From mboxrd@z Thu Jan  1 00:00:00 1970
From: agraf@suse.de (Alexander Graf)
Date: Thu, 15 Jan 2015 13:39:03 +0100
Subject: [PATCH net-next v13 0/3] add hisilicon hip04 ethernet driver
In-Reply-To: <14674609.dERhba4yMV@wuerfel>
References: <1421217254-12008-1-git-send-email-dingtianhong@huawei.com>
 <54B642B4.8010807@suse.de> <54B77C43.6010200@huawei.com>
 <14674609.dERhba4yMV@wuerfel>
Message-ID: <54B7B4E7.8030108@suse.de>
To: linux-arm-kernel@lists.infradead.org
List-Id: linux-arm-kernel.lists.infradead.org



On 15.01.15 10:42, Arnd Bergmann wrote:
> On Thursday 15 January 2015 16:37:23 Ding Tianhong wrote:
>> On 2015/1/14 18:19, Alexander Graf wrote:
>>>
>>> >From a 10000 feet perspective it looks like two problems to me
>>>
>>>   1) Allocation failure doesn't get handled properly somewhere
> 
> This is the bug that Eric pointed out as well.
> 
>>>   2) We fail to allocate with order=0 - I don't see why
> 
> GFP_ATOMIC. When allocating from a the napi poll function in softirq
> context, you have to use nonblocking allocations, which occasionally
> fail. This should not cause any harm other than dropped packets.
> 
>> is it easy to repetition this bug? how big is your memory on your board,
>> is it happened in your previous hip04 driver?
> 
> It should be independent of memory size, but may be more likely if you
> don't have swap space configured.

With the previous driver I was unable to get this far - I ended up in
random memory corruption and had a ~90% packet loss after about an hour
of uptime.

I'm not sure whether it's easy to reproduce, I merely started up a few
VMs, did some disk I/O and started to compile QEMU in the background ;).

I'll happily give your follow up patch that's going to fix the memory
allocation problems a try though.


Alex