The flag pmu->pg->golden_image_initialized is set to
true during initial GPU context creation and is not
cleared while the GPU goes into pm_suspend (during railgate).
Hence, when the GPU resumes after un-railgate it retains
the previous value which can cause ELPG to kick in immediately.
Due to this, when ELPG and Railgating are enabled, IDLE_SNAP
is seen for read access of gr_gpc0_tpc0_sm_arch_r reg.
To resolve this, if golden image is ready set the
pmu->pg->golden_image_initialized to suspend state during railgate,
to delay the early enable of ELPG. Add a new
pmu_init_golden_img_state hal in the NVGPU_INIT_TABLE_ENTRY.
This will be called after all the GR access is done and GPU resumes
completely after un-railgate. This hal will then check if
golden_image_initialized flag is in suspend state, it will set it
to ready state and then re-enable ELPG.
Bug 3431798
Change-Id: I1fee83e66e09b6b78d385bbe60529d0724f79e79
Signed-off-by: Divya <dsinghatwari@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2639188
Reviewed-by: Mahantesh Kumbar <mkumbar@nvidia.com>
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: svc-mobile-cert <svc-mobile-cert@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
Most of the Orin chip specific code is compiled out of safety build
with CONFIG_NVGPU_NON_FUSA and CONFIG_NVGPU_HAL_NON_FUSA. Remove the
config protection from Orin/GA10B specific code. Currently all code
is enabled. Code not required in safety will be compiled out later
in separate activity.
Other noteworthy changes in this patch related to safety build:
- In ga10b_ce_request_idle(), add a log print to dump num_pce so that
compiler does not complain about unused variable num_pce.
- In ga10b_fifo_ctxsw_timeout_isr(), protect variables active_eng_id and
recover under CONFIG_NVGPU_KERNEL_MODE_SUBMIT to fix compilation
errors of unused variables.
- Compile out HAL gops.pbdma.force_ce_split() from safety since this HAL
is GA100 specific and not required for GA10B.
- Compile out gr_ga100_process_context_buffer_priv_segment() with
CONFIG_NVGPU_DEBUGGER.
- Compile out VAB support with CONFIG_NVGPU_HAL_NON_FUSA.
- In ga10b_gr_intr_handle_sw_method(), protect left_shift_by_2 variable
with appropriate configs to fix unused variable compilation error.
- In ga10b_intr_isr_stall_host2soc_3(), compile ELPG function calls
with CONFIG_NVGPU_POWER_PG.
- In ga10b_pmu_handle_swgen1_irq(), move whole function body under
CONFIG_NVGPU_FALCON_DEBUG to fix unused variable compilation errors.
- Add below TU104 specific files in safety build since some of the code
in those files is required for GA10B. Unnecessary code will be
compiled out later on.
hal/gr/init/gr_init_tu104.c
hal/class/class_tu104.c
hal/mc/mc_tu104.c
hal/fifo/usermode_tu104.c
hal/gr/falcon/gr_falcon_tu104.c
- Compile out GA10B specific debugger/profiler related files from
safety build.
- Disable CONFIG_NVGPU_FALCON_DEBUG from safety debug build temporarily
to work around compilation errors seen with keeping this config
enabled. Config will be re-enabled in safety debug build later.
Jira NVGPU-7276
Change-Id: I35f2489830ac083d52504ca411c3f1d96e72fc48
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2627048
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Introduce HAL function gops.mssnvlink.get_links, this function retrieves
the number of nvlinks supported by the chip along with their base
addresses.
Update ga10b_mssnvlink_init_soc_credits to call mssnvlink.get_links.
Jira NVGPU-6641
Change-Id: I4ff857925f126bf41dc83eebc5723403244f66b0
Signed-off-by: Antony Clince Alex <aalex@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2618368
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Make ga10b_init_nvlink_soc_credits OS agnostic by replacing OS
specific functions with corresponding nvgpu wrappers. This function is now
assigned to gops.mssnvlink.init_soc_credits HAL.
Introduce nvgpu wrapper, nvgpu_io_map/unmap to map/unmap specified
physical address range.
Jira NVGPU-6641
Change-Id: I337bc75b8ec36552fe471bf5e42f62c19f67ed4a
Signed-off-by: Antony Clince Alex <aalex@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2618237
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Earlier, buffer metadata support was made dependent on compression.
However that is not required.
Update the enabled flag NVGPU_SUPPORT_BUFFER_METADATA setup for
various hals. Enable it for all from linux characteristics init.
Update REGISTER_BUFFER and GET_BUFFER_INFO ioctls to seggregate
the compile/runtime compression functionality.
If compression is disabled, return error in case comptags are
required else don't fail the REGISTER_BUFFER ioctl.
Bug 200767700
Change-Id: I3850ccc879f180c97b830fb3d652c094b9d28a5b
Signed-off-by: Sagar Kamble <skamble@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2614378
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
GPU HW expects physically contiguous addresses when clearing
the compression bit store in memory. Currently on hypervisor setup,
the DMA_ATTR_FORCE_CONTIGUOUS flag ensures contiguous IPA, but it
is not possible to ensure contiguous physical memory.Disable
compression on virtualized environments until physically contiguous
memory is feasible.
Buffer Metadata support is dependent on compression support.
Move the initialization of NVGPU_SUPPORT_BUFFER_METADATA flag to
common code where NVGPU_SUPPORT_COMPRESSION is initialized.
Bug 200780546
Change-Id: Id94bffc878e275a80948880f0475162d0bb4ddae
Signed-off-by: Tejal Kudav <tkudav@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2607830
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
- For older chips, nvgpu used to set some registers for ELPG
sequencer settings
- These writes are no longer required post-Turing as these
programming have been updated into the HW default values
itself and the register definitions have been changed to
help improve security as well.
- Set the pmu_setup_elpg HAL to NULL
Bug 200766930
Bug 3389932
Change-Id: I3820b14c8491f8180b2feb28cb38e23462546655
Signed-off-by: Divya <dsinghatwari@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2607599
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
- On pre-silicon platform, static pg will be
done by nvgpu driver. For this, retain structs
and HALs of static pg.
- Add the static pg support under pre-silicon code.
- On silicon, the static pg will be done by BPMP.
- Rename variables used in static pg for better
readability and consistency
Bug 200768322
JIRA NVGPU-6433
Change-Id: Ib31c0f83b751c2b1563a36bd51af78a0bd12a117
Signed-off-by: Divya <dsinghatwari@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2594801
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
common.fbp has two interfaces to initialize FBP:
1. Public API nvgpu_fbp_init_support
2. HAL fbp.fbp_init_support
nvgpu_fbp_init_support() is only used to initialize HAL
fbp.fbp_init_support. Remove the HAL and use the API directly.
JIRA NVGPU-6644
Change-Id: I2c455e09dbcf5e4fb1dc370b284e4f0d5c678b40
Signed-off-by: Tejal Kudav <tkudav@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2592047
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
This patch fixes a bug in the cbc initialization code for ga10b,
where it was erroneously assumed that a fixed ltc count of only one
should be used for historical reasons. For volta and later, the full
ltc count should be used in cbc-related computation.
Ensure
- CBC base address is 64K aligned
- CBC start address lies within CBC allocated memory
Check CBC is marked safe only for silicon platform.
Bug 3353418
Change-Id: I5edee2a78dc9e8c149e111a9f088a57e0154f5c2
Signed-off-by: Vedashree Vidwans <vvidwans@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2585778
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: Seema Khowala <seemaj@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
After CONFIG_UBSAN kernel compilation flag to know any shifting
cause overflow or not enablement ,this is identified.
The register "gr_fe_tpc_fs_r(gpc_index)" is read only after
Volta. The gops where we are computing the index is not needed.
Bug 200727116
Change-Id: Ib2306103389ba9df77fd59d012ec70e775104989
Signed-off-by: dt <dt@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2573296
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
There is HW specific limit on number of channel entries that can be
added for each TSG entry in runlist. Right now there is no checking
to enforce this from SW and hence if User binds more than supported
channels to same TSG, invalid TSG formation error interrupts are
generated.
Fix this by adding appropriate checks in below steps :
- Add new field ch_count to struct nvgpu_tsg to keep track of
channels bound to TSG.
- Define new hal gops.runlist.get_max_channels_per_tsg() to retrieve
HW specific maximum channel count per TSG.
- Implement the HAL for gk20a and gv11b chips, and assign new HALs for
all chips appropriately.
- Increment ch_count while binding the channel to TSG and decrement it
while unbinding.
- While binding channel to TSG, Check if current channel count is
already equal to max channel count. If yes, print an error and bail
out.
Bug 200763991
Change-Id: Ic5f17a52e0fb171d1c020bf4f085f57cdb95f923
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2582095
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: Konsta Holtta <kholtta@nvidia.com>
Reviewed-by: svc_kernel_abi <svc_kernel_abi@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
ECC counter structures are freed without removing the node from the
stats_list. This can lead to invalid access due to dangling pointers.
Update the ecc counter free logic to set them to NULL upon free, to
remove them from stats_list and free them by validation.
Also updated some of the ecc init paths where error was not propa-
gated to callers and full ecc counters deallocation was not done.
Now, calling unit ecc_free from any context (with counters alloc-
ated or not) is harmless as requisite checks are in place.
bug 3326612
bug 3345977
Change-Id: I05eb6ed226cff9197ad37776912da9dcb7e0716d
Signed-off-by: Sagar Kamble <skamble@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2565264
Tested-by: Ashish Mhetre <amhetre@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: svc_kernel_abi <svc_kernel_abi@nvidia.com>
Reviewed-by: Konsta Holtta <kholtta@nvidia.com>
Reviewed-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
Currently, cbc init and compression tests are failing because MMU marks
cbc to be not safe.
- Modify cbc.get_base_divisor hal to use ltc_count = 1 for Tegra devices
- Update fb.cbc_configure to write compbit_backing_size value to
fb_mmu_cbc_top register.
- After config confirm that CBC is marked safe.
Bug 3353418
Change-Id: I1e9c27f47f7bfcf476f2499231951382e2e8653d
Signed-off-by: Vedashree Vidwans <vvidwans@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2570550
Reviewed-by: Sami Kiminki <skiminki@nvidia.com>
Reviewed-by: svc_kernel_abi <svc_kernel_abi@nvidia.com>
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: V M S Seeta Rama Raju Mudundi <srajum@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
Enable nvriscv debug buffer feature in NVGPU.
Debug buffer is a feature to print the debug log from ucode onto console
in real time.
Debug buffer feature uses the DMEM, queue and SWGEN1 interrupt to share
ucode debug data with NVGPU.
Ucode writes debug message to DMEM and updates offset in queue to trigger
interrupt to NVGPU.
NVGPU copies the debug message from DMEM to local buffer to process and
print onto console.
Debug buffer feature is added under falcon unit and required engine
can utilize the feature by providing required param through public
functions.
Currently GA10B NVRISCV NS/LS PMU ucode has support for this feature
and enabled support on NVGPU side by adding required changes, with this
feature enabled, it is now possible to see prints in real time.
JIRA NVGPU-6959
Change-Id: I9d46020470285b490b6bc876204f62698055b1ec
Signed-off-by: mkumbar <mkumbar@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2548951
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: svc_kernel_abi <svc_kernel_abi@nvidia.com>
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Changes:
- This change will only init gsp software
state, nvgpu_gsp_bootstrap need to be called.
- CONFIG_NVGPU_GSP_SCHEDULER flag is created to
compile out the gsp scheduler code when needed.
- Created GSP engine reset which is needed when
ACR completed execution and need to load gsp fw.
NVGPU-6783
Signed-off-by: Ramesh Mylavarapu <rmylavarapu@nvidia.com>
Change-Id: I2ce43e512b01df59443559eab621ed39868ad158
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2554267
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Issue observed:
- In GA10B, it was observed that after recovery happens
ELPG does not engage.
- It was because, after CE reset, when nvgpu_submit_twod test
was run to engage ELPG, IDLE_FLIPPED_PWR_OFF signal was asserted.
- This means that when ELPG was engaged (engine is in PWR_OFF),
some idle signal flips (becomes non-idle) and this causes
IDLE_SNAP. After IDLE_SNAP is hit, ELPG will not engage further.
- After debugging from WAVES, it was observed that:
LCE0/LCE1 are not done with the reset sequence.
- The state of these LCE is RESET0. A pri request (pri read
to NV_CE_PCE_MAP register in CE) is seen that kicks it out of
RESET0. After this state, it goes through few states to update
some internal states (states RESET1/RESET2/PCE_MAP etc) and then
eventually settles down to IDLE state.
Solution:
- Read ce_pce_map_r register in recovery sequence (after ce reset).
- It is observed that when this read is added recovery is complete
and post that when nvgpu_submit_two test is executed, ELPG is engaging.
- This means that a pri read is needed after CE reset so that it settles
to idle state properly and post that ELPG can engage properly.
Bug 200734258
Change-Id: I5bb84921ca62a740fde81ffe6c29ccde4ebb341b
Signed-off-by: Divya Singhatwaria <dsinghatwari@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2554493
Reviewed-by: svcacv <svcacv@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: svc_kernel_abi <svc_kernel_abi@nvidia.com>
Reviewed-by: Mahantesh Kumbar <mkumbar@nvidia.com>
Reviewed-by: Seema Khowala <seemaj@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
On all chips except ga10b, the number of ROP, L2 units per FBP
were in sync, hence, their FS masks could be represented by a single
fuse register NV_FUSE_STATUS_OPT_ROP_L2_FBP. However, on ga10b, the ROP
unit was moved out from FBP to GPC and it no longer matches the number
of L2 units, so the previous fuse register was broken into two -
NV_FUSE_CTRL_OPT_LTC_FBP, NV_FUSE_CTRL_OPT_ROP_GPC.
At present, the driver reads the NV_FUSE_CTRL_OPT_ROP_GPC register
and reports incorrect L2 mask. Introduce HAL function
ga10b_fuse_status_opt_l2_fbp to fix this.
In addition, rename fields and functions to exclusively fetch L2 masks,
this should help accommadate ga10b and future chips in which L2 and ROP units
are not in same. As part of this, the following functions and
fields have been renamed.
- nvgpu_fbp_get_rop_l2_en_mask => nvgpu_fbp_get_l2_en_mask
- fuse.fuse_status_opt_rop_l2_fbp => fuse.fuse_status_opt_l2_fbp
- nvgpu_fbp.fbp_rop_l2_en_mask => nvgpu_fbp.fbp_l2_en_mask
The HAL ga10b_fuse_status_opt_rop_gpc is removed as rop mask is not
used anywhere in the driver nor exposed to userspace.
Bug 200737717
Bug 200747149
Change-Id: If40fe7ecd1f47c23f7683369a60d8dd686590ca4
Signed-off-by: Antony Clince Alex <aalex@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2551998
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
This CL covers the following modifications,
1) Added logic to skip the graphics unit specific sw context load
register write during context creation when MIG is enabled.
2) Added logic to skip the graphics unit specific sw method
register write when MIG is enabled.
3) Added logic to skip the graphics unit specific slcg and blcg gr
register write when MIG is enabled.
4) Fixed some priv errors observed during MIG boot.
5) Added MIG Physical support for GPU count < 1.
6) Host clk register access is not allowed for GA100.
So skipped to access host clk register.
7) Added utiliy api - nvgpu_gr_exec_with_ret_for_all_instances()
8) Added gr_pri_mme_shadow_ram_index_nvclass_v() reg field
to identify the sw method class number.
Bug 200649233
Change-Id: Ie434226f007ee5df75a506fedeeb10c3d6e227a3
Signed-off-by: Lakshmanan M <lm@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2549811
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Replace all nvgpu_next functions/structs either by 1) collapsing them
into nvgpu legacy functions/structs 2) renaming them as follows:
- nvgpu_next_*() => nvgpu_(ga10b/ga100)_*()
- nvgpu_next_*() => (ga10b/ga100)_*()
- nvgpu_next_*() => nvgpu_*() [only if this doesn't cause collision]
- nvgpu_next_*() = > nvgpu_*_extra()
Create hal.sim unit and move Ampere+ SIM code into it.
Jira NVGPU-4771
Change-Id: I215594a0d0df4bd663bd875a0d0db47bcb9ff6a2
Signed-off-by: Antony Clince Alex <aalex@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2548056
Reviewed-by: svc_kernel_abi <svc_kernel_abi@nvidia.com>
Reviewed-by: Mahantesh Kumbar <mkumbar@nvidia.com>
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
