[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [Xen-changelog] [xen-3.0-testing] [LINUX] Fix IRQ SMP affinity logic for event channels.
# HG changeset patch # User kaf24@xxxxxxxxxxxxxxxxxxxx # Node ID df221e310e2b4f162167147aa3d7656af3969417 # Parent d4312b52c3f88ff4dc4d872f9f0f1c7b556be0e2 [LINUX] Fix IRQ SMP affinity logic for event channels. The logic now mimics native x86 behaviour: a request to change affinity via /proc is held until the next interrupt on that event channel. So /proc/irq/n/smp_affinity may not change immediately! Other notes: 1. CPU-specific interrupts silently ignore requests to change affinity. For example, resched0, timer0, callfunc0, ... 2. Reading smp_affinity always returns a cpumask containing a single cpu. An event channel can only be bound to a single cpu at a time. Neither Xen nor XenLinux implement IRQ balancing: requires a user-space balancing daemon. Signed-off-by: Keir Fraser <keir@xxxxxxxxxxxxx> xen-unstable changeset: 10317:be05097d5d69925a72405201140df8da4c1cfa5c xen-unstable date: Sun Jun 11 09:54:35 2006 +0100 --- linux-2.6-xen-sparse/drivers/xen/core/evtchn.c | 44 +++++++------------------ 1 files changed, 14 insertions(+), 30 deletions(-) diff -r d4312b52c3f8 -r df221e310e2b linux-2.6-xen-sparse/drivers/xen/core/evtchn.c --- a/linux-2.6-xen-sparse/drivers/xen/core/evtchn.c Fri Jun 09 17:06:09 2006 +0100 +++ b/linux-2.6-xen-sparse/drivers/xen/core/evtchn.c Sun Jun 11 10:10:59 2006 +0100 @@ -120,6 +120,11 @@ static inline unsigned long active_evtch static void bind_evtchn_to_cpu(unsigned int chn, unsigned int cpu) { + int irq = evtchn_to_irq[chn]; + + BUG_ON(irq == -1); + set_native_irq_info(irq, cpumask_of_cpu(cpu)); + clear_bit(chn, (unsigned long *)cpu_evtchn_mask[cpu_evtchn[chn]]); set_bit(chn, (unsigned long *)cpu_evtchn_mask[cpu]); cpu_evtchn[chn] = cpu; @@ -127,7 +132,12 @@ static void bind_evtchn_to_cpu(unsigned static void init_evtchn_cpu_bindings(void) { + int i; + /* By default all event channels notify CPU#0. */ + for (i = 0; i < NR_IRQS; i++) + set_native_irq_info(i, cpumask_of_cpu(0)); + memset(cpu_evtchn, 0, sizeof(cpu_evtchn)); memset(cpu_evtchn_mask[0], ~0, sizeof(cpu_evtchn_mask[0])); } @@ -419,25 +429,14 @@ void unbind_from_irqhandler(unsigned int } EXPORT_SYMBOL_GPL(unbind_from_irqhandler); -#ifdef CONFIG_SMP -static void do_nothing_function(void *ign) -{ -} -#endif - /* Rebind an evtchn so that it gets delivered to a specific cpu */ static void rebind_irq_to_cpu(unsigned irq, unsigned tcpu) { evtchn_op_t op = { .cmd = EVTCHNOP_bind_vcpu }; - int evtchn; - - spin_lock(&irq_mapping_update_lock); - - evtchn = evtchn_from_irq(irq); - if (!VALID_EVTCHN(evtchn)) { - spin_unlock(&irq_mapping_update_lock); + int evtchn = evtchn_from_irq(irq); + + if (!VALID_EVTCHN(evtchn)) return; - } /* Send future instances of this interrupt to other vcpu. */ op.u.bind_vcpu.port = evtchn; @@ -450,21 +449,6 @@ static void rebind_irq_to_cpu(unsigned i */ if (HYPERVISOR_event_channel_op(&op) >= 0) bind_evtchn_to_cpu(evtchn, tcpu); - - spin_unlock(&irq_mapping_update_lock); - - /* - * Now send the new target processor a NOP IPI. When this returns, it - * will check for any pending interrupts, and so service any that got - * delivered to the wrong processor by mistake. - * - * XXX: The only time this is called with interrupts disabled is from - * the hotplug/hotunplug path. In that case, all cpus are stopped with - * interrupts disabled, and the missed interrupts will be picked up - * when they start again. This is kind of a hack. - */ - if (!irqs_disabled()) - smp_call_function(do_nothing_function, NULL, 0, 0); } @@ -589,8 +573,8 @@ static unsigned int startup_pirq(unsigne pirq_query_unmask(irq_to_pirq(irq)); + evtchn_to_irq[evtchn] = irq; bind_evtchn_to_cpu(evtchn, 0); - evtchn_to_irq[evtchn] = irq; irq_info[irq] = mk_irq_info(IRQT_PIRQ, irq, evtchn); out: _______________________________________________ Xen-changelog mailing list Xen-changelog@xxxxxxxxxxxxxxxxxxx http://lists.xensource.com/xen-changelog
|
Lists.xenproject.org is hosted with RackSpace, monitoring our |