multiboot2: Clean up the example kernel
v3 cover multiboot2: Clean up the example kernel
This cleans up the Multiboot2 example kernel in a few ways:
* It fixes the build with current gnulib.
* Change from obsolete .bzrignore to .gitignore.
* It fixes the build on non-MIPS systems such as x86.
* It fixes the Multiboot2 header tag alignment in the assembly language boot_*.S files to 8 byte boundaries.
* It fixes the alignment in boot_mips.S to align to 8 byte boundaries instead of 256 byte boundaries.
This is v3 of a previous patch set. The differences mainly are
* improved .gitignore
* fix the case logic for the ix86 matching pattern ([[]])
* fix mb2 header tag alignment
* fix alignment on mips
I am posting this right now to make clear that my previous patch set version v1/v2 had issues, but this v3 certainly can use some improvement as well. Also, this is not very urgent, so putting this off until after the next GRUB release is very reasonable.
Hans Ulrich Niedermann (4):
multiboot2: Allow autogen.sh to run with current gnulib
multiboot2: Use .gitignore files
multiboot2: Make example kernel build at least for i386
multiboot2: fix example kernel header tag alignment
.bzrignore | 30 -------------------
.gitignore | 51 ++++++++++++++++++++++++++++++++
configure.ac | 37 +++++++++++++++++++----
doc/.gitignore | 2 ++
doc/Makefile.am | 9 +++---
doc/{boot.S => boot_i386.S} | 9 ++++--
doc/boot_mips.S | 17 +++++++----
doc/multiboot.texi | 59 +++++++++++++++++++++++--------------
8 files changed, 144 insertions(+), 70 deletions(-)
delete mode 100644 .bzrignore
create mode 100644 .gitignore
create mode 100644 doc/.gitignore
rename doc/{boot.S => boot_i386.S} (93%)Interdiff:
.gitignore +13 -5
@@ -1,4 +1,4 @@ -# Generated by autogen.sh +# Generated by "autogen.sh" /gendocs.sh /lib/ /m4/
@@ -6,6 +6,8 @@ /NEWS /README +Makefile.in + /aclocal.m4 /autom4te.cache/ /compile
@@ -20,9 +22,7 @@ /INSTALL /missing -Makefile.in - -# Generated by configure +# Generated by "configure" Makefile .deps/
@@ -31,7 +31,7 @@ Makefile /config.status /stamp-h1 -# Generated by make +# Generated by "make" /doc/*.c.texi /doc/*.h.texi /doc/*.S.texi
@@ -40,4 +40,12 @@ Makefile /doc/version.texi *.o + /doc/kernel + +# Generated by "make web-manual" +/doc/manual/ +/doc/multiboot.aux +/doc/multiboot.cp +/doc/multiboot.log +/doc/multiboot.toc
configure.ac +17 -11
@@ -34,17 +34,23 @@ AC_SUBST(CCAS) dnl Build the example Multiboot kernel (if possible on this host) AC_CANONICAL_HOST -case "$host" in #( - i[3456]86-*) kernel_boot_arch=i386 - ;; #( - x86_64-*) kernel_boot_arch=i386 - kernel_ccasflags="-m32" - kernel_cflags="-m32" - ;; #( - mips-*) kernel_boot_arch=mips - ;; #( - *) kernel_boot_arch=unsupported - ;; +kernel_ccasflags= +kernel_cflags= +case "$host_cpu" in #( + i[[3456]]86) + kernel_boot_arch=i386 + ;; #( + x86_64) + kernel_boot_arch=i386 + kernel_ccasflags="-m32" + kernel_cflags="-m32" + ;; #( + mips*) + kernel_boot_arch=mips + ;; #( + *) + kernel_boot_arch=unsupported + ;; esac AC_SUBST([kernel_boot_arch]) AC_SUBST([kernel_ccasflags])
doc/boot_i386.S +5 -0
@@ -56,6 +56,7 @@ multiboot_header: /* checksum */ .long -(MULTIBOOT2_HEADER_MAGIC + MULTIBOOT_ARCHITECTURE_I386 + (multiboot_header_end - multiboot_header)) #ifndef __ELF__ + .align 8 address_tag_start: .short MULTIBOOT_HEADER_TAG_ADDRESS .short MULTIBOOT_HEADER_TAG_OPTIONAL
@@ -69,6 +70,7 @@ address_tag_start: /* bss_end_addr */ .long _end address_tag_end: + .align 8 entry_address_tag_start: .short MULTIBOOT_HEADER_TAG_ENTRY_ADDRESS .short MULTIBOOT_HEADER_TAG_OPTIONAL
@@ -77,6 +79,7 @@ entry_address_tag_start: .long multiboot_entry entry_address_tag_end: #endif /* __ELF__ */ + .align 8 framebuffer_tag_start: .short MULTIBOOT_HEADER_TAG_FRAMEBUFFER .short MULTIBOOT_HEADER_TAG_OPTIONAL
@@ -85,10 +88,12 @@ framebuffer_tag_start: .long 768 .long 32 framebuffer_tag_end: + .align 8 .short MULTIBOOT_HEADER_TAG_END .short 0 .long 8 multiboot_header_end: + multiboot_entry: /* Initialize the stack pointer. */ movl $(stack + STACK_SIZE), %esp
doc/boot_mips.S +11 -6
@@ -46,8 +46,8 @@ _start: nop /* Align 64 bits boundary. */ - .align 8 - + .balign 8 + /* Multiboot header. */ multiboot_header: /* magic */
@@ -59,7 +59,8 @@ multiboot_header: /* checksum */ .long -(MULTIBOOT2_HEADER_MAGIC + MULTIBOOT_ARCHITECTURE_MIPS32 + (multiboot_header_end - multiboot_header)) #ifndef __ELF__ -address_tag_start: + .balign 8 +address_tag_start: .short MULTIBOOT_HEADER_TAG_ADDRESS .short MULTIBOOT_HEADER_TAG_OPTIONAL .long address_tag_end - address_tag_start
@@ -72,7 +73,8 @@ address_tag_start: /* bss_end_addr */ .long _end address_tag_end: -entry_address_tag_start: + .balign 8 +entry_address_tag_start: .short MULTIBOOT_HEADER_TAG_ENTRY_ADDRESS .short MULTIBOOT_HEADER_TAG_OPTIONAL .long entry_address_tag_end - entry_address_tag_start
@@ -80,7 +82,8 @@ entry_address_tag_start: .long multiboot_entry entry_address_tag_end: #endif /* __ELF__ */ -framebuffer_tag_start: + .balign 8 +framebuffer_tag_start: .short MULTIBOOT_HEADER_TAG_FRAMEBUFFER .short MULTIBOOT_HEADER_TAG_OPTIONAL .long framebuffer_tag_end - framebuffer_tag_start
@@ -88,10 +91,12 @@ framebuffer_tag_start: .long 768 .long 32 framebuffer_tag_end: + .balign 8 .short MULTIBOOT_HEADER_TAG_END .short 0 .long 8 multiboot_header_end: + multiboot_entry: /* Initialize the stack pointer. */ lui $sp, %hi (stack + STACK_SIZE)
@@ -114,6 +119,6 @@ loop: nop halt_message: .asciz "Halted." - .align 8 + .balign 8 /* Our stack area. */ .comm stack, STACK_SIZE
doc/multiboot.texi +4 -4
@@ -1566,12 +1566,12 @@ executes @file{boot_*.S}, it initializes the stack pointer and If @code{cmain} returns to the callee, then it shows a message to inform the user of the halt state and stops forever until you push the -reset key. The file @file{kernel.c} contains the function +system reset button. The file @file{kernel.c} contains the function @code{cmain}, which checks if the magic number passed by the boot loader is valid and so on, and some functions to print messages on the -screen. The file @file{multiboot2.h} defines some macros, such as the -magic number for the Multiboot2 header, the Multiboot2 header -structure and the Multiboot2 information structure. +screen. The file @file{multiboot2.h} defines some C preprocessor +macros, such as the magic number for the Multiboot2 header, the +Multiboot2 header structure and the Multiboot2 information structure. @menu * multiboot2.h::
Signature · 1 line
2.26.2
PATCH v3 1/4 multiboot2: Allow autogen.sh to run with current gnulib
Current gnulib's gnulib-tool (which is called by autogen.sh) only runs if configure.ac contains at least AC_PREREQ(2.63).
Requiring Autoconf 2.63 (release tagged in 2008-09) instead of Autoconf 2.59 (release 2.59 tagged in 2003-11, 2.59d in 2006-06) should not affect too many systems in 2020.
configure.ac +1 -1
@@ -12,7 +12,7 @@ dnl "AS IS" CONDITION. THE FREE SOFTWARE FOUNDATION DISCLAIMS ANY dnl LIABILITY OF ANY KIND FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE dnl USE OF THIS SOFTWARE. -AC_PREREQ(2.59) +AC_PREREQ(2.63) AC_INIT([Multiboot], [2.0], [[email protected]]) AC_CONFIG_SRCDIR([doc/multiboot.texi]) AC_CONFIG_HEADER([config.h])
Signature · 1 line
2.26.2
Re: [MULTIBOOT2 SPEC PATCH v3 1/4] multiboot2: Allow autogen.sh to run with current gnulib
Quoted 8 lines · Hans Ulrich Niedermann
Current gnulib's gnulib-tool (which is called by
autogen.sh) only runs if configure.ac contains at leastAC_PREREQ(2.63).Requiring Autoconf 2.63 (release tagged in 2008-09) instead of Autoconf 2.59 (release 2.59 tagged in 2003-11, 2.59d in 2006-06) should not affect too many systems in 2020.
Signed-off-by Hans Ulrich Niedermann <[email protected]>
Daniel
PATCH v3 2/4 multiboot2: Use .gitignore files
In reply to Hans Ulrich NiedermannAdd .gitignore file and remove the obsolete .bzrignore file.
The doc/.gitignore file is generated by gnulib-tool so we keep that file in addition to the top-level .gitignore file which covers everything else in the subdirectory tree.
delete mode 100644 .bzrignore create mode 100644 .gitignore create mode 100644 doc/.gitignore
.bzrignore deleted +0 -30
@@ -1,30 +0,0 @@ -gendocs.sh -lib -m4 -.deps -Makefile -autom4te.cache -config.cache -config.h -config.log -config.status -stamp-h -stamp-h1 -doc/Makefile -doc/*.info* -doc/.deps -doc/gendocs_template -doc/mdate-sh -doc/Makefile.in -doc/boot.S.texi -doc/kernel -doc/kernel.c.texi -doc/multiboot.h.texi -doc/texinfo.tex -Makefile.in -aclocal.m4 -config.h.in -configure -depcomp -install-sh -missing
.gitignore new file +49 -0
49 lines not on this page · raw message
doc/.gitignore new file +2 -0
@@ -0,0 +1,2 @@ +/gendocs_template +/gendocs_template_min
Signature · 1 line
2.26.2
Re: [MULTIBOOT2 SPEC PATCH v3 2/4] multiboot2: Use .gitignore files
Quoted 5 lines · Hans Ulrich Niedermann
Add .gitignore file and remove the obsolete .bzrignore file.
The doc/.gitignore file is generated by gnulib-tool so we keep that file in addition to the top-level .gitignore file which covers everything else in the subdirectory tree.
If it is generated by gnulib-tool why do we need to create it?
Quoted 63 lines
Signed-off-by Hans Ulrich Niedermann <[email protected]>delete mode 100644 .bzrignore create mode 100644 .gitignore create mode 100644 doc/.gitignore.bzrignore deleted +0
-30@@ -1,30 +0,0 @@ -gendocs.sh -lib -m4 -.deps -Makefile -autom4te.cache -config.cache -config.h -config.log -config.status -stamp-h -stamp-h1 -doc/Makefile -doc/*.info* -doc/.deps -doc/gendocs_template -doc/mdate-sh -doc/Makefile.in -doc/boot.S.texi -doc/kernel -doc/kernel.c.texi -doc/multiboot.h.texi -doc/texinfo.tex -Makefile.in -aclocal.m4 -config.h.in -configure -depcomp -install-sh -missing.gitignore new file +15
-0@@ -0,0 +1,49 @@ +# Generated by "autogen.sh" +/gendocs.sh +/lib/ +/m4/ + +/NEWS +/README + +Makefile.in + +/aclocal.m4 +/autom4te.cache/ +/compile +/config.h.in +/configure
.gitignore line 25
Quoted 10 lines
.gitignore new file +10
-0+/depcomp +/doc/mdate-sh +/doc/texinfo.tex +/install-sh +/INSTALL +/missing + +# Generated by "configure" +Makefile +.deps/
Slash, "/", prefix?
Quoted 5 lines
.gitignore new file +5
-0+ +/config.h +/config.log +/config.status +/stamp-h1
.gitignore line 40
Quoted 10 lines
.gitignore new file +10
-0+ +# Generated by "make" +/doc/*.c.texi +/doc/*.h.texi +/doc/*.S.texi +/doc/multiboot.info* +/doc/stamp-vti +/doc/version.texi + +*.o
Could you be more consistent and list prefixed and not prefixed variants in the same order in each section?
Daniel
Re: [MULTIBOOT2 SPEC PATCH v3 2/4] multiboot2: Use .gitignore files
Quoted 2 lines · On Wed, 13 May 2020 18:21:11 +0200 Daniel Kiper <[email protected]> wrote:
On Fri, May 08, 2020 at 06:50:47AM +0200, Hans Ulrich Niedermann wrote:
[... lots of gitignore stuff again ... ]
Quoted 11 lines
.gitignore line 25
Quoted 9 lines
.gitignore new file +9
-0+/doc/mdate-sh +/doc/texinfo.tex +/install-sh +/INSTALL +/missing + +# Generated by "configure" +Makefile +.deps/Slash, "/", prefix?
Trailing slash just means "match directory" as opposed to "match file or directory", so ".deps/" matches every directory named ".deps", while the ".deps" pattern also matches files called ".deps". I'd rather such a file showed up in "git status" so such a weird thing would be noticed.
The leading slash to match only relative to the .gitignore file's directory is only implied with slashes somewhere within the pattern, e.g. "doc/texinfo.tex" and "/doc/textinfo.tex" have the identical meaning.
Quoted 10 lines
Quoted 10 lines
.gitignore new file +10
-0+ +/config.h +/config.log +/config.status +/stamp-h1 + +# Generated by "make" +/doc/*.c.texi +/doc/*.h.texi +/doc/*.S.texi
Quoted 8 lines
.gitignore line 40
Quoted 5 lines
.gitignore new file +5
-0+/doc/multiboot.info* +/doc/stamp-vti +/doc/version.texi + +*.oCould you be more consistent and list prefixed and not prefixed variants in the same order in each section?
Good point.
Uli
Re: [MULTIBOOT2 SPEC PATCH v3 2/4] multiboot2: Use .gitignore files
In reply to Daniel KiperQuoted 9 lines · On Wed, 13 May 2020 18:21:11 +0200 Daniel Kiper <[email protected]> wrote:
Quoted 5 lines · Hans Ulrich Niedermann
Add .gitignore file and remove the obsolete .bzrignore file.
The doc/.gitignore file is generated by gnulib-tool so we keep that file in addition to the top-level .gitignore file which covers everything else in the subdirectory tree.
If it is generated by gnulib-tool why do we need to create it?
The alternatives to adding doc/.gitignore to the version controlled files are to either
a) have the top-level .gitignore file ignore "/doc/.gitignore"
b) have the gnulib-generated doc/.gitignore file just flying around, neither .gitignored like all other generated files, nor under version control
Option a) sounds like the beginning of a road into world of very weird and unexpected things, and option b) defeats the purpose of having any .gitignore or .bzrignore etc. files at all.
I much prefer ignoring a few patterns which gnulib is considering useful to be ignored at this time by adding doc/.gitignore. Well, now that I am thinking about this...
c) Maybe gnulib would recognize when the top-level .gitignore files does the job gnulib would create the doc/.gitignore file for. Then we would just have two more patterns in our top-level .gitignore file, and gnulib would not create a doc/.gitignore file.
I will have to investigate option c). I would prefer that over my original option from the patch (adding the doc/.gitignore file as generated by gnulib).
Uli
Re: [MULTIBOOT2 SPEC PATCH v3 2/4] multiboot2: Use .gitignore files
Quoted 30 lines · On Wed, 13 May 2020 23:11:03 +0200 Hans Ulrich Niedermann…
Quoted 9 lines · On Wed, 13 May 2020 18:21:11 +0200 Daniel Kiper <[email protected]> wrote:
Quoted 5 lines · Hans Ulrich Niedermann
Add .gitignore file and remove the obsolete .bzrignore file.
The doc/.gitignore file is generated by gnulib-tool so we keep that file in addition to the top-level .gitignore file which covers everything else in the subdirectory tree.
If it is generated by gnulib-tool why do we need to create it?
The alternatives to adding doc/.gitignore to the version controlled files are to either
a) have the top-level .gitignore file ignore "/doc/.gitignore"
b) have the gnulib-generated doc/.gitignore file just flying around, neither .gitignored like all other generated files, nor under version control
Option a) sounds like the beginning of a road into world of very weird and unexpected things, and option b) defeats the purpose of having any .gitignore or .bzrignore etc. files at all.
I much prefer ignoring a few patterns which gnulib is considering useful to be ignored at this time by adding doc/.gitignore. Well, now that I am thinking about this...
Quoted 8 lines
c) Maybe gnulib would recognize when the top-level .gitignore files does the job gnulib would create the doc/.gitignore file for.
Then we would just have two more patterns in our top-level .gitignore
file, and gnulib would not create a doc/.gitignore file.I will have to investigate option c). I would prefer that over my original option from the patch (adding the doc/.gitignore file as generated by gnulib).
No luck here. I have just examined what gnulib-tool does in the presence of the proper patterns having been added to the top-level .gitignore file with an absent doc/.gitignore file: gnulib-tool creates a new file doc/.gitignore to which it adds its patterns.
So option c) is out, option a) and option b) both create unnecessary friction in the day to day use of git, I am therefore still convinced the best way to go is to have a top-level .gitignore with our own stuff, and a doc/.gitignore file with whatever patterns gnulib-tool wants to put in there together with the scripts it puts into doc/.
Note that gnulib-tool creates a few more subdirectory .gitignore files in /lib/ and /m4/. We just do not notice those as we .gitignore the whole subdirectory for those two anyway. The only thing from gnulib we are interested in are the scripts generating HTML from doc/multiboot.texi, not m4 macros or C source files.
Uli
PATCH v3 3/4 multiboot2: Make example kernel build at least for i386
In reply to Hans Ulrich NiedermannThe example kernel has assembly language boot code for both i386 and mips, but the mips assembly code used to be built unconditionally, even if the build is using non-mips build tools such as for x86_64 or i386.
This makes the example kernel build at least for i386, both on i386 and on x86_64 hosts.
* renames the i386 boot code from boot.S to boot_i386.S to go along with the mips boot code in boot_mips.S
* adds AC_CANONICAL_HOST to select the proper boot code:
* i386 if building on x86_64 (adds -m32) or on i[3456]86
* mips if building for mips*
* do not build the kernel if building for another system
* adds m4 quoting and uses AS_HELP_STRING use in configure.ac
* fixes the name of the constants used in boot_i386.S to use the actual constant names from multiboot2.h
* documents both boot_i386.S and boot_mips.S in the multiboot.texi page
Building the example kernel for mips still needs some work, as I have not managed to build for mips yet with either my mips64-linux-gnu cross-compilation toolchain on Fedora 32 or on my Debian 10.3 MIPS install running in qemu-system-mips.
rename doc/{boot.S => boot_i386.S} (94%).gitignore +2 -0
@@ -11,7 +11,9 @@ Makefile.in /aclocal.m4 /autom4te.cache/ /compile +/config.guess /config.h.in +/config.sub /configure /depcomp /doc/mdate-sh
configure.ac +30 -5
@@ -32,11 +32,36 @@ dnl Because recent automake complains about AS, set it here. CCAS="$CC" AC_SUBST(CCAS) -dnl Build the example Multiboot kernel. -AC_ARG_ENABLE(example-kernel, - [ --enable-example-kernel - build the example Multiboot kernel]) -AM_CONDITIONAL(BUILD_EXAMPLE_KERNEL, test "x$enable_example_kernel" = xyes) +dnl Build the example Multiboot kernel (if possible on this host) +AC_CANONICAL_HOST +kernel_ccasflags= +kernel_cflags= +case "$host_cpu" in #( + i[[3456]]86) + kernel_boot_arch=i386 + ;; #( + x86_64) + kernel_boot_arch=i386 + kernel_ccasflags="-m32" + kernel_cflags="-m32" + ;; #( + mips*) + kernel_boot_arch=mips + ;; #( + *) + kernel_boot_arch=unsupported + ;; +esac +AC_SUBST([kernel_boot_arch]) +AC_SUBST([kernel_ccasflags]) +AC_SUBST([kernel_cflags]) + +AC_ARG_ENABLE([example-kernel], [dnl +AS_HELP_STRING([--enable-example-kernel], + [build the example Multiboot2 kernel])]) + +AM_CONDITIONAL([BUILD_EXAMPLE_KERNEL], + [test "x$enable_example_kernel" = xyes && test "x$kernel_boot_arch" != "xunsupported"]) dnl Because recent automake complains about CCASFLAGS, set it here. CCASFLAGS='$(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(CPPFLAGS) $(CFLAGS)'
doc/Makefile.am +5 -4
@@ -1,6 +1,6 @@ info_TEXINFOS = multiboot.texi -EXAMPLES = boot_mips.S kernel.c multiboot2.h -multiboot_TEXINFOS = boot.S.texi kernel.c.texi multiboot2.h.texi +EXAMPLES = boot_i386.S boot_mips.S kernel.c multiboot2.h +multiboot_TEXINFOS = boot_i386.S.texi boot_mips.S.texi kernel.c.texi multiboot2.h.texi SRC2TEXI = src2texi noinst_SCRIPTS = $(SRC2TEXI) EXTRA_PROGRAMS = kernel
@@ -9,8 +9,9 @@ pkginclude_HEADERS = multiboot2.h # The example kernel is built if you specify --enable-example-kernel. if BUILD_EXAMPLE_KERNEL noinst_PROGRAMS = kernel -kernel_SOURCES = $(EXAMPLES) -kernel_CFLAGS = -fno-builtin -nostdinc -O -g -Wall \ +kernel_SOURCES = boot_$(kernel_boot_arch).S kernel.c multiboot2.h +kernel_CCASFLAGS = $(kernel_ccasflags) +kernel_CFLAGS = $(kernel_cflags) -fno-builtin -nostdinc -O -g -Wall \ -imacros $(top_builddir)/config.h kernel_LDFLAGS = -nostdlib -Wl,-N -Wl,-Ttext -Wl,80100000 -Wl,--build-id=none
doc/boot_i386.S renamed from doc/boot.S +2 -2
@@ -50,11 +50,11 @@ multiboot_header: /* magic */ .long MULTIBOOT2_HEADER_MAGIC /* ISA: i386 */ - .long GRUB_MULTIBOOT_ARCHITECTURE_I386 + .long MULTIBOOT_ARCHITECTURE_I386 /* Header length. */ .long multiboot_header_end - multiboot_header /* checksum */ - .long -(MULTIBOOT2_HEADER_MAGIC + GRUB_MULTIBOOT_ARCHITECTURE_I386 + (multiboot_header_end - multiboot_header)) + .long -(MULTIBOOT2_HEADER_MAGIC + MULTIBOOT_ARCHITECTURE_I386 + (multiboot_header_end - multiboot_header)) #ifndef __ELF__ address_tag_start: .short MULTIBOOT_HEADER_TAG_ADDRESS
doc/multiboot.texi +37 -22
76 lines not on this page · raw message
Signature · 1 line
2.26.2
Re: [MULTIBOOT2 SPEC PATCH v3 3/4] multiboot2: Make example kernel build at least for i386
Quoted 17 lines · Hans Ulrich Niedermann
The example kernel has assembly language boot code for both i386 and mips, but the mips assembly code used to be built unconditionally, even if the build is using non-mips build tools such as for
x86_64or i386.This makes the example kernel build at least for i386, both on i386 and on
x86_64hosts.* renames the i386 boot code from
boot.Stoboot_i386.Sto go along with the mips boot code inboot_mips.S* adds
AC_CANONICAL_HOSTto select the proper boot code:* i386 if building on
x86_64(adds -m32) or on i[3456]86
* mips if building for mips*
* do not build the kernel if building for another system
Quoted 7 lines
* adds m4 quoting and uses
AS_HELP_STRINGuse in configure.ac* fixes the name of the constants used in
boot_i386.Sto use the actual constant names frommultiboot2.h* documents both
boot_i386.Sandboot_mips.Sin the multiboot.texi page
May I ask you to split this patch into logical parts?
Quoted 4 lines
Building the example kernel for mips still needs some work, as I have not managed to build for mips yet with either my mips64-linux-gnu cross-compilation toolchain on Fedora 32 or on my Debian 10.3 MIPS install running in qemu-system-mips.
This is not big deal. We can fix it later.
Daniel
Re: [MULTIBOOT2 SPEC PATCH v3 3/4] multiboot2: Make example kernel build at least for i386
Quoted 21 lines · On Wed, 13 May 2020 18:24:28 +0200 Daniel Kiper <[email protected]> wrote:
Quoted 19 lines · Hans Ulrich Niedermann
The example kernel has assembly language boot code for both i386 and mips, but the mips assembly code used to be built unconditionally, even if the build is using non-mips build tools such as for
x86_64or i386.This makes the example kernel build at least for i386, both on i386 and on
x86_64hosts.* renames the i386 boot code from
boot.Stoboot_i386.Sto go along with the mips boot code inboot_mips.S* adds
AC_CANONICAL_HOSTto select the proper boot code:* i386 if building on
x86_64(adds -m32) or on i[3456]86
* mips if building for mips*
* do not build the kernel if building for another system* adds m4 quoting and uses
AS_HELP_STRINGuse in configure.ac
Quoted 7 lines
Quoted 5 lines
* fixes the name of the constants used in
boot_i386.Sto use the actual constant names frommultiboot2.h* documents both
boot_i386.Sandboot_mips.Sin the multiboot.texi pageMay I ask you to split this patch into logical parts?
Being a little slow on the uptake, I only just figured out why I ended up with that big of a patch.
When given a building and working source tree, I usually make my commits such that they are small, self-contained and keep the source tree in a building and working state.
That is why I usually add changes until it finally builds/works.
However, that is a useless argument here, as the state of the source tree *before* my change is non-building maybe-working, so if the source tree does not build *after* my small, self-contained change, my change has not made it worse and can thusly still be considered good.
Changing my normal modus operandi accordingly; considering failing builds as normal for now.
Expect a good dozen patches in a day or two.
Uli
PATCH v3 4/4 multiboot2: fix example kernel header tag alignment
In reply to Hans Ulrich NiedermannProperly align all Multiboot2 header tags to 8 byte boundaries as per the Multiboot2 specification.
Note that the assembler directive ".align 8" is machine
dependent: On i386, it means "align to 8 byte boundary". On mips, it means that the lower 8 bits of address must be zero, i.e. ".align 8" aligns to a 256 byte boundary!
Therefore, this changes boot_mips.S to use ".balign 8" where it had mistakenly used ".align 8" before.
The boot_i386.S file keeps using ".align 8".
Note also that the Multiboot2 header termination tag (unfortunately unlabeled in the source) actually needs that extra alignment to an 8 byte boundary, as the preceding tag (framebuffer_tag_start) is 20 bytes long which is not a multiple of 8.
You can add "-Wa,-adhlns=$(@:.o=.lst)" to kernel_CCASFLAGS to generate a listing file which shows the offsets of the labels in the boot_*.S files to verify the alignment issues.
Note also that the "grub-file --is-x86-multiboot2" utility does not recognize the wrong alignment of the Multiboot2 header tags, as it only considers the beginning of the Multiboot2 header and completely ignores the Multiboot2 header tags.
doc/boot_i386.S +5 -0
@@ -56,6 +56,7 @@ multiboot_header: /* checksum */ .long -(MULTIBOOT2_HEADER_MAGIC + MULTIBOOT_ARCHITECTURE_I386 + (multiboot_header_end - multiboot_header)) #ifndef __ELF__ + .align 8 address_tag_start: .short MULTIBOOT_HEADER_TAG_ADDRESS .short MULTIBOOT_HEADER_TAG_OPTIONAL
@@ -69,6 +70,7 @@ address_tag_start: /* bss_end_addr */ .long _end address_tag_end: + .align 8 entry_address_tag_start: .short MULTIBOOT_HEADER_TAG_ENTRY_ADDRESS .short MULTIBOOT_HEADER_TAG_OPTIONAL
@@ -77,6 +79,7 @@ entry_address_tag_start: .long multiboot_entry entry_address_tag_end: #endif /* __ELF__ */ + .align 8 framebuffer_tag_start: .short MULTIBOOT_HEADER_TAG_FRAMEBUFFER .short MULTIBOOT_HEADER_TAG_OPTIONAL
@@ -85,10 +88,12 @@ framebuffer_tag_start: .long 768 .long 32 framebuffer_tag_end: + .align 8 .short MULTIBOOT_HEADER_TAG_END .short 0 .long 8 multiboot_header_end: + multiboot_entry: /* Initialize the stack pointer. */ movl $(stack + STACK_SIZE), %esp
doc/boot_mips.S +11 -6
@@ -46,8 +46,8 @@ _start: nop /* Align 64 bits boundary. */ - .align 8 - + .balign 8 + /* Multiboot header. */ multiboot_header: /* magic */
@@ -59,7 +59,8 @@ multiboot_header: /* checksum */ .long -(MULTIBOOT2_HEADER_MAGIC + MULTIBOOT_ARCHITECTURE_MIPS32 + (multiboot_header_end - multiboot_header)) #ifndef __ELF__ -address_tag_start: + .balign 8 +address_tag_start: .short MULTIBOOT_HEADER_TAG_ADDRESS .short MULTIBOOT_HEADER_TAG_OPTIONAL .long address_tag_end - address_tag_start
@@ -72,7 +73,8 @@ address_tag_start: /* bss_end_addr */ .long _end address_tag_end: -entry_address_tag_start: + .balign 8 +entry_address_tag_start: .short MULTIBOOT_HEADER_TAG_ENTRY_ADDRESS .short MULTIBOOT_HEADER_TAG_OPTIONAL .long entry_address_tag_end - entry_address_tag_start
@@ -80,7 +82,8 @@ entry_address_tag_start: .long multiboot_entry entry_address_tag_end: #endif /* __ELF__ */ -framebuffer_tag_start: + .balign 8 +framebuffer_tag_start: .short MULTIBOOT_HEADER_TAG_FRAMEBUFFER .short MULTIBOOT_HEADER_TAG_OPTIONAL .long framebuffer_tag_end - framebuffer_tag_start
@@ -88,10 +91,12 @@ framebuffer_tag_start: .long 768 .long 32 framebuffer_tag_end: + .balign 8 .short MULTIBOOT_HEADER_TAG_END .short 0 .long 8 multiboot_header_end: + multiboot_entry: /* Initialize the stack pointer. */ lui $sp, %hi (stack + STACK_SIZE)
@@ -114,6 +119,6 @@ loop: nop halt_message: .asciz "Halted." - .align 8 + .balign 8 /* Our stack area. */ .comm stack, STACK_SIZE
Signature · 1 line
2.26.2
Re: [MULTIBOOT2 SPEC PATCH v3 4/4] multiboot2: fix example kernel header tag alignment
Quoted 8 lines · Hans Ulrich Niedermann
Properly align all Multiboot2 header tags to 8 byte boundaries as per the Multiboot2 specification.
Note that the assembler directive ".align 8" is machine
dependent: On i386, it means "align to 8 byte boundary". On mips, it means that the lower 8 bits of address must be zero, i.e. ".align 8" aligns to a 256 byte boundary!
Quoted 4 lines
Therefore, this changes
boot_mips.Sto use ".balign 8" where it had mistakenly used ".align 8" before.The
boot_i386.Sfile keeps using ".align 8".
I would use ".balign 8" everywhere to avoid confusion.
Quoted 10 lines
Note also that the Multiboot2 header termination tag (unfortunately unlabeled in the source) actually needs that extra alignment to an 8 byte boundary, as the preceding tag (
framebuffer_tag_start) is 20 bytes long which is not a multiple of 8.You can add "-Wa,-adhlns=$(@:.o=.lst)" to
kernel_CCASFLAGSto generate a listing file which shows the offsets of the labels in the boot_*.S files to verify the alignment issues.
Quoted 5 lines
Note also that the "grub-file --is-x86-multiboot2" utility does not recognize the wrong alignment of the Multiboot2 header tags, as it only considers the beginning of the Multiboot2 header and completely ignores the Multiboot2 header tags.
Please also mention here that you are additionally fixing some formatting errors.
Daniel