Re: stuck with 2.6.23.14 on x86_64

On Wed, 13 Feb 2008 10:21:23 +0100 Fabio Coatti wrote:

Alle martedì 12 febbraio 2008, Rafael J. Wysocki ha scritto:
On Tuesday, 12 of February 2008, Fabio Coatti wrote:
Alle martedì 12 febbraio 2008, Randy Dunlap ha scritto:
On Tue, 12 Feb 2008 15:03:41 +0100 Fabio Coatti wrote:
Hi all,
I'm stuck in a weird situation: I'm unable to go beyond 2.6.23.14, so
to fix the splice bug I've had to apply by hand the patch. (x86_64)

Basically, with 2.6.24.2 (the same with 2.6.24 and .1), tha machine
won't boot due to a problem with cciss driver, that prevents to find
the / partition. (bug described here: Kernel Bug Tracker Bug 9859
http://bugzilla.kernel.org/show_bug.cgi?id=9859 );

With kernels 2.6.23, the lastest that I can compile is 2.6.23.14;
starting from .15 (and .16) I get this message:

======================
UPD include/linux/compile.h
CC init/version.o
LD init/built-in.o
LD .tmp_vmlinux1
drivers/built-in.o: In function `acpi_init':
bus.c:(.init.text+0x1713): undefined reference to `pm_flags'
bus.c:(.init.text+0x1756): undefined reference to `pm_flags'
======================

All .config are the same, (make oldconfig) beside the obvious
differences between .23 and .24

Hardware: x86_64 AMD 2216HE
SCSI controller: HP Smart Array E200i Controller
Compiler: gcc (GCC) 4.1.1
binutils: 2.16.1

On a x86 machine, Intel(R) Xeon(TM) CPU 3.20GHz
with a cciss0: HP Smart Array 6i Controller,
the 2.6.24.2 compiles just fine and works, so the cciss problems
seems related only to E200i controller.

Right now, on AMD64 machines, I'm forced to patch by hand the kernel,
that's quite uncomfortable :)

Can someone point me in the right direction to get out of this
situation? Of course I can provide any further information. (.config
not inlcuded now to avoid cluttering )

Thanks for any answer.

a/ send .config file for the build problem above
b/ How do you download and/or apply 2.6.23.{15,16} ?
Full tarball or base tarball + patches?
If patches, what base tree are they applied to?

full tarball from kernel.org (.16), tried also applying patches to 2.6.23
vanilla.(.15,.16) Same process leads to successful compilation for .14

You're not supposed to have CONFIG_PM unset and CONFIG_ACPI set at the same
time. The oldconfig generation must have gone wrong at one point.

Maybe it's not supposed to have this situation, but maybe you should tell this
to the kernel itself :)

# zcat /proc/config.gz | egrep "PM|ACPI"
CONFIG_X86_64_ACPI_NUMA=y
# CONFIG_PM is not set
CONFIG_ACPI=y
# CONFIG_ACPI_PROCFS is not set

# uname -rv
2.6.23.12 #3 SMP Tue Feb 12 11:22:16 CET 2008

And you can easily get this situation from menuconfig: just fire up make
menuconfig without any .config, go to power management options and turn
off "Power Management support". exit and look at .config:

# CONFIG_PM is not set
CONFIG_SUSPEND_SMP_POSSIBLE=y
CONFIG_HIBERNATION_SMP_POSSIBLE=y
CONFIG_ACPI=y

Maybe if this is not supposed to be the right situation, some dependencies are
not respected... (tested on .16)


Yes, this is yet another select problem. X86_64_NUMA_ACPI
selects ACPI, even though PM is not enabled. :(

So you have a manual fix: just enable CONFIG_PM.

Is this specific config important to you? I.e., do you want/need
NUMA support? If so, you'll just need to continue enabling CONFIG_PM.

I made a patch that prevents the select problem (see below [now added]).

The "select PCI" in the X86_64_ACPI_NUMA config option is evil. :(


Anyway, with .16 and CONFIG_PM set, I get this :
===================
WARNING: vmlinux.o(.text+0x1f6395): Section mismatch: reference
to .init.data.14:pxm_to_node_map (between 'pxm_to_node' and 'node_to_pxm')
WARNING: vmlinux.o(.text+0x1f63a7): Section mismatch: reference
to .init.data.14:node_to_pxm_map (between 'node_to_pxm'
and '__acpi_map_pxm_to_node')
WARNING: vmlinux.o(.text+0x1f63b2): Section mismatch: reference
to .init.data.14:pxm_to_node_map (between '__acpi_map_pxm_to_node'
and 'acpi_get_pxm')
WARNING: vmlinux.o(.text+0x1f63bc): Section mismatch: reference
to .init.data.14:node_to_pxm_map (between '__acpi_map_pxm_to_node'
and 'acpi_get_pxm')
WARNING: vmlinux.o(.text+0x1f6412): Section mismatch: reference
to .init.data.14:pxm_to_node_map (between 'acpi_map_pxm_to_node'
and 'acpi_get_node')
WARNING: vmlinux.o(.text+0x1f6458): Section mismatch: reference
to .init.data.14:pxm_to_node_map (between 'acpi_map_pxm_to_node'
and 'acpi_get_node')
WARNING: vmlinux.o(.text+0x1f645f): Section mismatch: reference
to .init.data.14:node_to_pxm_map (between 'acpi_map_pxm_to_node'
and 'acpi_get_node')
=========================

Anyway, it booted and so far it runs without "splice" problem, thanks.


---
From: Randy Dunlap <randy.dunlap@xxxxxxxxxx>

Fix a kconfig 'select' problem: X86_64_NUMA_ACPI selects ACPI,
so ACPI could be enabled even when PM is not enabled.
This causes a build failure, so make X86_64_NUMA_ACPI depend on PM.

Signed-off-by: Randy Dunlap <randy.dunlap@xxxxxxxxxx>
---
arch/x86_64/Kconfig | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

--- linux-2.6.23.16.orig/arch/x86_64/Kconfig
+++ linux-2.6.23.16/arch/x86_64/Kconfig
@@ -367,7 +367,7 @@ config NODES_SHIFT

config X86_64_ACPI_NUMA
bool "ACPI NUMA detection"
- depends on NUMA
+ depends on NUMA && PM
select ACPI
select PCI
select ACPI_NUMA

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Relevant Pages

  • Re: stuck with 2.6.23.14 on x86_64
    ... to fix the splice bug I've had to apply by hand the patch. ... All .config are the same, ... SCSI controller: ... Right now, on AMD64 machines, I'm forced to patch by hand the kernel, ...
    (Linux-Kernel)
  • Re: 2.6.5-rc3-mm4 x86_64 sched domains patch
    ... > The patch doesn't apply against 2.6.5-mm5 anymore. ... bool "Two level sched domains" ... depends on NUMA ... config ALPHA_LARGE_VMALLOC ...
    (Linux-Kernel)
  • Re: stuck with 2.6.23.14 on x86_64
    ... the splice bug I've had to apply by hand the patch. ... All .config are the same, beside the obvious differences ... SCSI controller: ... Right now, on AMD64 machines, I'm forced to patch by hand the kernel, that's ...
    (Linux-Kernel)
  • Re: [PATCH 7/12] acpi: fix another compile warning
    ... How were you able to produce a .config with CONFIG_ACPI_BLACKLIST_YEAR not defined? ... On x86_64 ACPI keeps enabled by default. ... Please test if this single patch fixes all your multiple ACPI related build errors. ... NUMA more user-friendly. ...
    (Linux-Kernel)
  • Re: [BUGFIX] NULL pointer dereference in __vm_enough_memory()
    ... | I'm not sure if this patch is the right fix for the bug. ... | I'm running debian/sid. ... The .config is attached. ...
    (Linux-Kernel)