Re: STRICT_KERNEL_RWX on PPC32 is broken on PowerMac G4

From: Christophe LEROY
Date: Tue Nov 21 2017 - 04:58:40 EST

Le 21/11/2017 Ã 03:28, Balbir Singh a ÃcritÂ:
On Sun, Nov 19, 2017 at 1:36 AM, LEROY Christophe
<christophe.leroy@xxxxxx> wrote:
Meelis Roos <mroos@xxxxxxxx> a Ãcrit :

How early does it hang ? Any oops or trace ?

Very early - instead oif kernel emssages, I see some repeated gibberish
of some characteers, and the background turns white.
I am booting from yaboot, background is normally black.

Ok, could you try by replacing #ifdef CONFIG_STRICT_KERNEL_RWX by #if 0
in arch/powerpc/lib/code-patching.c

With this change and CONFIG_STRICT_KERNEL_RWX=y, it still boots.

BTW, I get these warnings (sorry for the word wrap from screen paste) -
may they be related or rather not?

WRAP arch/powerpc/boot/zImage.pmac
WRAP arch/powerpc/boot/zImage.coff
WRAP arch/powerpc/boot/zImage.miboot
INFO: Uncompressed kernel (size 0x5d4c3c) overlaps the address of the
INFO: Fixing the link_address of wrapper to (0x600000)

Then i believe there is something wrong with commit

Balbir, do you have any idea ?

Hmm.. interesting, so nobats works, but code-patching has issues? Any
chance you can boot with xmon=on on the command line
when you drop into the xmon> prompt, type dl to get the kernel log.

I have the same issue with a 83xx board allthought it is booting well on my 8xx boards.

Indeed the issue is that patch_instruction() is called from early_init(), therefor all static data must be addressed using PTRRELOC()

I have submitted a fixup patch, can you review it ?
slab_is_available() cannot be called that early for the same reason but I believe it is not needed.


Balbir Singh.