netdev - Re: [PATCH resend 1/3] Revert "net: fec: fix missing napi

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Sun, 28 Apr 2013 12:31:56 +0800
From:	Frank Li <lznuaa@...il.com>
To:	Francois Romieu <romieu@...zoreil.com>
Cc:	Robert Schwebel <r.schwebel@...gutronix.de>,
	David Miller <davem@...emloft.net>,
	Lucas Stach <l.stach@...gutronix.de>,
	"netdev@...r.kernel.org" <netdev@...r.kernel.org>,
	"Frank.Li@...escale.com" <Frank.Li@...escale.com>,
	Fabio Estevam <festevam@...il.com>,
	Shawn Guo <shawn.guo@...aro.org>
Subject: Re: [PATCH resend 1/3] Revert "net: fec: fix missing napi_disable call"

2013/4/28 Frank Li <lznuaa@...il.com>:
> 2013/4/28 Francois Romieu <romieu@...zoreil.com>:
>> Frank Li <lznuaa@...il.com> :
>>> 2013/4/27, Francois Romieu <romieu@...zoreil.com>:
>> [...]
>>> > napi_disable may sleep.
>>> >
>>> > fec_restart can be called with spinlock held in fec_enet_adjust_link
>>>
>>> You are right. Spin lock in FEC enet adjust link can be removed when
>>> remove spinlock in tx and RX.
>>
>> fec_restart is also called from the netdev watchdog handler (fec_timeout)
>> and tasklet can't sleep either. You should imho schedule_work from
>> fec_timeout.
>
> I will add delay_work to fix this issue.
>
>>
>> How is the driver supposed to avoid the napi context
>> fec_enet_tx:netif_queue_stopped vs fec_enet_start_xmit:netif_stop_queue
>> race since both can run on different cpu without any read/write ordering
>> enforcement between one thread netif_* and its peer {dirty/cur}_tx ?
>
> This is lockless design if only one read and one write. It likes kfifo.
>
> Assume CPU1 run xmit.  CPU2 run napi fec_enet_tx
>
> (1)       if (fep->cur_tx == fep->dirty_tx)
> (2)                 netif_stop_queue(ndev);
>
> if CPU2 update fep->dirty_tx before CPU1 run (1).  the condition is false.
> if CPU2 update fep->dirty_tx after (1) before (2),   netif_stop_queue
> will be called.  There are not problem at this time even though queue
> is not full here.  queue is not empty for sure here. fec_enet_tx will
> be called again when NAPI trigger by one frame finished transfer,
> fec_enet_tx will wake up send queue.
>
> Most worse case is waste one BD script.

I just send out a patch to try to fix your problem.
Can you help test it in your sit?

>
>>
>> --
>> Ueimor
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html