Re: [PATCH] mm: vmscan: Avoid signedness error for GCC 5.4

[David Laight]

On Wed, 7 May 2025 12:06:11 +0800
WangYuli <wangyuli@xxxxxxxxxxxxx> wrote:

> Hi Andrew Morton,
> 
> On 2025/5/7 07:24, Andrew Morton wrote:
> > On Wed,  7 May 2025 00:02:38 +0800 WangYuli <wangyuli@xxxxxxxxxxxxx> wrote:
> >
> > Make `tier' unsigned?  After all, negative tier numbers are nonsensical.  
> 
> That point is well taken.
> 
> However, I've noticed that variables named "tier" seem to be commonly 
> defined as int rather than unsigned int throughout the mm subsystem, and 
> perhaps even the wider kernel code.
> 
> I was wondering if changing just this one instance might feel a little 
> inconsistent?
> 
> Perhaps a possible approach for now could be to change this line to for 
> (i = tier % MAX_NR_TIERS; i <= min_t(int, tier, MAX_NR_TIERS - 1); i++) 
> {, which would allow us to keep the signed int type for the tier 
> variable itself.

Remember that min_t(int, a, b) is just min((int)a, (int)b) and you really
wouldn't put casts like that in code.
Even if a cast can't be avoided only one side would normally need it.

There really ought to be a 'duck shoot' against min_t().
I'm trying very hard to stop any more being added.

	David

> 
> Regarding the potential for a more comprehensive change in the future to 
> redefine all these "tier" variables and related ones as unsigned int, I 
> would be very grateful for your guidance on whether that's a direction 
> we should consider.
> 
> But actually, whether it's signed or not likely won't affect its normal 
> operation...
> 
> Thanks,
>