On Mon, Jan 16, 2023 at 12:15:33PM -0800, Andrew Morton wrote:
> On Mon, 16 Jan 2023 19:39:00 +0000 Jiaqi Yan <jiaqiyan@xxxxxxxxxx> wrote:
>
> > +/*
> > + * Per NUMA node memory failure handling statistics.
> > + */
> > +struct memory_failure_stats {
> > + /*
> > + * Number of pages poisoned.
> > + * Cases not accounted: memory outside kernel control, offline page,
> > + * arch-specific memory_failure (SGX), and hwpoison_filter()
> > + * filtered error events.
> > + */
> > + unsigned long pages_poisoned;
> > + /*
> > + * Recovery results of poisoned pages handled by memory_failure,
> > + * in sync with mf_result.
> > + * pages_poisoned = pages_ignored + pages_failed +
> > + * pages_delayed + pages_recovered
> > + */
> > + unsigned long pages_ignored;
> > + unsigned long pages_failed;
> > + unsigned long pages_delayed;
> > + unsigned long pages_recovered;
> > +};
>
> I don't think the "pages_" here add much value - a simple code comment
> saying "everything counts in pages" should suffice. If you're feeling
> deprived of columnar space, maybe just remove? Or not ;)
I personally feel more like removing the "pages_". And I feel that the
main counter can be renamed with pages_total or total, because it show the
relationship among counters, and the structure name "memory_failure_stats"
already implies that the counter is about hwpoisoned pages.
Thanks,
Naoya Horiguchi
