The offset units in the nvgpu_pramin_access_batched() code changes
midway through the function. In the first section it is treated as
bytes but then in the while-loop iterating over the PRAMIN window
and page_alloc_chunks it becomes an offset in words.
This patch leaves the offset field in bytes and converts to words
where needed.
Change-Id: Iba964171679dfc27645238b297ed467a450b5cbc
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1537079
Reviewed-by: Automatic_Commit_Validation_User
Reviewed-by: svccoveritychecker <svccoveritychecker@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
The call to __set_pd_level() for vidmem allocs had the wrong
length being passed in. This was a silent error since the subsequent
__set_pd_level() calls overwrote the bad mappings. However this
caused significantly more PDE/PTE writes than necessary since
each chunk could be mapped N times where N is the number of chunks
in an SGL.
Change-Id: Ied7247b70825dc91b9eea1c3350f4ef370ab1a52
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1537078
Reviewed-by: svccoveritychecker <svccoveritychecker@nvidia.com>
Reviewed-by: Konsta Holtta <kholtta@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Automatic_Commit_Validation_User
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reorganize HAL initialization to remove inheritance and construct
the gpu_ops struct at compile time. This patch only covers the
mm sub-module of the gpu_ops struct.
Perform HAL function assignments in hal_gxxxx.c through the
population of a chip-specific copy of gpu_ops.
Jira NVGPU-74
Change-Id: I289284e6e528fc7951c959c8765ccf9349eec33b
Signed-off-by: Sunny He <suhe@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1533351
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
In gm20b_tegra_postscale(), we use platform->railgate_lock to check if GPU is
railgated or not
But platform->railgate_lock was introduced only to prevent unrailgating in midst
of gk20a_do_idle() sequence
This lock is not the right way to check railgate status since it is still
possible to railgate GPU with this lock being held
Hence remove acquire/release of platform->railgate_lock from
gm20b_tegra_postscale()
Bug 1962265
Change-Id: I6208063de3fa77ed71e8fb0c011367fb66151193
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1536573
(cherry picked from commit 68bce66be338e48f4921f645b10b3fa5994fe1d4)
Reviewed-on: https://git-master.nvidia.com/r/1537297
Reviewed-by: svccoveritychecker <svccoveritychecker@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Bharat Nihalani <bnihalani@nvidia.com>
After setting 'Y' in disable_bigpage, in native SMMU case,
we could still see 64K GMMU pages beeing used.
Fixed the following:
- enforce disable_bigpage in nvgpu_vm_map
- update GPU characteristics so that new clients know
whether or not big pages are enabled. For instance
this may affect how CUDA requests memory mapping.
JIRA EVLR-1694
Change-Id: I62841096add3bd798c5c11090054f82c8a2be832
Signed-off-by: Thomas Fleury <tfleury@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1532429
Reviewed-by: Richard Zhao <rizhao@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: svccoveritychecker <svccoveritychecker@nvidia.com>
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Remove the mm.get_iova_addr() HAL and replace it with a new HAL
called mm.gpu_phys_addr(). This new HAL provides the real phys
address that should be passed to the GPU from a physical address
obtained from a scatter list. It also provides a mechanism by
which the HAL code can add extra bits to a GPU physical address
based on the attributes passed in. This is necessary during GMMU
page table programming.
Also remove the flags argument from the various address functions.
This flag was used for adding an IO coherence bit to the GPU
physical address which is not supported.
JIRA NVGPU-30
Change-Id: I69af5b1c6bd905c4077c26c098fac101c6b41a33
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1530864
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Pass struct gk20a pointer instead of struct device to
gk20a_wait_for_idle(). The code is not Linux specific and does not
need pointer to struct device.
Change-Id: I2cafd6c7db019c9de76b6e68a1ae73f0b4cea37d
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1533173
GVS: Gerrit_Virtual_Submit
Reviewed-by: Alex Waterman <alexw@nvidia.com>
ACCESS_ONCE() is used for making sure that in a given place of code
access a variable exactly once. It prevents compiler rearranging the
read from happening earlier.
Remove its use from cases where rearranging of the read does not
create problems.
Change-Id: I340f375e8fecc31f3a3fab543256069cb4c682dc
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1531649
GVS: Gerrit_Virtual_Submit
Reviewed-by: Alex Waterman <alexw@nvidia.com>
Refactor the sync_debugfs LTC HAL op so that the logic to enable
or disable LTC goes to common code nvgpu_ltc_sync_enabled() and
the LTC HAL set_enabled only performs the hardware register access.
Create a new common function nvgpu_init_ltc_support() to initialize
the LTC software variable, and move hardware initialization of LTC to
be called from it.
JIRA NVGPU-62
Change-Id: Ib1cf4f5b83ca3dac08407464ed56a732e0a33923
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1528262
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Fix kernel warnings for GPUs with real vidmem:
- dma.c: in nvgpu_dma_alloc_flags, ignore incoming flags when using vidmem,
since anything but NVGPU_DMA_NO_KERNEL_MAPPING will end up generating
kernel warnings, and the vidmem mapping functions ignore the other flags
anyway.
- gmmu.c: in __nvgpu_gmmu_update_page_table, use appropriate function for
memory type to retrieve physical address
Bug 1967748
Change-Id: I6fc01fd5f2c5cd7b81cba70ab59cc3c8fe4cda19
Signed-off-by: Peter Daifuku <pdaifuku@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1530877
Reviewed-by: Alex Waterman <alexw@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Move fields in struct gk20a related to interrupt handling into
Linux specific nvgpu_os_linux. At the same time move the counter
logic from function in HAL into Linux specific code, and two Linux
specific power management functions from generic gk20a.c to Linux
specific module.c.
JIRA NVGPU-123
Change-Id: I0a08fd2e81297c8dff7a85c263ded928496c4de0
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1528177
Reviewed-by: Automatic_Commit_Validation_User
Reviewed-by: Sourab Gupta <sourabg@nvidia.com>
GVS: Gerrit_Virtual_Submit
In the PD caching code use a non-contiguous DMA alloc for PAGE_SIZE
and below allocations. There's no need for using the special contig
pool of mem for these page sized allocs so wasting said mem can lead
us to OOM problems pretty quickly (think large sparse textures, for
example).
Also turn several pd_dbg() statements for printing OOM errors into
nvgpu_err()s since knowing exactly where an alloc fails is very
convenient.
Bug 200326705
Change-Id: Ib7c45020894d4bdd73cc92179ef707e472714d61
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1527294
Reviewed-by: Automatic_Commit_Validation_User
GVS: Gerrit_Virtual_Submit
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
gk20a_channel_release can still get called even if the open_channel call failed
(e.g., if we ran out of hw chids), in which case priv is null. Check for this
case and return if null.
Bug 1964531
Change-Id: I48bc88e4dbd88a1c30fc399de629d8f8b344cfd9
Signed-off-by: Peter Daifuku <pdaifuku@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1526544
Reviewed-by: Automatic_Commit_Validation_User
GVS: Gerrit_Virtual_Submit
Reviewed-by: Alex Waterman <alexw@nvidia.com>
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
If an error occurs during an attempt to perform a runtime_resume,
the runtime power management framework sets an error flag that
prevents further attempts to resume until the error is cleared.
nvgpu currently does not clear the flag, which causes nvgpu to
lock up if an error occurs during runtime_resume.
This change explicitly sets the device pm status to suspended
on error, which clears the error flag so that subsequent attempts
to resume will not be blocked.
Bug 200324790
Change-Id: I3c875453670d3691ab01cff90ce31e797296662a
Signed-off-by: Sunny He <suhe@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1526478
Reviewed-by: Automatic_Commit_Validation_User
GVS: Gerrit_Virtual_Submit
Reviewed-by: Alex Waterman <alexw@nvidia.com>
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reorganize HAL initialization to remove inheritance and construct
the gpu_ops struct at compile time. This patch covers the debug
and dbg_session_ops sub-modules of the gpu_ops struct.
Perform HAL function assignments in hal_gxxxx.c through the
population of a chip-specific copy of gpu_ops.
Jira NVGPU-74
Change-Id: Id51feeccbea91f884a6057efc680566a7d5d0b6d
Signed-off-by: Sunny He <suhe@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1514822
Reviewed-by: Automatic_Commit_Validation_User
GVS: Gerrit_Virtual_Submit
Reviewed-by: Alex Waterman <alexw@nvidia.com>
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
Add new routines for accessing and modifying PTEs in situ. They are:
__nvgpu_pte_words()
__nvgpu_get_pte()
__nvgpu_set_pte()
All the details of modifying a page table entry are handled within.
Note, however, that these routines will not build page tables. If a PTE
does not exist then said PTE will not be created. Instead -EINVAL will
be returned. But, keep in mind, a PTE marked as invalid still exists.
So this API can be used to mark an invalid PTE valid.
JIRA NVGPU-30
Change-Id: Ic8615f209a0c4eb6fa64af9abadcfb3b2c11ee73
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1510447
Reviewed-by: Automatic_Commit_Validation_User
Tested-by: Seema Khowala <seemaj@nvidia.com>
Reviewed-by: Seema Khowala <seemaj@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
A negative value in the timeout duration does not have any special uses,
so change the duration type to u32 (from just int). Delete some
unnecessary typecasts to int.
Also change MAX_SCHEDULE_TIMEOUT to ULONG_MAX in default gr idle timeout
because the value is in milliseconds instead of scheduling units and to
drop unnecessary Linux dependency.
Change-Id: I5cf6febd4f1cb00c46fe159603436a9ac3b003ac
Signed-off-by: Konsta Holtta <kholtta@nvidia.com>
Reviewed-on: https://git-master/r/1512565
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Ensure that all debug prints are consistent from chip to chip
and function to function. The following maps letters in the
debug print to their meaning:
C Mapping is cachable
v Mapping is volatile
S Mapping is sparse
P Mapping is private (VPR/WPR)
c Mapping is coherent
V Mapping is valid
JIRA NVGPU-30
Change-Id: Ia890af88677c3e6d3fdd8c4fe266158c35b8afcd
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master/r/1514903
GVS: Gerrit_Virtual_Submit
Reviewed-by: Deepak Nibade <dnibade@nvidia.com>
Tested-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
Add gv11b implementation of gpu_phys_addr() that checks the t19x
GMMU attributes struct to determine if L3 allocation should be
enabled. If L3 alloc is enabled then a special physical address
bit is set.
Add flag NVGPU_AS_MAP_BUFFER_FLAGS_L3_ALLOC to struct
nvgpu_as_map_buffer_ex_args so that User space can add a hint to
allocate buffer in L3 cache
Jira GPUT19X-10
Bug 200279508
Change-Id: I1bb9876a670b252980922aa50e3e69b802be137f
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master/r/1512602
GVS: Gerrit_Virtual_Submit
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
On some GPUs certain physical address bits have special meaning. This
patch adds support for setting those bits based on the GMMU attributes
struct.
Jira GPUT19X-10
Bug 200279508
Change-Id: I32b8a028be7fd62af06a60c393a8c9251de0ef3c
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master/r/1512600
GVS: Gerrit_Virtual_Submit
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
Use sysmem_coherent aperture if the buffer mappings are requested
to be IO coherent. Use sysmem_noncoherent aperture otherwise. This
is implemented by adding a new coherent field to the GMMU attrs
struct.
Jira GPUT19X-17
Bug 1651331
Bug 200283998
Change-Id: I5cfb71b5913d4db50ebf10331b19f5a4216456bf
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master/r/1514438
GVS: Gerrit_Virtual_Submit
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
In some cases page directories require less than a full page of memory.
For example, on Pascal, the final PD level for large pages is only 256 bytes;
thus 16 PDs can fit in a single page. To allocate an entire page for each of
these 256 B PDs is extremely wasteful. This patch aims to alleviate the
wasted DMA memory from having small PDs in a full page by packing multiple
small PDs into a single page.
The packing is implemented as a slab allocator - each page is a slab and
from each page multiple PD instances can be allocated. Several modifications
to the nvgpu_gmmu_pd struct also needed to be made to support this. The
nvgpu_mem is now a pointer and there's an explicit offset into the nvgpu_mem
struct so that each nvgpu_gmmu_pd knows what portion of the memory it's
using.
The nvgpu_pde_phys_addr() function and the pd_write() functions also require
some changes since the PD no longer is always situated at the start of the
nvgpu_mem.
Initialization and cleanup of the page tables for each VM was slightly
modified to work through the new pd_cache implementation. Some PDs (i.e
the PDB), despite not being a full page, still require a full page for
alignment purposes (HW requirements). Thus a direct allocation method for
PDs is still provided. This is also used when a PD that could in principle
be cached is greater than a page in size.
Lastly a new debug flag was added for the pd_cache code.
JIRA NVGPU-30
Change-Id: I64c8037fc356783c1ef203cc143c4d71bbd5d77c
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master/r/1506610
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
GVS: Gerrit_Virtual_Submit
Update the high level mapping logic. Instead of iterating over the
GPU VA iterate over the scatter-gather table chunks. As a result
each GMMU page table update call is simplified dramatically.
This also modifies the chip level code to no longer require an SGL
as an argument. Each call to the chip level code will be guaranteed
to be contiguous so it only has to worry about making a mapping from
virt -> phys.
This removes the dependency on Linux that the chip code currently
has. With this patch the core GMMU code still uses the Linux SGL but
the logic is highly transferable to a different, nvgpu specific,
scatter gather list format in the near future.
The last major update is to push most of the page table attribute
arguments to a struct. That struct is passed on through the various
mapping levels. This makes the funtions calls more simple and
easier to follow.
JIRA NVGPU-30
Change-Id: Ibb6b11755f99818fe642622ca0bd4cbed054f602
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master/r/1484104
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
GVS: Gerrit_Virtual_Submit
If devfreq governor is called from a timer during boot it is
possible that gm20b_tegra_postscale() gets called before we have
called gk20a_tegra_scale_init(). This change adds an explicit null ptr
check to prevent any null ptr dereference.
Bug 1954269
Change-Id: I4ebb6c702175b99af2862169f76623e24256b1ed
Signed-off-by: Peter Boonstoppel <pboonstoppel@nvidia.com>
Reviewed-on: https://git-master/r/1514262
GVS: Gerrit_Virtual_Submit
Reviewed-by: Konsta Holtta <kholtta@nvidia.com>
Changes to enable 64 subcontexts: 1 SYNC + 63 ASYNC
Currently all subcontexts with in a tsg can have only
single address space.
Add support for NVGPU_TSG_IOCTL_BIND_CHANNEL_EX for
selecting subctx id by client.
Bug 1842197
Change-Id: Icf56a41303bd1ad7fc6f2a6fbc691bb7b4a01d22
Signed-off-by: seshendra Gadagottu <sgadagottu@nvidia.com>
Reviewed-on: https://git-master/r/1511145
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
GVS: Gerrit_Virtual_Submit
When two or more apps ran simultaneously
First app context sets power_on flag & starts init
Other app context check the power_on flag
and try to use GPU without init completed
Which makes aother apps to assert
Added mutex to synchronize poweron access
Bug 200297265
Change-Id: Ie138f7f43bb0dd3304ed91ae3649a6a4947bee91
Signed-off-by: skadamati <skadamati@nvidia.com>
Reviewed-on: https://git-master/r/1511436
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
- Added nvgpu_flcn_clear_halt_intr_status() to
Wait for halt interrupt status clear by
clear_halt_interrupt_status() HAL within timeout
- Added gk20a_flcn_clear_halt_interrupt_status()
to clear falcon controller halt interrupt status
- Replaced flacon halt interrupt clear with
nvgpu_flcn_clear_halt_intr_status() method
NVGPU JIRA-99
Change-Id: I762a3c01cd1d02028eb6aaa9898a50be94376619
Signed-off-by: Mahantesh Kumbar <mkumbar@nvidia.com>
Reviewed-on: https://git-master/r/1511333
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
Convert a few calls from dev_*() logging to nvgpu_*(). This reduces
dependency to Linux specific struct device pointer.
JIRA NVGPU-38
Change-Id: Ib51a6b1287db25b7dd4d164aec3ac75fa2801ebf
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: https://git-master/r/1507929
GVS: Gerrit_Virtual_Submit
Make default big page size query a HAL op instead of per-platform
constant. This allows querying for default big page size without
accessing Linux specific gk20a_platform structure.
JIRA NVGPU-38
Change-Id: Ibfbd1319764fdae5fdb06700fb64d23f6f3dd01a
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: https://git-master/r/1507928
Reviewed-by: Automatic_Commit_Validation_User
GVS: Gerrit_Virtual_Submit
Remove gk20a support. Leave only gk20a code which is reused by other
GPUs.
JIRA NVGPU-38
Change-Id: I3d5f2bc9f71cd9f161e64436561a5eadd5786a3b
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: https://git-master/r/1507927
GVS: Gerrit_Virtual_Submit
Rename files, as they are not directly related to PMU.
They just send commands to PMU, similar to all other clock
change codes.
Bug 1921094
Change-Id: I4a67d4c950d995c68cfce464108cd36104f44080
Signed-off-by: Thomas Fleury <tfleury@nvidia.com>
Reviewed-on: https://git-master/r/1508820
GVS: Gerrit_Virtual_Submit
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Remove mem_config_idx from platform data, and instead let
HAL determine which memory configuration to use. For this
purpose, HAL may use PCI device identifiers, VBIOS version
and possibly RAMCFG strap register.
Bug 1929155
Change-Id: I9fcd67ff407382839ff81470789043fae1c81283
Reviewed-on: http://git-master/r/1497813
Signed-off-by: Thomas Fleury <tfleury@nvidia.com>
(cherry picked from commit 3f722945213bacfc5f6707059b9baccebd92cef1)
Reviewed-on: https://git-master/r/1506583
Reviewed-by: Automatic_Commit_Validation_User
GVS: Gerrit_Virtual_Submit
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
