Re: watchdog infrastructure

[Arnd Bergmann]

On Montag, 5. Juli 2004 11:38, you wrote:
> Did you have a look allready at the different watchdog operations in 
> include/linux/watchdog.h ?

Yes, I already have a working driver, which does not use the
experimental infrastructure code. I'm just not allowed to
publish it until the hardware is available.

There are a couple of things I noticed about your new code:

- Is there any reason having an alloc_watchdogdev function in the
  common code? Simply statically allocating the structure in each
  device driver should be a lot simpler.
- Keeping watchdog_ops out of watchdog_device will simplify 
  the lifetime rules. Just put them in the same structure, add an
  owner field and get rid of the *private field.
- watchdog_is_open_sem can just be an atomic_t, you never
  actually down() it.
- You need to get the module reference count before calling any
  watchdog operation, the best place for this is probably the
  open() fop.
- Maybe its easier to always register the misc devices when
  watchdog.ko is loaded, and then deny opening them when no
  actual watchdog driver is registered to it.
- Why do you need seperate operations for start and keepalive?
- the reboot notifier and the nowayout parameter are probably
  common enough to be put into the generic module.

	Arnd <><

Attachment: pgp00000.pgp
Description: signature