[PATCH v2] arm64: disable kasan when accessing frame->fp in unwind_frame

[Shi, Yang]

On 2/9/2016 9:31 AM, Shi, Yang wrote:
> On 2/9/2016 9:23 AM, Will Deacon wrote:
>> On Tue, Feb 09, 2016 at 09:17:12AM -0800, Shi, Yang wrote:
>>> On 2/9/2016 8:54 AM, Will Deacon wrote:
>>>> On Mon, Feb 08, 2016 at 09:13:09AM -0800, Yang Shi wrote:
>>>>> When boot arm64 kernel with KASAN enabled, the below error is
>>>>> reported by
>>>>> kasan:
>>>>>
>>>>> BUG: KASAN: out-of-bounds in unwind_frame+0xec/0x260 at addr
>>>>> ffffffc064d57ba0
>>>>> Read of size 8 by task pidof/499
>>>>> page:ffffffbdc39355c0 count:0 mapcount:0 mapping:          (null)
>>>>> index:0x0
>>>>> flags: 0x0()
>>>>> page dumped because: kasan: bad access detected
>>>>> CPU: 2 PID: 499 Comm: pidof Not tainted 4.5.0-rc1 #119
>>>>> Hardware name: Freescale Layerscape 2085a RDB Board (DT)
>>>>> Call trace:
>>>>> [<ffffffc00008d078>] dump_backtrace+0x0/0x290
>>>>> [<ffffffc00008d32c>] show_stack+0x24/0x30
>>>>> [<ffffffc0006a981c>] dump_stack+0x8c/0xd8
>>>>> [<ffffffc0002e4400>] kasan_report_error+0x558/0x588
>>>>> [<ffffffc0002e4958>] kasan_report+0x60/0x70
>>>>> [<ffffffc0002e3188>] __asan_load8+0x60/0x78
>>>>> [<ffffffc00008c92c>] unwind_frame+0xec/0x260
>>>>> [<ffffffc000087e60>] get_wchan+0x110/0x160
>>>>> [<ffffffc0003b647c>] do_task_stat+0xb44/0xb68
>>>>> [<ffffffc0003b7730>] proc_tgid_stat+0x40/0x50
>>>>> [<ffffffc0003ac840>] proc_single_show+0x88/0xd8
>>>>> [<ffffffc000345be8>] seq_read+0x370/0x770
>>>>> [<ffffffc00030aba0>] __vfs_read+0xc8/0x1d8
>>>>> [<ffffffc00030c0ec>] vfs_read+0x94/0x168
>>>>> [<ffffffc00030d458>] SyS_read+0xb8/0x128
>>>>> [<ffffffc000086530>] el0_svc_naked+0x24/0x28
>>>>> Memory state around the buggy address:
>>>>>   ffffffc064d57a80: 00 00 00 00 00 00 00 00 f1 f1 f1 f1 00 00 f4 f4
>>>>>   ffffffc064d57b00: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
>>>>>> ffffffc064d57b80: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
>>>>>                                    ^
>>>>>   ffffffc064d57c00: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
>>>>>   ffffffc064d57c80: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
>>>>>
>>>>> Since the shadow byte pointed by the report is 0, so it may mean it
>>>>> is just hit
>>>>> oob in non-current task. So, disable the instrumentation to silence
>>>>> these
>>>>> warnings.
>>>>
>>>> Curious, but how did you trigger this? I'm just trying to confirm that
>>>> mainline is affected, but my machine boots happily with KASAN and
>>>> STACKTRACE
>>>> selected and I can cat /proc/self/{stack,stat} quite happily.
>>>>
>>>> What am I missing?
>>>
>>> I'm using mainline 4.5-rc1 kernel with gcc 5.2. And, my rootfs is NFS
>>> mounted.
>>>
>>> Not sure if other kernel configs, i.e tracing stuff will have impact
>>> on the
>>> trigger since I'm not using the defconfig.
>>
>> If you could put your .config somewhere, that would be helpful, please.
>
> Attached it. BTW, I run the test on LS2085a RDB board which has 8 A57
> cores. Not sure if it could be reproduced on other boards easily.

This config has GCOV enabled, my test is run with it disabled. So, you 
may need disable it to replicate the trigger.

Regards,
Yang

>
> Yang
>
>>
>> Will
>>
>