[U-Boot] Question regarding cpu_get_count(..)

[Christian Gmeiner]

2018-04-09 10:26 GMT+02:00 Bin Meng <bmeng.cn at gmail.com>:
> Hi Christian,
>
> On Mon, Apr 9, 2018 at 3:52 PM, Christian Gmeiner
> <christian.gmeiner at gmail.com> wrote:
>> Hi Bin
>>
>> 2018-04-06 10:30 GMT+02:00 Christian Gmeiner <christian.gmeiner at gmail.com>:
>>> Hi Bin
>>>
>>> 2018-04-06 9:47 GMT+02:00 Bin Meng <bmeng.cn at gmail.com>:
>>>> Hi Christian,
>>>>
>>>> On Fri, Apr 6, 2018 at 2:55 PM, Christian Gmeiner
>>>> <christian.gmeiner at gmail.com> wrote:
>>>>> Hi
>>>>>
>>>>> I tried my luck on the u-boot irc channel but nobody cared so I am
>>>>> asking this here again.
>>>>> Does cpu_get_count(..) return the number of physical CPUs or the
>>>>> number of logical CPUs?
>>>>>
>>>>> I am currently preparing a patch set which adds
>>>>> arch/x86/cpu/queensbay/cpu.c and there I need
>>>>> to specify a struct cpu_ops with a function called get_count(). But
>>>>> what must I return (physical vs. logical CPU count)?
>>>>>
>>>>
>>>> It's logical CPUs, say you have one physical CPU, each with
>>>> dual-threads, so get_count() returns 2.
>>>>
>>>
>>> Good to know. I am looking into an issue where vxworks 7 intel gen is not able
>>> to detect the correct amount of physical and logical CPUs (on a blob UEFI-Bios).
>>> I compared the new platform with an older queensbay based one the there are some
>>> discrepancies. As vx7 makes use of a handful of BIOS tables I thought I start at
>>> the root (get_count()) and move up. So as a result of this that means that the
>>> MP table has one "Processor Entry" per logical CPU - or? I tired to get that
>>> information for the spec but failed.
>>>
>>>> But Intel CrownBay already supports SMP in U-Boot. What patch are you
>>>> working on?
>>>
>>> I am out of office today and do not have access to my wip git rep :(
>>> But I needed
>>> to set some speed frequency registers.. but can tell you more on Monday.
>>>
>>
>> Okay.. yeah I need to configure speed: https://hastebin.com/toqihobali.cpp
>>
>> So the end result is to get all BIOS tables standard conform and vx7 should
>> be happy - I hope.
>>
>
> I am not sure I understand the mods here. Are you saying that vx7 is
> not happy due to U-Boot is missing some core speed register
> configuration?
>

No - I am happy with the code basis as is. In order to fix some
performance issues
regarding cache I ended adding cpu.c. That was some months ago.

Now some guys and me need to bring up vx7 smp on a handful of targets and we are
seeing some wired vxCpuIdTopology() outputs on different targets.
Looks like u-boot
is okay but a vendor binary bios has problems

https://hastebin.com/qojenuqoge.js

To sum it up:
* cpu_get_count(..) needs to return the number of logical CPUs.
* The generated MP-Table has one "Processor Entry" per logical CPU

I will prepare a proper patch for the cpu.c addition during the week.
-- 
greets
--
Christian Gmeiner, MSc

https://christian-gmeiner.info