Re: [PATCH v3 6/6] net: sh_eth: use NAPI

["Shimoda, Yoshihiro" <yoshihiro.shimoda.uh@renesas.com>]

2012/05/16 3:29, Francois Romieu wrote:
> David Miller <davem@davemloft.net> :
>> From: "Shimoda, Yoshihiro" <yoshihiro.shimoda.uh@renesas.com>
> [...]
>>> I will modify the code as the following. Is it correct?
>>>
>>> 	if (txfree_num) {
>>> 		netif_tx_lock(ndev);
>>> 		if (netif_queue_stopped(ndev))
>>> 			netif_wake_queue(ndev);
>>> 		netif_tx_unlock(ndev);
>>> 	}
>>
>> Yes, and then you don't need that private lock in the start_xmit()
>> method at all, since that method runs with the tx_lock held.
> 
> I agree that the lock should go but:
> 1. something must be done to prevent sh_eth_txfree() being called
>    at the same time from the xmit and poll handlers
> 2. while netif_tx locking above provides an implicit memory barrier,
>    there won't be one in sh_eth_start_xmit once the lock is removed.
>    mdp->dirty_tx changes may thus go unnoticed in sh_eth_start_xmit
> 

Thank you for the review.

I checked tg3.c, and I found a patch to fix the race condition:
commit ID 1b2a72050 : [TG3]: Fix tx race condition

I will check the patch, and I will write such a patch for the sh_eth
driver later.

Best regards,
Yoshihiro Shimoda