mac80211: improve ath12k recovery and link handling - #2
Open
KakatkarAkshay wants to merge 5 commits into
Open
Conversation
The AHB driver currently uses the Qualcomm SSR notifier only to wait for rootPD startup. Handle crash shutdown notifications as well so the actual failure is visible instead of only its later effects. Release an SCM-authenticated userPD before the remoteproc core collects the rootPD dump. This mirrors the ordering used by Qualcomm's rootPD crash handling without importing its downstream group reset and panic machinery. Adapted from Qualcomm's downstream rootPD crash handling by Sowmiya Sree Elavalagan and Avula Sri Charan. Link: https://github.com/Telecominfraproject/wlan-ap/blob/4dda11dcf7455828e03a753079c520701f2f7a7c/feeds/qca-wifi-7/mac80211/patches-qca/ath12k/0111-0017-QSDK-PLATFORM-wifi-ath12k-Register-and-handle-rootPD-crash.patch Signed-off-by: Akshay Kakatkar <me@kakatkarakshay.dev>
The rootPD crash notifier releases the AHB userPD so remoteproc can collect the complete q6wcss dump, but no ath12k recovery is started. The AHB userPD remains down and PCIe partners stay attached to the stale WSI group. Mark the AHB device crashed before releasing its userPD. Once rootPD power-up completes, queue the existing reset work for every registered device in the WSI group so wireless service recovers without a reboot. Avoid sending an unacknowledgeable userPD stop request after a rootPD crash. Reinitialize userPD completions before each request and always clear the spawn request bit to prevent stale acknowledgements from satisfying a restart. Adapt Qualcomm's downstream recovery ordering without its panic, forced partner assert, or direct remoteproc state manipulation. Signed-off-by: Sowmiya Sree Elavalagan <quic_ssreeela@quicinc.com> Signed-off-by: Avula Sri Charan <quic_asrichar@quicinc.com> Signed-off-by: Akshay Kakatkar <me@kakatkarakshay.dev>
An ieee80211 add event can arrive while a failed wireless setup is still running or teardown is active. Record the retry until teardown completes and reset the retry budget afterward. Devices already down still start immediately. This replaces the IPQ53xx-specific polling hook with the generic event-driven path. Signed-off-by: Akshay Kakatkar <me@kakatkarakshay.dev>
Most embedded devices with integrated ath12k radios store the MAC address in an NVMEM partition. Fetch it through the standard of_get_mac_address API and prefer it over firmware-provided addresses. Use the factory Wi-Fi MAC from the Flint 3 ART partition for its integrated and PCI radios. Signed-off-by: Ansuel Smith <ansuelsmth@gmail.com>
Group addressed frames reach ath12k_mac_op_tx() with no STA and with the TX control link field set to IEEE80211_LINK_UNSPECIFIED, so ath12k_mac_get_tx_link() falls back to ahvif->deflink.link_id. deflink is preallocated storage reused for whichever link of an ML VIF comes up first, and ath12k_mac_unassign_link_vif() memsets it when that link goes away. Its link_id is then zero while the remaining links keep their own ids, so the fallback selects link zero. On an MLD that does not include link zero, for example one covering only 5 GHz and 6 GHz, ahvif->link[0] is NULL and every such frame is warned about and dropped. Select the first driver-backed active link instead. The set of live links is the only valid source for this decision; deflink.link_id describes a storage slot, not a link. The check added in 113-wifi-ath12k-select-valid-link-when-link-zero-is- missing.patch matches an explicit zero in the TX control field and does not cover this path. Signed-off-by: Akshay Kakatkar <me@kakatkarakshay.dev>
Contributor
Author
|
I have tested all of these changes on the GL-BE9300, and they work as intended to resolve the corresponding issues. |
1 task
perceival
added a commit
that referenced
this pull request
Aug 6, 2026
Patch by Akshay Kakatkar, taken from PR #2. Group-addressed and other management frames reach ath12k_mac_op_tx() with no STA and with the TX control link set to IEEE80211_LINK_UNSPECIFIED, so ath12k_mac_get_tx_link() falls back to ahvif->deflink.link_id. deflink is preallocated storage reused by whichever link comes up first, and ath12k_mac_unassign_link_vif() memsets it when that link goes away, leaving link_id == 0 while the surviving links keep their own ids. On an MLD without link 0 every such frame is warned about and dropped. Verified on hardware here rather than by inspection. The trigger is a *partial* teardown - remove the link that owns deflink while another stays up: wifi down radio1 ; sleep 10 ; wifi up radio1 Two negative results pin the mechanism down: `wifi reload` does not reproduce it (both links go away, links_map reaches 0, deflink is legitimately reused), and removing link 2 does not either (link 1 owns deflink, being created first). That is why one of our APs sat at 0 occurrences for 16 hours while the other flooded on the same build. Same trigger, both units: ap2 ch36 non-DFS unpatched 6257 events (760 printed + 5497 suppressed), link id 0, still climbing ap2 ch36 non-DFS patched 0 ap1 ch100 DFS patched 10, all link id 1, one 0.3 s burst, then zero The link-0 flood is gone. The residual on the DFS unit is a different and narrower race: it requests link 1, a real link, during the window that the 60 s CAC opens between the link entering links_map/active_links and its arvif being installed. Bounded and self-limiting, but the read side still trusts the selected link without checking ahvif->link[id] is populated - reported upstream in issue #6. Note when re-testing: a fresh reboot proves nothing, because it clears the stale deflink and a patched and unpatched unit both read zero. And rebuilding the mac80211 package does not change the kernel build stamp, so `uname -v` is not proof the new module is running - compare the ath12k_wifi7.ko hash. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Yv2jeo6zrG622qSnuUC9M
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
2 participants
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.
Summary
Validation
flint3-be9300branchgit diff --checkpasses