[PATCH] D93731: scudo: Support memory tagging in the secondary allocator.

[Christopher Ferris via Phabricator via llvm-commits]

cferris requested changes to this revision.
cferris added a comment.
This revision now requires changes to proceed.

Sorry it took a long time to get back to this. Unfortunately, not only does this cause relatively large performance regressions, it also causes test failures. The test failures are related to malloc_iterate, but I haven't looked too closely at them yet. Is it possible that something needs to be updated in the code related to tracking the total allocated memory?

Benchmarks show as high as 10%, but many closer to 5% regression in other benchmarks.

As you mentioned, there are some large size calloc/mallocs that are 10x slower, for the 128KB sizes. These are all of the microbenchmarks, but even the trace benchmarks show large performance issues.  I see 15% to 20% regressions on a few of these.

Also, this increases the RSS significantly on some of the traces. For example, the 32 bit camera trace that does a lot of large allocations went from ~50MB of RSS to ~55MB. It also increased the VA space on a few of the traces. For example, the maps 32 bit trace goes from ~38MB to ~46.5MB. We already have issues in 32 bit land with taking too much VA space, this would likely cause additional issues.

If there is a way to enable/disable this that would be great, otherwise, the regression is too large on almost every axis to take. Even with a possible kernel patch, I don't know if that will fix the RSS and VA space issues.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D93731/new/

https://reviews.llvm.org/D93731