Place VM objects on the object list when created and never remove them.
ClosedPublic
Actions

Authored by jhb on May 1 2015, 5:00 PM.

Details

Reviewers

kib
alc

Commits

rS282660: Place VM objects on the object list when created and never remove them.

Summary

This is ok since objects come from a NOFREE zone and allows objects to
be locked while traversing the object list without triggering a LOR.

Ensure that objects on the list are marked DEAD while free or stillborn,
and that they have a refcount of zero.

Test Plan

booted a kernel with INVARIANTS and WITNESS

Diff Detail

Repository

rS FreeBSD src repository - subversion

Lint

No Lint Coverage

Unit

No Test Coverage

Event Timeline

jhb updated this revision to Diff 5140.May 1 2015, 5:00 PM

jhb retitled this revision from to Place VM objects on the object list when created and never remove them..

jhb edited the test plan for this revision. (Show Details)

jhb added reviewers: alc, kib.

jhb updated this object.

Herald added a subscriber: imp. · View Herald TranscriptMay 1 2015, 5:00 PM

jhb added a child revision: D2277: Export a list of VM objects via a sysctl..May 1 2015, 5:03 PM

kib accepted this revision.May 1 2015, 5:30 PM

kib edited edge metadata.

This revision is now accepted and ready to land.May 1 2015, 5:30 PM

alc added inline comments.May 1 2015, 6:30 PM

sys/vm/vm_object.c
191	I don't think that it is correct to remove this because of: obj_zone = uma_zcreate("VM OBJECT", sizeof (struct vm_object), NULL, #ifdef INVARIANTS vm_object_zdtor, #else NULL, #endif vm_object_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
192	I would suggest KASSERT()ing this here and moving the actual assignment to the backing object scan below.
1807	Add: backing_object->type = OBJT_DEAD;

Ok. I had actually meant to enable the zone dtor unconditionally (oops), but I'm fine with fixing it as you suggest instead.

Hmm, one other note is that most objects that are destroyed when the ref count drops to zero do not set OBJT_DEAD in their pager's dealloc routine. In fact, only the vnode pager seems to do this, so I think I will need to explicitly set OBJT_DEAD in vm_object_deallocate(). I should perhaps clear the handle to NULL as well?

Ya, you're right. Add the OBJT_DEAD assignment to the other pagers so that we're handling this in a uniform way.