Wake On Lan device semantics

Wake On Lan device semantics

[Stephen Hemminger]

I am working on getting WOL to work on sky2 (and then skge). But in the process I
noticed that the semantics of WOL seems to be device dependent. I assume that WOL
should work when device is suspended. But some drivers also support WOL when
the device is down (or even removed).

Now I know some distro's like Ubuntu take down and then remove every network
device on suspend. That's their problem, if they don't want to use suspend
as intended because they want to handle broken hardware, that's their problem.

It doesn't seem like a good idea for a network device to wake the system
if it is down. Maybe if the kernel fully supported dormant, maybe, but
when device is down it shouldn't impact the system.


-- 
Stephen Hemminger <shemminger@osdl.org>

Re: Wake On Lan device semantics

[Auke Kok]

Stephen Hemminger wrote:
> It doesn't seem like a good idea for a network device to wake the system
> if it is down.

before suspend existed this was the only useful case for WoL. Why does it not seem a 
good idea to wake up a machine that was shutdown (and thus the interface `downed`) ?

Auke

Re: Wake On Lan device semantics

[Stephen Hemminger]

On Fri, 03 Nov 2006 15:44:13 -0800
Auke Kok <auke-jan.h.kok@intel.com> wrote:

> Stephen Hemminger wrote:
> > It doesn't seem like a good idea for a network device to wake the system
> > if it is down.
> 
> before suspend existed this was the only useful case for WoL. Why does it not seem a 
> good idea to wake up a machine that was shutdown (and thus the interface `downed`) ?
> 
> Auke

It seems odd because that means you can never make a device fully deaf.

-- 
Stephen Hemminger <shemminger@osdl.org>

Re: Wake On Lan device semantics

[Stephen Hemminger]

On Fri, 03 Nov 2006 15:44:13 -0800
Auke Kok <auke-jan.h.kok@intel.com> wrote:

> Stephen Hemminger wrote:
> > It doesn't seem like a good idea for a network device to wake the system
> > if it is down.
> 
> before suspend existed this was the only useful case for WoL. Why does it not seem a 
> good idea to wake up a machine that was shutdown (and thus the interface `downed`) ?
> 
> Auke

Interestingly it looks like e100 is one of the ones that only wakes from suspend (not when down).


-- 
Stephen Hemminger <shemminger@osdl.org>

Re: Wake On Lan device semantics

[Jeff Garzik]

Stephen Hemminger wrote:
> I am working on getting WOL to work on sky2 (and then skge). But in the process I
> noticed that the semantics of WOL seems to be device dependent. I assume that WOL
> should work when device is suspended. But some drivers also support WOL when
> the device is down (or even removed).

[...]
> It doesn't seem like a good idea for a network device to wake the system
> if it is down. Maybe if the kernel fully supported dormant, maybe, but
> when device is down it shouldn't impact the system.


You seem to be muddling "device", "driver", and "system" together.

The purpose of WOL is being able to turn on a system remotely, if it is 
in a power-off or sleep state.

So, if the system is -on- and the interface is down and/or driver is 
unloaded, are you saying WOL is a problem somehow?

	Jeff

Re: Wake On Lan device semantics

[David Miller]

From: Jeff Garzik <jgarzik@pobox.com>
Date: Fri, 03 Nov 2006 18:51:25 -0500

> The purpose of WOL is being able to turn on a system remotely, if it is 
> in a power-off or sleep state.
> 
> So, if the system is -on- and the interface is down and/or driver is 
> unloaded, are you saying WOL is a problem somehow?

Stephen is saying that if you down an interface, it should disable
that WoL functionality.

I guess you can argue that, like IP addresses, this WoL thing is an
attribute of the "system".

Re: Wake On Lan device semantics

[Stephen Hemminger]

On Fri, 03 Nov 2006 16:02:30 -0800 (PST)
David Miller <davem@davemloft.net> wrote:

> From: Jeff Garzik <jgarzik@pobox.com>
> Date: Fri, 03 Nov 2006 18:51:25 -0500
> 
> > The purpose of WOL is being able to turn on a system remotely, if it is 
> > in a power-off or sleep state.
> > 
> > So, if the system is -on- and the interface is down and/or driver is 
> > unloaded, are you saying WOL is a problem somehow?
> 
> Stephen is saying that if you down an interface, it should disable
> that WoL functionality.
> 
> I guess you can argue that, like IP addresses, this WoL thing is an
> attribute of the "system".

Looking harder. The semantic needs to be WOL is okay if driver is loaded
and device is up or down. But the default for WOL should be disabled until
enabled by ethtool (or parameter).

-- 
Stephen Hemminger <shemminger@osdl.org>

Re: Wake On Lan device semantics

[Jeff Garzik]

David Miller wrote:
> From: Jeff Garzik <jgarzik@pobox.com>
> Date: Fri, 03 Nov 2006 18:51:25 -0500
> 
>> The purpose of WOL is being able to turn on a system remotely, if it is 
>> in a power-off or sleep state.
>>
>> So, if the system is -on- and the interface is down and/or driver is 
>> unloaded, are you saying WOL is a problem somehow?
> 
> Stephen is saying that if you down an interface, it should disable
> that WoL functionality.

Many distros down the interface on poweroff, a state from which WOL is 
often used, so we don't want this.


> I guess you can argue that, like IP addresses, this WoL thing is an
> attribute of the "system".

Yeah, it's definitely a system state.  When the magic packet arrives, 
the WOL wire on the motherboard is tickled, turning the machine on.

	Jeff

Re: Wake On Lan device semantics

[David Miller]

From: Jeff Garzik <jgarzik@pobox.com>
Date: Fri, 03 Nov 2006 19:42:49 -0500

> David Miller wrote:
> > I guess you can argue that, like IP addresses, this WoL thing is an
> > attribute of the "system".
> 
> Yeah, it's definitely a system state.  When the magic packet arrives, 
> the WOL wire on the motherboard is tickled, turning the machine on.

Ok, and Stephen seems to agree now too on this point.

I think there is merit to Stephen's assertion that WoL should
be off by default.  It allows remote entities to do something
to your computer.

I'm happy to hear counter-arguments, of course :-)

Re: Wake On Lan device semantics

[Auke Kok]

Stephen Hemminger wrote:
> On Fri, 03 Nov 2006 15:44:13 -0800
> Auke Kok <auke-jan.h.kok@intel.com> wrote:
> 
>> Stephen Hemminger wrote:
>>> It doesn't seem like a good idea for a network device to wake the system
>>> if it is down.
>> before suspend existed this was the only useful case for WoL. Why does it not seem a 
>> good idea to wake up a machine that was shutdown (and thus the interface `downed`) ?
>>
>> Auke
> 
> It seems odd because that means you can never make a device fully deaf.

sure you can, just turn off WoL and e1000 will really shutdown (at least, I hope :))

Auke

Re: Wake On Lan device semantics

[Auke Kok]

Stephen Hemminger wrote:
> On Fri, 03 Nov 2006 15:44:13 -0800
> Auke Kok <auke-jan.h.kok@intel.com> wrote:
> 
>> Stephen Hemminger wrote:
>>> It doesn't seem like a good idea for a network device to wake the system
>>> if it is down.
>> before suspend existed this was the only useful case for WoL. Why does it not seem a 
>> good idea to wake up a machine that was shutdown (and thus the interface `downed`) ?
>>
>> Auke
> 
> Interestingly it looks like e100 is one of the ones that only wakes from suspend (not when down).

that would be a bug, I'll have to get that checked especially after the latest changes 
to it.

Auke

Re: Wake On Lan device semantics

[Stephen Hemminger]

On Fri, 03 Nov 2006 17:36:45 -0800
Auke Kok <auke-jan.h.kok@intel.com> wrote:

> Stephen Hemminger wrote:
> > On Fri, 03 Nov 2006 15:44:13 -0800
> > Auke Kok <auke-jan.h.kok@intel.com> wrote:
> > 
> >> Stephen Hemminger wrote:
> >>> It doesn't seem like a good idea for a network device to wake the system
> >>> if it is down.
> >> before suspend existed this was the only useful case for WoL. Why does it not seem a 
> >> good idea to wake up a machine that was shutdown (and thus the interface `downed`) ?
> >>
> >> Auke
> > 
> > Interestingly it looks like e100 is one of the ones that only wakes from suspend (not when down).
> 
> that would be a bug, I'll have to get that checked especially after the latest changes 
> to it.
> 

Sorry, my bad my test machine was not setup properly.

Re: Wake On Lan device semantics

[Auke Kok]

Stephen Hemminger wrote:
> On Fri, 03 Nov 2006 17:36:45 -0800
> Auke Kok <auke-jan.h.kok@intel.com> wrote:
> 
>> Stephen Hemminger wrote:
>>> On Fri, 03 Nov 2006 15:44:13 -0800
>>> Auke Kok <auke-jan.h.kok@intel.com> wrote:
>>>
>>>> Stephen Hemminger wrote:
>>>>> It doesn't seem like a good idea for a network device to wake the system
>>>>> if it is down.
>>>> before suspend existed this was the only useful case for WoL. Why does it not seem a 
>>>> good idea to wake up a machine that was shutdown (and thus the interface `downed`) ?
>>>>
>>>> Auke
>>> Interestingly it looks like e100 is one of the ones that only wakes from suspend (not when down).
>> that would be a bug, I'll have to get that checked especially after the latest changes 
>> to it.
>>
> 
> Sorry, my bad my test machine was not setup properly.

I don't blame you, WoL is one of the hardest features to get right, especially with all 
the various e100{,0} hardware and options that influence it (manageability, eeproms, etc).

On top of that there's this blob called BIOS messing it all up after a powercycle :)

Cheers,

Auke

Re: Wake On Lan device semantics

[Auke Kok]

Stephen Hemminger wrote:
> On Fri, 03 Nov 2006 16:02:30 -0800 (PST)
> David Miller <davem@davemloft.net> wrote:
> 
>> From: Jeff Garzik <jgarzik@pobox.com>
>> Date: Fri, 03 Nov 2006 18:51:25 -0500
>>
>>> The purpose of WOL is being able to turn on a system remotely, if it is 
>>> in a power-off or sleep state.
>>>
>>> So, if the system is -on- and the interface is down and/or driver is 
>>> unloaded, are you saying WOL is a problem somehow?
>> Stephen is saying that if you down an interface, it should disable
>> that WoL functionality.
>>
>> I guess you can argue that, like IP addresses, this WoL thing is an
>> attribute of the "system".
> 
> Looking harder. The semantic needs to be WOL is okay if driver is loaded
> and device is up or down. But the default for WOL should be disabled until
> enabled by ethtool (or parameter).

Since this is enabled already in the BIOSes for most systems (or disabled, and 
switcheable by the user), I'd say we can let the driver handle the default (on for most 
cards) already. The BIOS can also handle a WoL call after a (long) powerfailure, before 
the OS boots, so disabling it in the OS would be quite useless in that case: the BIOS 
would override it anyway.

Doesn't sound like a problem you can solve in the driver layer...

Cheers,

Auke