Re: [PATCH v6 24/24] PCI/sysfs: Limit BAR resize attribute scope to platforms with PCI mmap

[Bjorn Helgaas <helgaas@kernel.org>]

On Thu, Apr 30, 2026 at 11:51:28AM +0900, Krzysztof Wilczyński wrote:
> Hello,
> 
> > The only platform without these aforementioned defines is Alpha, which is
> > conventional PCI only and cannot have ReBAR.  So this guard removes dead
> > sysfs code on platforms where it can never be executed.
> 
> Having a closer look:
> 
>   resource_resize_attr_is_visible()
>     pci_rebar_get_current_size()      <- returns -ENOTSUPP, so is_visible callback returns 0
>       pci_rebar_find_pos()
>         pos = pdev->rebar_cap         <- set to 0 on a conventional PCI
>         if (!pos)
>           return -ENOTSUPP            <- no ReBAR support
> 
> The pdev->rebar_cap is set during PCI enumeration:
> 
>   pci_init_capabilities()
>     pci_rebar_init()
>       pdev->rebar_cap = pci_find_ext_capability()
>         pci_find_next_ext_capability()
>           if (dev->cfg_size <= PCI_CFG_SPACE_SIZE)
>             return 0;                 <- dev->cfg_size set to 256 here for conventional PCI
> 
> The PCI_CFG_SPACE_SIZE is 256 here.
> 
> When the platform has support for PCI Express, the dev->cfg_size is then
> set to 4096.  On an architecture that supports conventional PCI only (such
> as Alpha), the pdev->rebar_cap will be set to 0, the is_visible callback
> will then return 0, and the resize sysfs attribute is never created, as
> such, the __resource_resize_store() callback will never be executed.
> 
> To make the connetion here to the #ifdef guards:
> 
> For the pci_rebar_get_current_size() to return >= 0, the device needs
> PCI Express and extended configuration space support.  As of today,
> every architecture with PCI Express support defines HAVE_PCI_MMAP or
> ARCH_GENERIC_PCI_MMAP_RESOURCE.
> 
> I hope the reasoning here works.

We're talking about this #ifdef:

  +#if defined(HAVE_PCI_MMAP) || defined(ARCH_GENERIC_PCI_MMAP_RESOURCE)
   static ssize_t __resource_resize_show(struct device *dev, int n, char *buf)
   ...
  +#endif

I follow the reasoning now but by next week I won't, so I think it
requires too much background knowledge.  Future changes involving
HAVE_PCI_MMAP or ARCH_GENERIC_PCI_MMAP_RESOURCE could easily break
this.

IIUC it's basically doing what "#ifdef CONFIG_PCI_REBAR" or even
"#ifdef CONFIG_PCI_EXPRESS" would do, if we had such a thing.

How terrible would it be if we just accepted this dead code on Alpha?