Re: [PATCH v3 1/4] x86/clear_page: extend clear_page*() for multi-page clearing

[Ankur Arora <ankur.a.arora@oracle.com>]

Matthew Wilcox <willy@infradead.org> writes:

> On Mon, Apr 14, 2025 at 12:52:37PM -0700, Ankur Arora wrote:
>> Ingo Molnar <mingo@kernel.org> writes:
>> >> +void clear_pages_orig(void *page, unsigned int length);
>> >> +void clear_pages_rep(void *page, unsigned int length);
>> >> +void clear_pages_erms(void *page, unsigned int length);
>> >
>> > What unit is 'length' in? If it's bytes, why is this interface
>> > artificially limiting itself to ~4GB? On x86-64 there's very little (if
>>
>> I was in two minds about the unit. Given that the largest page size is
>> 1GB, decided to go with 32bit. But, as you say below, there's no reason
>> to limit the x86-64 interface for MM reasons.  Will fix.
>
> Actually, I think there is (and we went through this with SPARC, if you
> remember?)

My google-fu is failing me. I don't think it was this thread:
 https://lore.kernel.org/lkml/1490310113-824438-1-git-send-email-pasha.tatashin@oracle.com/

> We _shouldn't_ be calling memset() with a large size (ie
> larger than 4GB).  If we have that much memory to clear, we should be
> doing something smarter, like using padata to get lots of CPUs clearing
> individual portions of the page.

Agreed. Or even offloading to an accelerator so as to not waste CPU time.

That said, whether to invoke clear_pages() in > 4GB seems like an MM
policy question. Not sure it makes sense to limit the low level interface.

> I don't know how relevant this is now that you're going to be using
> ALTERNATIVES.


--
ankur