Received-SPF: pass (google.com: domain of linux-wireless-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   venkatch@gmail.com
To:     johannes@sipsolutions.net
Cc:     linux-wireless@vger.kernel.org, victorg@ti.com,
        Venkat Chimata <venkata@shasta.cloud>
Subject: [PATCH] wifi: mac80211: Fix performance issue with mutex_lock
Date:   Wed, 14 Sep 2022 16:39:34 +0530
Message-Id: <20220914110934.617430-1-venkatch@gmail.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

From: Venkat Chimata <venkata@shasta.cloud>

OpenWiFi's ucentral agent periodically (typically 120 seconds)
issues nl80211 call to get associated client list from the
WLAN driver. Somehow this operation was causing tx/rx delays
sometimes and the video calls on connected clients are experiencing
jitter. The associated client list was protected by
a mutex lock. I saw that ieee80211_check_fast_xmit_all uses
rcu_read_lock and rcu_read_unlock to iterate through sta_list.
I took it as a refernce and changed the lock to rcu_read lock
from mutex.
Also saw this this comment just above sta_mutex declaration.
        /* Station data */
        /*
         * The mutex only protects the list, hash table and
         * counter, reads are done with RCU.
         */
        struct mutex sta_mtx;

Hence tried changing mutex_lock(/unlock) in ieee80211_dump_station
to rcu_read_lock(/unlock) and it resolved the jitter issue in the
video calls.

Tests:
We had this issue show up consistently and the patch fixed the issue.
We spent a good part of 2 weeks following up on this and with this
fix, the video calls are smooth.

Also tested if this could cause any crashes with below mentioned
process.

1. Connect 3 clients
2. A script running dumping clients in an infinite loop
3. Continuously disconnect / connect one client to see if
   there is any crash. No crash was observed.

Signed-off-by: Venkat Chimata <venkata@shasta.cloud>
---
 net/mac80211/cfg.c      | 6 ++----
 net/mac80211/sta_info.c | 3 +--
 2 files changed, 3 insertions(+), 6 deletions(-)

diff --git a/net/mac80211/cfg.c b/net/mac80211/cfg.c
index 6a8350d..0b526b5 100644
--- a/net/mac80211/cfg.c
+++ b/net/mac80211/cfg.c
@@ -794,16 +794,14 @@ static int ieee80211_dump_station(struct wiphy *wiphy, struct net_device *dev,
 	struct sta_info *sta;
 	int ret = -ENOENT;
 
-	mutex_lock(&local->sta_mtx);
-
+	rcu_read_lock();
 	sta = sta_info_get_by_idx(sdata, idx);
 	if (sta) {
 		ret = 0;
 		memcpy(mac, sta->sta.addr, ETH_ALEN);
 		sta_set_sinfo(sta, sinfo, true);
 	}
-
-	mutex_unlock(&local->sta_mtx);
+	rcu_read_unlock();
 
 	return ret;
 }
diff --git a/net/mac80211/sta_info.c b/net/mac80211/sta_info.c
index 550a610..af6fa75 100644
--- a/net/mac80211/sta_info.c
+++ b/net/mac80211/sta_info.c
@@ -231,8 +231,7 @@ struct sta_info *sta_info_get_by_idx(struct ieee80211_sub_if_data *sdata,
 	struct sta_info *sta;
 	int i = 0;
 
-	list_for_each_entry_rcu(sta, &local->sta_list, list,
-				lockdep_is_held(&local->sta_mtx)) {
+	list_for_each_entry_rcu(sta, &local->sta_list, list) {
 		if (sdata != sta->sdata)
 			continue;
 		if (i < idx) {
-- 
2.34.1