svn commit: r223218 - head/sys/dev/pccbb

[Warner Losh]

Author: imp
Date: Sat Jun 18 02:25:08 2011
New Revision: 223218
URL: http://svn.freebsd.org/changeset/base/223218

Log:
  After we get a good power signal, always wait about 10ms before
  proceeding.
  
  On boot, some laptops with certain cards in them sometimes fail on
  boot, but if the card is inserted after boot it works.  Experiments
  show that small delays here makes things more reliable.  It is
  believed that some combinations need a little more time before the
  power on the card is really stable enough to be reliable once the
  power is stable in the bridge.

Modified:
  head/sys/dev/pccbb/pccbb.c

Modified: head/sys/dev/pccbb/pccbb.c
==============================================================================
--- head/sys/dev/pccbb/pccbb.c	Sat Jun 18 00:53:51 2011	(r223217)
+++ head/sys/dev/pccbb/pccbb.c	Sat Jun 18 02:25:08 2011	(r223218)
@@ -800,24 +800,36 @@ cbb_power(device_t brdev, int volts)
 		 * We have a shortish timeout of 500ms here.  Some bridges do
 		 * not generate a POWER_CYCLE event for 16-bit cards.  In
 		 * those cases, we have to cope the best we can, and having
-		 * only a short delay is better than the alternatives.
+		 * only a short delay is better than the alternatives.  Others
+		 * raise the power cycle a smidge before it is really ready.
+		 * We deal with those below.
 		 */
 		sane = 10;
 		while (!(cbb_get(sc, CBB_SOCKET_STATE) & CBB_STATE_POWER_CYCLE) &&
 		    cnt == sc->powerintr && sane-- > 0)
 			msleep(&sc->powerintr, &sc->mtx, 0, "-", hz / 20);
 		mtx_unlock(&sc->mtx);
+
+		/*
+		 * Relax for 10ms.  Some bridges appear to assert this signal
+		 * right away, but before the card has stabilized.  Other
+		 * cards need need more time to cope up reliabily.
+		 * Experiments with troublesome setups show this to be a
+		 * "cheap" way to enhance reliabilty.  We need not do this for
+		 * "off" since we don't touch the card after we turn it off.
+		 */
+		pause("cbbPwr", min(hz / 100, 1));
+
 		/*
 		 * The TOPIC95B requires a little bit extra time to get its
 		 * act together, so delay for an additional 100ms.  Also as
 		 * documented below, it doesn't seem to set the POWER_CYCLE
 		 * bit, so don't whine if it never came on.
 		 */
-		if (sc->chipset == CB_TOPIC95) {
+		if (sc->chipset == CB_TOPIC95)
 			pause("cbb95B", hz / 10);
-		} else if (sane <= 0) {
+		else if (sane <= 0)
 			device_printf(sc->dev, "power timeout, doom?\n");
-		}
 	}
 
 	/*