Pull up following revision(s) (requested by wiz in ticket #1423):
external/bsd/tmux/usr.bin/tmux/Makefile: revision 1.47 (patch)
tmux: fix TMUX_CONF
Addresses PR 61034.
Pull up following revision(s) (requested by wiz in ticket #584):
external/bsd/tmux/usr.bin/tmux/Makefile: revision 1.47
tmux: fix TMUX_CONF
Addresses PR 61034.
The evaluation of resource shortage in wapbl_register_deallocation()
is incomplete as only "wl_dealloccnt" is checked. Replace this check with
the doflush check from wapbl_begin() testing for 75% instead 50% to
make sure the next wapbl_begin() will always flush.
It is still a mitigation, for a fix the file system must provide an
upper bound for every transaction to wapbl_begin().
PR kern/47146 user-triggerable crash in wapbl_register_deallocation (confi)
PR kern/60621 (panic: ufs_inactive: dirty filesystem?)
Implement a locking protocol for the 68k pmap structures. This essentially
replaces the old PMAP_CRIT_{ENTER,EXIT}() markers with PMAP_{LOCK,UNLOCK}()
calls.
On systems that aren't built for MP or for pmap lock debugging, this is
all a no-op as before.
Add UFS_WAPBL_UPDATE() after block allocations to group the
allocation and the inode update into the same transaction.
PR kern/60621 (panic: ufs_inactive: dirty filesystem?)
The 68010, 68020, and 68030 are have strictly ordered memory operations.
The 68040 and 68060 do not, however -- stores can be observed after
subsequent loads.
This is to say: all loads are load-acquire, all stores are store-release,
but store-before-load must be explicitly serialized. The 68040 User's
Manual states that a NOP insn forces this serialization.
And so: membar_aquire -> do nothing, membar_release -> do nothing,
membar_sync -> do nothing *with feeling*
Chapter and verse of the manual quoted in comments.
The MMU-specific functions that load the user root pointer already take
care of any necessary ATC / cache invalidations, so we don't need to do
it again here.