Re: [PATCH 2/4] MIPS: pci: Add driver for MT7621 PCIe controller

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

Hi Pali,

Thanks for your comments.

On Mon, May 31, 2021 at 3:14 PM Pali Rohár <pali@kernel.org> wrote:
>
> On Saturday 15 May 2021 14:40:53 Sergio Paracuellos wrote:
> > This patch adds a driver for the PCIe controller of MT7621 SoC.
> >
> > Signed-off-by: Sergio Paracuellos <sergio.paracuellos@gmail.com>
> > ---
> >  arch/mips/pci/Makefile     |   1 +
> >  arch/mips/pci/pci-mt7621.c | 624 +++++++++++++++++++++++++++++++++++++
> >  arch/mips/ralink/Kconfig   |   9 +-
> >  3 files changed, 633 insertions(+), 1 deletion(-)
> >  create mode 100644 arch/mips/pci/pci-mt7621.c
> >
> > diff --git a/arch/mips/pci/Makefile b/arch/mips/pci/Makefile
> > index f3eecc065e5c..178c550739c4 100644
> > --- a/arch/mips/pci/Makefile
> > +++ b/arch/mips/pci/Makefile
> > @@ -24,6 +24,7 @@ obj-$(CONFIG_PCI_AR2315)    += pci-ar2315.o
> >  obj-$(CONFIG_SOC_AR71XX)     += pci-ar71xx.o
> >  obj-$(CONFIG_PCI_AR724X)     += pci-ar724x.o
> >  obj-$(CONFIG_PCI_XTALK_BRIDGE)       += pci-xtalk-bridge.o
> > +obj-$(CONFIG_PCI_MT7621)     += pci-mt7621.o
> >  #
> >  # These are still pretty much in the old state, watch, go blind.
> >  #
> > diff --git a/arch/mips/pci/pci-mt7621.c b/arch/mips/pci/pci-mt7621.c
> > new file mode 100644
> > index 000000000000..fe1945819d25
> > --- /dev/null
> > +++ b/arch/mips/pci/pci-mt7621.c
> ...
> > +static int mt7621_pcie_enable_ports(struct mt7621_pcie *pcie)
> > +{
> > +     struct device *dev = pcie->dev;
> > +     struct mt7621_pcie_port *port;
> > +     u8 num_slots_enabled = 0;
> > +     u32 slot;
> > +     u32 val;
> > +     int err;
> > +
> > +     /* Setup MEMWIN and IOWIN */
> > +     pcie_write(pcie, 0xffffffff, RALINK_PCI_MEMBASE);
> > +     pcie_write(pcie, pcie->io.start, RALINK_PCI_IOBASE);
> > +
> > +     list_for_each_entry(port, &pcie->ports, list) {
> > +             if (port->enabled) {
> > +                     err = clk_prepare_enable(port->clk);
> > +                     if (err) {
> > +                             dev_err(dev, "enabling clk pcie%d\n", slot);
> > +                             return err;
> > +                     }
> > +
> > +                     mt7621_pcie_enable_port(port);
> > +                     dev_info(dev, "PCIE%d enabled\n", port->slot);
> > +                     num_slots_enabled++;
> > +             }
> > +     }
> > +
> > +     for (slot = 0; slot < num_slots_enabled; slot++) {
> > +             val = read_config(pcie, slot, PCI_COMMAND);
> > +             val |= PCI_COMMAND_MASTER;
> > +             write_config(pcie, slot, PCI_COMMAND, val);
>
> Hello! Is this part of code correct? Because it looks strange if PCIe
> controller driver automatically enables PCI bus mastering, prior device
> driver initialize itself.
>
> Moreover kernel has already function pci_set_master() for this purpose
> which is used by device drivers.
>
> So I think this code can confuse some device drivers...

I agree that we have pci_set_master() to be used in pci device driver
code. Original controller driver set this bit for enabled slots. Since
there is no documentation at all for the PCI in this SoC I have
maintained the setting in the driver in a cleaner way. See original
driver code and the setting here [0]. There is no other reason than
that. I am ok with removing this from here and testing with my two
devices that everything is still ok if having this setting in the pci
controller driver is a real problem.

[0]: https://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging.git/tree/drivers/staging/mt7621-pci/pci-mt7621.c?h=v4.18#n676

Best regards,
    Sergio Paracuellos
>
> > +             /* configure RC FTS number to 250 when it leaves L0s */
> > +             val = read_config(pcie, slot, PCIE_FTS_NUM);
> > +             val &= ~PCIE_FTS_NUM_MASK;
> > +             val |= PCIE_FTS_NUM_L0(0x50);
> > +             write_config(pcie, slot, PCIE_FTS_NUM, val);
> > +     }
> > +
> > +     return 0;
> > +}