Revert "gpu: nvgpu: Update CTXSW_TIMEOUT to 10ms"

This reverts commit 70b987dcf1.

Reason for revert: suspect for Bug 3563748.

Bug 3563748

Signed-off-by: shashank singh <shashsingh@nvidia.com>
Change-Id: Ibaeee5fe9ad5d58f4e2fba0f21559899a1f38d3f
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2678828
Reviewed-by: Alex Waterman <alexw@nvidia.com>
Reviewed-by: Aron Wong <awong@nvidia.com>
GVS: Gerrit_Virtual_Submit
Tested-by: Aron Wong <awong@nvidia.com>
This commit is contained in:
shashank singh
2022-03-09 16:18:57 +05:30
committed by Aron Wong
parent 329807b8f9
commit 830d5994d0

View File

@@ -216,21 +216,10 @@
*/
#define INVAL_ID (~U32(0U))
/**
* Timeout after which ctxsw timeout interrupt (if enabled by s/w)
* will be triggered by h/w if context fails to context switch.
* Typical ctxsw times should be in order of tens or hundreds of
* microseconds at least in Auto usecase with no vGPU etc. This is
* an assumption as we do not value data from HW on typical CTXSW times.
* For safety, we have FTTI budget of 100 ms. FTTI should include the
* diagnostic interrupt latency, and SW “fix” (timeout action may
* include recovery). The WCET for recovery is around 55ms.
* So, CTXSW timeout value should be somewhere in between hundreds of
* microseconds to 45 ms.
* Chooe CTXSW_TIMEOUT value to be 10ms -
* 1. It seems well within FTTI (with some budget for recovery if needed)
* 2. It can be easily updated if needed later.
* Timeout after which ctxsw timeout interrupt (if enabled by s/w) will be
* triggered by h/w if context fails to context switch.
*/
#define CTXSW_TIMEOUT_PERIOD_MS 10U
#define CTXSW_TIMEOUT_PERIOD_MS 100U
/** Subctx id 0 */
#define CHANNEL_INFO_VEID0 0U