AMD IDE oops in current 2.4-ac

Current 2.4.21-pre4-ac kernels oops in amd74xx.c with certain chipsets due to a table order problem. The problem is correctly detected by a BUG() in the pci probe routine, which should trigger for all non-nforce chipsets.

Regards
Andy

The fix, tested on a 7409 Viper board, is simple:

@@ -54,11 +54,11 @@
 	unsigned long base;
 	unsigned char flags;
 } amd_ide_chips[] = {
-	{ PCI_DEVICE_ID_AMD_8111_IDE,  0x00, 0x40, AMD_UDMA_100 },			/* AMD-8111 */
-	{ PCI_DEVICE_ID_AMD_OPUS_7441, 0x00, 0x40, AMD_UDMA_100 },			/* AMD-768 Opus */
-	{ PCI_DEVICE_ID_AMD_VIPER_7411, 0x00, 0x40, AMD_UDMA_100 | AMD_BAD_FIFO },	/* AMD-766 Viper */
-	{ PCI_DEVICE_ID_AMD_VIPER_7409, 0x00, 0x40, AMD_UDMA_66 | AMD_CHECK_SWDMA },	/* AMD-756 Viper */
 	{ PCI_DEVICE_ID_AMD_COBRA_7401, 0x00, 0x40, AMD_UDMA_33 | AMD_BAD_SWDMA },	/* AMD-755 Cobra */
+	{ PCI_DEVICE_ID_AMD_VIPER_7409, 0x00, 0x40, AMD_UDMA_66 | AMD_CHECK_SWDMA },	/* AMD-756 Viper */
+	{ PCI_DEVICE_ID_AMD_VIPER_7411, 0x00, 0x40, AMD_UDMA_100 | AMD_BAD_FIFO },	/* AMD-766 Viper */
+	{ PCI_DEVICE_ID_AMD_OPUS_7441, 0x00, 0x40, AMD_UDMA_100 },			/* AMD-768 Opus */
+	{ PCI_DEVICE_ID_AMD_8111_IDE,  0x00, 0x40, AMD_UDMA_100 },			/* AMD-8111 */
         { PCI_DEVICE_ID_NVIDIA_NFORCE_IDE, 0x00, 0x50, AMD_UDMA_100 },                  /* nVidia nForce */
         { PCI_DEVICE_ID_NVIDIA_NFORCE2_IDE, 0x00, 0x50, AMD_UDMA_100 },                  /* nVidia nForce */
 

If moving entries in the table caused a bug, adding new entries could do the same too perhaps ? This sounds quite fragile based on your description & diff.

Dave

| Dave Jones.        http://www.codemonkey.org.uk
| SuSE Labs

amd74xx_probe() compares the pci device ids of both the probe table and the settings table, so should be save from misprogramming.
To protect against using random memory after the end of the table, a bug check on the table size could help:

        if (dev->device != d->device) BUG();
+      if (sizeof(amd_ide_chips) >= id->driver_data) BUG();
        if (dev->device != amd_config->id) BUG();

Even with this it still looks error prone, especially since the probe table and the config table are several lines spread across the file.

Regards
Andy

Hello, Dave and other.
Same behavior was in 2.5.5x and I've send similar patch to Linus and it was apllied. Changeset 1.879.1.84. There is two tables and missorder of devices cause a BUG() during init.
Thanks, Ruslan.