On Sun, Sep 10, 2006 at 05:08:49AM +0000, Paul Stoeber wrote:
> Quoting sys/dev/vnd.c rev 1.62:
>
> 1121 /*
> 1122 * Wait interruptibly for an exclusive lock.
> 1123 *
> 1124 * XXX
> 1125 * Several drivers do this; it should be abstracted and made
> MP-safe.
> 1126 */
> 1127 int
> 1128 vndlock(sc)
> 1129 struct vnd_softc *sc;
> 1130 {
> 1131 int error;
> 1132
> 1133 while ((sc->sc_flags & VNF_LOCKED) != 0) {
> 1134 sc->sc_flags |= VNF_WANTED;
> 1135 if ((error = tsleep(sc, PRIBIO | PCATCH,
> "vndlck", 0)) != 0)
> 1136 return (error);
> 1137 }
> 1138 sc->sc_flags |= VNF_LOCKED;
> 1139 return (0);
> 1140 }
>
> Is it possible for a process to have the cpu taken away from it
> between lines 1137 and 1138?
Yes, by an unmasked interrupt, not by another process.
> If so, is the comment "Several drivers do this" serious?
Not really. The comment assumes processes can simultaneously be executed
in kernel mode on different processors. This is not the case in OpenBSD
(for most parts of the code).
-p.
