8822E: fix MCS4+/48M+ TX, resolve the 0x41e8 RX-desense quirk, port kernel channel/runtime obligations + quirks doc (#238)#268
Merged
Conversation
…nel/runtime obligations (#238) Root causes found by golden end-state diff (kernel via /proc read_reg live vs devourer via new DEVOURER_BB_DUMP, 0x000..0x4ffc) + subset bisection with a Jaguar3 DEVOURER_REPLAY_WSEQ port, kernel-TX ground truth on the same unit. Fixes (all bench-validated, 8812EU->8822CU ch36 + SDR): - MCS4+/48M+ whole-PPDU TX corruption: route the DPDT antenna switch to WL control (REG_LED_CFG 0x4c[24] = BIT_DPDT_WLBT_SEL, [22] clear; halmac pinmux family devourer's hand-rolled MAC init skipped) + single-path OFDM 1SS TX routing (0x820: 1ss-B@5G / 1ss-A@2G, kernel semantics; the old 1ss-on-AB mapping interacts with the DPDT fix and wedges MCS0). MCS7 goes 0 -> ~5000 clean/12s EVM -48; full ladder MCS0/4/7/54M clean. - 0x41e8 path-B RX-desense quirk RESOLVED: it was a downstream artifact of the same front-end mis-config, not TSSI asymmetry. Structural skip retired; path-B OFDM TX power is controllable in TX+RX mode again (full-duplex validated: 24k RX frames + 14.3k clean MCS7 simultaneously). - RF reg 0x0 writes: direct-window writes silently no-op; route through the legacy FON port 0x1808/0x4108 (addr<<20|data) in Halrf8822e, Halrf8822c and the radio-table loader (vendor tables carry reg-0 entries). - Construction-time TX-power state was clobbered by the FW power/coex H2Cs: authoritative TXAGC re-apply now runs post-coex (with readback diff log); PAD_CTRL1[29:28] re-asserted there too. - Channel-switch ports from the kernel switch_channel tail: spur elimination (14 harmonic channel/BW combos, NBI+CSI mask; FastRetune declines spur hops), CCK TX shaping filter (per-2G-channel, ch14 special), IGI toggle on BOTH Jaguar3 variants (fixes the intermittent post-switch RX deafness / MCS0 first-cell TX wedge; CU ch6 ground liveness now 5/5). - LCK synthesizer re-lock (AACK+RTK on >=4-unit thermal drift) wired into the coex-tick thermal tracker; fired live in a 22-min max-duty soak (1.10M clean MCS7 frames, EVM drift +3 dB). Verified non-gaps: TSSI subsystem (kernel runs THERMAL mode here, 0x1e7c[30] = 0 both drivers), thermal codeword LUT (end-state identical), TX descriptor (48B, byte-identical). 2.4 GHz TX airs undecodable frames under BOTH drivers (kernel-parity, module property) -- README row corrected, quirk documented. Docs: docs/8822e-quirks.md (the issue deliverable -- every quirk root-caused or kernel-parity-proven, with reproducers). Diag: replay hook at end of both bring-ups, DEVOURER_BB_DUMP, coex-tick TXAGC register watch, DEVOURER_TX_PWR_OFFSET_QDB on txdemo. Tests: 14 new harnesses (tests/eu_*.sh, cu_2g_ground_liveness.sh, sdr_burst_len.py); txpwr_offset_regcheck.sh path-B cell inverted to the new semantics. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…D/EDCCA) The vendor watchdog's RX-side adaptation loop, run every ~2 s like the kernel's: false-alarm/CCA window statistics (read + reset, phydm_fa_cnt_statistics_jgr3), unlinked DIG (IGI stepping in 0x1d70 by FA level, DFS channels pinned to 0x20), CCK packet-detection type4 (CCK-FA moving average driving the PD/CS ladder in 0x1ac8/0x1acc/0x1ad0), and EDCCA threshold tracking (0x84c, th_l2h = max(IGI+8, 48) / -8 — matches the kernel's end state; suppressed while SetCcaMode's EDCCA-disable is active). The DIG coverage window floors at 0x1e, not phydm's generic DIG_MIN_COVERAGE 0x1c: at IGI 0x1c the 8822CU's dense-QAM RX decodes NOTHING (0 of 65k kernel-injected MCS7 frames; 0x1e passes 65.9k) — hardware-bisected, value-specific, the same class of per-IC "For HW setting" floor exception the kernel carries for other ICs. TX/TX+RX sessions tick from the coex thread; RX-only sessions get a dedicated housekeeping thread — register I/O cannot run on the bulk-IN event thread (a sync control transfer there starves its own event loop). The kernel's remaining watchdog mechanisms self-disable without a link and are intentionally absent: CFO tracking (returns unless associated), RA and the beamforming watchdog (no station entries), antenna diversity (2T2R). Matched-power TX EVM parity verified (tests/eu_matched_power_evm.sh, one ground session, refs bit-identical at devourer offset -44 == kernel `txpower fixed 500`): MCS7 median EVM -60 both drivers; MCS0 -64 vs -66. Default-power ladder unchanged (MCS7 6407 clean / -48, MCS0 5680 / -27). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes the reproducible half of #238 on the reference hardware (LB-LINK BL-M8812EU2, rfe-21) and lands the issue's deliverable,
docs/8822e-quirks.md. Everything below is hardware-validated on the two-adapter bench (8812EU ↔ 8822CU, plus the vendor kernel and an in-tree rtw88 ground as controls); reproducers ship undertests/.Root causes & fixes
REG_LED_CFG 0x4c[24]=BIT_DPDT_WLBT_SELset,[22]clear, applied post-coex;txpath_1sssemantics):0x820=0x32(1ss-B) at 5 GHz /0x31at 2.4 GHz. The old 1ss-on-both mapping interacts with the DPDT fix (MCS0 wedges).Result: MCS7 goes 0 → ~5000 clean/12 s EVM −48; full ladder MCS0/4/7/54M clean; 40 MHz 5226 @ −53; 80 MHz TX verified against an independent kernel ground (5658 frames).
0x41e8path-B RX-desense) RESOLVED: it was a downstream artifact of the same front-end mis-config, not TSSI asymmetry (tests/eu_41e8_desense_recheck.sh: poking the ref with RX live = −2 %, noise). The structural skip is retired — path-B OFDM TX power is controllable in TX+RX mode again (full-duplex proof: 24k RX frames + 14.3k clean MCS7 simultaneously,tests/eu_fullduplex_pathb_check.sh).0x0writes silently no-op'd through the direct BB window — routed through the legacy FON port0x1808/0x4108per the kernel, in both halrf impls AND the radio-table loader (the vendor tables carry reg-0 entries). Verified via the TXGAPK gain-table readback (all-zero before, kernel-shaped ramps after).PAD_CTRL1[29:28]re-asserted there too.Kernel channel/runtime obligations ported (systematic source diff vs
reference/rtl88x2eu)PhydmRuntimeJaguar3): FA/CCA statistics, unlinked DIG, CCK-PD type4, EDCCA tracking — with one hardware-forced deviation: the DIG floor is0x1e, not phydm's generic0x1c(at0x1cthe 8822CU's dense-QAM RX decodes 0 of 65k MCS7 frames;0x1epasses 65.9k — same class as the kernel's per-IC "For HW setting" exceptions). RX-only sessions tick from a dedicated thread (register I/O on the bulk-IN event thread starves its own event loop).Verified non-gaps & closures
0x1e7c[30]=0under both drivers); thermal LUT already at parity; TX descriptor byte-identical (48 B).tests/eu_matched_power_evm.sh, refs bit-identical): MCS7 median −60 under both drivers.RF 0x1aTX_CCK_IND held post-IQK in full and fast paths).Out of scope / follow-ups
0bda:e822) and board variance: blocked on hardware (documented in the quirks doc).ctest 19/19 across the rebase onto current master; per-chip config subsets build (spot-checked no-Jaguar3).
🤖 Generated with Claude Code