x86, mce: fix a race condition about mce_callin and no_way_out
authorHuang Ying <ying.huang@intel.com>
Mon, 15 Jun 2009 07:37:07 +0000 (15:37 +0800)
committerH. Peter Anvin <hpa@zytor.com>
Tue, 16 Jun 2009 23:56:04 +0000 (16:56 -0700)
If one CPU has no_way_out == 1, all other CPUs should have no_way_out
== 1. But despite global_nwo is read after mce_callin, global_nwo is
updated after mce_callin too. So it is possible that some CPU read
global_nwo before some other CPU update global_nwo, so that no_way_out
== 1 for some CPU, while no_way_out == 0 for some other CPU.

This patch fixes this race condition via moving mce_callin updating
after global_nwo updating, with a smp_wmb in between. A smp_rmb is
added between their reading too.

Signed-off-by: Huang Ying <ying.huang@intel.com>
Acked-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
arch/x86/kernel/cpu/mcheck/mce.c

index fabba15e455864c6768eb958ac61fba3e2acf8a8..19294b8524cb281301271c7271465c14c5b07753 100644 (file)
@@ -703,6 +703,11 @@ static int mce_start(int no_way_out, int *order)
        }
 
        atomic_add(no_way_out, &global_nwo);
+       /*
+        * global_nwo should be updated before mce_callin
+        */
+       smp_wmb();
+       *order = atomic_add_return(1, &mce_callin);
 
        /*
         * Wait for everyone.
@@ -716,6 +721,10 @@ static int mce_start(int no_way_out, int *order)
                ndelay(SPINUNIT);
        }
 
+       /*
+        * mce_callin should be read before global_nwo
+        */
+       smp_rmb();
        /*
         * Cache the global no_way_out state.
         */
@@ -862,7 +871,7 @@ void do_machine_check(struct pt_regs *regs, long error_code)
         * Establish sequential order between the CPUs entering the machine
         * check handler.
         */
-       int order;
+       int order = -1;
 
        /*
         * If no_way_out gets set, there is no safe way to recover from this
@@ -887,7 +896,6 @@ void do_machine_check(struct pt_regs *regs, long error_code)
        if (!banks)
                goto out;
 
-       order = atomic_add_return(1, &mce_callin);
        mce_setup(&m);
 
        m.mcgstatus = mce_rdmsrl(MSR_IA32_MCG_STATUS);