Perform host address translation via QMP if available#12
Perform host address translation via QMP if available#12arcnmx wants to merge 1 commit intomemflow:mainfrom
Conversation
Heuristics using the largest allocation chunk are often insufficient
h33p
left a comment
h33p
left a comment
There was a problem hiding this comment.
I really like the conceptual idea. I think it would be better to not add host_start into mapping, but instead, rename remap_start to remap_base, and use it as an absolute value. Then, there's no difference between using QMP or not, you just know that remap_base is address within the QEMU process virtual address space. But aside from this, I'm quite excited to get this merged in.
| )?; | ||
|
|
||
| info!("qemu memory map found {:?}", qemu_map); | ||
| if let Some(qemu_map) = &biggest_map { |
| Ok((gpa, hva)) if gpa == mapping.range_start => | ||
| mapping.host_start = Some(hva), | ||
| _ => { | ||
| log::warn!("failed to parse host address from {line:?}"); |
There was a problem hiding this comment.
Does this ever happen? If it does happen rarely, it might be better to throw an error, instead of silently ignoring and falling back to the largest map offset.
ko1N
left a comment
ko1N
left a comment
There was a problem hiding this comment.
Thanks for the changes! Would be great if you could have a look at the suggestions and then we can merge it in 👍 🔥
| cmdline: &str, | ||
| qemu_map: &CTup2<Address, umem>, | ||
| ) -> Result<MemoryMap<(Address, umem)>> { | ||
| let qemu_map = || match qemu_map { |
There was a problem hiding this comment.
Why does this have to be a function that we resolve later in the different code paths? Can't we just error out immediately?
3 participants
Heuristics using the largest allocation chunk are often insufficient - for example if the VM has a large memory-mapped VFIO device attached.