witness(4)

NAME

witness - lock validation facility

SYNOPSIS

options WITNESS
options WITNESS_KDB
options WITNESS_SKIPSPIN

DESCRIPTION

The witness module keeps track of the locks acquired and re
leased by each
thread. It also keeps track of the order in which locks are
acquired
with respect to each other. Each time a lock is acquired,
witness uses
these two lists to verify that a lock is not being acquired
in the wrong
order. If a lock order violation is detected, then a mes
sage is output
to the kernel console detailing the locks involved and the
locations in
question. Witness can also be configured to drop into the
kernel debugger when an order violation occurs.
The witness code also checks various other conditions such
as verifying
that one does not recurse on a non-recursive lock. For
sleep locks,
witness verifies that a new process would not be switched to
when a lock
is released or a lock is blocked on during an acquire while
any spin
locks are held. If any of these checks fail, then the ker
nel will panic.
The flag that controls whether or not the kernel debugger is
entered when
a lock order violation is detected can be set in a variety
of ways. By
default, the flag is off, but if the WITNESS_KDB kernel op
tion is specified, then the flag will default to on. It can also be set
from the
loader(8) via the debug.witness.kdb environment variable or
after the
kernel has booted via the debug.witness.kdb sysctl. If the
flag is set
to zero, then the debugger will not be entered. If the flag
is non-zero,
then the debugger will be entered.
The witness code can also be configured to skip all checks
on spin
mutexes. By default, this flag defaults to off, but it can
be turned on
by specifying the WITNESS_SKIPSPIN kernel option. The flag
can also be
set via the loader(8) environment variable
debug.witness.skipspin. If
the variable is set to a non-zero value, then spin mutexes
are skipped.
Once the kernel has booted, the status of this flag can be
examined but
not set via the read-only sysctl debug.witness.skipspin.
The sysctl debug.witness.watch specifies the level of wit
ness involvement
in the system. A value of 1 specifies that witness is en
abled. A value
of 0 specifies that witness is disabled. This sysctl can be
written to
in order to disable witness, however it may not be used to
enable witness. The sysctl debug.witness.watch can be set via load
er(8).
The witness code also provides two extra ddb(4) commands if
both witness
and ddb(4) are compiled into the kernel:
show locks
Outputs the list of locks held by the current thread to the
kernel console along with the filename and line number at which each
lock was last
acquired by this thread.
show witness
Dump the current order list to the kernel console. The code
first displays the lock order tree for all of the sleep locks. Then
it displays
the lock order tree for all of the spin locks. Finally, it
displays a
list of locks that have not yet been acquired.

SEE ALSO

ddb(4), loader(8), sysctl(8), mutex(9)

HISTORY

The witness code first appeared in BSD/OS 5.0 and was im
ported from there
into FreeBSD 5.0.

BUGS

The witness code currently does not handle recursion of
shared sx(9)
locks properly.
BSD February 18, 2001
Copyright © 2010-2024 Platon Technologies, s.r.o.           Home | Man pages | tLDP | Documents | Utilities | About
Design by styleshout