agp: ensure GART has an address before enabling it

Some BIOSs (eg.  the AMI BIOS on the Asus P4P800 motherboard) don't
initialise the GART address, and pcibios_assign_resources() can ignore it
because it can be marked as a host bridge (see
https://bugzilla.kernel.org/show_bug.cgi?id=24392#c5 for details).  This
was handled correctly up to 2.6.35, but the pci_enable_device() cleanup in
2.6.36 96576a9e1a0cdb8 ("agp: intel-agp: do not use PCI resources before
pci_enable_device()") means that the kernel tries to enable the GART
before assigning it an address; in such cases the GART overlaps with other
device assignments and ends up being disabled.

This patch fixes https://bugzilla.kernel.org/show_bug.cgi?id=24392

Note that I imagine efficeon-agp.c probably has the same problem, but
I can't test that and I'd like to make sure this patch is suitable for
-stable (since 2.6.36 and 2.6.37 are affected).

Signed-off-by: Stephen Kitt <steve@sk2.org>
Cc: Bjorn Helgaas <bjorn.helgaas@hp.com>
Cc: Maciej Rutecki <maciej.rutecki@gmail.com>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Cc: Kulikov Vasiliy <segooon@gmail.com>
Cc: Florian Mickler <florian@mickler.org>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>

1 files changed, 16 insertions, 11 deletions

diff --git a/drivers/char/agp/intel-agp.c b/drivers/char/agp/intel-agp.c
index 857df10c0428..b0a0dccc98c1 100644
--- a/drivers/char/agp/intel-agp.c
+++ b/drivers/char/agp/intel-agp.c
@@ -774,20 +774,14 @@ static int __devinit agp_intel_probe(struct pci_dev *pdev,
         dev_info(&pdev->dev, "Intel %s Chipset\n", intel_agp_chipsets[i].name);
 
         /*
-        * If the device has not been properly setup, the following will catch
-        * the problem and should stop the system from crashing.
-        * 20030610 - hamish@zot.org
-        */
-        if (pci_enable_device(pdev)) {
-                dev_err(&pdev->dev, "can't enable PCI device\n");
-                agp_put_bridge(bridge);
-                return -ENODEV;
-        }
-
-        /*
         * The following fixes the case where the BIOS has "forgotten" to
         * provide an address range for the GART.
         * 20030610 - hamish@zot.org
+        * This happens before pci_enable_device() intentionally;
+        * calling pci_enable_device() before assigning the resource
+        * will result in the GART being disabled on machines with such
+        * BIOSs (the GART ends up with a BAR starting at 0, which
+        * conflicts a lot of other devices).
         */
         r = &pdev->resource[0];
         if (!r->start && r->end) {
@@ -798,6 +792,17 @@ static int __devinit agp_intel_probe(struct pci_dev *pdev,
                 }
         }
 
+        /*
+        * If the device has not been properly setup, the following will catch
+        * the problem and should stop the system from crashing.
+        * 20030610 - hamish@zot.org
+        */
+        if (pci_enable_device(pdev)) {
+                dev_err(&pdev->dev, "can't enable PCI device\n");
+                agp_put_bridge(bridge);
+                return -ENODEV;
+        }
+
         /* Fill in the mode register */
         if (cap_ptr) {
                 pci_read_config_dword(pdev,