On Thu, Sep 28, 2017 at 08:57:28AM -0700, Ard Biesheuvel wrote: > On 28 September 2017 at 02:03, Will Deacon <will.deacon@xxxxxxx> wrote: > > On Tue, Sep 26, 2017 at 12:32:00PM -0500, Bjorn Helgaas wrote: > >> [+cc Will] > >> > >> On Mon, Aug 28, 2017 at 07:04:36PM +0100, Ard Biesheuvel wrote: > >> > Some implementations of the Synopsys Designware PCIe controller implement > >> > a so-called ECAM shift mode, which allows a static memory window to be > >> > configured that covers the configuration space of the entire bus range. > >> > > >> > If the firmware performs all the low level configuration that is required > >> > to expose this controller in a fully ECAM compatible manner, we can > >> > simply describe it as "pci-host-ecam-generic" and be done with it. > >> > However, it appears that in some cases (one of which is the Armada 80x0), > >> > the IP is synthesized with an ATU window size that does not allow the > >> > first bus to be mapped in a way that prevents the device on the > >> > downstream port from appearing more than once. > >> > > >> > So implement a driver that relies on the firmware to perform all low > >> > level initialization, and drives the controller in ECAM mode, but > >> > overrides the config space accessors to take the above quirk into > >> > account. > >> > > >> > Note that, unlike most drivers for this IP, this driver does not expose > >> > a fake bridge device at B/D/F 00:00.0. There is no point in doing so, > >> > given that this is not a true bridge, and does not require any windows > >> > to be configured in order for the downstream device to operate correctly. > >> > Omitting it also prevents the PCI resource allocation routines from > >> > handing out BAR space to it unnecessarily. > >> > >> This is a tangent, but does this mean the other drivers do not need to > >> expose a fake 00:00.0 device either? > >> > >> s/Designware/DesignWare/ in comments, changelogs, Kconfig text, etc. > >> > >> > Cc: Bjorn Helgaas <bhelgaas@xxxxxxxxxx> > >> > Cc: Jingoo Han <jingoohan1@xxxxxxxxx> > >> > Cc: Joao Pinto <Joao.Pinto@xxxxxxxxxxxx> > >> > Signed-off-by: Ard Biesheuvel <ard.biesheuvel@xxxxxxxxxx> > >> > --- > >> > drivers/pci/dwc/Kconfig | 11 +++ > >> > drivers/pci/dwc/Makefile | 1 + > >> > drivers/pci/dwc/pcie-designware-ecam.c | 77 ++++++++++++++++++++ > >> > >> This really doesn't have any DesignWare specifics in it, and it seems > >> more related to drivers/pci/host/pci-host-generic.c than to anything > >> in drivers/pci/dwc. Maybe it should be > >> drivers/pci/host/pci-host-generic-quirks.c or something? That's > >> unwieldy, I admit. > >> > >> Putting it in pci/dwc would make Jingoo and Joao the default > >> maintainers; I don't know how they feel about that. We would probably > >> have to tweak MAINTAINERS if we *didn't* put it in pci/dwc. > >> > >> Any thoughts on this, Will? > > > > The idea of a "generic quirk" makes me smile, I must admit :) > > > > I think there are two options: > > > > 1. Use the full DWC driver, and don't rely on firmware > > -or- > > 2. Rely on firmware, but teach pci-host-generic to deal with the funny > > config space > > > > For (2), we probably want to describe this as generically as possible > > in case some other SoCs run into the same problem. > > > > I take it this implies a DT property. I could add one that consists of > an array of val/mask tuples or base/size tuples that allow us to > disable arbitrary subregions of the config space. I could also add a > simple boolean property that implements this exact quirk. Do you have > any preference? I'd say either a boolean property or a new compatible string. I think Rob prefers the latter, from what he said recently on an SMMU thread. Will -- To unsubscribe from this list: send the line "unsubscribe devicetree" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
