Change Details

This dequeue dates back to the CSRG sources: https://svnweb.freebsd.org/csrg/sys/vm/vm_fault.c?view=markup#l239 However, I don't believe it's necessary. If the page is in PQ_INACTIVE/PQ_LAUNDRY, the busy lock will cause the page daemon to ignore it. If the page is in PQ_ACTIVE, we can handle a concurrent scan by the page daemon, which is only counting references. If the fault is CoW, release_page() will call vm_page_deactivate() on the page, which handles the dequeue. Otherwise we will call vm_page_activate() on the page, which also dequeues as necessary. In particular, if the page is already in PQ_ACTIVE, we can avoid a needless requeue.

This dequeue dates back to the CSRG sources: https://svnweb.freebsd.org/csrg/sys/vm/vm_fault.c?view=markup#l239 However, I don't believe it's necessary. If the page is in PQ_INACTIVE/PQ_LAUNDRY, the busy lock will cause the page daemon to ignore it. If the page is in PQ_ACTIVE, we can handle a concurrent scan by the page daemon, which is only counting references. If the fault is CoW, release_page() will call vm_page_deactivate() on the page, which handles the dequeue. Otherwise we will call vm_page_activate() on the page, which also dequeues as necessary. In particular, if the page is already in PQ_ACTIVE, this change lets us avoid a needless pair of page queue lock acquisitions.

This dequeue dates back to the CSRG sources: https://svnweb.freebsd.org/csrg/sys/vm/vm_fault.c?view=markup#l239 However, I don't believe it's necessary. If the page is in PQ_INACTIVE/PQ_LAUNDRY, the busy lock will cause the page daemon to ignore it. If the page is in PQ_ACTIVE, we can handle a concurrent scan by the page daemon, which is only counting references. If the fault is CoW, release_page() will call vm_page_deactivate() on the page, which handles the dequeue. Otherwise we will call vm_page_activate() on the page, which also dequeues as necessary. In particular, if the page is already in PQ_ACTIVE, we can avoid a needlessthis change lets us reavoid a needless pair of page queue lock acquisitions.