[PATCH] wifi: mt76: mt7996: Fix possible token leak in mt7996_tx_prepare_skb()

Dylan Eskew dylan.eskew at candelatech.com
Wed Jun 3 08:43:42 PDT 2026


Hi Lore,

>> Hi Lore,
> Hi Dylan,
>
>> We have been seeing the token memory leak in our custom kernel. After
>> pulling your patch in, we are still getting the leak (validated with
>> kmemleak). How did you figure out where this potential leak was? I want to
>> determine if we are leaking because of our changes or if there's more areas
>> for token leakage.
> Can you please try to run kmemleak on Felix's tree to check if there are any
> leftover leaks not fixed yet?

Ran kmemleak with Felix's tree. I brought up only a few stations, no 
traffic run yet and kmemleak flagged a possible leak, same allocation 
location we've seen flagged in our custom kernel.

kmemleak trace:
```
unreferenced object 0xffff88811e6ca380 (size 128):
   comm "mt76-tx phy0", pid 1164, jiffies 4295044455
   hex dump (first 32 bytes):
     44 00 00 1a 0a a0 4c ae 0d 00 00 00 02 78 00 30 D.....L......x.0
     00 00 00 00 01 00 00 00 14 00 27 12 00 00 00 08 ..........'.....
   backtrace (crc fba2c5a3):
     __kmalloc_noprof+0x38e/0x480
     mt76_dma_tx_queue_skb+0x522/0x890 [mt76]
     __mt76_tx_queue_skb+0x3e/0xa0 [mt76]
     mt76_txq_schedule_pending_wcid+0x12b/0x200 [mt76]
     mt76_txq_schedule_pending+0x122/0x1b0 [mt76]
     mt76_tx_worker_run+0x1b/0xc0 [mt76]
     __mt76_worker_fn+0x49/0x90 [mt76]
     kthread+0xdc/0x110
     ret_from_fork+0x190/0x280
     ret_from_fork_asm+0x11/0x20
```

In dma.c, line 19 is the where the kmemleak trace points:
```
10 static struct mt76_txwi_cache *
11 mt76_alloc_txwi(struct mt76_dev *dev)
12 {
13         struct mt76_txwi_cache *t;
14         dma_addr_t addr;
15         u8 *txwi;
16         int size;
17
18         size = L1_CACHE_ALIGN(dev->drv->txwi_size + sizeof(*t));
19         txwi = kzalloc(size, GFP_ATOMIC);
20         if (!txwi)
21                 return NULL;
22
23         addr = dma_map_single(dev->dma_dev, txwi, dev->drv->txwi_size,
24                               DMA_TO_DEVICE);
25         if (unlikely(dma_mapping_error(dev->dma_dev, addr))) {
26                 kfree(txwi);
27                 return NULL;
28         }
29
30         t = (struct mt76_txwi_cache *)(txwi + dev->drv->txwi_size);
31         t->dma_addr = addr;
32
33         return t;
34 }
```

Let me know if you need any other information.

-- Dylan

>
> Regards,
> Lorenzo
>
>> -- Dylan
>>
>> On 5/31/26 2:10 AM, Lorenzo Bianconi wrote:
>>> If link_conf or link_sta lookup fails in mt7996_tx_prepare_skb routine,
>>> mt7996 driver leaks an already allocated tx token. Fix the issue
>>> releasing the token in case of error.
>>>
>>> Fixes: 7ef0c7ad735b0 ("wifi: mt76: mt7996: Implement MLD address translation for EAPOL")
>>> Signed-off-by: Lorenzo Bianconi <lorenzo at kernel.org>
>>> ---
>>>    drivers/net/wireless/mediatek/mt76/mt7996/mac.c | 8 ++++++--
>>>    drivers/net/wireless/mediatek/mt76/tx.c         | 2 +-
>>>    2 files changed, 7 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/drivers/net/wireless/mediatek/mt76/mt7996/mac.c b/drivers/net/wireless/mediatek/mt76/mt7996/mac.c
>>> index c98446057282..8c56344d211b 100644
>>> --- a/drivers/net/wireless/mediatek/mt76/mt7996/mac.c
>>> +++ b/drivers/net/wireless/mediatek/mt76/mt7996/mac.c
>>> @@ -1067,11 +1067,11 @@ int mt7996_tx_prepare_skb(struct mt76_dev *mdev, void *txwi_ptr,
>>>    		link_conf = rcu_dereference(vif->link_conf[wcid->link_id]);
>>>    		if (!link_conf)
>>> -			return -EINVAL;
>>> +			goto error_relase_token;
>>>    		link_sta = rcu_dereference(sta->link[wcid->link_id]);
>>>    		if (!link_sta)
>>> -			return -EINVAL;
>>> +			goto error_relase_token;
>>>    		dma_sync_single_for_cpu(mdev->dma_dev, tx_info->buf[1].addr,
>>>    					tx_info->buf[1].len, DMA_TO_DEVICE);
>>> @@ -1176,6 +1176,10 @@ int mt7996_tx_prepare_skb(struct mt76_dev *mdev, void *txwi_ptr,
>>>    	tx_info->nbuf = MT_CT_DMA_BUF_NUM;
>>>    	return 0;
>>> +
>>> +error_relase_token:
>>> +	mt76_token_release(mdev, id, NULL);
>>> +	return -EINVAL;
>>>    }
>>>    u32 mt7996_wed_init_buf(void *ptr, dma_addr_t phys, int token_id)
>>> diff --git a/drivers/net/wireless/mediatek/mt76/tx.c b/drivers/net/wireless/mediatek/mt76/tx.c
>>> index 22f9690634c9..f96d9c471853 100644
>>> --- a/drivers/net/wireless/mediatek/mt76/tx.c
>>> +++ b/drivers/net/wireless/mediatek/mt76/tx.c
>>> @@ -933,7 +933,7 @@ mt76_token_release(struct mt76_dev *dev, int token, bool *wake)
>>>    #endif
>>>    	}
>>> -	if (dev->token_count < dev->token_size - MT76_TOKEN_FREE_THR &&
>>> +	if (wake && dev->token_count < dev->token_size - MT76_TOKEN_FREE_THR &&
>>>    	    dev->phy.q_tx[0]->blocked)
>>>    		*wake = true;
>>>
>>> ---
>>> base-commit: 4913f44167cf35a9536e9eec7352e15b2de0c573
>>> change-id: 20260531-mt7996_tx_prepare_skb-token-leack-82e240d8c66f
>>>
>>> Best regards,



More information about the linux-arm-kernel mailing list