[PATCH v3 00/22] soundwire: code hardening and suspend-resume support

From: Pierre-Louis Bossart
Date: Thu Nov 14 2019 - 13:17:31 EST


this patchset applies on top of "[PATCH v3 00/15] soundwire: intel:
implement new ASoC interfaces".

It implements a series of improvements for:
a) interrupt handling on Intel platforms in MSI mode
b) race conditions on codec probe and enumeration
c) suspend-resume issues (clock-stop mode not supported for now)
d) underflow handling
e) updates to the stream state machine which did not support valid
ALSA transitions.

These patches were tested extensively on 4 different platforms and are
viewed as required for any sort of SoundWire-based product. While
tested extensively on Intel platforms only, they should also benefit
Qualcomm platforms who haven't yet enabled power management.

Changes since v2: (no feedback received since November 6)
Added idle scheduling to deal with pm_runtime issues when devices are
exposed in the DSDT, but are not populated on the board. A quirk is
introduced to deal with potential cases where the devices might be
powered at a later time, in which case it's legit to leave the bus
active.
Fixed .prepare callback to handle both underflow and resume cases. The
previous version was incorrect in the first case and did not follow
recommended programming sequence
Fixed an additional race condition leading to a timeout when the codec
device was suspended while the master remained active.
Fixed a couple of warnings reported by static analysis
Removed non-essential pr_err() traces in stream.c, left others when
useful
Changed subject of patches dealing with race conditions to make sure
reviewers can link with the interface changes.


Changes since v1: (no feedback received since October 23)
added support for initialization_complete, integration with Realtek
codecs exposed an additional race condition between the resume
operation and restoration of settings in separate thread triggered by
Slave status change.
No other functional change

Bard Liao (3):
soundwire: intel/cadence: fix timeouts in MSI mode
soundwire: stream: only prepare stream when it is configured.
soundwire: intel: reinitialize IP+DSP in .prepare(), but only when
resuming

Pierre-Louis Bossart (19):
soundwire: bus: fix race condition with probe_complete signaling
soundwire: bus: add PM/no-PM versions of read/write functions
soundwire: bus: write Slave Device Number without runtime_pm
soundwire: intel: add helpers for link power down and shim wake
soundwire: intel: Add basic power management support
soundwire: intel: add pm_runtime support
soundwire: intel: reset pm_runtime status during system resume
soundwire: bus: add helper to reset Slave status to UNATTACHED
soundwire: intel: call helper to reset Slave states on resume
soundwire: bus: check first if Slaves become UNATTACHED
soundwire: bus: fix race condition with enumeration_complete signaling
soundwire: bus: fix race condition with initialization_complete
signaling
soundwire: bus: fix race condition by tracking UNATTACHED transition
soundwire: intel: disable pm_runtime when removing a master
soundwire: bus: disable pm_runtime in sdw_slave_delete
soundwire: stream: remove redundant pr_err traces
soundwire: stream: update state machine and add state checks
soundwire: stream: do not update parameters during DISABLED-PREPARED
transition
soundwire: intel: pm_runtime idle scheduling

Documentation/driver-api/soundwire/stream.rst | 63 ++-
drivers/soundwire/bus.c | 169 +++++++-
drivers/soundwire/bus.h | 9 +
drivers/soundwire/bus_type.c | 5 +
drivers/soundwire/cadence_master.c | 17 +-
drivers/soundwire/cadence_master.h | 8 +
drivers/soundwire/intel.c | 400 ++++++++++++++++--
drivers/soundwire/intel.h | 2 +
drivers/soundwire/intel_init.c | 45 +-
drivers/soundwire/slave.c | 4 +
drivers/soundwire/stream.c | 71 +++-
include/linux/soundwire/sdw.h | 1 +
include/linux/soundwire/sdw_intel.h | 4 +
13 files changed, 714 insertions(+), 84 deletions(-)

--
2.20.1