On Sat, Jun 29, 2013 at 07:20:36PM +0200, Johannes Berg wrote:

On Tue, 2013-06-25 at 23:01 +0200, Johannes Stezenbach wrote:

quoted

On Tue, Jun 25, 2013 at 08:38:16PM +0200, Johannes Berg wrote:

quoted

Thanks to Ben's review mostly, here are some fixes/cleanups in
alx. I'm seriously considering removing WoWLAN as I have no use
for it and am not too motivated to debug it, anyone else want
to try debugging? :)

One thing I noticed is alx_select_powersaving_speed() will return 0
when there is no link but leaves *speed uninitialized.

Yeah, that's a bit odd.

quoted

Then __alx_shutdown() prints e.g.
   alx 0000:03:00.0 eth0: wol: ctrl=3, speed=FFFF8801
In this case the
  alx_write_phy_reg(hw, ALX_MII_IER, 0);
is skipped in alx_select_powersaving_speed().
However, it should still be done in alx_pre_suspend()
when WOL is disabled.  But it might be we need to move
the alx_clear_phy_intr() in __alx_shutdown() down.

I'm also not sure why alx_get_phy_link() reads MII_BMSR
twice, but the old driver from compat-wireless-3.6.8-1-snpc
also does it, it seems deliberate.

Yeah, all this looks like magic to me...

quoted

Unless someone beats me to it I will try to debug
on the weekend.

The whole PCI(e) stuff there looks pretty odd to me and I can't seem to
fix it right now, so I'm removing it. It really looks like all the
wakeup_enable() etc. is in the wrong places? Hmmm. Well the code is
still in the history ;-)

Personally I don't care about WOL, no problem if you remove it.

So far I didn't have time to test (planning to do it tomorrow),
but looking at the code in __alx_shutdown(), I think
__alx_stop() tears down the link (via alx_reset_mac()).
Thus alx_select_powersaving_speed() would need to be
split in two functions, one looking at the current link,
called before __alx_stop(), and one called after it
to re-enable the link.  And alx_clear_phy_intr() needs to be moved
after alx_pre_suspend().  And alx_config_wol() should be
before alx_pre_suspend() according to the alx_pre_suspend() comment.
And alx_select_powersaving_speed() should not be called
when WOL is disabled.


Thanks,
Johannes