| Message ID | 20250605-mt7996-sleep-while-atomic-v1-5-d46d15f9203c@kernel.org |
|---|---|
| State | New |
| Headers | show |
| Series | mt76: mt7996: Fix 'sleep while atomic' in mt7996_mac_sta_rc_work() | expand |
> On 6/5/25 04:14, Lorenzo Bianconi wrote: > > Since mt7996_mcu_add_rate_ctrl() and mt7996_mcu_set_fixed_field() can't > > run in atomic context, move RCU critical section in > > mt7996_mcu_add_rate_ctrl() and mt7996_mcu_set_fixed_field(). This patch > > fixes a 'sleep while atomic' issue in mt7996_mac_sta_rc_work(). > > > > Fixes: 0762bdd30279 ("wifi: mt76: mt7996: rework mt7996_mac_sta_rc_work to support MLO") > > Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org> > > Thanks for the quick fix on this, it appears to solve the problems we > saw related to this. ack, thx. Can you please add your 'tested-by' tag in this case? Regards, Lorenzo > > Thanks, > Ben > > -- > Ben Greear <greearb@candelatech.com> > Candela Technologies Inc http://www.candelatech.com > >
On 6/5/25 09:57, Lorenzo Bianconi wrote: >> On 6/5/25 04:14, Lorenzo Bianconi wrote: >>> Since mt7996_mcu_add_rate_ctrl() and mt7996_mcu_set_fixed_field() can't >>> run in atomic context, move RCU critical section in >>> mt7996_mcu_add_rate_ctrl() and mt7996_mcu_set_fixed_field(). This patch >>> fixes a 'sleep while atomic' issue in mt7996_mac_sta_rc_work(). >>> >>> Fixes: 0762bdd30279 ("wifi: mt76: mt7996: rework mt7996_mac_sta_rc_work to support MLO") >>> Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org> >> >> Thanks for the quick fix on this, it appears to solve the problems we >> saw related to this. > > ack, thx. Can you please add your 'tested-by' tag in this case? Tested-by: Ben Greear <greearb@candelatech.com> Thanks, Ben
On 6/5/25 10:15, Lorenzo Bianconi wrote: > On Jun 05, Ben Greear wrote: >> On 6/5/25 09:57, Lorenzo Bianconi wrote: >>>> On 6/5/25 04:14, Lorenzo Bianconi wrote: >>>>> Since mt7996_mcu_add_rate_ctrl() and mt7996_mcu_set_fixed_field() can't >>>>> run in atomic context, move RCU critical section in >>>>> mt7996_mcu_add_rate_ctrl() and mt7996_mcu_set_fixed_field(). This patch >>>>> fixes a 'sleep while atomic' issue in mt7996_mac_sta_rc_work(). >>>>> >>>>> Fixes: 0762bdd30279 ("wifi: mt76: mt7996: rework mt7996_mac_sta_rc_work to support MLO") >>>>> Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org> >>>> >>>> Thanks for the quick fix on this, it appears to solve the problems we >>>> saw related to this. >>> >>> ack, thx. Can you please add your 'tested-by' tag in this case? >> >> Tested-by: Ben Greear <greearb@candelatech.com> > > I am assuming this is for the full series ;) Yes, and to be clear, I had to re-work the patches a bit to get them into my tree, but that was just because I have a lot of mtk upstream patches and my own hacks in this code. I still think the tested-by is valid though. Thanks, Ben
diff --git a/drivers/net/wireless/mediatek/mt76/mt7996/mac.c b/drivers/net/wireless/mediatek/mt76/mt7996/mac.c index 5561b1098e3fb0d25551fde3cfdb3057217395fd..445fe149ac0d7e194497a2dc884a599b9546c9ef 100644 --- a/drivers/net/wireless/mediatek/mt76/mt7996/mac.c +++ b/drivers/net/wireless/mediatek/mt76/mt7996/mac.c @@ -2353,20 +2353,12 @@ void mt7996_mac_update_stats(struct mt7996_phy *phy) void mt7996_mac_sta_rc_work(struct work_struct *work) { struct mt7996_dev *dev = container_of(work, struct mt7996_dev, rc_work); - struct ieee80211_bss_conf *link_conf; - struct ieee80211_link_sta *link_sta; struct mt7996_sta_link *msta_link; - struct mt7996_vif_link *link; - struct mt76_vif_link *mlink; - struct ieee80211_sta *sta; struct ieee80211_vif *vif; - struct mt7996_sta *msta; struct mt7996_vif *mvif; LIST_HEAD(list); u32 changed; - u8 link_id; - rcu_read_lock(); spin_lock_bh(&dev->mt76.sta_poll_lock); list_splice_init(&dev->sta_rc_list, &list); @@ -2377,44 +2369,28 @@ void mt7996_mac_sta_rc_work(struct work_struct *work) changed = msta_link->changed; msta_link->changed = 0; - - sta = wcid_to_sta(&msta_link->wcid); - link_id = msta_link->wcid.link_id; - msta = msta_link->sta; - mvif = msta->vif; - vif = container_of((void *)mvif, struct ieee80211_vif, drv_priv); - - mlink = rcu_dereference(mvif->mt76.link[link_id]); - if (!mlink) - continue; - - link_sta = rcu_dereference(sta->link[link_id]); - if (!link_sta) - continue; - - link_conf = rcu_dereference(vif->link_conf[link_id]); - if (!link_conf) - continue; + mvif = msta_link->sta->vif; + vif = container_of((void *)mvif, struct ieee80211_vif, + drv_priv); spin_unlock_bh(&dev->mt76.sta_poll_lock); - link = (struct mt7996_vif_link *)mlink; - if (changed & (IEEE80211_RC_SUPP_RATES_CHANGED | IEEE80211_RC_NSS_CHANGED | IEEE80211_RC_BW_CHANGED)) mt7996_mcu_add_rate_ctrl(dev, msta_link->sta, vif, - link_id, true); + msta_link->wcid.link_id, + true); if (changed & IEEE80211_RC_SMPS_CHANGED) - mt7996_mcu_set_fixed_field(dev, msta, NULL, link_id, + mt7996_mcu_set_fixed_field(dev, msta_link->sta, NULL, + msta_link->wcid.link_id, RATE_PARAM_MMPS_UPDATE); spin_lock_bh(&dev->mt76.sta_poll_lock); } spin_unlock_bh(&dev->mt76.sta_poll_lock); - rcu_read_unlock(); } void mt7996_mac_work(struct work_struct *work)
Since mt7996_mcu_add_rate_ctrl() and mt7996_mcu_set_fixed_field() can't run in atomic context, move RCU critical section in mt7996_mcu_add_rate_ctrl() and mt7996_mcu_set_fixed_field(). This patch fixes a 'sleep while atomic' issue in mt7996_mac_sta_rc_work(). Fixes: 0762bdd30279 ("wifi: mt76: mt7996: rework mt7996_mac_sta_rc_work to support MLO") Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org> --- drivers/net/wireless/mediatek/mt76/mt7996/mac.c | 38 +++++-------------------- 1 file changed, 7 insertions(+), 31 deletions(-)