Skip to contentSkip to conversation

x86: early_quirk check all bus/dev/func in domain 0

PATCH In review Yinghai Lu 2 participants 2 messages Feb 7, 2011 – Mar 4, 2011

Subsystems (2)
lore
Share & tools
lore

Yinghai Lu

author
#0 lore raw

PATCH 1/1 x86: early_quirk check all bus/dev/func in domain 0

When we have 00:1a.0, 00:1a.1, 00:1a.2, 00:1a.7
Found only 00:1a.0, and 00:1a.1 get processed. and 00:1a.2 and 00:1a.7 get skipped.

We should only check header type on func0.
Also should not bail out when have gap.

At last try to process bus 00 - 0xff instead of only bus 00.

-v2: check with VENDOR_ID instead of CLASS_DEVICE for invalid device.

---
 arch/x86/kernel/early-quirks.c |   44 +++++++++++++++++++++++------------------
 1 file changed, 25 insertions(+), 19 deletions(-)
arch/x86/kernel/early-quirks.c +25 -19
Index: linux-2.6/arch/x86/kernel/early-quirks.c
===================================================================
@@ -234,20 +234,22 @@ static struct chipset early_qrk[] __init
  */
 static int __init check_dev_quirk(int num, int slot, int func)
 {
+	u32 l;
 	u16 class;
 	u16 vendor;
 	u16 device;
 	u8 type;
 	int i;
 
-	class = read_pci_config_16(num, slot, func, PCI_CLASS_DEVICE);
-
-	if (class == 0xffff)
-		return -1; /* no class, treat as single function */
-
-	vendor = read_pci_config_16(num, slot, func, PCI_VENDOR_ID);
+	l = read_pci_config(num, slot, func, PCI_VENDOR_ID);
+	/* some broken boards return 0 or ~0 if a slot is empty: */
+	if (l == 0xffffffff || l == 0x00000000 ||
+	    l == 0x0000ffff || l == 0xffff0000)
+		return 0;
 
-	device = read_pci_config_16(num, slot, func, PCI_DEVICE_ID);
+	class = read_pci_config_16(num, slot, func, PCI_CLASS_DEVICE);
+	vendor = l & 0xffff;
+	device = l >> 16;
 
 	for (i = 0; early_qrk[i].f != NULL; i++) {
 		if (((early_qrk[i].vendor == PCI_ANY_ID) ||
@@ -263,27 +265,31 @@ static int __init check_dev_quirk(int nu
 			}
 	}
 
-	type = read_pci_config_byte(num, slot, func,
-				    PCI_HEADER_TYPE);
-	if (!(type & 0x80))
-		return -1;
+	/* only check header type on func 0 */
+	if (func == 0) {
+		type = read_pci_config_byte(num, slot, func,
+					    PCI_HEADER_TYPE);
+		if (!(type & 0x80))
+			return -1;
+	}
 
 	return 0;
 }
 
 void __init early_quirks(void)
 {
-	int slot, func;
+	int num, slot, func;
 
 	if (!early_pci_allowed())
 		return;
 
 	/* Poor man's PCI discovery */
-	/* Only scan the root bus */
-	for (slot = 0; slot < 32; slot++)
-		for (func = 0; func < 8; func++) {
-			/* Only probe function 0 on single fn devices */
-			if (check_dev_quirk(0, slot, func))
-				break;
-		}
+	/* Only can scan first domain */
+	for (num = 0; num < 256; num++)
+		for (slot = 0; slot < 32; slot++)
+			for (func = 0; func < 8; func++) {
+				/* Only probe func 0 on single fn devices */
+				if (check_dev_quirk(num, slot, func))
+					break;
+			}
 }
#1 lore raw
Quoted 4 lines · On Mon, 07 Feb 2011 14:14:52 -0800 Yinghai Lu <[email protected]> wrote:

When we have 00:1a.0, 00:1a.1, 00:1a.2, 00:1a.7
Found only 00:1a.0, and 00:1a.1 get processed. and 00:1a.2 and 00:1a.7 get skipped.

Quoted 8 lines

We should only check header type on func0.
Also should not bail out when have gap.

At last try to process bus 00 - 0xff instead of only bus 00.

-v2: check with VENDOR_ID instead of CLASS_DEVICE for invalid device.

I need a much better changelog here. It sounds (and looks) like the patch is doing several things:
- processing more than the first two functions on a device
- processing the full set of busses in the first domain
- checking for nonexistent devices

Why are you making these changes? Do you have a system that needs early quirks but violates the existing assumptions? Can you apply a standard quirk instead?

Signature · 1 line
Jesse Barnes, Intel Open Source Technology Center