1024805 - Use tagged anonymous memory for Nuwa thread stacks

Status: RESOLVED FIXED

(Core :: DOM: Content Processes, defect)

Description

[Jed Davis [:jld] ⟨⏰|UTC-8⟩ ⟦he/him⟧]

Currently we malloc() Nuwa thread stacks, meaning that the entire virtual memory area containing the stack — and probably a lot of unrelated malloc'ed data — is marked as being the stack of one of the threads in question in procfs maps files.

Also currently, we consider thread stacks as untagged anonymous memory in the system memory reporter, meaning that a bunch of the jemalloc heap will be misreported after bug 1011350 lands.  (The name is available in that case, but I didn't want to complicate the parser to extract it when we could fix this problem directly instead.)

If we mmap Nuwa stacks directly with the tagged anonymous memory functionality from bug 1011350, then this can't happen: a thread stack can't be merged with anything else (it has a different tag), and the guard page (different protection) prevents two adjacent stacks from being merged.

Regular thread stacks aren't quite there unless/until we patch bionic (which already has an internal interface to PR_SET_VMA_ANON_NAME, and uses it for its own malloc), but empirically these stacks seem to be followed by jemalloc or something else that isn't also r/w and untagged and anonymous.

So, with this fixed we might be able to (if tagged anonymous memory is supported) report non-main thread stacks as such.  It would even be possible to obtain the thread name (probably make sense only for Gecko processes) from procfs.

Comment 1

[Jed Davis [:jld] ⟨⏰|UTC-8⟩ ⟦he/him⟧]

Attachment: bug1024805-nuwa-named-stacks-hg0.diff

Trying: https://tbpl.mozilla.org/?tree=Try&rev=bf8c703c7b5d

Also tested locally on Flame.

Comment 2

[Cervantes Yu [:cyu] [:cervantes]]

Comment on attachment 8442518 [details] [diff] [review]
bug1024805-nuwa-named-stacks-hg0.diff

Review of attachment 8442518 [details] [diff] [review]:
-----------------------------------------------------------------

Also thanks for the bonus fixes :)

Comment 3

[Jed Davis [:jld] ⟨⏰|UTC-8⟩ ⟦he/him⟧]

The oranges on the try run in comment #1 seem to not be my fault.

Comment 4

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/20c915f8f762

Comment 5

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/20c915f8f762