Am 06.03.2017 um 09:11 schrieb Johannes Berg:
> On Sun, 2017-03-05 at 16:12 +0100, Oliver Freyermuth wrote:
>
> [snip]
>
>> With tcp-mode, and using the small testing code I found from you
>> somewhere on the web:
>> https://bugs.chromium.org/p/chromium/issues/detail?id=361435#c21
>> indeed I get:
>> "new client!"
>> after the machine has suspended.
>> It even seems to receive the "WAKEUPNOW" when I close the server,
>> since it does not make further reconnection attempts after.
>
> Heh. I think you're quite possibly the first person (other than me) to
> ever use this :)
>
It seems the script itself is really working as it should, see below ;-).
> [snip]
>
>> So nothing which looks bad as far as I can see.
>
> Agree, nothing looks bad. Try running "iw event" while you suspend - if
> the NIC thought it woke up the system there will be an event indicating
> so.
>
Indeed, I get:
1488836432.153465: wlan0 (phy #0): WoWLAN wakeup
* packet (might be truncated):
ac:fd:ce:de:e1:58:b8:27:eb:f4:3b:4c:08:00:45:00:00:31:78:d1:40:00:3f:06:3d:55:c0:a8:02:23:c0:a8:02:2d:30:39:30:39:f2:04:6f:88:8b:74:4c:c2:50:18:72:10:00:00:00:00:57:41:4b:45:55:50:4e:4f:57
* TCP connection wakeup received
Note the trailing "57:41:4b:45:55:50:4e:4f:57" which is your "WAKEUPNOW".
Also magic packet "works" this way, i.e. shows up via "iw event".
>> Maybe this UEFI firmware is broken in a different, albeit "more
>> stupid" way:
>> Discarding ACPI wakeup from PCI devices, even though it leaves them
>> powered?
>
> I guess it's possible. I guess there's also a chance that we're missing
> some setup in the driver though. I vaguely remember a patch in ChromeOS
> that somehow we never merged ... ah yes, must've been this one:
> https://chromium-review.googlesource.com/311101
> though that means it's not relevant in your system.
>
> So basically I don't really see anything missing in the driver, hmm.
I realized I also have a Win10 installation on that machine - and may as well
try WoWLAN with it.
Enabling (in the dreaded device manager) wake via the WiFi device, magic packet
mode, I also cannot wake the machine.
So if it was a driver bug, it would have to be present both in Windows and
Linux drivers - but I guess it's just another case
of broken system firmware *sigh*.
And that's even though it's a pretty recent Clevo W230SD-based machine (which I
considered decently widespread)
with unlocked UEFI (so I see almost all options), but of course nothing related
to ACPI wakeup.
But I guess widespread alone does not help ;-).
>
> johannes
>
Thanks for all!
Oliver
