We currently check that no job is pending before proceeding with
suspend. This prevents suspend, when we could simply disable and
preempt all channels. Moreover, pending jobs accounting is done
using pm_runtime usage count, which is not updated for GPUs with
pm_runtime disabled (e.g. vgpu).
Replaced the check on pm_runtime usage count, with a check on
gk20a handle usage count. Suspend is allowed when there is no
task inside a busy/idle sequence.
JIRA EVLR-1852
Change-Id: I79e71c8112182622dbd1c7c46cd84befa61a5c4d
Signed-off-by: Thomas Fleury <tfleury@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1552348
Reviewed-by: Automatic_Commit_Validation_User
GVS: Gerrit_Virtual_Submit
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
We currently check that engine is idle before proceeding with
suspend. This prevents suspend when we could simply disable and
preempt all channels. Moreover, doing such a check in virtualization
case, would require to query engine status from RM server, before
proceeding with suspend.
Removed check on engine idle for system suspend.
JIRA EVLR-1852
Change-Id: Ic6dc65af14b00f236db20038dfc04fa0928c1fe2
Signed-off-by: Thomas Fleury <tfleury@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1552347
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Delte the Linux headers and make some modifications to get rid of the
minor compilation issues that resulted.
- Add <linux/iommu.h> to os_linux.h
- Delete #if 0 code that "flushed" a buffer in gr_gk20a.c
- Delete FLUSH_CPU_DCACHE() macro
- Move the cache flush definitions to <nvgpu/linux/vm.h>
and include this header in sim_gk20a.c. This file will
not be used by QNX so this should be fine.
- Add <linux/pci_ids.h> to gp106/bios_gp106.c and
gp106/mclk_gp106.c.
- Move function to common/linux/dmabuf.h since it is a
dmabuf related function and uses a struct device pointer
as an argument.
JIRA NVGPU-30
Change-Id: I11f56b98524c7fac3efa91b4686592130e5f8a46
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1585510
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
clk_gk20a.h is used for dGPU and iGPU clocks. Because in gm20b the
clocks are owned by nvgpu, it has references to Linux CCF. Protect
the references behind #ifdef CONFIG_COMMON_CLK to compile it out on
non-Linux platforms.
JIRA NVGPU-259
Change-Id: I6ff095de7acaf1f828897cf3416acfaf050f8b51
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1586414
Reviewed-by: Automatic_Commit_Validation_User
Reviewed-by: Deepak Nibade <dnibade@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Add a sysfs node to allow root user to set PRI_FE_GFXP_WFI_TIMEOUT, for gp10b
only, in units of sysclk cycles. Store the set value in a variable, and write
the set value to register after GPU is un-railgated.
NV_PGRAPH_PRI_FE_GFXP_WFI_TIMEOUT is engine_reset after Bug 1623341.
Change default value to be specified in cycles, rather than time. This value
is almost the current value in cycles calculated each boot.
Bug 1932782
Change-Id: I0a4207e637cd1413a1be95abe2bcce3adccf76fa
Reviewed-on: https://git-master.nvidia.com/r/1540939
Signed-off-by: Jonathan McCaffrey <jmccaffrey@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1580999
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Make GPU characteristics flags specific to Linux code only. The
rest of driver is moved to using nvgpu_is_enabled() API.
JIRA NVGPU-259
Change-Id: I46a5a90bb34f170e9e755e7683be142ed6b18cce
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1583992
GVS: Gerrit_Virtual_Submit
In gk20a_fifo_tsg_unbind_channel(), we disable/preempt TSG, unbind one channel
from TSG, and then re-enable rest of the channels in TSG
But it is possible that TSG has already timed out due to some error and is
already disabled
If we re-enable all channels in such case, it can cause random issues right
after re-enabling faulted channel
Hence do not re-enable TSG if it has timedout
Since we disable all channels of TSG if one channel encounters fatal error,
it is safe to assume that TSG has timed out if one channel has timed out
Bug 1958308
Bug 200327095
Change-Id: I958ca6a2b408ff1338f2e551a79c072f1e203eda
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1585421
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
- in the native case, replace calls for init_cyclestats with
the gm20b version, as each chip had identical versions of the code.
- in the virtual case, use the vgpu version of the function in order
to get the new max_css_buffer_size characteristic set to the mempool
size.
JIRA ESRM-54
Bug 200296210
Change-Id: I475876cb392978fb1350ede58e37d0962ae095c3
Signed-off-by: Peter Daifuku <pdaifuku@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1578934
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Add max_css_buffer_size to gpu characteristics. In the virtual
case, the size of the cycle stats snapshot buffer is constrained
by the size of the mempool shared between the guest OS and the
RM server, so tools need to find out what is the maximum size
allowed.
In the native case, we return 0xffffffff to indicate
that the buffer size is unbounded (subject to memory availability),
in the virtual case we return the size of the mempool.
Also collapse native init_cyclestats functions to a single version,
as each chip had identical versions of the code.
JIRA ESRM-54
Bug 200296210
Change-Id: I71764d32c6e71a0d101bd40f274eaa4bea3e5b11
Signed-off-by: Peter Daifuku <pdaifuku@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1578930
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
- Clock arbiter has lot of linux dependent code
so moved clk_arb.c to common/linux folder &
clk_arb.h to include/nvgpu/clk_arb.h, this move
helps to unblock QNX.
- QNX must implement functions present
under clk_arb.h as needed.
JIRA NVGPU-33
Change-Id: I38369fafda9c2cb9ba2175b3e530e40d0c746601
Signed-off-by: Mahantesh Kumbar <mkumbar@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1582473
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
For SCG to work, smid numbering needs to be done
based on scg performance of tpcs. For gv11b and
gv11b vgpu, reuse gv100 function "gr_gv100_init_sm_id_table"
to do this.
Used local variable "index" to avoid multiple computations in
the function: gr_gv100_init_sm_id_table
index = sm_id + sm
Add deug info for printing initialized gpc/tpc/sm/global_tpc
indexs.
Bug 1842197
Change-Id: Ibf10f47f10a8ca58b86c307a22e159b2cc0d0f43
Signed-off-by: seshendra Gadagottu <sgadagottu@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1583916
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
We right now initialize czf_bypass value in gr_gp10b_init_preemption_state()
which is run at every rail ungate
And that results in any user specified value through sysfs getting lost after
railgate
To fix this, move initialization of czf_bypass to gk20a_init_gr_setup_sw() so
that it gets initialized only once
Add new HAL g->ops.gr.init_czf_bypass to initialize same and define it for
gp10b/gp106/vgpu-gp10b
Bug 2008262
Change-Id: I80a38ef527c86e32c6d64d0626b867239db9ea51
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1585224
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
PMU response(intr callback for messages) can run faster
than the kthread posting commands to PMU.
This causes the PMU message callback to skip important pmu
state change(which happens just after the PMU command is posted).
Solution:
State change should be triggered from only inside the intr callback.
Other places can only update the pmu_state variable.
This change also adds error check to print in case command post fails.
JIRA GPUT19X-20
Change-Id: Ib0a4275440455342a898c93ea9d86c5822e039a7
Signed-off-by: Deepak Goyal <dgoyal@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1583577
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
BAR1 test could occasionally fail when doing CPU write through userd
then reading back through BAR1. This is because nvgpu_smp_mb() only
guarantees ordering between cores.
Replaced with nvgpu_mb() to ensure the write will be visible to all
bus masters in the system.
JIRA EVLR-1959
Bug 200352099
Change-Id: Id002e73d135e0805fca2f153a6de77e210a7b226
Signed-off-by: Thomas Fleury <tfleury@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1582928
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Move vm_priv.h to <nvgpu/linux/vm.h> and rename nvgpu_vm_map()
to nvgpu_vm_map_linux(). Also remove a redundant unmap function
from the unmap path. These changes are the beginning of reworking
the nvgpu Linux mapping and unmapping code.
The rest of this patch is just the necessary changes to use the
new map function naming and the new path to the Linux vm header.
Patch Series Goal
-----------------
There's two major goals for this patch series. Note that these
goals are not achieved in this patch. There will be subsequent
patches.
1. Remove all last vestiges of Linux code from common/mm/vm.c
2. Implement map caching in the common/mm/vm.c code
To accomplish this firstly the VM mapping code needs to have the
struct nvgpu_mapped_buf data struct be completely Linux free. That
means implementing an abstraction for this to hold the Linux stuff
that mapped buffers carry about (SGT, dma_buf). This is why the
vm_priv.h code has been moved: it will need to be included by the
<nvgpu/vm.h> header so that the OS specific struct can be pulled
into struct nvgpu_mapped_buf.
Next renaming the nvgpu_vm_map() to nvgpu_vm_map_linux() is in
preparation for adding a new nvgpu_vm_map() that handles the
map caching with nvgpu_mapped_buf. The mapping code is fairly
straight forward: nvgpu_vm_map does OS generic stuff; each OS
then calls this function from an nvgpu_vm_map_<OS>() or the like
that does any OS specific adjustments/management.
Freeing buffers is much more tricky however. The maps are all
reference counted since userspace does not track buffers and
expects us to handle this instead. Ugh! Since there's ref-counts
the free code will require a callback into the OS specific code
since the OS specific code cannot free a buffer directly. THis
make's the path for freeing a buffer quite convoluted.
JIRA NVGPU-30
JIRA NVGPU-71
Change-Id: I5e0975f60663a0d6cf0a6bd90e099f51e02c2395
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1578896
GVS: Gerrit_Virtual_Submit
Reviewed-by: David Martinez Nieto <dmartineznie@nvidia.com>
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Move much of the remaining generic MM code to a new common location:
common/mm/mm.c. Also add a corresponding <nvgpu/mm.h> header. This
mostly consists of init and cleanup code to handle the common MM
data structures like the VIDMEM code, address spaces for various
engines, etc.
A few more indepth changes were made as well.
1. alloc_inst_block() has been added to the MM HAL. This used to be
defined directly in the gk20a code but it used a register. As a
result, if this register hypothetically changes in the future,
it would need to become a HAL anyway. This path preempts that
and for now just defines all HALs to use the gk20a version.
2. Rename as much as possible: global functions are, for the most
part, prepended with nvgpu (there are a few exceptions which I
have yet to decide what to do with). Functions that are static
are renamed to be as consistent with their functionality as
possible since in some cases function effect and function name
have diverged.
JIRA NVGPU-30
Change-Id: Ic948f1ecc2f7976eba4bb7169a44b7226bb7c0b5
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1574499
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
In t19x, gv11b semaphore read and write operations are
translated to host1x syncpoint read and write operations
using semaphore syncpoint shim aperture. Implement relevant
vgpu hal functions for this in fifo hal.
Jira EVLR-1571
Change-Id: I6296cc6e592ea991e1c01bc9662d02fb063ff3c7
Signed-off-by: Aparna Das <aparnad@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1516367
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Move functions to query preemption type names to the user of the
function: ioctl_channel.c. This removes a dependency to
<uapi/linux/nvgpu.h> from gr_gk20a.h.
JIRA NVGPU-259
Change-Id: I6cafda986eb4659fcfc1b19eac77e43aaaeaec76
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1577248
Function to query interleave name depends on IOCTL flag definition.
Move that code to fifo_gk20a.c to remove Linux dependency in header.
Change-Id: I6d6a80e550bf30973b2be09febc2347890b77d25
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1577249
Reviewed-by: Automatic_Commit_Validation_User
Reviewed-by: Alex Waterman <alexw@nvidia.com>
GVS: Gerrit_Virtual_Submit
Updated gk20a_gr_handle_class_error with
sub channel info, mme related info.
Also printing the correct method info from
isr_data->offset by left shifting it by 2.
Generated following hw definitions for gk20a/gm20b/gp10b/gp106
to dump relevant data in gk20a_gr_handle_class_error:
gr_trapped_addr_mme_generated_v
gr_trapped_addr_datahigh_v
gr_trapped_addr_priv_v
gr_trapped_data_lo_r
gr_trapped_data_mme_r
gr_trapped_data_mme_pc_v
Bug 2003671
Change-Id: I02e15ef16d7498b6a7dc2af547a14e84d570e8a7
Signed-off-by: seshendra Gadagottu <sgadagottu@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1574061
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
With coherency issues solved, it is no longer needed to allocate the fault
buffer in vidmem as a workaround.
JIRA: NVGPUGV100-36
Change-Id: I1c83e9bac61f27b75f38fce963899485afeed009
Signed-off-by: David Nieto <dmartineznie@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1582769
Reviewed-by: Automatic_Commit_Validation_User
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-by: svccoveritychecker <svccoveritychecker@nvidia.com>
GVS: Gerrit_Virtual_Submit
- Added method to load mem unlock binary into
nvdec falcon & execute to perform mem unlock
if VPR enabled.
- Updated .mem_unlock gv100 HAL to point
method gv100_fb_memory_unlock().
- Updated .mem_unlock gv11b HAL to NULL.
- Added vpr info hw registers
- Added nvdec enable hw register
Change-Id: Ia4bf820ae103baede679d300d1d390fd748c919a
Signed-off-by: Mahantesh Kumbar <mkumbar@nvidia.com>
(cherry picked from commit 2e176ad9d47316bf4d001692a2ae07e6c1fb1ccb)
Reviewed-on: https://git-master.nvidia.com/r/1573101
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
- Add methods to read/write falcon mailbox
at interface layer
- Created falcon mailbox read/write HAL
- Added HAL methods to read/write mailbox
- Added macro to get next block based on address
- Added macro to get IMEM tag using IMEM address
- Added ucode header format
Change-Id: I879b1df4538d403cac40fd4ed6e723190f62922c
Signed-off-by: Mahantesh Kumbar <mkumbar@nvidia.com>
(cherry picked from commit 30e8b76a7be9d9e6d8225bdc08e441f408692f63)
Reviewed-on: https://git-master.nvidia.com/r/1509469
Reviewed-by: Automatic_Commit_Validation_User
GVS: Gerrit_Virtual_Submit
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Tested-by: Terje Bergstrom <tbergstrom@nvidia.com>
- Created "mem_unlock" HAL under fb to support memory
unlock
- Called as part of gk20a_finalize_poweron() if memory unlock
support needed by checking HAL
- Assigned "mem_unlock" HAL to NULL for chips which don't
need memory unlocks.
Change-Id: I68d0910f15d293feaacfcbf6bd17ecccd3b5219d
Signed-off-by: Mahantesh Kumbar <mkumbar@nvidia.com>
(cherry picked from commit 586894eb84860bbbe4c75dae4715bdf27432a480)
Reviewed-on: https://git-master.nvidia.com/r/1564703
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
- Added "nvgpu_flacon nvdec_flcn" member to gk20a
- Added base address & flacon id of NVDEC falcon
- Included nvdec falcon to access common falcon code
- Enabled nvdec falcon support for GP106
- Disabled nvdec falcon support for iGPU
- Made call to enable nvdec falcon support if supported
Change-Id: Ia928d082275a720e4e8c6852384e489c8ec444f8
Signed-off-by: Mahantesh Kumbar <mkumbar@nvidia.com>
(cherry picked from commit 3d80aeff295bad8365af6022555ad151f1a32cf6)
Reviewed-on: https://git-master.nvidia.com/r/1564305
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Patch buffer can hold 128 u32 entries. Each patch write
takes total of 2 u32 entries, 1 u32 for addr and 1 u32
for data. Ideally 64 entries could be written before buffer
overflows. Driver patch some things when creating the channel,
and later when context switch type is changed after channel is loaded.
Reset patch_ctx.data_count before beginning patch
write otherwise system might not be in a state to accept all
patch writes even if patch buffer has valid entries.
If the patch buffer has non-zero entries, then the patch buffer
would be read and all pri writes would be sent out. Once done,
ucode updates the main header patch buffer count to 0.
Without this fix, below priv errors seen on t186 platforms
SYS Write error for ADR 0, INFO 0d000200 and CODE badf1100
Error info decodes as:
NV_PPRIV_SYS_PRIV_ERROR_INFO R[0x00122128]
SUBID [29:24] 13 (?)
LOCAL_ORDERING [22:22] 0 (I)
PRIV_LEVEL [21:20] 0 (I)
SENDING_RS [17:12] 0 (I)
PENDING [ 9: 9] 1 (?)
ORPHAN [ 8: 8] 0 (I)
PRIV_MASTER [ 5: 0] 0 (I)
Ctxsw ucode(subid 13 i.e. 0xd) makes only few pri transactions
at priv level 0. Patch buffer pri writes are one of those.
Bug 200350539
Change-Id: If9e71b5fef4d85600d72a8a633a082d9261c3e1b
Signed-off-by: Seema Khowala <seemaj@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1581591
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Convert the work_struct used by the vidmem background clearing to
a thread to make it more cross platform. The thread waits on a
condition variable to determine when work needs to be done. The
signal comes from the DMA API when it enqueues a new nvgpu_mem that
needs clearing.
Add logic for handling suspend: the CE cannot be accessed while
the GPU is suspended. As such the background thread must be paused
while the GPU is suspended and the CE is not available.
Several other changes were also made:
o Move the code that enqueues a nvgpu_mem from the DMA API
code to a function in the VIDMEM code.
o Move nvgpu_vidmem_get_pending_alloc() to the Linux specific
code as this function is only used there. It's a trivial
function that QNX can easily implement as well.
o Remove the was_empty logic from the enqueue. Now just always
signal the condition variable when anew nvgpu_mem comes in.
o Move CE suspend to after MM suspend.
JIRA NVGPU-30
JIRA NVGPU-138
Change-Id: Ie9286ae5a127c3fced86dfb9794e7d81eab0491c
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1574498
Reviewed-by: Automatic_Commit_Validation_User
GVS: Gerrit_Virtual_Submit
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
