I noticed this while reviewing the non-transparent superpages patch.
vm_object_madvise() is a no-op for unmanaged objects, but we should
limit the scope of mappings on which we'll call pmap_remove(). In
particular, with largepage objects we must remove mappings along
superpage boundaries.
Details
Details
- Reviewers
alc kib - Commits
- rS364317: Skip Linux madvise(MADV_DONTNEED) on unmanaged objects.
Diff Detail
Diff Detail
- Lint
Lint Passed - Unit
No Test Coverage - Build Status
Buildable 33026 Build 30413: arc lint + arc unit
Event Timeline
Comment Actions
I do not object, but is there a way for linux process to ever mmap largepage ?
Hmm, perhaps freebsd process can shm_open, then pass fd to linux process over unix socket, and then linux process can mmap(fd) ?
Comment Actions
That's what I was thinking. Or, maybe the fd can be inherited over execve(). I am not sure if the implicit O_CLOEXEC for shm objects can be cleared.