When running several kvm processes with lots of memory overcommitment,
we have seen an oops during process shutdown:
------------[ cut here ]------------
Kernel BUG at 0000000000193434 [verbose debug info unavailable]
addressing exception: 0005 [#1] PREEMPT SMP
Modules linked in: kvm sunrpc qeth_l2 dm_mod qeth ccwgroup
CPU: 10 Not tainted 2.6.28-rc4-kvm-bigiron-00521-g0ccca08-dirty #8
Process kuli (pid: 14460, task: 0000000149822338, ksp: 0000000024f57650)
Krnl PSW : 0704e00180000000 0000000000193434 (unmap_vmas+0x884/0xf10)
R:0 T:1 IO:1 EX:1 Key:0 M:1 W:0 P:0 AS:3 CC:2 PM:0 EA:3
Krnl GPRS: 0000000000000002 0000000000000000 000000051008d000 000003e05e6034e0
00000000001933f6 00000000000001e9 0000000407259e0a 00000002be88c400
00000200001c1000 0000000407259608 0000000407259e08 0000000024f577f0
0000000407259e09 0000000000445fa8 00000000001933f6 0000000024f577f0
Krnl Code: 0000000000193426: eb22000c000d sllg %r2,%r2,12
000000000019342c: a7180000 lhi %r1,0
0000000000193430: b2290012 iske %r1,%r2
>0000000000193434: a7110002 tmll %r1,2
0000000000193438: a7840006 brc 8,193444
000000000019343c: 9602c000 oi 0(%r12),2
0000000000193440: 96806000 oi 0(%r6),128
0000000000193444: a7110004 tmll %r1,4
Call Trace:
([<00000000001933f6>] unmap_vmas+0x846/0xf10)
[<0000000000199680>] exit_mmap+0x210/0x458
[<000000000012a8f8>] mmput+0x54/0xfc
[<000000000012f714>] exit_mm+0x134/0x144
[<000000000013120c>] do_exit+0x240/0x878
[<00000000001318dc>] do_group_exit+0x98/0xc8
[<000000000013e6b0>] get_signal_to_deliver+0x30c/0x358
[<000000000010bee0>] do_signal+0xec/0x860
[<0000000000112e30>] sysc_sigpending+0xe/0x22
[<000002000013198a>] 0x2000013198a
INFO: lockdep is turned off.
Last Breaking-Event-Address:
[<00000000001a68d0>] free_swap_and_cache+0x1a0/0x1a4
<4>---[ end trace bc19f1d51ac9db7c ]---
The faulting instruction is the storage key operation (iske) in
ptep_rcp_copy (called by pte_clear, called by unmap_vmas). iske
reads dirty and reference bit information for a physical page and
requires a valid physical address. Since we are in pte_clear, we
cannot rely on the pte containing a valid address. Fortunately we
dont need these information in pte_clear - after all there is no
mapping. The best fix is to remove the needless call to ptep_rcp_copy
that contains the iske.
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
The current enable_sie code sets the mm->context.pgstes bit to tell
dup_mm that the new mm should have extended page tables. This bit is also
used by the s390 specific page table primitives to decide about the page
table layout - which means context.pgstes has two meanings. This can cause
any kind of bugs. For example - e.g. shrink_zone can call
ptep_clear_flush_young while enable_sie is running. ptep_clear_flush_young
will test for context.pgstes. Since enable_sie changed that value of the old
struct mm without changing the page table layout ptep_clear_flush_young will
do the wrong thing.
The solution is to split pgstes into two bits
- one for the allocation
- one for the current state
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
This patch defines a dirty bit in the PGSTE that can be used to implement
dirty pages logging for KVM's live migration. The bit is set in the
ptep_rcp_copy function, which is called to save dirty and referenced information
from the storage key in the PGSTE. The bit can be tested and reset by KVM using
the kvm_s390_test_and_clear_page_dirty function that is introduced by this patch.
Acked-by: Carsten Otte <cotte@de.ibm.com>
Signed-off-by: Florian Funke <ffunke@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
