Re: [PATCH v6 3/4] driver core: shut down devices asynchronously

[Jeremy Allison]

On Wed, Jun 12, 2024 at 02:55:11PM -0600, Keith Busch wrote:
> On Thu, May 16, 2024 at 10:49:19AM -0500, Stuart Hayes wrote:
> > Add code to shut down devices asynchronously, while ensuring that each
> > device is shut down before its parents & suppliers, and allowing devices
> > that share a driver to be shutdown one at a time if necessary.
> >
> > Add /sys/kernel/async_shutdown to allow user control of this feature:
> >
> >   safe: shut down all devices synchronously, unless driver prefers async
> > 	shutdown (driver opt-in) (default)
> >   on:	shut down all devices asynchronously, unless disabled by the driver
> > 	(driver opt-out)
> >   off:	shut down all devices synchronously
> >
> > Add async_shutdown to struct device_driver, and expose it via sysfs.
> > This will be used to view or change driver opt-in/opt-out of asynchronous
> > shutdown, if it is globally enabled.
> >
> >  async:   driver opt-in to async device shutdown (devices will be shut down
> >           asynchronously if async_shutdown is "on" or "safe")
> >  sync:    driver opt-out of async device shutdown (devices will always be
> >           shut down synchronously)
> >  default: devices will be shutdown asynchronously if async_shutdown is "on"
> >
> > This can dramatically reduce system shutdown/reboot time on systems that
> > have multiple devices that take many seconds to shut down (like certain
> > NVMe drives). On one system tested, the shutdown time went from 11 minutes
> > without this patch to 55 seconds with the patch.
>
> I've successfully tested this out on a few systems, and noticing a very
> decent shutdown time on my nvme systems. I also like the current
> solution here, as the two-pass method was harder to follow.
>
> So I think just remove the extra options that Christoph mentioned and
> always use the driver's preferred shutdown method, then this would all
> look good to me.

Yes, I have tested this patch on my systems and am greatly in
favour of this instead of the two-pass version I was trying
to make work. It is easy to understand and fixes the problem
for my NVME issue.

Jeremy.