Re: [PATCH] x86: Add check for number of available vectors before CPU down [v2]

[Prarit Bhargava <prarit@redhat.com>]

On 12/30/2013 07:56 AM, rui wang wrote:

> An irq can be mapped to only one vector number, but can have multiple
> destination CPUs. i.e. the same irq/vector can appear on multiple
> CPUs' vector_irq[]. So checking data->affinity is necessary I think.

That's true Rui -- but here's what I think the scenario actually is.

Suppose we have a 4-cpu system, and we have an IRQ that is mapped to multiple
cpu's vector_irq[].  For example, we have IRQ 200 that is mapped to CPU 2
vector_irq[50], and CPU 3 vector_irq[60].

Now I 'echo 0 > /sys/devices/system/cpu/cpu3/online'.

cpu_disable is called and the kernel migrates IRQs off to other cpus.
Regardless if IRQ 200 is already mapped to CPU2 vector_irq[50], the mapping for
CPU 3 vector_irq[60] *must be migrated* to another CPU.  It has a valid irq
handler and the IRQ is active.  It doesn't just disappear because the CPU went down.

ie) AFAICT we should not differentiate between a multiple mapped IRQ and a
singly mapped IRQ when traversing the vector_irq[] for CPU 3.

I'm probably being dense on this but I'm not seeing a problem with migrating the
IRQ.

> But notice that data->affinity is updated in chip->irq_set_affinity()
> inside fixup_irqs(), while cpu_online_mask is updated in
> remove_cpu_from_maps() inside cpu_disable_common(). 

It shouldn't matter that the maps are updated in different areas during the
execution as we're in stop_machine().

They are updated
> in different places. So the algorithm to check them against each other
> should be different, depending on where you put the check_vectors().
> That's my understanding.
> 

P.

> Thanks
> Rui