In gk20a_ctrl_ioctl_gpu_characteristics() we right now just calculate
GPC mask in s/w and return to user space
But this could give incorrect result as any GPC could be floorswept
in h/w
Add gops.fuse.fuse_status_opt_gpc() to read GPC floorsweep status
from fuse
Add gops.gr.get_gpc_mask() to get actual GPC mask from h/w
Set these HALs only for dGPUs right now. Fuse register to read GPC
mask is not yet supported in simulation and hence simulation boot fails
These HALs will be set for iGPU once simulation issue is resolved
Use gops.gr.get_gpc_mask() if it is defined in
gk20a_ctrl_ioctl_gpu_characteristics() to send the actual GPC mask
to user space
Jira NVGPUT-132
Change-Id: I3b552de07883328fcfa41d4334ec0d777e04bdd3
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1822811
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
-Created struct nvgpu_acr to hold acr module related member
within single struct which are currently spread across multiple structs
like nvgpu_pmu, pmu_ops & gk20a.
-Created struct hs_flcn_bl struct to hold ACR HS bootloader specific members
-Created struct hs_acr to hold ACR ucode specific members like bootloader data
using struct hs_flcn_bl, acr type & falcon info on which ACR ucode need to run.
-Created acr ops under struct nvgpu_acr to perform ACR specific operation,
currently ACR ops were part PMU which caused to have always dependence
on PMU even though ACR was not executing on PMU.
-Added acr_remove_support ops which will be called as part of
gk20a_remove_support() method, earlier acr cleanup was part of
pmu remove_support method.
-Created define for ACR types,
-Ops acr_sw_init() function helps to set ACR properties
statically for chip currently in execution & assign ops to point to
needed functions as per chip.
-Ops acr_sw_init execute at early as nvgpu_init_mm_support calls acr
function to alloc blob space.
-Created ops to fill bootloader descriptor & to patch WPR info to ACR uocde
based on interfaces used to bootstrap ACR ucode.
-Created function gm20b_bootstrap_hs_acr() function which is now common
HAL for all chips to bootstrap ACR, earlier had 3 different function for
gm20b/gp10b, gv11b & for all dgpu based on interface needed.
-Removed duplicate code for falcon engine wherever common falcon code can be used.
-Removed ACR code dependent on PMU & made changes to use from nvgpu_acr.
JIRA NVGPU-1148
Change-Id: I39951d2fc9a0bb7ee6057e0fa06da78045d47590
Signed-off-by: Mahantesh Kumbar <mkumbar@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1813231
GVS: Gerrit_Virtual_Submit
Reviewed-by: svc-misra-checker <svc-misra-checker@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>
gk20a/hal.c depends on HAL init functions in all chips. But all chips
also depend on gk20a. That creates a circular dependency. In order to
solve the above, move gpu_init_hal and gk20a_detect_chip to
common/init/hal_init.c. These methods are declared in
include/nvgpu/hal_init.h. Also, the above methods are renamed to
nvgpu_init_hal and nvgpu_detect_chip respectively.
Jira NVGPU-613
Change-Id: Ib0df90287d4491571e4751475739b75fabd1041b
Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1827576
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Move OS agnostic parts of vgpu clk code out of os/linux specific
path. This includes implementation sending rpc commands to
RM Server. Move Linux specific vgpu clk code to platform vgpu files
keeping it consistent with native implementation.
Bug 2363882
Jira EVLR-3254
Change-Id: I0aae014ef16415bb356c81e9bfd76bc65206d9fd
Signed-off-by: Aparna Das <aparnad@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1820674
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Restrict access to devfreq when there are active set requests in the
clk_arbiter. We make the following changes in the patch.
1) Add a global counter in the struct gk20a named clk_arb_global_nr which
is used to track the number of active set requests in the system.
2) Anytime a set request is successfully made by the userspace,
clk_arb_global_nr is incremented by 1 and during the completion of
request(releasing the corresponding file handle), clk_arb_global_nr is
decremented by 1.
3) gk20a_scale_target(invoked by devfreq to set the new frequency based
on load) atomically checks clk_arb_global_nr. If the value = 0, the
code simply continue or else if its > 0, it quits thus making devfreq
requests mutually exclusive with the clk_arbiter.
Bug 2061372
Change-Id: I5d19de03e45520f4ff8fccb97b1f1589d04c8ab8
Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1790002
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
This patch constructs clk_arbiter specific code for gp10b as well as
gv11b and does the necessary plumbing in the clk_arbiter code. The
changes made are as follows.
1) Constructed clk_arb_gp10b.* files which add support for clk_arb
related HALS including the nvgpu_clk_arb_init and nvgpu_clk_arb_cb.
This doesn't have support for debugfs nor the VFUpdateEvent yet and
consequently no support for arb->notifications.
2) Added gpcclk specific variables corresponding to every gpc2clk in
a given clk_arb related struct.
3) Linux specific support_clk_freq_controller is assigned true in
platform_gp10b.c and platform_gv11b.c files.
4) Incremented the clk_arb_worker.put atomic variable during
worker_deinit so as to allow the worker thread to be stopped.
5) Added the flag clk_arb_events_supported as part of struct
nvgpu_clk_arb. This flag is used to selectively account for the extra
refcounting present in OS specific code i.e.
nvgpu_clk_arb_commit_request_fd. For igpus, the extra refcount is
reduced during nvgpu_clk_arb_release_completion_dev.
Bug 2061372
Change-Id: Id00acb106db2b46e55aa0324034a16a73723c078
Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1774281
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Expose the linux specific clock implementations via the HAL
interface to allow nvgpu to use the controls globally. This patch
does the following.
1) Implement a new ops interface and a corresponding linux specific
implementation for allowing nvgpu to iterate through a list of
available clock frequencies via nvgpu_linux_clk_get_f_points().
2) Implement nvgpu_linux_clk_get_range().
Bug 2061372
Change-Id: I7ce9a999dbdcd9fafcc84301af148545f6ca97a9
Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1774280
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Using two separate locks (poweron_lock and poweroff_lock)
allows concurrent gpu power-on and power-off. This shall
not happen as driver won't be able to maintain correct
gpu state.
Use a single power_lock to manage gpu power state. This
lock will be used to manage gpu power state from multiple
triggers like gpu idle, gpu gc-off, etc.
JIRA NVGPU-1100
Change-Id: Ia9b4aeda024a5844ae9f182d453cd6341876680a
Signed-off-by: Nitin Kumbhar <nkumbhar@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1827812
Reviewed-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-by: svc-misra-checker <svc-misra-checker@nvidia.com>
Reviewed-by: Debarshi Dutta <ddutta@nvidia.com>
Reviewed-by: Automatic_Commit_Validation_User
GVS: Gerrit_Virtual_Submit
Reviewed-by: Alex Waterman <alexw@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
1) Move channel_sync_gk20a.* from gk20a/ to common/ directory as they
donot program any hardware registers. Also as an add-on rename
channel_sync_gk20a.* to channel_sync.* and update the headers
in required files.
2) Rename the struct gk20a_channel_sync to struct nvgpu_channel_sync. Also,
corresponding syncpt and semaphore versions of the struct alongwith
related methods are renamed by removing "gk20a" from their names and
adding "nvgpu".
3) Add misra-c cleanups
Jira NVGPU-1086
Change-Id: I4e0e21803ca3858dd7a5fc4d2454dba1f1bfcecd
Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1812594
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Move implementation of MC HAL to common/mc. Also bump gk20a
implementation to gm20b.
gk20a_mc_boot_0 was used via a HAL, but we have only one possible
implementation. It also has to be anyway called directly to detect
which HALs to assign, so make it a true common function.
mc_gk20a_handle_intr_nonstall was also used only in os/linux/intr.c
so move it there.
JIRA NVGPU-954
Change-Id: I79aedc9158f90d578db0edc17b714617b52690ac
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1813519
Reviewed-by: svc-misra-checker <svc-misra-checker@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Konsta Holtta <kholtta@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
NVGPU_DBG_GPU_IOCTL_REG_OPS currently doesn't return if the ctx was
resident in engine or not.
Regops are broken down into batches of 128 and each batch is executed
together. Since there only 32 bits were available in IOCTL args, returning
is ctx was resident isn't possible for all batches.
Hence return if the ctx was resident for the first batch.
Bug 200445575
Change-Id: Iff950be25893de0afadd523d4ea04842a8ddf2af
Signed-off-by: Anup Mahindre <amahindre@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1812975
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
page_idx is an element of the struct nvgpu_semaphore_pool, defined in
include/nvgpu/semaphore.h file.
page_idx can not be negative so changing it from int to u64 and its
related changes in various files.
This also fixes MISRA 10.4 violations in these files.
Jira NVGPU-992
Change-Id: Ie9696dab7da9e139bc31563783b422c84144f18b
Signed-off-by: Sai Nikhil <snikhil@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1801632
Reviewed-by: Adeel Raza <araza@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Alex Waterman <alexw@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Add sysfs to indicate if GPU has ever been powered on.
This allows an application to avoid booting dGPU, when there is no
need for it. For instance the fan controller can check this value,
and refrain from booting the GPU, only to check the temperature.
This will save power, and also avoid locking some registers which
currently prevent from using VBIOS update tool.
Bug 200408936
Change-Id: I6344001e91e9eee8762ff8b227ca6698d8c6786f
Signed-off-by: Thomas Fleury <tfleury@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1802098
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Below regops HALs are not being called from anywhere, so remove them
gops.regops.get_runcontrol_whitelist_ranges()
gops.regops.get_runcontrol_whitelist_ranges_count()
gops.regops.get_qctl_whitelist_ranges()
gops.regops.get_qctl_whitelist_ranges_count()
HAL gops.regops.apply_smpc_war() is unimplemented for all the chips, and it
was originally only needed for gk20a which is not unsupported
So remove this HAL and its call too
Jira NVGPU-620
Change-Id: Ia2c74883cd647a2e94ee740ffd040a40c442b939
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1813106
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
This migrates the common DMA code (os agnostic) to the
common directory. This new unit will be the common DMA
allocator that lets users allocate SYSMEM, VIDMEM, or
either. Other units will be responsible for actually
handling the mechanics of allocating VIDMEM or SYSMEM.
Also update the names of the DMA related files so that
tmake doesn't complain about duplicate C file names. To
do this call the common DMA file dma.c and prepend the
OS to the other DMA files. So now we have:
common/mm/dma.c
os/posix/posix-dma.c
os/linux/linux-dma.c
JIRA NVGPU-990
Change-Id: I22d2d41803ad89be7d9c28f87864ce4fedf10836
Signed-off-by: Alex Waterman <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1799807
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Move debugfs related code of therm from common driver to linux
specific part of the driver. gp106_therm_debugfs_init()
is updated to use nvgpu_os_linux_ops.
This also affects gv100 as gp106_therm_debugfs_init is used
for gv100 as well.
JIRA NVGPU-603
Change-Id: Ia293d14599bc0c91fd1e917b5a430bd8f3d96e56
Signed-off-by: Nitin Kumbhar <nkumbhar@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1797906
GVS: Gerrit_Virtual_Submit
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Move linux dependencies and CONFIG_DEBUG_FS to linux specific
code from common driver for gp106 clk debugfs. There is no
code change in functions moved from gp106/clk_gp106.c.
It uses nvgpu_os_linux_ops to add gp106 specific clk debugfs
ops. The linux specific part of nvgpu driver uses this op
to initialize gp106 clk debugfs.
As gv100 also uses gp106 clk debugfs ops, set up os ops for
gv100.
JIRA NVGPU-603
Change-Id: Ib55ef051b13366e5907e1d05376bb18bf42c8653
Signed-off-by: Nitin Kumbhar <nkumbhar@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1797904
Reviewed-by: svc-misra-checker <svc-misra-checker@nvidia.com>
Reviewed-by: Deepak Nibade <dnibade@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Adding a comptag_mem_deduct in the platform_gk20a has certain problems:
1) It's not really convenient for platform users to configure it.
2) All products using the same GPU have to share the same configuration.
So this patch moves this comptag_mem_deduct from struct platform_gk20a
to struct gr_gk20a (per device). And it adds an sysfs node for products
or platform users to easily configure from user space.
Note: The comptag memory will not be allocated until the GPU driver goes
through the final poweron routine. So the user space has a small window
to configure this sysfs node.
Bug 2327574
Bug 2284925
Change-Id: Ie7d00b082704e422645c0ea254b59e22f9fc3b7f
Signed-off-by: Nicolin Chen <nicolinc@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1810334
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
g->clk_arb is currently initialized as a part of gk20a_finalize_poweron().
Any subsequent call to gk20a_finalize_poweron reinitializes the clk_arb
and leading to memory leaks. This is resolved by protecting the
g->clk_arb initialization with a mutex clk_arb_enable_lock in struct
gk20a. We skip initializing the g->clk_arb if its not NULL.
Bug 2061372
Change-Id: I59158e0a5e4c827fdbd6d9ea2d04c78d0986347a
Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1811650
Reviewed-by: Deepak Nibade <dnibade@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
- Remove the usage of NVGPU_CTXSW_FILTER_ISSET splattered
across nvgpu, and replace with a MACRO defined in common code.
The usage is still inside Linux, but this helps the
subsequent unification efforts, e.g. to unify the fecs trace
path.
- Remove "uapi/linux/nvgpu.h" from common code.
EVLR-3078
Change-Id: I60b0e1627576a4b255671d58530d8c773ea6154c
Signed-off-by: Vaibhav Kachore <vkachore@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1803210
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
- Remove the usage of nvgpu_ctxsw_trace_entry splattered
across nvgpu, and replace with a struct defined in common code.
The usage is still inside Linux, but this helps the
subsequent unification efforts, e.g. to unify the fecs trace
path.
- Remove "asm/barrier.h" as "nvgpu/barrier.h" is already
included.
EVLR-3078
Change-Id: Iabfb105b891b0078ed326a8047ef14ebe1888cf2
Signed-off-by: Vaibhav Kachore <vkachore@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1803208
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
GM20B GPCPLL registers are accessed via sys address space with the
exception of the last dvfs2 register that can be accessed only through
bcast address. However, in debugfs loop dvfs2 sys address should be
used as loop terminator; otherwise, loop body is never executed, and
GPCPLL registers debugfs node is broken. Fixed it in this commit.
Change-Id: Ifea839757fda7d97b2c9238178227ab198526ab0
Signed-off-by: Alex Frid <afrid@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1808784
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Currently only cde uses nvgpu_os_linux_ops to set up linux
specific ops. Move nvgpu_os_linux_ops of a gpu to a common
file so that those can be reused for other os ops of that
gpu.
JIRA NVGPU-603
Change-Id: Icf1ff275d3832229137f730fe8183b8015e82673
Signed-off-by: Nitin Kumbhar <nkumbhar@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1797902
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
In nvgpu repository, we have multiple accesses to methods in
pmu_gk20a.h which have register accesses. Instead of directly invoking
these methods, these are now called via HALs. Some common methods such
as pmu_wait_message_cond which donot have any register accesses
are moved to pmu_ipc.c and the method declarations are moved
to pmu.h. Also, changed gm20b_pmu_dbg to
nvgpu_dbg_pmu all across the code base. This would remove all
indirect dependencies via gk20a.h into pmu_gk20a.h. As a result
pmu_gk20a.h is now removed from gk20a.h
JIRA-597
Change-Id: Id54b2684ca39362fda7626238c3116cd49e92080
Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1804283
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
The comptag allocates memory based on the available total RAM, which
theoretically should be the MAX physical RAM size however practically
should deduct the part being taken by the running system. Otherwise,
the taken memory part will never get used and wasted.
This change adds a comptag_mem_deduct to each platform and to allow
them to assign the deductible value based on their own use cases so
as to save memory.
Bug 2327574
Bug 2284925
Change-Id: I124e20a66183c22723c34a7ec6ce34832c12f02e
Signed-off-by: Nicolin Chen <nicolinc@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1804157
Reviewed-by: svc-misra-checker <svc-misra-checker@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Sachin Nikam <snikam@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
nvgpu_os_fence_syncpt_create returns ERR_PTR instead of NULL whenever
its unable to construct a valid struct sync_fence instance. The
current code is not handled to detect errors returned using
ERR_PTR. This patch replaces the !fence check with IS_ERR() that
handles the return correctly.
Bug 200421587
Change-Id: I0ecfa8a651b8d5b743d11aab80d7a4cd71be8b1a
Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1804509
Reviewed-by: svc-misra-checker <svc-misra-checker@nvidia.com>
Reviewed-by: Konsta Holtta <kholtta@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Add READ_SM_ERROR IOCTL support to TSG level.
Moved the struct to save the sm_error details
from gr to tsg as the sm_error support is context
based, not global.
Also corrected MISRA 21.1 error in header file.
nvgpu_dbg_gpu_ioctl_write_single_sm_error_state and
nvgpu_dbg_gpu_ioctl_read_single_sm_error_state
functions are modified to use the tsg struct
nvgpu_tsg_sm_error_state.
Bug 200412642
Change-Id: I9e334b059078a4bb0e360b945444cc4bf1cc56ec
Signed-off-by: Vinod G <vinodg@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1794856
Reviewed-by: svc-misra-checker <svc-misra-checker@nvidia.com>
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>
