number of LTCs from FB. These changes together
remove last direct uses of FB registers from GR.
JIRA NVGPU-714
Change-Id: I1b4b46fc2f636f5c1904e4174040a47a27948999
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1773076
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
In some use cases client will disable and preempt TSG and then re-enable it
using IOCTLs provided
In case there is only one context getting re-enabled and there is no other job
submission in parallel runlist fetcher will just sleep until doorbell is
received next time
This causes above mentioned test cases to stall after re-enabling TSG
until some one submits a new job and triggers a doorbell
Fix this by explicitly triggering doorbell in vgpu code
after we enable all channels in TSG.
Bug 2205192
Change-Id: I25d643e06152adc6aaf874baf610316f6cd8f13f
Signed-off-by: Aparna Das <aparnad@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1772948
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Enable replayable fault only for contexts where they are requested.
This required moving the code to initialize subcontexts to happen
later.
Fix signedness issues in definition of flags.
JIRA NVGPU-714
Change-Id: I472004e13b1ea46c1bd202f9b12d2ce221b756f9
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1773262
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
- Commit c61e21c868 fixed race codition in PMU
state transition.
- The race condition is such that PMU response(intr callback for messages) can
run faster than kthread posting commands to PMU and thus PMU message callback
may skip important pmu state change.
- Commit c61e21c868 introduced a fix where PMU
state change was only updated from callback while other places can only update
pmu_state variable
- However, this commit introduced a regression as follows:
- When PMU state is PMU_STATE_INIT_RECEIVED, we loop over every engine
supported by GPU --> If state = PMU_STATE_INIT_RECEIVED, change the state
to PMU_STATE_ELPG_BOOTING and init ELPG else If state != PMU_STATE_INIT_RECEIVED
throw an error saying "PMU INIT not received"
- Now, if GPU supports multiple engines, first engine will check that
pmu_state is PMU_STATE_INIT_RECEIVED and change it to PMU_STATE_ELPG_BOOTING
However, from second engine onwards, since state is already changed to
PMU_STATE_ELPG_BOOTING, all engines except first engine start throwing
error "PMU INIT not received"
- This patch fixes the issue by changing pmu state from
PMU_STATE_INIT_RECEIVED to PMU_STATE_ELPG_BOOTING only once.
Bug 200372838
JIRA EVLR-2164
Change-Id: Ic8c954d14acb1d6ec3adcbc4bcf4d4745542d9f0
Signed-off-by: Deepak Bhosale <dbhosale@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1769814
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Mahantesh Kumbar <mkumbar@nvidia.com>
Reviewed-by: Deepak Goyal <dgoyal@nvidia.com>
Reviewed-by: Aparna Das <aparnad@nvidia.com>
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-by: svccoveritychecker <svccoveritychecker@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
While removing nvgpu driver, devm mapped reg mappings
are released on driver_unregister. For iGPU, these
regs are explicitly unmapped with iounmap(). This
results in "Trying to vfree() nonexistent vm area"
warnings on driver removal.
Address this by using devm* variants to map all IO regions
of both iGPU and dGPU and let the driver unregister
release these mappings.
Also, lock out GPU regs in driver removal path.
Bug 1987855
Change-Id: I0388daf90bea3eaf8752255059cfd3ceabf66e7d
Signed-off-by: Nitin Kumbhar <nkumbhar@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1730539
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
In some use cases client will disable and preempt TSG and then re-enable it
using IOCTLs provided
In case there is only one context getting re-enabled and there is no other job
submission in parallel runlist fetcher will just sleep until doorbell is
received next time
This causes above mentioned test cases to stall after re-enabling TSG until
some one submits a new job and triggers a doorbell
Fix this by explicitly triggering doorbell from gv11b_fifo_enable_tsg() after
we enable all channels in TSG
Bug 2205192
Change-Id: I08e70e3d0f7e4dc6471e63809e246430cc4200c1
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1772378
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
gv11b_mm_fault_info_mem_destroy() and gv11b_mm_mmu_hw_fault_buf_deinit()
serve a similar purpose of disabling hub interrupts and deinitializing
memory related to MMU fault handling.
Out of the two the latter was called from BAR2 deinitialization, and the
former from nvgpu_remove_mm_support().
Combine the functions and leave the call from nvgpu_remove_mm_support().
This way BAR2 deinitialization can be combined with gp10b version.
JIRA NVGPU-714
Change-Id: I4050865eaba404b049c621ac2ce54c963e1aea44
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1769627
Reviewed-by: Alex Waterman <alexw@nvidia.com>
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Rename os/linux/vidmem.c to os/linux/dmabuf_vidmem.c. The code is
mainly dealing with interfacing with Linux dmabuf framework and its
responsibilities got confused with common/mm/vidmem.c.
Also move the header include/nvgpu/linux/vidmem.h to
os/linux/dmabuf_vidmem.h. It does not expose any interface to outside
Linux code.
Change-Id: I2cb1057a8934d5cb5c5860023aa12f8f048a6684
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1768261
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
vidmem.h had a forward declaration for a Linux specific struct
work_struct. Removed that.
vidmem.h also #included nvgpu_mem.h even though there was no use
for it. As a follow-up css_gr_gk20a.h did refer to nvgpu_mem but
did not #include it, so added that.
Change-Id: Ifea88adae86ed95302465641821fbb107d7cc233
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1768260
Reviewed-by: Alex Waterman <alexw@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
In nvgpu_dma_alloc_flags_vid_at(), we check pending bytes of vidmem which are
yet to be cleared by reading g->mm.vidmem.bytes_pending.atomic_var
If there is something to be cleared we return EAGAIN otherwise we return ENOMEM
But to store above variable we use "int before_pending" which evaluates to zero
for sizes like 4GB and we end up returning ENOMEM instead of EAGAIN
Fix this by declaring before_pending variable as u64
Bug 200427361
Change-Id: I6ffe977e3663a5135fa17699ecafe78ac90d9314
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1770384
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Konsta Holtta <kholtta@nvidia.com>
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Add new HAL gops.fb.mmu_invalidate_replay() to invalidate replay mmu fault
Use existing API gv11b_fb_mmu_invalidate_replay() to set to this HAL on all
Volta chips
Bug 2228914
Jira NVGPU-838
Jira NVGPUT-73
Change-Id: I394901857d41499f3ea44023393fe271fb664260
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1767970
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
In gr_gk20a_split_ppc_broadcast_addr() we convert a PPC broadcast address to
its corresponding unicast address list
But we consider gr.pe_count_per_gpc instead of actual number of PPCs and that
leads to generating incorrect list of addresses
Fix this by using gr.gpc_ppc_count[gpc_num] which gives correct number of
PPC count
Jira NVGPUT-117
Change-Id: If7e7c19244b90cb3c405dcba4ae7a86c782972f7
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1767838
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Konsta Holtta <kholtta@nvidia.com>
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
- Added prefix gp106_ to sec2_wait_for_halt()
& sec2_clear_halt_interrupt_status() for gp106
SEC2 HAL
- Made changes to gp106_sec2_wait_for_halt() to
read SEC2 falcon mailbox using common falcon
mailbox access functions.
- Add define for falcon mailbox
- These changes are done to reuse gp106 HAL's
for GPU_NEXT.
Change-Id: Id32a7636d775b482684212ed4ef5d01c8ea65335
Signed-off-by: Mahantesh Kumbar <mkumbar@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1755618
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
In gr_gk20a_find_priv_offset_in_buffer() we right now calculate
offset of a register in gpccs segment based on register address type
Separate out sequence to find offset in gpccs segment and move it to new API
gr_gk20a_get_offset_in_gpccs_segment()
Introduce new HAL gops.gr.get_offset_in_gpccs_segment() and set above API
to this HAL
Call HAL from gr_gk20a_find_priv_offset_in_buffer() instead of calling direct
API
Jira NVGPUT-118
Change-Id: I0df798456cf63e3c3a43131f3c4ca7990b89ede0
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1761669
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
nvgpu_mem_rd*() functions were implemented per OS. They also used
nvgpu_pramin_access_batched() and implemented a big portion of logic
for using PRAMIN in OS specific code.
Make the implementation for the functions generic. Move all PRAMIN
logic to PRAMIN and simplify the interface provided by PRAMIN.
Change-Id: I1acb9e8d7d424325dc73314d5738cb2c9ebf7692
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1753708
Reviewed-by: Konsta Holtta <kholtta@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
When requesting a gpc2clck below lowest V/f point, clock arbiter
did not properly adjust target value to nearest V/f point. This
could lead to lower than expected effective frequency. Fixed the
logic to adjust to nearest V/f point.
Bug 200412996
Change-Id: I36c24b4c081931e2ac54da14d49e46fcb14503e3
(cherry picked from commit 7ed1f8fb39f76208922daa91d00905cdb96b2304)
Signed-off-by: Thomas Fleury <tfleury@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1763641
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
