Am 22.12.23 um 09:59 schrieb Ping-Ke Shih:
On Fri, 2023-12-22 at 09:25 +0100, Martin Kaistra wrote:
Am 22.12.23 um 09:05 schrieb Martin Kaistra:
Am 22.12.23 um 02:45 schrieb Ping-Ke Shih:
On Fri, Dec 22, 2023 at 12:45 AM Martin Kaistra
<martin.kaistra@xxxxxxxxxxxxx> wrote:
Check first whether priv->vifs[0] exists and is of type STATION, then go
to priv->vifs[1]. Make sure to call refresh_rate_mask for both
interfaces.
Signed-off-by: Martin Kaistra <martin.kaistra@xxxxxxxxxxxxx>
---
.../wireless/realtek/rtl8xxxu/rtl8xxxu_core.c | 19 +++++++++++--------
1 file changed, 11 insertions(+), 8 deletions(-)
diff --git a/drivers/net/wireless/realtek/rtl8xxxu/rtl8xxxu_core.c
b/drivers/net/wireless/realtek/rtl8xxxu/rtl8xxxu_core.c
index c5b71892369c9..fd0108668bcda 100644
--- a/drivers/net/wireless/realtek/rtl8xxxu/rtl8xxxu_core.c
+++ b/drivers/net/wireless/realtek/rtl8xxxu/rtl8xxxu_core.c
@@ -7200,11 +7200,15 @@ static void rtl8xxxu_watchdog_callback(struct
work_struct *work)
{
struct ieee80211_vif *vif;
struct rtl8xxxu_priv *priv;
+ int i;
priv = container_of(work, struct rtl8xxxu_priv, ra_watchdog.work);
- vif = priv->vif;
+ for (i = 0; i < ARRAY_SIZE(priv->vifs); i++) {
+ vif = priv->vifs[i];
+
+ if (!vif || vif->type != NL80211_IFTYPE_STATION)
+ continue;
Currently, this loop becomes to get RSSI and update rate mask, but only for
station mode. That means we don't update them for peer stations in AP mode.
Maybe, we need ieee80211_iterate_stations_atomic() for all stations, but
ieee80211_ave_rssi() is only for 'vif', so we need to add a driver level
RSSI for all stations (macid). Also, we need to extend priv->rssi_level for
all macid as well.
I suppose _default_ value can work to stations in AP mode, so you can decide
if you will defer this support temporarily.
(Sorry, I don't dig rtl8xxxu very deeply, so I'm not able to tell you all
things in one go.)
It probably makes sense to fix this, however if it's ok for you, I would like to
do it separatly from this series.
Ok to me. :-)
- if (vif && vif->type == NL80211_IFTYPE_STATION) {
int signal;
struct ieee80211_sta *sta;
@@ -7215,22 +7219,21 @@ static void rtl8xxxu_watchdog_callback(struct
work_struct *work)
dev_dbg(dev, "%s: no sta found\n", __func__);
rcu_read_unlock();
- goto out;
+ continue;
}
rcu_read_unlock();
signal = ieee80211_ave_rssi(vif);
- priv->fops->report_rssi(priv, 0,
+ priv->fops->report_rssi(priv, rtl8xxxu_get_macid(priv, sta),
rtl8xxxu_signal_to_snr(signal));
- if (priv->fops->set_crystal_cap)
- rtl8xxxu_track_cfo(priv);
-
rtl8xxxu_refresh_rate_mask(priv, signal, sta, false);
It seems like 'sta' isn't protected by RCU read lock.
(an existing bug, not introduced by this patch)
I will add a patch which moves the rcu_read_unlock() to fix this.
Actually, looking at the code again, rtl8xxxu_refresh_rate_mask() seems to
already contain calls to rcu_read_lock(). Just the call to
rtl8xxxu_get_macid(priv, sta) in there might need additional protection.
We must use RCU lock to protect 'sta' from dereference to whole access, but
it looks like
rtl8xxxu_watchdog_callback()
rcu_read_lock();
sta = ...
rcu_read_unlock() // after this point, use 'sta' is unsafe..
rtl8xxxu_refresh_rate_mask(sta)
rcu_read_lock();
rate_bitmap = sta->...
rcu_read_unlock();
If it is not an easy fix, does it make sense to you to do this also separately
from this series?
