VER-895: wifi: ath12k: fix dp_peer stale entry on station removal failure

[GadiAvr]

wifi: ath12k: fix dp_peer stale entry on station removal failure

When stopping hostapd, mac80211 triggers a state transition for stations
from IEEE80211_STA_NONE to IEEE80211_STA_NOTEXIST.

In ath12k_mac_op_sta_state, if the WMI command to delete a peer fails
(e.g., due to a firmware timeout), the current error handler jumps
directly to the exit label. This skips the second stage of cleanup:
ath12k_dp_peer_delete().

As a result, an orphaned ath12k_dp_peer entry remains in dp_peers_list.
When the station tries to reconnect, ath12k_dp_peer_create() finds the
stale entry and returns -EEXIST, leading to the error:
"unable to create ath12k_dp_peer for sta".

Fix this by adding a peer_cleanup label that ensures ath12k_dp_peer_delete()
is called even if the firmware-side removal failed, as mac80211 will
remove the station regardless.

Tests: Automation test- 100 times of stop/start hostapd (by changing the RF configuration)

[michaeln-ceragon]

@dima1308 @adrian-nicolau guys please review and approve this fix. From my point of view fix is good, but I would like to have one more opinion

[Copilot]

Pull request overview

Fixes an ath12k station teardown error path where a firmware-side peer delete failure could skip datapath cleanup, leaving a stale dp_peer entry that blocks re-association with -EEXIST.

Changes:

• Add a peer_cleanup error path to ensure ath12k_dp_peer_delete() is executed on IEEE80211_STA_NONE -> IEEE80211_STA_NOTEXIST failures.
• Add a warning log when forced cleanup is performed.
• Minor whitespace-only change near ath12k_mac_register().

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[adrian-nicolau]

@dima1308 @adrian-nicolau guys please review and approve this fix. From my point of view fix is good, but I would like to have one more opinion

You can also search Qualcomm's driver to see if they handle it differently:
https://git.codelinaro.org/clo/qsdk/oss/src/mac80211/wlan-open/-/blob/win.wlan_host_opensource.4.0.r37/backports-6.1-9a0dddfb3/drivers/net/wireless/ath/ath12k/mac.c?ref_type=heads

[Copilot]

Pull request overview

Fixes ath12k station teardown so that driver-side dp_peer cleanup still occurs when per-link station removal fails (e.g., firmware timeout during peer delete), preventing stale dp_peer entries that can block subsequent reconnects.

Changes:

• Adjust error handling in ath12k_mac_op_sta_state() to treat IEEE80211_STA_NONE -> IEEE80211_STA_NOTEXIST teardown failures as non-fatal from mac80211’s perspective.
• Ensure the dp_peer deletion path is reached for teardown failures by redirecting to the existing peer_delete cleanup.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[GadiAvr]

resolve copilot comment

[GadiAvr]

Resolve comments

[Copilot]

Pull request overview

Fixes ath12k station teardown so driver-side datapath peer state is cleaned up even when firmware-side peer deletion fails during IEEE80211_STA_NONE -> IEEE80211_STA_NOTEXIST (e.g., hostapd stop with FW timeout), preventing stale dp_peer entries that block reconnects.

Changes:

• Adds a dedicated cleanup jump target for the STA_NONE -> STA_NOTEXIST failure path.
• Ensures ath12k_dp_peer_delete() is still executed when ath12k_mac_handle_link_sta_state() fails during station removal.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

This PR addresses a reconnect failure in ath12k caused by stale dp_peer entries when station removal proceeds in mac80211 but firmware-side peer deletion fails (e.g., WMI timeout), ensuring driver-side DP cleanup still runs.

Changes:

• Redirect an error goto path to a new peer_cleanup label so ath12k_dp_peer_delete() can still run after certain failures.
• Introduce peer_cleanup: label before the station removal cleanup block.

Comments suppressed due to low confidence (1)

drivers/net/wireless/ath/ath12k/mac.c:7794

• Because peer_cleanup: is now a jump target from an error path, the unconditional ret = 0; below this block will overwrite the earlier failure code. That prevents callers/logs from seeing the firmware-side delete failure. Consider only setting ret = 0 on the non-error fallthrough path (e.g., guard with if (!ret) or goto exit immediately after performing the dp peer cleanup when ret is already non-zero).

peer_cleanup:
	/* IEEE80211_STA_NONE -> IEEE80211_STA_NOTEXIST:
	 * Remove the station from driver (handle ML sta here since that
	 * needs special handling. Normal sta will be handled in generic
	 * handler below
	 */
	if (old_state == IEEE80211_STA_NONE &&
	    new_state == IEEE80211_STA_NOTEXIST) {
		if (sta->mlo)
			ath12k_mac_ml_station_remove(ahvif, ahsta);

		ath12k_dp_peer_delete(&ah->dp_hw, sta->addr, sta);
	}

	ret = 0;
	goto exit;

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

This PR addresses an ath12k station removal edge case where a firmware-side peer delete failure can skip datapath peer cleanup, leaving a stale dp_peer entry that later prevents the station from reconnecting.

Changes:

• Adds a peer_cleanup jump target for the IEEE80211_STA_NONE -> IEEE80211_STA_NOTEXIST transition when link STA state handling fails.
• Ensures ath12k_dp_peer_delete() is still invoked even if the WMI peer delete fails.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[GadiAvr]

Resolve comments

[Copilot]

Pull request overview

This PR addresses an ath12k teardown edge case where a firmware-side peer delete failure can skip host-side cleanup, leaving a stale dp_peer entry that prevents subsequent station reconnects.

Changes:

• Route the IEEE80211_STA_NONE -> IEEE80211_STA_NOTEXIST failure path to a new peer_cleanup label so ath12k_dp_peer_delete() still runs.
• Ensure the teardown path returns success to mac80211 to avoid warnings and state-machine stalls.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

This PR hardens station teardown in ath12k_mac_op_sta_state() by ensuring driver-side datapath peer cleanup still happens when firmware-side peer deletion fails during the IEEE80211_STA_NONE -> IEEE80211_STA_NOTEXIST transition (e.g., WMI timeout), preventing stale dp_peer entries that block reconnect.

Changes:

• Add a dedicated peer_cleanup path for NONE -> NOTEXIST teardown failures so ath12k_dp_peer_delete() still runs.
• Ensure teardown failures in that state transition do not propagate back to mac80211 (avoiding WARN/stuck state machine).
• Expand inline comments to document why errors are suppressed for this transition and what cleanup is still required.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.