Re: [PATCH] acpica events: Call acpi_os_hotplug_execute on Ejection Requests

[Matthew Garrett <mjg@redhat.com>]

On Fri, Sep 23, 2011 at 09:29:19AM -0600, Bjorn Helgaas wrote:
> On Thu, Sep 22, 2011 at 7:13 PM, Matthew Garrett <mjg@redhat.com> wrote:
> > We're seeing an increasing number of machines that use acpi
> > notifications to trigger rescans of devices - for instance, SD readers
> > that only appear on the PCI bus once a card has been inserted. These
> > often fail to flag themselves as removable or ejectable, so the acpiphp
> > core never binds. I think we need to fix this pretty urgently, for PCI
> > if nothing else.
> 
> Can you elaborate on this a bit?  I certainly agree that Prarit's
> process hang needs to be fixed, but it sounds like you're talking
> about the larger issue of the ACPI core not supporting hotplug.

Yup.

> I think that since the core doesn't do hotplug, we can't deal with
> hot-add of any ACPI devices unless we have already loaded a driver
> that has hotplug support for that device.  And acpiphp is not really a
> "driver," in the sense that it doesn't bind to a specific PNP ID and
> there isn't a good way to autoload it.

Indeed. Worse, it only binds to devices that have _RMV or _EJ* methods.

> If we moved most of the notification parts of acpiphp into the core,
> maybe that would be a step towards handling these SD readers and
> similar devices?

I think so, yes.

-- 
Matthew Garrett | mjg59@srcf.ucam.org