Re: [PATCH v2 1/5] mm: introduce num_pages_contiguous()

[lizhe.67@bytedance.com]

On Fri, 4 Jul 2025 14:10:15 -0300, jgg@ziepe.ca wrote:

> > diff --git a/include/linux/mm.h b/include/linux/mm.h
> > index 0ef2ba0c667a..1d26203d1ced 100644
> > --- a/include/linux/mm.h
> > +++ b/include/linux/mm.h
> > @@ -205,6 +205,26 @@ extern unsigned long sysctl_admin_reserve_kbytes;
> >  #define folio_page_idx(folio, p)	((p) - &(folio)->page)
> >  #endif
> >  
> > +/*
> > + * num_pages_contiguous() - determine the number of contiguous pages
> > + * starting from the first page.
> > + *
> > + * @pages: an array of page pointers
> > + * @nr_pages: length of the array
> > + */
> > +static inline unsigned long num_pages_contiguous(struct page **pages,
> > +						 unsigned long nr_pages)
> 
> Both longs should be size_t I think

Yes, size_t is a better choice.

> > +{
> > +	struct page *first_page = pages[0];
> > +	unsigned long i;
> 
> Size_t
> 
> > +
> > +	for (i = 1; i < nr_pages; i++)
> > +		if (pages[i] != nth_page(first_page, i))
> > +			break;
> 
> It seems OK. So the reasoning here is this is faster on
> CONFIG_SPARSEMEM_VMEMMAP/nonsparse

Yes.

> and about the same on sparse mem?
> (or we don't care?)

Regarding sparse memory, I'm not entirely certain. From my
understanding, VFIO is predominantly utilized in virtualization
scenarios, which typically have sufficient kernel resources. This
implies that CONFIG_SPARSEMEM_VMEMMAP is generally set to "y" in
such cases. Therefore, we need not overly concern ourselves with
this particular scenario. Of course, David has also proposed
optimization solutions for sparse memory scenarios[1]. If anyone
later complains about performance in this context, I would be happy
to assist with further optimization efforts. Currently, I only have
a x86_64 machine, on which CONFIG_SPARSEMEM_VMEMMAP is forcibly
enabled. Attempting to compile with CONFIG_SPARSEMEM &&
!CONFIG_SPARSEMEM_VMEMMAP results in compilation errors, preventing
me from conducting the relevant performance tests.

Thanks,
Zhe

[1]: https://lore.kernel.org/all/c1144447-6b67-48d3-b37c-5f1ca6a9b4a7@redhat.com/#t