From 1c596b4b30e7ecc2f525ae47b52a73b8d4de816d Mon Sep 17 00:00:00 2001 From: Arthur Hanson Rasmusson Date: Fri, 23 Oct 2026 18:54:53 -0400 Subject: r576b diag: witness the recovery-release target (sema-fake lines) The r575 atomics wedge: a UVM tracker wait for payload 34 on its channel tracking semaphore stalled at 25 while the pump consumed the channel's remaining ring entries via the unreadable-entry grace/fake ladder and the client's wait never cleared (a manual exact-slot delivery of 34 via poke_guest unblocked it and the client completed green). The recovery release chosen by svc_fake_release is the suspect: its pool target is last_rel_va, learned from readable pushes, and a foreign release VA appearing there would send every later recovery to the wrong semaphore. Log every authored/queued recovery with its chosen target and payload (bounded, gsp_sema_dbg) so the next draw names the divergence. --- kernel/gsp_sema_service.c | 17 +++++++++++++++++ 1 file changed, 17 insertions(+) (limited to 'kernel') diff --git a/kernel/gsp_sema_service.c b/kernel/gsp_sema_service.c index dfe7c3f..d2c7eaf 100644 --- a/kernel/gsp_sema_service.c +++ b/kernel/gsp_sema_service.c @@ -8503,6 +8503,7 @@ static u64 svc_rel_va_seen; * on every decoded release regardless of its target. */ static u32 svc_rel_count; +static u32 svc_fake_rel_log; /* r576b DIAG: sema-fake witness budget */ static u32 svc_dbg_streams; static u32 svc_chan_put_log; /* ISSUE-0002 (r55): per-rank put-snapshot throttle (ch is a local copy). */ @@ -16554,6 +16555,22 @@ static bool svc_fake_release(struct nv_vgpu_device *vgpu, struct svc_channel *ch svc_pending_add(pool_va, deliver, 0); svc.releases_forwarded++; } + /* + * r576b DIAG: witness every authored/queued recovery release and the + * target it chose. The r575 atomics wedge (a UVM tracker wait for + * payload 34 that stalled at 25 while the ring was consumed) needs + * this to prove which VA the fakes targeted when the client's pool + * stopped advancing - last_rel_va is learned from readable pushes and + * a foreign release VA there would send every later recovery to the + * wrong semaphore. + */ + if (gsp_sema_dbg && svc_fake_rel_log < 256) { + svc_fake_rel_log++; + pr_info("sema-fake: h=0x%x pool=0x%llx deliver=%u delivered=%d have_pa=%d last_rel=0x%llx exp=%u\n", + ch->h_object, pool_va, deliver, delivered ? 1 : 0, + have_pa ? 1 : 0, ch->last_rel_va, + ch->expected_sem_payload); + } svc.releases_faked++; svc_rel_count++; ch->expected_sem_payload++; -- cgit v1.2.3