[PATCH ath-next] wifi: ath11k: stop every vdev before starting them on a new channel

[PATCH ath-next] wifi: ath11k: stop every vdev before starting them on a new channel

[Christopher Griffin via B4 Relay]

From: Christopher Griffin <christophergriffin05@gmail.com>

On a channel switch, ath11k_mac_update_vif_chan() restarts each vdev
that is up on the new channel (WMI_VDEV_RESTART_REQUEST). On an IPQ8074
5 GHz radio the transmit rates collapse after that, and stay low until
the radio is stopped and started again. From a client heard at -70 dBm,
after a fresh start on channel 149 the downlink was 76 Mbit/s, with the
firmware sending 16-55 % of its PPDUs at MCS 0-4 (HTT PDEV_TX_RATE
stats). After two switches, 149 -> 157 -> 149, it was 7 Mbit/s three
times in a row, with 98-100 % at MCS 0-4, and the uplink fell from 120
to 16-65 Mbit/s. Stations that join after the switch are affected too.

Sending the HE MU and sounding modes again after the restart does not
help, nor does stopping and starting the vdevs one at a time, which
leaves the others started on the old channel. Taking every vdev down
and stopping it before any of them starts on the new channel, as a
fresh start of the radio does, helps: after the same two switches,
68.5, 107.8 and 86.2 Mbit/s down (ten runs after switches: median 84;
fresh starts: median 92).

Do that, and keep the restart for a vdev that could not be stopped. A
vdev start leaves the HE MU and sounding modes at the firmware's
defaults, so set them again, as enabling the beacon does on the first
start.

Tested-on: IPQ8074 hw2.0 AHB WLAN.HK.2.9.0.1-02146-QCAHKSWPL_SILICONZ-1

Fixes: d5c65159f289 ("ath11k: driver for Qualcomm IEEE 802.11ax devices")
Assisted-by: Claude:claude-opus-5-5
Signed-off-by: Christopher Griffin <christophergriffin05@gmail.com>
---
Found on the IPQ8074 5 GHz radios of an Arista C-360 and C-260 running
OpenWrt, each with four AP vdevs up, where a management system moves
the radio to another channel by CSA. On the C-260 the downlink after
the same two switches was 245.7, 234.6 and 241.1 Mbit/s with this patch,
against 257.3 from a fresh start.

Not tested: station, mesh or monitor vdevs, and any chip but IPQ8074.
a4146249a333 ("ath11k: skip sending vdev down for channel switch")
dropped a vdev down before the restart because QCA6390 firmware crashed
on it; here the down is followed by a stop and a start, not a restart,
but I could not try that on a QCA6390 or WCN6855.

The investigation and the patch were done with Claude Code
(claude-opus-5-5): it compared ath11k's WMI debug log of a fresh start
with that of a CSA, wrote and ran the variants on the APs, and wrote this
patch and its changelog. The downlink figures are iperf3 runs from a
QCN9072 client; the MCS shares are deltas of the HTT PDEV_TX_RATE stats.
Built with W=1 on ath-next.
---
 drivers/net/wireless/ath/ath11k/mac.c | 41 ++++++++++++++++++++++++++++++++++-
 1 file changed, 40 insertions(+), 1 deletion(-)

diff --git a/drivers/net/wireless/ath/ath11k/mac.c b/drivers/net/wireless/ath/ath11k/mac.c
index 1a237adaa37d..a9b940c2da41 100644
--- a/drivers/net/wireless/ath/ath11k/mac.c
+++ b/drivers/net/wireless/ath/ath11k/mac.c
@@ -7793,6 +7793,7 @@ ath11k_mac_update_vif_chan(struct ath11k *ar,
 	int ret;
 	int i;
 	bool monitor_vif = false;
+	unsigned long stopped = 0;
 
 	lockdep_assert_held(&ar->conf_mutex);
 
@@ -7802,6 +7803,33 @@ ath11k_mac_update_vif_chan(struct ath11k *ar,
 
 	/* TODO: Update ar->rx_channel */
 
+	/* Take every vdev that is up down and stop it before any of them
+	 * starts on the new channel, as a fresh start of the radio does.
+	 * Restarting the vdevs in place, or stopping and starting them one
+	 * at a time while the others stay started, leaves the radio sending
+	 * at far lower rates on the new channel until it is restarted.
+	 */
+	for (i = 0; i < n_vifs && i < BITS_PER_LONG; i++) {
+		arvif = ath11k_vif_to_arvif(vifs[i].vif);
+
+		if (!arvif->is_started || !arvif->is_up)
+			continue;
+
+		ret = ath11k_wmi_vdev_down(ar, arvif->vdev_id);
+		if (ret)
+			ath11k_warn(ab, "failed to down vdev %d: %d\n",
+				    arvif->vdev_id, ret);
+
+		ret = ath11k_mac_vdev_stop(arvif);
+		if (ret) {
+			ath11k_warn(ab, "failed to stop vdev %d: %d\n",
+				    arvif->vdev_id, ret);
+			continue;
+		}
+
+		stopped |= BIT(i);
+	}
+
 	for (i = 0; i < n_vifs; i++) {
 		arvif = ath11k_vif_to_arvif(vifs[i].vif);
 
@@ -7815,7 +7843,18 @@ ath11k_mac_update_vif_chan(struct ath11k *ar,
 		 * Firmware expect vdev_restart only if vdev is up.
 		 * If vdev is down then it expect vdev_stop->vdev_start.
 		 */
-		if (arvif->is_up) {
+		if (i < BITS_PER_LONG && (stopped & BIT(i))) {
+			ret = ath11k_mac_vdev_start(arvif, vifs[i].new_ctx);
+			if (ret) {
+				ath11k_warn(ab, "failed to start vdev %d: %d\n",
+					    arvif->vdev_id, ret);
+				continue;
+			}
+
+			/* as enabling the beacon did for the first start */
+			ath11k_mac_set_he_txbf_conf(arvif);
+		} else if (arvif->is_up) {
+			/* the vdev could not be stopped */
 			ret = ath11k_mac_vdev_restart(arvif, vifs[i].new_ctx);
 			if (ret) {
 				ath11k_warn(ab, "failed to restart vdev %d: %d\n",

---
base-commit: 7b14c8f4a9615579b09cbcdc4515b65ae6124753
change-id: 20261029-ath11k-csa-stop-vdevs-f09daa425259

Best regards,
--  
Christopher Griffin <christophergriffin05@gmail.com>