For bigger buffer allocation (e.g. 4GB, 5GB etc) from IOMMU heap, 70% of
the total time is consumed in cache cleaning. CUDA team confirmed that,
it is not always necessary to clean the CPU cache during allocation
flow. Hence provide an option to users of libnvrm_mem to skip cache
cleaning whenever required.
Bug 4628529
Change-Id: I9f4cdc930fcc673b69344f0167c8bc1378ec8d61
Signed-off-by: Ketan Patil <ketanp@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3192376
Reviewed-by: Sachin Nikam <snikam@nvidia.com>
Reviewed-by: Pritesh Raithatha <praithatha@nvidia.com>
GVS: buildbot_gerritrpt <buildbot_gerritrpt@nvidia.com>
A crash is observed when trying to access ch1 debugfs register in vi0
and vi1 folder. This is because the base reg address is not defined in
DT. This seems like an old legacy code which is ported over. The
registers accessed by this debugfs node are not longer directly
accessible using CCPLEX. While we wait on confirmation for this, adding
this check to not create debugfs node, if register base address is not
defined.
Bug 4626339
Change-Id: Ife714fa7e989bcaf187956920d757870e5fd9701
Signed-off-by: Mohit Ingale <mohiti@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3184450
(cherry picked from commit 1fb6c087c924a540b151183cb5be13311f2c4af1)
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3185971
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Create and register set and get operating point sysfs functions.
The operating point notification will contain the requested
operating point.
The nvcap resource manager will extract the operating point
from the notfication and send it to rce via the hsp command
interface.
In turn, rce will apply the operating point by adjusting the
functional clock frequencies for the following camera ip:
- rce hardware
- vi hardware
- isp hardware
- nvcsi
Jira CAMERASW-26378
Signed-off-by: Robert Kelly <rkelly@nvidia.com>
Change-Id: Ia814b2716d3738efb3cbc37307a267140b555f42
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3179593
(cherry picked from commit d259727a57ecf4b9463030b3418715891b3b1209)
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3189078
Reviewed-by: Semi Malinen <smalinen@nvidia.com>
Reviewed-by: Frank Chen <frankc@nvidia.com>
GVS: buildbot_gerritrpt <buildbot_gerritrpt@nvidia.com>
DCE FW will run the dma test using 512 bytes of transfer
between dram and tcm for 400 iterations and alu test
generating 100 prime numbers for 200 iterations.
DCE running above 600Mhz will take nearly 50msec
for each test.
Jira TDS-16211
Change-Id: I34570acd4db6b8103bd2451833b280dc8e32131a
Signed-off-by: vinodg <vinodg@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3192552
Reviewed-by: Mahesh Kumar <mahkumar@nvidia.com>
GVS: buildbot_gerritrpt <buildbot_gerritrpt@nvidia.com>
Reduce free memory reported by QueryHeapParams by 0.1% to avoid
OOM issues for Iovmm_heap. For the overall free memory reported
by the query heap params api, page align it to the previous page
to avoid any failures in allocattr due to page align of requested
size, when requested size is equal to reported free size. Update
allocattr api as well so that it checks the available free memory
before allocating.
Bug 4719292
Change-Id: I7ecd69216d58c36ee5d0750107546601400e722d
Signed-off-by: N V S Abhishek <nabhishek@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3165599
Reviewed-by: Ketan Patil <ketanp@nvidia.com>
GVS: buildbot_gerritrpt <buildbot_gerritrpt@nvidia.com>
Reviewed-by: Sachin Nikam <snikam@nvidia.com>
The current code can issue CDMA flushes (DMAPUT bumps) in the middle
of a job, before all opcodes have been written into the pushbuffer.
This can happen when pushbuffer fills up. Presumably this made sense
at some point in the past, but it doesn't anymore, as it cannot lead
to more space appearing in the pushbuffer as it is only cleaned full
jobs at a time.
Mid-job flushes can also cause problems, as in an extreme situation
(seen in practice), the hardware can run through the entire pushbuffer
including the prefix of a partially written job without the driver
being able to process any CDMA updates. This can cause the engine
MLOCK to be taken and held for extended periods as the tail of the
job is not yet available to hardware.
Bug 4606343
Signed-off-by: Mikko Perttunen <mperttunen@nvidia.com>
Change-Id: I68ca317150e917428d757d4ee8a68c2c15eaf8de
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3186138
GVS: buildbot_gerritrpt <buildbot_gerritrpt@nvidia.com>
Reviewed-by: Santosh BS <santoshb@nvidia.com>
For Linux v6.11, the 'pci_epc_event_ops' structure 'core_init' callback
was renamed to 'epc_init' and a new 'epc_deinit' callback was added.
Update the PCI endpoint drivers for Linux v6.11 by:
1. Adding tests to conftest to detect the presence of 'epc_init' and
'epc_deinit'.
2. Update the drivers as necessary to use the appropriate callback
depending upon what is present in the kernel.
Bug 4749580
Change-Id: Ife7e2d5d18332bf84f22142c9727dc1ece913c49
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3186686
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
- For NvRmHeap_GpuMem, we are switching from a carveout to huge pages
obtained from hugetlbfs.
- Allocate a handle from VA using get_user_pages, when allocation is
requested from GPU heap.
- Introduce a new field to indicate that the pages are allocated from
hugetlbfs. This field will be useful while returning the correct heap
while querying the handle params for a handle which was created using
hugetlbfs.
- Update the query heap API for GpuMem heap to return the huge pages
memory values from meminfo.
Bug 4510173
Change-Id: I0dbef4c4e95969f8e3975a6dc58f10255db0635e
Signed-off-by: Ketan Patil <ketanp@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3174720
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
This reverts the commit 20c43e79ff ("host1x-fence:Free
the kzalloc mem")
This change is causing the kernel crash when running the
ap_compue_level_1 on kernel 6.8 and kernel 6.9.
Reverting the change while more investigation is in progress
for proper fix of memory leak.
Bug 4779016
Change-Id: Ib7a1cf81cbc2f23d2e0a64115b4374da34a2961b
Signed-off-by: Laxman Dewangan <ldewangan@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3184970
GVS: buildbot_gerritrpt <buildbot_gerritrpt@nvidia.com>
The macro of_property_for_each_u32() was updated for Linux v6.11 to
remove two of the arguments that were only used internally in the macro
and do not need to be passed. Add a test to conftest to detect the
version of the of_property_for_each_u32() that is present in the kernel
and fix the f75308 driver accordingly.
Bug 4749580
Change-Id: If787a40d0af49989d93d9563dd6a84e906cb5209
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3184092
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
The Tegra virtual storage driver fails to build for Linux v6.11 because
of the following issues:
1. The functions blk_queue_logical_block_size(),
blk_queue_physical_block_size() and blk_queue_max_discard_sectors()
have been removed.
2. The function blk_queue_write_cache() has too many arguments for
Linux v6.11.
3. The flag QUEUE_FLAG_NONROT is not defined for Linux v6.11.
Add a dummy driver for this driver for Linux v6.11 until this driver is
fixed.
Bug 4311184
Bug 4749580
Change-Id: I56f811d6ee29dd0025a5dc199ae84e3fc0f37b21
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3162886
Reviewed-by: Laxman Dewangan <ldewangan@nvidia.com>
GVS: buildbot_gerritrpt <buildbot_gerritrpt@nvidia.com>
- Enable MODS kernel driver ADSP APIs
- Updated CONFIG_TEGRA_NVADSP to MODS_HAS_TEGRA
- get_handle call is invoked for each API
- To avoid regression on parallel WAT testing
- Fork nvidia-oot from nvidia-kernel
- Firmware APIs do not support 5.1 version.
- Updating the same IOCTLs as its being used by AON and ADSP tests.
Bug 4149877
Change-Id: I0063f6610c078395ce9ce3f348bbab360e7dc676
Signed-off-by: sitalluri <sitalluri@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3057290
Reviewed-by: Kuan Luo <kluo@nvidia.com>
GVS: buildbot_gerritrpt <buildbot_gerritrpt@nvidia.com>
Tested-by: John Lu <johlu@nvidia.com>
Reviewed-by: Bharat Nihalani <bnihalani@nvidia.com>
Reviewed-by: John Lu <johlu@nvidia.com>
Upstream commit cdd9a571b7d8 ("fs/proc: move page_mapcount() to
fs/proc/internal.h") made page_mapcount() an internal function for Linux
v6.11. This function is used by the NVMAP driver and so implement this
function in the NVMAP driver directly instead. Note that newer kernels
implement folio_entire_mapcount() for compound page counts where as
older kernels use compound_mapcount(). Use conftest to detect which of
these functions is present in the kernel.
Bug 4749580
Change-Id: I6504448727d6b3e9f2caa8581a66aad464ae2426
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3180097
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Platforms that implement suspend-resume as clean firmware
restart need to latch the shared mem address in every
suspend-resume cycle, but mbox value may be lost due to
module reset, so ensure that shared_mem_hwmbox is written
unconditionally in os_start flow.
Bug 3916054
Change-Id: I649d70871401475d66e0e50abeeaf4c347040119
Signed-off-by: Viswanath L <viswanathl@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3180285
GVS: buildbot_gerritrpt <buildbot_gerritrpt@nvidia.com>
Reviewed-by: Asha T <atalambedu@nvidia.com>
Reviewed-by: Dara Ramesh <dramesh@nvidia.com>
In Linux v6.11, the structure 'kernel_ethtool_ts_info' was added as a
copy of the existing 'ethtool_ts_info' for the kernel and the various
kernel functions were updated to use the new structure. This broke the
build for the nvethernet driver, so add a test to the conftest script to
detect the presence of this new strucutue and update the nvethernet
driver accordingly.
Bug 4749580
Change-Id: I2c5daa0132edee8c98eec3a66e19144484325ab0
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/3176868
GVS: buildbot_gerritrpt <buildbot_gerritrpt@nvidia.com>
Reviewed-by: Revanth Kumar Uppala <ruppala@nvidia.com>
