Re: Regression: 442ec4c04d1: PCI: dwc: all: Split struct pcie_port into host-only and core structures

[Lorenzo Pieralisi]

On Mon, May 08, 2017 at 10:33:25PM +0200, Peter Senna Tschudin wrote:
> On Mon, May 08, 2017 at 06:55:36PM +0100, Lorenzo Pieralisi wrote:
> > On Mon, May 08, 2017 at 06:34:08PM +0200, Lucas Stach wrote:
> > > Am Montag, den 08.05.2017, 17:02 +0200 schrieb Peter Senna Tschudin:
> > > > Hello Kishon,
> > > > 
> > > > Our iMX6 hardware (imx6q-b850v3.dts) is not booting with latest
> > > > linux-next and I could bisect until:
> > > > 
> > > > commit 442ec4c04d1235f8c664a74004dae54a7a574d18
> > > > Author: Kishon Vijay Abraham I <kishon@xxxxxx>
> > > > Date:   Wed Feb 15 18:48:14 2017 +0530
> > > > 
> > > >     PCI: dwc: all: Split struct pcie_port into host-only and core structures
> > > > 
> > > > Which seem to be causing our issues. Our device (imx6q-b850v3.dts) boots
> > > > fine with 4.10, and also boots if we disable pcie with:
> > > 
> > > Are you sure about this bisect? It seems more likely that the secondary
> > > bus scanning goes wrong. It seems Lorenzo Pieralisi's cfg space
> > > non-posting patches landed in next, so the root cause of this might just
> > > be the driver hooking the wrong fault code now.
> > > 
> > > Async and synchronous aborts are different fault codes on ARM. Can you
> > > see of removing the "16 +" in hook_fault_code() in
> > > drivers/pci/dwc/pci-imx6.c fixes this issue?
> > 
> > You can also s/pci_remap_cfgspace/ioremap in pcie-designware-host.c
> > to see if my patches are the issue, it is quite a mechanical change.
> 
> No change. Exactly same issue, but I noticed that
> devm_pci_remap_cfgspace() is called twice for the same device, can this
> be a problem?

No and anyway, given that you basically tested with pci_remap_cfgspace()
reverted I do not see how it may affect your platform but that's not
what you are asking, so yes, it is safe to call it for multiple
resources in a given device, think of it as ioremap for configuration
space resources.

I still do not understand if you manage to bisect and fix the issue,
PCI tree was merged in the mainline last night so you could give it
a shot, it may be easier to bisect than -next.

Thanks !
Lorenzo

> ...
> [    0.505069] OF: PCI: host bridge /soc/pcie@1ffc000 ranges:
> [    0.510586] OF: PCI:    IO 0x01f80000..0x01f8ffff -> 0x00000000
> [    0.516474] OF: PCI:   MEM 0x01000000..0x01efffff -> 0x01000000
> [    0.522420] imx6q-pcie 1ffc000.pcie: devm_pci_remap_cfgspace()
> [    0.528656] imx6q-pcie 1ffc000.pcie: devm_pci_remap_cfgspace()
> ...