Dump process map on mmap failure and fix typos
#1127
Conversation
src/util/memory.rs
Outdated
| #[cfg(target_os = "linux")] | ||
| info!("{}", get_process_memory_maps()); |
There was a problem hiding this comment.
- It is better if we make the call regardless of the current OS, but we may implement the function platform-specific. At this moment, we can make the function do nothing (or return empty string or "(unavailable)") for systems other than Linux.
- We should just print using
eprint!oreprintln!(or do the printing insideget_process_memory_maps, or introduce aprint_process_memory_maps). It will be very helpful to have the information when the release build fails. And if it fails, it doesn't matter whether we print more things to the console or less.
There was a problem hiding this comment.
Ideally we want more information (what address was MMTk trying to mmap), but our current internal API for all this is quite wonky.
| @@ -147,6 +147,7 @@ pub fn handle_mmap_error<VM: VMBinding>(error: Error, tls: VMThread) -> ! { | |||
| match error.kind() { | |||
| // From Rust nightly 2021-05-12, we started to see Rust added this ErrorKind. | |||
| ErrorKind::OutOfMemory => { | |||
| eprintln!("{}", get_process_memory_maps()); | |||
There was a problem hiding this comment.
Well, I just realised that the out-of-memory event is not always lethal because the VM binding can implement VM::VMCollection::out_of_memory. If the VM can recover from OOM (like OpenJDK), it is probably unnecessary to print the memory maps. But if it crashes (like when OpenJDK generates something like hs_err_pidXXXXX.log), the ability to print mmap would be helpful. I think if there is a call to VM::VMCollection::out_of_memory following this, we should move it to the default implementation of Collection::out_of_memory (because it panics anyway).
There was a problem hiding this comment.
No this is different. This is an mmap error, not a simple OOM error. The simple OOM error is AllocationError::HeapOutOfMemory. As it says in the comment below the inserted line: "Expect the VM to abort immediately"
| @@ -159,14 +160,18 @@ pub fn handle_mmap_error<VM: VMBinding>(error: Error, tls: VMThread) -> ! { | |||
| if let Some(os_errno) = error.raw_os_error() { | |||
| // If it is OOM, we invoke out_of_memory() through the VM interface. | |||
| if os_errno == libc::ENOMEM { | |||
| eprintln!("{}", get_process_memory_maps()); | |||
There was a problem hiding this comment.
This is OOM, too. We may rely on the default impl of Collection::out_of_memory to print the mmap.
No description provided.