lguest: the asm offsets

PATCH In review [email protected] 6 participants 11 messages May 9, 2007 – Jun 13, 2007

Subsystems (2)

Kbuild virtio core

lore

Share & tools

lore

[email protected]

author

2007-05-09 09:51 UTC #0 lore raw

PATCH 4/9 lguest: the asm offsets

From: Rusty Russell <[email protected]>

This is the structure offsets required by lg.ko's switcher.S.

Unfortunately we don't have infrastructure for private asm-offsets creation.

Signed-off-by Rusty Russell <[email protected]>

Cc Andi Kleen <[email protected]>

Signed-off-by Andrew Morton <[email protected]>

---

 arch/i386/kernel/asm-offsets.c |   19 +++++++++++++++++++
 1 file changed, 19 insertions(+)

arch/i386/kernel/asm-offsets.c renamed from arch/i386/kernel/asm-offsets.c~lguest-the-asm-offsets +19 -0

diff -puN arch/i386/kernel/asm-offsets.c~lguest-the-asm-offsets arch/i386/kernel/asm-offsets.c
@@ -17,6 +17,10 @@
 #include <asm/pgtable.h>
 #include <asm/thread_info.h>
 #include <asm/elf.h>
+#ifdef CONFIG_LGUEST_GUEST
+#include <linux/lguest.h>
+#include "../../../drivers/lguest/lg.h"
+#endif
 
 #define DEFINE(sym, val) \
         asm volatile("\n->" #sym " %0 " #val : : "i" (val))

@@ -116,4 +120,19 @@ void foo(void)
 	OFFSET(PARAVIRT_iret, paravirt_ops, iret);
 	OFFSET(PARAVIRT_read_cr0, paravirt_ops, read_cr0);
 #endif
+
+#ifdef CONFIG_LGUEST_GUEST
+	BLANK();
+	OFFSET(LGUEST_DATA_irq_enabled, lguest_data, irq_enabled);
+	OFFSET(LGUEST_PAGES_host_gdt_desc, lguest_pages, state.host_gdt_desc);
+	OFFSET(LGUEST_PAGES_host_idt_desc, lguest_pages, state.host_idt_desc);
+	OFFSET(LGUEST_PAGES_host_cr3, lguest_pages, state.host_cr3);
+	OFFSET(LGUEST_PAGES_host_sp, lguest_pages, state.host_sp);
+	OFFSET(LGUEST_PAGES_guest_gdt_desc, lguest_pages,state.guest_gdt_desc);
+	OFFSET(LGUEST_PAGES_guest_idt_desc, lguest_pages,state.guest_idt_desc);
+	OFFSET(LGUEST_PAGES_guest_gdt, lguest_pages, state.guest_gdt);
+	OFFSET(LGUEST_PAGES_regs_trapnum, lguest_pages, regs.trapnum);
+	OFFSET(LGUEST_PAGES_regs_errcode, lguest_pages, regs.errcode);
+	OFFSET(LGUEST_PAGES_regs, lguest_pages, regs);
+#endif
 }

Sam Ravnborg

2007-05-09 20:45 UTC #1 lore raw

Quoted 6 lines · On Wed, May 09, 2007 at 02:51:36AM -0700, [email protected] wrote:

From: Rusty Russell <[email protected]>
This is the structure offsets required by lg.ko's switcher.S.
Unfortunately we don't have infrastructure for private asm-offsets creation.

Someone maybe it was Oleg? did an OK job to abstract out asm-offset generation. I asked to have it rebased but never heard anything back.

I would like to get this generic approach in and then let lguest use it. But I do not see this as a reason to hold back inclusion in -linus.

But could we please get rid of the "../../../" in the include....

Sam

Rusty Russell

2007-05-10 00:09 UTC #2 lore raw

Quoted 5 lines · Sam Ravnborg

Quoted 4 lines · On Wed, May 09, 2007 at 02:51:36AM -0700, [email protected] wrote:
From: Rusty Russell <[email protected]>
This is the structure offsets required by lg.ko's switcher.S.

Quoted 7 lines

Quoted 2 lines
Unfortunately we don't have infrastructure for private asm-offsets creation.
Someone maybe it was Oleg? did an OK job to abstract out asm-offset generation. I asked to have it rebased but never heard anything back.
I would like to get this generic approach in and then let lguest use it. But I do not see this as a reason to hold back inclusion in -linus.

Hi Sam,

Yeah, I never heard back either. And I'm reluctant to try to fix it
myself: I've never managed to patch the build system without causing you to rewrite it better 8)

Quoted 1 line

But could we please get rid of the "../../../" in the include....

Sure: but how? It's a private internal include for the lg.ko module.

Thanks!
Rusty.

Oleg Verych

2007-06-11 12:38 UTC #3 lore raw

* Date: Thu, 10 May 2007 10:09:39 +1000

Quoted 12 lines · Sam Ravnborg

Quoted 10 lines · On Wed, May 09, 2007 at 02:51:36AM -0700, [email protected] wrote:
Quoted 6 lines
From: Rusty Russell <[email protected]>
This is the structure offsets required by lg.ko's switcher.S.
Unfortunately we don't have infrastructure for private asm-offsets creation.
Someone maybe it was Oleg? did an OK job to abstract out asm-offset generation. I asked to have it rebased but never heard anything back.

Quoted 8 lines

Quoted 2 lines
I would like to get this generic approach in and then let lguest use it. But I do not see this as a reason to hold back inclusion in -linus.
Hi Sam,
Yeah, I never heard back either. And I'm reluctant to try to fix it
myself: I've never managed to patch the build system without causing you to rewrite it better 8)

If it's still open, i can try, after getting back from RL and 28k LKML backlog ;)
____

Sam Ravnborg

2007-06-11 12:58 UTC #4 lore raw

Quoted 16 lines · Oleg Verych

* Date: Thu, 10 May 2007 10:09:39 +1000
Quoted 15 lines · Sam Ravnborg
Quoted 12 lines · On Wed, May 09, 2007 at 02:51:36AM -0700, [email protected] wrote:
Quoted 6 lines
From: Rusty Russell <[email protected]>
This is the structure offsets required by lg.ko's switcher.S.
Unfortunately we don't have infrastructure for private asm-offsets creation.
Someone maybe it was Oleg? did an OK job to abstract out asm-offset generation. I asked to have it rebased but never heard anything back.
I would like to get this generic approach in and then let lguest use it. But I do not see this as a reason to hold back inclusion in -linus.

Quoted 8 lines

Quoted 5 lines
Hi Sam,
Yeah, I never heard back either. And I'm reluctant to try to fix it
myself: I've never managed to patch the build system without causing you to rewrite it better 8)
If it's still open, i can try, after getting back from RL and 28k LKML backlog ;)

Hi Olg.

Please do so.
Thanks,
Sam

Oleg Verych

2007-06-12 10:20 UTC #5 lore raw

PATCH 1/1 kbuild: remember ARCH in the object directory

[patch] kbuild: remember ARCH in the object directory

- rephrase some related (misspelled) comments,
- remove all trailing whitespace in the top Makefile,

* remember ARCH in the output directory, thus making build of i386 on amd64 *actually* convenient,

- harmless refactoring of the Makefile generating.

Signed-off-by Oleg Verych <[email protected]>

---
-o--=O`C /. .\
#oo'L O o
<___=E M ^--

Just lazy typing make ARCH=i386 every time i want to check lguest.

 Makefile           |   17 ++++++++---------
 scripts/mkmakefile |   42 ++++++++++++++++++++++++------------------
 2 files changed, 32 insertions(+), 27 deletions(-)

scripts/mkmakefile +24 ~~-18~~

Index: linux-2.6.22-rc4-mm2/scripts/mkmakefile
===================================================================
@@ -1,6 +1,6 @@
-#!/bin/sh
+#!/bin/sh -e
 # Generates a small Makefile used in the root of the output
-# directory, to allow make to be started from there.
-# The Makefile also allow for more convinient build of external modules
+# directory, to allow kernel build to be started from there.
+# External modules can't built with it as well.
 
 # Usage

@@ -9,28 +9,34 @@
 # $3 - version
 # $4 - patchlevel
+# $5 - ARCH
 
+set -e
 
-test ! -r $2/Makefile -o -O $2/Makefile || exit 0
 echo "  GEN     $2/Makefile"
+echo "#
+# Automatically generated by
+# $0
+#
+VERSION		= $3
+PATCHLEVEL	= $4
+
+KERNELSRC	= $1
+KERNELOUTPUT	= $2
+ARCH		= $5
+" >$2/Makefile
 
-cat << EOF > $2/Makefile
-# Automatically generated by $0: don't edit
+cat << "EOF" >> $2/Makefile
+MAKEFLAGS	+= -rR --no-print-directory
 
-VERSION = $3
-PATCHLEVEL = $4
+export ARCH
 
-KERNELSRC    := $1
-KERNELOUTPUT := $2
-
-MAKEFLAGS += --no-print-directory
+all:
+	$(MAKE) -C $(KERNELSRC) O=$(KERNELOUTPUT)
 
-.PHONY: all \$(MAKECMDGOALS)
+$(filter-out all Makefile,$(MAKECMDGOALS)) %/:
+	$(MAKE) -C $(KERNELSRC) O=$(KERNELOUTPUT) $@
 
-all:
-	\$(MAKE) -C \$(KERNELSRC) O=\$(KERNELOUTPUT)
+.PHONY: all $(MAKECMDGOALS)
 
 Makefile:;
-
-\$(filter-out all Makefile,\$(MAKECMDGOALS)) %/:
-	\$(MAKE) -C \$(KERNELSRC) O=\$(KERNELOUTPUT) \$@
 EOF

Makefile +8 -9

Index: linux-2.6.22-rc4-mm2/Makefile
===================================================================
@@ -156,5 +156,5 @@ export srctree objtree VPATH TOPDIR
 # first, and if a usermode build is happening, the "ARCH=um" on the command
 # line overrides the setting of ARCH below.  If a native build is happening,
-# then ARCH is assigned, getting whatever value it gets normally, and 
+# then ARCH is assigned, getting whatever value it gets normally, and
 # SUBARCH is subsequently ignored.

@@ -242,5 +242,5 @@ export KBUILD_CHECKSRC KBUILD_SRC KBUILD
 #
 # If $(quiet) is empty, the whole command will be printed.
-# If it is set to "quiet_", only the short version will be printed. 
+# If it is set to "quiet_", only the short version will be printed.
 # If it is set to "silent_", nothing will be printed at all, since
 # the variable $(silent_cmd_cc_o_c) doesn't exist.

@@ -352,11 +352,10 @@ scripts/basic/%: scripts_basic ;
 
 PHONY += outputmakefile
-# outputmakefile generates a Makefile in the output directory, if using a
-# separate output directory. This allows convenient use of make in the
-# output directory.
 outputmakefile:
+# generate a Makefile in the output directory, if using a
+# separate output directory; allows convenient use of make there
 ifneq ($(KBUILD_SRC),)
 	$(Q)$(CONFIG_SHELL) $(srctree)/scripts/mkmakefile \
-	    $(srctree) $(objtree) $(VERSION) $(PATCHLEVEL)
+	    $(srctree) $(objtree) $(VERSION) $(PATCHLEVEL) $(ARCH)
 endif

@@ -699,5 +698,5 @@ endef
 cmd_ksym_ld = $(cmd_vmlinux__)
 define rule_ksym_ld
-	: 
+	:
 	+$(call cmd,vmlinux_version)
 	$(call cmd,vmlinux__)

@@ -750,5 +749,5 @@ endif
 	$(Q)rm -f .old_version
 
-# The actual objects are generated when descending, 
+# The actual objects are generated when descending,
 # make sure no implicit rule kicks in
 $(sort $(vmlinux-init) $(vmlinux-main)) $(vmlinux-lds): $(vmlinux-dirs) ;

@@ -1454,5 +1453,5 @@ endif
 	$(Q)$(MAKE) -f $(srctree)/scripts/Makefile.modpost
 
-# FIXME Should go into a make.lib or something 
+# FIXME Should go into a make.lib or something
 # ===========================================================================

Randy Dunlap

2007-06-12 15:04 UTC #6 lore raw

Re: [patch] kbuild: remember ARCH in the object directory

Quoted 8 lines · On Tue, 12 Jun 2007 12:31:56 +0200 Oleg Verych wrote:

[patch] kbuild: remember ARCH in the object directory
- rephrase some related (misspelled) comments,
- remove all trailing whitespace in the top Makefile,
* remember ARCH in the output directory, thus making build of i386 on amd64 *actually* convenient,

Quoted 6 lines

- harmless refactoring of the Makefile generating.
Signed-off-by Oleg Verych <[email protected]>
---
Just lazy typing make ARCH=i386 every time i want to check lguest.

Hi Oleg,

So does this only work when O= is used?

I had tried to modify Makefile(s) to use the include/asm symlink if it exists, so that 'make ARCH=i386' is not needed after the symlink is created, but I never got that to work.

Thanks, this is useful in any case (not yet tested though).

Quoted 58 lines

 Makefile           |   17 ++++++++---------
 scripts/mkmakefile |   42 ++++++++++++++++++++++++------------------
 2 files changed, 32 insertions(+), 27 deletions(-)

Index: linux-2.6.22-rc4-mm2/scripts/mkmakefile
===================================================================

scripts/mkmakefile +23 ~~-13~~

@@ -1,6 +1,6 @@
-#!/bin/sh
+#!/bin/sh -e
 # Generates a small Makefile used in the root of the output
-# directory, to allow make to be started from there.
-# The Makefile also allow for more convinient build of external modules
+# directory, to allow kernel build to be started from there.
+# External modules can't built with it as well.
 
 # Usage

@@ -9,28 +9,34 @@
 # $3 - version
 # $4 - patchlevel
+# $5 - ARCH
 
+set -e
 
-test ! -r $2/Makefile -o -O $2/Makefile || exit 0
 echo "  GEN     $2/Makefile"
+echo "#
+# Automatically generated by
+# $0
+#
+VERSION		= $3
+PATCHLEVEL	= $4
+
+KERNELSRC	= $1
+KERNELOUTPUT	= $2
+ARCH		= $5
+" >$2/Makefile
 
-cat << EOF > $2/Makefile
-# Automatically generated by $0: don't edit
+cat << "EOF" >> $2/Makefile
+MAKEFLAGS	+= -rR --no-print-directory
 
-VERSION = $3
-PATCHLEVEL = $4
+export ARCH
 
-KERNELSRC    := $1
-KERNELOUTPUT := $2
-
-MAKEFLAGS += --no-print-directory
+all:
+	$(MAKE) -C $(KERNELSRC) O=$(KERNELOUTPUT)
 
-.PHONY: all \$(MAKECMDGOALS)
+$(filter-out all Makefile,$(MAKECMDGOALS)) %/:
+	$(MAKE) -C $(KERNELSRC) O=$(KERNELOUTPUT) $@

scripts/mkmakefile line 42

Quoted 10 lines

scripts/mkmakefile +1 -5

 
-all:
-	\$(MAKE) -C \$(KERNELSRC) O=\$(KERNELOUTPUT)
+.PHONY: all $(MAKECMDGOALS)
 
 Makefile:;
-
-\$(filter-out all Makefile,\$(MAKECMDGOALS)) %/:
-	\$(MAKE) -C \$(KERNELSRC) O=\$(KERNELOUTPUT) \$@
 EOF

---
~Randy
*** Remember to use Documentation/SubmitChecklist when testing your code ***

GEN /scsi/linsrc/linux-2.6.22-rc4-git4/BLD/Makefile
scripts/kconfig/conf -s arch/x86_64/Kconfig #... many config warning lines
*
* Restart config...
*
*
* Processor type and features
*
Subarchitecture Type

Quoted 1 line

1. PC-compatible (X86_PC)

2. Support for ScaleMP vSMP (X86_VSMP) (NEW)
choice[1-2]: aborted!

Console input/output is redirected. Run 'make oldconfig' to update configuration.

make[3]: *** [silentoldconfig] Error 1 make[2]: *** [silentoldconfig] Error 2 make[1]: *** [include/config/auto.conf] Error 2
make: *** [all] Error 2

I.e., see the line above that says:
scripts/kconfig/conf -s arch/x86_64/Kconfig

It defaulted to ARCH=x86_64 even though BLD/Makefile contained:

ARCH = i386

What did I do wrong??

Quoted 112 lines

Index: linux-2.6.22-rc4-mm2/scripts/mkmakefile
===================================================================

scripts/mkmakefile +24 ~~-18~~

@@ -1,6 +1,6 @@
-#!/bin/sh
+#!/bin/sh -e
 # Generates a small Makefile used in the root of the output
-# directory, to allow make to be started from there.
-# The Makefile also allow for more convinient build of external modules
+# directory, to allow kernel build to be started from there.
+# External modules can't built with it as well.
 
 # Usage

@@ -9,28 +9,34 @@
 # $3 - version
 # $4 - patchlevel
+# $5 - ARCH
 
+set -e
 
-test ! -r $2/Makefile -o -O $2/Makefile || exit 0
 echo "  GEN     $2/Makefile"
+echo "#
+# Automatically generated by
+# $0
+#
+VERSION		= $3
+PATCHLEVEL	= $4
+
+KERNELSRC	= $1
+KERNELOUTPUT	= $2
+ARCH		= $5
+" >$2/Makefile
 
-cat << EOF > $2/Makefile
-# Automatically generated by $0: don't edit
+cat << "EOF" >> $2/Makefile
+MAKEFLAGS	+= -rR --no-print-directory
 
-VERSION = $3
-PATCHLEVEL = $4
+export ARCH
 
-KERNELSRC    := $1
-KERNELOUTPUT := $2
-
-MAKEFLAGS += --no-print-directory
+all:
+	$(MAKE) -C $(KERNELSRC) O=$(KERNELOUTPUT)
 
-.PHONY: all \$(MAKECMDGOALS)
+$(filter-out all Makefile,$(MAKECMDGOALS)) %/:
+	$(MAKE) -C $(KERNELSRC) O=$(KERNELOUTPUT) $@
 
-all:
-	\$(MAKE) -C \$(KERNELSRC) O=\$(KERNELOUTPUT)
+.PHONY: all $(MAKECMDGOALS)
 
 Makefile:;
-
-\$(filter-out all Makefile,\$(MAKECMDGOALS)) %/:
-	\$(MAKE) -C \$(KERNELSRC) O=\$(KERNELOUTPUT) \$@
 EOF

Index: linux-2.6.22-rc4-mm2/Makefile
===================================================================

Makefile +7 -8

@@ -156,5 +156,5 @@ export srctree objtree VPATH TOPDIR
 # first, and if a usermode build is happening, the "ARCH=um" on the command
 # line overrides the setting of ARCH below.  If a native build is happening,
-# then ARCH is assigned, getting whatever value it gets normally, and 
+# then ARCH is assigned, getting whatever value it gets normally, and
 # SUBARCH is subsequently ignored.

@@ -242,5 +242,5 @@ export KBUILD_CHECKSRC KBUILD_SRC KBUILD
 #
 # If $(quiet) is empty, the whole command will be printed.
-# If it is set to "quiet_", only the short version will be printed. 
+# If it is set to "quiet_", only the short version will be printed.
 # If it is set to "silent_", nothing will be printed at all, since
 # the variable $(silent_cmd_cc_o_c) doesn't exist.

@@ -352,11 +352,10 @@ scripts/basic/%: scripts_basic ;
 
 PHONY += outputmakefile
-# outputmakefile generates a Makefile in the output directory, if using a
-# separate output directory. This allows convenient use of make in the
-# output directory.
 outputmakefile:
+# generate a Makefile in the output directory, if using a
+# separate output directory; allows convenient use of make there
 ifneq ($(KBUILD_SRC),)
 	$(Q)$(CONFIG_SHELL) $(srctree)/scripts/mkmakefile \
-	    $(srctree) $(objtree) $(VERSION) $(PATCHLEVEL)
+	    $(srctree) $(objtree) $(VERSION) $(PATCHLEVEL) $(ARCH)
 endif

@@ -699,5 +698,5 @@ endef
 cmd_ksym_ld = $(cmd_vmlinux__)
 define rule_ksym_ld
-	: 
+	:
 	+$(call cmd,vmlinux_version)
 	$(call cmd,vmlinux__)

@@ -750,5 +749,5 @@ endif
 	$(Q)rm -f .old_version
 
-# The actual objects are generated when descending, 
+# The actual objects are generated when descending,
 # make sure no implicit rule kicks in
 $(sort $(vmlinux-init) $(vmlinux-main)) $(vmlinux-lds): $(vmlinux-dirs) ;

Makefile endif · line 1457

Quoted 6 lines

Makefile +1 -1

@@ -1454,5 +1453,5 @@ endif
 	$(Q)$(MAKE) -f $(srctree)/scripts/Makefile.modpost
 
-# FIXME Should go into a make.lib or something 
+# FIXME Should go into a make.lib or something
 # ===========================================================================

---
~Randy
*** Remember to use Documentation/SubmitChecklist when testing your code ***

Oleg Verych

2007-06-12 17:07 UTC #8 lore raw

Re: [patch] kbuild: remember ARCH in the object directory

On Tue, Jun 12, 2007 at 09:12:09AM -0700, Randy Dunlap wrote:
[]

Quoted 23 lines

I did:
# build a linux-2.6.22-rc4-git4 kernel source tree & cd into it
# mkdir BLD
$ make ARCH=i386 O=BLD defconfig
$ make -j4 O=BLD all >BLD/bld.out 2>&1
and it cried to me:
GEN /scsi/linsrc/linux-2.6.22-rc4-git4/BLD/Makefile
scripts/kconfig/conf -s arch/x86_64/Kconfig #... many config warning lines
*
* Restart config...
*
*
* Processor type and features
*
Subarchitecture Type
Quoted 1 line
1. PC-compatible (X86_PC)
2. Support for ScaleMP vSMP (X86_VSMP) (NEW)
choice[1-2]: aborted!

Quoted 7 lines

Console input/output is redirected. Run 'make oldconfig' to update configuration.
make[3]: *** [silentoldconfig] Error 1 make[2]: *** [silentoldconfig] Error 2 make[1]: *** [include/config/auto.conf] Error 2
make: *** [all] Error 2

[]

Quoted 1 line

What did I do wrong??

That Makefile is used only, if you are *in* that obj. directory. Thus cd BLD is what you've forget.

There must be a better way to have this kind of choice be saved, but it's chicken-and-egg issue from the kconfig POV.

As i'm always try to use separate obj. dir., i've made life easier *in* it.
____

Sam Ravnborg

2007-06-12 18:09 UTC #9 lore raw

Re: [patch] kbuild: remember ARCH in the object directory

Quoted 36 lines · Oleg Verych

On Tue, Jun 12, 2007 at 09:12:09AM -0700, Randy Dunlap wrote:
[]
Quoted 23 lines
I did:
# build a linux-2.6.22-rc4-git4 kernel source tree & cd into it
# mkdir BLD
$ make ARCH=i386 O=BLD defconfig
$ make -j4 O=BLD all >BLD/bld.out 2>&1
and it cried to me:
GEN /scsi/linsrc/linux-2.6.22-rc4-git4/BLD/Makefile
scripts/kconfig/conf -s arch/x86_64/Kconfig #... many config warning lines
*
* Restart config...
*
*
* Processor type and features
*
Subarchitecture Type
Quoted 1 line
1. PC-compatible (X86_PC)
2. Support for ScaleMP vSMP (X86_VSMP) (NEW)
choice[1-2]: aborted!
Quoted 7 lines
Console input/output is redirected. Run 'make oldconfig' to update configuration.
make[3]: *** [silentoldconfig] Error 1 make[2]: *** [silentoldconfig] Error 2 make[1]: *** [include/config/auto.conf] Error 2
make: *** [all] Error 2
[]
Quoted 1 line
What did I do wrong??

Quoted 8 lines

That Makefile is used only, if you are *in* that obj. directory. Thus cd BLD is what you've forget.
There must be a better way to have this kind of choice be saved, but it's chicken-and-egg issue from the kconfig POV.
As i'm always try to use separate obj. dir., i've made life easier *in* it.

I fully agree with the intent of the patch.
But it fails exactly because it introduce different behaviour dependent of actual usage.

For non O= build ARCH is not saved and does not work For make O= it does not work
For make in the output dir it works
For external modules it does not work

This will greatly confuse people.

If we go the "save important parts of the config" I prefer something along the suggestion by hpa with a config file. The config file should though be named along the lines of Kbuild.config and the syntax should be future proof. I like the syntax of the .git/config file and
it should be along these lines.

And the bahavior should be exactly the same for all uses listed above.

Sam

H. Peter Anvin

2007-06-13 02:12 UTC #10 lore raw

Re: [kbuild-devel] [patch] kbuild: remember ARCH in the object directory

Quoted 8 lines · Sam Ravnborg

If we go the "save important parts of the config" I prefer something along the suggestion by hpa with a config file. The config file should though be named along the lines of Kbuild.config and the syntax should be future proof. I like the syntax of the .git/config file and
it should be along these lines.

It's actually derived from the Windows 3.x config file. I think it's massive overkill for this purpose; since the variables can be set on the command line we're not dealing with multiple namespaces, and it would be better to have a make-like syntax.

-hpa