[lkml]   [2000]   [Jul]   [6]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
    SubjectRe: 2.2.15 driver OK on UP - no ints on SMP (SOLVED)


    When we were putting SMP on NetWare, we discovered that using the IOAPIC
    hardware for UP systems was almost 4% faster than the 8259-PIC for LAN
    drivers (since it has a dedicated ICC bus to pass interrupt messages to
    a local APIC). Something to consider....



    Linus Torvalds wrote:
    > In article <>,
    > William Montgomery <> wrote:
    > >
    > >So how can my driver my driver work fine in UP and not in SMP when
    > >using the pci_read_config* calls?
    > On UP, we don't generally use the IO-APIC and it's more advanced irq
    > routing features. So on a regular UP compile, there will be a 1:1
    > relationship between what the device _thinks_ the irq is, and what it
    > truly is..
    > >My workaround is to use
    > > struct pci_dev *dev;
    > >and when pci_find_device() inits dev, I use dev->irq instead of
    > > pci_read_config_byte(dev, PCI_INTERRUPT_LINE, &irq);
    > >
    > >Can anyone explain this?
    > The above is not a work-around. It's the way you are supposed to work.
    > In fact, drivers should generally _never_ look at the "generic" part of
    > the PCI config space, because that part has been parsed for them by the
    > kernel, which does a whole lot better job of it than most drivers can
    > ever do. This includes more than just figuring out what the _real_
    > interrupt routing is. The generic PCI layer does a good job of figuring
    > out what resources the device has, of what type they are, and how large
    > they are. And it enables them correctly. Something that is usually a
    > lot more code than is good for you in a regular driver.
    > For example, on multi-bus systems (especially 64-bit sparcs with
    > multiple PCI buses mapped in strange ways), you should not do the config
    > byte read to figure out where your memory aperture is: you should use
    > the data in "struct pci_dev *dev" instead, notably you should use the
    > functions especially created for them: things like
    > start = pci_resource_start(dev,nr);
    > to get at the resource start address. Because again, as with the
    > interrupt number, the device itself does not know what the global system
    > mapping is (it so happens that on PC's it's a 1:1 mapping, but that's
    > not always the case).
    > Linus
    > -
    > To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    > the body of a message to
    > Please read the FAQ at

    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to
    Please read the FAQ at

     \ /
      Last update: 2005-03-22 13:57    [W:0.021 / U:222.884 seconds]
    ©2003-2017 Jasper Spaans. hosted at Digital OceanAdvertise on this site