Skip to contentSkip to conversation

x86 git acpi issue?

PATCH Hiroshi Shimamoto 6 participants 9 messages May 3, 2008

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

Hi,

commit: afcab879e4044f952b7a031d5fd504e0feb7df41
visws: build fix

breaks something around acpi on 32bit.
In my test box scsi driver complains and the kernel doesn't boot.

I'm not sure, but when moving the acpi.o before legacy.o, it works.

arch/x86/pci/Makefile_32 +1 -1
@@ -6,11 +6,11 @@ obj-$(CONFIG_PCI_DIRECT)	+= direct.o
 obj-$(CONFIG_PCI_OLPC)		+= olpc.o
 
 pci-y				:= fixup.o
+pci-$(CONFIG_ACPI)		+= acpi.o
 pci-y				+= legacy.o irq.o
 
 pci-$(CONFIG_X86_VISWS)		:= visws.o fixup.o
 pci-$(CONFIG_X86_NUMAQ)		:= numa.o irq.o
-pci-$(CONFIG_ACPI)		+= acpi.o
 pci-$(CONFIG_NUMA)		+= mp_bus_to_node.o
 
 obj-y				+= $(pci-y) common.o early.o

of cause this isn't correct patch, it's just a workaround for me.

thanks,
Hiroshi Shimamoto

#1 lore raw
Quoted 25 lines · Hiroshi Shimamoto

Hi,

commit: afcab879e4044f952b7a031d5fd504e0feb7df41
visws: build fix

breaks something around acpi on 32bit.
In my test box scsi driver complains and the kernel doesn't boot.

I'm not sure, but when moving the acpi.o before legacy.o, it works.

arch/x86/pci/Makefile_32 +1 -1
@@ -6,11 +6,11 @@ obj-$(CONFIG_PCI_DIRECT)	+= direct.o
 obj-$(CONFIG_PCI_OLPC)		+= olpc.o
 
 pci-y				:= fixup.o
+pci-$(CONFIG_ACPI)		+= acpi.o
 pci-y				+= legacy.o irq.o
 
 pci-$(CONFIG_X86_VISWS)		:= visws.o fixup.o
 pci-$(CONFIG_X86_NUMAQ)		:= numa.o irq.o
-pci-$(CONFIG_ACPI)		+= acpi.o
 pci-$(CONFIG_NUMA)		+= mp_bus_to_node.o
 
 obj-y				+= $(pci-y) common.o early.o
Quoted 2 lines

of cause this isn't correct patch, it's just a workaround for me.

The problem is that := should be += for the VISWS and NUMAQ lines. There is already a patch queued up to fix that.

-hpa

#2 lore raw
Quoted 24 lines · H. Peter Anvin
Quoted 23 lines · Hiroshi Shimamoto

Hi,

commit: afcab879e4044f952b7a031d5fd504e0feb7df41
visws: build fix

breaks something around acpi on 32bit.
In my test box scsi driver complains and the kernel doesn't boot.

I'm not sure, but when moving the acpi.o before legacy.o, it works.

arch/x86/pci/Makefile_32 +1 -1
@@ -6,11 +6,11 @@ obj-$(CONFIG_PCI_DIRECT)	+= direct.o
 obj-$(CONFIG_PCI_OLPC)		+= olpc.o
 
 pci-y				:= fixup.o
+pci-$(CONFIG_ACPI)		+= acpi.o
 pci-y				+= legacy.o irq.o
 
 pci-$(CONFIG_X86_VISWS)		:= visws.o fixup.o
 pci-$(CONFIG_X86_NUMAQ)		:= numa.o irq.o
-pci-$(CONFIG_ACPI)		+= acpi.o
 pci-$(CONFIG_NUMA)		+= mp_bus_to_node.o
 
Quoted 8 lines
Quoted 4 lines
arch/x86/pci/Makefile_32 +0 -0
 obj-y				+= $(pci-y) common.o early.o

of cause this isn't correct patch, it's just a workaround for me.

The problem is that := should be += for the VISWS and NUMAQ lines. There is already a patch queued up to fix that.

OK, thanks.

Hiroshi Shimamoto

#3 lore raw
In reply to H. Peter Anvin
Quoted 23 lines · H. Peter Anvin
Quoted 22 lines · Hiroshi Shimamoto

Hi,

commit: afcab879e4044f952b7a031d5fd504e0feb7df41
visws: build fix

breaks something around acpi on 32bit.
In my test box scsi driver complains and the kernel doesn't boot.

I'm not sure, but when moving the acpi.o before legacy.o, it works.

arch/x86/pci/Makefile_32 +1 -1
@@ -6,11 +6,11 @@ obj-$(CONFIG_PCI_DIRECT)	+= direct.o
 obj-$(CONFIG_PCI_OLPC)		+= olpc.o
  pci-y				:= fixup.o
+pci-$(CONFIG_ACPI)		+= acpi.o
 pci-y				+= legacy.o irq.o
  pci-$(CONFIG_X86_VISWS)		:= visws.o fixup.o
 pci-$(CONFIG_X86_NUMAQ)		:= numa.o irq.o
-pci-$(CONFIG_ACPI)		+= acpi.o
 pci-$(CONFIG_NUMA)		+= mp_bus_to_node.o
  obj-y				+= $(pci-y) common.o early.o
 
Quoted 5 lines
Quoted 2 lines

of cause this isn't correct patch, it's just a workaround for me.

The problem is that := should be += for the VISWS and NUMAQ lines. There is already a patch queued up to fix that.

Commit afcab879e4044f952b7a031d5fd504e0feb7df41 is the commit where Ingo both _reverts_ his patch you describe in the x86 tree and additionally moves pci-$(CONFIG_ACPI) in exactly the way that causes this breakage.

In the 4 hours between Ingo doing the += change and Ingo reverting his own patch in the x86 tree it was in a pull request to Linus, so the "queued up" patch is in Linus' tree but reverted in the x86 tree.

I saw neither the patch changing the :='s to += on linux-kernel nor does "do not override the existing pci-y rule when adding visws or numaq rules." in the commit description give any indication what the actual problem was.

It's also not obvious why the revert with the subject "visws: build fix" also moved the pci-$(CONFIG_ACPI) causing this bug here.

Quoted 1 line

-hpa

cu
Adrian

Signature · 4 lines
       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed
#4 lore raw
Quoted 8 lines · Adrian Bunk

Commit afcab879e4044f952b7a031d5fd504e0feb7df41 is the commit where Ingo both _reverts_ his patch you describe in the x86 tree and additionally moves pci-$(CONFIG_ACPI) in exactly the way that causes this breakage.

In the 4 hours between Ingo doing the += change and Ingo reverting his own patch in the x86 tree it was in a pull request to Linus, so the "queued up" patch is in Linus' tree but reverted in the x86 tree.

Quoted 7 lines

I saw neither the patch changing the :='s to += on linux-kernel nor does "do not override the existing pci-y rule when adding visws or numaq rules." in the commit description give any indication what the actual problem was.

It's also not obvious why the revert with the subject "visws: build fix" also moved the pci-$(CONFIG_ACPI) causing this bug here.

Right. This Makefile is a nasty trap. I looked into it and the fix for now is below. Long term this Makefile trickery needs to be cleaned up to avoid trapping into this again.

Thanks,
tglx

--------->
Subject: x86: undo visws/numaq build changes
From: Thomas Gleixner <[email protected]>
Date: Fri, 02 May 2008 21:24:30 +0200

arch/x86/pci/Makefile_32 has a nasty detail. VISWS and NUMAQ builds override the generic pci-y rules. This needs a proper cleanup, but that needs more thoughts. Undo

commit 895d30935ebe05f192e844792668bf8d19deaae7
x86: numaq fix
do not override the existing pci-y rule when adding visws or numaq rules.

and add a comment to the Makefile.

Remove the srat stub code in discontig_32.c to allow a proper NUMAQ build.

---
 arch/x86/mm/discontig_32.c |   26 --------------------------
 arch/x86/pci/Makefile_32   |    8 ++++++--
 2 files changed, 6 insertions(+), 28 deletions(-)
arch/x86/mm/discontig_32.c +0 -26
Index: linux-2.6/arch/x86/mm/discontig_32.c
===================================================================
@@ -476,29 +476,3 @@ int memory_add_physaddr_to_nid(u64 addr)
 
 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
 #endif
-
-#ifndef CONFIG_HAVE_ARCH_PARSE_SRAT
-/*
- * XXX FIXME: Make SLIT table parsing available to 32-bit NUMA
- *
- * These stub functions are needed to compile 32-bit NUMA when SRAT is
- * not set. There are functions in srat_64.c for parsing this table
- * and it may be possible to make them common functions.
- */
-void acpi_numa_slit_init (struct acpi_table_slit *slit)
-{
-	printk(KERN_INFO "ACPI: No support for parsing SLIT table\n");
-}
-
-void acpi_numa_processor_affinity_init (struct acpi_srat_cpu_affinity *pa)
-{
-}
-
-void acpi_numa_memory_affinity_init (struct acpi_srat_mem_affinity *ma)
-{
-}
-
-void acpi_numa_arch_fixup(void)
-{
-}
-#endif /* CONFIG_HAVE_ARCH_PARSE_SRAT */
arch/x86/pci/Makefile_32 +6 -2
Index: linux-2.6/arch/x86/pci/Makefile_32
===================================================================
@@ -9,8 +9,12 @@ pci-y				:= fixup.o
 pci-$(CONFIG_ACPI)		+= acpi.o
 pci-y				+= legacy.o irq.o
 
-pci-$(CONFIG_X86_VISWS)		+= visws.o fixup.o
-pci-$(CONFIG_X86_NUMAQ)		+= numa.o irq.o
+# Careful: VISWS and NUMAQ overrule the pci-y above. The colons are
+# therefor correct. This needs a proper fix by distangling the code.
+pci-$(CONFIG_X86_VISWS)		:= visws.o fixup.o
+pci-$(CONFIG_X86_NUMAQ)		:= numa.o irq.o
+
+# Necessary for NUMAQ as well
 pci-$(CONFIG_NUMA)		+= mp_bus_to_node.o
 
 obj-y				+= $(pci-y) common.o early.o
#5 lore raw
Quoted 14 lines · Thomas Gleixner
Quoted 13 lines · Adrian Bunk

Commit afcab879e4044f952b7a031d5fd504e0feb7df41 is the commit where Ingo both _reverts_ his patch you describe in the x86 tree and additionally moves pci-$(CONFIG_ACPI) in exactly the way that causes this breakage.

In the 4 hours between Ingo doing the += change and Ingo reverting his own patch in the x86 tree it was in a pull request to Linus, so the "queued up" patch is in Linus' tree but reverted in the x86 tree.

I saw neither the patch changing the :='s to += on linux-kernel nor does "do not override the existing pci-y rule when adding visws or numaq rules." in the commit description give any indication what the actual problem was.

Quoted 6 lines
Quoted 2 lines

It's also not obvious why the revert with the subject "visws: build fix" also moved the pci-$(CONFIG_ACPI) causing this bug here.

Right. This Makefile is a nasty trap. I looked into it and the fix for now is below. Long term this Makefile trickery needs to be cleaned up to avoid trapping into this again.

Thanks Thomas, i messed up that commit - i should have known that the NUMAQ build fix looked _too_ easy ;-)

Ingo

#6 lore raw
Quoted 6 lines · Ingo Molnar
Quoted 3 lines

Right. This Makefile is a nasty trap. I looked into it and the fix for now is below. Long term this Makefile trickery needs to be cleaned up to avoid trapping into this again.

Thanks Thomas, i messed up that commit - i should have known that the NUMAQ build fix looked _too_ easy ;-)

Also the move of
pci-$(CONFIG_ACPI) += acpi.o
is what broke things.

There is a nasty init function ordering problem as well between

pci-$(CONFIG_ACPI) += acpi.o
pci-y += legacy.o irq.o

Sigh.

tglx

#7 lore raw
Quoted 12 lines · Thomas Gleixner
Quoted 6 lines · Ingo Molnar
Quoted 3 lines

Right. This Makefile is a nasty trap. I looked into it and the fix for now is below. Long term this Makefile trickery needs to be cleaned up to avoid trapping into this again.

Thanks Thomas, i messed up that commit - i should have known that the NUMAQ build fix looked _too_ easy ;-)

Also the move of
pci-$(CONFIG_ACPI) += acpi.o
is what broke things.

Quoted 6 lines

There is a nasty init function ordering problem as well between

pci-$(CONFIG_ACPI) += acpi.o
pci-y += legacy.o irq.o

Sigh.

This is the kind of stuff that has prevented me from unifying the two Makefiles.
But if no-one beats me I will give it a shot soon. It will be a microsteps because we will likeky hit such issues.

Sam

#8 lore raw
Quoted 15 lines · Sam Ravnborg
Quoted 14 lines · Thomas Gleixner
Quoted 6 lines · Ingo Molnar
Quoted 3 lines

Right. This Makefile is a nasty trap. I looked into it and the fix for now is below. Long term this Makefile trickery needs to be cleaned up to avoid trapping into this again.

Thanks Thomas, i messed up that commit - i should have known that the NUMAQ build fix looked _too_ easy ;-)

Also the move of
pci-$(CONFIG_ACPI) += acpi.o
is what broke things.

There is a nasty init function ordering problem as well between

Quoted 8 lines
Quoted 4 lines

pci-$(CONFIG_ACPI) += acpi.o
pci-y += legacy.o irq.o

Sigh.

This is the kind of stuff that has prevented me from unifying the two Makefiles. But if no-one beats me I will give it a shot soon. It will be a microsteps because we will likeky hit such issues.

yeah, it needs to be micro-steps (nice word! :) as this stuff is very subtle.

Ingo