[5.10] mm: make zone_to_nid() and zone_set_nid() available for DISCONTIGMEM

From: Mike Rapoport <rppt@linux.ibm.com>

Since the commit ce6ee46e0f39 ("mm/page_alloc: fix memory map
initialization for descending nodes") initialization of the memory map
relies on availability of zone_to_nid() and zone_set_nid methods to link
struct page to a node.

But in 5.10 zone_to_nid() is only defined for NUMA, but not for
DISCONTIGMEM which causes crashes on m68k systems with two memory banks.

For instance on ARAnyM with both ST-RAM and FastRAM atari_defconfig build
produces the following crash:

Unable to handle kernel access at virtual address (ptrval)
Oops: 00000000
Modules linked in:
PC: [<0005fbbc>] bpf_prog_alloc_no_stats+0x5c/0xba
SR: 2200  SP: (ptrval)  a2: 016daa90
d0: 0000000c    d1: 00000200    d2: 00000001    d3: 00000cc0
d4: 016d1f80    d5: 00034da6    a0: 305c2800    a1: 305c2a00
Process swapper (pid: 1, task=(ptrval))
Frame format=7 eff addr=31800000 ssw=0445 faddr=31800000
wb 1 stat/addr/data: 0000 00000000 00000000
wb 2 stat/addr/data: 0000 00000000 00000000
wb 3 stat/addr/data: 00c5 31800000 00000001
push data: 00000000 00000000 00000000 00000000
Stack from 3058fec8:
        00000dc0 00000000 004addc2 3058ff16 0005fc34 00000238 00000000 00000210
        004addc2 3058ff16 00281ae0 00000238 00000000 00000000 004addc2 004bc7ec
        004aea9e 0048b0c0 3058ff16 00460042 004ba4d2 3058ff8c 004ade6a 0000007e
        0000210e 0000007e 00000002 016d1f80 00034da6 000020b4 00000000 004b4764
        004bc7ec 00000000 004b4760 004bc7c0 004b4744 001e4cb2 00010001 016d1fe5
        016d1ff0 004994d2 003e1589 016d1f80 00412b8c 0000007e 00000001 00000001
Call Trace: [<004addc2>] sock_init+0x0/0xaa
 [<0005fc34>] bpf_prog_alloc+0x1a/0x66
 [<004addc2>] sock_init+0x0/0xaa
 [<00281ae0>] bpf_prog_create+0x2e/0x7c
 [<004addc2>] sock_init+0x0/0xaa
 [<004aea9e>] ptp_classifier_init+0x22/0x44
 [<004ade6a>] sock_init+0xa8/0xaa
 [<0000210e>] do_one_initcall+0x5a/0x150
 [<00034da6>] parse_args+0x0/0x208
 [<000020b4>] do_one_initcall+0x0/0x150
 [<001e4cb2>] strcpy+0x0/0x1c
 [<00010001>] stwotoxd+0x5/0x1c
 [<004994d2>] kernel_init_freeable+0x154/0x1a6
 [<001e4cb2>] strcpy+0x0/0x1c
 [<0049951a>] kernel_init_freeable+0x19c/0x1a6
 [<004addc2>] sock_init+0x0/0xaa
 [<00321510>] kernel_init+0x0/0xd8
 [<00321518>] kernel_init+0x8/0xd8
 [<00321510>] kernel_init+0x0/0xd8
 [<00002890>] ret_from_kernel_thread+0xc/0x14

Code: 204b 200b 4cdf 180c 4e75 700c e0aa 3682 <2748> 001c 214b 0140 022b
ffbf 0002 206b 001c 2008 0680 0000 0108 2140 0108 2140
Disabling lock debugging due to kernel taint
Kernel panic - not syncing: Attempted to kill init! exitcode=0x0000000b

Using CONFIG_NEED_MULTIPLE_NODES rather than CONFIG_NUMA to guard
definitions of zone_to_nid() and zone_set_nid() fixes the issue.

Reported-by: Mikael Pettersson <mikpelinux@gmail.com>
Fixes: ce6ee46e0f39 ("mm/page_alloc: fix memory map initialization for descending nodes")
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Tested-by: Mikael Pettersson <mikpelinux@gmail.com>
---
 include/linux/mmzone.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

On Wed, Aug 11, 2021 at 04:41:39PM +0300, Mike Rapoport wrote:
> From: Mike Rapoport <rppt@linux.ibm.com>

> 

> Since the commit ce6ee46e0f39 ("mm/page_alloc: fix memory map

> initialization for descending nodes") initialization of the memory map

> relies on availability of zone_to_nid() and zone_set_nid methods to link

> struct page to a node.

> 

> But in 5.10 zone_to_nid() is only defined for NUMA, but not for

> DISCONTIGMEM which causes crashes on m68k systems with two memory banks.

> 


Many thanks for the work done to figure this out and fix it.  Now queued
up.

greg k-h