[PATCH 0 of 2] partly fix cross-build of ioemu-remote

[PATCH 0 of 2] partly fix cross-build of ioemu-remote

[Aron Griffis]

These two patches partly repair cross-building of ioemu-dir.
There's still a problem where the translation layer is being
built, which doesn't exist for ia64, so it bombs out eventually.

The first patch applies to xen-unstable.hg, second patch applies
to the ioemu git tree.

Thanks,
Aron

[PATCH 1 of 2] pass IOEMU_CONFIGURE_CROSS to ioemu-remote/xen-setup

[Aron Griffis]

# HG changeset patch
# User Aron Griffis <aron@hp.com>
# Date 1217515818 14400
# Node ID be6b349abfb8d1d84618597c5304ab2e0500d5b4
# Parent  05391ca07588c45f5c8dd16c2a470b0d615767b8
pass IOEMU_CONFIGURE_CROSS to ioemu-remote/xen-setup

Add --cpu to IOEMU_CONFIGURE_CROSS since otherwise qemu uses the
output of `uname -m` even for cross-building.  Pass
IOEMU_CONFIGURE_CROSS to xen-setup which passes it on to
ioemu/configure.

Signed-off-by: Aron Griffis <aron@hp.com>

diff -r 05391ca07588 -r be6b349abfb8 tools/Makefile
--- a/tools/Makefile	Thu Jul 31 11:14:23 2008 +0100
+++ b/tools/Makefile	Thu Jul 31 10:50:18 2008 -0400
@@ -55,7 +55,8 @@
 clean distclean: subdirs-clean
 
 ifneq ($(XEN_COMPILE_ARCH),$(XEN_TARGET_ARCH))
-IOEMU_CONFIGURE_CROSS ?= --cross-prefix=$(CROSS_COMPILE) \
+IOEMU_CONFIGURE_CROSS ?= --cpu=$(XEN_TARGET_ARCH) \
+			 --cross-prefix=$(CROSS_COMPILE) \
 			 --interp-prefix=$(CROSS_SYS_ROOT)
 endif
 
@@ -96,7 +97,7 @@
 		esac; \
 		export XEN_ROOT; \
 		cd ioemu-dir; \
-		./xen-setup
+		./xen-setup $(IOEMU_CONFIGURE_CROSS)
 
 subdir-all-ioemu-dir subdir-install-ioemu-dir: ioemu-dir-find

[PATCH 2 of 2] cross-build fixes in ioemu-remote

[Aron Griffis]

Partly fix cross-building in ioemu-remote:

- Don't call install -s; it uses the host strip program which
  doesn't understand cross-built binaries.  This same change was
  in the legacy ioemu dir.

- The ioemu version of cc-option doesn't take $(CC) as the first
  argument, so all builds were getting -msse2 regardless of
  architecture.

Signed-off-by: Aron Griffis <aron@hp.com>

diff --git a/Makefile b/Makefile
index bebd244..b055b3f 100644
--- a/Makefile
+++ b/Makefile
@@ -205,7 +205,7 @@ endif
 install: all $(if $(BUILD_DOCS),install-doc)
 	mkdir -p "$(DESTDIR)$(bindir)"
 ifneq ($(TOOLS),)
-	$(INSTALL) -m 755 -s $(TOOLS) "$(DESTDIR)$(bindir)"
+	$(INSTALL) -m 755 $(TOOLS) "$(DESTDIR)$(bindir)"
 endif
 	mkdir -p "$(DESTDIR)$(datadir)"
 	set -e; for x in bios.bin vgabios.bin vgabios-cirrus.bin ppc_rom.bin \
diff --git a/Makefile.target b/Makefile.target
index 1aace9c..e796d61 100644
--- a/Makefile.target
+++ b/Makefile.target
@@ -707,7 +707,7 @@ clean:
 
 install: all install-hook
 ifneq ($(PROGS),)
-	$(INSTALL) -m 755 -s $(PROGS) "$(DESTDIR)$(bindir)"
+	$(INSTALL) -m 755 $(PROGS) "$(DESTDIR)$(bindir)"
 endif
 
 # Include automatically generated dependency files
diff --git a/xen-hooks.mak b/xen-hooks.mak
index af977ef..c90f546 100644
--- a/xen-hooks.mak
+++ b/xen-hooks.mak
@@ -3,7 +3,7 @@ CPPFLAGS+= -I$(XEN_ROOT)/tools/xenstore
 CPPFLAGS+= -I$(XEN_ROOT)/tools/include
 CPPFLAGS+= -I$(XEN_ROOT)/tools/blktap/lib
 
-SSE2 := $(call cc-option,$(CC),-msse2,)
+SSE2 := $(call cc-option,-msse2,)
 ifeq ($(SSE2),-msse2)
 CFLAGS += -DUSE_SSE2=1 -msse2
 endif

Re: [Xen-devel] [PATCH 2 of 2] cross-build fixes in ioemu-remote

[Ian Jackson]

Aron Griffis writes ("[Xen-devel] [PATCH 2 of 2] cross-build fixes in ioemu-remote"):
> Partly fix cross-building in ioemu-remote:
> 
> - The ioemu version of cc-option doesn't take $(CC) as the first
>   argument, so all builds were getting -msse2 regardless of
>   architecture.

I'll commit this in just a moment.

> - Don't call install -s; it uses the host strip program which
>   doesn't understand cross-built binaries.  This same change was
>   in the legacy ioemu dir.

I think this should be fixed by the build environment, or possibly
$(INSTALL), being arranged to not have this bug.  If $(INSTALL) -s
doesn't work then this should be fixed.  Options include putting a
symlink to /bin/true on the path somewhere under the name `strip', or
overriding INSTALL somehow to a more clever program.

I really don't want to apply these Makefile changes.  That is the kind
of thing which makes merging painful.

Ian.

Re: [Xen-ia64-devel] Re: [PATCH 2 of 2] cross-build fixes in ioemu-remote

[Samuel Thibault]

Ian Jackson, le Thu 31 Jul 2008 16:16:37 +0100, a écrit :
> Aron Griffis writes ("[Xen-devel] [PATCH 2 of 2] cross-build fixes in ioemu-remote"):
> > - Don't call install -s; it uses the host strip program which
> >   doesn't understand cross-built binaries.  This same change was
> >   in the legacy ioemu dir.
> 
> I think this should be fixed by the build environment, or possibly
> $(INSTALL), being arranged to not have this bug.  If $(INSTALL) -s
> doesn't work then this should be fixed.  Options include putting a
> symlink to /bin/true on the path somewhere under the name `strip', or
> overriding INSTALL somehow to a more clever program.

I have the same issue when cross-compiling some other projects.  I
wonder how install would be able to know which strip command it should
use.

Samuel

Re: Re: [Xen-devel] [PATCH 2 of 2] cross-build fixes in ioemu-remote

[Samuel Thibault]

Ian Jackson, le Thu 31 Jul 2008 16:16:37 +0100, a écrit :
> Aron Griffis writes ("[Xen-devel] [PATCH 2 of 2] cross-build fixes in ioemu-remote"):
> > Partly fix cross-building in ioemu-remote:
> > 
> > - The ioemu version of cc-option doesn't take $(CC) as the first
> >   argument, so all builds were getting -msse2 regardless of
> >   architecture.
> 
> I'll commit this in just a moment.

Mmm, actually I think we don't need that sse2 option any more: thanks
to dirty bit tracking, the sse2-optimized memcmp loop is gone. At least
-DUSE_SSE2 is not used anywhere.

Samuel

Re: [Xen-ia64-devel] Re: [PATCH 2 of 2] cross-build fixes in ioemu-remote

[Ian Jackson]

Samuel Thibault writes ("Re: [Xen-ia64-devel] Re: [Xen-devel] [PATCH 2 of 2] cross-build fixes in ioemu-remote"):
> I have the same issue when cross-compiling some other projects.  I
> wonder how install would be able to know which strip command it should
> use.

Perhaps it should call STRIP if it exists, or perhaps the
cross-building setup should provide a `strip' which will always DTRT
(perhaps by examining the file to decide which real strip to invoke).

Ian.

Re: [Xen-ia64-devel] Re: [PATCH 2 of 2] cross-build fixes in ioemu-remote

[Aron Griffis]

Ian Jackson wrote:  [Thu Jul 31 2008, 12:48:15PM EDT]
> Samuel Thibault writes ("Re: [Xen-ia64-devel] Re: [Xen-devel] [PATCH 2 of 2] cross-build fixes in ioemu-remote"):
> > I have the same issue when cross-compiling some other projects.  I
> > wonder how install would be able to know which strip command it should
> > use.
> 
> Perhaps it should call STRIP if it exists, or perhaps the
> cross-building setup should provide a `strip' which will always DTRT
> (perhaps by examining the file to decide which real strip to invoke).

I don't think it's necessary to provide one that will DTRT.  The
compiler and linker don't need that abstraction, and "install" is
only called for stuff that will be installed on the target
system.

qemu seems to have the ability to set the install command using
configure.  Seems something like the following should work,
except that (1) i386-dm/hookstarget.mak uses $(INSTALL_PROG)
which seems to be defined outside of the ioemu tree, and (2) I'm
having enough trouble cross-building qemu that it's hard to test.

Thanks,
Aron

diff -r 40daf3257cad tools/Makefile
--- a/tools/Makefile	Thu Jul 31 10:51:30 2008 -0400
+++ b/tools/Makefile	Thu Jul 31 15:18:24 2008 -0400
@@ -36,9 +36,11 @@ SUBDIRS-$(PYTHON_TOOLS) += pygrub
 SUBDIRS-$(PYTHON_TOOLS) += pygrub
 endif
 
-# For the sake of linking, set the sys-root
+# For the sake of linking, set the path to "strip" and the sys-root
 ifneq ($(CROSS_COMPILE),)
+CROSS_BIN_PATH ?= /usr/$(CROSS_COMPILE:-=)/bin
 CROSS_SYS_ROOT ?= /usr/$(CROSS_COMPILE:-=)/sys-root
+# export sys-root to environment for check/funcs.sh
 export CROSS_SYS_ROOT
 endif
 
@@ -57,7 +59,8 @@ ifneq ($(XEN_COMPILE_ARCH),$(XEN_TARGET_
 ifneq ($(XEN_COMPILE_ARCH),$(XEN_TARGET_ARCH))
 IOEMU_CONFIGURE_CROSS ?= --cpu=$(XEN_TARGET_ARCH) \
 			 --cross-prefix=$(CROSS_COMPILE) \
-			 --interp-prefix=$(CROSS_SYS_ROOT)
+			 --interp-prefix=$(CROSS_SYS_ROOT) \
+			 --install='env PATH="$(CROSS_BIN_PATH):$$$$PATH" install'
 endif
 
 ioemu/config-host.mak:

Re: [Xen-ia64-devel] Re: [PATCH 2 of 2] cross-build fixes in ioemu-remote

[Ian Jackson]

Aron Griffis writes ("Re: [Xen-ia64-devel] Re: [Xen-devel] [PATCH 2 of 2] cross-build fixes in ioemu-remote"):
> -			 --interp-prefix=$(CROSS_SYS_ROOT)
> +			 --interp-prefix=$(CROSS_SYS_ROOT) \
> +			 --install='env PATH="$(CROSS_BIN_PATH):$$$$PATH" install'

This is a bit bizarre (not to mention the fact that the number of $'s
will depend on undocumented properties of the way qemu's configure
inteprets that option).

Why not just set PATH directly and export it ?

But in general I think this is a reasonable approach.

Ian.

Re: Re: [Xen-devel] [PATCH 2 of 2] cross-build fixes in ioemu-remote

[Aron Griffis]

Ian Jackson wrote:  [Fri Aug 01 2008, 12:26:54PM EDT]
> Aron Griffis writes ("Re: [Xen-ia64-devel] Re: [Xen-devel] [PATCH 2 of 2] cross-build fixes in ioemu-remote"):
> > -			 --interp-prefix=$(CROSS_SYS_ROOT)
> > +			 --interp-prefix=$(CROSS_SYS_ROOT) \
> > +			 --install='env PATH="$(CROSS_BIN_PATH):$$$$PATH" install'
> 
> This is a bit bizarre (not to mention the fact that the number
> of $'s will depend on undocumented properties of the way qemu's
> configure inteprets that option).

qemu's configure uses quoting properly and applies no
interpretation on the way through.  The dollars are doubled once
for the Makefile where this is set, once for the inner Makefile
where it is used.

However... there is a Solaris hack in configure that will break
with any --install option other than a bare word.  And I agree
with you anyway, it's a fragile approach.

> Why not just set PATH directly and export it ?

qemu expects "gcc" and the like to refer to the host toolchain.
I think we only want the PATH set for install.  Ideally there
would be an /usr/bin/ia64-linux-gnu-install to accompany the
others and use the right strip, but oh well.

> But in general I think this is a reasonable approach.

How about...

# HG changeset patch
# User Aron Griffis <aron@hp.com>
# Date 1217610812 14400
# Node ID 714215c06a98ea845e3f25371945d62abb1c18c7
# Parent  f69a23b209c4ab04827051734e37f095607400ed
add cross-install to find the right "strip"

Signed-off-by: Aron Griffis <aron@hp.com>

diff --git a/tools/Makefile b/tools/Makefile
--- a/tools/Makefile
+++ b/tools/Makefile
@@ -38,8 +38,10 @@
 
 # For the sake of linking, set the sys-root
 ifneq ($(CROSS_COMPILE),)
+CROSS_BIN_PATH ?= /usr/$(CROSS_COMPILE:-=)/bin
 CROSS_SYS_ROOT ?= /usr/$(CROSS_COMPILE:-=)/sys-root
-export CROSS_SYS_ROOT
+export CROSS_SYS_ROOT # exported for check/funcs.sh
+export CROSS_BIN_PATH # exported for cross-install.sh
 endif
 
 .PHONY: all
@@ -57,7 +59,8 @@
 ifneq ($(XEN_COMPILE_ARCH),$(XEN_TARGET_ARCH))
 IOEMU_CONFIGURE_CROSS ?= --cpu=$(XEN_TARGET_ARCH) \
 			 --cross-prefix=$(CROSS_COMPILE) \
-			 --interp-prefix=$(CROSS_SYS_ROOT)
+			 --interp-prefix=$(CROSS_SYS_ROOT) \
+			 --install=$(CURDIR)/cross-install
 endif
 
 ioemu/config-host.mak:
diff --git a/tools/cross-install b/tools/cross-install
new file mode 100755
--- /dev/null
+++ b/tools/cross-install
@@ -0,0 +1,8 @@
+#!/bin/sh
+
+# prepend CROSS_BIN_PATH to find the right "strip"
+if [ -n "$CROSS_BIN_PATH" ]; then
+    PATH="$CROSS_BIN_PATH:$PATH"
+fi
+
+exec install "$@"

Re: [Xen-ia64-devel] Re: [PATCH 2 of 2] cross-build fixes in ioemu-remote

[Aron Griffis]

Any responses to my last email and patch in this thread?

It would be nice to get cross-install (or something like it) in
the tree, along with fixing the build on ia64.

The build problem is that translate-all.c doesn't build on ia64,
because there's no support for instruction translation on ia64.
But it shouldn't need to build translate-all.c at all for ioemu.
It didn't on the legacy ioemu, and it's only working right now on
ia32 and x86_64 because those arches can build translate-all.c,
however unnecessary it is.

I took a quick look but I'm not quickly seeing how to change the
build to omit translate-all.c.  It would be great if somebody
more familiar with ioemu could take a glance.

Thanks,
Aron

Re: [Xen-ia64-devel] Re: [PATCH 2 of 2] cross-build fixes in ioemu-remote

[Keir Fraser]

Unfortunately Ian is away this week, so these changes aren't likely to get
picked up or commented on until Monday.

 -- Keir

On 4/8/08 23:31, "Aron Griffis" <aron@hp.com> wrote:

> Any responses to my last email and patch in this thread?
> 
> It would be nice to get cross-install (or something like it) in
> the tree, along with fixing the build on ia64.
> 
> The build problem is that translate-all.c doesn't build on ia64,
> because there's no support for instruction translation on ia64.
> But it shouldn't need to build translate-all.c at all for ioemu.
> It didn't on the legacy ioemu, and it's only working right now on
> ia32 and x86_64 because those arches can build translate-all.c,
> however unnecessary it is.
> 
> I took a quick look but I'm not quickly seeing how to change the
> build to omit translate-all.c.  It would be great if somebody
> more familiar with ioemu could take a glance.
> 
> Thanks,
> Aron
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel

Re: [Xen-ia64-devel] Re: [PATCH 2 of 2] cross-build fixes in ioemu-remote

[Aron Griffis]

(cc'd to lists)

Isaku Yamahata wrote:  [Tue Aug 05 2008, 10:02:49PM EDT]
> On Tue, Aug 05, 2008 at 10:50:32AM -0400, Aron Griffis wrote:
> > Keir Fraser wrote:  [Tue Aug 05 2008, 08:47:40AM EDT]
> > > I suggest we turn off ioemu-remote for ia64, at least by
> > > default. Even if we get it building next week, it still needs
> > > testing. And we plan to release next Friday (15th August).
> > 
> > I'd defer to Isaku (cc'd) on that question.  I'm not quite
> > connected enough to Xen/ia64 development to know the best answer.
> 
> Given that no one hasn't seemed to test ioemu-remote,
> I think default off is a reasonable choice.

Okay, here's the patch.

Thanks,
Aron

# HG changeset patch
# User Aron Griffis <aron@hp.com>
# Date 1218028500 14400
# Node ID 7071b1990a15ee14e2ccb9d199764237962c7e57
# Parent  c329dcc9df04ade9c183bfb696bca37296d80a1f
Use legacy ioemu on ia64 for Xen 4.0

Signed-off-by: Aron Griffis <aron@hp.com>

diff -r c329dcc9df04 -r 7071b1990a15 Config.mk
--- a/Config.mk	Wed Aug 06 11:38:42 2008 +0100
+++ b/Config.mk	Wed Aug 06 09:15:00 2008 -0400
@@ -86,7 +86,11 @@
 # Mercurial in-tree version, or a local directory, or a git URL.
 # CONFIG_QEMU   ?= ioemu
 # CONFIG_QEMU   ?= ../qemu-xen.git
+ifeq ($(XEN_TARGET_ARCH),ia64)
+CONFIG_QEMU   ?= ioemu
+else
 CONFIG_QEMU   ?= $(QEMU_REMOTE)
+endif
 
 # Optional components
 XENSTAT_XENTOP     ?= y

Re: Re: [Xen-devel] [PATCH 2 of 2] cross-build fixes in ioemu-remote

[Ian Jackson]

Aron Griffis writes ("Re: [Xen-ia64-devel] Re: [Xen-devel] [PATCH 2 of 2] cross-build fixes in ioemu-remote"):
> # HG changeset patch
> # User Aron Griffis <aron@hp.com>
> # Date 1217610812 14400
> # Node ID 714215c06a98ea845e3f25371945d62abb1c18c7
> # Parent  f69a23b209c4ab04827051734e37f095607400ed
> add cross-install to find the right "strip"

I think this is a good patch.

Ian.