[IA64] Fix race when multiple cpus go through MCA

Additional testing uncovered a situation where the MCA recovery code could
hang due to a race condition.

According to the SAL spec, SAL sends a rendezvous interrupt to all but the first
CPU that goes into MCA.  This includes other CPUs that go into MCA at the same
time.  Those other CPUs will go into the linux MCA handler (rather than the
slave loop) with the rendezvous interrupt pending.  When all the CPUs have
completed MCA processing and the last monarch completes, freeing all the CPUs,
the CPUs with the pended rendezvous interrupt then go into the
ia64_mca_rendez_int_handler().  In ia64_mca_rendez_int_handler() the CPUs
get marked as rendezvoused, but then leave the handler (due to no MCA).
That leaves the CPUs marked as rendezvoused _before_ the next MCA event.

When the next MCA hits, the monarch will mistakenly believe that all the CPUs
are rendezvoused when they are not, opening up a window where a CPU can get
stuck in the slave loop.

This patch avoids leaving CPUs marked as rendezvoused when they are not.

Signed-off-by: Russ Anderson <rja@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>
This commit is contained in:
Russ Anderson 2007-09-19 16:58:31 -05:00 committed by Tony Luck
parent 2bc5c28299
commit e1b1eb011e
1 changed files with 25 additions and 22 deletions

View File

@ -701,8 +701,7 @@ ia64_mca_cmc_vector_enable_keventd(struct work_struct *unused)
/* /*
* ia64_mca_wakeup * ia64_mca_wakeup
* *
* Send an inter-cpu interrupt to wake-up a particular cpu * Send an inter-cpu interrupt to wake-up a particular cpu.
* and mark that cpu to be out of rendez.
* *
* Inputs : cpuid * Inputs : cpuid
* Outputs : None * Outputs : None
@ -711,14 +710,12 @@ static void
ia64_mca_wakeup(int cpu) ia64_mca_wakeup(int cpu)
{ {
platform_send_ipi(cpu, IA64_MCA_WAKEUP_VECTOR, IA64_IPI_DM_INT, 0); platform_send_ipi(cpu, IA64_MCA_WAKEUP_VECTOR, IA64_IPI_DM_INT, 0);
ia64_mc_info.imi_rendez_checkin[cpu] = IA64_MCA_RENDEZ_CHECKIN_NOTDONE;
} }
/* /*
* ia64_mca_wakeup_all * ia64_mca_wakeup_all
* *
* Wakeup all the cpus which have rendez'ed previously. * Wakeup all the slave cpus which have rendez'ed previously.
* *
* Inputs : None * Inputs : None
* Outputs : None * Outputs : None
@ -741,7 +738,10 @@ ia64_mca_wakeup_all(void)
* *
* This is handler used to put slave processors into spinloop * This is handler used to put slave processors into spinloop
* while the monarch processor does the mca handling and later * while the monarch processor does the mca handling and later
* wake each slave up once the monarch is done. * wake each slave up once the monarch is done. The state
* IA64_MCA_RENDEZ_CHECKIN_DONE indicates the cpu is rendez'ed
* in SAL. The state IA64_MCA_RENDEZ_CHECKIN_NOTDONE indicates
* the cpu has come out of OS rendezvous.
* *
* Inputs : None * Inputs : None
* Outputs : None * Outputs : None
@ -778,6 +778,7 @@ ia64_mca_rendez_int_handler(int rendez_irq, void *arg)
(long)&nd, 0, 0) == NOTIFY_STOP) (long)&nd, 0, 0) == NOTIFY_STOP)
ia64_mca_spin(__FUNCTION__); ia64_mca_spin(__FUNCTION__);
ia64_mc_info.imi_rendez_checkin[cpu] = IA64_MCA_RENDEZ_CHECKIN_NOTDONE;
/* Enable all interrupts */ /* Enable all interrupts */
local_irq_restore(flags); local_irq_restore(flags);
return IRQ_HANDLED; return IRQ_HANDLED;
@ -1221,26 +1222,27 @@ ia64_mca_handler(struct pt_regs *regs, struct switch_stack *sw,
if (notify_die(DIE_MCA_MONARCH_ENTER, "MCA", regs, (long)&nd, 0, 0) if (notify_die(DIE_MCA_MONARCH_ENTER, "MCA", regs, (long)&nd, 0, 0)
== NOTIFY_STOP) == NOTIFY_STOP)
ia64_mca_spin(__FUNCTION__); ia64_mca_spin(__FUNCTION__);
ia64_mc_info.imi_rendez_checkin[cpu] = IA64_MCA_RENDEZ_CHECKIN_CONCURRENT_MCA;
if (sos->monarch) { if (sos->monarch) {
ia64_wait_for_slaves(cpu, "MCA"); ia64_wait_for_slaves(cpu, "MCA");
/* Wakeup all the processors which are spinning in the
* rendezvous loop. They will leave SAL, then spin in the OS
* with interrupts disabled until this monarch cpu leaves the
* MCA handler. That gets control back to the OS so we can
* backtrace the other cpus, backtrace when spinning in SAL
* does not work.
*/
ia64_mca_wakeup_all();
if (notify_die(DIE_MCA_MONARCH_PROCESS, "MCA", regs, (long)&nd, 0, 0)
== NOTIFY_STOP)
ia64_mca_spin(__FUNCTION__);
} else { } else {
ia64_mc_info.imi_rendez_checkin[cpu] = IA64_MCA_RENDEZ_CHECKIN_CONCURRENT_MCA;
while (cpu_isset(cpu, mca_cpu)) while (cpu_isset(cpu, mca_cpu))
cpu_relax(); /* spin until monarch wakes us */ cpu_relax(); /* spin until monarch wakes us */
ia64_mc_info.imi_rendez_checkin[cpu] = IA64_MCA_RENDEZ_CHECKIN_NOTDONE;
} }
/* Wakeup all the processors which are spinning in the rendezvous loop.
* They will leave SAL, then spin in the OS with interrupts disabled
* until this monarch cpu leaves the MCA handler. That gets control
* back to the OS so we can backtrace the other cpus, backtrace when
* spinning in SAL does not work.
*/
ia64_mca_wakeup_all();
if (notify_die(DIE_MCA_MONARCH_PROCESS, "MCA", regs, (long)&nd, 0, 0)
== NOTIFY_STOP)
ia64_mca_spin(__FUNCTION__);
/* Get the MCA error record and log it */ /* Get the MCA error record and log it */
ia64_mca_log_sal_error_record(SAL_INFO_TYPE_MCA); ia64_mca_log_sal_error_record(SAL_INFO_TYPE_MCA);
@ -1274,21 +1276,22 @@ ia64_mca_handler(struct pt_regs *regs, struct switch_stack *sw,
/* wake up the next monarch cpu, /* wake up the next monarch cpu,
* and put this cpu in the rendez loop. * and put this cpu in the rendez loop.
*/ */
ia64_mc_info.imi_rendez_checkin[cpu] = IA64_MCA_RENDEZ_CHECKIN_CONCURRENT_MCA;
for_each_online_cpu(i) { for_each_online_cpu(i) {
if (cpu_isset(i, mca_cpu)) { if (cpu_isset(i, mca_cpu)) {
monarch_cpu = i; monarch_cpu = i;
cpu_clear(i, mca_cpu); /* wake next cpu */ cpu_clear(i, mca_cpu); /* wake next cpu */
while (monarch_cpu != -1) while (monarch_cpu != -1)
cpu_relax(); /* spin until last cpu leaves */ cpu_relax(); /* spin until last cpu leaves */
ia64_mc_info.imi_rendez_checkin[cpu] = IA64_MCA_RENDEZ_CHECKIN_NOTDONE;
set_curr_task(cpu, previous_current); set_curr_task(cpu, previous_current);
ia64_mc_info.imi_rendez_checkin[cpu]
= IA64_MCA_RENDEZ_CHECKIN_NOTDONE;
return; return;
} }
} }
} }
set_curr_task(cpu, previous_current); set_curr_task(cpu, previous_current);
monarch_cpu = -1; ia64_mc_info.imi_rendez_checkin[cpu] = IA64_MCA_RENDEZ_CHECKIN_NOTDONE;
monarch_cpu = -1; /* This frees the slaves and previous monarchs */
} }
static DECLARE_WORK(cmc_disable_work, ia64_mca_cmc_vector_disable_keventd); static DECLARE_WORK(cmc_disable_work, ia64_mca_cmc_vector_disable_keventd);