[PATCH] PCI probing status cleanup

[PATCH] PCI probing status cleanup

[Tim Hockin]

[-- Attachment #1: Type: text/plain, Size: 316 bytes --]

Martin et al,

Attached is a tiny patch that clears the master abort bit on bridges when
probing PCI devices.

Please let me know if there is any reason this would not be included in the
mainline kernel.

Thanks
Tim

-- 
Tim Hockin
Systems Software Engineer
Sun Microsystems, Cobalt Server Appliances
thockin@sun.com

[-- Attachment #2: pci_bridge.diff --]
[-- Type: text/plain, Size: 867 bytes --]

diff -ruN dist-2.4.6/drivers/pci/pci.c cobalt-2.4.6/drivers/pci/pci.c
--- dist-2.4.6/drivers/pci/pci.c	Mon Jul  2 14:42:53 2001
+++ cobalt-2.4.6/drivers/pci/pci.c	Mon Jul  9 11:04:01 2001
@@ -1229,8 +1233,19 @@
 		return NULL;
 
 	/* some broken boards return 0 or ~0 if a slot is empty: */
-	if (l == 0xffffffff || l == 0x00000000 || l == 0x0000ffff || l == 0xffff0000)
+	if (l == 0xffffffff || l == 0x00000000 
+	 || l == 0x0000ffff || l == 0xffff0000) {
+		/*
+		 * host/pci and pci/pci bridges will set Received Master Abort
+		 * (bit 13) on failed configuration access (happens when
+		 * searching for devices).  To be safe, clear the status
+		 * register.
+		 */
+		unsigned short st;
+		pci_read_config_word(temp, PCI_STATUS, &st);
+		pci_write_config_word(temp, PCI_STATUS, st);
 		return NULL;
+	}
 
 	dev = kmalloc(sizeof(*dev), GFP_KERNEL);
 	if (!dev)

Re: [PATCH] PCI probing status cleanup

[Jeff Garzik]

Tim Hockin wrote:
>         /* some broken boards return 0 or ~0 if a slot is empty: */
> -       if (l == 0xffffffff || l == 0x00000000 || l == 0x0000ffff || l == 0xffff0000)
> +       if (l == 0xffffffff || l == 0x00000000
> +        || l == 0x0000ffff || l == 0xffff0000) {
> +               /*
> +                * host/pci and pci/pci bridges will set Received Master Abort
> +                * (bit 13) on failed configuration access (happens when
> +                * searching for devices).  To be safe, clear the status
> +                * register.
> +                */
> +               unsigned short st;
> +               pci_read_config_word(temp, PCI_STATUS, &st);
> +               pci_write_config_word(temp, PCI_STATUS, st);
>                 return NULL;
> +       }

ok, though I wonder if we shouldn't just clear status for all PCI
devices on boot for paranoia's sake.

-- 
Jeff Garzik      | A recent study has shown that too much soup
Building 1024    | can cause malaise in laboratory mice.
MandrakeSoft     |