View Single Post
Old 05-07-04, 10:43 PM   #1
nfarce
Registered User
 
Join Date: May 2004
Posts: 3
Default nForce2 CPU C1 Disconnect Hang Fix (Semi-Official)

Fix for using APIC, IO-APIC and ACPI without any lockups caused by CPU transistioning to C1 disconnects causing the NForce2 hangs that everybody knows about. The fix is in current BitKeeper for 2.4.27 and 2.6.6 kernels.

From Allen Martin at nVidia to kernel mailing list, original mail available here. Conclusion, they've known about the problem forever but just didn't want to tell anybody and only included it in the small print for motherboard manufacturers to update their BIOS. (About bloody time, though Allen probably had to fight the evil sales force to get this released.)

Quote:
I'm happy to be able to make this information public to the Linux
community. This information has been previously released to BIOS /
board vendors as an appnote, but in the interest of getting a workaround
into the hands of users the quickest we're making it public for possible
inclusion into the Linux kernel.

Problem:
C1 Halt Disconnect problem on nForce2 systems

Description:
A hang is caused when the CPU generates a very fast CONNECT/HALT cycle
sequence.

Workaround:
Set the SYSTEM_IDLE_TIMEOUT to 80 ns. This allows the state-machine and
timer to return to a proper state within 80 ns of the CONNECT and probe
appearing together.

Since the CPU will not issue another HALT within 80 ns of the initial
HALT, the failure condition is avoided.

This will require changing the value for register at bus:0 dev:0 func:0
offset 6c.

Chip Current Value New Value
C17 1F0FFF01 1F01FF01
C18D 9F0FFF01 9F01FF01

Northbridge chip version may be determined by reading the PCI revision
ID (offset 8) of the host bridge at bus:0 dev:0 func:0. C1 or greater
is C18D.
Instant patch from the current IDE maintainer;

Quote:
diff -puN arch/i386/pci/fixup.c~nforce2_fix arch/i386/pci/fixup.c
--- linux-2.6.6-rc3-bk2/arch/i386/pci/fixup.c~nforce2_fix 2004-05-04 00:27:18.114421672 +0200
+++ linux-2.6.6-rc3-bk2-bzolnier/arch/i386/pci/fixup.c 2004-05-04 01:02:29.821393416 +0200
@@ -187,6 +187,39 @@ static void __devinit pci_fixup_transpar
dev->transparent = 1;
}

+/*
+ * Fixup for C1 Halt Disconnect problem on nForce2 systems.
+ *
+ * From information provided by "Allen Martin" <AMartin@nvidia.com>:
+ *
+ * A hang is caused when the CPU generates a very fast CONNECT/HALT cycle
+ * sequence. Workaround is to set the SYSTEM_IDLE_TIMEOUT to 80 ns.
+ * This allows the state-machine and timer to return to a proper state within
+ * 80 ns of the CONNECT and probe appearing together. Since the CPU will not
+ * issue another HALT within 80 ns of the initial HALT, the failure condition
+ * is avoided.
+ */
+static void __devinit pci_fixup_nforce2(struct pci_dev *dev)
+{
+ u32 val, fixed_val;
+ u8 rev;
+
+ pci_read_config_byte(dev, PCI_REVISION_ID, &rev);
+
+ /*
+ * Chip Old value New value
+ * C17 0x1F01FF01 0x1F0FFF01
+ * C18D 0x9F01FF01 0x9F0FFF01
+ */
+ fixed_val = rev < 0xC1 ? 0x1F01FF01 : 0x9F01FF01;
+
+ pci_read_config_dword(dev, 0x6c, &val);
+ if (val != fixed_val) {
+ printk(KERN_WARNING "PCI: nForce2 C1 Halt Disconnet fixup\n");
+ pci_write_config_dword(dev, 0x6c, fixed_val);
+ }
+}
+
struct pci_fixup pcibios_fixups[] = {
{
.pass = PCI_FIXUP_HEADER,
@@ -290,5 +323,11 @@ struct pci_fixup pcibios_fixups[] = {
.device = PCI_ANY_ID,
.hook = pci_fixup_transparent_bridge
},
+ {
+ .pass = PCI_FIXUP_HEADER,
+ .vendor = PCI_VENDOR_ID_NVIDIA,
+ .device = PCI_DEVICE_ID_NVIDIA_NFORCE2,
+ .hook = pci_fixup_nforce2
+ },
{ .pass = 0 }
};
nfarce is offline   Reply With Quote