Re: [Qemu-devel] [PULL v3 00/53] Misc changes for 2017-01-12

[Marc-André Lureau]

Hi

On Tue, Jan 16, 2018 at 1:06 PM, Peter Maydell <address@hidden> wrote:
> On 16 January 2018 at 11:58, Marc-André Lureau
> <address@hidden> wrote:
>> Hi
>>
>> On Tue, Jan 16, 2018 at 12:25 PM, Peter Maydell
>> <address@hidden> wrote:
>>> On 15 January 2018 at 23:35, Paolo Bonzini <address@hidden> wrote:
>>>> The following changes since commit 
>>>> 997eba28a3ed5400a80f754bf3a1c8044b75b9ff:
>>>>
>>>>   Merge remote-tracking branch 
>>>> 'remotes/pmaydell/tags/pull-target-arm-20180111' into staging (2018-01-11 
>>>> 14:34:41 +0000)
>>>>
>>>> are available in the git repository at:
>>>>
>>>>
>>>>   git://github.com/bonzini/qemu.git tags/for-upstream
>>>>
>>>> for you to fetch changes up to ff9adba50bf8a4c080b8aee9be2314ef179a7b5f:
>>>>
>>>>   ucontext: annotate coroutine stack for ASAN (2018-01-12 15:21:14 +0100)
>>>>
>>>> ----------------------------------------------------------------
>>>> * QemuMutex tracing improvements (Alex)
>>>> * ram_addr_t optimization (David)
>>>> * SCSI fixes (Fam, Stefan, me)
>>>> * do {} while (0) fixes (Eric)
>>>> * KVM fix for PMU (Jan)
>>>> * memory leak fixes from ASAN (Marc-André)
>>>> * migration fix for HPET, icount, loadvm (Maria, Pavel)
>>>> * hflags fixes (me, Tao)
>>>> * block/iscsi uninitialized variable (Peter L.)
>>>> * full support for GMainContexts in character devices (Peter Xu)
>>>> * more boot-serial-test (Thomas)
>>>> * Memory leak fix (Zhecheng)
>>>
>>> Various build failures, I'm afraid:
>>>
>>
>> damn, sorry..
>>
>>> x86-64/Linux/gcc:
>>>
>>> configure produces an error message:
>>>
>>> ERROR: ASAN build enabled, but ASAN header is too old.
>>>        Without code annotation, the report may be inferior.
>>>
>>> even though this configure did not explicitly request ASAN.
>>> Then configure seems to exit successfully anyway, since the
>>> build proceeds.
>>
>> ASAN is enabled by default if available when --enable-debug. We could
>> add more flags if that helps.
>
> Configure switches should work like this:
>  * default: use feature if present, but don't complain if not present
>    or not usable
>  * --enable-foo: use feature. if feature not present, complain and
>    fail configure
>  * --disable-foo: don't test for or use feature

Would that be enough to drop the "ERROR:" prefix ?

>>> For some reason the build creates config-target.h a lot:
>
>> One per target no? (the build should be more silent than before)
>
> Oh, I guess that makes sense. It's a bit odd that the message neither
> gives the target part of the filename nor has those entering/leaving
> directory messages...
>
>>>
>>> Then it runs configure again, this time without the ERROR message,
>>> and eventually fails with:
>>>
>>>   CC      hw/display/exynos4210_fimd.o
>>> /home/petmay01/linaro/qemu-for-merges/hw/display/exynos4210_fimd.c: In
>>> function ‘fimd_get_buffer_id’:
>>> /home/petmay01/linaro/qemu-for-merges/hw/display/exynos4210_fimd.c:1105:5:
>>> error: case label does not reduce to an integer constant
>>>      case FIMD_WINCON_BUF2_STAT:
>>
>> never saw that error, hmm interesting. This is related to
>> -fsanitize=address ? Is this on Debian stable?

Interesting, looks like a bug in gcc ubsan that doesn't happen with
recent versions (related to
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80550 but probably a
different bug). Adding a cast is enough:

-#define FIMD_WINCON_BUF2_STAT       ((0 << 21) | (1 << 31))
+#define FIMD_WINCON_BUF2_STAT       (uint32_t)((0 << 21) | (1 << 31))

It looks like there are no other cases like this



>
> Ubuntu xenial (16.04.3 LTS). No idea if it's related to the
> sanitizer or not.
>
>>
>>>      ^
>>>
>>> On sparc64 host configure fails:
>
>> Hmm, I guess the check -fsanitize=address doesn't return an error
>> unless -Werror is given. Perhaps it needs:
>>
>> diff --git a/configure b/configure
>> index f5550f3289..ba68c550c9 100755
>> --- a/configure
>> +++ b/configure
>> @@ -5190,7 +5190,7 @@ fi
>>
>>  have_asan=no
>>  write_c_skeleton
>> -if compile_prog "-fsanitize=address" ""; then
>> +if compile_prog "-Werror -fsanitize=address" ""; then
>>      have_asan=yes
>>  fi
>>
>> @@ -5207,7 +5207,7 @@ int main(void) {
>>    return 0;
>>  }
>>  EOF
>> -if compile_prog "-fsanitize=address" "" ; then
>> +if compile_prog "-Werror -fsanitize=address" "" ; then
>>      have_asan_iface_fiber=yes
>>  fi
>>
>
> Looks plausible.

Actually, it probably needs also $CPU_CFLAGS

Paolo, would you drop "build-sys: add some sanitizers when
--enable-debug if possible" & "ucontext: annotate coroutine stack for
ASAN" from the series? I'll send a new version for those 2.

Thanks


-- 
Marc-André Lureau