[PATCH] D96626: Support: mapped_file_region: Pass MAP_NORESERVE to mmap

[Joseph Tremoulet via Phabricator via llvm-commits]

JosephTremoulet added a subscriber: ruiu.
JosephTremoulet added a comment.

In D96626#2640715 <https://reviews.llvm.org/D96626#2640715>, @labath wrote:

> I haven't checked the manpage or the linux kernel behavior, but intuitively, I would expect that a shared mapping should not trigger an OOM, because at any point the kernel can decide to flush the dirty pages to disk (that's actually one technique for creating a per-process swap file).

I tried it out empirically and yes, I'm seeing success for a shared read/write mapping of a very large file without needing MAP_NORESERVE.  But the line in the man page saying "In kernels before 2.6, this flag had effect only for private writable mappings" was bugging me, so I went looking through kernel and manpage history.  What I found is that shared read/write mappings using "huge TLB pages" will do the reservation check (and this was added in 2.6) and thus can fail without MAP_NORESERVE.  And the docs for the huge TLB page feature [1] explain "Huge pages cannot be swapped out under memory pressure."  Since we're never passing MAP_HUGETLB in the mapped_file_region code, I believe this case is moot and it's only private read/write mappings that can have the prompt failure.

Perhaps @ruiu has thoughts about the linker's perspective? (though I realize LLD was brought up as just one example of a caller that might care about this flag)

Thanks.

1 - https://www.kernel.org/doc/Documentation/vm/hugetlbpage.txt


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D96626/new/

https://reviews.llvm.org/D96626