hisilicon-opensdk: bump to b922e19, so the watchdog can reset the board

[widgetii]

Bumps hisilicon-opensdk from 916e767 to b922e19, picking up OpenIPC/openhisilicon#223.

Fixes #1803. Touches #127 — see the end.

What it fixes

The watchdog driver did not honour the Linux watchdog API in six ways. The one
that matters most here is #1803: release() set the driver's state back to
SELFCLR, which handed the device to the driver's own kernel feeder thread. So
the moment majestic died and the kernel closed its descriptor for it, the driver
went on feeding the dog for the rest of the board's life, and nothing ever reset.
The only workaround available was nodeamon=1 — which removes the feeder
altogether, and with it the cover between module load and the first open.

Measured on the lab boards before the change:

board
hi3516ev300	majestic SIGKILLed at uptime 40195	still up at 40400
hi3516av300	SIGKILLed at 215043	still up at 215249
gk7205v200	probe died at 28050	still up 164 s later

The rest, briefly:

• The margin meant twice itself. The SP805 runs its load value twice before
  resetting, so a whole margin in the counter made every timeout double. 30 s
  reset the board at 58.
• A ping did nothing for half of every cycle. feed() returned without
  touching anything unless an interrupt was already pending, so the honest
  cadence of one ping per half-margin was worth one accepted ping per cycle —
  miss it and a process that had been feeding all along got reset under it.
• open() did not start the timer. After any deliberate stop the device was
  inert while looking entirely healthy: open succeeded, the margin was accepted
  and read back, pings were accepted, and nothing would ever reset.
• A reopen ran at the counter maximum, not the margin the caller had set —
  GETTIMEOUT said 30 while the hardware ran at 1431 s.
• rmmod racing an open() could wedge the board in D state holding
  module_mutex, past a clean reboot. It cost one lab board a power cycle.

hi3516cv300 gains a watchdog it never had

Its driver contained no module entry point and its build linked no init object,
so hi3516cv300_wdt.ko loaded, did nothing, and the board came up with no
/dev/watchdog. majestic logged Watchdog device (/dev/watchdog) not found
and every cv300 has been running unguarded. Confirmed on a lab cv300 before the
change; after it majestic logs Detected Watchdog / Watchdog timeout set to 300 seconds for the first time.

Verified

Four lab cameras, both binding styles. Crash path is open, SETTIMEOUT(30), die
without "V"; the board must reset one margin later and come back.

board	crash → reset	GETTIMELEFT	reopen keeps margin	clean majestic stop	majestic soak
hi3516cv300	30 s (was: no device)	299/300	yes	no reset	456 s
hi3516av300	30 s (was: never)	299/300	yes	no reset	491 s
hi3516ev300	30 s (was: never)	299/300	yes	no reset	1327 s
gk7205v200	30 s (was: never)	299/300	yes	no reset	—

All 35 checks green on the upstream PR, including the SDK builds and QEMU boots
for every affected chip.

Worth a release note

A cv300 whose majestic wedges will now reboot where it previously sat there
unguarded. That is the fix working, but it is a behaviour change in the field
and the first thing anyone will notice.

On #127

That issue asks for the watchdog module on hi3516cv200/hi3518ev200. It is
already insmoded there — the #nodeamon=1 beside it in load_hisilicon is a
comment, not a parameter — but the .ko is a prebuilt vendor blob shipped in
general/package/hisilicon-osdrv-hi3516cv200/files/kmod/wdt.ko, dated 2022. So
none of the above reaches those boards. The V2 generation also has no OSAL at
all, which is why upstream left its copy of the driver alone. Both are separate
pieces of work; noting it here so the issue is not closed on the strength of
this bump.

[qodo-free-for-open-source-projects]

PR Summary by Qodo

Bump hisilicon-opensdk to restore watchdog resets

🐞 Bug fix ✨ Enhancement 🕐 10-20 Minutes

AI Description

• Bump openhisilicon to b922e19, incorporating Linux watchdog API fixes.
• Restore crash-triggered resets and reliable timeout, ping, reopen, and unload behavior.
• Enable the previously nonfunctional watchdog on hi3516cv300.

Diagram

graph TD
  A["Buildroot Package"] -->|pins| B["OpenHiSilicon SDK"] -->|builds| C["Watchdog Module"] -->|exposes| D["/dev/watchdog"] -->|used by| E["Majestic"]
  D -->|timeout| F["Board Reset"]

Loading

High-Level Assessment

Pinning the tested upstream commit is the appropriate approach because it incorporates the complete, jointly validated watchdog fixes without maintaining a downstream patch stack. Cherry-picking individual fixes was considered but would risk omitting interacting lifecycle corrections and diverging from upstream.

Files changed (1) +1 / -1

Other (1) +1 / -1

hisilicon-opensdk.mkPin hisilicon-opensdk to watchdog-fixed revision b922e19 +1/-1

Pin hisilicon-opensdk to watchdog-fixed revision b922e19

• Updates the OpenHiSilicon source revision from 916e767 to b922e19. The new revision supplies corrected watchdog behavior and enables the hi3516cv300 watchdog module.

general/package/hisilicon-opensdk/hisilicon-opensdk.mk

[qodo-free-for-open-source-projects]

Code Review by Qodo

🐞 Bugs (0) 📘 Rule violations (0) 📎 Requirement gaps (0)

Great, no issues found!

Qodo reviewed your code and found no material issues that require review

Tip of the day

💡 Did you know, you can hide the parts of a finding you never read, like the evidence or the agent prompt

More tips ↗ | Customize Qodo ↗ | Qodo docs ↗