>From 931e5eb59e64804f3a90598b26c56643c43deb35 Mon Sep 17 00:00:00 2001
From: Sven Eckelmann <sven@narfation.org>
Date: Fri, 30 Oct 2020 12:02:21 +0100
Subject: [RFC PATCH] ath11k: Accept new regdomain during initialization
The driver is registering as iee80211_hw with its OPs and is then able to
be called by the upper layer. This for example happens early in the phase
when the correct regulary domain should be set. But the regulary domain
will only be accepted when the ATH11K_FLAG_REGISTERED flag was set after
the ATH11K_QMI_EVENT_FW_READY was processed. So it can easily happen that
the regularly domain is not correctly processed when
ATH11K_QMI_EVENT_FW_READY isn't handled immediately:
$ iw reg set CA
$ iw reg get|grep country
country CA: DFS-FCC
country CA: DFS-FCC
country CA: DFS-FCC
$ echo "c000000.wifi1" > /sys/bus/platform/drivers/ath11k/unbind
$ echo "c000000.wifi1" > /sys/bus/platform/drivers/ath11k/bind
$ iw reg get|grep country
country CA: DFS-FCC
country US: DFS-FCC
country US: DFS-FCC
It is therefore essential to accept the regulatory changes without having
seen the ATH11K_QMI_EVENT_FW_READY. And since there are also potentially
more problems in ath11k_debug_pdev_create, ath11k_mac_register and
ath11k_mac_allocate with their use of ATH11K_FLAG_REGISTERED, it is better
to move the ATH11K_QMI_EVENT_FW_READY. to a new flag.
Tested with WLAN.HK.2.4.0.1.r1-00019-QCAHKSWPL_SILICONZ-1
Fixes: d5c65159f289 ("ath11k: driver for Qualcomm IEEE 802.11ax devices")
Signed-off-by: Sven Eckelmann <sven@narfation.org>
---
drivers/net/wireless/ath/ath11k/core.h | 2 +-
drivers/net/wireless/ath/ath11k/mac.c | 2 ++
drivers/net/wireless/ath/ath11k/qmi.c | 4 ++--
3 files changed, 5 insertions(+), 3 deletions(-)
@@ -170,7 +170,7 @@ enum ath11k_scan_state {
enum ath11k_dev_flags {
ATH11K_CAC_RUNNING,
- ATH11K_FLAG_CORE_REGISTERED,
+ ATH11K_FLAG_FW_READY,
ATH11K_FLAG_CRASH_FLUSH,
ATH11K_FLAG_RAW_MODE,
ATH11K_FLAG_HW_CRYPTO_DISABLED,
@@ -6280,6 +6280,8 @@ static int __ath11k_mac_register(struct ath11k *ar)
*/
ar->hw->wiphy->interface_modes &= ~BIT(NL80211_IFTYPE_MONITOR);
+ set_bit(ATH11K_FLAG_REGISTERED, &ab->dev_flags);
+
/* Apply the regd received during initialization */
ret = ath11k_regd_update(ar, true);
if (ret) {
@@ -2612,7 +2612,7 @@ static void ath11k_qmi_driver_event_work(struct work_struct *work)
ath11k_qmi_event_load_bdf(qmi);
break;
case ATH11K_QMI_EVENT_FW_READY:
- if (test_bit(ATH11K_FLAG_REGISTERED, &ab->dev_flags)) {
+ if (test_bit(ATH11K_FLAG_FW_READY, &ab->dev_flags)) {
ath11k_hal_dump_srng_stats(ab);
queue_work(ab->workqueue, &ab->restart_work);
break;
@@ -2620,7 +2620,7 @@ static void ath11k_qmi_driver_event_work(struct work_struct *work)
ath11k_core_qmi_firmware_ready(ab);
ab->qmi.cal_done = 1;
- set_bit(ATH11K_FLAG_REGISTERED, &ab->dev_flags);
+ set_bit(ATH11K_FLAG_FW_READY, &ab->dev_flags);
break;
case ATH11K_QMI_EVENT_COLD_BOOT_CAL_DONE:
--
2.28.0
Hi, just tested here with HK01 based board. The firwmare WLAN.HK.2.1.0.1-01228-QCAHKSWPL_SILICONZ-1 doesn't show this problematic behavior all the time but 2.4.0.1.r1-00026-QCAHKSWPL_SILICONZ-2 or WLAN.HK. 2.4.0.1.r1-00019-QCAHKSWPL_SILICONZ-1 does. With 2.1.0.1 I could basically do following: iw reg set CA echo "c000000.wifi1" > /sys/bus/platform/drivers/ath11k/unbind echo "c000000.wifi1" > /sys/bus/platform/drivers/ath11k/bind and the global and self managed country codes were still all CA (in many tests). But with 2.4.0.1, they will end up with CA as global and US as self managed. I can see that the country code change is correctly send by cfg80211 when the HW is initializes but the 2.4.0.1 firmware just ignores it. I can see following send command by ath11k in ath11k_wmi_send_init_country_cmd: 00000000: 38 00 61 02 00 00 00 00 00 00 00 00 43 41 00 00 and I also receive a CA answer for that. But the response from the firmware comes in when the ATH11K_FLAG_REGISTERED bit is still not set. Thus it is rejected by ath11k_reg_chan_list_event for the regd_update_work. This seems to be a race between receiving the "ATH11K_QMI_EVENT_FW_READY" event and the ieee80211_register_hw() in __ath11k_mac_register. The ieee80211_register_hw will make the hw responsible for handling the regd updates but they will not be correctly processed until the ATH11K_QMI_EVENT_FW_READY was handled by the firmware. And if the regd change was processed after ath11k_regd_update was called in __ath11k_mac_register (but before the ATH11K_QMI_EVENT_FW_READY was processed) then we will just loose the regd change. So I am a little bit baffled why ATH11K_FLAG_REGISTERED is bound to something which is not related to the IEEE802211 registration code. Sounds to me like somebody wanted a FW_READY flag but misused the ATH11K_FLAG_REGISTERED flag for this purpose. Btw. there are similar problems with the use of ATH11K_FLAG_REGISTERED in context of ath11k_debug_pdev_create, ath11k_mac_register and ath11k_mac_allocate. Maybe something like the attached patch could be used to improve this situation. Kind regards, Sven