| Message ID | 1438027789-1637-1-git-send-email-eric@anholt.net (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Mon, 27 Jul 2015, Eric Anholt wrote: > +/* Unmasks the IPI on the CPU wen it's first brought online. */ when > +static int bcm2836_arm_irqchip_cpu_notify(struct notifier_block *nfb, > + unsigned long action, void *hcpu) > +{ > + unsigned int cpu = (unsigned long)hcpu; > + unsigned int int_reg = LOCAL_MAILBOX_INT_CONTROL0; > + unsigned int mailbox = 0; > + > + if (action == CPU_STARTING || action == CPU_STARTING_FROZEN) > + bcm2836_arm_irqchip_unmask_per_cpu_irq(int_reg, mailbox, cpu); Shouldn't you mask the irq on CPU_DYING? Thanks, tglx
Thomas Gleixner <tglx@linutronix.de> writes: > On Mon, 27 Jul 2015, Eric Anholt wrote: >> +/* Unmasks the IPI on the CPU wen it's first brought online. */ > > when > >> +static int bcm2836_arm_irqchip_cpu_notify(struct notifier_block *nfb, >> + unsigned long action, void *hcpu) >> +{ >> + unsigned int cpu = (unsigned long)hcpu; >> + unsigned int int_reg = LOCAL_MAILBOX_INT_CONTROL0; >> + unsigned int mailbox = 0; >> + >> + if (action == CPU_STARTING || action == CPU_STARTING_FROZEN) >> + bcm2836_arm_irqchip_unmask_per_cpu_irq(int_reg, mailbox, cpu); > > Shouldn't you mask the irq on CPU_DYING? I was just following what other drivers were doing. Is CPU_DYING the only thing that needs masking?
On Mon, 3 Aug 2015, Eric Anholt wrote: > Thomas Gleixner <tglx@linutronix.de> writes: > > > On Mon, 27 Jul 2015, Eric Anholt wrote: > >> +/* Unmasks the IPI on the CPU wen it's first brought online. */ > > > > when > > > >> +static int bcm2836_arm_irqchip_cpu_notify(struct notifier_block *nfb, > >> + unsigned long action, void *hcpu) > >> +{ > >> + unsigned int cpu = (unsigned long)hcpu; > >> + unsigned int int_reg = LOCAL_MAILBOX_INT_CONTROL0; > >> + unsigned int mailbox = 0; > >> + > >> + if (action == CPU_STARTING || action == CPU_STARTING_FROZEN) > >> + bcm2836_arm_irqchip_unmask_per_cpu_irq(int_reg, mailbox, cpu); > > > > Shouldn't you mask the irq on CPU_DYING? > > I was just following what other drivers were doing. Is CPU_DYING the > only thing that needs masking? CPPU_DYING is the counterpart of CPU_STARTING. It's called from the CPU which goes down. Thanks, tglx
diff --git a/drivers/irqchip/irq-bcm2836.c b/drivers/irqchip/irq-bcm2836.c index 87340b0..0bb0552 100644 --- a/drivers/irqchip/irq-bcm2836.c +++ b/drivers/irqchip/irq-bcm2836.c @@ -14,6 +14,7 @@ * GNU General Public License for more details. */ +#include <linux/cpu.h> #include <linux/of_address.h> #include <linux/of_irq.h> #include <linux/irqchip.h> @@ -195,6 +196,25 @@ static void bcm2836_arm_irqchip_send_ipi(const struct cpumask *mask, writel(1 << ipi, mailbox0_base + 16 * cpu); } } + +/* Unmasks the IPI on the CPU wen it's first brought online. */ +static int bcm2836_arm_irqchip_cpu_notify(struct notifier_block *nfb, + unsigned long action, void *hcpu) +{ + unsigned int cpu = (unsigned long)hcpu; + unsigned int int_reg = LOCAL_MAILBOX_INT_CONTROL0; + unsigned int mailbox = 0; + + if (action == CPU_STARTING || action == CPU_STARTING_FROZEN) + bcm2836_arm_irqchip_unmask_per_cpu_irq(int_reg, mailbox, cpu); + + return NOTIFY_OK; +} + +static struct notifier_block bcm2836_arm_irqchip_cpu_notifier = { + .notifier_call = bcm2836_arm_irqchip_cpu_notify, + .priority = 100, +}; #endif static const struct irq_domain_ops bcm2836_arm_irqchip_intc_ops = { @@ -205,13 +225,12 @@ static void bcm2836_arm_irqchip_smp_init(void) { #ifdef CONFIG_SMP - int i; + /* Unmask IPIs to the boot CPU. */ + bcm2836_arm_irqchip_cpu_notify(&bcm2836_arm_irqchip_cpu_notifier, + CPU_STARTING, + (void *)smp_processor_id()); + register_cpu_notifier(&bcm2836_arm_irqchip_cpu_notifier); - /* unmask IPIs */ - for_each_possible_cpu(i) { - bcm2836_arm_irqchip_unmask_per_cpu_irq( - LOCAL_MAILBOX_INT_CONTROL0, 0, i); - } set_smp_cross_call(bcm2836_arm_irqchip_send_ipi); #endif }
Signed-off-by: Eric Anholt <eric@anholt.net> --- Florian: Thanks! That looks like what we need to squash for Thomas's feedback. It feels a bit silly to me to avoid enabling IPIs on not-yet-booted CPUs, despite the fact that Linux itself is what generates IPIs. However, I've tested it on the 2836 series and it does work. drivers/irqchip/irq-bcm2836.c | 31 +++++++++++++++++++++++++------ 1 file changed, 25 insertions(+), 6 deletions(-)