diff options
| author | Arthur Hanson Rasmusson <arthur@vgpu.io> | 2026-10-21 12:30:54 -0400 |
|---|---|---|
| committer | Arthur Hanson Rasmusson <arthur@vgpu.io> | 2026-10-21 12:30:54 -0400 |
| commit | 67a85365fe3b0da7647edcbaa1d8a529a01a3d94 (patch) | |
| tree | 1e9891586d4645e3f0c7451516d9168f1974d66a | |
| parent | 2ae643432b2d1e37e90191afa6fbac042967bfe7 (diff) | |
| download | open-iov-67a85365fe3b0da7647edcbaa1d8a529a01a3d94.tar.gz open-iov-67a85365fe3b0da7647edcbaa1d8a529a01a3d94.zip | |
gr: r411 sticky wedge episode - quiet the first retry of any zero-write launch once the boot has seen a quiet heal (or a no-write accept)
r418 with the r410 quiet path showed the wedge alternating one launch at a
time: every second push no-wrote and was accepted (streak was reset by
the healed launch in between, so the quiet trigger never armed), e.g.
c584 NOWRITE / c92c healed / d424 NOWRITE / d7cc healed / ... 6 holds,
6 heals, 7 accepted no-writes. A zero-write accept (or a successful quiet
heal) now arms a sticky wedge episode for the boot; any later zero-write
launch takes the quiet path on its first retry, so in an episode every
wedged launch either heals (engine writes, client data correct) or is
classified loudly instead of riding through the quick-retry accept.
| -rw-r--r-- | kernel/gsp_phys_booter.c | 23 |
1 files changed, 21 insertions, 2 deletions
diff --git a/kernel/gsp_phys_booter.c b/kernel/gsp_phys_booter.c index b966ac3..d20e042 100644 --- a/kernel/gsp_phys_booter.c +++ b/kernel/gsp_phys_booter.c @@ -16314,6 +16314,16 @@ static u64 gsp_phys_gr_stall_retry; static u32 gsp_phys_gr_nw_streak; static u64 gsp_phys_gr_wedge_quiet; static u64 gsp_phys_gr_wedge_heals; +/* + * r411: once a quiet heal has succeeded on this boot, the lane is in a + * wedge-prone episode. r418 showed the wedge can alternate one-no-write / + * one-heal, and the healed launch resets nw_streak, so a streak-only trigger + * quieted only every second wedged launch and the other was accepted. The + * heal count makes the episode sticky for the boot: any later zero-write + * launch gets the quiet path on its first retry, so every wedged launch + * either heals (engine writes, client data correct) or is loudly classified. + */ +static bool gsp_phys_gr_wedge_episode; /* m86: program-page override blob (see gsp-phys-gr-override). */ static u8 gsp_phys_gr_override[GSP_PHYS_PAGE_SIZE]; static u32 gsp_phys_gr_override_len; @@ -20529,7 +20539,8 @@ gr_page_unmapped: while (wrote == 0 && nw_attempts < nw_retries) { u32 quiet_ms = 0; - if (READ_ONCE(gsp_phys_gr_nw_streak) > 0 && + if ((READ_ONCE(gsp_phys_gr_nw_streak) > 0 || + READ_ONCE(gsp_phys_gr_wedge_episode)) && nw_attempts <= 1) { quiet_ms = gsp_phys_gr_wedge_quiet_ms_value(); @@ -20833,6 +20844,8 @@ gr_page_unmapped: if (quiet_used) { if (wrote) { gsp_phys_gr_wedge_heals++; + WRITE_ONCE(gsp_phys_gr_wedge_episode, + true); pr_warn("vgsp-gr: WEDGE-QUIET healed: engine wrote %u page(s) after the %u ms hold (streak was %u)\n", wrote, quiet_used, READ_ONCE( @@ -21003,10 +21016,16 @@ gr_page_unmapped: * even when the retry loop is disabled so the streak always * reflects the accepted-outcome history. */ - if (n_tracked) + if (n_tracked) { WRITE_ONCE(gsp_phys_gr_nw_streak, wrote ? 0u : READ_ONCE(gsp_phys_gr_nw_streak) + 1u); + /* r411: arm the wedge episode on any accepted + * zero-write launch, so the next zero-write launch + * quiets even after a heal reset the streak. */ + if (!wrote) + WRITE_ONCE(gsp_phys_gr_wedge_episode, true); + } } /* * m114: the kernel ran against the pre-commit bounce window; copy the |