RX: keep multiple bulk-IN URBs in flight (mirror kernel's always-posted model)#107
Merged
Conversation
…ed model) The monitor RX loop did one synchronous libusb_bulk_transfer at a time, so between transfers no URB is posted. usbmon ground-truth shows the kernel (rtw88) instead keeps 4 bulk-IN URBs always posted and re-submits each ~20us after completion. With a single posted URB the 8814's RX-DMA ring pauses in the parse gap and, under sparse traffic, intermittently wedges after a short burst — the "8814 RX stalls after ~10 frames" symptom (8812/8821 tolerate it). Run infinite_read() from a small worker pool (default 4) instead of a single loop. infinite_read() is already reentrant — local 32KB buffer + local FrameParser, and libusb is thread-safe — so the workers just give the same always-posted behaviour the kernel has; packetProcessor is serialised under a mutex. Frame order across workers is not preserved, which is fine for monitor-mode RX. DEVOURER_RX_URBS overrides the worker count (1 restores the old single-transfer behaviour). Validated: usbmon confirms 4 bulk-IN URBs in flight; regression-clean on RTL8812AU / RTL8821AU (300+ frames) and RTL8814AU. ctest green. Note: the intermittent 8814 stall it targets is timing-dependent and was dormant at commit time, so this hardens the RX path to match the kernel's proven model rather than being confirmed against a live repro. Co-Authored-By: Claude Opus 4.8 (1M context) <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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What
The monitor-mode RX loop did one synchronous
libusb_bulk_transferat a time, leaving a window between transfers where no URB is posted. usbmon ground-truth on the same hardware shows the kernel (rtw88) instead keeps 4 bulk-IN URBs always posted, re-submitting each ~20 µs after completion.With a single posted URB, the RTL8814AU's RX-DMA ring pauses in the parse gap and, under sparse traffic, intermittently wedges after a short burst — the "8814 RX stalls after ~10 frames" symptom (the 8812/8821 tolerate it; the 8814 does not).
Change
Run
infinite_read()from a small worker pool (default 4) instead of a single loop.infinite_read()is already reentrant — local 32 KB buffer + localFrameParser, and libusb is thread-safe — so the workers simply restore the always-posted behaviour the kernel has.packetProcessoris serialised under a mutex. Frame order across workers isn't preserved, which is fine for monitor-mode RX.DEVOURER_RX_URBSoverrides the worker count (1restores the old single-transfer behaviour).Validation
ctestgreen.88XXau/rtw88_8814au) which receives thousands of frames where devourer's single-URB path stalled — confirming the 8814 RX hardware is healthy and the stall was a host-side single-URB issue, not RF/rig degradation.Honest caveat
The intermittent 8814 stall this targets is timing-dependent and was dormant at commit time, so this hardens the RX path to match the kernel's proven always-posted model rather than being confirmed against a live reproduction. It is low-risk (reentrant call, mutex-serialised delivery, env escape hatch) and regression-validated on all three Jaguar chips.
🤖 Generated with Claude Code