I have created some code in lm78.c to handle the W83792D watchdog portion
of the chip. A diff is pasted at the end of this message. There is one
problem with it though. Under OpenBSD 4.1, if sysctl.kern.period is set to
anything other than 0 while sysctl.kern.auto is set to 1, after a few
seconds a panic occurs with the following trace:
panic: process context required
lockmgr
ichiic_ic2_acquire_bus
lm_i2c_writereg
wb_w83792d_wdg_cb
wdog_tickle
softclock
Bad Frame Pointer: ...
The problem is that lockmgr requires the caller to have a process id, but
the wdog tickler is not associated with a process. So either lockmgr needs
to be changed to handle a caller with no process, or the tickler needs to
be associated with a context.
The problem is different on current. The trace I get for that is:
kernel: page fault trap, code = 0
Stooped at sleep_setup + 0x1d: cmpb $0x7, 0x35(%ebx)
ddb> trace
sleep_setup(d0xd9cdc, d1901700, 10, d071e656, cba8d6aa) at sleep_setup+0x1d
tsleep(d1901700, 10, d071e656, 64) at tsleep+0x5c
ichiic_i2c_exec(d1901700, 1, 2f, d08d9dae, 1) at ichiic_i2c_exec+0x27d
iic_exec(d1901740, 1, 2f, d08d9dae, 1, d08d9daf, 1, 9, d1901764, 10) at
iic_exec+0x177
lm_i2c_writereg(d191b800, 4, d07555f0d0, d08d8000, 0) at
lm_i2c_readreg+0x40
wb_w83792d_wdg_cb(d191b800) at wb_w83792_wdb_cb+0x17
wdog_tickle(0, 0, 0, 5305, 0) at wdog_tickle+0x25
sofclock(d0650058, 10, 380010, d080010, d08d9e6c) at softclock+0x23c
Bad Frame Pointer: 0xd08d9e14
I only looked at the problem on current for a short period of time. I'm
going to be using the watchdogd so I don't actually need the
sysctl.watchdog.auto functionality. But if this is going to get committed,
than I would be more than happy in testing any changes that may arise due
to the issues above.
On another note, about five months I submitted a driver for the W83627HF
watchdog in a new file called wbwdg.c. I have been coordinating with an
OpenBSD developer to get it committed, and last I heard it was going to
get committed. But it has not been committed, and I seem to have lost
contact with the developer. So I was wondering if this code could get
committed?
Thanks
Jonathan Steel
Diff for W83792D watchdog
--- lm78.c Mon Jun 25 16:50:18 2007
+++ lm78.c_old Mon Jan 29 08:06:01 2007
@@ -69,6 +69,9 @@
void as_refresh_temp(struct lm_softc *, int);
+void wb_w83792d_wdg_init(struct lm_softc *);
+int wb_w83792d_wdg_cb(void *, int);
+
struct lm_chip {
int (*chip_match)(struct lm_softc *);
};
@@ -555,6 +558,7 @@
else
printf(": W83792D rev 0x%x\n", devid);
lm_setup_sensors(sc, w83792d_sensors);
+ wb_w83792d_wdg_init(sc);
break;
case WB_CHIPID_AS99127F:
if (vendid == WB_VENDID_ASUS) {
@@ -885,3 +889,36 @@
sensor->value = sdata * 500000 + 273150000;
}
}
+
+void
+wb_w83792d_wdg_init(struct lm_softc *sc)
+{
+ int control = sc->lm_readreg(sc, 0x40);
+
+ /* Enable watchdog in configuration register */
+ control |= 0x10;
+ sc->lm_writereg(sc, 0x40, control);
+
+ /*
+ * Enable hard watch dog timer with period 0
+ * so that its initially disabled
+ */
+ sc->lm_writereg(sc, 0x04, 0x00);
+ sc->lm_writereg(sc, 0x01, 0xAA);
+ sc->lm_writereg(sc, 0x01, 0x33);
+
+ wdog_register(sc, wb_w83792d_wdg_cb);
+
+ return;
+}
+
+int
+wb_w83792d_wdg_cb(void *arg, int period)
+{
+ struct lm_softc *sc = arg;
+
+ sc->lm_writereg(sc, 0x04, period);
+
+ return period;
+}
--
Using Opera's revolutionary e-mail client: http://www.opera.com/mail/
