Good day all,
For the first time I begin working on PPC, and on low level, and right start
from boot sequence, one issue puzzled me.
After bootstrap code (zImage) uncompressed the kernel vmLinux to physical
memory (say from addr 0),
it jumps to the kernel entry point, _start, using physically address.
At this time, the MMU is not yet setup to map the kernel virtual address
(which is statically linked against base address KERNELBASE) to the
physically address.
$ nm vmlinux |grep early_init
c038b8e0 T early_init
_start calls early_init before mmu is on to map the KERNEL_BASE to
physically address
The question is how "bl early_init" can branch to the early_init entry
point, properly, as early_init is still a virtual address?
Thanks
Jim
--
View this message in context: http://www.nabble.com/boottime-kernel-relocation%2C-what-I-missed--tf4072673.html#a11574529
Sent from the linuxppc-embedded mailing list archive at Nabble.com.
Figure that out, the bootstrap actually mapped the first 16M from C000000 to
the physicall address,
so calling a c routine, as long as it is in the first 16M, is OK
meerkat wrote:
Good day all,
For the first time I begin working on PPC, and on low level, and right
start from boot sequence, one issue puzzled me.
After bootstrap code (zImage) uncompressed the kernel vmLinux to physical
memory (say from addr 0),
it jumps to the kernel entry point, _start, using physically address.
At this time, the MMU is not yet setup to map the kernel virtual address
(which is statically linked against base address KERNELBASE) to the
physically address.
$ nm vmlinux |grep early_init
c038b8e0 T early_init
_start calls early_init before mmu is on to map the KERNEL_BASE to
physically address
The question is how "bl early_init" can branch to the early_init entry
point, properly, as early_init is still a virtual address?
Thanks
Jim
From: Becky Bruce <hidden> Date: 2007-07-16 19:04:41
On Jul 13, 2007, at 5:07 PM, meerkat wrote:
Figure that out, the bootstrap actually mapped the first 16M from
C000000 to
the physicall address,
so calling a c routine, as long as it is in the first 16M, is OK
I think you're still not understanding the fact that "bl" is a
*relative* branch - the branch target in the instruction encoding is
just an offset from the current address, not an effective address.
The bl should work correctly whether the code is actually running at
the link address reported by nm (0xcxxxxxxx in this case), or if it
has been loaded and executed elsewhere.
Refer to the 32-bit Programming Environments Manual for PowerPC, or
in the EREF (if you're using a BookE part - e500/e200) for more
details. If you're just learning PowerPC assembler, you should
really give this book a good thourough read.
meerkat wrote:
quoted
Good day all,
For the first time I begin working on PPC, and on low level, and
right
start from boot sequence, one issue puzzled me.
After bootstrap code (zImage) uncompressed the kernel vmLinux to
physical
memory (say from addr 0),
it jumps to the kernel entry point, _start, using physically address.
If you read the book specified above, you will see that branch
instructions always specify an effective address, not a physical
address. You can disable translation or map the address so EA=PA,
but that's a different issue.
-Becky
Becky:
Thanks for the reply. Agree I should read bookE, which given the short
schedule, it is still on my TODO list yet.
So are you saying linker does all the relative reference, that the bl foo
is always relative even though the foo is defined in a different module? (In
my case, bl call is made in head.S, early_init() is defined in another file
setup.c).
Regards,
Jim
Becky Bruce wrote:
On Jul 13, 2007, at 5:07 PM, meerkat wrote:
quoted
Figure that out, the bootstrap actually mapped the first 16M from
C000000 to
the physicall address,
so calling a c routine, as long as it is in the first 16M, is OK
I think you're still not understanding the fact that "bl" is a
*relative* branch - the branch target in the instruction encoding is
just an offset from the current address, not an effective address.
The bl should work correctly whether the code is actually running at
the link address reported by nm (0xcxxxxxxx in this case), or if it
has been loaded and executed elsewhere.
Refer to the 32-bit Programming Environments Manual for PowerPC, or
in the EREF (if you're using a BookE part - e500/e200) for more
details. If you're just learning PowerPC assembler, you should
really give this book a good thourough read.
quoted
meerkat wrote:
quoted
Good day all,
For the first time I begin working on PPC, and on low level, and
right
start from boot sequence, one issue puzzled me.
After bootstrap code (zImage) uncompressed the kernel vmLinux to
physical
memory (say from addr 0),
it jumps to the kernel entry point, _start, using physically address.
If you read the book specified above, you will see that branch
instructions always specify an effective address, not a physical
address. You can disable translation or map the address so EA=PA,
but that's a different issue.
-Becky
_______________________________________________
Linuxppc-embedded mailing list
Linuxppc-embedded@ozlabs.org
https://ozlabs.org/mailman/listinfo/linuxppc-embedded
From: Becky Bruce <hidden> Date: 2007-07-16 20:45:34
On Jul 16, 2007, at 2:36 PM, meerkat wrote:
So are you saying linker does all the relative reference, that the
bl foo
is always relative even though the foo is defined in a different
module? (In
my case, bl call is made in head.S, early_init() is defined in
another file
setup.c).
If you see "bl foo", it's relative. It has to be, because that's how
the instruction works. Also, it's not really in a different module,
it's just a different file, but it's all linked into a single
executable image.
Cheers,
-Becky
Regards,
Jim
Becky Bruce wrote:
quoted
On Jul 13, 2007, at 5:07 PM, meerkat wrote:
quoted
Figure that out, the bootstrap actually mapped the first 16M from
C000000 to
the physicall address,
so calling a c routine, as long as it is in the first 16M, is OK
I think you're still not understanding the fact that "bl" is a
*relative* branch - the branch target in the instruction encoding is
just an offset from the current address, not an effective address.
The bl should work correctly whether the code is actually running at
the link address reported by nm (0xcxxxxxxx in this case), or if it
has been loaded and executed elsewhere.
Refer to the 32-bit Programming Environments Manual for PowerPC, or
in the EREF (if you're using a BookE part - e500/e200) for more
details. If you're just learning PowerPC assembler, you should
really give this book a good thourough read.
quoted
meerkat wrote:
quoted
Good day all,
For the first time I begin working on PPC, and on low level, and
right
start from boot sequence, one issue puzzled me.
After bootstrap code (zImage) uncompressed the kernel vmLinux to
physical
memory (say from addr 0),
it jumps to the kernel entry point, _start, using physically
address.
If you read the book specified above, you will see that branch
instructions always specify an effective address, not a physical
address. You can disable translation or map the address so EA=PA,
but that's a different issue.
-Becky
_______________________________________________
Linuxppc-embedded mailing list
Linuxppc-embedded@ozlabs.org
https://ozlabs.org/mailman/listinfo/linuxppc-embedded