git: kernel - Attempt to fix a rare vm_page_queues panic
dillon at crater.dragonflybsd.org
Fri Dec 27 21:44:44 PST 2013
Author: Matthew Dillon <dillon at apollo.backplane.com>
Date: Fri Dec 27 21:37:42 2013 -0800
kernel - Attempt to fix a rare vm_page_queues panic
* Change the way the vm_page_t markers are used in the pageout daemon.
The structures are not stable because they are declared on the stack,
but the vm pageout scan functions were assuming that a TAILQ_NEXT()
call would yield a stable vm_page_t structure that could then be locked
* Fix the problem by holding the appropriate vm_page_queues spinlock
across the TAILQ_NEXT() lookup.
* The panic could not be definitively traced to this issue so we will have
to see if it reoccurs.
* This also fixes a potential infinite marker hop-over case when multiple
cpus are trying to scan the same queue at the same time. However, it
would be exceedingly rare for this case to actually occur since only
the pageout and swapcache daemons could compete in this way.
Summary of changes:
sys/vm/vm_fault.c | 10 ++++
sys/vm/vm_page.c | 2 +-
sys/vm/vm_pageout.c | 153 ++++++++++++++++++++++++++--------------------------
3 files changed, 86 insertions(+), 79 deletions(-)
DragonFly BSD source repository
More information about the Commits