Increase the size of the sigaltstack. #1298
Conversation
Rust's stack overflow handler installs a sigaltstack stack with size SIGSTKSZ, which is too small for some of the things we do in signal handlers. Install bigger sigaltstack stacks so that we have enough space.
There was a problem hiding this comment.
Would it be possible to migrate this to Rust instead of C++? I'm hoping eventually to remove the need for C++ except for FFI definitions of setjmp/longjmp in the long term, and I don't think there's any fundamental reason why this needs to be in C++ rather than Rust.
Additionally could you add some comments about how libstd is configuring a sigaltstack, but it's too small for our purposes (w/ references to the two issues so far) so we make sure that whenever wasm is called we have a bigger sigaltstack.
| if (mprotect(stack_ptr, sigaltstack_size, PROT_READ | PROT_WRITE) != 0 || | ||
| sigaltstack(&new_stack, &old_stack) != 0 || | ||
| old_stack.ss_flags != 0 || | ||
| old_stack.ss_size > sigaltstack_size) |
There was a problem hiding this comment.
Could this test be done before we start mmap'ing? We could check to see if the sigaltstack size is already big enough or if it's disabled then we we can act before we mmap our own and set it.
| sigaltstack(&new_stack, &old_stack) != 0 || | ||
| old_stack.ss_flags != 0 || | ||
| old_stack.ss_size > sigaltstack_size) | ||
| abort(); |
There was a problem hiding this comment.
How come there's an OOM trap, but then this is an abort? Could this translate to a trap as well?
|
Also FWIW I see this as mostly a temporary patch. #900 is sort of fundamentaly un-fixable with our current strategy of stack overflow detection because we'll always have the case that wasm calls a ton of functions then calls some native code which aborts the whole process instead of returning a wasm trap. For example you could have something like: (module
(import "" "" (func $f))
(func $recurse
call $f
call $recurse)
(start $recurse))That would test calling a native function on every single wasm stack frame, and eventually the native code will abort the process due to stack overflow when in reality this is a wasm stack overflow condition. This'll make it so stack overflow doesn't trigger a second stack overflow when generating a backtrace, but we'll still want to handle this recursive case. |
This'll get fixed in bytecodealliance#1298 but for now let's get CI working again
This'll get fixed in #1298 but for now let's get CI working again
Subscribe to Label ActionThis issue or pull request has been labeled: "wasmtime", "wasmtime:api" Users Subscribed to "wasmtime:api"To subscribe or unsubscribe from this label, edit the |
|
Closing in favor of #1315. |
…ytecodealliance#1298) This patch adds a third mode for templates: REX inference is requestable at template instantiation time. This reduces the number of recipes by removing rex()/nonrex() redundancy for many instructions.
Rust's stack overflow handler installs a sigaltstack stack with size
SIGSTKSZ, which is too small for some of the things we do in signal
handlers. Install bigger sigaltstack stacks so that we have enough
space.
cc #900. The code here is a sketch of a possible
alternative. Advantages of this approach include not needing extra
prologue checks, not disabling the Rust stack overflow message, and
ensuring that we always have a guard page.