Support unwinding after a panic #693

Aaron1011 · 2019-04-17T01:12:30Z

Fixes #658

This commit adds support for unwinding after a panic. It requires a
companion rustc PR to be merged, in order for the necessary hooks to
work properly.

Currently implemented:

Selecting between unwind/abort mode based on the rustc Session
Properly popping off stack frames, unwinding back the caller
Running 'unwind' blocks in Mir terminators

Not yet implemented:

'Abort' terminators

This PR was getting fairly large, so I decided to open it for review without
implementing 'Abort' terminator support. This could either be added on
to this PR, or merged separately.

I've a test to exercise several different aspects of unwind panicking. Ideally, we would run Miri against the libstd panic tests, but I haven't yet figured out how to do that.

This depends on rust-lang/rust#60026

src/fn_call.rs

oli-obk · 2019-04-17T08:24:40Z

Please also add a test that check whether double panics cause an abort and don't confuse miri. Something like

struct Foo;
impl Drop for Foo {
    fn drop(&mut self) {
        panic!("second");
    }
}
fn main() {
    let foo = Foo;
    panic!("first");
}

Aaron1011 · 2019-04-17T19:32:39Z

@oli-obk: Double-panics are handled correctly, but end up causing an error due to uw::_Unwind_Backtrace being unhandled. This is due to the fact that when a double panic occurs, libstd unconditionally tries to print a backtrace.

Should we look into building libstd without the backtrace feature, or should we try to implement it (e.g adding a cfg(mir) into libstd's backtrace code?

oli-obk · 2019-04-18T07:56:13Z

Should we look into building libstd without the backtrace feature

That seems doable since we're building a libstd anyway. But you don't have to do that in this PR, adding the test so we see the failure is enough for me right now.

src/fn_call.rs

src/lib.rs

tests/run-pass/catch_panic.rs

RalfJung · 2019-04-29T14:14:14Z

Running 'unwind' blocks in Mir terminators

Where is that happening? The cleanup field of the Call terminator still seems to be ignored, and that's where things should have to jump to during unwinding.

Aaron1011 · 2019-04-29T22:37:07Z

@RalfJung: 'unwind' processing is handled here

src/fn_call.rs

Aaron1011 · 2019-05-30T03:12:16Z

@RalfJung: I've updated this PR (and the corresponding rustc PR) to move more of the unwinding logic into rustc. Now, the implementation doesn't manually step through box_me_up - it's handled via the normal main loop.

src/fn_call.rs

src/lib.rs

src/fn_call.rs

src/lib.rs

tests/run-pass/catch_panic.rs

Aaron1011 · 2019-11-17T17:28:26Z

This is going to require the ability to read wide strings, which appears to be currently being worked on in rust-lang/rust#66470.

@RalfJung: Is there any chance that we could just disable unwinding tests on Windows for now? Getting this working is going to require rust-lang/rust#66470 to be merged, after which we will need to implement GetModuleHandleW and GetProcAddress in a manner analogous to the current dlsym handling.

RalfJung · 2019-11-17T18:08:06Z

Is there any chance that we could just disable unwinding tests on Windows for now?

Seems fine to me, if you open an issue.
I also seem to recall that the reason RwLock uses GetProcAddress is to support Windows XP or so, but support for that was since dropped from Rust... so I think (but you might want to confirm that with some Rust team members) that we could patch RwLock on Windows to statically link against the platform methods, and not require GetProcAddress any more.

Miri currently does not support `GetProcAddress` and `GetModuleHandleW`, both of which end up getting invoked by the libstd panic hook.

Aaron1011 · 2019-11-17T18:52:07Z

@RalfJung; I've opened #1059

src/helpers.rs

src/shims/foreign_items.rs

RalfJung · 2019-11-19T13:53:03Z

@Aaron1011 I pushed a commit with a bunch of comments, but also some simplifications in panic.rs. Could you take a look whether the comments all still seem correct to you?

Then this is ready to land IMO, modulo my question about __rust_start_panic above.

@oli-obk any comments/concerns?

Aaron1011 · 2019-11-19T15:13:16Z

@RalfJung: This should now be ready to merge.

src/shims/foreign_items.rs

src/shims/panic.rs

Aaron1011 · 2019-11-19T20:33:44Z

@RalfJung: I've fixed the two comments you mentioned.

oli-obk · 2019-11-19T20:39:13Z

@bors r+

bors · 2019-11-19T20:39:14Z

📌 Commit 2750f60 has been approved by oli-obk

bors · 2019-11-19T20:39:21Z

⌛ Testing commit 2750f60 with merge e64caeb...

Support unwinding after a panic Fixes #658 This commit adds support for unwinding after a panic. It requires a companion rustc PR to be merged, in order for the necessary hooks to work properly. Currently implemented: * Selecting between unwind/abort mode based on the rustc Session * Properly popping off stack frames, unwinding back the caller * Running 'unwind' blocks in Mir terminators Not yet implemented: * 'Abort' terminators This PR was getting fairly large, so I decided to open it for review without implementing 'Abort' terminator support. This could either be added on to this PR, or merged separately. I've a test to exercise several different aspects of unwind panicking. Ideally, we would run Miri against the libstd panic tests, but I haven't yet figured out how to do that. This depends on rust-lang/rust#60026

bors · 2019-11-19T20:49:52Z

💔 Test failed - status-appveyor

We should re-enable this once rust-lang#1059 is fixed

Aaron1011 · 2019-11-19T21:00:32Z

@oli-obk: I had forgotten to disable one of the panic tests on Windows. Can you re-try the build?

oli-obk · 2019-11-19T21:07:22Z

@bors r+

bors · 2019-11-19T21:07:23Z

📌 Commit 6fe89e4 has been approved by oli-obk

bors · 2019-11-19T21:07:30Z

⌛ Testing commit 6fe89e4 with merge e588d95...

Support unwinding after a panic Fixes #658 This commit adds support for unwinding after a panic. It requires a companion rustc PR to be merged, in order for the necessary hooks to work properly. Currently implemented: * Selecting between unwind/abort mode based on the rustc Session * Properly popping off stack frames, unwinding back the caller * Running 'unwind' blocks in Mir terminators Not yet implemented: * 'Abort' terminators This PR was getting fairly large, so I decided to open it for review without implementing 'Abort' terminator support. This could either be added on to this PR, or merged separately. I've a test to exercise several different aspects of unwind panicking. Ideally, we would run Miri against the libstd panic tests, but I haven't yet figured out how to do that. This depends on rust-lang/rust#60026

bors · 2019-11-19T21:32:16Z

☀️ Test successful - checks-travis, status-appveyor
Approved by: oli-obk
Pushing e588d95 to master...

RalfJung · 2019-11-19T21:34:00Z

Finally, after around half a year, this landed. Thanks @Aaron1011, this is awesome!

Aaron1011 mentioned this pull request Apr 17, 2019

Add hooks for Miri panic unwinding rust-lang/rust#60026

Merged

oli-obk reviewed Apr 17, 2019

View reviewed changes

RalfJung reviewed Apr 29, 2019

View reviewed changes

src/fn_call.rs Outdated Show resolved Hide resolved

RalfJung reviewed Apr 29, 2019

View reviewed changes

src/fn_call.rs Outdated Show resolved Hide resolved

RalfJung reviewed Apr 29, 2019

View reviewed changes

src/fn_call.rs Outdated Show resolved Hide resolved

RalfJung reviewed Apr 29, 2019

View reviewed changes

src/fn_call.rs Outdated Show resolved Hide resolved

RalfJung reviewed Apr 29, 2019

View reviewed changes

src/lib.rs Outdated Show resolved Hide resolved

RalfJung reviewed Apr 29, 2019

View reviewed changes

src/lib.rs Outdated Show resolved Hide resolved

RalfJung reviewed Apr 29, 2019

View reviewed changes

src/lib.rs Outdated Show resolved Hide resolved

RalfJung reviewed Apr 29, 2019

View reviewed changes

tests/run-pass/catch_panic.rs Outdated Show resolved Hide resolved

Aaron1011 force-pushed the feature/panic_unwind_final branch from 1d26324 to 4893036 Compare April 30, 2019 02:02

RalfJung reviewed May 5, 2019

View reviewed changes

src/fn_call.rs Outdated Show resolved Hide resolved

Aaron1011 force-pushed the feature/panic_unwind_final branch 2 times, most recently from 859366f to 6d195bc Compare May 29, 2019 01:55

Aaron1011 force-pushed the feature/panic_unwind_final branch from 634bf34 to b909669 Compare May 30, 2019 03:12

RalfJung reviewed Jun 4, 2019

View reviewed changes

src/fn_call.rs Outdated Show resolved Hide resolved

RalfJung reviewed Jun 4, 2019

View reviewed changes

src/fn_call.rs Outdated Show resolved Hide resolved

RalfJung reviewed Jun 4, 2019

View reviewed changes

src/lib.rs Outdated Show resolved Hide resolved

RalfJung reviewed Jun 4, 2019

View reviewed changes

src/lib.rs Outdated Show resolved Hide resolved

RalfJung reviewed Jun 4, 2019

View reviewed changes

src/fn_call.rs Outdated Show resolved Hide resolved

RalfJung reviewed Jun 4, 2019

View reviewed changes

src/fn_call.rs Outdated Show resolved Hide resolved

RalfJung reviewed Jun 4, 2019

View reviewed changes

src/lib.rs Outdated Show resolved Hide resolved

RalfJung reviewed Jun 4, 2019

View reviewed changes

src/lib.rs Outdated Show resolved Hide resolved

RalfJung reviewed Jun 4, 2019

View reviewed changes

tests/run-pass/catch_panic.rs Outdated Show resolved Hide resolved

Disable panic tests on Windows

80f9484

Miri currently does not support `GetProcAddress` and `GetModuleHandleW`, both of which end up getting invoked by the libstd panic hook.

RalfJung reviewed Nov 19, 2019

View reviewed changes

src/helpers.rs Show resolved Hide resolved

RalfJung reviewed Nov 19, 2019

View reviewed changes

src/shims/foreign_items.rs Outdated Show resolved Hide resolved

comments and slight refactoring

82374ad

Delegate to the actual panic runtime crate

8936d67

RalfJung reviewed Nov 19, 2019

View reviewed changes

src/shims/foreign_items.rs Outdated Show resolved Hide resolved

RalfJung reviewed Nov 19, 2019

View reviewed changes

src/shims/panic.rs Outdated Show resolved Hide resolved

Aaron1011 added 2 commits November 19, 2019 15:31

Re-add comment

e02dc4a

Update panic runtime comment

2750f60

Disable #[should_panic] test on Windows

6fe89e4

We should re-enable this once rust-lang#1059 is fixed

bors merged commit 6fe89e4 into rust-lang:master Nov 19, 2019

Aaron1011 deleted the feature/panic_unwind_final branch November 19, 2019 21:32

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Support unwinding after a panic #693

Support unwinding after a panic #693

Aaron1011 commented Apr 17, 2019 •

edited

Loading

oli-obk commented Apr 17, 2019

Aaron1011 commented Apr 17, 2019

oli-obk commented Apr 18, 2019

RalfJung commented Apr 29, 2019

Aaron1011 commented Apr 29, 2019

Aaron1011 commented May 30, 2019

Aaron1011 commented Nov 17, 2019

RalfJung commented Nov 17, 2019 •

edited

Loading

Aaron1011 commented Nov 17, 2019

RalfJung commented Nov 19, 2019

Aaron1011 commented Nov 19, 2019

Aaron1011 commented Nov 19, 2019

oli-obk commented Nov 19, 2019

bors commented Nov 19, 2019

bors commented Nov 19, 2019

bors commented Nov 19, 2019

Aaron1011 commented Nov 19, 2019

oli-obk commented Nov 19, 2019

bors commented Nov 19, 2019

bors commented Nov 19, 2019

bors commented Nov 19, 2019

RalfJung commented Nov 19, 2019

Support unwinding after a panic #693

Support unwinding after a panic #693

Conversation

Aaron1011 commented Apr 17, 2019 • edited Loading

oli-obk commented Apr 17, 2019

Aaron1011 commented Apr 17, 2019

oli-obk commented Apr 18, 2019

RalfJung commented Apr 29, 2019

Aaron1011 commented Apr 29, 2019

Aaron1011 commented May 30, 2019

Aaron1011 commented Nov 17, 2019

RalfJung commented Nov 17, 2019 • edited Loading

Aaron1011 commented Nov 17, 2019

RalfJung commented Nov 19, 2019

Aaron1011 commented Nov 19, 2019

Aaron1011 commented Nov 19, 2019

oli-obk commented Nov 19, 2019

bors commented Nov 19, 2019

bors commented Nov 19, 2019

bors commented Nov 19, 2019

Aaron1011 commented Nov 19, 2019

oli-obk commented Nov 19, 2019

bors commented Nov 19, 2019

bors commented Nov 19, 2019

bors commented Nov 19, 2019

RalfJung commented Nov 19, 2019

Aaron1011 commented Apr 17, 2019 •

edited

Loading

RalfJung commented Nov 17, 2019 •

edited

Loading