Re: [PATCH] powerpc: Use sysconf (_SC_SIGSTKSZ) to set SIGSTKSZ and MINSIGSTKSZ.

[Manjunath S Matti <mmatti@linux.vnet.ibm.com>]

On 25/05/23 12:38 am, Florian Weimer wrote:
> * Manjunath Matti via Libc-alpha:
>
>> diff --git a/sysdeps/unix/sysv/linux/powerpc/bits/sigstksz.h b/sysdeps/unix/sysv/linux/powerpc/bits/sigstksz.h
>> new file mode 100644
>> index 0000000000..399a49c1b0
>> --- /dev/null
>> +++ b/sysdeps/unix/sysv/linux/powerpc/bits/sigstksz.h
>> @@ -0,0 +1,33 @@
>> +#ifndef _SIGNAL_H
>> +# error "Never include <bits/sigstksz.h> directly; use <signal.h> instead."
>> +#endif
>> +
>> +#if defined __USE_DYNAMIC_STACK_SIZE && __USE_DYNAMIC_STACK_SIZE
>> +# include <unistd.h>
>> +
>> +/* Default stack size for a signal handler: sysconf (SC_SIGSTKSZ).  */
>> +# undef SIGSTKSZ
>> +# define SIGSTKSZ sysconf (_SC_SIGSTKSZ)
>> +
>> +/* Minimum stack size for a signal handler: SIGSTKSZ/4.  */
>> +# undef MINSIGSTKSZ
>> +# define MINSIGSTKSZ (SIGSTKSZ >> 2)
>> +#endif
> I'm not sure if it is a good idea to hard-code that constant 4 here.
> Similarly, it's not great to encode the constant 1 in the default case.
>
> This is a subtle portability hazard because with these changes,
> MINSIGSTKSZ is not enough on POWER to do anything on the signal stack
> (because it's exactly the kernel supplied value).  On other
> architectures, it's at least possible to do some minor stuff and call
> e.g. siglongjmp (at least if lazy binding is not required).

After deliberate thinking, internal discussions and the below comment by 
H J Lu.

https://patchwork.sourceware.org/project/glibc/patch/20230424105208.301614-1-mmatti@linux.ibm.com/#144525

I think we can drop this patch.

> The factor 4 in the implementation is somewhat x86-64-specific because
> it accounts for the fact that we do a signal-like context switch
> (pushing XSAVE data on the stack) in the lazy binding trampoline.
>
> Thanks,
> Florian
>
Thank you Florian for your time and support.

Regards,

Manjunath S Matti.