Re: [Ilw] Intel Wireless 7260 hardware timed out randomly

[Emmanuel Grumbach <egrumbach@gmail.com>]

On 11/15/2013 05:06 AM, wzyboy wrote:
> 2013/11/15 Bjorn Helgaas <bhelgaas@google.com>:
>> Why would it be unlikely to fix the driver?  Do people think the
>> problem is not actually in the driver?
>>
>> Asking Lenovo how to disable L1 PM substates is really a non-answer.
>> Only the extremely technical and extremely patient user (hi wzyboy :))
>> will even bother to investigate why wifi works fine with Windows but
>> not with Linux.  The only thing Lenovo *could* do is to release a new
>> BIOS with a switch to control L1 PM Substates.  If I were Lenovo, I
>> would never do that because then I would have to tell customers
>> "disable this for Linux, enable this for Windows," and I'd have to
>> deal with support calls about devices using more power than they
>> should, battery life being shorter, etc.  Plus you'd have to ask every
>> Linux user to upgrade their BIOS.  That's all just a terrible user
>> experience.
> 
> 
> I am a little confused. There are two sets of "setpci" commands, both
> of which can make me use my NIC reliably. But you two say they are
> just workarounds, not real fixes.
> 

Right - because they force a mode that the BIOS doesn't allow. The BIOS
doesn't allow the OS (the driver) to decide in what mode to work - so we
cannot reach the same effect as the setpci command from the OS / driver
level. setpci just directly accesses the HW without asking the
permissions of anyone.

> I know the "side effect" of first two "setpci" commands is consuming
> more power. (Actually by my experience of running on battery, I did
> not notice ...)
> 
> But Grumbach said after the second two "setpci" commands enables "L1".
> Does it mean it saves power? So what's the "side effect" of second two
> "setpci" commands?
> 

They are both the same in terms of side-effects. The first set of setpci
commands will disable L1 altogether - meaning you don't save any power.
The second set of setpci doesn't disable L1, but disable a more subtle
power state (actually several) which are defined as L1 PM substates. In
theses substates, you save less power than in L1 (I think) but you are
more likely to be able to reach them. After all, it is always the same
story - the deeper you sleep, you longer it takes to wake up. And if it
takes longer to wake up, it also means that in several cases you won't
chose to go to sleep. So the way PCI folks help to save power even in
case where you cannot go to a deep sleep is to define states in the
middle in which you save less power, but in which you are more likely to
be. Again - time spent in each state and power saved in each state trade
off.
Now:
L1 - deep sleep
L1 PM substate - something in the middle.

First setpci command - disable both features.
Second setpci command - disable only the second feature.

Regarding side effects... I don't think this is really "dangerous". But
this is not a fix in the way that I wouldn't like to deploy millions of
machines like that. The risk you have here is probably to have a bad
timing and have the setpci commands run exactly when the link is in a
state that setpci disables. That would be bad. How bad? Probably would
just require a reboot - or worst case G3 (take the battery off).

> IMHO, if this could user use their NIC reliably, maybe Grumbach may
> write these commands to iwlwifi driver and run them when 7260 is
> detected...

I can't as exlained above.

> 
> BTW, no replies from Lenovo, yet.
> 


> Or maybe you could add an option, which enables this "workaround" if
> user wants. A user could simply write a /etc/modprobe.d/iwlwifi.conf
> and enable this "workaround", to use their NICs without having to
> reboot from time to time...

same.