summaryrefslogtreecommitdiffstats
path: root/kernel/gsp_sema_service.c
Commit message (Collapse)AuthorAge
* Reapply "vgsp-gr r590b: the stream-order watermark must watch every ring, ↵Arthur Hanson Rasmusson16 hours
| | | | | | not engine_type!=0" This reverts commit 64cca43699d6b74a1f2f2c5e087d33c7aba59de1.
* vgsp-sema r595: run a plausible GR push's ops - the driver's relocation ↵Arthur Hanson Rasmusson16 hours
| | | | | | | | | | | | | | | | copies live on a compute channel The r568 b-copy class is rooted in the op gate: allow_ops was (ch->engine_type != 0), so a GR-class channel's own memory copies (LAUNCH_DMA on h=0x5c000059, eng=0) never executed unless the push page came from an injected pin. Their releases were still applied by the op-hold path, the client's blocking cudaMemcpy returned, and the launch read the pre-copy page (c = a + 0). Admit ops when the push content is a plausible method stream - the same content arbiter (cap_trust) that already gates GR launch capture - never on a replay or a rehold. Measured: the d_a/d_b relocation copies now run ~0.5-1.5 s after the driver submits them instead of tens of seconds.
* Revert "vgsp-gr r590b: the stream-order watermark must watch every ring, not ↵Arthur Hanson Rasmusson16 hours
| | | | | | engine_type!=0" This reverts commit f3225a81ca384248180d77f78c3614c73dcf289d.
* vgsp-gr r590b: the stream-order watermark must watch every ring, not ↵Arthur Hanson Rasmusson17 hours
| | | | | | | | | | | | | | | engine_type!=0 The r591 draw reproduced the r568 signature (VADD bad=32735 csum=8139264, c = a + 0) with ceback=0: the driver submits its relocation/H2D copies on a COMPUTE channel (h=0x5c000059 registers eng=0), so the engine_type filter skipped exactly the backlog the guard exists for. Consider every other channel of the device (the launching channel is excluded: its own ring legitimately runs ahead with future entries), and read each one's live USERD GP_PUT against this pass's consumed position. r590's validated part is unchanged: same pass snapshot, same bounded defer, historical forward at the absolute cap (never a refusal).
* vgsp-gr r590: CE submission watermark - read the ring's live GP_PUT against ↵Arthur Hanson Rasmusson17 hours
| | | | | | | | | | | | | | | | | | this pass's consumption before forwarding a launch The r589 mail VADD drew c = a + 0: the h_b H2D entries sat behind a translation-held CE entry while the pump's last read of the ring's GP_PUT was 92, so the GR launch (processed later in the same pass) forwarded against the still-zero b window. The r541 guard needs a decoded copy and the r568 opdef gate needs a decoded hold; an entry the pump has not reached yet is invisible to both. Record this pass's channel snapshot (CE channels first), read each CE ring's live USERD GP_PUT at the GR decision, and defer while it differs from the snapshot's consumed position. Both sides of the comparison come from the same channel and the same pass, so it cannot reproduce r588's stale-epoch over-fire; the guard only defers (bounded, witnessed) and at the absolute cap the historical forward stands, never a refusal.
* Revert "vgsp-sema r588: copy-pending sees a CE ring's undecoded backlog ↵Arthur Hanson Rasmusson17 hours
| | | | | | (stream-order)" This reverts commit c88a7f6b2658ea2d49b67cd63179a5a1fbce3663.
* vgsp-sema r588: copy-pending sees a CE ring's undecoded backlog (stream-order)Arthur Hanson Rasmusson17 hours
| | | | | | | | | | | | | | | | | | | | | The r568 copy-pending guard in svc_gr_request could only see decoded state: an active entry hold (svc_copy_pending) or a recorded device-write target (svc_dstpend_overlap). A CE copy entry sitting BEHIND a translation-held entry in the ring has neither - it was never decoded - so copypend read 0 while the ring still owed that copy. A GR launch decoded in that window captured the copy's destination before its data landed: r587 mail VADD computed c = a + 0 (h_c == h_a exactly: bad = 32735 = 32768 - 33, the count of i where b[i] == 0; csum = sum(a) = 8139264), because the d_b H2D bigcopy executed after the launch's copy-in. Extend svc_copy_pending() (now per-device) to also report a CE channel whose ring has submitted-but-unconsumed entries (last_get != put_prev - the same outstanding relation the completion publish uses). With the r507o CE-before-GR pass order the CE channel's write-back for the pass has already run when the GR decision reads it, so the guard defers (bounded by the existing defer/expiry machinery) until the ring drains and the re-fill captures the post-copy pages.
* r586: demanded staging VAs resolve via the client pagemap; fail-closed at ↵Arthur Hanson Rasmusson20 hours
| | | | | | | | | | | | | | | | | | | | | | | the r543 argument defer cap Two halves of the staging-pin coverage class (r584/r585 boots): 1. guest_pin_stream demand handler: resolve_va_by_content rejects any VA outside the CE pushbuffer window (PB_CE_LO..PB_CE_HI), so the demands for the GRFILL dropped staging pages (high client VAs like 0x703304e00000) were silent no-ops - no pin ever landed and the fills kept dropping them. r586 falls back to the client's own pagemap (scan_range on the exact page), the same walk the window scans use. 2. vgsp-work: the r543 argument-defer cap expiry now takes the r583 fail-closed path instead of forwarding: the r585 boot's saxpy was forwarded with n_data=969 dropped=7 partial=1 at the cap and read stale operands (bad=1047527 csum=130523290); the r584 matmul (n_data=2 dropped=18) reached the r564 degenerate refusal. Refuse, author the QMD release in software (rel_done latched), consume, witness. The dropped pages were demanded first (r585), so the retry window had its chance; a launch that would run with unresolved operands is never silently forwarded.
* vgsp-gr r585: demand the fill's dropped pages through the pin channel (the ↵Arthur Hanson Rasmusson20 hours
| | | | | | | | | | | | | | | | | | | | staging-pin coverage class) The r584 boot's matmul draw: GR-DEC push=0x20043be64 n_data=2 dropped=18 partial=1 stalled=1 -> the r543 argument defer expired -> r564 degenerate refusal -> SW completion -> MATMUL bad=65536 csum=0. The dropped pages (the client's staging buffers 0x71b662c00000/c40000/c80000, 0x7dd0cac00000, kind=a) had NO injected mapping: the host pin stream's demand loop only parses "stalled VA #N: 0x..." lines, and the GRFILL-DROP line format is invisible to it - so the fill's retries never saw pins and the launch was SW-completed without running. r585: svc_log_stalled_va() for each dropped page (up to the 4 detailed per episode; the helper dedups per boot and the episode marker bounds the rate). The host demand loop (<=3 s) then pushes the VA to the guest scanner, whose pin lands inside the launch's bounded defer window on the retry - the demand path that already heals every other unresolved-VA class, now connected to the fill drops.
* vgsp-sema r584: the vidmem-window fill decoder - resolve window VAs ↵Arthur Hanson Rasmusson20 hours
| | | | | | | | | | | | | | | | | | | | | | | | | structurally (the last per-draw GREEN blocker) The r583 boot's fail-closed refusal traced to a fill that never executes: the CE channel-setup push's 8 KB zero fill to out=0x204820000 decoded fine (type=0 op, dma=0x586, line=8192, value=0) but its EXECUTION resolved the window VA through the generic chain (pin -> ledger -> translate), which answered SYS with a stale landing pa=0x1300b0000; the r369 below-floor guard then refused it ("LOWFILL-REFUSED ... below-floor fill resolved outside vidmem"), the fill never ran, its DSTPEND note stayed live ("DSTPEND note va=0x204820000 type=0", dma-trace denied every pass), and every GR launch touching that page deferred to the r556 cap and (since r583) took the fail-closed no-write path. The vidmem window is offset-mapped by construction (SVC_VIDMEM_VA_BASE 0x200080000 rationale: the client's vidmem VAs are its window base plus the vidmem offset into the emulated BAR1 buffer; the same convention svc_lookup already applies). svc_fill_1d now resolves a VID-aperture fill whose destination is a window VA structurally - addr = VA - window base, aperture VID - so the existing contiguous shadow-write block executes it. Witnessed once per boot ("sema-lowfill-window va=... -> fb offset ..."). No guest page table can express the window offset; the structural rule is faster AND safer than the stale-looking generic chain.
* vgsp-work r583: fail-closed at the defer cap when a launch's data page still ↵Arthur Hanson Rasmusson20 hours
| | | | | | | | | | | | | | | | | | | | | has an unexecuted device write The r581 boot's saxpy red draw (bad=524121 csum=874874129) shows the residual hole of the r541 dstpend guard: GR-DEC "push=0x20043c198 n_data=2505 dropped=8 partial=1 dstpend=1 copypend=0 defer_exp=0" and on the final dump defer_exp=1 - the launch overlapped a page whose device write had never executed (live dstpend records 9.5-85 s old), the guard deferred through its bounded retry, the r556 15 s absolute cap expired, and the historical path forwarded the launch anyway: the engine read the known-stale operand and produced the ~50%-wrong result. r583: at cap expiry with the overlap still live the forward is REFUSED. Fail-closed: the candidate's QMD completion is authored in software (forward-only; rel_done latched so the consume tail cannot double-author) and the candidate is consumed without an engine launch - the client completes, the engine never touches a page whose writer did not run, and the event is witnessed loudly ("GR forward REFUSED at defer cap ..."). This cannot hang the guest: the refusal consumes instead of holding.
* vgsp-sema r581: do not pay the translation grace for provably-foreign ↵Arthur Hanson Rasmusson21 hours
| | | | | | | | | | | | | | | | rescued content r577's foreign-release gate made unreadable entries take the !rel_seen path, where each entry pays the full ce_grace_ms hold before the exact-slot fallback delivers its pool step ('skipping stale entry ... after 8000 ms grace' every ~8 s on the r577/r580 boots; a client's init crawls for minutes). When the rescue decoded a release but the gate refused it as a tracking slot, the content is provably stale/foreign: no work may run from it (rescued work is never executed anyway) and the 8 s wait buys only the small chance that the real page lands inside the window. Backdate defer_since so the consume path delivers the pool step immediately; the translation grace still applies to entries whose rescue decoded nothing (their content may still be the real push).
* vgsp r580: walk-over-bootcache for pool pages + event burst dedup and ↵Arthur Hanson Rasmusson21 hours
| | | | | | | | | | | | | | | | | | | | | | 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).
* vgsp-sema r577: never let a foreign release VA claim a CE channel's tracking ↵Arthur Hanson Rasmusson22 hours
| | | | | | | | | | | | | | | | | | | | | | | | slot; event-record A/B knobs; GR-DEC dump deepened r575 starved-release wedge (live): an unreadable UVM CE entry's rescued stale page decoded a release to a foreign VA (0x12006c004, pay=60); the rescue paths adopted it as the channel's tracking slot (last_rel_va) and every later recovery release targeted it, while the client's real pool (0x121000000) stopped at 25 - the client spun in uvm_tracker_wait for 34 (exact-slot poke delivered it and the client completed green). For CE channels only a pool-shaped release target (0x121000000..0x122000000) may become last_rel_va; a foreign rescued release is refused (witnessed, bounded) and the pool step is delivered through the exact-slot fallback / deferred-release machinery instead. The fake's read-before-write keeps the delivery monotonic even when a legit push carries multiple releases. Also: gsp_rm_ev_{seq,nlist,cl} runtime knobs for the guest-RM event drop A/B (the r576 experiment: guest consumes every POST_EVENT record and never enters nv_post_event), and the GR-DEC launch dump now covers 256 data pages (the mail VADD fill carries n_data=969). New: tools/cuda-ab/evhold.cu - deterministic CUDA event holder/wait probe (context + cudaEventSynchronize + a hold window) for the event experiments.
* r576b diag: witness the recovery-release target (sema-fake lines)Arthur Hanson Rasmusson23 hours
| | | | | | | | | | | | | 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.
* r576 diag: launch-decision + event-post payload witnesses ↵Arthur Hanson Rasmusson24 hours
| | | | | | | | | | | | | | | | | | | | | (gsp_phys_gr_dec_log, gsp_rm_ev_log) The GR launch-decision dump logs, for the first N draws whose fill saw dropped pages / a partial argument set or whose decision point carries a dstpend/copy-pending verdict: the decision verdicts, the launch's resolved data pages (VA -> guest PA), the live deferred-write records and the pending CE opdef entries - the operand evidence needed to explain a red draw of the readback lottery (the wrong-but-similar page hypothesis). The event-post witness logs every EVENT_POST_EVENT record actually sent with its full payload (hClient/hEvent/notifyIndex/data) and send result, plus a sentinel when post_all_events found valid events but posted none while none were dedup-skipped. This is the evidence needed when a client parks waiting for a completion wake-up that never arrives (the 2nd-client device-state spin): it shows whether the post was generated, for whom, and whether the status-queue write accepted it or dropped it (burst cap / ring full). Both are inert by default (budget 0).
* vgsp-sema r568: gate a partially-resolved launch on any unexecuted CE writeArthur Hanson Rasmusson24 hours
| | | | | | | | | | | | | | | | | The page-level dstpend guard is blind when the launch's argument base was dropped (GRFILL-DROP): the launch's data set then lacks the buffer's pages. r571 mail: the VADD's copy-in ran at 319818.68 while the b H2D executed at 319820.16 (its base dropped, 9 pages unresolved) - the engine computed c = a + 0 again, same signature as r567/r568. r570 passed only because the page-level guard happened to catch it there. svc_copy_pending(): true when any active opdef slot belongs to a CE channel (a withheld CE release = a device write that has not executed; GR channels excluded so a launch's own withheld release cannot defer itself). A launch that is partially resolved (gr_arg_partial or gr_dropped > 0) now defers while any such write is pending, bounded by the same busy-defer + absolute 15 s cap machinery. Fully-resolved launches are unaffected (the page-level guard already covers them).
* vgsp-sema r567: a transfer-shaped push with no op notes its destination and ↵Arthur Hanson Rasmusson25 hours
| | | | | | | | | | | | | | | | | | | | | stays owed The r569 mail failure's missing half: the b H2D's COPY op needs have_in (the staging upload), which was not resolved on the early passes, so the push decoded NO op; the release was still applied (the client's cudaMemcpy(b) returned at 318935) and the slot was marked complete, while the copy itself only executed at 318952 (17 s later) - the launch's copy-in (318945.7) read b = 0 and the engine computed c = a + 0 (readback src16 = a's values). The r541 guard had nothing to defer on (no op, no dstpend). A trusted, allow_ops, transfer-shaped push that decodes no op with a device destination now: - notes the destination via svc_dstpend_note (the launch guard defers until the copy runs or the 10 s window expires), and - sets svc.op_owed (the slot stays incomplete, so a later fresh pass runs the copy instead of replay-suppressing it forever). No hold is armed, so a push whose trust never arrives cannot wedge the pump; the guard provides the ordering, exactly like the r541 design.
* vgsp-sema r566: note a held entry's unexecuted device-dst copies for the ↵Arthur Hanson Rasmusson25 hours
| | | | | | | | | | | | | | | | | | | launch guard Root cause (r567/r568, deterministic): the mail client's b H2D COPY op was COLLECTED but never attempted while its entry was held for another operand; the VADD launch's copy-in ran 0.6 s BEFORE the copy executed, so the engine read b = 0 and computed c = a + 0 (the readback page held a's values {0.0,0.5,1.0,1.5} vs the correct {0.0,0.75,1.5,2.25}). The r541 guard only saw copies that FAILED with -ENOENT (svc_dstpend_note at the failure site), so a merely PENDING copy was invisible to it. svc_dstpend_note_ops() runs after each svc_run_ops() call in svc_decode_stream whenever svc.op_tr_pending is set: every collected device-dst copy whose per-op done bit is not set (i.e. did not execute this pass or earlier) gets its destination noted. The launch guard then defers until the copy runs (the copy's success path clears the record), bounded by the existing dstpend TTL + busy-defer + 15 s cap machinery.
* vgsp-sema r565: session-scope the GR exactly-once caches (cross-client VA reuse)Arthur Hanson Rasmusson25 hours
| | | | | | | | | | | | | The UVM pool reuses the same launch push VAs for every client of a boot. svc_gr_cap_hist[8] matches on the push VA ALONE (no content hash) and was boot-scoped, so a previous client's capture could classify a later client's fresh launch as already captured and skip it (LAUNCH-SKIP (replay; already captured=1) from a dead session's record). svc_gr_seen (hash-keyed) and svc_inline_once (content-keyed) share the same lifetime and the same class of risk. Reset all three in svc_session_flush; they refill within the session from the live client's own pushes, so same-session dedup is unchanged.
* vgsp-sema r563c: the stall watchdog gates on ACTIVE withheld work, not ↵Arthur Hanson Rasmusson25 hours
| | | | | | | | | | | | | | cumulative counters r566 evidence: the dump fired 4x with done=8929 and opdef=215/7/5 frozen while the ring was EMPTY (every channel last_get == put - the clients were waiting, no entry pending) and the forced flush had nothing to deliver: is a lifetime counter (215 holds applied/expired long ago), not proof of a current hold. A real stall has either a pending translation (svc_pending_n) or an ACTIVE opdef slot with releases recorded (e->active && e->n); the dump (and with it the self-heal arm) now requires one of those. The r564-class true stalls (active slots holding releases) still dump and still heal.
* vgsp-sema r563b: the stall self-heal clears the walk-state blockers (force ↵Arthur Hanson Rasmusson26 hours
| | | | | | | | | | | | | | | | | | | opdef flush + per-channel state) r564 evidence: the STALL-DUMPs fired (4x, done=12700 held=727 frozen) but ZERO STALL-HEAL lines - the force-consume budget never engaged because the walk breaks BEFORE the instrumented hold sites (the cheap-wait unread window at the loop top, or a stale GR hold/nocap index). The self-heal now acts on the state every break site consults: - svc_opdef_force_flush(): deliver every withheld release forward-only, record the delivered marks (so a re-decode cannot re-hold), clear the slots - the same edge the 3 s bound would run, at the watchdog's discretion; - per channel (under chan_lock): unread_va/unread_until = 0 (the cheap-wait break), gr_hold_idx/gr_hold_until/gr_nocap_idx/gr_nocap_va reset; - the force-consume budget stays as the belt for the hold sites. One loud STALL-HEAL line per firing.
* vgsp-sema r563: stall-watchdog self-heal (forced consume after 20 s of no ↵Arthur Hanson Rasmusson26 hours
| | | | | | | | | | | | | | | | | | | | | progress) Every remaining frozen-pump shape (r562: STALL-DUMP no entry consumed for 20 s, held=99 opdef=150/11/3 frozen, pool_dual=0 conflicts=1) has the same end state: some hold condition keeps winning and the walk breaks on the same entry forever, whichever hole was closed last (r548..r558 each closed one). Add the single backstop that cannot be out-run: when the r539 stall watchdog fires (20 s without consuming a GPFIFO entry while withheld work exists), arm a budget of 4 forced consumes; the next hold decisions that would park an entry are bypassed (opdef hold, no-capture hold, busy-defer re-hold, data-stalled hold), the entry is consumed with its withheld releases applied forward-only and the GR QMD completion authored by the existing tail, and the client's wait advances. Bounded by the dump cadence (<=4/boot) and the per-arming budget, so a healthy boot never sees it; each bypass logs one STALL-HEAL line. This turns 'the client stalls forever' into 'the client completes with the best-effort data the guards would have fallen back to anyway'.
* vgsp-sema r561: bound the pin-injection drain per service pass (starved ring ↵Arthur Hanson Rasmusson26 hours
| | | | | | | | | | | | | | | | | | walk) r559 soak2 evidence (victory_saxpy, 167 s at 100% CPU, module alive and printing stats but consuming no GPFIFO entries): the guest pin stream flooded the injection queue (svc_handle_map_add: 1867-6101 callbacks suppressed per burst, several bursts per 10 s) and the unbounded for(;;) drain spent every service pass processing injections, so the ring walk never ran - work_guard counters froze (deferred=513 applied=50 unchanged for minutes) and the client spun. Process at most 64 queued requests per svc_handle_map_add() call. The drain runs again on every service pass and from svc_gr_request/ svc_run_ops, so nothing is lost; under a flood the ring walk now gets its cycles every pass. Bounded batches are the same fairness rule the ring walk itself already uses.
* vgsp-sema r558: the opdef hold path honors the absolute per-launch boundArthur Hanson Rasmusson27 hours
| | | | | | | | | | | | | | r557 STALL-DUMP evidence (victory_matmul): "no GPFIFO entry consumed for 20s" repeated while opdef=3880/16/64 and held=155 climbed - the pump froze on one held entry whose operand never resolved, re-holding it every pass. The opdef hold was the only pump hold path without the absolute per-launch-VA bound (svc_gr_hold_bound_expired, the r509 one-shot used by every GR hold path); it could therefore extend indefinitely through fresh slot allocations even though each individual slot window is 3 s. Consult the same absolute bound before holding: once it fires the entry flushes its withheld releases and is consumed, so no operand-pending entry can freeze the pump's entry loop.
* vgsp-gr r556: absolute 15 s per-launch deferral cap (single stall backstop)Arthur Hanson Rasmusson27 hours
| | | | | | | | | | | | | | | | r553/r555 boots still show clients walking the individual defer/hold bounds for 400+ s: victory_reduce 430 s (same push VA re-deferred every pass, r554's 64-slot latch), victory_saxpy 254 s at write time with work_guard deferred climbing by the thousand. Each guard is bounded on its own, but their bounds interact and re-arm across re-decode passes. Arm an ABSOLUTE per-push-VA deadline on first sighting (64-slot table); once 15 s have passed, svc_gr_defer_expired() returns true and every defer site forwards best-effort: the r541 dstpend guard, the r543 arg-partial guard and the m150 data-stalled hold. 15 s sits far above the 0.5-2 s pin latency the deferrals exist to cover and far below every client-side timeout (>= 60 s), so the pin-late cases keep their window and no launch can be parked longer than the cap, whichever guard keeps winning.
* vgsp-gr r554: the per-VA launch-defer latch needs 64 slots, not 8Arthur Hanson Rasmusson27 hours
| | | | | | | | | | | | | r553 soak evidence (victory_reduce, 430 s stall, 0 hangs): the same launch push VA (0x20043be64) was deferred on every pass for 7+ minutes ("partial argument set: 7-8 pages unresolved"). A client's UVM pool allocates ~8-10 channels and each command buffer reuses two launch push slots, so more than 8 distinct push VAs rotate through the guard; with 8 latch slots each new VA evicted another and the one-deferral-per-VA guarantee thrashed - the second sighting of any VA found its record gone and deferred again, so a partial argument set that never resolves held the launch forever. 64 slots cover a full client with margin (1 KB of static state).
* vgsp-gr r553: the dstpend launch-defer window is absolute, not slidingArthur Hanson Rasmusson28 hours
| | | | | | | | | | | | | | | | | r551 soak evidence (victory_strided, same boot as a green battery): the client's init copies keep failing -ENOENT and are retried by the pump; every retry re-noted its destination page and re-freshened the r541 dstpend record, so svc_gr_dstpend_overlap() never lapsed and the launch guard deferred the launch forever: "GR busy-defer timeout -> hold 8000 ms" every ~4 s for 427 s, zero GR LAUNCH submitted, client at 100% CPU. Stamp a page's record once per failure episode: an existing record inside its 10 s window is not re-stamped (the window stays absolute from the first unexecuted write); an expired record is re-stamped for a genuinely new episode. The launch then forwards best-effort at the bound exactly as the guard's own comment promises ("bounded by the same busy-defer + gr_hold machinery ... it can never hang the launch").
* vgsp-sema r551: a delivered opdef release is not withheld again (kill the ↵Arthur Hanson Rasmusson29 hours
| | | | | | | | | | | | | | | | | | | | | | | | | | | | re-hold cycle) r550 evidence (victory_strided, reused CE channel 0x5c000059, init sequence with ~40 device-aperture fills whose walk fails -ENOENT): the sweep delivered each entry's withheld counter release every ~3 s and cleared the slot, the pump re-decoded the still-parked entry, svc_opdef_release recorded a FRESH slot and held it again - SWEEP-EXPIRED repeated six times for the same entry (entries 41..43 at ~18-21 s each) before OPDEF-EXPIRED consumed it; the client span at 100% CPU for 12+ min and the battery timed out (r549's 3 s bound alone only shrank, did not remove, the cycle). Remember every delivered release by (channel, entry, va, payload) in a bounded 256-slot / 60 s ring: - svc_opdef_release: a re-decode of an already-delivered release claims it (rel_seen bookkeeping stays correct) and does NOT re-record or hold; - svc_opdef_hold: an entry with any recent delivery is consumed, never held again; - svc_opdef_flush / expire_sweep record what they deliver; - the ring is per-session (svc_session_flush), because handles, ring slots and payload sequences are reused by the next client. Releases are monotonic completion signals and every delivery path is forward-only, so re-applying one is a no-op - the change cannot double anything, it only removes the re-hold.
* vgsp-sema r549: cap the dead-operand walk at 3 s per entry (opdef hold + ↵Arthur Hanson Rasmusson30 hours
| | | | | | | | | | | | | | | | | | | opfail window) r548 evidence (matmul, reused channel handle 0x5c000059 after the previous client exited): the new client's init sequence carries ~65 small device-destination ops whose operands no resolver covers (memcopy/memset rc=-2, SVC_VAS_FLOOR destinations like 0x2000c0100). Each one armed the opdef hold, expired after exactly 10 s (OPDEF-EXPIRED idx 7..18 at ~10.5 s spacing) and only then released the counter increment the client waits on: the pump walked one entry per 10 s while the client spun at 100 % CPU (12+ min, battery timeout). The two bounds were both 10000 ms. Ops that DO resolve show 0.5-2 s pin latency (SVC_OPDEF_HOLD_MS comment), so 3000 ms keeps the m163 guarantee (a failed device copy still gets its retry once the pins land) and the op-failure latch now fires on the first 3 s retry instead of a second 10 s window. Both are pure bounds - no path loses its retry.
* vgsp-sema r548: never consume a GR launch whose forward is still owed (r547 ↵Arthur Hanson Rasmusson30 hours
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | saxpy stall) Live evidence (r547, 16+ min stuck victory_saxpy): both launches were captured, deferred by the r541 (unexecuted device write) and r543 (partial argument set) guards, and left as unoffered candidates - then every later pass re-decoded the entries as replays (LAUNCH-SKIP; already captured=1) and the consume path dropped them with no GR LAUNCH and no GR-COMPLETE-SW for the client, whose cudaDeviceSynchronize waited forever (client-side strace: RM-control poll loop, libcuda condvar 500 ms retimer; host log: consume idx=105/106 every ~2 s, releases never authored for rel=0x204827f7c). Three coordinated changes: - svc_gr_arg_defer_ok: per-VA latch table (8 slots) instead of a single slot. Two interleaved launches made the single slot ping-pong (each call saw the other VA and re-deferred both forever); each VA now gets exactly one deferral inside the 30 s window and the next sighting proceeds, so the guard delays without ever hanging. - svc_decode_stream: a fresh decode pass no longer wipes a valid, never-offered candidate (a deferred forward is not stale). The capture path already rotates to a free candidate-ring slot. - pump no-capture hold: fires for a REPLAY whose push VA still has an unoffered candidate in the ring (owed forward), forcing a fresh capture + request per bounded window; and when the window expires with the forward still owed, svc_gr_owed_resolve authors the QMD completion release from that candidate and marks it consumed, so an entry is never consumed without a completion (no late double-run). New counter gr_owed in the sema-svc drain line; hold log carries owed=.
* vgsp-sema r543: hold a GR forward with a partially-resolved argument set ↵Arthur Hanson Rasmusson31 hours
| | | | | | | | | | | | | | | | | | | | (bounded) The r542 matmul failure was the NOWRITE wedge: the fill resolved 969 pages but 2 of 13 high argument bases resolved 0 pages; with gsp_phys_gr_fill_hold=0 the fill is log-only for dropped pages and only the aggregate all-arguments-unresolved rule holds, so the launch forwarded, the engine fetched the QMD (qmd_rel=0x600df00d, engine fetch OK) and wrote nothing (retirement wait 0/1032, NOWRITE-AMBIGUOUS fe_esr=0x80000000, trap=0x80010250); the client read zeros (MATMUL bad=65536 csum=0). Fix: the fill now records gr_arg_partial when an ARGUMENT candidate page was dropped while other arguments resolved. svc_gr_request defers such a launch once per push VA (SVC_GR_ARG_DEFER_MS window, svc_gr_arg_defer_ok latch) through the existing bounded busy-defer retry, so the pins can land before the engine reads the operands. A genuinely junk pointer word pays the window once per push, not per retry; after the window the historical forward stands, so the guard can only delay a launch, never hang it.
* vgsp-sema r541: hold a GR forward whose data page has an unexecuted device writeArthur Hanson Rasmusson32 hours
| | | | | | | | | | | | | | | | | | | | | Live root cause of the smiley/streams2/streams_events readback lottery (r539): smiley pass 2 verified a2 + b1 and pass 4 verified a4 + b2 - each even pass's launch read the PREVIOUS even pass's b page. The b H2D's release completed the client's synchronous cudaMemcpy while the copy op itself was still held (b's svc_copy executed at 294026.818; launch 2's engine read at ~294026.3), so the client legitimately submitted the launch against a shadow page the H2D had not written yet. Passes 1/3 looked 'correct' only because passes 2 and 3 share the same b formula. Fix, as recommended in CONT.100/94: a device-dst copy/fill that fails to execute this pass (rc=-ENOENT) records its destination VA page in svc_dstpend[] with a 10 s stamp. svc_gr_request, right after the fill and before any submit, checks the launch's filled data pages against the table (svc_gr_dstpend_overlap); a hit returns the busy-defer verdict, so the pump holds the entry in its bounded retry window and re-captures once the copy has run - the record clears on the copy's success. The 10 s window restores the historical forward when the copy never lands, so the guard can only delay a launch, never hang it.
* vgsp-sema r540: a denied transfer op keeps its slot incomplete past the pend capArthur Hanson Rasmusson32 hours
| | | | | | | | | | | | | | | | | | | | | | The r539 boot showed the 5-minute device-init stall class live: a GRCE channel push (eng=0) carrying an H2D copy is decoded before the pin stream reaches its operands, so allow_ops=0 and the copy is denied; the r308 path sets op_tr_pending and holds the entry. That path is capped at 4 tries per slot ('so a slot whose trust never arrives cannot wedge the channel'), but once the cap is hit op_tr_pending stays 0 and svc_decode_stream's tail marks the slot COMPLETE - so every later pass classifies the entry as a replay and suppresses the copy forever. The pin arrives seconds later; the copy still never runs and the client spins at DEVICE until UVM's own retry. Evidence (r539, /tmp/opencode/r539-live.log): held idx 89/91..94 on h=0x5c000059 with 'denied ... replay=1 trusted=1' and 'sema-pend: hold ... idx=89' for minutes while the client printed DEVICE. Fix: new svc.op_owed flag. When the pend cap blocks the retry, the pass is marked owed so slot_ops_complete stays 0; the next fresh pass runs the copy as soon as the translation/injection exists. No hold is armed for an owed pass, so an operand that never translates still cannot wedge the pump (the denied path's liveness guarantee is preserved - only the freshness is).
* vgsp-sema r539: pool release fan-out to a second candidate page + STALL-DUMP ↵Arthur Hanson Rasmusson32 hours
| | | | | | | | | | | | | | | | | | | | | | | | | | watchdog The init-stall boots (r530-r538) share one shape: the pump loops but the entry loop stops consuming, and a client spins in uvm channel_pool_add -> uvm_push_end_and_wait while its pool tracker word is frozen. The r538 live boot showed the cause live: va=0x121000000 flapped between the injected pin (the guest kernel's tracker->page->rm_mem chain) and the host GMMU walk on every resolve (pool-boot conflict ping-pong, ~50/s), and the release write follows whichever source answered that pass - a release written to the losing page strands the client that polls the other one. Fix: svc_pool_boot_store keeps BOTH most-recent distinct pages (primary + secondary, each with its own proof stamp). svc_pool_boot_fanout applies every successful pool release to the secondary as well (forward-only, read-back proven, 10 s freshness so a freed/reused page is never written). Wired into all three svc_sem_write success paths and svc_fake_release's recovery write. The r536 walk-over-pin preference stays as the primary choice; the pin is now recorded as the secondary there. Also: STALL-DUMP - when the entry loop consumes no entry for 20 s while withheld work exists (pending releases or opdef slots), dump the pending list, opdef slots, pool candidates and counters once (bounded 4/boot), so the next stall is named in one place instead of N bounded logs.
* vgsp-sema: bounded fallback registration for a parked VAS candidate (15 s)Arthur Hanson Rasmusson33 hours
| | | | | | | | | | | | The VAS candidate verification needs the client's ring to be found by the RAM content scan; on a boot where that never anchors the ring the candidate stays parked forever and every launch operand walk fails rc=-2 (observed: pd_found=0, vas[0] holding an internal pdb, walks stuck at 9555 successes with 114 K attempts while the client spun for minutes). After 15 s parked, register the candidate unverified with a loud log: it came from the client's own RM publish and the newest-first walk order keeps an older stale directory from winning, so the bounded wait's registration is the lesser evil.
* vgsp-sema: prefer the client's page tables over a disagreeing pool pin ↵Arthur Hanson Rasmusson33 hours
| | | | | | | | | | | | | | | | (moved-but-readable pool) The pool resolver trusted an injected pin after a read-back proof. That proof does not catch a pool page that MOVED: the old page is still readable, so the release write lands on the old page while the client polls the new one. Observed: 'pending release starved va=0x12100000x tries=64' for ~45 s with the stale pin (pa=0x120b30000) readable, until the fresh pin (pa=0x130e9d000) arrived; the client's tracker never woke and its init hung (multikernel battery timeout, copies stuck rc=-2). The client's own page tables are the GPU/UVM ground truth for where the tracker reads, so when a structural walk resolves a different page than the pin (and it is not a degenerate identity), the walk wins - both sources are read-back proven at the exact offset.
* vgsp-sema: chunk the discovery scan's guest reads (64 KB) - the rebuild was ↵Arthur Hanson Rasmusson34 hours
| | | | | | | | | | | | | starving the pump The full-RAM discovery scan read one 4 KB page per vfio_dma_rw call: 1.5 M guest reads over the 6 GB range, ~10-11 s per rebuild, and the rebuild runs every 10 s - so the pump got roughly one pass per rebuild. A client whose resolutions were dropped by a session flush then waited minutes for the scan to re-find its pages (the observed post-flush stall: data-VA copies returning -ENOENT with the pump counters frozen). Reading 64 KB chunks amortizes the iommu walk + pin over 16 pages; the per-page detection logic is unchanged.
* Revert "vgsp-sema: keep the learned VA map across the session flush ↵Arthur Hanson Rasmusson34 hours
| | | | | | (re-compact, not wipe)" This reverts commit 1a49c7eb69c03f58bc51939828de222a2ac42fea.
* vgsp-sema: keep the learned VA map across the session flush (re-compact, not ↵Arthur Hanson Rasmusson35 hours
| | | | | | | | | | | | | | | | | | wipe) The flush fires when the channel table momentarily empties - a live client's UVM channel cycling triggers it - and wiping the map left the client's already-resolved data VAs unresolvable until a rebuild scan re-found them, which it may not. Evidence: a mid-battery boot where multikernel's operands went NO ENTRY 40+ s after a flush (the rebuild restored 14-31 K entries without them); the pump stalled for minutes and the next clients read back whole-page garbage. The map is a cache with a merge-on-rebuild contract, so re-compact it (dedup + sort) and keep it as the merge base; reused-VA correctness stays with the ledger landing, the pin-wins rule and the fresh scan. Also: push_battery.sh now pushes inplace_scale (the battery list had it, the push list did not).
* r526 THE XDIFF ROOT CAUSE: the LOAD_INLINE_DATA destination cursor must ↵Arthur Hanson Rasmusson38 hours
| | | | advance even when the r391f once-only gate suppresses the write. A re-decode pass that skipped the FIRST run of a split upload left the cursor at the first run's destination, so the SECOND run's data landed at offset 0 (the deterministic tgemm X[0..1]<-X[2..3] 8-byte overwrite, present in every pass); the gate suppresses the write, not the method stream's positional semantics
* r510 keep the page-directory discovery attempts for coherent evidence: a ↵Arthur Hanson Rasmusson42 hours
| | | | | | | | | | | | | | large mixed-space map no longer burns them r508/r509 witness: 'page directory discovery: not found (base=0x0 ap=ram cands=39621 best=0/8)' during the very first client - eight RAM walks spent on a stride subsample of the boot-scoped map, which holds pairs from several VA spaces and (per the code's own comment) no single directory reproduces. By the time the push-buffer pairs - the strongest samples - formed, the attempts budget was gone; every later client's high argument pointers stayed unresolvable and the boot ended in the held-launch class. Only a small map (one session's own pairs) still falls back; a larger one waits for the pairs and keeps the attempts.
* r509 the GR data-hold bound can report expiry: one-shot verdict + latch, and ↵Arthur Hanson Rasmusson42 hours
| | | | | | | | | | | | | | | | | | the r351 retry-forward consumes it svc_gr_hold_bound_expired() re-armed the deadline on the very call that detected expiry and then compared jiffies against the NEW deadline - the function could never return true. A data-stalled launch whose arguments never resolve therefore held its entry forever (r508 witness: matmul plus ten sibling channels in 'GR hold ... data args unresolved' with zero GR-STALL-hold-bound lines for 10+ minutes; the designed one retry-forward plus consume never ran). The expiry is now one-shot: push the deadline, latch gr_hold_bound_hit (snapshot-persisted) for the second call site of the same pass, return true; the r351 retry-forward block consumes the latch and forces its single best-effort forward, and the NEXT expiry cycle (gr_hold_exp_logged now set) falls through to the historical consume with the software completion - a never-resolving launch now costs ~6 s and the client sees a stale-data FAIL instead of a hang.
* r507o walk the CE channels before the GR channels: enforce the client's ↵Arthur Hanson Rasmusson42 hours
| | | | | | | | | | | | | | | stream order in the pump itself The streams order their H2D copies before the kernel that consumes them, but the channels register in handle order (the RM's GR channel early, the copy-CE channels later) and the pump walked them in that order; a pass that decoded the GR launch before the CE channel's entries ran let the forward workqueue execute the launch's copy-in against pre-copy data (r480-sep: copy-in 115 ms before its stream's H2D soft copies; r481 proved no host semaphore links the two). The CE copies execute on the same service thread, so consuming the CE entries first fixes the inversion without any cross-channel hold. Stable partition of the pass snapshot; relative order within each class is preserved.
* r507a op-failure latch is time-based: a full opdef window before it stops ↵Arthur Hanson Rasmusson42 hours
| | | | | | | | | | | | holding (r507 matmul lost a pin-late 1 MB UVA copy and verified FAIL) r506w's attempt-count latch tripped after 8 fast passes (~100 ms) while the legitimate pin-late retry needs 0.5-2 s (r507 witness: memcopy dst=0x71365ee02200 src=0x71365e500000 bytes=1048576 rc=-2 skipped; matmul completed with STATUS FAIL instead of hang). The first failure of a VA now arms a full 10000 ms window during which the op holds its entry exactly as before; only after a full bound with no success does the op stop forcing the hold. A success clears the record (fresh window on a later failure).
* r506w kill the r505/r506 post-DEVICE grind: bounded op-failure latch + exact ↵Arthur Hanson Rasmusson42 hours
| | | | | | | | | | | | | | | | | | | | | | | | | | | release routing Two independent defects turned one unresolvable virtual op into a client-wide hang (r505 multikernel, r506 matmul: 'client at DEVICE forever' while the pump ground its stream at ~10 s per push): - an op that keeps failing -ENOENT re-runs on every decode pass and re-set svc.op_tr_pending each time, so its push held for the full SVC_OPDEF_HOLD_MS opdef bound every pass (r506 witness: the RM's channel-progress zeroing at 0x2000c0xxx, 1913 failed attempts, 65 ring entries in ~90 s). New per-VA failure latch: after 8 failures the op still attempts its write but stops forcing the hold; the entry consumes and its withheld release applies - exactly the outcome the expired bound already sanctioned, minus the 10 s tax. - svc_rel_apply_launch routed EVERY release through the exact-slot counter fallback (release_pa ?: counter_pa), which writes (and forward-only-skips against) the channel's registered completion slot - a different word. The tracker releases to 0x20480fef0 were skipped against the polluted PUT counter (cur=96) and lost: rel-apply h=0x5c000059 pa=0x10809def0 pay=1 cur=96 -> skip. The fallback is now restricted to counter VAs; generic releases are forward-only by value (payload > fresh cur) so a decode blinded by a stale landing can no longer drop them.
* r505b opdef flush marks the slot complete; bound the slot arrays for large ringsArthur Hanson Rasmusson47 hours
| | | | | | | | | | | | | | | Two fixes for the r505 multikernel livelock (opdef held=18359, expired=61, client at DEVICE forever while the pump looped on ring slot 1): - after the bounded opdef hold expires and svc_opdef_flush delivers the withheld releases, the slot's ops are permanently skipped - but the decode-time bookkeeping had stored slot_ops_complete=0 (op_tr_pending was set when it ran), so every pin-batch rewind replay of the slot re-decoded it, re-set op_tr_pending and re-held the entry forever. Mark the slot complete on flush so a replay suppresses the ops (entry_replay) and the entry consumes normally. - the per-slot op arrays are SVC_GPFIFO_MAX_BITS (1024) but the X/GL channel's ring is 8192 slots: bound every slot-index write/read (entry_ident is zeroed past the array size, so the bookkeeping skips).
* r496j VAS candidates are registered only after a content-verified ring walk ↵Arthur Hanson Rasmusson47 hours
| | | | | | | | | | | | | | | | | | | | (+3 fixes) The r496g blast registration of every COPY_SERVER_PDES directory let a stale kernel directory answer another client's VA in the walker's first-success order (the r459 stale-directory class) - measured as a CUDA client hanging after DEVICE. Candidates are now parked and registered only when a walk of the client's own content-validated ring VA through them reproduces that ring's PA; the r496f vidmem sweep remains as a fallback. Also in this batch: - svc_chan_userd's VRAM shape scan now prefers a block with work PENDING (get < put) and then the one nearest the registered memdesc offset: on a channel re-registration the stale previous USERD still validates (it holds the same put/entry values) and the old first-match adoption made the pump write GP_GET into the dead block (r504 second DMA init). - The USERD/FB scan and the CE caps byte-1 BL bits (r496i) are unchanged; TU104_CE_CAPS1 = 0x03 needs the CUDA battery re-verification.
* r496h do not pin a linear window for content-scanned rings; the VAS walk ↵Arthur Hanson Rasmusson48 hours
| | | | | | | | | | | resolves the rest The registered COPY_SERVER_PDES directory walks the ring VA correctly (trace: L0 -> L1 -> L2 idx 9 -> PD0), but the linear window pin installed around the ring shadows it in svc_lookup_trusted (pin_lookup_any returns the win-flagged entry first), so the pushbuffer/tracker reads kept using the wrong noncontiguous guess. Pin only the content-validated ring page; every other VA of the client now resolves through the walk.
* r496f discover the kernel-RM client VAS page directory by verificationArthur Hanson Rasmusson2 days
| | | | | | | | | | | | | | The X/GL channel's pushbuffer pages and progress semaphore live in a kernel-RM internal VAS whose page directory is never published (fn 54 is UVM/external-VAS only), so no translation source can resolve them: the tracker push reads zeros/garbage and the driver's DMA sanity poll (which waits on the progress-tracker semaphore the GPU writes when it consumes the tracker push) never completes. With the ring content-validated, try each 4 KB page of the vidmem shadow as a walk root and register the first whose walk reproduces the ring VA -> ring PA translation. The walker then resolves every other VA of the same client - pushbuffer pages, the tracker push, the progress semaphore.
generated by cgit v1.2.3 (git 2.39.1) at 2026-10-24 22:16:23 +0000