netdev - Re: [net-next PATCH 2/6] enic: Bug fix: try harder to fill Rx ring on skb allocation failures

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: Windows password security audit tool. GUI, reports in PDF.

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

Message-ID: <C75279E7.1B001%scofeldm@cisco.com>
Date:	Sat, 19 Dec 2009 12:39:03 -0800
From:	Scott Feldman <scofeldm@...co.com>
To:	Simon Horman <horms@...ge.net.au>
CC:	<davem@...emloft.net>, <netdev@...r.kernel.org>
Subject: Re: [net-next PATCH 2/6] enic: Bug fix: try harder to fill Rx ring on
 skb allocation failures

On 12/19/09 2:41 AM, "Simon Horman" <horms@...ge.net.au> wrote:

> On Fri, Dec 18, 2009 at 06:09:46PM -0800, Scott Feldman wrote:
>> enic: Bug fix: try harder to fill Rx ring on skb allocation failures
>> 
>> During probe(), make sure we get at least one skb on the Rx ring.
>> Otherwise abort the interface load.  Also, if we get skb allocation
>> failures in NAPI poll while trying to replenish the ring, try again
>> later so we don't end up starving out the Rx ring completely.
>> 
>> @@ -1115,16 +1116,19 @@ static int enic_poll(struct napi_struct *napi, int
>> budget)
>> 0 /* don't unmask intr */,
>> 0 /* don't reset intr timer */);
>>  
>> - if (rq_work_done > 0) {
>> + err = vnic_rq_fill(&enic->rq[0], enic->rq_alloc_buf);
>>  
>> -  /* Replenish RQ
>> -   */
>> + /* Buffer allocation failed. Stay in polling
>> +  * mode so we can try to fill the ring again.
>> +  */
>>  
>> -  vnic_rq_fill(&enic->rq[0], enic->rq_alloc_buf);
>> + if (err)
>> +  rq_work_done = rq_work_to_do;
> 
> Is it intentional for rq_work_done = rq_work_to_do to become the
> return value?

That was intentional.  If the replacement skb allocation fails, we're
returning like we did a full budget's worth of work so we stay scheduled and
hopefully the next polling pass we'll get the allocations.  Before this fix,
there is a corner case which isn't covered: if hw has used all descs and
gens an intr and we get into polling and the replacement alloc fails, then
Rx is hung.  Hw is desc starved and we're not going to get any more intrs:
game over.

I was looking at tg3.c and it looks like it does napi_schedule() if the
allocation fails.  Would this be a better option than setting work_done =
budget?
 
>> @@ -1333,11 +1343,13 @@ static int enic_open(struct net_device *netdev)
>> }
>>  
>> for (i = 0; i < enic->rq_count; i++) {
>> -  err = vnic_rq_fill(&enic->rq[i], enic->rq_alloc_buf);
>> -  if (err) {
>> +  vnic_rq_fill(&enic->rq[i], enic->rq_alloc_buf);
>> +  /* Need at least one buffer on ring to get going */
>> +  if (vnic_rq_desc_used(&enic->rq[i]) == 0) {
>> printk(KERN_ERR PFX
>> "%s: Unable to alloc receive buffers.\n",
>> netdev->name);
>> +   err = -ENOMEM;
>> goto err_out_notify_unset;
>> }
>> }
> 
> My brain may well have switched off for the day,
> but its unclear to me how &enic->rq[i] could ever be NULL.

You missed a ")"
 
> Also, in the case where a failure occurs for i > 0,
> it it necessary to unwind the previous rq allocations?

Yes, good catch.  We'll fix that.

-scott

--
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