Re: [PATCH] mm/mmap: Tighten up cmdline_parse_stack_guard_gap()

[Matthew Wilcox]

On Wed, Aug 30, 2023 at 08:47:12AM +0530, Anshuman Khandual wrote:
> 
> 
> On 8/29/23 18:21, Matthew Wilcox wrote:
> > On Mon, Aug 28, 2023 at 10:52:12AM +0530, Anshuman Khandual wrote:
> >> -static int __init cmdline_parse_stack_guard_gap(char *p)
> >> +static int __init cmdline_parse_stack_guard_gap(char *str)
> >>  {
> >>  	unsigned long val;
> >> -	char *endptr;
> >>  
> >> -	val = simple_strtoul(p, &endptr, 10);
> >> -	if (!*endptr)
> >> -		stack_guard_gap = val << PAGE_SHIFT;
> >> +	if (!str)
> >> +		return 0;
> > 
> > Please explain how this function can be called with a NULL pointer.
> 
> This is an additional check just in case. We have similar constructs
> in the following __setup() functions as well.

In case of _what_?  Somebody goes insane and decides to start calling
__setup functions with NULL pointers?  We don't test "Did the VFS call
this filesystem with a NULL inode pointer" because that would make
ZERO sense.  Defensive programming doesn't need to defend against an
insane kernel core.

> __setup("hashdist=", set_hashdist)
> __setup("numa_balancing=", setup_numabalancing)
> __setup("transparent_hugepage=", setup_transparent_hugepage)

Those should have this stupid NULL check removed.

> > Now you've removed the abillity for someone to say stack_guard_gap=0,
> > which seems potentially useful.
> 
> In that case, should the following two scenarios be differentiated ?
> 
> * stack_guard_gap=	- Retains DEFAULT_STACK_GUARD_GAP
> * stack_guard_gap=0	- Changes to 0 pages

I don't know.  You appear to have run into the scenario where
'stack_guard_gap=' was specified.  What did you expect it to do?