Am 12.07.2016 um 16:36 schrieb Peter Lieven:
> Am 11.07.2016 um 18:39 schrieb Eric Blake:
>> On 07/11/2016 03:07 AM, Peter Lieven wrote:
>>> the allocated stack will be adjusted to the minimum supported stack size
>>> by the OS and rounded up to be a multiple of the system pagesize.
>>> Additionally an architecture dependent guard page is added to the stack
>>> to catch stack overflows.
>>>
>>> Signed-off-by: Peter Lieven <p...@kamp.de>
>>> ---
>>> include/sysemu/os-posix.h | 23 +++++++++++++++++++++++
>>> util/oslib-posix.c | 44 ++++++++++++++++++++++++++++++++++++++++++++
>>> 2 files changed, 67 insertions(+)
>>>
>>> +
>>> +static size_t adjust_stack_size(size_t sz)
>>> +{
>>> + /* avoid stacks smaller than _SC_THREAD_STACK_MIN */
>>> + sz = MAX(sz, sysconf(_SC_THREAD_STACK_MIN));
>> sz is unsigned, but sysconf() is signed. Furthermore, sysconf() is
>> permitted to return -1 if there is no such minimum. MAX() would then
>> operate on the common integral promotion between the two arguments,
>> which may treat (unsigned)(-1) as the larger of the two values, and give
>> you the wrong results.
>>
>> I think it is theoretical (all platforms that we compile on have a
>> working sysconf(_SC_THREAD_STACK_MIN), right?), but still may be worth
>> being sure that sysconf() returned a positive value before computing MAX().
>>
> If you feel more comfortable I can surround it by a
>
> if (sysconf(_SC_THREAD_STACK_MIN) > 0) { }
>
> I wonder if the _SC_THREAD_STACK_MIN constant exists if there is no minimum?
Update:
glibc basically does the following:
static gulong g_thread_min_stack_size = 0;
#ifdef _SC_THREAD_STACK_MIN
g_thread_min_stack_size = MAX (sysconf (_SC_THREAD_STACK_MIN), 0);
#endif /* _SC_THREAD_STACK_MIN */
stack_size = MAX (g_thread_min_stack_size, stack_size);
So we should do sth similar, I think?!
Peter
