mbox series

[v3,0/5] Thermal devfreq cooling improvements with Energy Model

Message ID 20201209103016.10442-1-lukasz.luba@arm.com
Headers show
Series Thermal devfreq cooling improvements with Energy Model | expand

Message

Lukasz Luba Dec. 9, 2020, 10:30 a.m. UTC
Hi all,

This patch set is a continuation of my previous work, which aimed
to add Energy Model to all devices [1]. This series is a follow up
for the patches which got merged to v5.9-rc1. It aims to change
the thermal devfreq cooling and use the Energy Model instead of
private power table and structures. The power model is now simplified,
static power and dynamic power are removed. The new registration interface
in the patch 3/5 helps to register devfreq cooling and the EM in one call.
There is also small improvement, patch 2/5 is changing the way how
thermal gets the device status (now uses a copy) and normalize the values.
The last patch is here for consistency and will probably go through drm tree.

The patch set is based on current next-20201208, because it depends on EM
API change which is queued in the pm/linux-next tree as v5.11 material.

changes:
v3:
- dropped direct check of device status and used just a copy of 'status';
  a separate patch set will be proposed to address this issue
- modified _normalize_load() and used 1024 scale to handle ms, us, ns
- removed 'em_registered' and called em_dev_unregister_perf_domain()
  unconditionally, so the drivers will have to make sure the right order of
  all unregister calls to frameworks which might use EM; this call must be last
  one; a proper comment added
- removed 'em' pointer from struct devfreq_cooling_device, 'dev->em_pd' is used
- removed of_node_get/put(), since the code can handle it
- removed dfc_em_get_requested_power() (as missed to do it in v2)
- collected all Reviewed-by tags
v2 [3]:
- renamed freq_get_state() and related to perf_idx pattern as
  suggested by Ionela
v1 [2]

Regards,
Lukasz Luba

[1] https://lkml.org/lkml/2020/5/11/326
[2] https://lore.kernel.org/linux-pm/20200921122007.29610-1-lukasz.luba@arm.com/
[3] https://lore.kernel.org/linux-pm/20201118120358.17150-1-lukasz.luba@arm.com/

Lukasz Luba (5):
  thermal: devfreq_cooling: change tracing function and arguments
  thermal: devfreq_cooling: use a copy of device status
  thermal: devfreq_cooling: add new registration functions with Energy
    Model
  thermal: devfreq_cooling: remove old power model and use EM
  drm/panfrost: Register devfreq cooling and attempt to add Energy Model

 drivers/gpu/drm/panfrost/panfrost_devfreq.c |   2 +-
 drivers/thermal/devfreq_cooling.c           | 420 ++++++++++----------
 include/linux/devfreq_cooling.h             |  40 +-
 include/trace/events/thermal.h              |  19 +-
 4 files changed, 240 insertions(+), 241 deletions(-)

Comments

Lukasz Luba Dec. 10, 2020, 1:22 p.m. UTC | #1
Hi Daniel,

On 12/9/20 10:30 AM, Lukasz Luba wrote:
> Hi all,
> 
> This patch set is a continuation of my previous work, which aimed
> to add Energy Model to all devices [1]. This series is a follow up
> for the patches which got merged to v5.9-rc1. It aims to change
> the thermal devfreq cooling and use the Energy Model instead of
> private power table and structures. The power model is now simplified,
> static power and dynamic power are removed. The new registration interface
> in the patch 3/5 helps to register devfreq cooling and the EM in one call.
> There is also small improvement, patch 2/5 is changing the way how
> thermal gets the device status (now uses a copy) and normalize the values.
> The last patch is here for consistency and will probably go through drm tree.
> 
> The patch set is based on current next-20201208, because it depends on EM
> API change which is queued in the pm/linux-next tree as v5.11 material.
> 
> changes:
> v3:
> - dropped direct check of device status and used just a copy of 'status';
>    a separate patch set will be proposed to address this issue
> - modified _normalize_load() and used 1024 scale to handle ms, us, ns
> - removed 'em_registered' and called em_dev_unregister_perf_domain()
>    unconditionally, so the drivers will have to make sure the right order of
>    all unregister calls to frameworks which might use EM; this call must be last
>    one; a proper comment added
> - removed 'em' pointer from struct devfreq_cooling_device, 'dev->em_pd' is used
> - removed of_node_get/put(), since the code can handle it
> - removed dfc_em_get_requested_power() (as missed to do it in v2)
> - collected all Reviewed-by tags
> v2 [3]:
> - renamed freq_get_state() and related to perf_idx pattern as
>    suggested by Ionela
> v1 [2]
> 
> Regards,
> Lukasz Luba
> 
> [1] https://lkml.org/lkml/2020/5/11/326
> [2] https://lore.kernel.org/linux-pm/20200921122007.29610-1-lukasz.luba@arm.com/
> [3] https://lore.kernel.org/linux-pm/20201118120358.17150-1-lukasz.luba@arm.com/
> 
> Lukasz Luba (5):
>    thermal: devfreq_cooling: change tracing function and arguments
>    thermal: devfreq_cooling: use a copy of device status
>    thermal: devfreq_cooling: add new registration functions with Energy
>      Model
>    thermal: devfreq_cooling: remove old power model and use EM
>    drm/panfrost: Register devfreq cooling and attempt to add Energy Model
> 
>   drivers/gpu/drm/panfrost/panfrost_devfreq.c |   2 +-
>   drivers/thermal/devfreq_cooling.c           | 420 ++++++++++----------
>   include/linux/devfreq_cooling.h             |  40 +-
>   include/trace/events/thermal.h              |  19 +-
>   4 files changed, 240 insertions(+), 241 deletions(-)
> 

If you consider to take it, please don't. I am going to send a v4 which
does not have this em_dev_register_perf_domain() dependency due to API
change. Then it could go via your thermal tree without issues.

It will be a small change in the patch 3/5, which will simplify
registration function (use only dev_pm_opp_of_register_em()) and also
instead of two registration function, have only one (which was also
suggested by Ionela during review).

Regards,
Lukasz