sparc64: Make sure %pil interrupts are enabled during hypervisor yield.
authorDavid S. Miller <davem@davemloft.net>
Mon, 24 Mar 2014 18:45:12 +0000 (14:45 -0400)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sun, 27 Apr 2014 00:15:36 +0000 (17:15 -0700)
[ Upstream commit cb3042d609e30e6144024801c89be3925106752b ]

In arch_cpu_idle() we must enable %pil based interrupts before
potentially invoking the hypervisor cpu yield call.

As per the Hypervisor API documentation for cpu_yield:

Interrupts which are blocked by some mechanism other that
pstate.ie (for example %pil) are not guaranteed to cause
a return from this service.

It seems that only first generation Niagara chips are hit by this
bug.  My best guess is that later chips implement this in hardware
and wake up anyways from %pil events, whereas in first generation
chips the yield is implemented completely in hypervisor code and
requires %pil to be enabled in order to wake properly from this
call.

Fixes: 87fa05aeb3a5 ("sparc: Use generic idle loop")
Reported-by: Fabio M. Di Nitto <fabbione@fabbione.net>
Reported-by: Jan Engelhardt <jengelh@inai.de>
Tested-by: Jan Engelhardt <jengelh@inai.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
arch/sparc/kernel/process_64.c

index baebab215492966718fc8846e9f251ae3d20af82..b9cc9763faf4aa9ef7570867d788c191f9494ec7 100644 (file)
@@ -57,9 +57,12 @@ void arch_cpu_idle(void)
 {
        if (tlb_type != hypervisor) {
                touch_nmi_watchdog();
+               local_irq_enable();
        } else {
                unsigned long pstate;
 
+               local_irq_enable();
+
                 /* The sun4v sleeping code requires that we have PSTATE.IE cleared over
                  * the cpu sleep hypervisor call.
                  */
@@ -81,7 +84,6 @@ void arch_cpu_idle(void)
                        : "=&r" (pstate)
                        : "i" (PSTATE_IE));
        }
-       local_irq_enable();
 }
 
 #ifdef CONFIG_HOTPLUG_CPU