diff options
| author | Arthur Hanson Rasmusson <arthur@vgpu.io> | 2026-10-15 12:41:14 -0400 |
|---|---|---|
| committer | Arthur Hanson Rasmusson <arthur@vgpu.io> | 2026-10-15 12:41:14 -0400 |
| commit | ecfb1278086f98161402f93bca8d3df602fbcb45 (patch) | |
| tree | c8febb2cd8b89fb8907411ad9276832bd304a74e /kernel/gsp_phys_booter.c | |
| parent | e07c62b1fab2f2dbc9981458c0697829b287b01e (diff) | |
| download | open-iov-ecfb1278086f98161402f93bca8d3df602fbcb45.tar.gz open-iov-ecfb1278086f98161402f93bca8d3df602fbcb45.zip | |
r154: after the second client's idle period the FECS saved/unloaded the
context (SKED rec: cwd0=0x12486 fecs_cur=b0633ee3 -> cwd0=0x2086 fecs_cur=0
and no change ever after) and every later forwarded launch acked
(qmd_rel/ch_rel) while dispatching nothing - including all three outcome
retries with fresh ring entries (0/205 changed, 720 ms no-writes, x4).
Re-issuing only GPFIFO_SCHEDULE+TSG_GPFIFO_SCHEDULE+doorbell did not
revive dispatch. The extended re-arm now replays the mode-35 lane-enable
tail first (GR_SET_CTXSW_PREEMPTION_MODE CILP + SET_TIMESLICE +
DEVICE_SCHED - the proven-accepted stage-4 sequence, HF-3 design), then
the schedules and the doorbell. Poison variants (explicit PREEMPT/tail,
BIND, PROMOTE) remain excluded.
Diffstat (limited to 'kernel/gsp_phys_booter.c')
| -rw-r--r-- | kernel/gsp_phys_booter.c | 18 |
1 files changed, 18 insertions, 0 deletions
diff --git a/kernel/gsp_phys_booter.c b/kernel/gsp_phys_booter.c index 123f93d..9eb1179 100644 --- a/kernel/gsp_phys_booter.c +++ b/kernel/gsp_phys_booter.c @@ -16741,6 +16741,24 @@ static int gsp_phys_rmcall_gr_resched_locked(void __iomem *bar0) u32 runlist_id, ch_id, doorbell; int ret; + /* + * m155 (HF-3 extended re-arm): after the channel has been idle long + * enough for the FECS to save/unload the context, re-issuing only the + * two GPFIFO_SCHEDULE verbs does not revive dispatch - r154: matmul's + * post-smiley launches acked (qmd_rel/ch_rel) but SKED stayed idle + * (cwd0=0x2086/fecs_cur=0) for every attempt including outcome + * retries, while the same channel executed mail/smiley earlier in the + * boot. Replay the mode-35 lane-enable tail (CILP preempt mode, + * timeslice, device sched) that stage 4 ran when the lane first went + * live - it is proven accepted and is the design's extended re-arm. + * The poison variants (explicit PREEMPT/tail, BIND, PROMOTE) stay out. + */ + ret = gsp_phys_rmcall_user_lane_lane_enable(bar0); + if (ret) + pr_info("vgsp-gr: rearm lane-enable rc=%d (continuing to schedules)\n", + ret); + rpc_result = 0; + status = 0; ret = gsp_phys_rmcall_control(bar0, boot_stage.rmcall_channel, GSP_PHYS_RMCALL_CTRL_GPFIFO_SCHEDULE, sched_params, sizeof(sched_params), 16, |