Country code setting ignored on bootup

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

From 931e5eb59e64804f3a90598b26c56643c43deb35 Mon Sep 17 00:00:00 2001
From: Sven Eckelmann <[email protected]>
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

---
 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(-)

index 18b97420f0d8..1d214eed9ea1 100644
@@ -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,

index 7f8dd47d2333..00aca46505a6 100644
@@ -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) {

index c2b165158225..417f63d9fb6d 100644
@@ -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

ath11k mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/ath11k

Agree that there is a possibility of a race between FW reg event for user regd update and where the ATH11K_REGISTERED flag gets set in the above scenario.

With and without this patch, i also see there are slight chances the default regd (provided by fw during init) can get overwritten by new reg rules from FW in similar scenario you had mentioned.
But, Overwriting of default regd is not allowed.

I get a feeling that we can remove the use of ATH11k_FLAG_REGISTERED for the purpose of checking whether to update default or new_regd.

Since the default reg rules wil be received only once per pdev on FW load, the above flag based check can be replaced with a check to see
if default_regd is already set, so that we can now update the new_regd. Also if the new_regd is set, irrespective of the argument in ath11k_regd_update() we can use the new_regd data then to update the reg info for the phy.

This would avoid both overwriting of the default regd as well as the race you had mentioned and also a possibility of another case where default_regd overwrites new_regd after the previous call to regulatory_set_wiphy_regd_sync_rtnl.

Do you see any issues?

btw, i dont see any possible issues in current usage of ATH11k_FLAG_REGISTERED in
ath11k_debug_pdev_create, ath11k_mac_register and ath11k_mac_allocate
since they sequenced and fall within the same context. Do you face any issues?

Thanks,
Sriram.R

ath11k mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/ath11k

A firmware crash before ATH11K_QMI_EVENT_FW_READY was processed/received. So we might want to recreate the mac/debugfs/... even when it was already created
- at least this is the stuff mentioned in the commit(s) which introduce the ATH11k_FLAG_REGISTERED in these positions.

Kind regards,
Sven

ath11k mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/ath11k

Yeah right, i think it was added to avoid multiple driver registration. The ATH11K_FLAG_REGISTERED currently implies core registration complete where mac registration is one part of it. Agree that checking this flag in reg path is not the right approach to decide if mac is registered and default rules are applied for it.
I can also prepare a patch to avoid the issue you had mentioned and a similar race as below (which i noted with this patch in my setup)and
see if it works fine for all use cases.
Please note below on the 4th line, the Default regd for pdev 0 gets overwritten from US to IN(the country set before ath11k bind), which is not desirable.

[  949.056244] ath11k c000000.wifi1: Default REGD update pdev 0 US
[  949.056493] ath11k c000000.wifi1: Default REGD update pdev 1 US
[  949.056595] ath11k c000000.wifi1: Default REGD update pdev 2 US
[  949.061231] ath11k c000000.wifi1: Default REGD update pdev 0 IN
[  949.063915] ath11k c000000.wifi1: User REGD update pdev 1 IN
[  949.066156] ath11k c000000.wifi1: User REGD update pdev 2 IN

Thanks and Regards,
Sriram.R

ath11k mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/ath11k