Re: [PATCH v4 0/3] thermal/netlink/intel_hfi: Enable HFI feature only when required

[Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>]

On Thu, Feb 29, 2024 at 04:18:50PM +0100, Rafael J. Wysocki wrote:
> On Mon, Feb 12, 2024 at 5:16 PM Stanislaw Gruszka
> <stanislaw.gruszka@linux.intel.com> wrote:
> >
> > The patchset introduces a new genetlink family bind/unbind callbacks
> > and thermal/netlink notifications, which allow drivers to send netlink
> > multicast events based on the presence of actual user-space consumers.
> > This functionality optimizes resource usage by allowing disabling
> > of features when not needed.
> >
> > Then implement the notification mechanism in the intel_hif driver,
> > it is utilized to disable the Hardware Feedback Interface (HFI)
> > dynamically. By implementing a thermal genl notify callback, the driver
> > can now enable or disable the HFI based on actual demand, particularly
> > when user-space applications like intel-speed-select or Intel Low Power
> > daemon utilize events related to performance and energy efficiency
> > capabilities.
> >
> > On machines where Intel HFI is present, but there are no user-space
> > components installed, we can save tons of CPU cycles.
> >
> > Changes v3 -> v4:
> >
> > - Add 'static inline' in patch2
> >
> > Changes v2 -> v3:
> >
> > - Fix unused variable compilation warning
> > - Add missed Suggested by tag to patch2
> >
> > Changes v1 -> v2:
> >
> > - Rewrite using netlink_bind/netlink_unbind callbacks.
> >
> > - Minor changelog tweaks.
> >
> > - Add missing check in intel hfi syscore resume (had it on my testing,
> > but somehow missed in post).
> >
> > - Do not use netlink_has_listeners() any longer, use custom counter instead.
> > To keep using netlink_has_listners() would be required to rearrange
> > netlink_setsockopt() and possibly netlink_bind() functions, to call
> > nlk->netlink_bind() after listeners are updated. So I decided to custom
> > counter. This have potential issue as thermal netlink registers before
> > intel_hif, so theoretically intel_hif can miss events. But since both
> > are required to be kernel build-in (if CONFIG_INTEL_HFI_THERMAL is
> > configured), they start before any user-space.
> >
> > v1: https://lore.kernel.org/linux-pm/20240131120535.933424-1-stanislaw.gruszka@linux.intel.com//
> > v2: https://lore.kernel.org/linux-pm/20240206133605.1518373-1-stanislaw.gruszka@linux.intel.com/
> > v3: https://lore.kernel.org/linux-pm/20240209120625.1775017-1-stanislaw.gruszka@linux.intel.com/
> >
> > Stanislaw Gruszka (3):
> >   genetlink: Add per family bind/unbind callbacks
> >   thermal: netlink: Add genetlink bind/unbind notifications
> >   thermal: intel: hfi: Enable interface only when required
> >
> >  drivers/thermal/intel/intel_hfi.c | 95 +++++++++++++++++++++++++++----
> >  drivers/thermal/thermal_netlink.c | 40 +++++++++++--
> >  drivers/thermal/thermal_netlink.h | 26 +++++++++
> >  include/net/genetlink.h           |  4 ++
> >  net/netlink/genetlink.c           | 30 ++++++++++
> >  5 files changed, 180 insertions(+), 15 deletions(-)
> >
> > --
> 
> What tree is this based on?

v5: https://patchwork.kernel.org/project/linux-pm/list/?series=829159
it's on top of linux-pm master, but require additional net dependency,
which can be added by:

git fetch https://git.kernel.org/pub/scm/linux/kernel/git/kuba/linux.git for-thermal-genetlink-family-bind-unbind-callbacks
git merge FETCH_HEAD

and will be merged mainline in the next merge window.
So at this point would be probably better just wait for 6.9-rc1 
when the dependency will be in the mainline, before applying this set.

Regards
Stanislaw