On 18.12.19 04:21, Dan Williams wrote:
> On Tue, Dec 17, 2019 at 2:47 AM David Hildenbrand <david@xxxxxxxxxx> wrote:
>>
>> When we remove an early section, we don't free the usage map, as the
>> usage maps of other sections are placed into the same page. Once the
>> section is removed, it is no longer an early section (especially, the
>> memmap is freed). When we re-add that section, the usage map is reused,
>> however, it is no longer an early section. When removing that section
>> again, we try to kfree() a usage map that was allocated during early
>> boot - bad.
>>
>> Let's check against PageReserved() to see if we are dealing with an usage
>> map that was allocated during boot. We could also check against
>> !(PageSlab(usage_page) || PageCompound(usage_page)), but PageReserved()
>> is cleaner.
>>
>> Can be triggered using memtrace under ppc64/powernv:
>>
>> $ mount -t debugfs none /sys/kernel/debug/
>> $ echo 0x20000000 > /sys/kernel/debug/powerpc/memtrace/enable
>> $ echo 0x20000000 > /sys/kernel/debug/powerpc/memtrace/enable
>> [ 12.093442] ------------[ cut here ]------------
>> [ 12.093469] kernel BUG at mm/slub.c:3969!
>> [ 12.093656] Oops: Exception in kernel mode, sig: 5 [#1]
>> [ 12.093961] LE PAGE_SIZE=64K MMU=Hash SMP NR_CPUS=2048 NUMA PowerNV
>> [ 12.094320] Modules linked in:
>> [ 12.094615] CPU: 0 PID: 154 Comm: sh Not tainted 5.5.0-rc2-next-20191216-00005-g0be1dba7b7c0 #61
>> [ 12.095066] NIP: c000000000396b38 LR: c000000000385848 CTR: c000000000143d30
>> [ 12.095427] REGS: c000000073077680 TRAP: 0700 Not tainted (5.5.0-rc2-next-20191216-00005-g0be1dba7b7c0)
>> [ 12.095886] MSR: 900000000282b033 <SF,HV,VEC,VSX,EE,FP,ME,IR,DR,RI,LE> CR: 28004828 XER: 20000000
>> [ 12.096395] CFAR: c000000000396b9c IRQMASK: 0
>> [ 12.096395] GPR00: c000000000385848 c000000073077910 c00000000110f300 c00000007ffffc00
>> [ 12.096395] GPR04: 0000000000000000 ffffffffffffffff 0000000000000000 0000000000000000
>> [ 12.096395] GPR08: 0000000000000000 0000000000000001 0000000000000000 ffffffffffffffc8
>> [ 12.096395] GPR12: 0000000000004000 c0000000012d0000 0000000000001000 c000000000d33c78
>> [ 12.096395] GPR16: 0000000000000000 c0000000011bfeb0 ffffffffffffe000 c0000000000b6370
>> [ 12.096395] GPR20: ffffffffe0000000 c0000000011411c0 0000000000006000 c0000000000b6390
>> [ 12.096395] GPR24: 0000000010000000 0000000000000040 0000000000000000 0000000000000000
>> [ 12.096395] GPR28: c000000000385848 c00c0000001fffc0 0000000000004000 0000000000000000
>> [ 12.099882] NIP [c000000000396b38] kfree+0x338/0x3b0
>> [ 12.100135] LR [c000000000385848] section_deactivate+0x138/0x200
>> [ 12.100508] Call Trace:
>> [ 12.100927] [c000000073077910] [c0000000010599a8] 0xc0000000010599a8 (unreliable)
>> [ 12.101338] [c000000073077960] [c000000000385848] section_deactivate+0x138/0x200
>> [ 12.101696] [c000000073077a10] [c00000000039b9f4] __remove_pages+0x114/0x150
>> [ 12.102025] [c000000073077a60] [c00000000006793c] arch_remove_memory+0x3c/0x160
>> [ 12.102381] [c000000073077ae0] [c00000000039c154] try_remove_memory+0x114/0x1a0
>> [ 12.102715] [c000000073077b90] [c00000000039c020] __remove_memory+0x20/0x40
>> [ 12.103041] [c000000073077bb0] [c0000000000b6714] memtrace_enable_set+0x254/0x850
>> [ 12.103402] [c000000073077cb0] [c0000000004197e8] simple_attr_write+0x138/0x160
>> [ 12.103751] [c000000073077d10] [c000000000558c9c] full_proxy_write+0x8c/0x110
>> [ 12.104100] [c000000073077d60] [c0000000003d02a8] __vfs_write+0x38/0x70
>> [ 12.104409] [c000000073077d80] [c0000000003d3c5c] vfs_write+0x11c/0x2a0
>> [ 12.104711] [c000000073077dd0] [c0000000003d4054] ksys_write+0x84/0x140
>> [ 12.105011] [c000000073077e20] [c00000000000b594] system_call+0x5c/0x68
>> [ 12.105357] Instruction dump:
>> [ 12.105606] e93d0000 75290001 41820090 8bfd0051 38a0ffff 7ca5f830 7bff0020 7ca507b4
>> [ 12.105993] e95d0000 39200000 754a0001 4182005c <0b090000> 893d0007 3d42000b 38800006
>> [ 12.106583] ---[ end trace 4b053cbd84e0db62 ]---
>>
>> The first invocation will offline+remove memory blocks. The second
>> invocation will first add+online them again, in order to offline+remove
>> them again (usually we are lucky and the exact same memory blocks will
>> get "reallocated").
>>
>> Tested on powernv with boot memory: The usage map will not get freed.
>> Tested on x86-64 with DIMMs: The usage map will get freed.
>>
>> Fixes: 326e1b8f83a4 ("mm/sparsemem: introduce a SECTION_IS_EARLY flag")
>> Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
>> Cc: Dan Williams <dan.j.williams@xxxxxxxxx>
>> Cc: Oscar Salvador <osalvador@xxxxxxx>
>> Cc: Michal Hocko <mhocko@xxxxxxxxxx>
>> Signed-off-by: David Hildenbrand <david@xxxxxxxxxx>
>> ---
>> mm/sparse.c | 9 ++++++++-
>> 1 file changed, 8 insertions(+), 1 deletion(-)
>>
>> diff --git a/mm/sparse.c b/mm/sparse.c
>> index b20ab7cdac86..3822ecbd8a1f 100644
>> --- a/mm/sparse.c
>> +++ b/mm/sparse.c
>> @@ -777,7 +777,14 @@ static void section_deactivate(unsigned long pfn, unsigned long nr_pages,
>> if (bitmap_empty(subsection_map, SUBSECTIONS_PER_SECTION)) {
>> unsigned long section_nr = pfn_to_section_nr(pfn);
>>
>> - if (!section_is_early) {
>
> With this change this function no longer needs a local
> section_is_early variable...
Right. I'll send a cleanup patch and we can decide if we want to squash
them in Andrews tree.
>
>> + /*
>> + * When removing an early section, the usage map is kept (as the
>> + * usage maps of other sections fall into the same page). It
>> + * will be re-used when re-adding the section - which is then no
>> + * longer an early section. If the usage map is PageReserved, it
>> + * was allocated during boot.
>> + */
>> + if (!PageReserved(virt_to_page(ms->usage))) {
>
> This looks good. However, if this was wrong then I wonder why you are
> not seeing problems with free_map_bootmem()?
AFAIKS, in contrast to the usage map, the memmap will get freed. So when
you readd memory, you actually allocate a new one and mark the section
!early.
I was thinking about replacing the section flag by a simple
PageReserved(virt_to_page(memmap)) check. Might be a little bit slower.
Maybe something for the new year :)
Thanks!
--
Thanks,
David / dhildenb
