svn commit: r191405 - in head/sys: amd64/amd64 i386/i386

[Ivan Voras]

2009/4/22 John Baldwin <jhb at freebsd.org>:
> On Wednesday 22 April 2009 5:40:37 pm John Baldwin wrote:
>> Author: jhb
>> Date: Wed Apr 22 21:40:37 2009
>> New Revision: 191405
>> URL: http://svn.freebsd.org/changeset/base/191405
>>
>> Log:
>>   Adjust the way we number CPUs on x86 so that we attempt to "group" all
>>   logical CPUs in a package.  We do this by numbering the non-boot CPUs
>>   by starting with the first CPU whose APIC ID is after the boot CPU and
>>   wrapping back around to APIC ID 0 if needed rather than always starting
>>   at APIC ID 0.  While here, adjust the cpu_mp_announce() routine to list
>>   CPUs based on the mapping established by assign_cpu_ids() rather than
>>   making assumptions about the algorithm assign_cpu_ids() uses.
>
> An example is probably in order for this to make sense.  Suppose you have a
> system with two quad-core CPUs.  Package 0 has CPUs numbered 0, 1, 2, and 3.
> Package 1 has CPUs numbered 4, 5, 6, and 7.  With the old code, if package 0
> won the election to be the boot processor, then CPU 0 would be the BSP and
> the logical IDs would match the APIC IDs.  However, if package 1 won the
> election during POST, then CPU 0 would be APIC ID 4 on package 0 followed by
> CPU 1 being APIC ID 0, CPU 2 being APIC ID 1, etc.  Thus, when CPU 0 was the
> boot CPU you had a nice grouping where CPUs 0-3 were a single package and
> CPUs 4-7 were another package.  However, when CPU 4 was the boot CPU, CPUs 0
> and 5-7 where one package, and CPUs 1-4 where the second package.  The effect
> of this patch is to change the case when CPU 4 is the boot CPU such that CPUs
> 0-3 are now all from CPU 4's package (APIC IDs 4-7), and CPUs 4-7 are from
> the other package (APIC IDs 0-3).  What this means, in turn, is that in both
> cases you now always have CPUs 0-3 as one package and CPUs 4-7 as another
> package regardless of which CPU wins the boot-time election.

I like that the new numbering is more elegant, but this is orthogonal
to ULE topology detection, right?