Skip to contentSkip to conversation

Yet another 2.6.35 regression (AGP)? (resend)

DISCUSSION Woody Suwalski 2 participants 2 messages Jul 9, 2010 Agpgart driver

lore
Share & tools
lore
#0 lore raw

I have found one system, where 2.6.35 does not work (as tested with rc3 and rc4) That Intel system has no problems in 2.6.33.x nor 2.6.34.0.

The problem seems to be in AGP - I can boot if I specify "agp=off" - but of course only in text mode... There seems to be a hard lock-up, so the only way to show the crash is by picture 8-)

Since I do not build kernel on that machine, I did not do any bisect tests, however if someone is interested in digging deeper, I can try... Preferably a patch to try out ;-)

This bug seems to be different then https://bugzilla.kernel.org/show_bug.cgi?id=16179

Should it be blamed on BIOS (the conflict indicated just before the crash)?

Thanks, Woody

PS. Resending, as original post cc'd [email protected], what seems to be bad...

#1 lore raw
Quoted 16 lines · Woody Suwalski

I have found one system, where 2.6.35 does not work (as tested with rc3 and rc4)
That Intel system has no problems in 2.6.33.x nor 2.6.34.0.

The problem seems to be in AGP - I can boot if I specify "agp=off" - but of course only in text mode...
There seems to be a hard lock-up, so the only way to show the crash is by picture 8-)

Since I do not build kernel on that machine, I did not do any bisect tests, however if someone is interested in digging deeper, I can try... Preferably a patch to try out ;-)

This bug seems to be different then
https://bugzilla.kernel.org/show_bug.cgi?id=16179

Quoted 7 lines

Should it be blamed on BIOS (the conflict indicated just before the crash)?

Thanks, Woody

PS. Resending, as original post cc'd [email protected], what seems to be bad...

How about this patch? I'm waiting confirmation from another user, but your problem looks quite similar.

rtg

Signature · 1 line
Tim Gardner [email protected]
Quoted 1 line

From 684925d4ba303f9e23b428831c2386f0203d6622 Mon Sep 17 00:00:00 2001

From: Tim Gardner <[email protected]>
Date: Fri, 9 Jul 2010 11:14:41 -0600
Subject: [PATCH] i915: Use the correct mask to detect i830 aperture size.

commit f1befe71fa7a79ab733011b045639d8d809924ad introduced a regression when detecting aperture size of some i915 adapters, e.g., those on the Intel Q35 shipset.

The original report: https://bugzilla.kernel.org/show_bug.cgi?id=15733
The regression report: https://bugzilla.kernel.org/show_bug.cgi?id=16294

According to the specification found at http://intellinuxgraphics.org/VOL_1_graphics_core.pdf,
the PCI config space register I830_GMCH_CTRL is a mirror of GMCH Graphics Control. The correct macro for isolating the aperture size bits is therefore I830_GMCH_GMS_MASK along with the attendant changes to the case statement.

Cc Chris Wilson <[email protected]>
Cc Eric Anholt <[email protected]>
---
 drivers/char/agp/intel-gtt.c |   13 ++++++++-----
 1 files changed, 8 insertions(+), 5 deletions(-)
drivers/char/agp/intel-gtt.c +8 -5
index 9344216..a754715 100644
@@ -1216,17 +1216,20 @@ static int intel_i915_get_gtt_size(void)
 
 		/* G33's GTT size defined in gmch_ctrl */
 		pci_read_config_word(agp_bridge->dev, I830_GMCH_CTRL, &gmch_ctrl);
-		switch (gmch_ctrl & G33_PGETBL_SIZE_MASK) {
-		case G33_PGETBL_SIZE_1M:
+		switch (gmch_ctrl & I830_GMCH_GMS_MASK) {
+		case I830_GMCH_GMS_STOLEN_512:
+			size = 512;
+			break;
+		case I830_GMCH_GMS_STOLEN_1024:
 			size = 1024;
 			break;
-		case G33_PGETBL_SIZE_2M:
-			size = 2048;
+		case I830_GMCH_GMS_STOLEN_8192:
+			size = 8*1024;
 			break;
 		default:
 			dev_info(&agp_bridge->dev->dev,
 				 "unknown page table size 0x%x, assuming 512KB\n",
-				(gmch_ctrl & G33_PGETBL_SIZE_MASK));
+				(gmch_ctrl & I830_GMCH_GMS_MASK));
 			size = 512;
 		}
 	} else {
Signature · 1 line
1.7.0.4