From: Matthew Dillon <dillon@dragonflybsd.org>
Date: Mon, 8 Nov 2004 16:50:33 +0000 (+0000)
Subject: Do not explicitly set PCIM_CMD_SERRESPEN or PCIM_CMD_PERRESPEN.  This was
X-Git-Tag: v2.0.1~9821
X-Git-Url: https://gitweb.dragonflybsd.org/dragonfly.git/commitdiff_plain/60955de8ed4d81bdfa25167b4506f1f1b7c80754

Do not explicitly set PCIM_CMD_SERRESPEN or PCIM_CMD_PERRESPEN.  This was
set in FreeBSD to try to catch device problems, but it looks like we actually
found a device that blows up with an NMI during fwohci_init() so we no longer
set these bits.  OpenBSD doesn't set it.

My theory is that the device is not clearing its internal memory leaving
some of it in a parity-adverse state, causing accesses to the uninitialized
memory to fail.  Alternatively the hardware is just plain broken, but it
seems to probe fine without the bit set so...

Gabor's-hardware:
AMD Athlon 64 3000+ based system.
fwohci0: <Texas Instruments TSB43AB21/A/AI/A-EP> mem
0xd0000000-0xd0003fff,0xd0008000-0xd00087ff irq 11 at device 6.0 on pci0
fwohci0: OHCI version 1.10 (ROM=1)
fwohci0: No. of Isochronous channel is 4.
fwohci0: EUI64 00:90:f5:00:00:32:10:37
fwohci0: Phy 1394a available S400, 1 ports.
fwohci0: Link S400, max_rec 2048 bytes.

NMI-Reported-by: Gabor MICSKO <trey@hup.hu>
---

diff --git a/sys/bus/firewire/fwohci_pci.c b/sys/bus/firewire/fwohci_pci.c
index 10368ba0f7..61ee30c0f1 100644
--- a/sys/bus/firewire/fwohci_pci.c
+++ b/sys/bus/firewire/fwohci_pci.c
@@ -32,7 +32,7 @@
  * POSSIBILITY OF SUCH DAMAGE.
  * 
  * $FreeBSD: src/sys/dev/firewire/fwohci_pci.c,v 1.38 2004/01/23 17:37:09 simokawa Exp $
- * $DragonFly: src/sys/bus/firewire/fwohci_pci.c,v 1.15 2004/07/18 12:37:03 asmodai Exp $
+ * $DragonFly: src/sys/bus/firewire/fwohci_pci.c,v 1.16 2004/11/08 16:50:33 dillon Exp $
  */
 
 #define BOUNCE_BUFFER_TEST	0
@@ -233,12 +233,23 @@ fwohci_pci_init(device_t self)
 	int olatency, latency, ocache_line, cache_line;
 	u_int16_t cmd;
 
+	/*
+	 * Clear PCIM_CMD_MWRICEN as per FreeBSD/1.20, but note that the
+	 * problem may have been related to SERR and PERR being
+	 * unconditionally enabled in that rev.
+	 *
+	 * Do not change the SERRESPEN or PERRESPEN bits.  Use the BIOS
+	 * values (probably off).  This crashes some machines in fwohci_init().
+	 *
+	 * The theory here is that the device may not be properly initializing
+	 * its on-chip memory, leaving some of it in a parity errored state,
+	 * and enabling parity check may result in the device blowing up.
+	 * It's also possible that some hardware is just plain broken.  OpenBSD
+	 * does not turn on SERRESPEN or PERRESPEN so we won't either.
+	 */
 	cmd = pci_read_config(self, PCIR_COMMAND, 2);
-	cmd |= PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN | PCIM_CMD_MWRICEN |
-		PCIM_CMD_SERRESPEN | PCIM_CMD_PERRESPEN;
-#if 1
+	cmd |= PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN;
 	cmd &= ~PCIM_CMD_MWRICEN; 
-#endif
 	pci_write_config(self, PCIR_COMMAND, cmd, 2);
 
 	latency = olatency = pci_read_config(self, PCIR_LATTIMER, 1);