(vm object 5) Move PGA_NOSYNC to VPA_NOSYNC so that the object lock is not required.

[jeff]

Summary

This patch simply moves the flag field and makes it atomic to eliminate the final dependency on the object lock in vm_page_set_validclean().

I do not fully understand the race described in vm_fault.c:~225 that requires the page lock. This may be the final vestige of page_lock after Mark and I are done so it would be good for someone to suggest a new approach and especially to review this patch in that light. I believe what I have done is correct however.

[kib]

We have MAP_NOSYNC flag for mmap(2). Suppose we have one mapping with the flag set, and other with the flag clear. Then write fault for the page might come from either mappings, and we either want or not want to set VPO_NOSYNC depending from which mapping the write fault is coming. More, if the VPO_NOSYNC page is re-dirtied from non-NOSYNC mapping, we want to remove VPO_NOSYNC.

Of course this is very weird because of contradictory requirements and the fact that write mappings are not invalidated when the NOSYNC is set. Nonetheless, MAP_NOSYNC is apparently used by users, see where it all started at https://lists.freebsd.org/pipermail/freebsd-hackers/2014-September/045981.html

[markj]

sys/vm/vm_page.c

2485 ↗

(On Diff #61912)

See my comment below.

2629 ↗

(On Diff #61912)

The comment was referring to stray oflags, so it doesn't really make sense here anymore. I think you can just delete it.