Re: [PATCH v8 4/5] PCI: andes: Add Andes QiLai SoC PCIe host driver support

[Randolph Lin <randolph@andestech.com>]

Hello Bjorn,
On Tue, Oct 21, 2025 at 12:05:16PM -0500, Bjorn Helgaas wrote:
> [EXTERNAL MAIL]
> 
> On Tue, Oct 14, 2025 at 08:03:48PM +0800, Randolph Lin wrote:
> > Add driver support for DesignWare based PCIe controller in Andes
> > QiLai SoC. The driver only supports the Root Complex mode.
> 
> > + * Setup the Qilai PCIe IOCP (IO Coherence Port) Read/Write Behaviors to the
> > + * Write-Back, Read and Write Allocate mode.
> 
> s/Setup/Set up/
> s/Qilai/QiLai/
> 

ok

> > + * The QiLai SoC PCIe controller's outbound iATU region supports
> > + * a maximum size of SZ_4G - 1. To prevent programming failures,
> > + * only consider bridge->windows with sizes within this limit.
> > + *
> > + * To ensure compatibility with most endpoint devices, at least
> > + * one memory region must be mapped within the 32-bits address space.
> > + */
> > +static int qilai_pcie_host_fix_ob_iatu_count(struct dw_pcie_rp *pp)
> > +{
> > +     struct dw_pcie *pci = to_dw_pcie_from_pp(pp);
> > +     struct device *dev = pci->dev;
> > +     struct resource_entry *entry;
> > +     /* Reserved 1 ob iATU for config space */
> > +     int count = 1;
> > +     bool ranges_32bits = false;
> > +     u64 pci_addr;
> > +     u64 size;
> > +
> > +     resource_list_for_each_entry(entry, &pp->bridge->windows) {
> > +             if (resource_type(entry->res) != IORESOURCE_MEM)
> > +                     continue;
> > +
> > +             size = resource_size(entry->res);
> > +             if (size < SZ_4G)
> > +                     count++;
> > +
> > +             pci_addr = entry->res->start - entry->offset;
> > +             if (pci_addr < SZ_4G)
> > +                     ranges_32bits = true;
> > +     }
> > +
> > +     if (!ranges_32bits) {
> > +             dev_err(dev, "Bridge window must contain 32-bits address\n");
> > +             return -EINVAL;
> 
> Is this really a PCI host controller driver probe failure?  I assume
> there are devices that only have 64-bit BARs and could work fine
> without a 32-bit window?
> 
> If a device requires a 32-bit BAR, and the PCI core can't assign such
> an address, and gracefully decline to enable a device where we
> couldn't assign the BAR, I think that would be preferable and would
> identify the specific device that doesn't work.
>

I have a clear understanding of the meaning behind this.
However, based on reviewer Niklas's suggestion, I have decided to use patch [1].
As a result, the function that adjusts the number of ib/ob window is no longer needed.

[1]: https://lore.kernel.org/linux-pci/aPDObXsvMoz1OYso@ryzen/T/#m11c3d95215982411d0bbd36940e70122b70ae820

> > +     }
> > +
> > +     pci->num_ob_windows = count;
> > +
> > +     return 0;
> > +}
> 
Sincerely,
Randolph Lin