Re: s390-ccw: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]

[Philippe Mathieu-Daudé]

On 4/22/21 12:30 PM, Peter Maydell wrote:
> On Thu, 22 Apr 2021 at 11:18, Daniel P. Berrangé <berrange@redhat.com> wrote:
>>
>> On Thu, Apr 22, 2021 at 06:47:30AM +0200, Thomas Huth wrote:
>>> On 22/04/2021 06.18, Philippe Mathieu-Daudé wrote:
>>>> Hi Thomas, Daniel, Stefano,
>>>>
>>>> Regarding the following warning (GCC 11 on Fedora 34):
>>>>
>>>> In file included from pc-bios/s390-ccw/main.c:11:
>>>>
>>>> In function ‘memset’,
>>>>
>>>>      inlined from ‘boot_setup’ at pc-bios/s390-ccw/main.c:185:5,
>>>>
>>>>      inlined from ‘main’ at pc-bios/s390-ccw/main.c:288:5:
>>>>
>>>> pc-bios/s390-ccw/libc.h:28:14: warning: writing 1 byte into a region of
>>>> size 0 [-Wstringop-overflow=]
>>>>
>>>>     28 |         p[i] = c;
>>>>
>>>>        |         ~~~~~^~~
>>>>
>>>> Daniel were right on IRC:
>>>>
>>>> danpb: it is from a call  memset((char *)S390EP, 0, 6)     where  S390EP
>>>> is just a constant address 0x10008
>>>> danpb: the compiler doesn't now how big that is, so it seems to assume
>>>> it is zero length
>>>>
>>>> This is a known GCC issue:
>>>> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99578
>>>> "gcc-11 -Warray-bounds or -Wstringop-overread warning when accessing a
>>>> pointer from integer literal"
>>>
>>>  Hi Philippe,
>>>
>>> thanks for following up with the gcc bugzilla!
>>>
>>> ... so the problem is that GCC thinks we're in fact dereferencing a NULL
>>> pointer at offset 0x10008 here? Wow, that's ... crazy.
>>>
>>> Not sure what to do now - wait for the bug to get resolved? Compile the
>>> s390-ccw bios with -Wno-stringop-overread ? Add "volatiles" here and there
>>> to hope that these silence the compiler warnings? ... I tend to wait for the
>>> bug ticket to see whether the GCC folks change the behavior of the compiler
>>> again, but I'm open for other suggestions.
>>
>> Assuming it is just this one place in the code ,then we should just
>> use "pragma" to temporarily disable/re-enable that single warning flag
>> either side of the problem.
> 
> The gcc bug report suggests that use of 'volatile' also sidesteps
> the warning. Is that a sensible approach here ?

I'm not sure I got it right... I tried:

-    memset((char *)S390EP, 0, 6);
+    memset((char *)(volatile char *)S390EP, 0, 6);

But no change (still -Wstringop-overflow=).

If I use:

-    memset((char *)S390EP, 0, 6);
+    memset((volatile char *)S390EP, 0, 6);

I still have -Wstringop-overflow=, but also:

pc-bios/s390-ccw/main.c:185:12: warning: passing argument 1 of ‘memset’
discards ‘volatile’ qualifier from pointer target type
[-Wdiscarded-qualifiers]
pc-bios/s390-ccw/libc.h:22:34: note: expected ‘void *’ but argument is
of type ‘volatile char *’

This silents the warning however:

-- >8 --
diff --git a/pc-bios/s390-ccw/libc.h b/pc-bios/s390-ccw/libc.h
index bcdc45732d..2dea399904 100644
--- a/pc-bios/s390-ccw/libc.h
+++ b/pc-bios/s390-ccw/libc.h
@@ -19,6 +19,8 @@ typedef unsigned short     uint16_t;
 typedef unsigned int       uint32_t;
 typedef unsigned long long uint64_t;

+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wstringop-overflow"
 static inline void *memset(void *s, int c, size_t n)
 {
     size_t i;
@@ -30,6 +32,7 @@ static inline void *memset(void *s, int c, size_t n)

     return s;
 }
+#pragma GCC diagnostic pop

 static inline void *memcpy(void *s1, const void *s2, size_t n)
 {
---

Oddly this code doesn't emit any warning:

  if (!memcmp((char *)S390EP, "S390EP", 6)) {
  ...

Regards,

Phil.