powerpc/mce: Fix a bug where mce loops on memory UE.
authorMahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Mon, 23 Apr 2018 04:59:27 +0000 (10:29 +0530)
committerMichael Ellerman <mpe@ellerman.id.au>
Tue, 24 Apr 2018 03:54:51 +0000 (13:54 +1000)
The current code extracts the physical address for UE errors and then
hooks it up into memory failure infrastructure. On successful
extraction of physical address it wrongly sets "handled = 1" which
means this UE error has been recovered. Since MCE handler gets return
value as handled = 1, it assumes that error has been recovered and
goes back to same NIP. This causes MCE interrupt again and again in a
loop leading to hard lockup.

Also, initialize phys_addr to ULONG_MAX so that we don't end up
queuing undesired page to hwpoison.

Without this patch we see:
  Severe Machine check interrupt [Recovered]
    NIP: [000000001002588c] PID: 7109 Comm: find
    Initiator: CPU
    Error type: UE [Load/Store]
      Effective address: 00007fffd2755940
      Physical address:  000020181a080000
  ...
  Severe Machine check interrupt [Recovered]
    NIP: [000000001002588c] PID: 7109 Comm: find
    Initiator: CPU
    Error type: UE [Load/Store]
      Effective address: 00007fffd2755940
      Physical address:  000020181a080000
  Severe Machine check interrupt [Recovered]
    NIP: [000000001002588c] PID: 7109 Comm: find
    Initiator: CPU
    Error type: UE [Load/Store]
      Effective address: 00007fffd2755940
      Physical address:  000020181a080000
  Memory failure: 0x20181a08: recovery action for dirty LRU page: Recovered
  Memory failure: 0x20181a08: already hardware poisoned
  Memory failure: 0x20181a08: already hardware poisoned
  Memory failure: 0x20181a08: already hardware poisoned
  Memory failure: 0x20181a08: already hardware poisoned
  Memory failure: 0x20181a08: already hardware poisoned
  Memory failure: 0x20181a08: already hardware poisoned
  ...
  Watchdog CPU:38 Hard LOCKUP

After this patch we see:

  Severe Machine check interrupt [Not recovered]
    NIP: [00007fffaae585f4] PID: 7168 Comm: find
    Initiator: CPU
    Error type: UE [Load/Store]
      Effective address: 00007fffaafe28ac
      Physical address:  00002017c0bd0000
  find[7168]: unhandled signal 7 at 00007fffaae585f4 nip 00007fffaae585f4 lr 00007fffaae585e0 code 4
  Memory failure: 0x2017c0bd: recovery action for dirty LRU page: Recovered

Fixes: 01eaac2b0591 ("powerpc/mce: Hookup ierror (instruction) UE errors")
Fixes: ba41e1e1ccb9 ("powerpc/mce: Hookup derror (load/store) UE errors")
Cc: stable@vger.kernel.org # v4.15+
Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Balbir Singh <bsingharora@gmail.com>
Reviewed-by: Balbir Singh <bsingharora@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
arch/powerpc/kernel/mce_power.c

index fe6fc63251fec70e7cf2dee7f6deab47fe8ab256..38c5b4764bfed0e0aeb647418f53875bb0f7b2a0 100644 (file)
@@ -441,7 +441,6 @@ static int mce_handle_ierror(struct pt_regs *regs,
                                        if (pfn != ULONG_MAX) {
                                                *phys_addr =
                                                        (pfn << PAGE_SHIFT);
-                                               handled = 1;
                                        }
                                }
                        }
@@ -532,9 +531,7 @@ static int mce_handle_derror(struct pt_regs *regs,
                         * kernel/exception-64s.h
                         */
                        if (get_paca()->in_mce < MAX_MCE_DEPTH)
-                               if (!mce_find_instr_ea_and_pfn(regs, addr,
-                                                               phys_addr))
-                                       handled = 1;
+                               mce_find_instr_ea_and_pfn(regs, addr, phys_addr);
                }
                found = 1;
        }
@@ -572,7 +569,7 @@ static long mce_handle_error(struct pt_regs *regs,
                const struct mce_ierror_table itable[])
 {
        struct mce_error_info mce_err = { 0 };
-       uint64_t addr, phys_addr;
+       uint64_t addr, phys_addr = ULONG_MAX;
        uint64_t srr1 = regs->msr;
        long handled;