1391636 - We can support wasm on a platform only if it has 8-byte atomics

Status: RESOLVED FIXED

(Core :: JavaScript Engine: JIT, enhancement)

Description

[Lars T Hansen [:lth]]

8-byte atomics are required to be lock free, and 8-byte racy load and store are required to be access-atomic.  On ARM, the atomics require LDREXD and STREXD, which are supported from ARMv6K and forward, this should not be a hardship.  The loads and stores can be emulated with LDREXD+CLREX and LDREXD+STREXD, respectively, but it's better to use LDRD and STRD if they are access-atomic.  They are access-atomic if the Large Physical Address Extension is present, which it is in some ARMv7-A implementations and I believe in all ARMv8 aarch32 implementations.  It's possible to test for it.  But it is in any case subsumbed by the test for LDREXD and STREXD as far as testing for lock-freedom goes; it will simply be an optimization to use LDRD and STRD for access-atomic loads and stores when possible.

Anyhow, wasm::hasCompilerSupport() should call jit::AtomicOperations::isLockfree8() and only return true if the latter returns true.

Comment 1

[Benjamin Bouvier [:bbouvier] (inactive)]

Would it make sense to have this condition in wasmSupportsThreads() instead, and merge it into wasm::hasCompilerSupport only when the feature rides the trains?

Comment 2

[Lars T Hansen [:lth]]

(In reply to Benjamin Bouvier [:bbouvier] from comment #1)
> Would it make sense to have this condition in wasmSupportsThreads() instead,
> and merge it into wasm::hasCompilerSupport only when the feature rides the
> trains?

Yes, that seems meaningful.

Comment 3

[Lars T Hansen [:lth]]

Attachment: bug1391636-gate-on-lockfree8.patch

Comment 4

[Benjamin Bouvier [:bbouvier] (inactive)]

Comment on attachment 8898829 [details] [diff] [review]
bug1391636-gate-on-lockfree8.patch

Review of attachment 8898829 [details] [diff] [review]:
-----------------------------------------------------------------

Looks good, thanks.

Comment 5

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/mozilla-inbound/rev/9bf5648598b0
gate the availability of wasm threads on 8-byte lock-free atomics. r=bbouvier

Comment 6

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/mozilla-inbound/rev/e013d1324ea8
make some locals DEBUG-only.  r=me

Comment 7

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

https://hg.mozilla.org/mozilla-central/rev/9bf5648598b0
https://hg.mozilla.org/mozilla-central/rev/e013d1324ea8

Comment 8

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/mozilla-inbound/rev/2d6227004721
Move a guard from TestingFunctions into Wasm proper. r=me

Comment 9

[Stefan Hindli [:stefan_hindli]]

https://hg.mozilla.org/mozilla-central/rev/2d6227004721