Re: [PATCH v2] memcpy: use bhs/bls instead of bge/blt [BZ #25620]

[Florian Weimer <fw@deneb.enyo.de>]

* Joseph Myers:

> On Tue, 21 Apr 2020, Florian Weimer wrote:
>
>> * Joseph Myers:
>> 
>> > On Mon, 13 Apr 2020, zhuyan (M) wrote:
>> >
>> >> In ARMv7, the memcpy() implementation allows for program execution to 
>> >> continue in scenarios where a segmentation fault or crash should have 
>> >> occurred. The dangers occur in that subsequent execution and iterations 
>> >> of this code will be executed with this corrupted data.
>> >> 
>> >> Such as, we use 'memcpy' copy 0x80000000 byte to buffer(The buffer size 
>> >> is 100 bytes), it didn't crash.
>> >
>> > This patch includes an architecture-specific test, specific to memcpy.  
>> > My understanding of Wilco's request in bug 25620 was for an 
>> > architecture-independent test or tests, covering all string functions with 
>> > such large arguments, so we can ensure we're consistent across 
>> > architectures.
>> 
>> Sure, that would be great, but can we make this independent of this
>> fix, please?
>
> I think the minimum for this fix should be an architecture-independent 
> test for memcpy (but not other string functions, and not necessarily 
> testing all IFUNC variants of memcpy).

Fair enough.  I'll try to write one using the RAM compression feature
I added.  But I can't get a 32-bit Arm machine from the labs at work
right now, and it may be a while until I get a device delivered, so it
will be at least a few days more until I have a test that's been
confirmed to reproduce the bug.