From e25b9b9e37bb0612bc17d2e8c1710e1c4950cd80 Mon Sep 17 00:00:00 2001 From: Arthur Hanson Rasmusson Date: Fri, 23 Oct 2026 20:49:08 -0400 Subject: vgsp r580: walk-over-bootcache for pool pages + event burst dedup and fast-path hint Pool resolution (thread: the starve wedge surviving the r577 fix): the boot-scoped pool cache survives session flushes, but a pool page MOVES between sessions and the old page often stays readable, so the read-back proof cannot tell - the resolver wrote releases to the stale cache PA while the client's tracker read its current page. Live evidence (r579 boot): pool-wr src=bootcache pa=0x11e6d9000 for va=0x121000000 while the injected pin/current page was pa=0x134730000; the client spun in uvm_tracker_wait for minutes. svc_pool_pa_resolve now tries the structural walk (the client's own page tables) BEFORE trusting the cache: a differing, read-proven walk page wins (src=walk-over-boot) and the cached page is kept as the fan-out secondary. Event burst (r579 boot: events_dropped=724, the 'no postable event (events_n=8 dropped=724)' sentinel every 0.5 s): apply the 250 ms per-event dedup to the per-completion shotgun (8 records per completion burned the 48/s budget), and hoist the completion hint above the pump fast path so it runs under load too (a dropped wake is re-posted within 0.5 s). --- kernel/gsp_rm_emulation.c | 14 ++++++++++++++ 1 file changed, 14 insertions(+) (limited to 'kernel/gsp_rm_emulation.c') diff --git a/kernel/gsp_rm_emulation.c b/kernel/gsp_rm_emulation.c index fb0f31d..2a957b3 100644 --- a/kernel/gsp_rm_emulation.c +++ b/kernel/gsp_rm_emulation.c @@ -882,6 +882,20 @@ int gsp_rm_post_channel_event(struct nv_vgpu_device *vgpu, u32 h_src) for (i = 0; i < rm->events_n; i++) { if (!rm->events[i].valid) continue; + /* + * r580: apply the hint dedup to the per-completion + * shotgun. Every completion posts ALL registered + * events (the ABI gives no source association), so a + * completion burst burns the burst budget 8 records + * per completion and real wake-ups get dropped + * (r579 boot: events_dropped=724). A redundant + * re-post within 250 ms adds no wake-up - the + * client's dispatcher re-reads channel state on any + * event - and the 0.5 s hint re-posts late waiters. + */ + if (time_before(jiffies, rm->events[i].last_post + + GSPRM_EVENT_HINT_DEDUP)) + continue; gsp_rm_fill_ev_payload(rm, i, payload); { int rc = gsp_rm_send_flags(ctx, -- cgit v1.2.3