New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Fix race condition in feature cache on 32 platforms #837

Merged

alexcrichton merged 3 commits into rust-lang:master from matklad:fix-data-race

Jan 28, 2020

Member

matklad commented Jan 5, 2020

If we observe that the second word is initialized, we can't really
assume that the first is initialized as well. So check each word
separately.

I've found this by just looking at the code, and I didn't actually test this (cargo test --target i686-unknown-linux-gnu failed with some build.rs error which I haven't looked into).

rust-highfive assigned gnzlbg

rust-highfive commented Jan 6, 2020

Thanks for the pull request, and welcome! The Rust team is excited to review your changes, and you should hear from @gnzlbg (or someone else) soon.

If any changes to this PR are deemed necessary, please add them as extra commits. This ensures that the reviewer can see what has changed since they last reviewed the code. Due to the way GitHub handles out-of-date commits, this should also make it reasonably obvious what issues have or haven't been addressed. Large or tricky changes may require several passes of review and changes.

Please see the contribution instructions for more information.

matklad force-pushed the fix-data-race branch from e12a860 to 19fa496 Compare

January 6, 2020 00:00

matklad mentioned this pull request

spin-rs no longer maintained (dependency) briansmith/ring#921

Closed

Member Author

matklad commented Jan 6, 2020

Oh, this is another spin on #811 actually! But this version of the seems simpler.

Contributor

lu-zero commented Jan 27, 2020

Can you please try to rebase and see if now the tests pass?

matklad force-pushed the fix-data-race branch from 19fa496 to d7c4c0f Compare

January 27, 2020 09:49

Member Author

matklad commented Jan 27, 2020

rebased!

lu-zero approved these changes

View reviewed changes

Contributor

lu-zero left a comment

It looks fine to me, @gnzlbg and @alexcrichton do you agree?

alexcrichton reviewed

View reviewed changes

crates/std_detect/src/detect/cache.rs

-              /// Feature cache with capacity for `CACHE_CAPACITY` features.
+              /// This global variable is a cache of the features supported by the CPU.
+              #[cfg(target_pointer_width = "32")]
+              static CACHE: [Cache; 2] = [Cache::uninitialized(), Cache::uninitialized()];

Member

alexcrichton Jan 27, 2020

Instead of defining this differently on each platform, could this perhaps be unified a bit like so?

#[cfg(target_pointer_width = "64")]
 static CACHE: [Cache; 1] = [Cache::uninitialized()];
#[cfg(target_pointer_width = "32")]
 static CACHE: [Cache; 2] = [Cache::uninitialized(), Cache::uninitialized()];

I think that'd help cut down on the #[cfg] traffic

Member Author

matklad Jan 27, 2020

Awesome hint, thanks!

alexcrichton reviewed

View reviewed changes

crates/std_detect/src/detect/cache.rs Outdated Show resolved Hide resolved

matklad force-pushed the fix-data-race branch from d7c4c0f to 34b6e63 Compare

January 27, 2020 20:59

matklad commented

View reviewed changes

Member Author

matklad left a comment

Updated!

crates/std_detect/src/detect/cache.rs

-              /// Feature cache with capacity for `CACHE_CAPACITY` features.
+              /// This global variable is a cache of the features supported by the CPU.
+              #[cfg(target_pointer_width = "32")]
+              static CACHE: [Cache; 2] = [Cache::uninitialized(), Cache::uninitialized()];

Member Author

matklad Jan 27, 2020

Awesome hint, thanks!

crates/std_detect/src/detect/cache.rs Outdated Show resolved Hide resolved

matklad force-pushed the fix-data-race branch 3 times, most recently from 7125312 to 4418cda Compare

January 27, 2020 21:16

alexcrichton reviewed

View reviewed changes

crates/std_detect/src/detect/cache.rs Outdated

+              #[cfg(target_pointer_width = "32")]
+              const CACHE_BITS: u32 = 31;
+              #[cfg(target_pointer_width = "64")]
+              const CACHE_BITS: u32 = 63;

Member

alexcrichton Jan 28, 2020

Could this constant either be removed or defined in terms of mem::size_of::<usize>() instead of using #[cfg]?

Member Author

matklad Jan 28, 2020

Actually it could!

crates/std_detect/src/detect/cache.rs Outdated

@@ @@ -194,8 +171,18 @@ pub(crate) fn test<F>(bit: u32, f: F) -> bool @@
               where
                   F: FnOnce() -> Initializer,
               {
-                  if CACHE.is_uninitialized() {
-                      initialize(f());
+                  #[cfg(target_pointer_width = "32")]

Member

alexcrichton Jan 28, 2020

I think this #[cfg] could be avoided and the math here could be done with mem::size_of::<usize>() as well

crates/std_detect/src/detect/cache.rs Outdated

+              #[inline]
+              fn do_initialize(value: Initializer) {
+                  CACHE[0].initialize((value.0) as usize & CACHE_MASK);
+                  #[cfg(target_pointer_width = "32")]

Member

alexcrichton Jan 28, 2020

Like elsewhere, can the #[cfg] be removed here to work with shifts instead?

matklad force-pushed the fix-data-race branch 4 times, most recently from c978abc to b5eb5f9 Compare

January 28, 2020 13:47

matklad added 2 commits

January 28, 2020 14:48


          Fix race condition in feature cache on 32 platforms

2749c89

If we observe that the second word is initialized, we can't really
assume that the first is initialized as well. So check each word
separately.


          Use stronger atomic ordering

22231ea

Better SeqCst than sorry!

matklad force-pushed the fix-data-race branch from b5eb5f9 to 22231ea Compare

January 28, 2020 13:48

Member Author

matklad commented Jan 28, 2020

I've tried to get rid of cfgs using clever indexing, but unfortunately rustc was just smart enough to figure out that I try to index out of bounds into array and just dumb enough to not realize that it's dead code :)

In the last commit, I've just made both paths use two slots.


          Use two caches on x64 for simplicity

9b9cf9f

matklad force-pushed the fix-data-race branch from 65b65b8 to 9b9cf9f Compare

January 28, 2020 14:08

alexcrichton merged commit 9fa96fa into rust-lang:master

Member

alexcrichton commented Jan 28, 2020

👍

TethysSvensson mentioned this pull request

Soundness fix on 32-bit to cache.rs #811

Closed

matklad mentioned this pull request

minor improvements everywhere rust-analyzer/text-size#18

Merged

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet