nv-tegra-mirror/linux-nvgpu
1
0
Fork 0
mirror of git://nv-tegra.nvidia.com/linux-nvgpu.git synced 2025-12-25 02:52:51 +03:00
Code Issues Packages Projects Releases Wiki Activity

gpu: nvgpu: Enter Quiesce if GPU drops off the bus

Browse Source 
Currently, we reboot the entire system using kernel_restart() if
the GPU registers become inaccessible due to GPU disappearing
from the bus. GPU hitting high temperatures is one of the reasons
we might end up in above scenario.
Replace kernel_restart() with quiesce call as a more graceful way
of notifying about GPU's unavailability. While entering quiesce
state, make sure we do not trigger any register accesses which are
bound to fail in this case.

Bug 2919899

Change-Id: Ia9d413e04c7d205752414ff3e892f055c4363cce
Signed-off-by: Tejal Kudav <tkudav@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2398801
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
This commit is contained in:
Tejal Kudav
2020-08-14 06:32:24 +00:00
committed by Alex Waterman
parent 9963b94b4b
commit 71b005c1ef
2 changed files with 55 additions and 27 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
drivers/gpu/nvgpu/common/init/nvgpu_init.c
View File

@@ -61,8 +61,8 @@ bool is_nvgpu_gpu_state_valid(struct gk20a *g)
void nvgpu_check_gpu_state(struct gk20a *g)
{
if (!is_nvgpu_gpu_state_valid(g)) {
nvgpu_err(g, "Rebooting system!!");
nvgpu_kernel_restart(NULL);
nvgpu_err(g, "Entering SW Quiesce!!");
nvgpu_sw_quiesce(g);
}
}
@@ -201,9 +201,15 @@ void nvgpu_sw_quiesce(struct gk20a *g)
g->sw_quiesce_pending = true;
nvgpu_cond_signal_interruptible(&g->sw_quiesce_cond);
gk20a_mask_interrupts(g);
nvgpu_start_gpu_idle(g);
nvgpu_fifo_sw_quiesce(g);
/*
* Avoid register accesses when GPU had disappeared
* from the bus.
*/
if (is_nvgpu_gpu_state_valid(g)) {
gk20a_mask_interrupts(g);
nvgpu_fifo_sw_quiesce(g);
}
}
/* init interface layer support for all falcons */