[LEDE-DEV] FW: UDP throughput caused kernel panic if configured bridge mode in /etc/config/network

Y.B. Lu yangbo.lu at nxp.com
Wed Apr 26 01:47:33 PDT 2017


> -----Original Message-----
> From: Yousong Zhou [mailto:yszhou4tech at gmail.com]
> Sent: Monday, April 24, 2017 5:15 PM
> To: Y.B. Lu
> Cc: John Crispin; lede-dev at lists.infradead.org; jo at mein.io; Wes Li;
> Xiaobo Xie
> Subject: Re: [LEDE-DEV] FW: UDP throughput caused kernel panic if
> configured bridge mode in /etc/config/network
> 
> On 24 April 2017 at 15:39, Y.B. Lu <yangbo.lu at nxp.com> wrote:
> > Hi John,
> >
> > Thank you very much.
> >
> > But I still feel it's strange the crash didn't happen if used brctl to
> configure instead of /etc/config/network.
> > Much memory(about 700MB) was consumed in UDP throughput test only when
> used /etc/config/network.
> >
> > As I know, both ls1043a with DPAA ethernet driver and ls1012a with ppfe
> ethernet driver had this issue.
> > I think maybe I should focus on deep studying in /etc/config/network.
> But we had a deadline by this month to resolve it.
> >
> > Is there any possibility the issue was caused by OpenWrt?
> > Thanks again.
> >
> 
> Most parts of /etc/config/network will be parsed and executed by netifd.
> Comparing strace output of both brctl and netifd may reveal something
> useful.
> 
> The kernel is supposed to be resilient against whatever ways applications
> may use its exposed userspace interface which is driver and device-
> agnostic.  If the driver fails, there is very likely an issue within the
> driver itself.
> 
>                 yousong

[Lu Yangbo-B47093] Hi Yousong, thank you so much for your comments and suggestion.

It's now very strange. The driver failed because of there was no enough memory.
But this only happened with /etc/config/network bridge configure.
Both ls1012a and ls1043a which had different ethernet drivers got this problem.

I think your suggestion comparing strace output of both brctl and netifd probably is a good method.
We will have a try.

Thanks.

- Yangbo Lu
















More information about the Lede-dev mailing list