[V4,0/4] firmware: arm_scmi: Misc Fixes

Message

Sibi Sankar Oct. 23, 2024, 10:21 a.m. UTC

The series addresses the kernel warnings reported by Johan at [1] and are
are required to X1E cpufreq device tree changes [2] to land.

[1] - https://lore.kernel.org/lkml/[e-mail]/
[2] - https://lore.kernel.org/lkml/[e-mail]/

The following warnings remain unadressed:
arm-scmi arm-scmi.0.auto: Failed to add opps_by_lvl at 3417600 for NCC - ret:-16
arm-scmi arm-scmi.0.auto: Failed to add opps_by_lvl at 3417600 for NCC - ret:-16

They indicate that duplicate opps are reported by the SCP firmware and
they are seen during probe. They will get addressed by firmware updates. 

Duplicate levels:
arm-scmi arm-scmi.0.auto: Level 2976000 Power 218062 Latency 30us Ifreq 2976000 Index 10
arm-scmi arm-scmi.0.auto: Level 3206400 Power 264356 Latency 30us Ifreq 3206400 Index 11
arm-scmi arm-scmi.0.auto: Level 3417600 Power 314966 Latency 30us Ifreq 3417600 Index 12
arm-scmi arm-scmi.0.auto: Failed to add opps_by_lvl at 3417600 for NCC - ret:-16
arm-scmi arm-scmi.0.auto: Failed to add opps_by_lvl at 3417600 for NCC - ret:-16
arm-scmi arm-scmi.0.auto: Level 4012800 Power 528848 Latency 30us Ifreq 4012800 Index 15

^^ exist because SCP reports duplicate values for the highest sustainable
freq for perf domains 1 and 2. These are the only freqs that appear as
duplicates and will be fixed with a firmware update. FWIW the warnings
that we are addressing in this series will also get fixed by a firmware
update but they still have to land for devices already out in the wild.

V3:
* Pick up R-b, T-b from the list.
* Pick up the updated patch from Cristian for skipping opps.
* Update device names only when a name collision occurs [Dmitry/Ulf]
* Drop Johan's T-b from "fix debugfs node creation failure"
* Move scmi_protocol_msg_check to the top [Sudeep]

V2:
* Include the fix for do_xfer timeout
* Include the fix debugfs node creation failure
* Include Cristian's fix for skipping opp duplication

V1:
* add missing MSG_SUPPORTS_FASTCHANNEL definition.

Base branch: next-20241023

Cristian Marussi (1):
  firmware: arm_scmi: Skip opp duplicates

Sibi Sankar (3):
  firmware: arm_scmi: Ensure that the message-id supports fastchannel
  pmdomain: core: Fix debugfs node creation failure
  mailbox: qcom-cpucp: Mark the irq with IRQF_NO_SUSPEND flag

 drivers/firmware/arm_scmi/driver.c    | 72 +++++++++++++++------------
 drivers/firmware/arm_scmi/perf.c      | 40 +++++++++++----
 drivers/firmware/arm_scmi/protocols.h |  2 +
 drivers/mailbox/qcom-cpucp-mbox.c     |  2 +-
 drivers/pmdomain/core.c               | 65 ++++++++++++++++++------
 include/linux/pm_domain.h             |  1 +
 6 files changed, 123 insertions(+), 59 deletions(-)

Comments

Johan Hovold Oct. 25, 2024, 2:02 p.m. UTC | #1

On Wed, Oct 23, 2024 at 03:51:44PM +0530, Sibi Sankar wrote:
> The series addresses the kernel warnings reported by Johan at [1] and are
> are required to X1E cpufreq device tree changes [2] to land.
> 
> [1] - https://lore.kernel.org/lkml/[e-mail]/
> [2] - https://lore.kernel.org/lkml/[e-mail]/
> 
> The following warnings remain unadressed:
> arm-scmi arm-scmi.0.auto: Failed to add opps_by_lvl at 3417600 for NCC - ret:-16
> arm-scmi arm-scmi.0.auto: Failed to add opps_by_lvl at 3417600 for NCC - ret:-16
> 
> They indicate that duplicate opps are reported by the SCP firmware and
> they are seen during probe. They will get addressed by firmware updates. 
> 
> Duplicate levels:
> arm-scmi arm-scmi.0.auto: Level 2976000 Power 218062 Latency 30us Ifreq 2976000 Index 10
> arm-scmi arm-scmi.0.auto: Level 3206400 Power 264356 Latency 30us Ifreq 3206400 Index 11
> arm-scmi arm-scmi.0.auto: Level 3417600 Power 314966 Latency 30us Ifreq 3417600 Index 12
> arm-scmi arm-scmi.0.auto: Failed to add opps_by_lvl at 3417600 for NCC - ret:-16
> arm-scmi arm-scmi.0.auto: Failed to add opps_by_lvl at 3417600 for NCC - ret:-16
> arm-scmi arm-scmi.0.auto: Level 4012800 Power 528848 Latency 30us Ifreq 4012800 Index 15
> 
> ^^ exist because SCP reports duplicate values for the highest sustainable
> freq for perf domains 1 and 2. These are the only freqs that appear as
> duplicates and will be fixed with a firmware update. FWIW the warnings
> that we are addressing in this series will also get fixed by a firmware
> update but they still have to land for devices already out in the wild.

Thanks for sorting this out, Sibi!

I guess the above description is clear enough too as to why the
opps_by_lvl warnings should stay (in some form).

Johan