Changing kernel uncompressing address

Changing kernel uncompressing address

[Alexander Simon]

Hi!

I have an old Toshiba Satellie Pro Laptop with broken RAM.
I thought it could be no problem getting it to work with the BadRAM patch.
But first RAM Errors occur at 2M and last until 32M
Unfornately, the RAM Chips are on board, so no chance of replacing them.
When I try to load a kernel image from diskette, it unpacks the kernel 
image without errors. But when it tries to start that kernel it stops or 
reboots.
If I keep the kernel very very small, it starts, but I would have to 
exclude TCP/IP code, causing the system unusable.

After studying arch/i386/boot/compressed/head.S and misc.c of 2.4.24 for 
a long time, i found out that the kernel is uncompressed to 0x100000.
Stupidly, I'm not familiar with assembler code. So I just changed the 
0x100000 to 0xF00000 (should be 16M?!? memtest86 reported the range 
15M-18M OK, however...) in line 77 in head.S and line 309 in misc.c.
Of couse it did NOT work :[.
I would need to high loaded kernel anyway, again because of TCP/IP.

Thanks for any ideas

Re: Changing kernel uncompressing address

[Matt Mackall]

On Sun, Mar 21, 2004 at 11:52:07AM +0100, Alexander Simon wrote:
> Hi!
> 
> I have an old Toshiba Satellie Pro Laptop with broken RAM.
> I thought it could be no problem getting it to work with the BadRAM patch.
> But first RAM Errors occur at 2M and last until 32M
> Unfornately, the RAM Chips are on board, so no chance of replacing them.
> When I try to load a kernel image from diskette, it unpacks the kernel 
> image without errors. But when it tries to start that kernel it stops or 
> reboots.
> If I keep the kernel very very small, it starts, but I would have to 
> exclude TCP/IP code, causing the system unusable.
> 
> After studying arch/i386/boot/compressed/head.S and misc.c of 2.4.24 for 
> a long time, i found out that the kernel is uncompressed to 0x100000.
> Stupidly, I'm not familiar with assembler code. So I just changed the 
> 0x100000 to 0xF00000 (should be 16M?!? memtest86 reported the range 
> 15M-18M OK, however...) in line 77 in head.S and line 309 in misc.c.
> Of couse it did NOT work :[.
> I would need to high loaded kernel anyway, again because of TCP/IP.

That's part of it, you'll also need to tweak the boot-time page tables
and whatnot to cover all of the space you need.

-- 
Matt Mackall : http://www.selenic.com : Linux development and consulting