Stream: git-wasmtime

Topic: wasmtime / issue #4760 cranelift-fuzzgen fuzzbug: "Floati...

view this post on Zulip Wasmtime GitHub notifications bot (Aug 23 2022 at 15:56):

cfallin opened issue #4760:

https://oss-fuzz.com/testcase-detail/5258173635756032

ERROR: AddressSanitizer: FPE on unknown address 0x627000024418 (pc 0x627000024418 bp 0x7ffece9cd500 sp 0x7ffece9cd3e0 T0)
SCARINESS: 10 (signal)
    #0 0x627000024418  (<unknown module>)
    #1 0x6270000252f7  (<unknown module>)
    #0 0x561a4f7ef6c4 in cranelift_filetests::function_runner::CompiledFunction::call::ha6b9162cd2e21784 [wasmtime/cranelift/filetests/src/function_runner.rs:183](https://github.com/bytecodealliance/wasmtime/blob/418dbc15bd2a5269b338587661387e05fc77b983/cranelift/filetests/src/function_runner.rs#L183):9

input: here

cc @afonso360

view this post on Zulip Wasmtime GitHub notifications bot (Aug 23 2022 at 16:22):

afonso360 commented on issue #4760:

Huh, I can't format this input:

ubuntu@instance-20220805-0848:~/git/wasmtime/fuzz$ cargo fuzz fmt cranelift-fuzzgen ./4760.in --no-default-features

Output of `std::fmt::Debug`:

Arbitrary Error: The raw data is not of the correct format to construct this type

view this post on Zulip Wasmtime GitHub notifications bot (Aug 23 2022 at 16:25):

afonso360 commented on issue #4760:

Nevermind, forgot to base64 decode it :sweat_smile:

<details>
<summary> Formatted </summary>

ubuntu@instance-20220805-0848:~/git/wasmtime/fuzz$ cargo fuzz fmt cranelift-fuzzgen ./4760.in --no-default-features

Output of `std::fmt::Debug`:

;; Fuzzgen test case

test interpret
test run
set enable_llvm_abi_extensions
target aarch64
target s390x
target x86_64

function u0:1(i8, i32, f32 sext, f32 sext, i64 uext, b1, b1, b1, b1 uext, b1, i128 uext, i64, i16 sext, i8, f64 sext, i8 sext) -> i8, f64, f64 sext, i8, i32, f32 sext, b1 sext, i64 uext, i128, i128, i128 uext, i128 uext, b1 sext, i8 sext, i128 uext, i64 system_v {
    ss0 = explicit_slot 54
    ss1 = explicit_slot 54
    ss2 = explicit_slot 54
    sig0 = (i64, i64) -> i64 fast
    sig1 = () -> b1, f64, f64, f64, f64, f64, f64, f64, f64, f64, f64, f64, f64, f64 system_v
    sig2 = (i64, i64) -> i64 fast
    fn0 = %UdivI64 sig0
    fn1 = colocated u0:0 sig1
    fn2 = colocated %UdivI64 sig2
    jt0 = jump_table [block11, block12, block12, block4, block3, block3, block3, block2, block12, block12, block12, block2, block9, block5]
    jt1 = jump_table [block3, block12, block12, block4, block3, block3, block3]
    jt2 = jump_table [block12, block12, block12, block2, block9, block5, block11, block3, block12, block12, block12, block12, block11, block12]
    jt3 = jump_table [block2, block2, block5, block4, block11, block2, block2, block2, block11, block2, block2, block2, block2, block3, block3, block4, block12, block12, block12, block12, block4, block9, block5, block3, block2, block3, block12, block12]
    jt4 = jump_table [block2, block4, block12, block4, block9, block5, block3, block2, block3, block12, block12, block5, block12, block12, block12]
    jt5 = jump_table [block2, block3, block4, block5, block2, block2, block2, block3, block2]
    jt6 = jump_table [block2, block4]
    jt7 = jump_table [block4, block4, block4, block3, block2, block12, block3, block3, block3, block3, block3, block3, block3, block3]
    jt8 = jump_table [block2, block2, block2]
    jt9 = jump_table [block3, block3]
    jt10 = jump_table [block12, block12]
    jt11 = jump_table [block12, block3, block3, block3, block12, block12, block12, block12, block5, block5, block5, block5, block3, block3, block12, block12, block5, block12, block12]
    jt12 = jump_table [block2, block3, block4, block9, block5, block3, block3, block12, block12, block12, block3, block11, block12, block12, block4, block3, block3, block3, block5]
    jt13 = jump_table [block2, block2]
    jt14 = jump_table [block3, block3, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2]
    jt15 = jump_table [block12, block11, block12, block12, block12, block3, block3, block3, block12, block12, block12, block12, block2, block12, block12, block12, block12, block3, block4]
    jt16 = jump_table [block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2]
    jt17 = jump_table [block12, block4, block9, block5, block3, block3, block12, block12, block5, block12, block12, block12, block12, block2, block3, block4, block2, block4, block12]
    jt18 = jump_table [block9, block5, block3, block3, block12, block12, block4, block3, block3, block3, block12, block12, block12, block12, block2, block9, block5, block11, block3, block12]
    jt19 = jump_table [block2, block2, block4, block2, block3, block3, block3, block3, block2]
    jt20 = jump_table [block5, block2, block2, block2, block2, block11, block2, block2, block2, block2, block2, block2, block2, block4, block2, block3, block4, block2, block4, block12, block4, block9, block5, block3, block2, block3, block12, block12]
    jt21 = jump_table [block2, block11]
    jt22 = jump_table [block4, block4, block4, block4, block4, block4, block3, block2, block12, block3, block3, block3, block3, block3]
    jt23 = jump_table [block3, block3, block12, block12, block12, block12, block2, block9, block5, block11, block3, block12, block12, block12, block12, block11, block12, block12, block12, block3, block3, block3, block12, block12, block12, block12, block2, block12, block12, block12, block12, block3]
    jt24 = jump_table [block3, block2, block2, block2, block2, block2, block2, block2, block2, block9, block4, block2, block9, block5, block2, block3, block4, block5, block2, block2, block2, block3, block2, block2, block2, block2, block2, block2, block2]
    jt25 = jump_table [block2, block2]
    jt26 = jump_table [block5, block3, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2]
    jt27 = jump_table [block5, block2]
    jt28 = jump_table [block3, block3, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2, block2]
    jt29 = jump_table [block3, block12, block12, block5, block12, block12, block4, block9, block5, block3]
    jt30 = jump_table [block12, block12, block2, block3, block4, block4, block4, block12, block4, block9, block5, block3, block3, block12, block12, block5, block12, block12, block12]
    jt31 = jump_table [block11, block12, block12, block4, block3, block3, block3, block5, block5, block5, block5, block5, block5, block5, block5, block5, block5, block5, block5, block5, block5, block5, block5]
    jt32 = jump_table [block12, block2, block2, block2, block2, block2, block2, block2, block5, block2, block2, block2, block3, block2, block4, block2, block2, block2, block2]
    jt33 = jump_table [block5, block11, block12, block3, block12, block4, block12, block3, block2, block9, block5, block11, block9, block12, block3]
    jt34 = jump_table [block3, block3, block3, block3, block3, block3, block3, block3, block3, block3, block3, block3, block3, block3, block5, block11, block2, block11, block9, block11, block9, block11, block5, block5, block11, block2, block11, block4, block2, block2]
    jt35 = jump_table [block4, block2]
    jt36 = jump_table [block5, block2, block9, block12, block12, block12, block2, block2, block2, block12, block12]
    jt37 = jump_table [block2, block2]
    jt38 = jump_table [block5, block4, block11, block2, block12, block12, block12, block2, block2, block2, block12, block12, block12, block12, block2, block2, block2, block2, block3, block3, block3, block3, block3]
    jt39 = jump_table [block4, block12, block9, block4, block9, block5, block3, block2, block3, block12, block12, block5, block12, block12, block12, block12, block5, block3, block3, block2, block2, block2, block2, block2, block2, block2, block2]
    jt40 = jump_table [block4, block2, block9, block11, block11, block11, block11, block12, block4, block5, block5, block5, block5, block5, block5, block5, block5, block5, block5, block5, block5, block5, block5, block5, block5, block5, block5, block5, block5, block5, block5, block5]
    jt41 = jump_table [block11, block5]
    jt42 = jump_table [block12, block11, block12, block12, block12, block3, block3, block3, block12, block12, block12, block12, block5, block5, block5, block5, block3, block3, block12]
    jt43 = jump_table [block4, block2]
    jt44 = jump_table [block9, block9, block12, block2, block2, block2, block2, block2, block2, block2, block5, block2, block2, block2, block3, block2, block4, block2, block2, block2, block2, block2, block2, block2]
    jt45 = jump_table [block2, block4]
    jt46 = jump_table [block2, block9, block5, block3, block3, block12, block12, block4, block3, block3, block3, block12, block12, block12, block12, block2, block9, block5, block11]
    jt47 = jump_table [block2, block5, block5, block5, block5, block9, block2, block2, block2, block2, block2, block2, block3, block9, block5]
    jt48 = jump_table [block4, block5, block5, block2, block2, block2, block4, block2]
    jt49 = jump_table [block2, block2, block4]
    jt50 = jump_table [block12, block2, block2, block2, block12, block12, block12, block12, block2, block2, block2, block2, block3, block3, block3, block3, block3, block3, block3]
    jt51 = jump_table [block9, block4, block9, block5, block3, block2, block3, block12, block12, block5, block12, block12, block12, block12, block5, block3, block3, block2, block2]
    jt52 = jump_table [block11, block4, block2, block2, block2, block2, block12, block12, block12, block12, block12, block12, block12, block2, block2, block2, block2, block2, block2, block2, block2, block2]
    jt53 = jump_table [block12, block11, block12, block12, block12, block3, block3, block3, block12, block12, block12, block12, block5, block5, block5, block5, block3, block3, block12]
    jt54 = jump_table [block2, block11, block5, block11, block9, block5, block5, block5, block5, block11, block5, block9, block9, block9, block9, block3, block3, block3]
    jt55 = jump_table [block2, block4]
    jt56 = jump_table [block5, block5, block3, block2, block3, block12, block12, block11, block3, block9, block5, block2, block2, block5]
    jt57 = jump_table [block2, block2, block2, block2, block2, block5, block5, block5, block5, block9, block2, block2, block2, block2, block2]
    jt58 = jump_table [block3, block3, block3, block3, block3, block3, block3, block3, block3, block3, block3, block3, block3, block3, block5, block11, block2, block11, block9, block11, block9, block11, block5, block5, block11, block2, block3, block11, block2, block2]
    jt59 = jump_table [block4, block3, block2, block
[message truncated]

view this post on Zulip Wasmtime GitHub notifications bot (Aug 23 2022 at 16:28):

abrown commented on issue #4760:

Ah, fuzzing... good to know we can use jump tables:laughing:

view this post on Zulip Wasmtime GitHub notifications bot (Aug 23 2022 at 18:35):

afonso360 commented on issue #4760:

Interesting, trying to reproduce this on windows (via the generated clif test case) gives me a different result (STATUS_INTEGER_DIVIDE_BY_ZERO).

PS C:\Users\Afonso\CLionProjects\wasmtime\cranelift> cargo run -- test .\lmao.clif
    Finished dev [unoptimized + debuginfo] target(s) in 0.36s
     Running `C:\Users\Afonso\CLionProjects\wasmtime\target\debug\clif-util.exe test .\lmao.clif`
error: process didn't exit successfully: `C:\Users\Afonso\CLionProjects\wasmtime\target\debug\clif-util.exe test .\lmao.clif` (exit code: 0xc0000094, STATUS_INTEGER_DIVIDE_BY_ZERO)

And I did confirm, the debugger stops in a div instruction. I'll have to continue on a linux machine.

view this post on Zulip Wasmtime GitHub notifications bot (Aug 23 2022 at 18:36):

afonso360 edited a comment on issue #4760:

Interesting, trying to reproduce this on windows (via the generated clif test case) gives me a different result (STATUS_INTEGER_DIVIDE_BY_ZERO).

The only change I did to the test case was altering the calling convention from system_v to windows_fastcall.

PS C:\Users\Afonso\CLionProjects\wasmtime\cranelift> cargo run -- test .\lmao.clif
    Finished dev [unoptimized + debuginfo] target(s) in 0.36s
     Running `C:\Users\Afonso\CLionProjects\wasmtime\target\debug\clif-util.exe test .\lmao.clif`
error: process didn't exit successfully: `C:\Users\Afonso\CLionProjects\wasmtime\target\debug\clif-util.exe test .\lmao.clif` (exit code: 0xc0000094, STATUS_INTEGER_DIVIDE_BY_ZERO)

And I did confirm, the debugger stops in a div instruction. I'll have to continue on a linux machine.

view this post on Zulip Wasmtime GitHub notifications bot (Aug 23 2022 at 18:37):

afonso360 edited a comment on issue #4760:

Interesting, trying to reproduce this on windows (via the generated clif test case) gives me a different result (STATUS_INTEGER_DIVIDE_BY_ZERO).

The only change I did to the test case was altering the calling convention from system_v to windows_fastcall.

PS C:\Users\Afonso\CLionProjects\wasmtime\cranelift> cargo run -- test .\lmao.clif
    Finished dev [unoptimized + debuginfo] target(s) in 0.36s
     Running `C:\Users\Afonso\CLionProjects\wasmtime\target\debug\clif-util.exe test .\lmao.clif`
error: process didn't exit successfully: `C:\Users\Afonso\CLionProjects\wasmtime\target\debug\clif-util.exe test .\lmao.clif` (exit code: 0xc0000094, STATUS_INTEGER_DIVIDE_BY_ZERO)

And I did confirm, the debugger stops in a div instruction. I'll have to continue on a linux machine (which I just confirmed, can reproduce the issue via the clif test case).

view this post on Zulip Wasmtime GitHub notifications bot (Aug 23 2022 at 18:40):

afonso360 edited a comment on issue #4760:

Interesting, trying to reproduce this on windows (via the generated clif test case) gives me a different result (STATUS_INTEGER_DIVIDE_BY_ZERO).

The only change I did to the test case was altering the calling convention from system_v to windows_fastcall.

PS C:\Users\Afonso\CLionProjects\wasmtime\cranelift> cargo run -- test .\lmao.clif
    Finished dev [unoptimized + debuginfo] target(s) in 0.36s
     Running `C:\Users\Afonso\CLionProjects\wasmtime\target\debug\clif-util.exe test .\lmao.clif`
error: process didn't exit successfully: `C:\Users\Afonso\CLionProjects\wasmtime\target\debug\clif-util.exe test .\lmao.clif` (exit code: 0xc0000094, STATUS_INTEGER_DIVIDE_BY_ZERO)

And I did confirm, the debugger stops in a div instruction. I'll have to continue on a linux machine (which I just confirmed, can reproduce the issue via the clif test case).

Edit: Ah, seems to be a POSIX thing

view this post on Zulip Wasmtime GitHub notifications bot (Aug 23 2022 at 18:40):

afonso360 edited a comment on issue #4760:

Interesting, trying to reproduce this on windows (via the generated clif test case) gives me a different result (STATUS_INTEGER_DIVIDE_BY_ZERO).

The only change I did to the test case was altering the calling convention from system_v to windows_fastcall.

PS C:\Users\Afonso\CLionProjects\wasmtime\cranelift> cargo run -- test .\lmao.clif
    Finished dev [unoptimized + debuginfo] target(s) in 0.36s
     Running `C:\Users\Afonso\CLionProjects\wasmtime\target\debug\clif-util.exe test .\lmao.clif`
error: process didn't exit successfully: `C:\Users\Afonso\CLionProjects\wasmtime\target\debug\clif-util.exe test .\lmao.clif` (exit code: 0xc0000094, STATUS_INTEGER_DIVIDE_BY_ZERO)

And I did confirm, the debugger stops in a div instruction. I'll have to continue on a linux machine (which I just confirmed, can reproduce the issue via the clif test case).

Edit: Ah, seems to be a POSIX thing, its probably the same issue.

view this post on Zulip Wasmtime GitHub notifications bot (Aug 24 2022 at 15:56):

afonso360 edited a comment on issue #4760:

Interesting, trying to reproduce this on windows (via the generated clif test case) gives me a different result (STATUS_INTEGER_DIVIDE_BY_ZERO).

The only change I did to the test case was altering the calling convention from system_v to windows_fastcall.

PS C:\Users\Afonso\CLionProjects\wasmtime\cranelift> cargo run -- test .\lmao.clif
    Finished dev [unoptimized + debuginfo] target(s) in 0.36s
     Running `C:\Users\Afonso\CLionProjects\wasmtime\target\debug\clif-util.exe test .\lmao.clif`
error: process didn't exit successfully: `C:\Users\Afonso\CLionProjects\wasmtime\target\debug\clif-util.exe test .\lmao.clif` (exit code: 0xc0000094, STATUS_INTEGER_DIVIDE_BY_ZERO)

And I did confirm, the debugger stops in a div instruction. I'll have to continue on a linux machine (which I just confirmed, can reproduce the issue via the clif test case).

Edit: Ah, seems to be a POSIX thing, its probably the same issue.

view this post on Zulip Wasmtime GitHub notifications bot (Aug 26 2022 at 20:44):

jameysharp commented on issue #4760:

OSS-Fuzz thinks this was fixed, and indeed, I can reproduce the SIGFPE on 9cb987c67 but not on d394edcef. However I don't immediately see why #4752 would have actually fixed this.

On top of that, the same input produces a new error now:

thread '<unnamed>' panicked at 'called `Result::unwrap()` on an `Err` value: Undeclared function u0:0 is referenced by u0:1!', fuzz/fuzz_targets/cranelift-fuzzgen.rs:73:53

git bisect points at #4667 as the commit that introduced this new failure mode.

view this post on Zulip Wasmtime GitHub notifications bot (Aug 26 2022 at 21:36):

jameysharp commented on issue #4760:

Okay, #4752 does not change the CLIF that this fuzz target generates, which is good: we didn't accidentally "fix" the fuzz bug by changing the input format.

It does dramatically change the assembly generated from that CLIF, including making the stack frame slightly smaller and the function slightly larger. But both versions have almost 1,200 instructions and I don't have any hope of getting to a root cause from there.

There's a rotl instruction in the generated CLIF whose result (v216) is the discriminant for what looks like a "switch" construct, so it could be that this divide-by-zero is on a branch that isn't executed if the rotate is performed correctly. There's also at least one udiv that operates on the result of an ishl. There's probably plenty of other ways that shifts or rotates could affect the inputs to the many division instructions in this test case.

So it's plausible enough that fixing shifts/rotates made this test case pass that I'm going to just declare that OSS-Fuzz is right, and this bug is fixed by that PR. Hooray!

The new error is another matter which I'll bring up in #4667.

view this post on Zulip Wasmtime GitHub notifications bot (Aug 26 2022 at 21:36):

jameysharp closed issue #4760:

https://oss-fuzz.com/testcase-detail/5258173635756032

ERROR: AddressSanitizer: FPE on unknown address 0x627000024418 (pc 0x627000024418 bp 0x7ffece9cd500 sp 0x7ffece9cd3e0 T0)
SCARINESS: 10 (signal)
    #0 0x627000024418  (<unknown module>)
    #1 0x6270000252f7  (<unknown module>)
    #0 0x561a4f7ef6c4 in cranelift_filetests::function_runner::CompiledFunction::call::ha6b9162cd2e21784 [wasmtime/cranelift/filetests/src/function_runner.rs:183](https://github.com/bytecodealliance/wasmtime/blob/418dbc15bd2a5269b338587661387e05fc77b983/cranelift/filetests/src/function_runner.rs#L183):9

input: here

cc @afonso360

Last updated: Nov 22 2024 at 17:03 UTC