diff options
| -rw-r--r-- | kernel/gsp_sema_service.c | 107 | ||||
| -rw-r--r-- | kernel/gsp_sema_service.h | 4 | ||||
| -rw-r--r-- | kernel/main.c | 8 |
3 files changed, 118 insertions, 1 deletions
diff --git a/kernel/gsp_sema_service.c b/kernel/gsp_sema_service.c index f90ed2c..cb9d731 100644 --- a/kernel/gsp_sema_service.c +++ b/kernel/gsp_sema_service.c @@ -3530,6 +3530,9 @@ static inline bool svc_pool_pa_identity(u64 va, u64 pa) static void svc_pool_boot_store(u64 va, u64 pa, int ap); /* m168 */ static bool svc_dst_guard_hold(u64 va); /* m169b */ +static bool svc_dst_guard_should_hold(u64 va); /* r609 */ +static void svc_dst_guard_age(void); /* r609 */ +static void svc_dst_guard_reset(void); /* r609 */ static bool svc_dst_guard_pending(u64 va); /* r231 */ static u32 svc_read_order_log; /* r231 budget */ static u32 svc_atrace_log; /* r234 A-TRACE budget */ @@ -6890,7 +6893,7 @@ static int svc_copy_1d(struct nv_vgpu_device *vgpu, u64 dst, u64 src, size_t tot if (dst_virt && dst_aperture == CE_APERTURE_VID && dst_prov >= 3 && dst_orig >= SVC_VAS_FLOOR && gsp_sema_copy_dst_pin_guard && - svc_dst_guard_hold(dst_orig)) { + svc_dst_guard_should_hold(dst_orig)) { svc.copy_dst_hold++; if (svc.copy_dst_hold <= 16) pr_info("sema-svc: device-dst copy held: dst=0x%llx prov=%u (no pin yet; stale-table guard)\n", @@ -12305,6 +12308,7 @@ static void svc_session_flush(void) svc_gr_diag_reset(); /* m150: per-session diagnostic budgets */ svc_gr_defer_reset(); /* r604: per-session defer latches */ + svc_dst_guard_reset(); /* r609: per-session device-dst windows */ svc.map_n = 0; svc_pb_n = 0; svc_pd_found = 0; @@ -14833,6 +14837,88 @@ bool gsp_sema_wait_gate = 1; unsigned gsp_sema_copy_dst_guard_ms = 3000; /* + * r609: the m169 first-touch window is per PAGE, but the stale-table risk it + * covers is per CLIENT TABLE BUILD. The driver's H2D relocation burst on the + * eng==0 copy channel writes N distinct first-write pages back to back, and + * because the pump holds the FIRST held entry and the rest of the chain + * decodes only after it drains, each page paid its own full window, + * serialized (r606 red draw: `sema-pend: hold h=0x5c000059 idx=90` held + * 362167.7->362170.8, idx=92 held 362177.7->362182.9; the b H2D entry ran + * 362182.882, 326 ms AFTER the launch forward -> `VADD bad=32735 + * csum=8139264`, c = a + 0). The guard's bet - a fresh pin lands inside the + * window - is a property of the client's table state, not of one page: once + * ONE device-dst page of a channel has waited the full window with NO pin + * arriving (every observed release logs "without a pin"), the same bet for + * the next page of the same burst is already lost. Age the channel for + * gsp_sema_copy_dst_age_ms after such a release; further first-write + * device-dst pages of that channel inside the window skip the hold and run + * with the walk translation - the exact translation the expired guard would + * have used anyway. The first page of every burst still pays the full + * window, so the anti-stale bet is unchanged; only the N x window + * serialization is removed (r606 class: ~20 s of chain -> one window). + * Keyed by h_object like every channel table; the session flush clears it, + * so a recycled handle can never inherit the previous client's ageing + * (the r604 lifetime lesson). 0 disables (= per-page windows, pre-r609). + */ +unsigned gsp_sema_copy_dst_age_ms = 3000; +#define SVC_DST_AGE_SLOTS 16 +static u32 svc_dst_age_ch[SVC_DST_AGE_SLOTS]; +static unsigned long svc_dst_age_j[SVC_DST_AGE_SLOTS]; +static u32 svc_dst_age_pos; +static u32 svc_dst_age_log; + +/* r609: the current decode channel's guard window expired without a pin. */ +static void svc_dst_guard_age(void) +{ + u32 i; + + if (!gsp_sema_copy_dst_age_ms) + return; + for (i = 0; i < SVC_DST_AGE_SLOTS; i++) + if (svc_dst_age_ch[i] == svc.cur_ch) { + svc_dst_age_j[i] = jiffies; + return; + } + i = svc_dst_age_pos++ % SVC_DST_AGE_SLOTS; + svc_dst_age_ch[i] = svc.cur_ch; + svc_dst_age_j[i] = jiffies; +} + +/* r609: is the current decode channel inside its post-expiry ageing window? */ +static bool svc_dst_guard_aged(void) +{ + unsigned long now = jiffies; + u32 i; + + if (!gsp_sema_copy_dst_age_ms || !svc.cur_ch) + return false; + for (i = 0; i < SVC_DST_AGE_SLOTS; i++) + if (svc_dst_age_ch[i] == svc.cur_ch && + time_before(now, svc_dst_age_j[i] + + msecs_to_jiffies(gsp_sema_copy_dst_age_ms))) + return true; + return false; +} + +/* + * r609: decide whether a first-write device-dst page pays the m169 window. + * An aged channel skips it (the burst's first page already paid and no pin + * arrived); otherwise the normal per-page window runs. + */ +static bool svc_dst_guard_should_hold(u64 va) +{ + if (svc_dst_guard_aged()) { + if (svc_dst_age_log < 16) { + svc_dst_age_log++; + pr_info("sema-svc: device-dst guard age-skipped dst=0x%llx h=0x%x (same-channel burst after an un-pinned release; walk translation)\n", + va, svc.cur_ch); + } + return false; + } + return svc_dst_guard_hold(va); +} + +/* * r231 READ-ORDER guard (default 0: witness only). A device-src copy whose * source VA has a PENDING device-dst write hold (the m169 guard window is * still open) reads the PRE-write content: the writer's bytes land only @@ -14881,6 +14967,7 @@ static bool svc_dst_guard_hold(u64 va) if (svc_dst_guard[i].first) { svc_dst_guard[i].first = 0; svc.copy_dst_relaxed++; + svc_dst_guard_age(); /* r609: burst ageing */ pr_info("sema-svc: device-dst guard released without a pin va=0x%llx (using the walk translation)\n", page); } @@ -14897,6 +14984,24 @@ static bool svc_dst_guard_hold(u64 va) } /* + * r609: clear the m169 page windows and the r609 ageing ring with the + * session. Both are keyed by VA page / h_object alone, and both would + * otherwise pre-age the next client's first touch (a page slot still inside + * its window makes the new client's first device-dst write skip the + * anti-stale wait; an ageing stamp makes the new channel's whole first + * burst skip it). The stale-table risk this guard covers is precisely the + * new-session condition, so the windows must start fresh. + */ +static void svc_dst_guard_reset(void) +{ + memset(svc_dst_guard, 0, sizeof(svc_dst_guard)); + svc_dst_guard_next = 0; + memset(svc_dst_age_ch, 0, sizeof(svc_dst_age_ch)); + memset(svc_dst_age_j, 0, sizeof(svc_dst_age_j)); + svc_dst_age_pos = 0; +} + +/* * r231 READ-ORDER guard: true while a held device-dst write for this VA * page is still inside its bounded window (the writer's bytes have not * landed yet). See the gsp_sema_read_order_guard declaration. diff --git a/kernel/gsp_sema_service.h b/kernel/gsp_sema_service.h index 8664954..f891de1 100644 --- a/kernel/gsp_sema_service.h +++ b/kernel/gsp_sema_service.h @@ -329,6 +329,10 @@ extern bool gsp_sema_pool_hold_fresh; * walk/RAM scan until a pin confirms (default on). */ extern bool gsp_sema_copy_dst_pin_guard; +/* r609: after a device-dst guard window expires WITHOUT a pin, skip the + * window for further first-write device-dst pages of the same channel for + * this long (ms). 0 restores the per-page windows (pre-r609). */ +extern unsigned gsp_sema_copy_dst_age_ms; extern bool gsp_sema_wait_gate; /* r401: bounded acquire-wait hold (fail-open) in ms; must exceed the * emulated launch completion latency for copy-after-launch ordering. */ diff --git a/kernel/main.c b/kernel/main.c index 81bae08..e340b7b 100644 --- a/kernel/main.c +++ b/kernel/main.c @@ -2876,6 +2876,14 @@ MODULE_PARM_DESC(sema_wait_hold_ms, "bounded acquire-wait hold before fail-open, module_param_named(sema_push_raw, gsp_sema_push_raw, uint, 0644); MODULE_PARM_DESC(sema_push_raw, "r481 diagnostic: dump the decoded method stream of the first N large GR pushes (default 0 off)"); MODULE_PARM_DESC(sema_copy_dst_pin_guard, "hold device-dst copies without a pin-resolved translation (default 1)"); +/* + * r609: after a device-dst guard window expires WITHOUT a pin, skip the + * window for further first-write device-dst pages of the same channel for + * this long (ms) - the r606 held-chain serialization fix. 0 restores the + * per-page windows (pre-r609). + */ +module_param_named(sema_copy_dst_age_ms, gsp_sema_copy_dst_age_ms, uint, 0644); +MODULE_PARM_DESC(sema_copy_dst_age_ms, "device-dst guard burst-ageing window after an un-pinned release, ms (r609, default 3000; 0 off)"); /* * r384 (guest boot-corruption lottery): fill-destination guards. A large |