* docs/grub.texi (Platform limitations): New section.
(Platform-specific operations): Likewise. * docs/grub-dev.texi (Porting): Likewise.
This commit is contained in:
parent
7626111087
commit
ca1dacea3c
3 changed files with 483 additions and 3 deletions
|
@ -77,6 +77,7 @@ This edition documents version @value{VERSION}.
|
|||
* Finding your way around::
|
||||
* Coding style::
|
||||
* Contributing Changes::
|
||||
* Porting::
|
||||
* Error Handling::
|
||||
* CIA::
|
||||
* BIOS port memory map::
|
||||
|
@ -447,6 +448,315 @@ If your intention is to just get started, please do not submit a inclusion
|
|||
request. Instead, please subscribe to the mailing list, and communicate first
|
||||
(e.g. sending a patch, asking a question, commenting on another message...).
|
||||
|
||||
@node Porting
|
||||
@chapter Porting
|
||||
|
||||
GRUB2 is designed to be easily portable accross platforms. But because of the
|
||||
nature of bootloader every new port must be done separately. Here is how I did
|
||||
MIPS (loongson and ARC) and Xen ports. Note than this is more of suggestions,
|
||||
not absolute truth.
|
||||
|
||||
First of all grab any architecture specifications you can find in public
|
||||
(please avoid NDA).
|
||||
|
||||
First stage is ``Hello world''. I've done it outside of GRUB for simplicity.
|
||||
Your task is to have a small program which is loadable as bootloader and
|
||||
clearly shows its presence to you. If you have easily accessible console
|
||||
you can just print a message. If you have a mapped framebuffer you know address
|
||||
of, you can draw a square. If you have a debug facility, just hanging without
|
||||
crashing might be enough. For the first stage you can choose to load the
|
||||
bootloader across the network since format for network image is often easier
|
||||
than for local boot and it skips the need of small intermediary stages and
|
||||
nvram handling. Additionally you can often have a good idea of the needed
|
||||
format by running ``file'' on any netbootable executable for given platform.
|
||||
|
||||
This program should probably have 2 parts: an assembler and C one. Assembler one
|
||||
handles BSS cleaning and other needed setup (on some platforms you may need
|
||||
to switch modes or copy the executable to its definitive position). So your code
|
||||
may look like (x86 assembly for illustration purposes)
|
||||
|
||||
@example
|
||||
.globl _start
|
||||
_start:
|
||||
movl $_bss_start, %edi
|
||||
movl $_end, %ecx
|
||||
subl %edi, %ecx
|
||||
xorl %eax, %eax
|
||||
cld
|
||||
rep
|
||||
stosb
|
||||
call main
|
||||
@end example
|
||||
|
||||
@example
|
||||
|
||||
static const char msg[] = "Hello, world";
|
||||
|
||||
void
|
||||
putchar (int c)
|
||||
@{
|
||||
...
|
||||
@}
|
||||
|
||||
void
|
||||
main (void)
|
||||
@{
|
||||
const char *ptr = msg;
|
||||
while (*ptr)
|
||||
putchar (*ptr++);
|
||||
while (1);
|
||||
@}
|
||||
@end example
|
||||
|
||||
Sometimes you need a third file: assembly stubs for ABI-compatibility.
|
||||
|
||||
Once this file is functional it's time to move it into GRUB2. The startup
|
||||
assembly file goes to grub-core/kern/$cpu/$platform/startup.S. You should also
|
||||
include grub/symbol.h and replace call to entry point with call to
|
||||
EXT_C(grub_main). The C file goes to grub-core/kern/$cpu/$platform/init.c
|
||||
and its entry point is renamed to void grub_machine_init (void). Keep final
|
||||
infinite loop for now. Stubs file if any goes to
|
||||
grub-core/kern/$cpu/$platform/callwrap.S. Sometimes either $cpu or $platform
|
||||
is dropped if file is used on several cpus respectivelyplatforms.
|
||||
Check those locations if they already have what you're looking for.
|
||||
|
||||
Then modify in configure.ac the following parts:
|
||||
|
||||
CPU names:
|
||||
|
||||
@example
|
||||
case "$target_cpu" in
|
||||
i[[3456]]86) target_cpu=i386 ;;
|
||||
amd64) target_cpu=x86_64 ;;
|
||||
sparc) target_cpu=sparc64 ;;
|
||||
s390x) target_cpu=s390 ;;
|
||||
...
|
||||
esac
|
||||
@end example
|
||||
|
||||
Sometimes CPU have additional architecture names which don't influence booting.
|
||||
You might want to have some canonical name to avoid having bunch of identical
|
||||
platforms with different names.
|
||||
|
||||
NOTE: it doesn't influence compile optimisations which depend solely on
|
||||
chosen compiler and compile options.
|
||||
|
||||
@example
|
||||
if test "x$with_platform" = x; then
|
||||
case "$target_cpu"-"$target_vendor" in
|
||||
i386-apple) platform=efi ;;
|
||||
i386-*) platform=pc ;;
|
||||
x86_64-apple) platform=efi ;;
|
||||
x86_64-*) platform=pc ;;
|
||||
powerpc-*) platform=ieee1275 ;;
|
||||
...
|
||||
esac
|
||||
else
|
||||
...
|
||||
fi
|
||||
@end example
|
||||
|
||||
This part deals with guessing the platform from CPU and vendor. Sometimes you
|
||||
need to use 32-bit mode for booting even if OS runs in 64-bit one. If so add
|
||||
your platform to:
|
||||
|
||||
@example
|
||||
case "$target_cpu"-"$platform" in
|
||||
x86_64-efi) ;;
|
||||
x86_64-emu) ;;
|
||||
x86_64-*) target_cpu=i386 ;;
|
||||
powerpc64-ieee1275) target_cpu=powerpc ;;
|
||||
esac
|
||||
@end example
|
||||
|
||||
Add your platform to the list of supported ones:
|
||||
|
||||
@example
|
||||
case "$target_cpu"-"$platform" in
|
||||
i386-efi) ;;
|
||||
x86_64-efi) ;;
|
||||
i386-pc) ;;
|
||||
i386-multiboot) ;;
|
||||
i386-coreboot) ;;
|
||||
...
|
||||
esac
|
||||
@end example
|
||||
|
||||
If explicit -m32 or -m64 is needed add it to:
|
||||
|
||||
@example
|
||||
case "$target_cpu" in
|
||||
i386 | powerpc) target_m32=1 ;;
|
||||
x86_64 | sparc64) target_m64=1 ;;
|
||||
esac
|
||||
@end example
|
||||
|
||||
Finally you need to add a conditional to the following block:
|
||||
|
||||
@example
|
||||
AM_CONDITIONAL([COND_mips_arc], [test x$target_cpu = xmips -a x$platform = xarc])
|
||||
AM_CONDITIONAL([COND_sparc64_ieee1275], [test x$target_cpu = xsparc64 -a x$platform = xieee1275])
|
||||
AM_CONDITIONAL([COND_powerpc_ieee1275], [test x$target_cpu = xpowerpc -a x$platform = xieee1275])
|
||||
@end example
|
||||
|
||||
Next stop is gentpl.py. You need to add your platform to the list of supported
|
||||
ones (sorry that this list is duplicated):
|
||||
|
||||
@example
|
||||
GRUB_PLATFORMS = [ "emu", "i386_pc", "i386_efi", "i386_qemu", "i386_coreboot",
|
||||
"i386_multiboot", "i386_ieee1275", "x86_64_efi",
|
||||
"mips_loongson", "sparc64_ieee1275",
|
||||
"powerpc_ieee1275", "mips_arc", "ia64_efi",
|
||||
"mips_qemu_mips", "s390_mainframe" ]
|
||||
@end example
|
||||
|
||||
You may also want already to add new platform to one or several of available
|
||||
groups. In particular we always have a group for each CPU even when only
|
||||
one platform for given CPU is available.
|
||||
|
||||
Then comes grub-core/Makefile.core.def. In the block ``kernel'' you'll need
|
||||
to define ldflags for your platform ($cpu_$platform_ldflags). You also need to
|
||||
declare startup asm file ($cpu_$platform_startup) as well as any other files
|
||||
(e.g. init.c and callwrap.S) (e.g. $cpu_$platform = kern/$cpu/$platform/init.c).
|
||||
At this stage you will also need to add dummy dl.c and cache.S with functions
|
||||
grub_err_t grub_arch_dl_check_header (void *ehdr), grub_err_t
|
||||
grub_arch_dl_relocate_symbols (grub_dl_t mod, void *ehdr) (dl.c) and
|
||||
void grub_arch_sync_caches (void *address, grub_size_t len) (cache.S). They
|
||||
won't be used for now.
|
||||
|
||||
You will need to create directory include/$cpu/$platform and a file
|
||||
include/$cpu/types.h. The later folowing this template:
|
||||
|
||||
@example
|
||||
#ifndef GRUB_TYPES_CPU_HEADER
|
||||
#define GRUB_TYPES_CPU_HEADER 1
|
||||
|
||||
/* The size of void *. */
|
||||
#define GRUB_TARGET_SIZEOF_VOID_P 4
|
||||
|
||||
/* The size of long. */
|
||||
#define GRUB_TARGET_SIZEOF_LONG 4
|
||||
|
||||
/* mycpu is big-endian. */
|
||||
#define GRUB_TARGET_WORDS_BIGENDIAN 1
|
||||
/* Alternatively: mycpu is little-endian. */
|
||||
#undef GRUB_TARGET_WORDS_BIGENDIAN
|
||||
|
||||
#endif /* ! GRUB_TYPES_CPU_HEADER */
|
||||
@end example
|
||||
|
||||
You will also need to add a dummy file to datetime and setjmp modules to
|
||||
avoid any of it having no files. It can be just completely empty at this stage.
|
||||
|
||||
You'll need to make grub-mkimage.c (util/grub_mkimage.c) aware of the needed
|
||||
format. For most commonly used formats like ELF, PE, aout or raw the support
|
||||
is already present and you'll need to make it follow the existant code paths
|
||||
for your platform adding adjustments if necessary. When done compile:
|
||||
|
||||
@example
|
||||
./autogen.sh
|
||||
./configure --target=$cpu --with-platform=$platform TARGET_CC=.. OBJCOPY=... STRIP=...
|
||||
make > /dev/null
|
||||
@end example
|
||||
|
||||
And create image
|
||||
|
||||
@example
|
||||
./grub-mkimage -d grub-core -O $format_id -o test.img
|
||||
@end example
|
||||
|
||||
And it's time to test your test.img.
|
||||
|
||||
If it works next stage is to have heap, console and timer.
|
||||
|
||||
To have the heap working you need to determine which regions are suitable for
|
||||
heap usage, allocate them from firmware and map (if applicable). Then call
|
||||
grub_mm_init_region (vois *start, grub_size_t s) for every of this region.
|
||||
As a shortcut for early port you can allocate right after _end or have
|
||||
a big static array for heap. If you do you'll probably need to come back to
|
||||
this later. As for output console you should distinguish between an array of
|
||||
text, terminfo or graphics-based console. Many of real-world examples don't
|
||||
fit perfectly into any of these categories but one of the models is easier
|
||||
to be used as base. In second and third case you should add your platform to
|
||||
terminfokernel respectively videoinkernel group. A good example of array of
|
||||
text is i386-pc (kern/i386/pc/init.c and term/i386/pc/console.c).
|
||||
Of terminfo is ieee1275 (kern/ieee1275/init.c and term/ieee1275/console.c).
|
||||
Of video is loongson (kern/mips/loongson/init.c). Note that terminfo has
|
||||
to be inited in 2 stages: one before (to get at least rudimentary console
|
||||
as early as possible) and another after the heap (to get full-featured console).
|
||||
For the input there are string of keys, terminfo and direct hardware. For string
|
||||
of keys look at i386-pc (same files), for termino ieee1275 (same files) and for
|
||||
hardware loongson (kern/mips/loongson/init.c and term/at_keyboard.c).
|
||||
|
||||
For the timer you'll need to call grub_install_get_time_ms (...) with as sole
|
||||
argument a function returning a grub_uint64_t of a number of milliseconds
|
||||
elapsed since arbitrary point in the past.
|
||||
|
||||
Once these steps accomplished you can remove the inifinite loop and you should
|
||||
be able to get to the minimal console. Next step is to have module loading
|
||||
working. For this you'll need to fill kern/$cpu/dl.c and kern/$cpu/cache.S
|
||||
with real handling of relocations and respectively the real sync of I and D
|
||||
caches. Also you'll need to decide where in the image to store the modules.
|
||||
Usual way is to have it concatenated at the end. In this case you'll need to
|
||||
modify startup.S to copy modules out of bss to let's say ALIGN_UP (_end, 8)
|
||||
before cleaning out bss. You'll probably find useful to add total_module_size
|
||||
field to startup.S. In init.c you need to set grub_modbase to the address
|
||||
where modules can be found. You may need grub_modules_get_end () to avoid
|
||||
declaring the space occupied by modules as usable for heap. You can test modules
|
||||
with:
|
||||
|
||||
@example
|
||||
./grub-mkimage -d grub-core -O $format_id -o test.img hello
|
||||
@end example
|
||||
|
||||
and then running ``hello'' in the shell.
|
||||
|
||||
Once this works, you should think of implementing disk access. Look around
|
||||
disk/ for examples.
|
||||
|
||||
Then, very importantly, you probably need to implement the actual loader
|
||||
(examples available in loader/)
|
||||
|
||||
Last step to have minimally usable port is to add support to grub-install to
|
||||
put GRUB in a place where firmware or platform will pick it up.
|
||||
|
||||
Next steps are: filling datetime.c, setjmp.S, network (net/drivers),
|
||||
video (video/), halt (lib/), reboot (lib/).
|
||||
|
||||
Please add your platform to Platform limitations and Supported kernels chapter
|
||||
in user documentation and mention any steps you skipped which result in reduced
|
||||
features or performance. Here is the quick checklist of features. Some of them
|
||||
are less important than others and skipping them is completely ok, just needs
|
||||
to be mentioned in user documentation.
|
||||
|
||||
Checklist:
|
||||
@itemize
|
||||
@item Is heap big enough?
|
||||
@item Which charset is supported by console?
|
||||
@item Does platform have disk driver?
|
||||
@item Do you have network card support?
|
||||
@item Are you able to retrieve datetime (with date)?
|
||||
@item Are you able to set datetime (with date)?
|
||||
@item Is serial supported?
|
||||
@item Do you have direct disk support?
|
||||
@item Do you have direct keyboard support?
|
||||
@item Do you have USB support?
|
||||
@item Do you support loading through network?
|
||||
@item Do you support loading from disk?
|
||||
@item Do you support chainloading?
|
||||
@item Do you support network chainloading?
|
||||
@item Does cpuid command supports checking all
|
||||
CPU features that the user might want conditionalise on
|
||||
(64-bit mode, hypervisor,...)
|
||||
@item Do you support hints? How reliable are they?
|
||||
@item Does platform have ACPI? If so do ``acpi'' and ``lsacpi'' modules work?
|
||||
@item Do any of platform-specific operations mentioned in the relevant section of
|
||||
user manual makes sense on your platform?
|
||||
@item Does your platform support PCI? If so is there an appropriate driver for
|
||||
GRUB?
|
||||
@item Do you support badram?
|
||||
@end itemize
|
||||
|
||||
@node Error Handling
|
||||
@chapter Error Handling
|
||||
|
||||
|
|
Loading…
Add table
Add a link
Reference in a new issue