Re: [PATCH v2 1/3] MIPS: ralink: Define PCI_IOBASE

[Sergio Paracuellos <sergio.paracuellos@gmail.com>]

Hi Thomas,

On Fri, Jul 30, 2021 at 10:30 AM Thomas Bogendoerfer
<tsbogend@alpha.franken.de> wrote:
>
> On Thu, Jul 29, 2021 at 01:21:45PM +0200, Sergio Paracuellos wrote:
> > Hi Thomas,
> >
> > On Thu, Jul 29, 2021 at 12:02 PM Thomas Bogendoerfer
> > <tsbogend@alpha.franken.de> wrote:
> > >
> > > On Mon, Jun 14, 2021 at 12:06:15PM +0200, Sergio Paracuellos wrote:
> > > > PCI_IOBASE is used to create VM maps for PCI I/O ports, it is
> > > > required by generic PCI drivers to make memory mapped I/O range
> > > > work. Hence define it for ralink architectures to be able to
> > > > avoid parsing manually IO ranges in PCI generic driver code.
> > > > Function 'plat_mem_setup' for ralink is using 'set_io_port_base'
> > > > call using '0xa0000000' as address, so use the same address in
> > > > the definition to align things.
> > > >
> > > > Signed-off-by: Sergio Paracuellos <sergio.paracuellos@gmail.com>
> > > > ---
> > > >  arch/mips/include/asm/mach-ralink/spaces.h | 10 ++++++++++
> > > >  1 file changed, 10 insertions(+)
> > > >  create mode 100644 arch/mips/include/asm/mach-ralink/spaces.h
> > > >
> > > > diff --git a/arch/mips/include/asm/mach-ralink/spaces.h b/arch/mips/include/asm/mach-ralink/spaces.h
> > > > new file mode 100644
> > > > index 000000000000..87d085c9ad61
> > > > --- /dev/null
> > > > +++ b/arch/mips/include/asm/mach-ralink/spaces.h
> > > > @@ -0,0 +1,10 @@
> > > > +/* SPDX-License-Identifier: GPL-2.0 */
> > > > +#ifndef __ASM_MACH_RALINK_SPACES_H_
> > > > +#define __ASM_MACH_RALINK_SPACES_H_
> > > > +
> > > > +#define PCI_IOBASE   _AC(0xa0000000, UL)
> > > > +#define PCI_IOSIZE   SZ_16M
> > > > +#define IO_SPACE_LIMIT       (PCI_IOSIZE - 1)
> > > > +
> > > > +#include <asm/mach-generic/spaces.h>
> > > > +#endif
> > >
> > > does this really work for you ? I tried the same trick for RB532
> > > and the generated IO addresses are wrong...
> >
> > I got pci io resources assigned without complaints from the pci core
> > code. I don't have real pci card that uses I/O bars but this is what I
> > see in the boot (I added some traces when I was testing this):
>
> resource handling works, but the addresses generated for IO access
> are wrong, because the iomap tries to ioremap it to a fixed
> virtual address (PCI_IOBASE), which can't work for KSEG1 addresses.
>
> > Is this wrong?
>
> to get it working this way, we would need to put PCI_IOBASE somewhere
> into KSEG2, which I don't like since it will create TLB entries for IO
> addresses, which most of the time isn't needed on MIPS because of
> access via KSEG1.
>
> I'd much prefer to make the devm_pci_remap_iospace() in drivers/pci/of.c
> optional. Something like this
>
> diff --git a/drivers/pci/of.c b/drivers/pci/of.c
> index a143b02b2dcd..657aef39bf63 100644
> --- a/drivers/pci/of.c
> +++ b/drivers/pci/of.c
> @@ -564,12 +564,14 @@ static int pci_parse_request_of_pci_ranges(struct device *dev,
>
>                 switch (resource_type(res)) {
>                 case IORESOURCE_IO:
> +#ifdef PCI_IOBASE
>                         err = devm_pci_remap_iospace(dev, res, iobase);
>                         if (err) {
>                                 dev_warn(dev, "error %d: failed to map resource %pR\n",
>                                          err, res);
>                                 resource_list_destroy_entry(win);
>                         }
> +#endif
>                         break;
>                 case IORESOURCE_MEM:
>                         res_valid |= !(res->flags & IORESOURCE_PREFETCH);
>
>
> This together with an increased IO space via
>
> #define IO_SPACE_LIMIT 0x1fffffff
>
> gives me a working PCI bus on the RB532.

BTW, I have tested your changes and they result in a no working pci
for mt7621. I get a resource collision error:

mt7621-pci 1e140000.pcie: resource collision: [io
0x1e160000-0x1e16ffff] conflicts with PCI IO [io  0x0000-0xffff]

My changes:
 - avoid PCI_IOBASE to be defined.
 - avoid map in pci_parse_request_of_pci_ranges
 - Change spaces.h to have the new IO_SPACE_LIMIT to 0x1fffffff

Am I missing something?

Thanks,
   Sergio Paracuellos
>
> No idea, if the patch would be accepted by the PCI maintainers.
>
> Thomas.
>
> --
> Crap can work. Given enough thrust pigs will fly, but it's not necessarily a
> good idea.                                                [ RFC1925, 2.3 ]