Skip to contentSkip to conversation

ia64-cpu-hotplug-cpu_present.patch?

DISCUSSION Paul Jackson 2 participants 3 messages Apr 28, 2004

lore
Share & tools
lore
#0 lore raw

Andrew, Ashok,

How did the following patch, named ia64-cpu-hotplug-cpu_present.patch, dated 25 April 2004, end up in Andrew's patch set?

I cannot find any mention of it on lkml or any other email list that either Google or I track.

Is there some backdoor path to Andrew's patch directory I don't know of ;)?

The patch comment was:

This patch adds cpu_present_map, cpu_present() and for_each_cpu_present() to distinguish between possible cpu's in a system and cpu's physically present in a system. Before cpu hotplug was introduced cpu_possible() represented cpu's physically present in the system. With hotplug capable Kernel, there is a requirement to distinguish a cpu as possible verses a CPU physically present in the system. This is required so thta when smp_init() attempts to start all cpu's it should now only attempt to start cpu's present in the system. When a hotplug cpu is physically inserted cpu_present_map will have bits updated dynamically.

This patch is cutting across the bitmap+cpumask rework that I'm working on. I'm going to try to refit it to my new code, but may well break something.

I'll try to remember to include Ashok explicitly on the next version of my patch set, so that he can review my changes, and catch what I break.

Signature · 3 lines
                          I won't rest till it's the best ...
                          Programmer, Linux Scalability
                          Paul Jackson <[email protected]> 1.650.933.1373
#1 lore raw

On Tue, Apr 27, 2004 at 10:57:41PM -0700, Paul Jackson wrote:

Hi Paul,

sorry for the trouble. I will try to address them, and will try to resolve the compile problem you are having.

Quoted 4 lines

I cannot find any mention of it on lkml or any other email list that either Google or I track.

Is there some backdoor path to Andrew's patch directory I don't know of ;)?

I sent the patch to linux-ia64, but forgot to copy lkml. No... i did not buy crispy cream for andrew.. although that did cross my mine :-)

Some of the points you raise, i will look into, but here is a quick summary fo Paul's top 10 list!.

cpu_present is required so that when cpu_up is called in smp_init() we dont call for all possible cpu's but just for cpu's physically present.

cpu_possible - indicates cpu may be available, but may not be present in the system. Hence different from phys_cpu_present_mask

cpu_possible wont change dynamically but present_map can change depending on the arrival of cpu's in a system. I implemented it for only ia64, and expection is that ACPI or its equilvalent would populate cpu_present_map dynamically.

In architectures that dont have cpu_hotplug capability, i have kept cpu_present and possible the same.

Once i take a look at the failures, i will send you an updated patch. again really sorry for the inconvenience.

Cheers,
ashok

#2 lore raw
Quoted 1 line

sorry for the trouble. I will try to address them

Ok - that's no problem. Thanks for looking into it.

Signature · 3 lines
                          I won't rest till it's the best ...
                          Programmer, Linux Scalability
                          Paul Jackson <[email protected]> 1.650.933.1373