On 9/14/2026 1:34 AM, Daniel P. Berrangé wrote:
> On Sun, Sep 13, 2026 at 10:08:52PM -0700, Pierrick Bouvier wrote:
>> On 9/10/26 11:14 AM, Alex Bennée wrote:
>>> From: Daniel P. Berrangé <[email protected]>
>>>
>>> When it finds a cross compiler, tests/tcg/meson.build tries
>>> building test_cc.c to validate that the compiler works. This
>>> includes "stdint.h" as a witness for working C library, however,
>>> that is not a good choice as it can be provided by the toolchain
>>> alone:
>>>
>>>    $ rpm -ql gcc-s390x-linux-gnu  | grep /stdint.h
>>>    /usr/lib/gcc/s390x-linux-gnu/16/include/stdint.h
>>>
>>> As a result, if you install the s390x GCC cross compiler, but
>>> not any C library, meson decides to use the host compiler instead
>>> of the container:
>>>
>>>      cc for aarch64-softmmu          : aarch64-linux-gnu-gcc (from 
>>> 'debian-all-test-cross' container)
>>>      cc for loongarch64-softmmu      : loongarch64-unknown-linux-gnu-gcc 
>>> (from 'debian-loongarch-cross' container)
>>>      cc for riscv64-softmmu          : riscv64-linux-gnu-gcc (from 
>>> 'debian-all-test-cross' container)
>>>      cc for s390x-softmmu            : s390x-linux-gnu-gcc
>>>      cc for x86_64-softmmu           : x86_64-linux-gnu-gcc (from 
>>> 'debian-amd64-cross' container)
>>>
>>> Eventually this results in failure to run 'make check':
>>>
>>> [33/40] Generating tests/tcg/s390x-softmmu-hello with a custom command
>>> FAILED: [code=1] tests/tcg/s390x-softmmu-hello.test
>>> /usr/bin/s390x-linux-gnu-gcc 
>>> /home/berrange/src/virt/qemu/tests/tcg/multiarch/system/hello.c -o 
>>> tests/tcg/s390x-softmmu-hello.test -static -MMD -MF 
>>> tests/tcg/s390x-softmmu-hello.d -Wall -Werror -O0 -g -fno-strict-aliasing 
>>> -nostdlib -ffreestanding -Wa,--noexecstack -I 
>>> /home/berrange/src/virt/qemu/tests/tcg/s390x/system/../../minilib 
>>> /home/berrange/src/virt/qemu/tests/tcg/s390x/system/../../minilib/printf.c 
>>> -I /home/berrange/src/virt/qemu/include/hw/s390x/ipl -march=z13 
>>> ../tests/tcg/s390x/system/../head64.S ../tests/tcg/s390x/system/../console.c
>>> In file included from ../tests/tcg/s390x/system/../console.c:8:
>>> ../tests/tcg/s390x/system/../../../../pc-bios/s390-ccw/sclp.c:11:10: fatal 
>>> error: string.h: No such file or directory
>>>     11 | #include <string.h>
>>>        |          ^~~~~~~~~~
>>> compilation terminated.
>>>
>>> Adding more include files to the test program ensures the
>>> meson probe makes a better decision.
>>>
>>> Fixes: e5d084d622b1 (tests/tcg/meson.build: check host cross cc is working)
>>> Signed-off-by: Daniel P. Berrangé <[email protected]>
>>> Message-ID: <[email protected]>
>>> Signed-off-by: Alex Bennée <[email protected]>
>>> ---
>>>   tests/tcg/test_cc.c | 6 +++++-
>>>   1 file changed, 5 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/tests/tcg/test_cc.c b/tests/tcg/test_cc.c
>>> index d3614a01fe6..fafee055b40 100644
>>> --- a/tests/tcg/test_cc.c
>>> +++ b/tests/tcg/test_cc.c
>>> @@ -1,7 +1,11 @@
>>>   /* SPDX-License-Identifier: GPL-2.0-or-later */
>>> -/* Include a standard header to make sure cross compiler provides them */
>>> +/* Use some standard headers to ensure the cross compiler provides them */
>>>   #include <stdint.h>
>>> +#include <string.h>
>>> +#include <stdbool.h>
>>> +#include <stddef.h>
>>> +#include <stdio.h>
>>>   int main(void)
>>>   {
>>
>> Ideally, we should do separate tests for system vs user tests.
>> System ones are always compiled with -ffreestanding, while user ones should
>> provide a C library.
>>
>> From this standard:
>>
>> https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2023/p2338r3.html
>> ```
>> A freestanding C implementation is required to provide the entirety of the
>> following headers:
>>
>> <float.h>
>> <iso646.h>
>> <limits.h>
>> <stdalign.h>
>> <stdarg.h>
>> <stdbool.h>
>> <stddef.h>
>> <stdint.h>
>> <stdnoreturn.h>
>> Most of <string.h> is required (strdup, strndup, strcoll, strxfrm, and
>> strerror are excluded). This includes strtok, which requires global data.
>> ```
>>
>> Maybe for now we can just remove stdio.h from current patch, as it's not
>> supposed to be provided in a freestanding environment.
> 
> A huge number of files under tests/tcg/ have a #include <stdio.h>
> currently.
> 
> In the example build failure in the commit message, we have
> 
>   tests/tcg/s390x/system/../console.c
> 
> which does
> 
>   #include "../../../pc-bios/s390-ccw/sclp.c"
> 
> and sclp.c has
> 
>   #include "s390-ccw.h"
> 
> and that header has:
> 
>   #include <stdbool.h>
>   #include <stddef.h>
>   #include <stdint.h>
>   #include <stdio.h>
> 
> hence the 4 headers I added to the meson check.
> 
> So AFAICT, our code is expecting stdio.h to be provided in a
> -ffreestanding build. Have we got a more fundamental mistake
> wrt our use of -ffreestanding, such that it is relying on
> unintended functionality ?
>

Looking a bit further, it seems that this is the only example of system
code pulling this header (through pc-bios), which might be a specificity
of s390x firmware.
That said, it's not the end of the world, and as we mostly use
-linux-gcc variants of cross compilers, it should be safe to expect them
to come with a libc. Let's keep the patch as it is.

> With regards,
> Daniel

Thanks,
Pierrick

Reply via email to