[PATCH] D77330: Consider increasing the default ARM32 page size to 64k.

Fri Apr 3 01:02:31 PDT 2020

psmith added reviewers: MaskRay, grimar, ruiu.
psmith added a comment.

Some references:
The GNU ld change to 64k max page size was made in 2014 https://patches.linaro.org/patch/32565/ [RFC] ld/ARM: Increase maximum page size to 64kB
My understanding from conversations at Arm and Linaro was that this was made primarily for the benefit of people running AArch32 executables on a AArch64 kernel built for 64k page sizes. In practice I don't think there ended up being a widely used distribution that simultaneously used 64K page size and offered a 32-bit user space. Ubuntu's decision to use 4k citing compatibility with old Arm v7 binaries https://wiki.ubuntu.com/ARM64/performance

The change to 64k page size for AArch64 in LLD D25079 <https://reviews.llvm.org/D25079>
Android driver forcing 4k page size in AArch64 D55029 <https://reviews.llvm.org/D55029>
Small number of Arm systems using non 4k page size in context of Gold https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=844467

As mentioned on llvm-dev the tests will need updating, I'm sorry this will be a rather tedious job.

My understanding is that Android and ChromeOs are the largest users of LLD on Arm that are most likely to be hurt by any image size changes. It is true that the change to adding a max-page-size in virtual memory rather than aligning to the next max-page-size will have helped mitigate this a bit, but I think we may still see an increase either side of RELRO. I'd very much like to hear from these folks to see if this hurts them, or if they are willing to force 4k alignment in the clang driver and if so how much time do they need to prepare for it?

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D77330/new/

https://reviews.llvm.org/D77330