Re: Info: mapping multiple BARs. Your kernel is fine.

From: "Rafael J. Wysocki" <rjw@rjwysocki.net>
To: Borislav Petkov <bp@alien8.de>
Cc: Stephane Eranian <eranian@google.com>,
	lkml <linux-kernel@vger.kernel.org>, x86 <x86@kernel.org>,
	Bjorn Helgaas <bhelgaas@google.com>,
	Linux PCI <linux-pci@vger.kernel.org>,
	ACPI Devel Maling List <linux-acpi@vger.kernel.org>,
	Zhang Rui <rui.zhang@intel.com>, Yinghai Lu <yinghai@kernel.org>,
	"H. Peter Anvin" <hpa@zytor.com>
Subject: Re: Info: mapping multiple BARs. Your kernel is fine.
Date: Mon, 17 Mar 2014 01:23:37 +0100	[thread overview]
Message-ID: <6878235.CLHeQJJUpt@vostro.rjw.lan> (raw)
In-Reply-To: <1425376.F2qgSAxAbW@vostro.rjw.lan>

On Monday, March 17, 2014 01:09:39 AM Rafael J. Wysocki wrote:
> On Sunday, March 16, 2014 02:08:16 PM Stephane Eranian wrote:
> > Rafael,
> > 
> > Thanks for the analysis.
> > 
> > On Sun, Mar 16, 2014 at 12:55 PM, Borislav Petkov <bp@alien8.de> wrote:
> > > On Sat, Mar 15, 2014 at 03:15:04PM +0100, Rafael J. Wysocki wrote:
> > >> I've just gone throught this.
> > >
> > > Thanks.
> > >
> > >> So the problem is that we have the PNP "system" driver whose only purpose seems
> > >> to be to reserve system resources so that the PCI layer doesn't assign them to
> > >> new devices on hotplug (disclaimer: I didn't invent it, I only read the code and
> > >> comments in there).
> > >>
> > >> It does that for ACPI device objects having the "PNP0C02" and "PNP0C01" IDs.
> > >
> > > Right, pnp 00:01 is PNP0C02.
> > >
> > >> Apparently, snb_uncore_imc_init_box() steps on a range already reserved by that
> > >> driver on your box.  And this doesn't seem to be a coincidence, because the ACPI
> > >> device object in question probably *does* correspond to the memory controller
> > >> that the uncore driver attempts to use.
> > >>
> > >> I'm not sure how to address that right now to be honest.  Arguably, the PNP
> > >> "system" driver should be replaced with something saner, but still the
> > >> resources it claims need to be kept out of reach of the PCI's resource
> > >> allocation code.
> > >
> > > Well, I'm only conjecturing here but there should be a way for the
> > > uncore code to tell the PNP "system" driver to free this resource
> > > because uncore is going to use it now. Or something to that effect.
> > >
> > I agree. The snb_uncore_imc() is making real (good) use of the device.
> > It needs to own it. So we need a way to free the resource from the PNP
> > system or  a way to tell PNP need to grab it on systems with the
> > snb_uncore_imc() support. Does that kind of API exist?
> > 
> > Where do I look to prevent PNP from grabbing the IMC?
> 
> drivers/pnp/system.c is the driver in question and system_pnp_probe() makes
> the reservations via reserve_resources_of_dev(), so you'd need to modify that.
> 
> I'm not sure what's the right way to go here, though.

Boris, can you please sent the acpidump output from that machine?

-- 
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.