mbox series

[v3,0/4] Add support for hwspinlock bust

Message ID 20240529-hwspinlock-bust-v3-0-c8b924ffa5a2@quicinc.com (mailing list archive)
Headers show
Series Add support for hwspinlock bust | expand

Message

Chris Lew May 29, 2024, 6:09 p.m. UTC 
hwspinlocks can be acquired by many devices on the SoC. If any of these
devices go into a bad state before the device releases the hwspinlock,
then that hwspinlock may end up in an unusable state.

In the case of smem, each remoteproc takes a hwspinlock before trying to
allocate an smem item. If the remoteproc were to suddenly crash without
releasing this, it would be impossible for other remoteprocs to allocate
any smem items.

We propose a new api to bust a hwspinlock. A driver can use the
the bust api if it detects the device has gone into an error state, thus
allowing other entities in the system to use the hwspinlock.

These patches were tested on an sm8650 mtp using engineering cdsp
firmware that triggers a watchdog with the smem hwspinlock acquired.

Checked for error in dt-bindings with below.
 - make DT_CHECKER_FLAGS=-m DT_SCHEMA_FILES=remoteproc/qcom,pas-common.yaml dt_binding_check
 - make qcom/sm8650-mtp.dtb CHECK_DTBS=1

Signed-off-by: Chris Lew <quic_clew@quicinc.com>
---
Changes in v3:
- Fixed 80 char comment formatting and missing space suggested by Bjorn
- Changed unsigned to unsigned int in smem apis from checkpatch warning
- Removed null pointer check in smem bust api
- Picked up reviewed-by trailers from Bjorn
- Link to v2: https://lore.kernel.org/r/20240524-hwspinlock-bust-v2-0-fb88fd17ca0b@quicinc.com

Changes in v2:
- Remove extra print in qcom_q6v5_pas as suggested by Bryan
- Expose SMEM API that remotproc can call to bust the smem hwlock as suggested by Krzysztof
- Drop patches related to multiple references on a hwlock
- Link to v1: https://lore.kernel.org/r/20240516-hwspinlock-bust-v1-0-47a90a859238@quicinc.com

---
Chris Lew (1):
      soc: qcom: smem: Add qcom_smem_bust_hwspin_lock_by_host()

Richard Maina (3):
      hwspinlock: Introduce hwspin_lock_bust()
      hwspinlock: qcom: implement bust operation
      remoteproc: qcom_q6v5_pas: Add hwspinlock bust on stop

 Documentation/locking/hwspinlock.rst     | 11 +++++++++++
 drivers/hwspinlock/hwspinlock_core.c     | 28 ++++++++++++++++++++++++++++
 drivers/hwspinlock/hwspinlock_internal.h |  3 +++
 drivers/hwspinlock/qcom_hwspinlock.c     | 25 +++++++++++++++++++++++++
 drivers/remoteproc/qcom_q6v5_pas.c       | 11 +++++++++++
 drivers/soc/qcom/smem.c                  | 26 ++++++++++++++++++++++++++
 include/linux/hwspinlock.h               |  6 ++++++
 include/linux/soc/qcom/smem.h            |  2 ++
 8 files changed, 112 insertions(+)
---
base-commit: e7b4ef8fffaca247809337bb78daceb406659f2d
change-id: 20240509-hwspinlock-bust-d497a70c1a3a

Best regards,

Comments

Bjorn Andersson May 29, 2024, 9:28 p.m. UTC | #1 
On Wed, 29 May 2024 11:09:54 -0700, Chris Lew wrote:
> hwspinlocks can be acquired by many devices on the SoC. If any of these
> devices go into a bad state before the device releases the hwspinlock,
> then that hwspinlock may end up in an unusable state.
> 
> In the case of smem, each remoteproc takes a hwspinlock before trying to
> allocate an smem item. If the remoteproc were to suddenly crash without
> releasing this, it would be impossible for other remoteprocs to allocate
> any smem items.
> 
> [...]

Applied, thanks!

[3/4] soc: qcom: smem: Add qcom_smem_bust_hwspin_lock_by_host()
      commit: 2e3f0d693875db698891ffe89a18121bda5b95b8
[4/4] remoteproc: qcom_q6v5_pas: Add hwspinlock bust on stop
      commit: 568b13b65078e2b557ccf47674a354cecd1db641

Best regards,