On 4/1/20 4:00 PM, Jason Gunthorpe wrote:
> On Wed, Apr 01, 2020 at 03:58:31PM -0700, Andrew Morton wrote:
>> On Tue, 31 Mar 2020 19:56:12 +1100 Stephen Rothwell <sfr@xxxxxxxxxxxxxxxx> wrote:
>>
>>> Hi all,
>>>
>>> After merging the akpm-current tree, today's linux-next build (i386
>>> defconfig) produced this warning:
>>>
>>> mm/hugetlb.c: In function 'huge_pte_offset':
>>> cc1: warning: function may return address of local variable [-Wreturn-local-addr]
>>> mm/hugetlb.c:5361:14: note: declared here
>>> 5361 | pud_t *pud, pud_entry;
>>> | ^~~~~~~~~
>>> cc1: warning: function may return address of local variable [-Wreturn-local-addr]
>>> mm/hugetlb.c:5360:14: note: declared here
>>> 5360 | p4d_t *p4d, p4d_entry;
>>> | ^~~~~~~~~
>>>
>>> Introduced by commit
>>>
>>> 826ddc88e2cf ("mm/hugetlb: fix a addressing exception caused by huge_pte_offset")
>>
>> I can reproduce this (i386 defconfig, gcc-7.2.0).
>>
>> I can see no way in which this makes any sense. Hopefully it's a gcc
>> bug but it's hard to see how it could have messed up this fairly simple
>> code.
>
> It is a code bug, there was a little thread about this. It happens
> because the address of a stack variable is passed into the pXX_offset
> functions which return that address when the page levels are folded.
>
> I recommend to drop this patch until it is adjusted..
Yes, this patch causes at least the BUG here,
https://lore.kernel.org/linux-mm/CA+G9fYsJgZhhWLMzUxu_ZQ+THdCcJmFbHQ2ETA_YPP8M6yxOYA@xxxxxxxxxxxxxx/
--
Mike Kravetz
