gpu: nvgpu: Set nvdec mailbox reg 0 to nonzero val

The scrubber binary after completion updates its return
code in mailbox register 0. The memory unlock code reads
this registers to determine the success of memory scrubbing.
This register is initialized to 0 during nvdec falcon reset.
If the scrubber binary halts due to an error condition, the
return code is not updated and it stays at 0.
Initialize the status register explicitly to non-zero value
helps avoid just false positives.
Add falcon register dump and PC trace to help debug the memory
unlock failures.

Change-Id: I3086dda2a9719c2d0b8a7ae898f1a03bedfa21b0
Signed-off-by: Tejal Kudav <tkudav@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1808899
Reviewed-by: Deepak Goyal <dgoyal@nvidia.com>
Reviewed-by: svc-misra-checker <svc-misra-checker@nvidia.com>
Reviewed-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-by: Mahantesh Kumbar <mkumbar@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>

1 files changed, 6 insertions, 0 deletions

diff --git a/drivers/gpu/nvgpu/common/fb/fb_gv100.c b/drivers/gpu/nvgpu/common/fb/fb_gv100.c
index d092ff4a..ce51a2c6 100644
--- a/drivers/gpu/nvgpu/common/fb/fb_gv100.c
+++ b/drivers/gpu/nvgpu/common/fb/fb_gv100.c
@@ -192,6 +192,11 @@ int gv100_fb_memory_unlock(struct gk20a *g)
                 mem_unlock_ucode_header[OS_DATA_OFFSET] >> 2],
                 mem_unlock_ucode_header[OS_DATA_SIZE], 0);
 
+        /* Write non-zero value to mailbox register which is updated by
+         * mem_unlock bin to denote its return status.
+         */
+        nvgpu_flcn_mailbox_write(&g->nvdec_flcn, 0, 0xdeadbeef);
+
         /* set BOOTVEC to start of non-secure code */
         nvgpu_flcn_bootstrap(&g->nvdec_flcn, 0);
 
@@ -202,6 +207,7 @@ int gv100_fb_memory_unlock(struct gk20a *g)
         val = nvgpu_flcn_mailbox_read(&g->nvdec_flcn, 0);
         if (val) {
                 nvgpu_err(g, "memory unlock failed, err %x", val);
+                nvgpu_flcn_dump_stats(&g->nvdec_flcn);
                 err = -1;
                 goto exit;
         }