[POWERPC] Fix PMU + soft interrupt disable bug
authorAnton Blanchard <anton@samba.org>
Mon, 17 Mar 2008 04:27:09 +0000 (15:27 +1100)
committerPaul Mackerras <paulus@samba.org>
Wed, 19 Mar 2008 23:14:55 +0000 (10:14 +1100)
Since the PMU is an NMI now, it can come at any time we are only soft
disabled.  We must hard disable around the two places we allow the kernel
stack SLB and r1 to go out of sync.  Otherwise the PMU exception can
force a kernel stack SLB into another slot, which can lead to it
getting evicted, which can lead to a nasty unrecoverable SLB miss
in the exception entry code.

Signed-off-by: Anton Blanchard <anton@samba.org>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: Olof Johansson <olof@lixom.net>
Signed-off-by: Paul Mackerras <paulus@samba.org>
arch/powerpc/kernel/process.c
arch/powerpc/mm/slb.c

index 4846bf543a8c5cf9dd2f95a611c4deae04cd9863..59311ec0d422d3eca2f94b358808bde287765fbc 100644 (file)
@@ -353,6 +353,12 @@ struct task_struct *__switch_to(struct task_struct *prev,
        account_process_vtime(current);
        calculate_steal_time();
 
+       /*
+        * We can't take a PMU exception inside _switch() since there is a
+        * window where the kernel stack SLB and the kernel stack are out
+        * of sync. Hard disable here.
+        */
+       hard_irq_disable();
        last = _switch(old_thread, new_thread);
 
        local_irq_restore(flags);
index 47b06bad24adbb8e4c857f782df2fdc30ac779fb..906daeda59a8ff431d6bee026af51f55e4318512 100644 (file)
@@ -124,6 +124,12 @@ void slb_flush_and_rebolt(void)
                ksp_vsid_data = get_slb_shadow()->save_area[2].vsid;
        }
 
+       /*
+        * We can't take a PMU exception in the following code, so hard
+        * disable interrupts.
+        */
+       hard_irq_disable();
+
        /* We need to do this all in asm, so we're sure we don't touch
         * the stack between the slbia and rebolting it. */
        asm volatile("isync\n"