NACK/Cmnt: [SRU][OEM-6.8][Noble][PATCH V2 0/1] UBUNTU: SAUCE: cpufreq: intel_pstate: Use HWP to initialize ITMT if CPPC is missing

[Stefan Bader]

On 24.06.24 20:03, Aaron Rainbolt wrote:
> BugLink: https://bugs.launchpad.net/bugs/2069988
> 
> V2: Resubmitted using a different email application in the hopes of
> not mangling things this time.
> 
> SRU Justification:
> 
> [Impact]
> 
> The new EEVDF scheduler, introduced in Linux 6.6 and present in the
> Noble 6.8 kernel, requires ITMT in order to determine which cores are
> P-cores and which cores are E-cores on hybrid-core Intel-based systems.
> Currently, the intel_pstate requires CPPC v2 to be present in order to
> enable ITMT. Not all systems that support ITMT also support CPPC v2
> however. On such systems, ITMT will be mistakenly disabled and the
> scheduler will no longer be able to schedule intensive single-core
> workloads on P-cores reliably. This results in a serious drop in
> single-core performance (machines capable of getting Geekbench 5
> single-core scores of over 1800 get scores of less than 1000).
> 
> [Fix]
> 
> The proposed patch (which I have shamelessly "stolen" from the
> linux-acpi mailing list where Rafael J. Wysocki proposed it) fixes this
> issue by using information from MSR_HWP_CAPABILITIES to determine
> whether or not to enable ITMT, in the event CPPC v2 is not present.
> 
> [Test Plan]
> 
> Internally, Kubuntu Focus will be using a thorough kernel testing suite
> on our hardware to ensure the kernel functions correctly in a wide
> variety of different use case scenarios, while also benchmarking the
> systems to ensure that single-core Geekbench 5 scores rise to their
> expected level. For public testing, it should be noted that this patch
> will only provide a performance improvement on hardware without CPPC v2
> and with a hybrid-core Intel processor. If you do not have an affected
> system, the patch should provide no change in performance at all. With
> this in mind, the following plan should be sufficient to ensure the
> kernel is functional and that performance is not negatively impacted:
> 
> * Install `7z`.
> * Using an older kernel, run `7z b -mmt1` and record the results.
> * Repeat the above three times.
> * Install the kernel from proposed.
> * Again, run `7z b -mmt1` and record the results three times.
> * Compare these results and ensure that performance is approximately
>    the same both before and after patching.
> * Run the system with usual day-to-day workloads for a day or two to
>    ensure that it is stable.
> 
> [Where problems could occur]
> 
> It is theoretically possible that kernel code that deals with ITMT could
> behave strangely when ITMT is active without CPPC v2 present. This could
> result in kernel instability, or (if the intel_pstate driver itself
> misbehaves) performance degredation. The above test plan is intended to
> catch either of these issues.
> 

Rejected for the following reasons:
Thanks for submitting again as a thread. And sorry for turning it down 
once more. At least for the time being. For inclusion into the distro 
kernel we stick to changes from upstream. That gets more important for 
LTS releases and from personal experience even more than that when it 
comes to anything related to HW/ACPI drivers.
So minimally this should be included in linux-next. By then it got 
through reviews on the sub-system lists and should be more or less 
final. It would not need the SAUCE markup and if something still is 
wrong (one easily can have good results with one type of HW and it 
horribly goes wrong with other platforms) there will be fixups which 
have a good chance to land via stable updates. So overall I would delay 
until this is linux-next. I just checked this morning but it has not 
landed, yet. Once this happens this fix could be submitted again 
(dropping SAUCE and adding cherry picked/backported reference).

-Stefan
-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_0xE8675DEECBEECEA3.asc
Type: application/pgp-keys
Size: 48643 bytes
Desc: OpenPGP public key
URL: <https://lists.ubuntu.com/archives/kernel-team/attachments/20240625/93ae9f57/attachment-0001.key>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: OpenPGP digital signature
URL: <https://lists.ubuntu.com/archives/kernel-team/attachments/20240625/93ae9f57/attachment-0001.sig>