summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorMiod Vallat <miod@cvs.openbsd.org>2010-02-24 22:33:21 +0000
committerMiod Vallat <miod@cvs.openbsd.org>2010-02-24 22:33:21 +0000
commit2252b8c11a5ca4eb448a721332b65a5811d452b0 (patch)
treedd1b8a24d44c7875619ac793143b711e72d4d124
parent27eac0f18d023d2641718dc61a7941f49fcc5ac0 (diff)
Update comments: the bit we flip in ISR4C on LS2F systems is not undocumented,
but documentation for it only exists in Chinese. I still don't have the slightest idea why flipping it is important, but I am more confident flipping it now (-:
-rw-r--r--sys/arch/loongson/dev/bonito.c15
1 files changed, 9 insertions, 6 deletions
diff --git a/sys/arch/loongson/dev/bonito.c b/sys/arch/loongson/dev/bonito.c
index 6762eec2025..9e141c78cc5 100644
--- a/sys/arch/loongson/dev/bonito.c
+++ b/sys/arch/loongson/dev/bonito.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: bonito.c,v 1.10 2010/02/12 08:14:02 miod Exp $ */
+/* $OpenBSD: bonito.c,v 1.11 2010/02/24 22:33:20 miod Exp $ */
/* $NetBSD: bonito_mainbus.c,v 1.11 2008/04/28 20:23:10 martin Exp $ */
/* $NetBSD: bonito_pci.c,v 1.5 2008/04/28 20:23:28 martin Exp $ */
@@ -289,11 +289,14 @@ bonito_attach(struct device *parent, struct device *self, void *aux)
if (!sc->sc_compatible) {
/*
- * According to Linux, changing the value of this
- * undocumented register ``avoids deadlock of PCI
- * reading/writing lock operation''.
- * Is this really necessary, and if so, does it
- * matter on other designs?
+ * According to Linux, changing the value of this register
+ * ``avoids deadlock of PCI reading/writing lock operation''.
+ *
+ * Unfortunately, documentation for the Implementation
+ * Specific Registers (ISR40 to ISR5C) is only found in the
+ * chinese version of the Loongson 2F documentation.
+ *
+ * The particular bit we set here is ``mas_read_defer''.
*/
/* c2000001 -> d2000001 */
REGVAL(BONITO_PCI_REG(0x4c)) |= 0x10000000;