Re: [PATCH] x86: fix: make PCI ECS for AMD CPUs hotplug capable

[Robert Richter <robert.richter@amd.com>]

On 22.08.08 12:10:04, Yinghai Lu wrote:
> On Fri, Aug 22, 2008 at 12:07 PM, Robert Richter <robert.richter@amd.com> wrote:
> > On 22.08.08 11:56:08, Yinghai Lu wrote:
> >
> > [...]
> >
> >> > +static int __init pci_io_ecs_init(void)
> >> > +{
> >> > +       int cpu;
> >> > +
> >> >        /* assume all cpus from fam10h have IO ECS */
> >> >         if (boot_cpu_data.x86 < 0x10)
> >> >                return 0;
> >> > -       on_each_cpu(enable_pci_io_ecs_per_cpu, NULL, 1);
> >> > +
> >> > +       register_cpu_notifier(&amd_cpu_notifier);
> >> > +       for_each_online_cpu(cpu)
> >> > +               amd_cpu_notify(&amd_cpu_notifier, (unsigned long)CPU_ONLINE,
> >> > +                              (void *)(long)cpu);
> >>
> >> wonder if those two lines should be reversed.
> >
> > Do you mean setting PCI_HAS_IO_ECS before for_each_online_cpu(...)?
> > PCI_HAS_IO_ECS will be used only in pci_direct_init(). PCI is
> > initialized in a later init stage, so it doesn't matter. My intention
> > was to set the bit after the setup of all online cpus is finished.
> 
> 
> ==>
> +       for_each_online_cpu(cpu)
> +               amd_cpu_notify(&amd_cpu_notifier, (unsigned long)CPU_ONLINE,
> +                              (void *)(long)cpu);
> +       register_cpu_notifier(&amd_cpu_notifier);

Hmm, most code in the kernel registers first. Theoretical, with this
code, if a cpu would go online in between, it wouldn't be initialized.

-Robert

> 
> YH
> 

-- 
Advanced Micro Devices, Inc.
Operating System Research Center
email: robert.richter@amd.com