Re: crisv32 runtime failure in -next due to 'page-flags: define behavior SL*B-related flags on compound pages'

From: Mikael Starvik <mikael.starvik@axis.com>
To: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Guenter Roeck <linux@roeck-us.net>,
	"Paul E. McKenney" <paulmck@linux.vnet.ibm.com>,
	Mikael Starvik <starvik@axis.com>,
	Jesper Nilsson <jespern@axis.com>,
	Hugh Dickins <hughd@google.com>,
	"Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>,
	"linux-next@vger.kernel.org" <linux-next@vger.kernel.org>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
	Andrew Morton <akpm@linux-foundation.org>,
	Minchan Kim <minchan@kernel.org>,
	linux-cris-kernel <linux-cris-kernel@axis.com>,
	Hans-Peter Nilsson <hp@axis.com>
Subject: Re: crisv32 runtime failure in -next due to 'page-flags: define behavior SL*B-related flags on compound pages'
Date: Tue, 22 Sep 2015 14:19:38 +0200	[thread overview]
Message-ID: <5023113A-80F3-4C3A-AF50-6B08134A070F@axis.com> (raw)
In-Reply-To: <20150922120344.GA17969@node.dhcp.inet.fi>

For cris it is completely valid to do that. It has been an issue before. If you for some reason really require dword alignment there should be an align in the struct.

CC:ing the compiler guy for further comments.

Best regard
/Mikael

> 22 sep 2015 kl. 14:03 skrev Kirill A. Shutemov <kirill@shutemov.name>:
> 
>> On Mon, Sep 21, 2015 at 06:17:34PM -0700, Guenter Roeck wrote:
>>> On 09/21/2015 08:34 AM, Kirill A. Shutemov wrote:
>>> Guenter Roeck wrote:
>>>>> On 09/18/2015 07:53 AM, Jesper Nilsson wrote:
>>>>>> On Fri, Sep 18, 2015 at 05:25:07PM +0300, Kirill A. Shutemov wrote:
>>>>>>> On Thu, Sep 17, 2015 at 09:29:27AM -0700, Guenter Roeck wrote:
>>>>>>> Hi,
>>>>>>> 
>>>>>>> my crisv32 qemu test fails with next-20150917 as follows.
>>>>>>> 
>>>>>>> NET: Registered protocol family 16
>>>>>>> kernel BUG at mm/slab.c:1648!
>>>>>>> Linux 4.3.0-rc1-next-20150917 #1 Wed Sep 16 23:56:59 PDT 2015
>>>>>>> Oops: 0000
>>>>>>> 
>>>>>>> [ register dump follows ]
>>>>>>> 
>>>>>>> See http://server.roeck-us.net:8010/builders/qemu-crisv32-next/builds/83/steps/qemubuildcommand/logs/stdio
>>>>>>> for a complete log.
>>>>>> 
>>>>>> Is there a chance to get proper backtrace?
>>>>> 
>>>>> Yes, it should be possible with CONFIG_KALLSYMS=y in the kconfig.
>>>> 
>>>> Good to know. I added it to my configuration.
>>>> 
>>>> Here it is:
>>>> 
>>>> kernel BUG at mm/slab.c:1648!
>>> 
>>> I still don't understand what's going on :(
>>> Could you try with this instrumentation:
>>> 
>>> diff --git a/mm/slab.c b/mm/slab.c
>>> index ce9c6531e6f7..10035d1a06d3 100644
>>> --- a/mm/slab.c
>>> +++ b/mm/slab.c
>>> @@ -1645,7 +1645,11 @@ static void kmem_freepages(struct kmem_cache *cachep, struct page *page)
>>>                sub_zone_page_state(page_zone(page),
>>>                                NR_SLAB_UNRECLAIMABLE, nr_freed);
>>> 
>>> -       BUG_ON(!PageSlab(page));
>>> +       if (!PageSlab(page)) {
>>> +               dump_page(page, "page");
>>> +               dump_page(compound_head(page), "compound_head(page)");
>>> +               BUG();
>>> +       }
>>>        __ClearPageSlabPfmemalloc(page);
>>>        __ClearPageSlab(page);
>>>        page_mapcount_reset(page);
>> 
>> page:c04a5340 count:1 mapcount:1 mapping:c1f34080 index:0xc1f34060
>> flags: 0x80(slab)
>> page dumped because: page
>> page:c1f17a04 count:0 mapcount:1 mapping:00d13600 index:0xc0
>> flags: 0x0()
>> page dumped because: compound_head(page)
>> 
>> Does that help ?
> 
> Kinda. It's false positive PageTail() due low bit set in
> page->rcu_head.next.
> 
> It happens (at least) due broken alignment of 'rcu' field within
> task_struct -- offsetof(struct task_struct, rcu): 773.
> 
> That's looks veery broken. I would guess compiler does something horribly
> wrong. I hope it's not an ABI issue. :-/
> 
> Mikael? Jesper?
> 
> -- 
> Kirill A. Shutemov