Skip to contentSkip to conversation

remove target_os

PATCH In review Robert Millan 4 participants 14 messages Jan 25, 2009 – Feb 8, 2009 Kbuild

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 remove target_os

Hi,

Based on the description of host/target triplets in configure.ac:

dnl   build  -- the environment for building GRUB
dnl   host   -- the environment for running utilities
dnl   target -- the environment for running GRUB

it seems that target_os is an oxymoron. There's no OS in the environment where GRUB will run (well, there's the firmware, but we already use target_vendor for that, and _os has a well-defined meaning).

Attached patch fixes that by supressing all references to target_os, and replacing them with host_os where suitable.

Christian: since this mostly affects Cygwin, could you verify that it doesn't cause breakage before I commit it?

Signature · 5 lines
Robert Millan

  The DRM opt-in fallacy: "Your data belongs to us. We will decide when (and
  how) you may access your data; but nobody's threatening your freedom: we
  still allow you to remove your data and not access it at all."

2009-01-26 Robert Millan <[email protected]>

* Makefile.in (host_os, host_cpu): New variables. (target_os): Remove. Update all users.

Makefile.in +3 -1
Index: Makefile.in
===================================================================
@@ -47,8 +47,10 @@ PACKAGE_VERSION = @PACKAGE_VERSION@
 PACKAGE_STRING = @PACKAGE_STRING@
 PACKAGE_BUGREPORT = @PACKAGE_BUGREPORT@
 
+host_os = @host_os@
+host_cpu = @host_cpu@
+
 target_cpu = @target_cpu@
-target_os = @target_os@
 platform = @platform@
 
 INSTALL = @INSTALL@
common.rmk +1 -1
Index: conf/common.rmk
===================================================================
@@ -146,7 +146,7 @@ CLEANFILES += update-grub_lib
 	./config.status --file=$@:$<
 	chmod +x $@
 grub-mkconfig_SCRIPTS = 00_header 10_linux 10_hurd 10_freebsd 30_os-prober 40_custom
-ifeq ($(target_os), cygwin)
+ifeq ($(host_os), cygwin)
 grub-mkconfig_SCRIPTS += 10_windows
 endif
 
configure.ac +7 -9
Index: configure.ac
===================================================================
@@ -103,12 +103,10 @@ case "$host_os" in
   mingw32) host_os=cygwin ;;
 esac
 
-case "$target_os" in
-  mingw32) target_os=cygwin ;;
-esac
+AC_SUBST(host_cpu)
+AC_SUBST(host_os)
 
 AC_SUBST(target_cpu)
-AC_SUBST(target_os)
 AC_SUBST(platform)
 
 #
@@ -200,10 +198,10 @@ AC_CHECK_FUNCS(posix_memalign memalign a
 
 # Use linker script if present, otherwise use builtin -N script.
 AC_MSG_CHECKING([for option to link raw image])
-if test -f "${srcdir}/conf/${target_cpu}-${platform}-${target_os}-img-ld.sc"; then
-  TARGET_IMG_LDSCRIPT='$(top_srcdir)'"/conf/${target_cpu}-${platform}-${target_os}-img-ld.sc"
+if test -f "${srcdir}/conf/${target_cpu}-${platform}-${host_os}-img-ld.sc"; then
+  TARGET_IMG_LDSCRIPT='$(top_srcdir)'"/conf/${target_cpu}-${platform}-${host_os}-img-ld.sc"
   TARGET_IMG_LDFLAGS="-Wl,-T${TARGET_IMG_LDSCRIPT}"
-  TARGET_IMG_LDFLAGS_AC="-Wl,-T${srcdir}/conf/${target_cpu}-${platform}-${target_os}-img-ld.sc"
+  TARGET_IMG_LDFLAGS_AC="-Wl,-T${srcdir}/conf/${target_cpu}-${platform}-${host_os}-img-ld.sc"
 else
   TARGET_IMG_LDSCRIPT=
   TARGET_IMG_LDFLAGS='-Wl,-N'
@@ -215,8 +213,8 @@ AC_MSG_RESULT([$TARGET_IMG_LDFLAGS_AC])
 
 # For platforms where ELF is not the default link format.
 AC_MSG_CHECKING([for command to convert module to ELF format])
-case "${host_os}:${target_os}" in
-  cygwin:cygwin) TARGET_OBJ2ELF='grub-pe2elf' ;;
+case "${host_os}" in
+  cygwin) TARGET_OBJ2ELF='grub-pe2elf' ;;
   *) ;;
 esac
 AC_SUBST(TARGET_OBJ2ELF)
#1 lore raw
Quoted 12 lines · Robert Millan

Hi,

Based on the description of host/target triplets in configure.ac:

dnl   build  -- the environment for building GRUB
dnl   host   -- the environment for running utilities
dnl   target -- the environment for running GRUB

it seems that target_os is an oxymoron. There's no OS in the environment where GRUB will run (well, there's the firmware, but we already use target_vendor for that, and _os has a well-defined meaning).

Quoted 7 lines

Attached patch fixes that by supressing all references to target_os, and replacing them with host_os where suitable.

Christian: since this mostly affects Cygwin, could you verify that it doesn't cause breakage before I commit it?

Hi Robert,

thanks for sending the patch first. Cygwin build looks good.

But even if GRUB itself is build for some $target_cpu-$target_vendor (i386-pc), the target_os is not useless if cross-compilation support is desired:

- GRUB tools like grub-setup are build for the target_os
- Package structure may depend on target_os.

- The gcc code generator may emit special code tailored for the assumed target_os Examples:
Stack frame checks
Nested function trampolines (__enable_execute_stack())

Unfortunately, gcc has no '-fno_os' option to specify the bare CPU as target. Fortunately, all current issues can be handled by autoconf checks.

configure.ac AC_MSG_RESULT([$TARGET_IMG_LDFLAGS_AC]) · line 219

Quoted 8 lines

AC_MSG_CHECKING([for command to convert module to ELF format])

configure.ac +2 -2
-case "${host_os}:${target_os}" in
-  cygwin:cygwin) TARGET_OBJ2ELF='grub-pe2elf' ;;
+case "${host_os}" in
+  cygwin) TARGET_OBJ2ELF='grub-pe2elf' ;;
   *) ;;
 esac

This won't work for a Linux cross compiler hosted on Cygwin. It would emit ELF format and does not need pe2elf.

But all this is theoretical unless cross compilation is really needed.

Christian

#2 lore raw

Hi GRUB devs,

Quoted 33 lines · Christian Franke
Quoted 12 lines · Robert Millan

Hi,

Based on the description of host/target triplets in configure.ac:

dnl   build  -- the environment for building GRUB
dnl   host   -- the environment for running utilities
dnl   target -- the environment for running GRUB

it seems that target_os is an oxymoron. There's no OS in the environment where GRUB will run (well, there's the firmware, but we already use target_vendor for that, and _os has a well-defined meaning).

Quoted 7 lines

Attached patch fixes that by supressing all references to target_os, and replacing them with host_os where suitable.

Christian: since this mostly affects Cygwin, could you verify that it doesn't cause breakage before I commit it?

Hi Robert,

thanks for sending the patch first. Cygwin build looks good.

But even if GRUB itself is build for some $target_cpu-$target_vendor (i386-pc), the target_os is not useless if cross-compilation support is desired:

- GRUB tools like grub-setup are build for the target_os
- Package structure may depend on target_os.

Quoted 8 lines

- The gcc code generator may emit special code tailored for the assumed target_os
Examples:
Stack frame checks
Nested function trampolines (__enable_execute_stack())

Unfortunately, gcc has no '-fno_os' option to specify the bare CPU as target.

Might -ffreestanding be what you are looking for?

Quoted 14 lines

Fortunately, all current issues can be handled by autoconf checks.

configure.ac AC_MSG_RESULT([$TARGET_IMG_LDFLAGS_AC]) · line 219

Quoted 8 lines

AC_MSG_CHECKING([for command to convert module to ELF format])

configure.ac +2 -2
-case "${host_os}:${target_os}" in
-  cygwin:cygwin) TARGET_OBJ2ELF='grub-pe2elf' ;;
+case "${host_os}" in
+  cygwin) TARGET_OBJ2ELF='grub-pe2elf' ;;
   *) ;;
 esac

This won't work for a Linux cross compiler hosted on Cygwin. It would emit ELF format and does not need pe2elf.

A, say, AMD64 Linux cross compiler hosted on x86 Cygwin would have $build=i686-pc-cygwin and $host=amd64-linux-gnu. Thus, no conflict ought to arise even with cross compilation enabled.

AFAIK, the full power of $build+$host+$target only matters when building _compilers_ (and binutils, etc.), because you might want to use an AMD64/Linux machine to build a compiler that will run on PPC/Darwin but produce executables for a x86/Cygwin machine

Quoted 11 lines

But all this is theoretical unless cross compilation is really needed.

Christian

_______________________________________________
Grub-devel mailing list
[email protected]
http://lists.gnu.org/mailman/listinfo/grub-devel

#3 lore raw
Quoted 8 lines · Javier Martín

A, say, AMD64 Linux cross compiler hosted on x86 Cygwin would have $build=i686-pc-cygwin and $host=amd64-linux-gnu. Thus, no conflict ought to arise even with cross compilation enabled.

AFAIK, the full power of $build+$host+$target only matters when building _compilers_ (and binutils, etc.), because you might want to use an AMD64/Linux machine to build a compiler that will run on PPC/Darwin but produce executables for a x86/Cygwin machine

Now lets move this idea to GRUB 2:

There is a build server is running on PPC/Linux and it wants to build all architectures supported by GRUB 2 (provide proper boot code for all arch and tools) and then makes software bundles (rpm,deb, or so) for all architectures.

In example if all shell tools are to be ran on x86-64/Linux and then boot code needs to be compatible with x86-32 as there is PC-BIOS present on target system.

#6 lore raw
In reply to Javier Martín
Quoted 6 lines · Javier Martín
Quoted 2 lines

Unfortunately, gcc has no '-fno_os' option to specify the bare CPU as target.

Might -ffreestanding be what you are looking for?

The option '-ffreestanding' is the same as '-fno-hosted'.

According to gcc (4.3.1) source, '-fno-hosted' clears variable 'flag_hosted' and sets '-fno-builtin'. The latter is already set within GRUB build.

A cleared 'flag_hosted' apparently has only 2 effects:
- disable the special handling of 'main()'.
- #define __STDC_HOSTED__ to 0 instead of 1

There is no effect on the target_os dependent parts of the gcc code generation. For example, on i386, __enable_execute_stack() calls are generated for target_os netbsd, openbsd and cygwin, but not for linux. The emit call is hard-coded in gcc/configs/i386/i386.c:x86_initialize_trampoline().

Other workarounds are needed to support building GRUB with code generators tailored for various target_os.

Quoted 13 lines
Quoted 12 lines

configure.ac AC_MSG_RESULT([$TARGET_IMG_LDFLAGS_AC]) · line 219

Quoted 10 lines

AC_MSG_CHECKING([for command to convert module to ELF format])

configure.ac +2 -2
-case "${host_os}:${target_os}" in
-  cygwin:cygwin) TARGET_OBJ2ELF='grub-pe2elf' ;;
+case "${host_os}" in
+  cygwin) TARGET_OBJ2ELF='grub-pe2elf' ;;
   *) ;;
 esac

This won't work for a Linux cross compiler hosted on Cygwin. It would emit ELF format and does not need pe2elf.

Quoted 5 lines

A, say, AMD64 Linux cross compiler hosted on x86 Cygwin would have $build=i686-pc-cygwin and $host=amd64-linux-gnu. Thus, no conflict ought to arise even with cross compilation enabled.

But the opposite won't work: $host=i686-pc-cygwin would enable grub-pe2elf, even if this gcc emits ELF for a linux target.

Christian

#4 lore raw
In reply to Vesa Jääskeläinen
Quoted 12 lines · Vesa Jääskeläinen
Quoted 8 lines · Javier Martín

A, say, AMD64 Linux cross compiler hosted on x86 Cygwin would have $build=i686-pc-cygwin and $host=amd64-linux-gnu. Thus, no conflict ought to arise even with cross compilation enabled.

AFAIK, the full power of $build+$host+$target only matters when building _compilers_ (and binutils, etc.), because you might want to use an AMD64/Linux machine to build a compiler that will run on PPC/Darwin but produce executables for a x86/Cygwin machine

Now lets move this idea to GRUB 2:

Quoted 8 lines

There is a build server is running on PPC/Linux and it wants to build all architectures supported by GRUB 2 (provide proper boot code for all arch and tools) and then makes software bundles (rpm,deb, or so) for all architectures.

In example if all shell tools are to be ran on x86-64/Linux and then boot code needs to be compatible with x86-32 as there is PC-BIOS present on target system.

So you propose --target to be reused like this (a PS3 building GRUB2 for an amd64 BIOS PC with Linux) :
./configure --build=ppc64-linux --host=x86_64-linux-gnu --target=i686-pc
^--compiling system ^--tools system ^--boot arch

I think we already have a better-geared switch for that (--with-platform it is?), which is currently used to distinguish between BIOS and EFI boot code on both x86 and x86-64.

Quoted 6 lines

_______________________________________________
Grub-devel mailing list
[email protected]
http://lists.gnu.org/mailman/listinfo/grub-devel

#7 lore raw
In reply to Christian Franke
Quoted 19 lines · Christian Franke
Quoted 7 lines · Javier Martín
Quoted 3 lines

Unfortunately, gcc has no '-fno_os' option to specify the bare CPU as target.

Might -ffreestanding be what you are looking for?

The option '-ffreestanding' is the same as '-fno-hosted'.

According to gcc (4.3.1) source, '-fno-hosted' clears variable 'flag_hosted' and sets '-fno-builtin'. The latter is already set within GRUB build.

A cleared 'flag_hosted' apparently has only 2 effects:
- disable the special handling of 'main()'.
- #define __STDC_HOSTED__ to 0 instead of 1

Quoted 5 lines

There is no effect on the target_os dependent parts of the gcc code generation. For example, on i386, __enable_execute_stack() calls are generated for target_os netbsd, openbsd and cygwin, but not for linux. The emit call is hard-coded in gcc/configs/i386/i386.c:x86_initialize_trampoline().

In that case, we are dealing with a GCC bug. We might want to require the user to create a bare "no-OS" cross compiler. These thingies are most likely only known to OS developers, but you can build a perfectly good "i686-pc-elf" GCC (only with no libc, of course). This might be the Right Way (TM) of building boot code, instead of the current way of using the OS-targetter compiler and trying to tell it not to do what it was built to do.

Quoted 19 lines

Other workarounds are needed to support building GRUB with code generators tailored for various target_os.

Quoted 14 lines
Quoted 13 lines

configure.ac AC_MSG_RESULT([$TARGET_IMG_LDFLAGS_AC]) · line 219

Quoted 10 lines

AC_MSG_CHECKING([for command to convert module to ELF format])

configure.ac +2 -2
-case "${host_os}:${target_os}" in
-  cygwin:cygwin) TARGET_OBJ2ELF='grub-pe2elf' ;;
+case "${host_os}" in
+  cygwin) TARGET_OBJ2ELF='grub-pe2elf' ;;
   *) ;;
 esac

This won't work for a Linux cross compiler hosted on Cygwin. It would emit ELF format and does not need pe2elf.

Quoted 8 lines
Quoted 5 lines

A, say, AMD64 Linux cross compiler hosted on x86 Cygwin would have $build=i686-pc-cygwin and $host=amd64-linux-gnu. Thus, no conflict ought to arise even with cross compilation enabled.

But the opposite won't work: $host=i686-pc-cygwin would enable grub-pe2elf, even if this gcc emits ELF for a linux target.

"The opposite" of a the described situation would be a Linux AMD64 machine cross compiling for a Cygwin x86 machine, that is $build=amd64-linux-gnu and $host=i686-pc-cygwin. In that case, i686-pc-cygwin-gcc would generate PE executables, and thus grub-pe2elf _is_ required indeed.

Quoted 9 lines

Christian

_______________________________________________
Grub-devel mailing list
[email protected]
http://lists.gnu.org/mailman/listinfo/grub-devel

#8 lore raw
Quoted 19 lines · Javier Martín
Quoted 17 lines · Christian Franke
Quoted 7 lines · Javier Martín
Quoted 3 lines

Unfortunately, gcc has no '-fno_os' option to specify the bare CPU as target.

Might -ffreestanding be what you are looking for?

The option '-ffreestanding' is the same as '-fno-hosted'.

According to gcc (4.3.1) source, '-fno-hosted' clears variable 'flag_hosted' and sets '-fno-builtin'. The latter is already set within GRUB build.

A cleared 'flag_hosted' apparently has only 2 effects:
- disable the special handling of 'main()'.
- #define __STDC_HOSTED__ to 0 instead of 1

Quoted 7 lines
Quoted 2 lines

There is no effect on the target_os dependent parts of the gcc code generation. For example, on i386, __enable_execute_stack() calls are generated for target_os netbsd, openbsd and cygwin, but not for linux. The emit call is hard-coded in gcc/configs/i386/i386.c:x86_initialize_trampoline().

In that case, we are dealing with a GCC bug. We might want to require the user to create a bare "no-OS" cross compiler.

Is building a bare 'no-OS' compiler supported by the upstream GCC sources? Probably a too strict prerequisite for building GRUB.

Quoted 19 lines
Quoted 19 lines
Quoted 19 lines
Quoted 13 lines

configure.ac AC_MSG_RESULT([$TARGET_IMG_LDFLAGS_AC]) · line 219

Quoted 10 lines

AC_MSG_CHECKING([for command to convert module to ELF format])

configure.ac +2 -2
-case "${host_os}:${target_os}" in
-  cygwin:cygwin) TARGET_OBJ2ELF='grub-pe2elf' ;;
+case "${host_os}" in
+  cygwin) TARGET_OBJ2ELF='grub-pe2elf' ;;
   *) ;;
 esac

This won't work for a Linux cross compiler hosted on Cygwin. It would emit ELF format and does not need pe2elf.

A, say, AMD64 Linux cross compiler hosted on x86 Cygwin would have $build=i686-pc-cygwin and $host=amd64-linux-gnu. Thus, no conflict ought to arise even with cross compilation enabled.

Quoted 9 lines
Quoted 2 lines

But the opposite won't work: $host=i686-pc-cygwin would enable grub-pe2elf, even if this gcc emits ELF for a linux target.

"The opposite" of a the described situation would be a Linux AMD64 machine cross compiling for a Cygwin x86 machine, that is $build=amd64-linux-gnu and $host=i686-pc-cygwin. In that case, i686-pc-cygwin-gcc would generate PE executables, and thus grub-pe2elf _is_ required indeed.

Yes - I was wrong, sorry.

Robert: Patch is IMO GTG.

Christian

#9 lore raw
Quoted 30 lines · Christian Franke
Quoted 29 lines · Javier Martín
Quoted 26 lines · Christian Franke
Quoted 8 lines · Javier Martín
Quoted 4 lines

Unfortunately, gcc has no '-fno_os' option to specify the bare CPU as target.

Might -ffreestanding be what you are looking for?

The option '-ffreestanding' is the same as '-fno-hosted'.

According to gcc (4.3.1) source, '-fno-hosted' clears variable 'flag_hosted' and sets '-fno-builtin'. The latter is already set within GRUB build.

A cleared 'flag_hosted' apparently has only 2 effects:
- disable the special handling of 'main()'.
- #define __STDC_HOSTED__ to 0 instead of 1

There is no effect on the target_os dependent parts of the gcc code generation. For example, on i386, __enable_execute_stack() calls are generated for target_os netbsd, openbsd and cygwin, but not for linux. The emit call is hard-coded in gcc/configs/i386/i386.c:x86_initialize_trampoline().

Quoted 7 lines
Quoted 4 lines

In that case, we are dealing with a GCC bug. We might want to require the user to create a bare "no-OS" cross compiler.

Is building a bare 'no-OS' compiler supported by the upstream GCC sources? Probably a too strict prerequisite for building GRUB.

I can't vouch so for all possible environments, but I can say that "clean" no-OS targets like i386-pc-elf exist at least for x86 and x86_64 (this last being added as of GCC 4.3.2). However, as I said, if -ffreestanding does not do its job of providing a "true" freestanding environment (i.e. without relying on _any_ external libraries or syscalls, as opposite to the usual hosted environment), a bug report should be posted.

#10 lore raw
Quoted 6 lines · Javier Martín
Quoted 3 lines · Christian Franke

Is building a bare 'no-OS' compiler supported by the upstream GCC sources? Probably a too strict prerequisite for building GRUB.

Quoted 9 lines

I can't vouch so for all possible environments, but I can say that "clean" no-OS targets like i386-pc-elf exist at least for x86 and x86_64 (this last being added as of GCC 4.3.2). However, as I said, if -ffreestanding does not do its job of providing a "true" freestanding environment (i.e. without relying on _any_ external libraries or syscalls, as opposite to the usual hosted environment), a bug report should be posted.

I haven't examined more recent gcc sources yet, the extra effect of -ffreestanding on 4.3.1 is a no-op if the program does not use main().

How do those no-OS targets handle the platform independent libgcc functions, e.g. __divdi3 which is used for 64bit integer division at least on i386 ? Is a no-OS version of libgcc provided or should the programmer handle these cases ?

Christian

#13 lore raw
In reply to Christian Franke
Quoted 6 lines · Christian Franke

But even if GRUB itself is build for some $target_cpu-$target_vendor (i386-pc), the target_os is not useless if cross-compilation support is desired:

- GRUB tools like grub-setup are build for the target_os
- Package structure may depend on target_os.

That would be host_os. Consider this example: a system running on cpu FOO cross-compiles grub (i.e. stuff like grub-probe or grub-mkimage) so that it will run on cpu BAR. However, grub-mkimage should generate images for cpu BAZ (accordingly, the modules are built for BAZ too).

This means build_cpu=FOO host_cpu=BAR target_cpu=BAZ.

host_os, then, is the OS that runs on BAR. It isn't necessarily related to BAZ.

Signature · 5 lines
Robert Millan

  The DRM opt-in fallacy: "Your data belongs to us. We will decide when (and
  how) you may access your data; but nobody's threatening your freedom: we
  still allow you to remove your data and not access it at all."
#5 lore raw

remove OS part of --target=xxx (Re: [PATCH] remove target_os)

In reply to Javier Martín
Quoted 5 lines · Javier Martín

So you propose --target to be reused like this (a PS3 building GRUB2 for an amd64 BIOS PC with Linux) :
./configure --build=ppc64-linux --host=x86_64-linux-gnu --target=i686-pc
^--compiling system ^--tools system ^--boot arch

I'm not proposing to change the way --target is used, only describing what its argument is supposed to mean, based on a comment from configure.ac.

I agree that having to specify an OS for --target in command-line is odd too, but it's a different problem. Maybe we could ask the config.sub maintainer for a dummy value, such as "unknown" or so (I think Glibc had a --host for standalone -no OS- arm builds, perhaps we could use the same).

Quoted 3 lines

I think we already have a better-geared switch for that (--with-platform it is?), which is currently used to distinguish between BIOS and EFI boot code on both x86 and x86-64.

--with-platform is an override for the default platform that is otherwise derived from target_vendor, AFAIK.

Signature · 5 lines
Robert Millan

  The DRM opt-in fallacy: "Your data belongs to us. We will decide when (and
  how) you may access your data; but nobody's threatening your freedom: we
  still allow you to remove your data and not access it at all."
#11 lore raw
In reply to Christian Franke
Quoted 2 lines · Christian Franke

Robert: Patch is IMO GTG.

Ok. If noone objects I'll check it in soon.

Signature · 5 lines
Robert Millan

  The DRM opt-in fallacy: "Your data belongs to us. We will decide when (and
  how) you may access your data; but nobody's threatening your freedom: we
  still allow you to remove your data and not access it at all."
#12 lore raw
Quoted 5 lines · Robert Millan
Quoted 2 lines · Christian Franke

Robert: Patch is IMO GTG.

Ok. If noone objects I'll check it in soon.

Committed.

Signature · 5 lines
Robert Millan

  The DRM opt-in fallacy: "Your data belongs to us. We will decide when (and
  how) you may access your data; but nobody's threatening your freedom: we
  still allow you to remove your data and not access it at all."