Support new IOCTL to manage PMA stream meta data by adding below API
nvgpu_prof_ioctl_pma_stream_update_get_put()
Add nvgpu_perfbuf_update_get_put() to handle all the updates coming
from userspace and to pass all required information.
Add gops.perf.update_get_put() to handle all HW accesses required in
perf HW unit.
Add gops.perf.bind_mem_bytes_buffer_addr() to bind the available bytes
buffer while binding HWPM streamout.
Bug 2510974
Jira NVGPU-5360
Change-Id: Ibacc2299b845e47776babc081759dfc4afde34fe
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2406484
Reviewed-by: automaticguardword <automaticguardword@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: Antony Clince Alex <aalex@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Add two new IOCTL APIs to allocate/free pma stream. Add two new
functions to handle this :
nvgpu_prof_ioctl_alloc_pma_stream()
nvgpu_prof_ioctl_free_pma_stream()
Allocation of pma stream includes below steps :
- Initializing perfbuf VM
- Mapping PMA buffer into perfbuf VM
- Mapping PMA byte buffer into perfbuf VM
- Mapping PMA byte buffer to CPU virtual address space
Store all of above data in struct nvgpu_profiler_object for
reference. OS specific data is stored in struct
nvgpu_profiler_object_priv
Update HWPM streamout bind/unbind sequence to enable/disable perfbuf
respectively.
Also take care of releasing the pma stream resources in profiler object
close path if they are not explicitly released by user space by IOCTL
call.
Bug 2510974
Jira NVGPU-5360
Change-Id: I126633746cabc4e293c7ad7c49806866a897949d
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2406483
Reviewed-by: automaticguardword <automaticguardword@nvidia.com>
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: svc-mobile-cert <svc-mobile-cert@nvidia.com>
Reviewed-by: Antony Clince Alex <aalex@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
This CL covers the following code changes,
1) Added api to get the gr_config per gr_instance_id basis.
2) Added api to covert from gpu_instance_id to gr_instance_id.
3) Modified nvgpu_gr_exec_with_ret_for_instance() utility to handle
generic data return type.
JIRA NVGPU-5662
JIRA NVGPU-5663
Change-Id: I4ab732e15cdbda25672975f99e23b5e5d27decb0
Signed-off-by: Lakshmanan M <lm@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2413195
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Add macros to get current GR instance id and the pointer
nvgpu_gr_get_cur_instance_ptr()
nvgpu_gr_get_cur_instance_id()
This approach makes sure that the caller is getting GR instance pointer
under mutex g->mig.gr_syspipe_lock in MIG mode. Trying to access
current GR instance outside of this lock in MIG mode dumps a warning.
Return 0th instance in case MIG mode is disabled.
Use these macros in nvgpu instead of direct reference to
g->mig.cur_gr_instance.
Store instance id in struct nvgpu_gr. This is to retrieve GR instance
id in functions where struct nvgpu_gr pointer is already available.
Jira NVGPU-5648
Change-Id: Ibfef6a22371bfdccfdc2a7d636b0a3e8d0eff6d9
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2413140
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
gr_gv100_init_hwpm_pmm_register() and gr_gv100_set_pmm_register() right
now assume common chiplet stride for all sys/fbp/gpc and use common API
g->ops.perf.get_pmm_per_chiplet_offset() to get the stride.
Chiplet strides are same for all partitions only by chance, and future
chip might change that.
Hence add and use below 3 separate HALs to get appropriate strides.
g->ops.perf.get_pmmsys_per_chiplet_offset()
g->ops.perf.get_pmmgpc_per_chiplet_offset()
g->ops.perf.get_pmmfbp_per_chiplet_offset()
Also store sys/fbp/gpc perfmon count in struct gk20a after first query
instead of querying them again and again. Querying the counts from HW
is time consuming.
Bug 2510974
Jira NVGPU-5360
Change-Id: I186009221009780d561617c0cd6f535854db585f
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2413108
Reviewed-by: automaticguardword <automaticguardword@nvidia.com>
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: svc-mobile-misra <svc-mobile-misra@nvidia.com>
Reviewed-by: Antony Clince Alex <aalex@nvidia.com>
Reviewed-by: Alex Waterman <alexw@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Add separate flag gpu_dbg_gr to enable common.gr specific debugging.
Add this flag to all the existing debug logs that use gpu_dbg_fn or
gpu_dbg_info for debugging. Also add many other debugging logs that
might be helpful in debugging.
Removing debug log in gv11b_gr_init_get_nonpes_aware_tpc() as it dumps
too much data that does not seem useful.
Batch all interrupt enable functions in gr_init_setup_hw() together for
readability.
Jira NVGPU-5648
Change-Id: I0b857650122cdb1f974b452d28c26e7f142baf61
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2411740
Reviewed-by: automaticguardword <automaticguardword@nvidia.com>
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: svc-mobile-cert <svc-mobile-cert@nvidia.com>
Reviewed-by: Lakshmanan M <lm@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Remove current mc hals
- mc.reset()
- mc.enable()
- mc.disable()
- mc.reset_mask()
- mc.reset_engine()
- mc.reset_engine_enable()
Add new mc hals
- mc.enable_units(g, units, enable)
> enable/disable given unit(s)
- mc.enable_dev(g, dev, enable)
> enable/disable engine represented by given device pointer
- mc.enable_devtype(g, devtype)
> enable/disable all engines of given devtype
Move common mc intr functions to common/mc/mc_intr.c.
Add below common mc functions
- nvgpu_mc_reset_units(g, units)
> reset given logical OR of nvgpu unit bitmap
- nvgpu_mc_reset_dev(g, dev)
> reset given single engine via dev
> if engine is graphics, reset gpcs for nvgpu_next
- nvgpu_mc_reset_devtype(g, devtype)
> reset all engines of given devtype
> if devtype is graphics, reset gpcs for nvgpu_next
Bug 200648985
Bug 3109773
Change-Id: Idc67a14a0a7cde83de44fbfbec13007fead3ed5c
Signed-off-by: Vedashree Vidwans <vvidwans@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2408523
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
nvgpu_gr_init_support() right now executes each of its function for each
GR instance separately. Instead of looping for each function, move the
GR engine initialization sequence to a separate gr_init_support_impl()
and execute this function for each instance.
Update below functions to take nvgpu_gr pointer as parameter. These
functions need not worry about GR instance, instead they'll just operate
on provided instance pointer.
gr_init_setup_hw
gr_init_config
gr_init_setup_sw
gr_init_sm_id_config_early
gr_init_ctxsw_falcon_support
Add new static function gr_init_support_finalize() to set the ready
status and invoke waiters. Execute this per GR instance.
gr_init_ecc_init() and nvgpu_cg_elcg_enable_no_wait() are not needed to
be run per instance.
gr_init_ecc_init() will be later updated to allocate meta data for all
instances
Jira NVGPU-5648
Change-Id: Ia6860f2bdfe0080aebf8930266d3f51bfd805e36
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2410703
Reviewed-by: automaticguardword <automaticguardword@nvidia.com>
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: svc-mobile-cert <svc-mobile-cert@nvidia.com>
Reviewed-by: Lakshmanan M <lm@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
Get number of SMs from GR instance specific nvgpu_gr_config pointer
instead of global SM count in below functions :
nvgpu_gr_fs_state_init()
gv11b_gr_init_sm_id_config()
Update nvgpu_gr_config_get_gpc_skip_mask() to return 0 in case gpc_index
is greater than available gpc_count. This is not MIG specific, but based
on code review possible even today for existing chips.
See gm20b_gr_init_pd_skip_table_gpc()
Update nvgpu_gr_get_override_ecc_val() to return GR instance specific
value.
Execute gr_init_setup_hw() for each GR instance.
Disable below failing unit tests:
nvgpu_gr_fs_state.test_gr_fs_state_error_injection
nvgpu_gr_init.test_gr_init_hal_config_error_injection
Jira NVGPU-5648
Change-Id: Ie8f1c0c304c634756786d85facf336a5c9ae8195
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2410702
Reviewed-by: automaticguardword <automaticguardword@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: Lakshmanan M <lm@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
Execute gr_init_setup_sw() for each GR instance.
Update all of the functions called from this function to receive
nvgpu_gr pointer explicitly.
Separate out nvgpu_gr_zbc_init() call to gr_init_setup_sw() and rename
gr_init_ctx_and_map_zbc() to gr_init_ctx_bufs() for more clarity.
Call gr_init_ecc_init() from nvgpu_gr_init_support() since this does not
need to be executed per GR instance.
Initialize mutex etc in nvgpu_gr_alloc() for consistency.
Jira NVGPU-5648
Change-Id: I8e990e11458c05c1b53a4d6710cc2ec3545762a8
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2410701
Reviewed-by: automaticguardword <automaticguardword@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: Lakshmanan M <lm@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
Ctxsw state will be lost after gr reset. After gr reset
in recovery sequence, re-initialize ctxsw state to send
below fecs methods:
gr_fecs_method_push_adr_discover_image_size_v()
gr_fecs_method_push_adr_discover_pm_image_size_v()
gr_fecs_method_push_adr_discover_zcull_image_size_v()
gr_fecs_method_push_adr_discover_preemption_image_size_v()
Without these methods sent to ctxsw, fecs will generate
host error interrupts indicating mismatches in ctxsw
image. Above fecs methods needs to be sent even if they
are already sent during golden context creation.
Bug 3109773
Change-Id: I2aeb92da8fa1961903ab95ef90f47906a1bb32c4
Signed-off-by: Seshendra Gadagottu <sgadagottu@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2406685
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Expose below two new APIs from common.grmgr unit
nvgpu_grmgr_get_gr_num_gpcs() - get per instance number of GPCs
nvgpu_grmgr_get_gr_gpc_phys_id() - get physical GPC id for MIG engine
local id in corresponding instance
Execute gr_init_config() for each GR instance.
Add gr_config_init_mig_gpcs() to initialize GPC data in case MIG is
enabled. Separate out gr_config_init_gpcs() for legacy GPC data
initialization.
These functions will inititialize below data in struct nvgpu_gr_config:
max_gpc_count
gpc_count
gpc_mask
gpc_tpc_mask[gpc_count]
max_tpc_per_gpc_count
Rest of the values in struct nvgpu_gr_config are either based on above
values, or read from HW after setting GPC PRI window.
In gr_config_alloc_struct_mem(), rename total_gpc_cnt to total_tpc_cnt
since it represents total TPC count and not GPC. Remove use of temp3
variable since it does not give any idea on usage.
Jira NVGPU-5648
Change-Id: I646cac2ddc312e72b241b1b2a0e51a5cce141535
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2406390
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: Lakshmanan M <lm@nvidia.com>
Reviewed-by: automaticguardword <automaticguardword@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Add a macro to iterate over a device list; it is just a wrapper to
the nvgpu_list_for_each() macro. It lets code iterate over the
list of detected devices without being aware of the underlying
instance IDs.
This also removes the need to do a separate nvgpu_device_get()
and subsequent NULL checking. This will reduce overhead for
unit testing!
Change-Id: If41dbee30a743d29ab62ce930a819160265b9351
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2404914
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
The pbdma_id field stored in struct nvgpu_device is bitmask and not
bit position as implied by the name. This field is incorrectly used as
bit position in nvgpu_engine_disable_activity(), causing PRI timeout
errors during iGPU and dGPU shutdown path.
PRI timeout errors-
nvgpu: 17000000.gv11b gk20a_ptimer_isr:54 [ERR]
PRI timeout: ADR 0x0000308c READ DATA 0x00000000
Here the pbdma_id stored in struct nvgpu_device for runlist_0 on
gv11b is 0x3(bitmask corresponding to PBDMA_0 and PBDMA_1).
nvgpu_engine_disable_activity() interprets this as PBDMA_3 and adds
incorrect offset to access PBDMA_STATUS register, causing PRI error.
Modify nvgpu_engine_disable_activity() to treat pbdma_id as bitmask
and loop through set bits.
JIRA NVGPU-5991
Change-Id: Iaffb974cddaa375a329e70f3b5903b9ef2a222c4
Signed-off-by: Tejal Kudav <tkudav@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2397954
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
nvgpu_gr_engine_interrupt_mask() earlier returned mask of all GR engine
instance interrupts. During device refactor series, this got changed to
return interrupt of only first instance.
Change this again to return interrupt mask of all the GR engine
instances since common.mc unit does not yet support APIs to enable
interrupt of individual GR instance.
Update nvgpu_gr_get_syspipe_id() API to take gr_instance_id as parameter
instead of struct nvgpu_gr pointer. Definition of struct nvgpu_gr is not
available outside of common.gr unit.
Jira NVGPU-5648
Change-Id: I5320d1515eea6054150dc14706a16475bd650da7
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2405409
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: svc-mobile-cert <svc-mobile-cert@nvidia.com>
Reviewed-by: Lakshmanan M <lm@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
Rename gr_init_reset_enable_hw() to gr_init_prepare_hw() since this
function does not actually do reset, but just prepares the HW
after reset for other SW/HW initialization.
Add a new function gr_init_prepare_hw_impl() that executes per-instance
sequence to prepare GR hardware. Execute this inside
nvgpu_gr_exec_with_ret_for_each_instance().
Note that enabling GR engine interrupts in MC is still expected to
be done in one shot hence keep that code outside of
gr_init_prepare_hw_impl()
Remove redundant calls to gops.gr.init.fifo_access() and
enable_gr_interrupts() from gr_init_setup_hw().
gr_init_prepare_hw() does this already and executes before
gr_init_setup_hw()
Jira NVGPU-5648
Change-Id: If0b7207f80c2fb00d894afebce04b06b7b61d432
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2405408
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: svc-mobile-cert <svc-mobile-cert@nvidia.com>
Reviewed-by: Lakshmanan M <lm@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
API nvgpu_grmgr_get_gr_syspipe_id() right now traverses all the GPU
instances to find requested gr_instance_id. But logically,
gr_instance_id is always going to be same as gpu_instance_id since
nvgpu only supports one GR engine instace per GPU instance.
Fix this function by extracting GPU instance based on gr_instance_id
and then fetching syspipe_id stored for that GPU instance.
Jira NVGPU-5648
Change-Id: Ie7b86d765006353d0571e786a8089e7f75f779c3
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2405406
Reviewed-by: Automatic_Commit_Validation_User
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: Lakshmanan M <lm@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
Add a new header file <nvgpu/gr/gr_instances.h> that supports below
macros to execute various functions for GR instances
1) nvgpu_gr_exec_for_each_instance
Execute a function for each GR instance by configuring GR remap
window for that instance. Function being executed returns void.
2) nvgpu_gr_exec_with_ret_for_each_instance
Execute a function for each GR instance by configuring GR remap
window for that instance. Function being executed returns an error.
3) nvgpu_gr_exec_for_all_instances
Execute a function for all GR instances at once. For this GR remap
window needs to be disabled temporarily.
If CONFIG_NVGPU_MIG is disabled, all above macros will turn into simple
funciton calls.
If CONFIG_NVGPU_MIG is disabled or if runtime flag NVGPU_SUPPORT_MIG is
disabled, all above macros will turn into simple function calls that
configure single GR instance.
Separate out GR engine reset code into new API gr_reset_engine() and
execute it with nvgpu_gr_exec_with_ret_for_each_instance().
PROD values need to be loaded in legacy mode, hence call
nvgpu_cg_init_gr_load_gating_prod() inside
nvgpu_gr_exec_for_all_instances().
Rename gr_init_prepare_hw() to more appropriate
gr_reset_hw_and_load_prod()
Moe gops.gr.init.fifo_access() call to gr_init_reset_enable_hw().
Add new API nvgpu_grmgr_get_gr_syspipe_id() to query GR instance syspipe
id from common.grmgr unit. Add nvgpu_gr_get_syspipe_id() that returns
same value stored in nvgpu_gr struct.
Add cur_gr_instance field to struct nvgpu_gr to track current GR
instance being programmed under remap window.
Jira NVGPU-5648
Change-Id: I86920303427a6e6547ebf195daa37438365bb38e
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2403550
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
GPC reset is right now triggered from common.mc unit for NVGPU_NEXT.
Move the triggers to common code in common.gr unit. This way it is much
more cleaner to handle multiple GR instances (added in subsequent patch)
Hardcode GR engine instance to 0 for now since by default there is only
one GR engine instance.
Jira NVGPU-5648
Change-Id: I3fd4d0a50db5a8c4b3decf1df881af323cea50c1
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2403549
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Following changes are made in this patch.
1) Change unnamed structs within gpu_ops to named structs
with the prefix gops_*.
2) Each named struct gops_ are moved into a separate gops specific file
under include/nvgpu/gops/
3) struct gpu_ops is moved into a separate file include/nvgpu/gpu_ops.h
and all other dependent struct gops_* are included in this header.
4) Direct references to include/nvgpu/gops are removed from files as its enough
to include gk20a.h.
Change-Id: Ieb22cb853be567e3bef14f5f8a04674eebd902ea
Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2398776
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: Rajesh Devaraj <rdevaraj@nvidia.com>
Reviewed-by: Alex Waterman <alexw@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
The copyengine code to do big transfers or memsets supports a 64-bit
size. Each copy is done as a rectangle with either side being at most
2GB, so a size that does not align nicely is split into multiple ops. It
turns out that there are at most two of these ops, so structure the code
to not loop but do two ops explicitly.
The first copyengine operation works with the first chunk that is less
than two gigabytes long. That leaves the remaining size to be a multiple
of two gigabytes, so it's sufficient to do just another operation as a
2D rectangle whose width is two gigabytes; the remaining size determines
the height, i.e. the number of two-gig lines.
The loop did just this already, but now with at most two operations per
submit the required pushbuf length is seen more easily from the code.
Jira NVGPU-5172
Change-Id: I6bca3b1204db3b79e131898c07018a1337d85774
Signed-off-by: Konsta Hölttä <kholtta@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2402028
Reviewed-by: automaticguardword <automaticguardword@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: Alex Waterman <alexw@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Continue making the incoherent channel functionality more structured by
moving the worker thread business to one file and the channel watchdog
logic to another. This is channel-internal restructuring; the interface
to other units does not change.
The watchdog logic is called from the worker thread and as such these
are rather tightly coupled but it's possible to have the thread and not
the watchdog.
Jira NVGPU-5582
Change-Id: I70f334dd15c9aca0eed75393b99e2f080d133015
Signed-off-by: Konsta Hölttä <kholtta@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2398921
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
The max values that the Linux nvhost driver tracks are adding some
complexity to our wrapper APIs. Max values are used only for internal
submit syncpoint tracking, so implement that tracking in the sync code
by just storing the last value that the syncpoing will reach after all
jobs are complete.
The value is a simple u32. It's accessed from functions in the submit
path that already is serialized, so there's no worrying about atomic
modifications.
Previously nvhost_syncpt_set_min_eq_max_ext() was used to reset the
syncpoint when necessary. Now with the internal max value we'll use
nvhost_syncpt_set_minval(), so add a wrapper for it.
The maxval reported with the user syncpoint allocation is just the
current value at allocation time since no jobs have affected it yet;
there is no means for the kernel to track the max value of user
syncpoints.
Jira NVGPU-5506
Change-Id: I34672eaa7fe3af36b2fbac92d11babe2bc6a2d2b
Signed-off-by: Konsta Hölttä <kholtta@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2400635
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
The last steps to finish job cleanup for both deterministic and
nondeterministic submits are the same: put away preallocated job
resources that the job had consumed. Avoid duplicated code by moving
this code to a function that's shared with both paths.
Jira NVGPU-5998
Change-Id: Ic278b0bc8f0f05895f5c24340a60c1ce3eade0b3
Signed-off-by: Konsta Hölttä <kholtta@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2401468
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
The joblist cleanup lock exists to synchronize the submit job cleanup
and the abort cleanup that may run in separate threads concurrently.
This concurrency is no problem anymore, so delete the lock.
The lock was added in commit f1072a28be ("gpu: nvgpu: add worker for
watchdog and job cleanup") when the abort cleanup still went through
each job in the pending list and released their semaphores; ordinary job
cleanup from the worker thread also accesses the jobs. Commit d20a501dcb
("gpu: nvgpu: simplify job semaphore release in abort") deleted the
entire loop because the semaphore, if any, is now reset in one go (via
the "set_min_eq_max" ch sync op), but the lock stayed.
With aggressive sync destroy enabled the sync object under the cleanup
lock can still disappear if the job cleanup runs, but that's already
guarded with the sync lock.
Jira NVGPU-5998
Change-Id: I6554eb2065b003c6fdf83f66f97067b59aa272f5
Signed-off-by: Konsta Hölttä <kholtta@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2401467
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
common.gr unit exports a separate API nvgpu_gr_prepare_sw to
initialize some SW pieces required for nvgpu_gr_enable_hw().
A separate API is really unnecessary since same initialization
can be performed in nvgpu_gr_alloc().
Remove nvgpu_gr_prepare_sw() and HAL gops.gr.gr_prepare_sw().
Initialize falcon and interrupt structures in loop from
nvgpu_gr_alloc().
Move nvgpu_netlist_init_ctx_vars() from nvgpu_gr_prepare_sw() to
common init path since netlist parsing need not be done from
common.gr unit. It just needs to happen before nvgpu_gr_enable_hw().
Also, trigger nvgpu_gr_free() from gr_remove_support() instead
of OS specific paths. Also remove nvgpu_gr_free() calls from
probe error paths since nvgpu_gr_alloc is no longer called in
probe path.
Move interrupt and falcon data structure free calls to nvgpu_gr_free().
Also remove corresponding unit testing code that tests
nvgpu_gr_prepare_sw() specifically.
Update some unit tests to initialize ecc counters and netlist.
Disable some unit tests that fail for reasons unknown.
Jira NVGPU-5648
Change-Id: I82ec8160f76530bc40e0c11a9f26ba1c8f9cf643
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2400166
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Add new API nvgpu_grmgr_get_num_gr_instances() that returns number of
GR instance enumerated by GR manager. This just returns number of sys
pipes enabled since it is same as number of GR instances.
For consistency until common.gr supports multiple GR instances
completely, add a temporary macro NVGPU_GR_NUM_INSTANCES and set it
to 1. If this macro is changed to 0 (for local MIG testing), fall
back to use nvgpu_grmgr_get_num_gr_instances() to get enumerated number
of GR instances.
Use a for loop to initialize other variables of struct nvgpu_gr.
Remove unnecessary NULL check in nvgpu_gr_alloc() since struct gk20a
pointer can never be NULL in this path. Also remove corresponding unit
test code.
Jira NVGPU-5648
Change-Id: Id151d634a23235381229044f2a9af89e390886f2
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2400151
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
