Created attachment 1975048 [details]
sosreport files

Hi, 

May I know why there is no test option 'profiler_hardware_uncore' in the certification plan, when we add it and test it, we got a FAIL result, but others platform has this testing option, please check and review the logs.

Thank.

Hi，

Does it the same situation with https://bugzilla.redhat.com/show_bug.cgi?id=2181947？

I got the same result:

[root@localhost ~]# lsmod | grep -i uncore
[root@localhost ~]# modprobe intel_uncore
modprobe: ERROR: could not insert 'intel_uncore': No such device
[root@localhost ~]# perf list | grep -P '^\s{1,3}unc' | awk '{print $1}'
[root@localhost ~]# 

[testing@localhost ~]$ uname -a
Linux localhost.localdomain 5.14.0-284.18.1.el9_2.x86_64 #1 SMP PREEMPT_DYNAMIC Wed May 31 10:39:18 EDT 2023 x86_64 x86_64 x86_64 GNU/Linux

If it is the same situation, what we should do for SE10 certification? Pass it or other?

Thanks.

Hi Kean,

I think the uncore events are not supported in the CPU "Intel Atom(R) x6425RE Processor" which is based on Tremont microarchitecture having codename Elkhart(EHL). Also, I did a little digging and found this https://github.com/intel/perfmon/tree/main/EHL/events

Have you tried executing the listed 3 commands on Fedora, just like you did in https://bugzilla.redhat.com/show_bug.cgi?id=2181947?

Thanks,
Akshay

Hi asaswadk,

No, I wasn't exec those commands on Fedora38 due to the device being away from me, we will carry all devices to a new office, so maybe will get testing on next week as quickly as possible.

Thanks. I also believe the Atom does not support the uncore events, so in this case, does the testing option become not necessary testing? And not be a blocker of cert of SE10?

Thanks.

I just take a little dive into the upstream kernel code. 

This is the output of boot_cpu_data structure on "Intel Atom(R) x6425RE Processor":
crash> p boot_cpu_data
boot_cpu_data = $1 = {
  x86 = 6 '\006', 
  x86_vendor = 0 '\000', 
  x86_model = 150 '\226', ====> 0x96
  x86_stepping = 1 '\001', 
  x86_tlbsize = 0,

According to https://elixir.bootlin.com/linux/v6.5-rc1/source/arch/x86/include/asm/intel-family.h, the cpu module 0x96 is INTEL_FAM6_ATOM_TREMONT.

And in arch/x86/events/intel/uncore.c:intel_uncore_init -> x86_match_cpu, which will check cpu module against intel_uncore_match[] table. And there is no INTEL_FAM6_ATOM_TREMONT in the table, only INTEL_FAM6_ATOM_TREMONT_D.

So from the code view, the INTEL_FAM6_ATOM_TREMONT, or the "Intel Atom(R) x6425RE Processor" doesn't support intel uncore, the intel_uncore module will not get initialized successfully.

Thanks,
Tao Liu

(In reply to ltao from comment #6)
> I just take a little dive into the upstream kernel code. 
> 
> This is the output of boot_cpu_data structure on "Intel Atom(R) x6425RE
> Processor":
> crash> p boot_cpu_data
> boot_cpu_data = $1 = {
>   x86 = 6 '\006', 
>   x86_vendor = 0 '\000', 
>   x86_model = 150 '\226', ====> 0x96
>   x86_stepping = 1 '\001', 
>   x86_tlbsize = 0,
> 
> According to
> https://elixir.bootlin.com/linux/v6.5-rc1/source/arch/x86/include/asm/intel-
> family.h, the cpu module 0x96 is INTEL_FAM6_ATOM_TREMONT.
> 
> And in arch/x86/events/intel/uncore.c:intel_uncore_init -> x86_match_cpu,
> which will check cpu module against intel_uncore_match[] table. And there is
> no INTEL_FAM6_ATOM_TREMONT in the table, only INTEL_FAM6_ATOM_TREMONT_D.
> 
> So from the code view, the INTEL_FAM6_ATOM_TREMONT, or the "Intel Atom(R)
> x6425RE Processor" doesn't support intel uncore, the intel_uncore module
> will not get initialized successfully.
> 
> Thanks,
> Tao Liu

s/cpu module/cpu model/g

Hi Tao,

Thank you for the information & confirming the "Intel Atom(R) x6425RE Processor" doesn't support uncore.


Hi @renhai2,

In this case, where the CPU doesn't support uncore, the uncore test won't be planned and is not required to be tested, i.e. not a blocker.

Thanks,
Akshay

Thanks @ltao, @asaswadk. 
Let close it.
Kean