Re: [PATCH 7/7] sparc32,leon: implement genirq CPU affinity

[Daniel Hellstrom <daniel@gaisler.com>]

Sam Ravnborg wrote:

>On Tue, Apr 19, 2011 at 06:07:19PM +0200, Daniel Hellstrom wrote:
>  
>
>>A simple implementation of CPU affinity, the first CPU in
>>the affinity CPU mask always takes the IRQ.
>>
>>Signed-off-by: Daniel Hellstrom <daniel@gaisler.com>
>>---
>> arch/sparc/kernel/leon_kernel.c |   66 +++++++++++++++++++++++++++++++++------
>> 1 files changed, 56 insertions(+), 10 deletions(-)
>>
>>diff --git a/arch/sparc/kernel/leon_kernel.c b/arch/sparc/kernel/leon_kernel.c
>>index 26acc75..e12f4e8 100644
>>--- a/arch/sparc/kernel/leon_kernel.c
>>+++ b/arch/sparc/kernel/leon_kernel.c
>>@@ -100,25 +100,68 @@ static inline unsigned long get_irqmask(unsigned int irq)
>> 	return mask;
>> }
>> 
>>+#ifdef CONFIG_SMP
>>+static int irq_choose_cpu(const struct cpumask *affinity)
>>+{
>>+	cpumask_t mask;
>>+
>>+	cpus_and(mask, cpu_online_map, *affinity);
>>+	if (cpus_equal(mask, cpu_online_map) || cpus_empty(mask))
>>+		return leon3_boot_cpu;
>>+	else
>>+		return first_cpu(mask);
>>+}
>>+
>>+int leon_set_affinity(struct irq_data *data, const struct cpumask *dest,
>>+		      bool force)
>>+{
>>+	unsigned long mask, oldmask, flags;
>>+	int oldcpu, newcpu;
>>+
>>+	mask = (unsigned long)data->chip_data;
>>+	oldcpu = irq_choose_cpu(data->affinity);
>>+	newcpu = irq_choose_cpu(dest);
>>+
>>+	if (oldcpu = newcpu)
>>+		goto out;
>>+
>>+	/* unmask on old CPU first before enabling on the selected CPU */
>>+	spin_lock_irqsave(&leon_irq_lock, flags);
>>+	oldmask = LEON3_BYPASS_LOAD_PA(LEON_IMASK(oldcpu));
>>+	LEON3_BYPASS_STORE_PA(LEON_IMASK(oldcpu), (oldmask & ~mask));
>>+	oldmask = LEON3_BYPASS_LOAD_PA(LEON_IMASK(newcpu));
>>+	LEON3_BYPASS_STORE_PA(LEON_IMASK(newcpu), (oldmask | mask));
>>+	spin_unlock_irqrestore(&leon_irq_lock, flags);
>>+out:
>>+	return IRQ_SET_MASK_OK;
>>+}
>>+#else
>>+#define irq_choose_cpu(affinity) leon3_boot_cpu
>>+#endif
>>    
>>
>
>We could always define the leon_set_affinity() function
>to avoid ifdeffery in irq_chip definition.
>  
>
Yes, that is what sparc64 does.

>The expense is that we define this function in both the
>UP and SMP case - it is NOT called by the generic irq
>layer in the UP case.
>  
>
Yes, it would increase the footprint to remove one ifdef. I could add a 
"#define leon_set_affinity NULL" in the already present ifdef above to 
avoid the extra ifdef in chip declaration?

Thanks,
Daniel