linux-kernel - Re: 4.9.28: VIA velocity hang: how to find cause?

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] [day] [month] [year] [list]

Message-ID: <20170525134617.286bcb5b@alans-desktop>
Date:   Thu, 25 May 2017 13:46:17 +0100
From:   Alan Cox <gnomes@...rguk.ukuu.org.uk>
To:     Udo van den Heuvel <udovdh@...all.nl>
Cc:     "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: Re: 4.9.28: VIA velocity hang: how to find cause?

On Thu, 25 May 2017 13:52:08 +0200
Udo van den Heuvel <udovdh@...all.nl> wrote:

> Hello,
> 
> Forcing the VIA Velocity interface to 1000 Mbps FDX appears to help a
> bit but still the connection stops transferring data after a while.
> I added some printk's and in or after velocity_tx_service it ends.
> 
> Where should I focus with my debug printk's?
> 
> It appears I can reproduce the issue by running a `find /` over ssh to
> the box. So perhaps it is an interrupt related issue?
> 
> Please help find the root cause.
> 

netdev@...r.kernel.org would be a better place to ask

But some things to play with

1. See if an scp of a very large file from a local machine can trigger it

2. See if you can trigger it with a small number of ping -f commands

3. Try both of them the other way around to see if its rx or tx related

Also see if you can duplicate the problem on another machine - just in
case. Saves spending days chasing some kind of insane hardware
compatibility problem 8)

My first guess given that more speed helps would be that something is
mishandling the ring buffer filling on receive (and that in turn causes
tcp to back off) so I would look at logging ring full events. Also dump
the device state after it hangs and see what the hardware thinks is going
on (eg IRQ's pending, rings full, errors logged, link state ....)

tcpdump might also give you some clues providing it doesn't change the
performance too much and hide the problem. You'll get to see if there are
any dropped frames/back offs/where it gets stuck.

Alan