[PATCH ath-current] wifi: ath12k: Fix failure to create ath12k_dp_peer on MLO reconnect

[PATCH ath-current] wifi: ath12k: Fix failure to create ath12k_dp_peer on MLO reconnect

[Miaoqing Pan]

In ath12k_mac_op_sta_state(), when transitioning an MLO station from
IEEE80211_STA_NOTEXIST to IEEE80211_STA_NONE, after ath12k_dp_peer_create()
succeeds, on is_sta_assoc_link verification failure, it takes the
'goto exit' path instead of 'goto peer_delete', leaving a stale dp_peer on
the dp_peers_list.

On the next connection attempt, ath12k_dp_peer_create() finds the stale
entry and returns -EEXIST (-17):

    wlP4p1s0: failed to insert STA entry for the AP (error -17)

Fix it by changing 'goto exit' to 'goto free_unassign_link'.

Also, there is another error path in ath12k_mac_op_sta_state() that
jumps to peer_delete after ath12k_mac_assign_link_sta() succeeds, without
undoing the assignment through ath12k_mac_free_unassign_link_sta().
As a result, the link RCU pointer remains assigned and a stale bit is
left in ahsta->links_map, which can trigger the same -EEXIST failure
on subsequent reconnect attempts. Fix this by redirecting that error
path to free_unassign_link instead.

Tested-on: WCN7850 hw2.0 PCI WLAN.HMT.1.1.c7-00108-QCAHMTSWPL_V1.0_V2.0_SILICONZ_UPSTREAM-3

Fixes: 0cd46d1f36ce ("wifi: ath12k: identify assoc link vif in station mode")
Fixes: ee16dcf573d5 ("wifi: ath12k: Define ath12k_dp_peer structure & APIs for create & delete")
Signed-off-by: Miaoqing Pan <miaoqing.pan@oss.qualcomm.com>
---
 drivers/net/wireless/ath/ath12k/mac.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/drivers/net/wireless/ath/ath12k/mac.c b/drivers/net/wireless/ath/ath12k/mac.c
index d4116ba0da08..57861003c15b 100644
--- a/drivers/net/wireless/ath/ath12k/mac.c
+++ b/drivers/net/wireless/ath/ath12k/mac.c
@@ -7845,7 +7845,7 @@ int ath12k_mac_op_sta_state(struct ieee80211_hw *hw,
 				ath12k_hw_warn(ah, "failed to verify assoc link setting with link id %u\n",
 					       link_id);
 				ret = -EINVAL;
-				goto exit;
+				goto free_unassign_link;
 			}
 
 			arsta->is_assoc_link = true;
@@ -7903,7 +7903,7 @@ int ath12k_mac_op_sta_state(struct ieee80211_hw *hw,
 
 			if (old_state == IEEE80211_STA_NOTEXIST &&
 			    new_state == IEEE80211_STA_NONE)
-				goto peer_delete;
+				goto free_unassign_link;
 			else
 				goto exit;
 		}
@@ -7943,6 +7943,8 @@ int ath12k_mac_op_sta_state(struct ieee80211_hw *hw,
 	ret = 0;
 	goto exit;
 
+free_unassign_link:
+	ath12k_mac_free_unassign_link_sta(ah, ahsta, link_id);
 peer_delete:
 	ath12k_dp_peer_delete(&ah->dp_hw, sta->addr, sta);
 ml_peer_id_clear:

base-commit: 91b552fe898e2c46edb363390211fac47b329308
-- 
2.34.1

Re: [PATCH ath-current] wifi: ath12k: Fix failure to create ath12k_dp_peer on MLO reconnect

[Jeff Johnson]

On 9/15/2026 1:00 AM, Miaoqing Pan wrote:
> In ath12k_mac_op_sta_state(), when transitioning an MLO station from
> IEEE80211_STA_NOTEXIST to IEEE80211_STA_NONE, after ath12k_dp_peer_create()
> succeeds, on is_sta_assoc_link verification failure, it takes the
> 'goto exit' path instead of 'goto peer_delete', leaving a stale dp_peer on
> the dp_peers_list.
> 
> On the next connection attempt, ath12k_dp_peer_create() finds the stale
> entry and returns -EEXIST (-17):
> 
>     wlP4p1s0: failed to insert STA entry for the AP (error -17)
> 
> Fix it by changing 'goto exit' to 'goto free_unassign_link'.
> 
> Also, there is another error path in ath12k_mac_op_sta_state() that
> jumps to peer_delete after ath12k_mac_assign_link_sta() succeeds, without
> undoing the assignment through ath12k_mac_free_unassign_link_sta().
> As a result, the link RCU pointer remains assigned and a stale bit is
> left in ahsta->links_map, which can trigger the same -EEXIST failure
> on subsequent reconnect attempts. Fix this by redirecting that error
> path to free_unassign_link instead.
> 
> Tested-on: WCN7850 hw2.0 PCI WLAN.HMT.1.1.c7-00108-QCAHMTSWPL_V1.0_V2.0_SILICONZ_UPSTREAM-3
> 
> Fixes: 0cd46d1f36ce ("wifi: ath12k: identify assoc link vif in station mode")
> Fixes: ee16dcf573d5 ("wifi: ath12k: Define ath12k_dp_peer structure & APIs for create & delete")
> Signed-off-by: Miaoqing Pan <miaoqing.pan@oss.qualcomm.com>
> ---

Baochen has a more holistic approach to this problem in the works, so dropping
this patch.

/jeff

Re: [PATCH ath-current] wifi: ath12k: Fix failure to create ath12k_dp_peer on MLO reconnect

[Miaoqing Pan]

On 9/29/2026 9:42 PM, Jeff Johnson wrote:
> On 9/15/2026 1:00 AM, Miaoqing Pan wrote:
>> In ath12k_mac_op_sta_state(), when transitioning an MLO station from
>> IEEE80211_STA_NOTEXIST to IEEE80211_STA_NONE, after ath12k_dp_peer_create()
>> succeeds, on is_sta_assoc_link verification failure, it takes the
>> 'goto exit' path instead of 'goto peer_delete', leaving a stale dp_peer on
>> the dp_peers_list.
>>
>> On the next connection attempt, ath12k_dp_peer_create() finds the stale
>> entry and returns -EEXIST (-17):
>>
>>      wlP4p1s0: failed to insert STA entry for the AP (error -17)
>>
>> Fix it by changing 'goto exit' to 'goto free_unassign_link'.
>>
>> Also, there is another error path in ath12k_mac_op_sta_state() that
>> jumps to peer_delete after ath12k_mac_assign_link_sta() succeeds, without
>> undoing the assignment through ath12k_mac_free_unassign_link_sta().
>> As a result, the link RCU pointer remains assigned and a stale bit is
>> left in ahsta->links_map, which can trigger the same -EEXIST failure
>> on subsequent reconnect attempts. Fix this by redirecting that error
>> path to free_unassign_link instead.
>>
>> Tested-on: WCN7850 hw2.0 PCI WLAN.HMT.1.1.c7-00108-QCAHMTSWPL_V1.0_V2.0_SILICONZ_UPSTREAM-3
>>
>> Fixes: 0cd46d1f36ce ("wifi: ath12k: identify assoc link vif in station mode")
>> Fixes: ee16dcf573d5 ("wifi: ath12k: Define ath12k_dp_peer structure & APIs for create & delete")
>> Signed-off-by: Miaoqing Pan <miaoqing.pan@oss.qualcomm.com>
>> ---
> 
> Baochen has a more holistic approach to this problem in the works, so dropping
> this patch.
> 
> /jeff
> 

I believe the patch that needs to be dropped is this one: 
https://lore.kernel.org/linux-wireless/20260815014329.3748059-1-miaoqing.pan@oss.qualcomm.com/.

However, the patch itself is still required. @Baochen, could you please 
help confirm?

Re: [PATCH ath-current] wifi: ath12k: Fix failure to create ath12k_dp_peer on MLO reconnect

[Baochen Qiang]

On 10/15/2026 12:09 PM, Miaoqing Pan wrote:
> 
> 
> On 9/29/2026 9:42 PM, Jeff Johnson wrote:
>> On 9/15/2026 1:00 AM, Miaoqing Pan wrote:
>>> In ath12k_mac_op_sta_state(), when transitioning an MLO station from
>>> IEEE80211_STA_NOTEXIST to IEEE80211_STA_NONE, after ath12k_dp_peer_create()
>>> succeeds, on is_sta_assoc_link verification failure, it takes the
>>> 'goto exit' path instead of 'goto peer_delete', leaving a stale dp_peer on
>>> the dp_peers_list.
>>>
>>> On the next connection attempt, ath12k_dp_peer_create() finds the stale
>>> entry and returns -EEXIST (-17):
>>>
>>>      wlP4p1s0: failed to insert STA entry for the AP (error -17)
>>>
>>> Fix it by changing 'goto exit' to 'goto free_unassign_link'.
>>>
>>> Also, there is another error path in ath12k_mac_op_sta_state() that
>>> jumps to peer_delete after ath12k_mac_assign_link_sta() succeeds, without
>>> undoing the assignment through ath12k_mac_free_unassign_link_sta().
>>> As a result, the link RCU pointer remains assigned and a stale bit is
>>> left in ahsta->links_map, which can trigger the same -EEXIST failure
>>> on subsequent reconnect attempts. Fix this by redirecting that error
>>> path to free_unassign_link instead.
>>>
>>> Tested-on: WCN7850 hw2.0 PCI WLAN.HMT.1.1.c7-00108-
>>> QCAHMTSWPL_V1.0_V2.0_SILICONZ_UPSTREAM-3
>>>
>>> Fixes: 0cd46d1f36ce ("wifi: ath12k: identify assoc link vif in station mode")
>>> Fixes: ee16dcf573d5 ("wifi: ath12k: Define ath12k_dp_peer structure & APIs for create &
>>> delete")
>>> Signed-off-by: Miaoqing Pan <miaoqing.pan@oss.qualcomm.com>
>>> ---
>>
>> Baochen has a more holistic approach to this problem in the works, so dropping
>> this patch.
>>
>> /jeff
>>
> 
> I believe the patch that needs to be dropped is this one: https://lore.kernel.org/linux-
> wireless/20260815014329.3748059-1-miaoqing.pan@oss.qualcomm.com/.
> 
> However, the patch itself is still required. @Baochen, could you please help confirm?

Yeah, exactly.

Jeff, I was wrong in the internal teams communication. Miaoqing is right on this topic.