[S390] pgtable.h: Fix oops in unmap_vmas for KVM processes
authorChristian Borntraeger <borntraeger@de.ibm.com>
Thu, 27 Nov 2008 10:05:57 +0000 (11:05 +0100)
committerMartin Schwidefsky <schwidefsky@de.ibm.com>
Thu, 27 Nov 2008 10:06:57 +0000 (11:06 +0100)
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: 0000000000193426eb22000c000d sllg %r2,%r2,12
           000000000019342ca7180000 lhi %r1,0
           0000000000193430b2290012 iske %r1,%r2
          >0000000000193434a7110002 tmll %r1,2
           0000000000193438a7840006 brc 8,193444
           000000000019343c9602c000 oi 0(%r12),2
           000000000019344096806000 oi 0(%r6),128
           0000000000193444a7110004 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>
arch/s390/include/asm/pgtable.h

index 7fc76133b3e46da10eddb1ca115b7ac69a4a80c0..5caddd4f7bedd74ed7fa9078c8cf4d97bdfc2227 100644 (file)
@@ -679,8 +679,6 @@ static inline void pmd_clear(pmd_t *pmd)
 
 static inline void pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
 {
-       if (mm->context.has_pgste)
-               ptep_rcp_copy(ptep);
        pte_val(*ptep) = _PAGE_TYPE_EMPTY;
        if (mm->context.noexec)
                pte_val(ptep[PTRS_PER_PTE]) = _PAGE_TYPE_EMPTY;