Re: [RFC 1/3] mm: add thp_utilization metrics to debugfs

[Alex Zhu (Kernel)]

> On Aug 26, 2022, at 5:11 PM, Zi Yan <ziy@xxxxxxxxxx> wrote:
> 
> On 25 Aug 2022, at 17:30, alexlzhu@xxxxxx wrote:
> 
> How large is the memory? Just wonder the scanning speed.
> Also, it might be better to explicitly add the time unit, second,
> in the output.

The size of memory was 65GB on the test machine I obtained these 
numbers on. I’ll take note of adding the time unit. Thanks!


> Is it possible to use cache-bypassing read to avoid cache
> pollution? You are scanning for 256*2M at a time. Wouldn’t that
> wipe out all the useful data in the cache?

I have only found non-temporal writes in arch/x86/, not non-temporal reads (with MOVNTDQA). I suppose we should figure out why nobody ever bothered using non-temporal reads on x86 before trying to make this code use them.

A quick search of the internet suggests they non-temporal reads are not being used on x86 because people could not show a performance improvement by using them, but maybe somebody here has more insight?