Hi Bjorn,
On Fri, Feb 14, 2014 at 9:31 AM, Bjorn Helgaas <bhelgaas@xxxxxxxxxx> wrote:
> On Fri, Feb 14, 2014 at 12:21 AM, Izumi, Taku <izumi.taku@xxxxxxxxxxxxxx> wrote:
>>> Hmm... I see that the Slot has a power controller. Which means that the power to the slot shall not be turned on automatically
>>> (by HW) when the card is plugged in. Also meaning that the link will not come up automatically - so this does not seem
>>> like the Link state based hotplug kicking in.
>>>
>>> What I suspect is this one:
>>>
>>> f02d1843d83b "PCI: pciehp: Remove surprise bit checks"
>>
>> You are right.
>> In case of omitting comit-f02d1843, it worked as expected.
>> Slot power doesn't become ON automatically when PCIe card is inserted.
>
> OK, I dropped f02d1843d83b ("PCI: pciehp: Remove surprise bit
> checks").
I think part of this commit was still good (The part that drops the
surprise check when a card is yanked out). That is because when a card
is yanked out, it shouldn't matter whether the surprise bit is set or
not - its gotta go.
Functionally, that scenario is already covered by my patches (because
yanking out a card will make the link go down, hence kicking off link
state based unplug) - thus no functional change. But just thought I'll
mention since you were already at this cleanup phase. (If you agree, I
can send a separate clean patch or you can use Takashi's one).
> Thanks for testing, everybody!
>
> I think our handling of slot capabilities and control is a bit
> haphazard. It seems like we're mostly responding to things that
> break, and we don't really have a coherent explanation of how things
> *should* work in different configurations. I think it would be good
> if somebody wrote up something for Documentation/PCI, with references
> to the relevant specifications, that describes how we think things
> should work.
Sure, I think I'll take a stab.
Thanks,
Rajat
> For example, I think we have at least four models:
>
> 1) ExpressCard
> 2) Slots with attention button
> 3) Slots with no button where we expect a software UI, e.g., Taku's box
> 4) Devices with no actual slot, no button, etc., e.g., Rajat's system
>
> We handle these slightly differently, implicitly relying on various
> capability tests scattered throughout the code. I think we should be
> able to come up with a scheme that would make this more explicit and
> make pciehp simpler and more consistent.
>
> Bjorn
--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
