Re: [PATCH v7 0/2] Improve VM CPUfreq and task placement behavior

[Rafael J. Wysocki]

On Mon, Oct 28, 2024 at 1:33 PM Sudeep Holla <sudeep.holla@xxxxxxx> wrote:
>
> On Mon, Oct 28, 2024 at 12:39:31PM +0100, Rafael J. Wysocki wrote:
> > On Sat, Oct 26, 2024 at 12:26 AM Saravana Kannan <saravanak@xxxxxxxxxx> wrote:
> > >
> > > On Tue, Oct 1, 2024 at 2:25 AM Viresh Kumar <viresh.kumar@xxxxxxxxxx> wrote:
> > > >
> > > > On 18-09-24, 17:08, David Dai wrote:
> > > > > Hi,
> > > > >
> > > > > This patch series is a continuation of the talk Saravana gave at LPC 2022
> > > > > titled "CPUfreq/sched and VM guest workload problems" [1][2][3]. The gist
> > > > > of the talk is that workloads running in a guest VM get terrible task
> > > > > placement and CPUfreq behavior when compared to running the same workload
> > > > > in the host. Effectively, no EAS(Energy Aware Scheduling) for threads
> > > > > inside VMs. This would make power and performance terrible just by running
> > > > > the workload in a VM even if we assume there is zero virtualization
> > > > > overhead.
> > > >
> > > > > David Dai (2):
> > > > >   dt-bindings: cpufreq: add virtual cpufreq device
> > > > >   cpufreq: add virtual-cpufreq driver
> > > > >
> > > > >  .../cpufreq/qemu,virtual-cpufreq.yaml         |  48 +++
> > > > >  drivers/cpufreq/Kconfig                       |  14 +
> > > > >  drivers/cpufreq/Makefile                      |   1 +
> > > > >  drivers/cpufreq/virtual-cpufreq.c             | 333 ++++++++++++++++++
> > > > >  include/linux/arch_topology.h                 |   1 +
> > > > >  5 files changed, 397 insertions(+)
> > > > >  create mode 100644 Documentation/devicetree/bindings/cpufreq/qemu,virtual-cpufreq.yaml
> > > > >  create mode 100644 drivers/cpufreq/virtual-cpufreq.c
> > > >
> > > > LGTM.
> > > >
> > > > Acked-by: Viresh Kumar <viresh.kumar@xxxxxxxxxx>
> > >
> > > Rafael/Viresh,
> > >
> > > Nudge... Any chance this will get pulled into 6.12?
> >
> > This is not a fix AFAICS, so 6.12 is out of the question.
> >
> > As for 6.13, Viresh thinks that this change is a good idea (or he
> > wouldn't have ACKed it), so it's up to him.  I'm still not convinced
> > that it will work on x86 or anything that doesn't use DT.
> >
>
> +1, I was about to comment on DT bindings patch, but then I assumed it is
> accepted to have a device object with similar CID and CRS(for register address)
> in ACPI for example.

Well, where would the device ID be defined for this?  The spec or
somewhere else?  If the latter, then where again?

> But yes, the patch itself is not adding support for that
> yet. If not is not the way, then we need to come up with a way that works
> for both ACPI and DT.

The DT use case is there I think and so I don't want to block it just
because there is no ACPI counterpart.  It can be added later if the
use case is relevant enough.