netdev - Re: [net-next PATCH 2/7] mlx4: use napi_consume

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: <20160309.160320.1535201013338177802.davem@davemloft.net>
Date:	Wed, 09 Mar 2016 16:03:20 -0500 (EST)
From:	David Miller <davem@...emloft.net>
To:	alexander.duyck@...il.com
Cc:	brouer@...hat.com, netdev@...r.kernel.org, eugenia@...lanox.com,
	alexei.starovoitov@...il.com, saeedm@...lanox.com,
	gerlitz.or@...il.com
Subject: Re: [net-next PATCH 2/7] mlx4: use napi_consume_skb API to get
 bulk free operations

From: Alexander Duyck <alexander.duyck@...il.com>
Date: Wed, 9 Mar 2016 08:47:58 -0800

> On Wed, Mar 9, 2016 at 3:00 AM, Jesper Dangaard Brouer
> <brouer@...hat.com> wrote:
>> Passing the budget down was Alex'es design.  Axel any thoughts?
> 
> I'd say just use dev_consume_skb_any in the bulk free instead of
> dev_consume_skb_irq.  This is slow path, as you said, so it shouldn't
> come up often.

Agreed.

>> I do wonder how expensive this check is... as it goes into a code
>> hotpath, which is very unlikely.  The good thing would be, that we
>> handle if buggy drivers call this function from a none softirq context
>> (as these bugs could be hard to catch).
>>
>> Can netpoll ever be called from softirq or with BH disabled? (It
>> disables IRQs, which would break calling kmem_cache_free_bulk).
> 
> It is better for us to switch things out so that the napi_consume_skb
> is the fast path with dev_consume_skb_any as the slow.  There are too
> many scenarios where we could be invoking something that makes use of
> this within the Tx path so it is probably easiest to just solve it
> that way so we don't have to deal with it again in the future.

Indeed.