[PATCH] pci: fixup parent subordinate busnr
authorIvan Kokshaysky <ink@jurassic.park.msu.ru>
Fri, 23 Sep 2005 04:06:31 +0000 (21:06 -0700)
committerLinus Torvalds <torvalds@g5.osdl.org>
Fri, 23 Sep 2005 15:05:16 +0000 (08:05 -0700)
I believe the change that broke things is introduction of
pci_fixup_parent_subordinate_busnr().

The patch here does two things:
- hunk #1 should fix the problems you've seen when you boot without
  additional "pci" kernel options;
- hunk #2 supposedly fixes boot with "pci=assign-busses" option which
  otherwise hangs Acer TM81xx machines as reported.

Please try this with and without "pci=assign-busses". If it boots,
I'd like to see 'lspci -vvx' for both cases.

Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
drivers/pci/probe.c

index c77d5b1bbff6ccd751111e5f15fc7c5604e418b3..005786416bb5be5cc7a99f010f9ed0662f64706c 100644 (file)
@@ -402,6 +402,12 @@ static void pci_enable_crs(struct pci_dev *dev)
 static void __devinit pci_fixup_parent_subordinate_busnr(struct pci_bus *child, int max)
 {
        struct pci_bus *parent = child->parent;
+
+       /* Attempts to fix that up are really dangerous unless
+          we're going to re-assign all bus numbers. */
+       if (!pcibios_assign_all_busses())
+               return;
+
        while (parent->parent && parent->subordinate < max) {
                parent->subordinate = max;
                pci_write_config_byte(parent->self, PCI_SUBORDINATE_BUS, max);
@@ -478,8 +484,18 @@ int __devinit pci_scan_bridge(struct pci_bus *bus, struct pci_dev * dev, int max
                 * We need to assign a number to this bus which we always
                 * do in the second pass.
                 */
-               if (!pass)
+               if (!pass) {
+                       if (pcibios_assign_all_busses())
+                               /* Temporarily disable forwarding of the
+                                  configuration cycles on all bridges in
+                                  this bus segment to avoid possible
+                                  conflicts in the second pass between two
+                                  bridges programmed with overlapping
+                                  bus ranges. */
+                               pci_write_config_dword(dev, PCI_PRIMARY_BUS,
+                                                      buses & ~0xffffff);
                        return max;
+               }
 
                /* Clear errors */
                pci_write_config_word(dev, PCI_STATUS, 0xffff);