a. LAUNCH_ERR
- Userspace error.
- Triggered due to faulty launch.
- Handle using recovery to reset CE engine and teardown the
faulty channel.
b. An INVALID_CONFIG -
- Triggered when LCE is mapped to floorswept PCE.
- On iGPU, we use the default PCE 2 LCE HW mapping.
The default mapping can be read from NV_CE_PCE2LCE_CONFIG
INIT value in CE refmanual.
- NvGPU driver configures the mapping on dGPUs (currently only on
Turing).
- So, this interrupt can only be triggered if there is
kernel or HW error
- Recovery ( which is killing the context + engine reset) will
not help resolve this error.
- Trigger Quiesce as part of handling.
c. A MTHD_BUFFER_FAULT -
- NvGPU driver allocates fault buffers for all TSGs or contexts,
maps them in BAR2 VA space and writes the VA into channel
instance block.
- Can be triggered only due to kernel bug
- Recovery will not help, need quiesce
d. FBUF_CRC_FAIL
- Triggered when the CRC entry read from the method fault buffer
does not match the computed CRC from the methods contained in
the buffer.
- This indicates memory corruption and is a fatal interrupt which
at least requires the LCE to be reset before operations can
start again, if not the entire GPU.
- Better to quiesce on memory corruption
CE Engine reset (via recovery) will not help.
e. FBUF_MAGIC_CHK_FAIL
- Triggered when the MAGIC_NUM entry read from the method fault
buf does not match NV_CE_MTHD_BUFFER_GLOBAL_HDR_MAGIC_NUM_VAL
- This indicates memory corruption and is a fatal interrupt
- Better to quiesce on memory corruption
f. STALLING_DEBUG
- Only triggered with SW write for debug purposes
- Debug interrupt, currently ignored
Move launch error handling from GP10b to GV11b HAL as -
1. LAUNCHERR_REPORT errcode METHOD_BUFFER_ACCESS_FAULT is not
defined on Pascal
2. We do not support GP10b on dev-main ToT
JIRA NVGPU-8102
Change-Id: Idc84119bc23b5e85f3479fe62cc8720e98b627a5
Signed-off-by: Tejal Kudav <tkudav@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2678893
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
This patch does the following changes:
- Compiles-out unused error reporting APIs and the related
data structures from safety build. For this purpose, it
introduces the new flag: CONFIG_NVGPU_INTR_DEBUG
- Updates nvgpu_report_err_to_sdl() API with one more argument,
hw_unit_id. This aids in finding whether an error to be reported
is corrected or uncorrected from LUT.
- Triggers SW quiesce, if an uncorrected error is reported to
Safety_Services, in safety build.
- Renames files in cic folder by replacing gv11b with ga10b,
since error reporting for gv11b is not supported in dev-main.
JIRA NVGPU-8002
Change-Id: Ic01e73b0208252abba1f615a2c98d770cdf41ca4
Signed-off-by: Rajesh Devaraj <rdevaraj@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2668466
Reviewed-by: Tejal Kudav <tkudav@nvidia.com>
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: svc-mobile-misra <svc-mobile-misra@nvidia.com>
Reviewed-by: svc-mobile-cert <svc-mobile-cert@nvidia.com>
Reviewed-by: Vaibhav Kachore <vkachore@nvidia.com>
GVS: Gerrit_Virtual_Submit
In DRIVE 6.0, NvGPU needs to support error reporting in QNX-Safety,
QNX-Standard, and Linux. To support error reporting in all these
platform variants, SDL unit will be moved from QNX to common code.
As part of this refactoring activity, this patch removes ops assignment
for report error. Also, it removes API calls that are used to take
time-stamp for stall interrupt thread. This time-stamp APIs will be
brought back later, if required to support periodic diagnostics.
JIRA NVGPU-7353
Change-Id: I38536019dc7165e6a97674863b37d009854af948
Signed-off-by: Rajesh Devaraj <rdevaraj@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2655958
Reviewed-by: svcacv <svcacv@nvidia.com>
Reviewed-by: Shashank Singh <shashsingh@nvidia.com>
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
common.cic unit is divided into common.cic.mon and common.cic.rm
based on rm and mon process split.
CIC-mon subunit includes the code which is utilized in critical
interrupt handling path like initialization, error detection and
error reporting path. CIC-rm subunit includes the code corresponding
to rest of interrupt handling(like collecting error debug data from
registers) and ISR status management (status of deferred interrupts).
Split the CIC APIs and data-members into above two subunits.
JIRA NVGPU-6899
Change-Id: I151b59105ff570607c4a62e974785e9c1323ef69
Signed-off-by: Tejal Kudav <tkudav@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2551897
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Add a new Central Interrupt Controller(CIC) unit in common code.
The interrupt handling is done in a distributed manner currently.
The error handling policy for different errors resides in each unit's
ISR code. The goal is to converge this data under one central place -
the CIC unit.
This patch creates framework for CIC unit and moves the gv11b QNX
safety LUT to CIC unit. All the error reporting APIs from different
units are also moved to CIC.
New APIs are exposed by CIC unit to access its internal data like:
1. Struct err_desc - the static err handling /injection data per
error id
2. Num_hw_modules - the number of error reporting HW units
supported by CIC
Init and deinit of CIC unit:
1. CIC unit should be initialized earlyon during boot so that it
is available for any interrupt handling.
2. Initialize CIC just before the interrupts are enabled during
boot.
3. Similarly, CIC is disabled late during deinit cycle; right
after the interrupts are masked.
LUT:
1. LUT is currently used only for reporting error to safety
services in gv11b QNX safety build.
2. This error handling policy LUT currently has only two levels
of handing - correctable and quiecse.
3. Once, the error handling policy decision is moved from leaf
unit nodes to CIC, LUT will be updated to have additional levels
like fast recovery and full recovery.
4. Also, then a separate LUT will be added for each platform/build.
5. In current framework, the LUT is set to NULL for all
configurations except gv11b.
report_err() ops is added to report error to safety services.
This ops is only effective for gv11b qnx build; and set to NULL for
other configurations.
NVGPU-6521
NVGPU-6523
NVGPU-6750
NVGPU-6758
NVGPU-6760
NVGPU-6754
Change-Id: I24be7836a96d787741e37b732e19863ed8014635
Signed-off-by: Tejal Kudav <tkudav@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2518683
Reviewed-by: Ajesh K V <akv@nvidia.com>
Reviewed-by: Alex Waterman <alexw@nvidia.com>
Reviewed-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
The function gk20a_mm_l2_flush incorrectly returns an error value
when it skips l2_flush when hardware is powered off.
This causes the following prints to occur even when the behavior is expected.
gv11b_mm_l2_flush:43 [ERR] gk20a_mm_l2_flush failed
nvgpu_gmmu_unmap_locked:1043 [ERR] gk20a_mm_l2_flush[1] failed
The above errors occur from the following paths
1) gk20a_remove -> gk20a_free_cb -> gk20a_remove_support ->
nvgpu_pmu_remove_support -> nvgpu_pmu_pg_deinit ->
nvgpu_dma_unmap_free
2) gk20a_remove -> gk20a_free_cb -> gk20a_remove_support ->
nvgpu_remove_mm_support -> gv11b_mm_mmu_fault_info_mem_destroy ->
nvgpu_dma_unmap_free
Since, these do not belong in the Poweron/Poweroff path, its okay to
skip flushing them when the hardware has powered off.
Fixed the userspace tests by allocating g->mm.bar1.vm to prevent NULL access
in gv11b_mm_l2_flush->tlb_invalidate.
Jira LS-77
Change-Id: I3ca71f5118daf4b2eeacfe5bf83d94317f29d446
Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2523751
Reviewed-by: svc_kernel_abi <svc_kernel_abi@nvidia.com>
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: svc-mobile-cert <svc-mobile-cert@nvidia.com>
Reviewed-by: Sagar Kamble <skamble@nvidia.com>
Reviewed-by: Vaibhav Kachore <vkachore@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
Mapping of large buffers to GMMU end up needing many
pages for the PTE tables. Allocating these one by one
can end up being a performance bottleneck, particularly
in the virtualized case.
This is adding the following changes:
- As the TLB invalidation doesn't have access to mem_off,
allow top-level allocation by alloc_cache_direct().
- Define NVGPU_PD_CACHE_SIZE, the allocation size for a new slab
for the PD cache, effectively set to 64K bytes
- Use the PD cache for any allocation < NVGPU_PD_CACHE_SIZE
When freeing up cached entries, avoid prefetch errors by
invalidating the entry (memset to 0).
- Try to fall back to direct allocation of smaller chunk for
contiguous allocation failures.
- Unit test changes.
Bug 200649243
Change-Id: I0a667af0ba01d9147c703e64fc970880e52a8fbc
Signed-off-by: dt <dt@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2404371
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
The FIFO pbdma map is an array of bit maps that link PBDMAs to runlists.
This array allows other software to query what PBDMA(s) serves a given
runlist. The PBDMA map is read verbatim from an array of host registers.
These registers are stored in a kmalloc()'ed array.
This causes a problem for the device management code. The device
management initialization executes well before the rest of the FIFO
PBDMA initialization occurs. Thus, if the device management code
queries the PBDMA mapping for a given device/runlist, the mapping has
yet to be populated.
In the next patches in this series the engine management code is subsumed
into the device management code. In other words the device struct is
reused by the engine management and all host SW does is pull pointers to
the host managed devices from the device manager. This means that all
engine initialization that used to be done on top of the device
management needs to move to the device code.
So, long story short, the PBDMA map needs to be read from the registers
directly, instead of an array that gets allocated long after the device
code has run.
This patch removes the pbdma map array, deletes two HALs that managed
that, and instead provides a new HAL to query this map directly from
the registers so that the device code can use it.
JIRA NVGPU-5421
Change-Id: I5966d440903faee640e3b41494d2caf4cd177b6d
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2361134
Tested-by: mobile promotions <svcmobile_promotions@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: Deepak Nibade <dnibade@nvidia.com>
Reviewed-by: Konsta Holtta <kholtta@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
Many tests used various incarnations of the mock register framework.
This was based on a dump of gv11b registers. Tests that greatly
benefitted from having generally sane register values all rely
heavily on this framework.
However, every test essentially did their own thing. This was not
efficient and has caused a some issues in cleaning up the device and
host code.
Therefore introduce a much leaner and simplified register framework.
All unit tests now automatically get a good subset of the gv11b
registers auto-populated. As part of this also populate the HAL with
a nvgpu_detect_chip() call. Many tests can now _probably_ have all
their HAL init (except dummy HAL stuff) deleted. But this does
require a few fixups here and there to set HALs to NULL where tests
expect HALs to be NULL by default.
Where necessary HALs are cleared with a memset to prevent unwanted
code from executing.
Overall, this imposes a far smaller burden on tests to initialize
their environments.
Something to consider for the future, though, is how to handle
supporting multiple chips in the unit test world.
JIRA NVGPU-5422
Change-Id: Icf1a63f728e9c5671ee0fdb726c235ffbd2843e2
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2335334
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Currently, setup_sw_s2 case of test_gv11b_mm_mmu_fault_setup_sw() in
hal.mm.mmu_fault fails for L4T OS. setup_sw_s2 test case executes
gv11b_mm_mmu_fault_info_mem_destroy() twice consecutively. This makes
the test acquire hub_isr_mutex after being destroyed; leading to failure
This patch removes setup_sw_s2 case to resolve this error. Local GCOV
report shows that there is no change in coverage with this change.
Jira NVGPU-4780
Change-Id: I7b9470180d8f3146ac09318ed34281f723cc7e1a
Signed-off-by: Vedashree Vidwans <vvidwans@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2302212
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Currently, GVS is failing intermittently for some tests in nvgpu-runlist
and hal/mm/mmu_fault.
This patch resets gk20a structure at the end of each mmu_fault test. The
test_runlist_reload_ids and test_runlist_update_locked tests are
modified to use fifo support environment initialized for nvgpu-runlist
unit test.
Bug 2791755
Change-Id: I0b69b4f216f8f820b0a480ed76170b523b434bef
Signed-off-by: Vedashree Vidwans <vvidwans@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/2265676
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
