On Tue, Mar 30, 2021 at 09:58:37AM +0200, David Hildenbrand wrote:

On 29.03.21 22:12, Minchan Kim wrote:

quoted

On Mon, Mar 29, 2021 at 07:44:31PM +0100, Matthew Wilcox wrote:

quoted

On Mon, Mar 29, 2021 at 11:25:53AM -0700, Minchan Kim wrote:

quoted

size_t in cma_alloc is confusing since it makes people think
it's byte count, not pages. Fix it.

i think it has to be unsigned long.

67a2e213e7e937c41c52ab5bc46bf3f4de469f6e

Right.

Fortunately, we don't have such large allocations yet via
CMA/alloc_contig_range

quoted

Thanks for the pinter. I wanted to have the smallest change.
The commit leads me to change cma_release, trace_cma_alloc,
cma_clear_bitmap and front_contig_range as well.(Not sure
we have more. Will check).

Ccing david@redhat.com for upcoming changing free_contig_range.

While at it, we might want to convert free_contig_range() to eat
"unsigned long start, unsigned long end" like alloc_contig_range(), instead
of "unsigned long pfn, unsigned int nr_pages" like alloc_contig_pages() ...

Well, I personllay tempted to change alloc_contig_range, not
free_contig_range because base_pfn with nr_pages was more
straightforward than base_pfn and end_pfn in that we don't
need to tell whether end_pfn is inclusive or exclusive.

When I look at callers of [alloc|free]_contig_range, many of them
already have used nr_pages based approach rather than start_pfn,
end_pfn. If your suggestion come from that "it's *range* API",
I'd like to rename it with "alloc_contig_pages|free_contig_pages".

Since it's beyond the goal of this patch and might be controversial, 
I will not deal with it in this patch.