Use pointers in `cell::{Ref,RefMut}` to avoid `noalias` by cuviper · Pull Request #97027 · rust-lang/rust

cuviper · 2022-05-13T19:09:32Z

When Ref and RefMut were based on references, they would get LLVM noalias attributes that were incorrect, because that alias guarantee is only true until the guard drops. A &RefCell on the same value can get a new borrow that aliases the previous guard, possibly leading to miscompilation. Using NonNull pointers in Ref and RefCell avoids noalias.

Fixes the library side of #63787, but we still might want to explore language solutions there.

rust-highfive · 2022-05-13T19:09:36Z

r? @thomcc

(rust-highfive has picked a reviewer for you, use r? to override)

CAD97

Ref doesn't need a PhantomData<&T> for correctness, but it might make sense to include it for explicitness / symmetry with RefMut?

Also, I forget how NonNull acts w.r.t. UnwindSafe; this is a reminder to someone to check that it's not changed by this PR.

RalfJung · 2022-05-13T21:26:34Z

 where
 F: FnOnce(&mut T) -> Option<&mut U>,
 {
 // FIXME(nll-rfc#40): fix borrow-check


I assume this no longer applies now?

Maybe? I'm not really sure what needed fixing -- was it that orig had to be destructured, then rebuilt in the Err case?

cuviper · 2022-05-13T23:04:00Z

@CAD97 :

Ref doesn't need a PhantomData<&T> for correctness, but it might make sense to include it for explicitness / symmetry with RefMut?

I could just add a comment addressing covariance, but are there more effects you want from PhantomData?

Also, I forget how NonNull acts w.r.t. UnwindSafe; this is a reminder to someone to check that it's not changed by this PR.

All of the auto traits are unchanged, at least according to rustdoc. NonNull does implement UnwindSafe for T: RefUnwindSafe, but Ref and RefMut do not.

thomcc · 2022-05-14T04:01:24Z

I think this is fine as-is, no need for additional PhantomData.

@bors r+

bors · 2022-05-14T04:01:26Z

📌 Commit 15d8c00 has been approved by thomcc

Use pointers in `cell::{Ref,RefMut}` to avoid `noalias` When `Ref` and `RefMut` were based on references, they would get LLVM `noalias` attributes that were incorrect, because that alias guarantee is only true until the guard drops. A `&RefCell` on the same value can get a new borrow that aliases the previous guard, possibly leading to miscompilation. Using `NonNull` pointers in `Ref` and `RefCell` avoids `noalias`. Fixes the library side of rust-lang#63787, but we still might want to explore language solutions there.

bors · 2022-05-15T10:46:32Z

⌛ Testing commit 15d8c00 with merge 0b2b6a751bfaaef94c621925159a481da0356b9d...

bors · 2022-05-15T11:19:23Z

💔 Test failed - checks-actions

cuviper · 2022-05-17T00:40:08Z

I took a stab at fixing natvis, but I'm just guessing...

thomcc · 2022-05-19T15:34:54Z

(I'm assuming I need to r+ this again)

@bors r+

bors · 2022-05-19T15:34:56Z

📌 Commit 1c3921f has been approved by thomcc

bors · 2022-05-19T15:34:57Z

🌲 The tree is currently closed for pull requests below priority 1000. This pull request will be tested once the tree is reopened.

bors · 2022-05-20T01:05:56Z

⌛ Testing commit 1c3921f with merge 4d6992b...

bors · 2022-05-20T03:26:56Z

☀️ Test successful - checks-actions
Approved by: thomcc
Pushing 4d6992b to master...

rust-highfive · 2022-05-20T03:27:25Z

📣 Toolstate changed by #97027!

Tested on commit 4d6992b.
Direct link to PR: #97027

💔 miri on windows: test-pass → test-fail (cc @oli-obk @eddyb @RalfJung).
💔 miri on linux: test-pass → test-fail (cc @oli-obk @eddyb @RalfJung).

@oli-obk

Tested on commit rust-lang/rust@4d6992b. Direct link to PR: <rust-lang/rust#97027> 💔 miri on windows: test-pass → test-fail (cc @oli-obk @eddyb @RalfJung). 💔 miri on linux: test-pass → test-fail (cc @oli-obk @eddyb @RalfJung).

rust-timer · 2022-05-20T05:15:04Z

Finished benchmarking commit (4d6992b): comparison url.

Instruction count

This benchmark run did not return any relevant results for this metric.

Max RSS (memory usage)

Results

Primary benchmarks: 😿 relevant regression found
Secondary benchmarks: mixed results

	Regressions 😿 (primary)	Regressions 😿 (secondary)	Improvements 🎉 (primary)	Improvements 🎉 (secondary)	All 😿 🎉 (primary)
count¹	1	1	0	2	1
mean²	0.1%	3.3%	N/A	-3.1%	0.1%
max	0.1%	3.3%	N/A	-3.2%	0.1%

Cycles

Results

Primary benchmarks: no relevant changes found
Secondary benchmarks: mixed results

	Regressions 😿 (primary)	Regressions 😿 (secondary)	Improvements 🎉 (primary)	Improvements 🎉 (secondary)	All 😿 🎉 (primary)
count¹	0	1	0	2	0
mean²	N/A	1.7%	N/A	-2.4%	N/A
max	N/A	1.7%	N/A	-3.0%	N/A

If you disagree with this performance assessment, please file an issue in rust-lang/rustc-perf.

@rustbot label: -perf-regression

number of relevant changes↩↩²
the arithmetic mean of the percent change↩↩²

RalfJung · 2022-05-20T06:07:48Z

The Miri tests fail because this affects the Display output of Ref and RefMut: they used to print the value, now they print the address of the RefCell. That is probably a bug?

rustup `Display` of `Ref`/`RefMut` was broken by rust-lang/rust#97027, let's deref them to use the underlying reference `Display`. Cc rust-lang/rust#97204

cuviper · 2022-05-20T18:05:50Z

Ah, yes, I'll fix that. The Debug impls use Deref, but Display is directly reaching for self.value (now a pointer).

These guards changed to pointers in rust-lang#97027, but their `Display` was formatting that field directly, which made it show the raw pointer value. Now we go through `Deref` to display the real value again.

Fix `Display` for `cell::{Ref,RefMut}` These guards changed to pointers in rust-lang#97027, but their `Display` was formatting that field directly, which made it show the raw pointer value. Now we go through `Deref` to display the real value again. Miri noticed this change, rust-lang#97204, so hopefully that will be fixed.

cuviper added 3 commits May 13, 2022 11:42

Use a pointer in cell::Ref so it's not noalias
d369045

Use a pointer in cell::RefMut so it's not noalias
2b8041f

Test RefCell aliasing
15d8c00

This comment was marked as resolved.
Sign in to view

rust-highfive assigned thomcc May 13, 2022

rustbot added T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. T-libs Relevant to the library team, which will review and decide on the PR/issue. labels May 13, 2022

rust-highfive added the S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. label May 13, 2022

cuviper mentioned this pull request May 13, 2022
Ref parameter incorrectly decorated with noalias attribute #63787
Closed

CAD97 reviewed May 13, 2022
View reviewed changes

RalfJung reviewed May 13, 2022
View reviewed changes

bors added S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels May 14, 2022

This comment was marked as spam.
Sign in to view

GuillaumeGomez mentioned this pull request May 14, 2022
Rollup of 4 pull requests #97050
Closed

Dylan-DPC mentioned this pull request May 15, 2022
Rollup of 4 pull requests #97054
Closed

bors added S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. and removed S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. labels May 15, 2022

cuviper added 3 commits May 16, 2022 17:22

Remove outdated references to nll-rfc#40
1e53fab

Add a comment for covariant Ref
1f33c92

Read the Ref/RefMut pointer in natvis
1c3921f

bors added S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels May 19, 2022

bors added the merged-by-bors This PR was explicitly merged by bors. label May 20, 2022

bors merged commit 4d6992b into rust-lang:masterMay 20, 2022

rustbot added this to the 1.63.0 milestone May 20, 2022

rust-highfive mentioned this pull request May 20, 2022
miri no longer builds after rust-lang/rust#97027 #97204
Closed

RalfJung mentioned this pull request May 20, 2022
rustup rust-lang/miri#2130
Merged

bors added a commit to rust-lang/miri that referenced this pull request May 20, 2022
Auto merge of#2130- RalfJung:rustup, r=RalfJung
d419efd
rustup `Display` of `Ref`/`RefMut` was broken by rust-lang/rust#97027, let's deref them to use the underlying reference `Display`. Cc rust-lang/rust#97204

cuviper mentioned this pull request May 20, 2022
Fix Display for cell::{Ref,RefMut}#97225
Merged

m-ou-se mentioned this pull request Jun 11, 2022
Make RwLockReadGuard covariant #96820
Merged

cuviper deleted the yesalias-refcell branch October 15, 2022 00:16

Ten0 mentioned this pull request Nov 23, 2022
Possible approach to fix #71 Kimundi/owning-ref-rs#72
Draft

yvt mentioned this pull request Nov 23, 2022
Stop storing a reference in a lock guard r3-os/r3#15
Open

Imberflur mentioned this pull request Feb 4, 2023
Use NonNull pointer in the place of references in AtomicRef and AtomicRefMut to avoid noalias related soundness bug. mozilla/atomic_refcell#18
Merged

Uh oh!

Uh oh!

Conversation

cuviper commented May 13, 2022

Uh oh!

This comment was marked as resolved.

rust-highfive commented May 13, 2022

Uh oh!

CAD97 left a comment

Choose a reason for hiding this comment

Uh oh!

RalfJungMay 13, 2022

Choose a reason for hiding this comment

Uh oh!

cuviperMay 13, 2022

Choose a reason for hiding this comment

Uh oh!

cuviper commented May 13, 2022

Uh oh!

thomcc commented May 14, 2022

Uh oh!

bors commented May 14, 2022

Uh oh!

This comment was marked as spam.

This comment was marked as spam.

This comment was marked as spam.

bors commented May 15, 2022

Uh oh!

bors commented May 15, 2022

Uh oh!

cuviper commented May 17, 2022

Uh oh!

thomcc commented May 19, 2022

Uh oh!

bors commented May 19, 2022

Uh oh!

bors commented May 19, 2022

Uh oh!

bors commented May 20, 2022

Uh oh!

bors commented May 20, 2022

Uh oh!

rust-highfive commented May 20, 2022

Uh oh!

rust-timer commented May 20, 2022

Instruction count

Max RSS (memory usage)

Cycles

Footnotes

Uh oh!

RalfJung commented May 20, 2022

Uh oh!

cuviper commented May 20, 2022

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

11 participants