From mboxrd@z Thu Jan  1 00:00:00 1970
From: Kay Sievers <kay.sievers@vrfy.org>
Date: Mon, 03 Oct 2005 12:21:30 +0000
Subject: Re: [RFC] coldplug - emit hotplug events from sysfs
Message-Id: <20051003122130.GA10321@vrfy.org>
List-Id: <linux-hotplug.vger.kernel.org>
References: <20051001124943.GA26076@vrfy.org>
In-Reply-To: <20051001124943.GA26076@vrfy.org>
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit
To: linux-hotplug@vger.kernel.org

On Mon, Oct 03, 2005 at 12:29:03PM +0200, Olivier Blin wrote:
> Kay Sievers <kay.sievers@vrfy.org> writes:
> 
> > Here is another approach to address the coldplug problem. After playing
> > around with "udevsythesize" it felt stupid to try to compose all the events
> > from sysfs values which will never be exactly the same as the kernel
> > event.
> > This patch adds a "uevent" file to all class/block/bus-devices. Writing to
> > this file, will just emit the hotplug event from the kernel again. udevstart
> > and all the coldplug logic can be 3 lines of shell-script now. Boots fine
> > and is _very_ fast on my box.
> >
> > What do you think?
> 
> It looks really nice, but will we still be able to differientate
> coldplug events in udev rules?
> For example, in Mandriva, we do PCI coldplug for USB controllers only,
> because we handle other stuff (sound, network, ...) later in
> services. We used to rely on UDEV_START in udev rules.

Not with this simple patch. It would be possible by adding an additional
parameter to kobject_hotplug() to add an aditional key. Or we could use a
different ACTION value and map it with udev...

But unlike udevstart/udevsynthesize, you have complete control which
events are generated, cause you will do it from a shell script and not
with a binary. Wouldn't it work, to just trigger the events for the
usb controllers and and leave the others alone?

To play with it, I use this stupid hack in an early init-script and
it works without udevstart or any other coldplug logic.
After mounting tmpfs, creating /dev/null, disabling /sbin/hotplug and
starting udevd, it creates the tty devices, waits for the events to
finish, then sends out all remaining events to finish asynchronously:
  ...
  # regenerate events by triggering sysfs
  for i in /sys/class/t*/*/uevent; do echo 1 > $i; done
  # wait for async tty events to finish
  while [ $(cat /proc/*/status 2> /dev/null | grep -c -E '^Name:.udevd?$') -gt 1 ]; do
      sleep 0.1
  done
  for i in /sys/class/[!t]*/*/uevent; do echo 1 > $i; done
  for i in /sys/block/*/uevent; do echo 1 > $i; done
  for i in /sys/block/*/*[1-9]/uevent; do echo 1 > $i; done
  for i in /sys/bus/*/devices/*/uevent; do echo 1 > $i; done
  ...

To make this reliable, we would need to trigger some events in a defined
order, like the device mapper events after the block device events and
similar, and may need some checkpoints to make sure that certain
devices are available at a specific point...

Anyway, it works on two of my boxes and is the fastest udev/coldplug
bootup, I've ever seen. :)

Thanks,
Kay


-------------------------------------------------------
This SF.Net email is sponsored by:
Power Architecture Resource Center: Free content, downloads, discussions,
and more. http://solutions.newsforge.com/ibmarch.tmpl
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel