Bug 11087 - MMCONFIG isn't used under 2.6.26 on second generation MacBook Pro
Summary: MMCONFIG isn't used under 2.6.26 on second generation MacBook Pro
Status: CLOSED INSUFFICIENT_DATA
Alias: None
Product: Platform Specific/Hardware
Classification: Unclassified
Component: x86-64 (show other bugs)
Hardware: All Linux
: P1 high
Assignee: platform_x86_64@kernel-bugs.osdl.org
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2008-07-14 19:40 UTC by Jack Howarth
Modified: 2010-01-19 19:43 UTC (History)
3 users (show)

See Also:
Kernel Version: 2.6.26
Subsystem:
Regression: Yes
Bisected commit-id:


Attachments
dmesg file from successful linux 2.6.25.10 boot (33.88 KB, text/plain)
2008-07-15 22:40 UTC, Jack Howarth
Details
dmesg file for linux 2.6.25.10 boot with apic=verbose (37.62 KB, text/plain)
2008-07-16 15:48 UTC, Jack Howarth
Details
lspci -vvxxx output from under 2.6.25.10 on MacBook Pro v2 (41.34 KB, application/octet-stream)
2008-07-16 19:43 UTC, Jack Howarth
Details
.config used for linux 2.6.26 kernel builds on MacBookPro (81.16 KB, text/plain)
2008-07-17 05:07 UTC, Jack Howarth
Details

Description Jack Howarth 2008-07-14 19:40:42 UTC
Latest working kernel version: 2.6.25.10
Earliest failing kernel version: 2.6.26
Distribution: Fedora (also with stock kernel build)
Hardware Environment: MacBook Pro (second generation) with 3GB RAM
Software Environment: Fedora 9 x86_64
Problem Description: The new MMCONFIG detection code is not working on a second generation MacBook Pro under 2.6.26. The boot process hangs with...

ACPI: bus type pci registered
PCI: MCFG configuration 0: base f0000000 segment 0 buses 0 - 255
PCI: Not using MMCONFIG
PCI: Using configuration type 1 for base access
ACPI: EC: EC description table is found, configuring boot EC
ACPI: EC: non-query interrupt received switching to interrupt mode
ACPI: BIOS_OSI (Linux) query ignored via DMI
ACPI: Interpreter enabled
ACPI: (supports S0 S3 S4 S5)
ACPI: Using IOAPIC for interrupt routing
PCI: MCFG configuration 0: base f0000000 segment 0 buses 0 - 255
PCI: BIOS Bug: MCFG area at f0000000 is not reserved in ACPI motherboard resources
PCI: Not using MMCONFIG
ACPI: EC:GPE=0x17, I/O: command/status=0x66, data=0x62
ACPI: EC: driver started in interrupt mode
ACPI: PCI Root Bridge [PCI0] (0000:00)
pci 0000:00:1f.0: quirk: region 0400-047f claimed by ICH6 ACPI/GPIO/TCO
pci 0000:00:1f.0: quirk: region 0500-053f claimed by ICH6 GPIO

This compares with the results reported on the linux-kernel mailing list
for a Mac-Mini...

ACPI: bus type pci registered
PCI: MCFG configuration 0: base e0000000 segment 0 buses 0 - 255
PCI: MCFG area at e0000000 reserved in E820
PCI: Using MMCONFIG at e0000000 - efffffff
PCI: Using configuration type 1 for base access
ACPI: EC: EC description table is found, configuring boot EC
ACPI: BIOS _OSI(Linux) query ignored via DMI
ACPI: Interpreter enabled
ACPI: (supports S0 S3 S4 S5)
ACPI: Using IOAPIC for interrupt routing
ACPI: EC: GPE = 0x17, I/O: command/status = 0x66, data = 0x62
ACPI: EC: driver started in poll mode
ACPI: PCI Root Bridge [PCI0] (0000:00)
pci 0000:00:1f.0: quirk: region 0400-047f claimed by ICH6 ACPI/GPIO/TCO
pci 0000:00:1f.0: quirk: region 0500-053f claimed by ICH6 GPIO
PCI: Transparent bridge - 0000:00:1e.0
ACPI: PCI Interrupt Routing Table [\_SB_.PCI0._PRT]
[...]

The critical difference is that on a 2nd generation MacBook Pro, we see...

PCI: MCFG configuration 0: base f0000000 segment 0 buses 0 - 255
PCI: Not using MMCONFIG

instead of...

PCI: MCFG configuration 0: base e0000000 segment 0 buses 0 - 255
PCI: MCFG area at e0000000 reserved in E820
PCI: Using MMCONFIG at e0000000 - efffffff

seen on a Mac-Mini. Could the failure be related to higher base address
on the MacBook Pro (which has 3GB)? Any suggestions on how to debug this
further would be most welcome.
 
Steps to reproduce:
Installed Fedora 9 x86_64
Built and installed stock unpatched 2.6.26 kernel. Same bug is seen in
Fedora's 2.6.26-131 kernel srpm built under Fedora 9.
Comment 1 Andrew Morton 2008-07-14 20:06:59 UTC
Marking as a regression.
Comment 2 Thomas Gleixner 2008-07-14 23:23:54 UTC
> Latest working kernel version: 2.6.25.10

Can you please provide the boot log for this kernel on the MacBook Pro
(second generation). The Mac-Mini one is not really helpful. 

> PCI: MCFG configuration 0: base f0000000 segment 0 buses 0 - 255
> PCI: BIOS Bug: MCFG area at f0000000 is not reserved in ACPI motherboard
> resources

This looks like the real problem.
Comment 3 Jack Howarth 2008-07-15 06:58:53 UTC
How can I generate a boot log when the boot process hangs? The first log section in the description of the bug is all of the boot information that I see. FYI, the use of pci=noacpi has no effect on the boot hang. I haven't had any luck finding
a vga=XXX setting that would decrease the font size and let me see more of the
boot messages. However I think I am seeing the critical ones starting with

ACPI: bus type pci registered
Comment 4 Jack Howarth 2008-07-15 07:22:28 UTC
Is this possibly due to the fact that I have 3 GB of ram installed? Is there
any kernel options that would fake out the kernel to believe the machine has
less than 3 GB (to avoid having to physically remove the ram)?
Comment 5 Thomas Gleixner 2008-07-15 14:21:04 UTC
> Is this possibly due to the fact that I have 3 GB of ram installed? Is there
> any kernel options that would fake out the kernel to believe the machine has
> less than 3 GB (to avoid having to physically remove the ram)?

mem=2G should do the trick
Comment 6 Jack Howarth 2008-07-15 22:34:37 UTC
In an attempt to figure out why MMCONFIG is not in use on a MacBook Pro v2
under 2.6.26-git2, I applied the following patch...


--- arch/x86/kernel/e820.c.org	2008-07-16 00:00:18.000000000 -0400
+++ arch/x86/kernel/e820.c	2008-07-16 01:00:57.000000000 -0400
@@ -87,10 +87,22 @@
 		struct e820entry *ei = &e820.map[i];
 
 		if (type && ei->type != type)
+                {
+                        printk(KERN_ERR "%d conflict for type %u vs %u in map\n", i, type, e820.map[i].type);
 			continue;
+                }
 		/* is the region (part) in overlap with the current region ?*/
 		if (ei->addr >= end || ei->addr + ei->size <= start)
+                {
+                        printk(KERN_ERR "%d addr %016Lx addr+size %016Lx start %016Lx end %016Lx \n", i,
+                       (unsigned long long) e820.map[i].addr,
+                       (unsigned long long)
+                       (e820.map[i].addr + e820.map[i].size),
+                       (unsigned long long) start,
+                       (unsigned long long) end);
+ 
 			continue;
+                }
 
 		/* if the region is at the beginning of <start,end> we move
 		 * start to the end of the region since it's ok until there

This produced the following output when booting the 2.6.26-git2 kernel...

PCI: BIOS Bug: MCFG area at f00000000 is not reserved in ACPI motherboard resources.
0 conflict for type 2 vs 1 in map
1 addr 000000000009fc00 addr+size 00000000000a0000 start 00000000f0000000 end 00000000ffffffff
2 addr 00000000000e0000 addr+size 0000000000100000 start 00000000f0000000 end 00000000ffffffff
3 conflict for type 2 vs 1 in map
4 conflict for type 2 vs 4 in map
5 conflict for type 2 vs 3 in map
6 conflict for type 2 vs 4 in map
7 conflict for type 2 vs 3 in map
8 addr 00000000bff00000 addr+size 00000000c0000000 start 00000000f0000000 end 00000000ffffffff

At this point the call to  e820_all_mapped() must be returning 0 because start wasn't greater than end.
Let me know if I can add any other printk's to help debug this problem of MMCONFIG not starting up.
Comment 7 Jack Howarth 2008-07-15 22:40:38 UTC
Created attachment 16831 [details]
dmesg file from successful linux 2.6.25.10 boot

I am attaching the dmesg file from a booting linux 2.6.25.10 on the same MacBook Pro v2 which fails to boot 2.6.26.
Comment 8 Jack Howarth 2008-07-16 15:48:02 UTC
Created attachment 16851 [details]
dmesg file for linux 2.6.25.10 boot with apic=verbose
Comment 9 Jack Howarth 2008-07-16 19:43:48 UTC
Created attachment 16853 [details]
 lspci -vvxxx output from under 2.6.25.10 on MacBook Pro v2
Comment 10 Jack Howarth 2008-07-16 23:23:23 UTC
Using the debug patches and your mmconfig patch as well as the kernel
options 'debug initcall_debug apic=verbose pci=routeirq", I see the following
messages that don't scroll off screen...

PCI: 0000:00:1f.1 reg 10 io port: [ 40d8, 40df]
PCI: 0000:00:1f.1 reg 14 io port: [ 40ec, 40ef]
PCI: 0000:00:1f.1 reg 18 io port: [ 40d0, 40d7]
PCI: 0000:00:1f.1 reg 1c io port: [ 40e8, 40eb]
PCI: 0000:00:1f.1 reg 20 io port: [ 40b0, 40bf]
pci 0000:00.1f.1: calling pci_fixup_transparent_bridge+0x0/0x2a
PCI: Found 0000:00:1f.2 [8086/27c4] 000101 00
PCI: 0000:00:1f.2 reg 10 io port: [ 40c8, 40cf]
PCI: 0000:00:1f.2 reg 14 io port: [ 40e4, 40e7]
PCI: 0000:00:1f.2 reg 18 io port: [ 40c0, 40c7]
PCI: 0000:00:1f.2 reg 1c io port: [ 40e0, 40e3]
PCI: 0000:00:1f.2 reg 20 io port: [ 40a0, 40af]
PCI: 0000:00:1f.2 reg 24 32bit mmio: [d8405000, d84053ff]
pci: 0000:00:1f.2 calling pci_fixup_transparent_bridge+0x0/0x2a
PCI: Found 0000:00:1f.3 [8086/27da] 000c05 00
PCI: 0000:00:1f.3 reg 20 io port: [ efa0, efbf]
PCI: 0000:00:1f.3: calling pci_fixup_transparent_bridge+0x0/0x2a
PCI: Fixups for bus 0000:00
PCI: Scanning behind PCI bridge 0000:00:01.0, config 010100, pass 0
PCI: Scanning bus 0000:01
PCI: Found 0000:01:00.0 [1002/71c5] 000300 00
PCI: 0000:00:01:00.0 reg 10 32bit mmio: [c0000000, cfffffff]
PCI: 0000:00:01:00.0 reg 14 io port: [ 3000, 30ff]
PCI: 0000:00:01:00.0 reg 18 32bit mmio: [d8300000, d830ffff]

...at which point the boot freezes.
Comment 11 Jack Howarth 2008-07-17 05:07:15 UTC
Created attachment 16864 [details]
.config used for linux 2.6.26 kernel builds on MacBookPro

This config file was generated from the .config file created by the Fedora development kernel-2.6.26-136 srpm and used with 'make oldconfig'.
Comment 12 Yinghai Lu 2008-09-04 13:44:36 UTC
please check if the problem still happen with current mainline or tip/master.
Comment 13 Alan 2010-01-19 19:43:25 UTC
Closing stale bug

Note You need to log in before you can comment on or make changes to this bug.