On Thu, Feb 20, 2025 at 06:32:22PM +0800, Hillf Danton wrote:
> On Thu, 20 Feb 2025 14:20:01 +0900 Byungchul Park <byungchul@xxxxxx>
> > To check luf's stability, I ran a heavy LLM inference workload consuming
> > 210GiB over 7 days on a machine with 140GiB memory, and decided it's
> > stable enough.
> >
> > I'm posting the latest version so that anyone can try luf mechanism if
> > wanted by any chance. However, I tagged RFC again because there are
> > still issues that should be resolved to merge to mainline:
> >
> > 1. Even though system wide total cpu time for TLB shootdown is
> > reduced over 95%, page allocation paths should take additional cpu
> > time shifted from page reclaim to perform TLB shootdown.
> >
> > 2. We need luf debug feature to detect when luf goes wrong by any
> > chance. I implemented just a draft version that checks the sanity
> > on mkwrite(), kmap(), and so on. I need to gather better ideas
> > to improve the debug feature.
> >
> > ---
> >
> > Hi everyone,
> >
> > While I'm working with a tiered memory system e.g. CXL memory, I have
> > been facing migration overhead esp. tlb shootdown on promotion or
> > demotion between different tiers. Yeah.. most tlb shootdowns on
> > migration through hinting fault can be avoided thanks to Huang Ying's
> > work, commit 4d4b6d66db ("mm,unmap: avoid flushing tlb in batch if PTE
> > is inaccessible").
> >
> > However, it's only for migration through hinting fault. I thought it'd
> > be much better if we have a general mechanism to reduce all the tlb
> > numbers that we can apply to any unmap code, that we normally believe
> > tlb flush should be followed.
> >
> > I'm suggesting a new mechanism, LUF(Lazy Unmap Flush), that defers tlb
> > flush until folios that have been unmapped and freed, eventually get
> > allocated again. It's safe for folios that had been mapped read-only
> > and were unmapped, as long as the contents of the folios don't change
> > while staying in pcp or buddy so we can still read the data through the
> > stale tlb entries.
> >
> Given pcp or buddy, you are opening window for use after free which makes
> no sense in 99% cases.
Just in case that I don't understand what you meant and for better
understanding, can you provide a simple and problematic example from
the u-a-f?
Byungchul
