]> git.ipfire.org Git - thirdparty/kernel/stable.git/commitdiff
MIPS: loongsoon64: Reserve memory below starting pfn to prevent Oops
authorzhanglianjie <zhanglianjie@uniontech.com>
Mon, 28 Jun 2021 09:11:05 +0000 (17:11 +0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Mon, 19 Jul 2021 08:01:12 +0000 (10:01 +0200)
[ Upstream commit 6817c944430d00f71ccaa9c99ff5b0096aeb7873 ]

The cause of the problem is as follows:
1. when cat /sys/devices/system/memory/memory0/valid_zones,
   test_pages_in_a_zone() will be called.
2. test_pages_in_a_zone() finds the zone according to stat_pfn = 0.
   The smallest pfn of the numa node in the mips architecture is 128,
   and the page corresponding to the previous 0~127 pfn is not
   initialized (page->flags is 0xFFFFFFFF)
3. The nid and zonenum obtained using page_zone(pfn_to_page(0)) are out
   of bounds in the corresponding array,
   &NODE_DATA(page_to_nid(page))->node_zones[page_zonenum(page)],
   access to the out-of-bounds zone member variables appear abnormal,
   resulting in Oops.
Therefore, it is necessary to keep the page between 0 and the minimum
pfn to prevent Oops from appearing.

Signed-off-by: zhanglianjie <zhanglianjie@uniontech.com>
Signed-off-by: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
arch/mips/loongson64/numa.c

index a8f57bf012854c055e99da6eaa3255bf7456c702..a791863ed352f6e9bda8d9c3c4e7065b48fec3c7 100644 (file)
@@ -118,6 +118,9 @@ static void __init node_mem_init(unsigned int node)
                if (node_end_pfn(0) >= (0xffffffff >> PAGE_SHIFT))
                        memblock_reserve((node_addrspace_offset | 0xfe000000),
                                         32 << 20);
+
+               /* Reserve pfn range 0~node[0]->node_start_pfn */
+               memblock_reserve(0, PAGE_SIZE * start_pfn);
        }
 }