Fail to boot 5.15 on mpc8347 with either debug_pagealloc or nobats

[Maxime Bizon]

On Saturday 04 Dec 2021 à 17:42:44 (+0000), Christophe Leroy wrote:

> I guess all the guard is in the comment ...
> 
> /*
>   * Set up one of the I/D BAT (block address translation) register pairs.
>   * The parameters are not checked; in particular size must be a power
>   * of 2 between 128k and 256M.
>   */

It's missing the aligment rule you just taught me, but it's arguably
not the right place to teach ppc 101.

> Not sure it is that simple.
> 
> I'm cooking a patch reusing the block_size() function in mm/book3s32/mmu.c

Indeed it will handle cases that need more than 2 BATs.

Also when mem=2G, I have physical memory mapped twice:

CONFIG_PAGE_OFFSET=0x80000000
CONFIG_LOWMEM_SIZE=0x60000000

0: 0x80000000-0x8fffffff 0x00000000       256M Kernel rw      m   
1: 0x90000000-0x9fffffff 0x10000000       256M Kernel rw      m   
2: 0xa0000000-0xafffffff 0x20000000       256M Kernel rw      m   
3: 0xb0000000-0xbfffffff 0x30000000       256M Kernel rw      m   
4: 0xc0000000-0xcfffffff 0x40000000       256M Kernel rw      m   
5: 0xd0000000-0xdfffffff 0x50000000       256M Kernel rw      m   
6: 0xf0000000-0xf7ffffff 0x50000000       128M Kernel rw      m   

BAT5 comes from __mmu_mapin_ram(), BAT6 from kasan init

Is BAT5 needed here ?


Last one, with KASAN and the following layout, I have an non working
kernel with VMALLOC_START > VMALLOC_END:

mem=2G
CONFIG_PAGE_OFFSET=0x80000000
CONFIG_LOWMEM_SIZE=0x70000000

[    0.000000]   * 0xf0000000..0xfe000000  : kasan shadow mem
[    0.000000]   * 0xef7ff000..0xeffff000  : fixmap
[    0.000000]   * 0xf1000000..0xef7ff000  : vmalloc & ioremap


IIUC the safeguard is here:

arch/powerpc/mm/init_32.c:
/* The amount of lowmem must be within 0xF0000000 - KERNELBASE. */
#if (CONFIG_LOWMEM_SIZE > (0xF0000000 - PAGE_OFFSET))
#error "You must adjust CONFIG_LOWMEM_SIZE or CONFIG_KERNEL_START"


but the definition needs to be adapted for KASAN=y and require 256
more MB.


-- 
Maxime