Re: [PATCH] NPTL proposed large page fix for pthread_create.

[Ulrich Drepper <drepper at redhat dot com>]

Steven Munroe wrote:
> +	{
> +	/* The stack is too small (or the guard too large), but this might not
> +	   be an error.  If we are on a system that supports multiple page
> +	   sizes,  the current page size may be larger then PTHREAD_STACK_MIN,
> +	   but we need to allocate a minimum of two pages for guard and stack.
> +	   In this case we can allocate the two page minimum.  */
> +	  if ((size >= PTHREAD_STACK_MIN)
> +	    && (PTHREAD_STACK_MIN< ((pagesize_m1 + 1)  * 2)))
> +	    /* The requested size is >= PTHREAD_STACK_MIN,  so the application
> +	       expect this to work. But PTHREAD_STACK_MIN is < 2 of the
> +	       current pages,  so bump the stack size up to two pages at the
> +	       current page size.  */
> +	    size = (pagesize_m1 + 1) * 2;
> +	  else
> +	    /* Not the large page case so return EINVAL.  */
> +	    return EINVAL;
> +	}

This is completely bogus.  The if expression change is OK, but if the 
stack is too small, it is too small.  Why on earth would you think that 
if the test for the minimally required stack size fails you can just add 
a test like this to work around it?  This second test is completely 
independent of the actual size needed.

If the modified test I checked in fails there is *nothing* which can be 
done.  There is no negotiating about the required stack size.  You 
cannot just ignore this fact.  If all this means programs which work 
with normal page sizes now fail, then that's a sad outcome.  Just 
another piece of evidence of the poor ABI design.


And as usual, you completely ignore the coding conventions.  Well, I can 
ignore, too.  I'll ignore every patch you sent which does not use the 
correct coding conventions.

--
â Ulrich Drepper â Red Hat, Inc. â 444 Castro St â Mountain View, CA â