Re: [PATCH] kvm-unit-tests: Fix GCC's 4.8 labels as values for nVMX tests.

[Jan Kiszka <jan.kiszka@siemens.com>]

On 2014-02-25 15:26, Paolo Bonzini wrote:
> Il 24/02/2014 16:58, Jan Kiszka ha scritto:
>> On 2014-02-24 16:25, Marius Vlad wrote:
>>> Commit 3b1274463fa8d074dd3bc77efe25b59a4ddd491e uses GCCs extension
>>> labels as values to handle exceptions, but GCC 4.8 ``mistakingly''
>>> uses the next body function as a jump label, for functions which
>>> do not return. Fixed by returning a int value for those functions.
>>>
>>> See http://thread.gmane.org/gmane.comp.emulators.kvm.devel/119186
>>>
>>> Signed-off-by: Marius Vlad <mv@sec.uni-passau.de>
>>> ---
>>>  x86/vmx.c | 10 +++++-----
>>>  1 file changed, 5 insertions(+), 5 deletions(-)
>>>
>>> diff --git a/x86/vmx.c b/x86/vmx.c
>>> index fe950e6..0c895af 100644
>>> --- a/x86/vmx.c
>>> +++ b/x86/vmx.c
>>> @@ -548,7 +548,7 @@ static void exception_handler(struct ex_regs *regs)
>>>      regs->rip = (u64)exception_return;
>>>  }
>>>
>>> -static int test_for_exception(unsigned int ex, void (*func)(void))
>>> +static int test_for_exception(unsigned int ex, int (*func)(void))
>>>  {
>>>      handle_exception(ex, exception_handler);
>>>      exception = false;
>>> @@ -557,23 +557,23 @@ static int test_for_exception(unsigned int ex,
>>> void (*func)(void))
>>>      return exception;
>>>  }
>>>
>>> -static void do_vmxon_off(void)
>>> +static int do_vmxon_off(void)
>>>  {
>>>      exception_return = &&resume;
>>>      barrier();
>>>      vmx_on();
>>>      vmx_off();
>>>  resume:
>>> -    return;
>>> +    return 0;
>>>  }
>>>
>>> -static void do_write_feature_control(void)
>>> +static int do_write_feature_control(void)
>>>  {
>>>      exception_return = &&resume;
>>>      barrier();
>>>      wrmsr(MSR_IA32_FEATURE_CONTROL, 0);
>>>  resume:
>>> -    return;
>>> +    return 0;
>>>  }
>>>
>>>  static int test_vmx_feature_control(void)
>>>
>>
>> Argh, getting old. I remembered that issue but forgot that I already had
>> a fix for this queued:
>>
>> http://thread.gmane.org/gmane.comp.emulators.kvm.devel/117866
>>
>> I don't mind which version to pick, but maybe Paolo has mine already in
>> his queue.
> 
> Yeah (I hadn't, but now I have).  According to GCC developers this is
> invalid code.
> 
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28581

Oh - how nicely intuitive.

> 
> Perhaps we should rewrite this using something more similar to "normal"
> C setjmp/longjmp.

Hmm, maybe. If barrier workaround wouldn't work already...

Jan

-- 
Siemens AG, Corporate Technology, CT RTC ITP SES-DE
Corporate Competence Center Embedded Linux