From: Angelo Rizzi <angelo.rizzi@3dautomazione.it>
To: Eric Dumazet <eric.dumazet@gmail.com>
Cc: netdev@vger.kernel.org,
"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>
Subject: Re: net_tx_action race condition?
Date: Mon, 24 Nov 2014 18:19:00 +0100 [thread overview]
Message-ID: <54736884.5040600@3dautomazione.it> (raw)
In-Reply-To: <1416843200.17888.53.camel@edumazet-glaptop2.roam.corp.google.com>
Hi Eric,
Thanks for your reply
You are right, it seems a bug in the NIOS2 architecture port.
I will check how local_irq_disable()/local_irq_enable() is implemented
on this kind of architecture.
Regards,
Angelo
Il 24/11/2014 16:33, Eric Dumazet ha scritto:
> On Mon, 2014-11-24 at 14:29 +0100, Angelo Rizzi wrote:
>> Hi Daniel,
>> Here attached the patch file you required.
>> The problem i've found is on the declaration of 'struct softnet_data
>> *sd' in function 'net_tx_action'
>> What happens to me (i have an embedded system based on FPGA and a NIOS2
>> cpu) is that, due to compiler optimization, the content of
>> 'sd->completion_queue' is saved in a CPU register before interrupt
>> disabling (when the instruction 'if (sd->completion_queue) {' is
>> executed) and then the register contents is used for interrupt-disabled
>> assignment ('clist = sd->completion_queue') instead of re-read the
>> variable contents.
>> This seems to lead to a race condition when an interrupt modifies the
>> content of 'sd->completion_queue' between these two instructions.
>> What i have done to avoid this situation is to change the declaration of
>> 'struct softnet_data *sd' to 'volatile struct softnet_data *sd' and now
>> everything seems to be ok.
>> I hope this will help.
>>
>> Regards,
>> Angelo
>>
> Do not add volatile in the kernel, this is not how we solve this kind of
> problems. Documentation/memory-barriers.txt and
>
> Documentation/00-INDEX:476:volatile-considered-harmful.txt
> Documentation/00-INDEX:477: - Why the "volatile" type class should not be used
>
>
> I am surprised this patch is needed. Many other 'bugs' would need
> similar fixes.
>
> local_irq_disable() MUST have a memory barrier. This looks like a bug in
> one particular arch implementation.
>
>
> On x86 for example, native_irq_disable() is really :
>
> asm volatile("cli": : :"memory");
>
>
> diff --git a/net/core/dev.c b/net/core/dev.c
> index ac4836241a965a71952469ba054f87d8dfca2d32..fa73072e515aa07fa8ae1bc39174b7d59c7a31a5 100644
> --- a/net/core/dev.c
> +++ b/net/core/dev.c
> @@ -3406,7 +3406,7 @@ static void net_tx_action(struct softirq_action *h)
> struct sk_buff *clist;
>
> local_irq_disable();
> - clist = sd->completion_queue;
> + clist = ACCESS_ONCE(sd->completion_queue);
> sd->completion_queue = NULL;
> local_irq_enable();
>
>
>
>
next prev parent reply other threads:[~2014-11-24 17:34 UTC|newest]
Thread overview: 3+ messages / expand[flat|nested] mbox.gz Atom feed top
[not found] <547332D7.7050104@3dautomazione.it>
2014-11-24 15:33 ` Eric Dumazet
2014-11-24 17:19 ` Angelo Rizzi [this message]
[not found] <BY2PR05MB712796CB064F3202FEFA328B3A30@BY2PR05MB712.namprd05.prod.outlook.com>
2018-03-28 16:32 ` Eric Dumazet
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=54736884.5040600@3dautomazione.it \
--to=angelo.rizzi@3dautomazione.it \
--cc=eric.dumazet@gmail.com \
--cc=linux-kernel@vger.kernel.org \
--cc=netdev@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®