Hi,
The screen shot in Attilio's bug report was done by me on my Indy. I
am not experienced in this beginning of the universe kind of code. But
I found the following clues which may be useful.
Please refer to bug 452798
(http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=452798). Here is
message #15 from Guido:
On Sun, Nov 25, 2007 at 01:02:25PM +0100, Martin Michlmayr wrote:
How much work is 64 bit support?
It's not the 64bit support, things are simply not fitting into the
memory segment we use anymore:
Tip22 finds this memory map on my Indy:
1) Free Memory(3) segment found at (0x88002000,0x8874)
2) Free Memory(3) segment found at (0x8880,0x89002000)
3) Free Memory(3) segment found at (0x896f2000,0x9400)
(1) is where the running kernel + initrd goes. (2) is where we tell the
prom to download the tftpboot image to and (3) is lots of free space.
The 64bit kernel (7MB) is just so much bigger than the 32bit one.
Attached is a hack to get things booting again (so installer testing can
proceed). It puts the tftp boot file into (3) and the initrd into (2) -
it succesfully boots a kernel with an initramfs on my R5K Indy. I'll
have to change some of the reservation code so we don't waste so much
ram before I can fix this for real though.
Cheers,
-- Guido
Here is the objdump -x output of the netboot-boot.img from Lenny:
debbootL.img: file format ecoff-bigmips
debbootL.img
architecture: mips:3000, flags 0x0032:
EXEC_P, HAS_SYMS, HAS_LOCALS
start address 0x897020cc
Sections:
Idx Name Size VMA LMA File off
Algn
0 .text 1e00 89702000 89702000 00d0 2**4
CONTENTS, ALLOC, LOAD, CODE
1 .data 00ad1500 89703e00 89703e00 1ed0 2**4
CONTENTS, ALLOC, LOAD, DATA
2 .bss 0040 8a1d5300 8a1d5300 2**4
ALLOC
SYMBOL TABLE:
[ 0] e 897038b8 st 1 sc 1 indx f strcpy
[ 1] e 89703e00 st 1 sc 2 indx f kernel_load
[ 2] e 89702a30 st 1 sc 1 indx f ArcGetPeer
[ 3] e 897033a8 st 1 sc 1 indx f printf
[ 4] e 89703e34 st 1 sc 2 indx f stdout
[ 5] e 89702de4 st 1 sc 1 indx f vsprintf
[ 6] e 89703e00 st 1 sc 2 indx f _fdata
And here is the objdump -x output of the netboot-boot.img from
Squeeze:
debbootS.img: file format ecoff-bigmips
debbootS.img
architecture: mips:3000, flags 0x0032:
EXEC_P, HAS_SYMS, HAS_LOCALS
start address 0x888020cc
Sections:
Idx Name Size VMA LMA File off
Algn
0 .text 1ea0 88802000 88802000 00d0 2**4
CONTENTS, ALLOC, LOAD, CODE
1 .data 00982140 88803ea0 88803ea0 1f70 2**4
CONTENTS, ALLOC, LOAD, DATA
2 .bss 0040 89185fe0 89185fe0 2**4
ALLOC
SYMBOL TABLE:
[ 0] e 88803914 st 1 sc 1 indx f strcpy
[ 1] e 88803ea0 st 1 sc 2 indx f kernel_load
[ 2] e 88802a30 st 1 sc 1 indx f ArcGetPeer
[ 3] e 888033a8 st 1 sc 1 indx f printf
[ 4] e 88803ed4 st 1 sc 2 indx f stdout
[ 5] e 88802de4 st 1 sc 1 indx f vsprintf
[ 6] e 88803ea0 st 1 sc 2 indx f _fdata
It seems to me that what Guido wanted was done in Lenny but not in
Squeeze. In the screen shot, initrd was being copied into 0x8880
by the Squeeze image. Wouldn't it overwrite the loader?
I downloaded the Squeeze tip22 package, and here is the beginning of
the ld.script.IP22 script:
OUTPUT_FORMAT(ecoff-bigmips)
OUTPUT_ARCH(mips)
ENTRY(_start)
SECTIONS
{
/* XXX: place the loader after the kernel */
. = 0x88802000;
/* read-only */
_text = .;/* Text and read-only data */
.text : {
_ftext = .;
*(.text .text.*)
*(.rodata*)
} =0
_etext = .; /* End of text section */
PROVIDE (etext = .);
Thanks for your help!
Edwin
--
To UNSUBSCRIBE, email to debian-bugs-rc-requ...@lists.debian.org
with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org
