Stefan Sperling <stsp <at> stsp.name> writes:
>
> Can you please clarify with which code exactly you're seeing this problem?
> What are you running?
> Does it happens with the if_xl_pci.c hunk of my proposed diff, or without it?
> Or does it always happen even in plain -current?
>
ok, of course.
1) everything works fine with your latest patches, if i manually do an
"ifconfig xl0 wol".
2) now i'd like to set the interface stuff in /etc/hostname.xl0. therefore,
i go with "inet 192.168.0.2 255.255.255.0 192.168.0.255 wol".
(re)booting then results in those error messages i presented before
now what i found out so far is, that the problem seems to be with the
"pci_conf_write" call inside "xl_pci_power" in if_xl_pci.c
both the XL_PME_EN and XL_PSTATE_D3 state are to be set. i narrowed it down to
that call by try and error. so when you disable "pci_conf_write", the error
disappears (of course no WOL then).
the problem is setting the XL_PSTATE_D3 bit.
it think, the error appears, because "xl_stop" is called inside "xl_init"
(both in xl.c). and "xl_init" is called _before_ "xl_attach" inside
"xl_pci_attach" (if_xl_pci.c), so the "xl_wol(ifp, 0)" of "xl_attach" call to
clear WOL flag doesn't work. somehow, this flag is set, and that's why it goes
down all the chain and "pci_conf_write" is called. however, since the system is
trying to bring the IF up, there's a conflict b/w bringing up and D3 power
state.
at least that's what i think...
the error appears with your patches only. that is, because -current
(checkout from yesterday) still has
if_xl_pci.c:
> command = pci_conf_read(pc, pa->pa_tag, XL_PCI_PWRMGMTCAP);
> if (command & XL_PME_CAP_D3_HOT) {
> sc->wol_power = xl_pci_wol_power;
> sc->wol_power_arg = psc;
> }
so it never comes to register "xl_pci_wol_power".
> This doesn't sound like a good idea.
> The PCI-specific layer of XL and the low-level layer in ic/ have to
> cooperate. Else you'd have to call PCI-specific routines from the code
> in ic/ which harms use of this driver with other buses.
>
>
i see your point. just got the idea from FreeBSD, it seems less confusing. :)
hope, this clears up the problem i see?
cheers
