Re: [PATCH v2] PCI: Batch BAR sizing operations

[David Laight <david.laight.linux@gmail.com>]

On Mon, 10 Feb 2025 13:48:48 -0700
Alex Williamson <alex.williamson@redhat.com> wrote:

> On Mon, 10 Feb 2025 21:08:19 +0100
> Oleg Nesterov <oleg@redhat.com> wrote:
> 
> > On 02/10, Oleg Nesterov wrote:  
> > >
> > > On 02/10, Alex Williamson wrote:    
> > > >
> > > > --- a/drivers/pci/probe.c
> > > > +++ b/drivers/pci/probe.c
> > > > @@ -345,7 +345,8 @@ static void pci_read_bases(struct pci_dev *dev, unsigned int howmany, int rom)

You probably need __always_inline to have any chance of a compile-time
test of howmany.

> > > >  	unsigned int pos, reg;
> > > >  	u16 orig_cmd;
> > > >
> > > > -	BUILD_BUG_ON(howmany > PCI_STD_NUM_BARS);
> > > > +	if (__builtin_constant_p(howmany))
> > > > +		BUILD_BUG_ON(howmany > PCI_STD_NUM_BARS);    
> > >
> > > Or just
> > >
> > > 	BUILD_BUG_ON(__builtin_constant_p(howmany) && howmany > PCI_STD_NUM_BARS);

	statically_true(howmany > PCI_STD_NUM_BARS)

> > >
> > > I dunno... Works for me in any case.
> > >
> > > I don't know if this is "right" solution though,    
> > 
> > I mean, atm I simply don't know if with the newer compiler
> > __builtin_constant_p(howmany) will be true in this case.  
> 
> That I can confirm.  With gcc 14.2.1, if I change the call in
> pci_setup_device() to use 7 rather than 6 for howmany, the build fails.
> If you confirm this resolves the build on older gcc, it seems like a
> valid fix.  I installed a VM with Fedora 23, which uses gcc 3.5.1, but
> I haven't yet been able to successfully build any kernel with it.  I'll
> post a patch if you want to provide a Tested-by.  Thanks,

Far too old a version to be supported.

	David

> 
> Alex
> 
>