Why do we have this distinction of what causes a crashdump?

@syntactically

Why do we have this distinction? We should be clear about exactly which exceptions result in dumps and which do not, and whether there is any relationship between that and which faults poison the sandbox.

Originally posted by @syntactically in #1264 (comment)

It is coming from here:

hyperlight/src/hyperlight_host/src/hypervisor/hyperlight_vm.rs

Lines 950 to 961 in 64aa448

    
           match result { 
        
               Ok(_) => Ok(()), 
        
               Err(RunVmError::ExecutionCancelledByHost) => { 
        
                   // no need to crashdump this 
        
                   Err(RunVmError::ExecutionCancelledByHost) 
        
               } 
        
               Err(e) => { 
        
                   #[cfg(crashdump)] 
        
                   if self.rt_cfg.guest_core_dump { 
        
                       crashdump::generate_crashdump(self) 
        
                           .map_err(|e| RunVmError::CrashdumpGeneration(Box::new(e)))?; 
        
                   }

and

hyperlight/src/hyperlight_host/src/hypervisor/hyperlight_vm.rs

Lines 870 to 893 in 64aa448

    
           Ok(VmExit::Halt()) => { 
        
               break Ok(()); 
        
           } 
        
           Ok(VmExit::IoOut(port, data)) => { 
        
               self.handle_io(mem_mgr, host_funcs, port, data)?; 
        
           } 
        
           Ok(VmExit::MmioRead(addr)) => { 
        
               let all_regions = self.get_mapped_regions(); 
        
               match get_memory_access_violation( 
        
                   addr as usize, 
        
                   MemoryRegionFlags::WRITE, 
        
                   all_regions, 
        
               ) { 
        
                   Some(MemoryAccess::AccessViolation(region_flags)) => { 
        
                       break Err(RunVmError::MemoryAccessViolation { 
        
                           addr, 
        
                           access_type: MemoryRegionFlags::READ, 
        
                           region_flags, 
        
                       }); 
        
                   } 
        
                   None => { 
        
                       break Err(RunVmError::MmioReadUnmapped(addr)); 
        
                   } 
        
               }

I can create an issue for this if we believe this is wrong

two scenarios we should consider

hyperlight isn't working as expected
things that try and breach security

This may have changed when we re-classified some of the errors. We should collect more feedback from end users.

Navigation Menu

Repository navigation

Why do we have this distinction of what causes a crashdump? #1280

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

	match result {
	Ok(_) => Ok(()),
	Err(RunVmError::ExecutionCancelledByHost) => {
	// no need to crashdump this
	Err(RunVmError::ExecutionCancelledByHost)
	}
	Err(e) => {
	#[cfg(crashdump)]
	if self.rt_cfg.guest_core_dump {
	crashdump::generate_crashdump(self)
	.map_err(\|e\| RunVmError::CrashdumpGeneration(Box::new(e)))?;
	}

	Ok(VmExit::Halt()) => {
	break Ok(());
	}
	Ok(VmExit::IoOut(port, data)) => {
	self.handle_io(mem_mgr, host_funcs, port, data)?;
	}
	Ok(VmExit::MmioRead(addr)) => {
	let all_regions = self.get_mapped_regions();
	match get_memory_access_violation(
	addr as usize,
	MemoryRegionFlags::WRITE,
	all_regions,
	) {
	Some(MemoryAccess::AccessViolation(region_flags)) => {
	break Err(RunVmError::MemoryAccessViolation {
	addr,
	access_type: MemoryRegionFlags::READ,
	region_flags,
	});
	}
	None => {
	break Err(RunVmError::MmioReadUnmapped(addr));
	}
	}

Why do we have this distinction of what causes a crashdump? #1280

Description

Activity

jsturtevant commented on Mar 6, 2026

jsturtevant commented on Mar 11, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions