| Message ID | 20090604233521.GA12900@ethanol (mailing list archive) |
|---|---|
| State | Not Applicable, archived |
| Headers | show |
On Thursday 04 June 2009 05:35:21 pm Alex Chiang wrote: > * Kenji Kaneshige <kaneshige.kenji@jp.fujitsu.com>: > > > > I have a concern about this change. > > > > The acpi_pci_irq_del_prt() against dev->bus removes not only > > the _PRT entries for PCI function corresponding to specified > > acpi_device, but also other _PRT entries for working PCI > > devices/functions on the same bus. As a result, interrupt > > initialization for those PCI functions would no longer work > > properly after that. > > > > So I think we should not call acpi_pci_irq_del_prt() against > > dev->bus. > > Thanks for the review. I agree with you. I agree that this respun version makes things more the way they were, so in that sense, it should do no harm. But I still have the niggling concern that .bind() adds _PRT info for non-bridges, and there's no corresponding removal. There should be some path that makes this more symmetric. > Here is a respun version of this patch. > > From: Alex Chiang <achiang@hp.com> > > ACPI: acpi_pci_unbind should clean up properly after acpi_pci_bind > > In acpi_pci_bind, we set device->ops.bind and device->ops.unbind, but > never clear them out. > > Cc: Bjorn Helgaas <bjorn.helgaas@hp.com> > Signed-off-by: Alex Chiang <achiang@hp.com> > --- > diff --git a/drivers/acpi/pci_bind.c b/drivers/acpi/pci_bind.c > index 62cb383..c9cc650 100644 > --- a/drivers/acpi/pci_bind.c > +++ b/drivers/acpi/pci_bind.c > @@ -109,11 +109,13 @@ static int acpi_pci_unbind(struct acpi_device *device) > struct pci_dev *dev; > > dev = acpi_get_pci_dev(device->handle); > - if (!dev) > + if (!dev || !dev->subordinate) > return 0; > > - if (dev->subordinate) > - acpi_pci_irq_del_prt(dev->subordinate); > + acpi_pci_irq_del_prt(dev->subordinate); > + > + device->ops.bind = NULL; > + device->ops.unbind = NULL; > > pci_dev_put(dev); > return 0; > -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
* Bjorn Helgaas <bjorn.helgaas@hp.com>: > On Thursday 04 June 2009 05:35:21 pm Alex Chiang wrote: > > * Kenji Kaneshige <kaneshige.kenji@jp.fujitsu.com>: > > > > > > I have a concern about this change. > > > > > > The acpi_pci_irq_del_prt() against dev->bus removes not only > > > the _PRT entries for PCI function corresponding to specified > > > acpi_device, but also other _PRT entries for working PCI > > > devices/functions on the same bus. As a result, interrupt > > > initialization for those PCI functions would no longer work > > > properly after that. > > > > > > So I think we should not call acpi_pci_irq_del_prt() against > > > dev->bus. > > > > Thanks for the review. I agree with you. > > I agree that this respun version makes things more the way they were, > so in that sense, it should do no harm. But I still have the niggling > concern that .bind() adds _PRT info for non-bridges, and there's no > corresponding removal. There should be some path that makes this > more symmetric. The comment in acpi_pci_bind doesn't seem to know if you can even have a _PRT for non-bridges. The spec (3.0b) says that _PRT is required for all root bridges, but doesn't mention anything about non-bridge devices. My gut feeling is the way to cure the symmetry is to remove the path in .bind() that adds _PRT for non-bridges, but I'm a little hesitant to do that without knowing for sure. I think for this patch, the Hippocratic approach is the right one for now. Thanks. /ac > > Here is a respun version of this patch. > > > > From: Alex Chiang <achiang@hp.com> > > > > ACPI: acpi_pci_unbind should clean up properly after acpi_pci_bind > > > > In acpi_pci_bind, we set device->ops.bind and device->ops.unbind, but > > never clear them out. > > > > Cc: Bjorn Helgaas <bjorn.helgaas@hp.com> > > Signed-off-by: Alex Chiang <achiang@hp.com> > > --- > > diff --git a/drivers/acpi/pci_bind.c b/drivers/acpi/pci_bind.c > > index 62cb383..c9cc650 100644 > > --- a/drivers/acpi/pci_bind.c > > +++ b/drivers/acpi/pci_bind.c > > @@ -109,11 +109,13 @@ static int acpi_pci_unbind(struct acpi_device *device) > > struct pci_dev *dev; > > > > dev = acpi_get_pci_dev(device->handle); > > - if (!dev) > > + if (!dev || !dev->subordinate) > > return 0; > > > > - if (dev->subordinate) > > - acpi_pci_irq_del_prt(dev->subordinate); > > + acpi_pci_irq_del_prt(dev->subordinate); > > + > > + device->ops.bind = NULL; > > + device->ops.unbind = NULL; > > > > pci_dev_put(dev); > > return 0; > > > > -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Alex Chiang wrote: > * Kenji Kaneshige <kaneshige.kenji@jp.fujitsu.com>: >>> >> I have a concern about this change. >> >> The acpi_pci_irq_del_prt() against dev->bus removes not only >> the _PRT entries for PCI function corresponding to specified >> acpi_device, but also other _PRT entries for working PCI >> devices/functions on the same bus. As a result, interrupt >> initialization for those PCI functions would no longer work >> properly after that. >> >> So I think we should not call acpi_pci_irq_del_prt() against >> dev->bus. > > Thanks for the review. I agree with you. > > Here is a respun version of this patch. > > From: Alex Chiang <achiang@hp.com> > > ACPI: acpi_pci_unbind should clean up properly after acpi_pci_bind > > In acpi_pci_bind, we set device->ops.bind and device->ops.unbind, but > never clear them out. > > Cc: Bjorn Helgaas <bjorn.helgaas@hp.com> > Signed-off-by: Alex Chiang <achiang@hp.com> > --- > diff --git a/drivers/acpi/pci_bind.c b/drivers/acpi/pci_bind.c > index 62cb383..c9cc650 100644 > --- a/drivers/acpi/pci_bind.c > +++ b/drivers/acpi/pci_bind.c > @@ -109,11 +109,13 @@ static int acpi_pci_unbind(struct acpi_device *device) > struct pci_dev *dev; > > dev = acpi_get_pci_dev(device->handle); > - if (!dev) > + if (!dev || !dev->subordinate) > return 0; This would leak the pci_dev's refcount if dev != NULL && dev->subordinate == NULL. Thanks, Kenji Kaneshige > > - if (dev->subordinate) > - acpi_pci_irq_del_prt(dev->subordinate); > + acpi_pci_irq_del_prt(dev->subordinate); > + > + device->ops.bind = NULL; > + device->ops.unbind = NULL; > > pci_dev_put(dev); > return 0; > -- > To unsubscribe from this list: send the line "unsubscribe linux-acpi" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html > > -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
* Bjorn Helgaas <bjorn.helgaas@hp.com>: > On Thursday 04 June 2009 05:35:21 pm Alex Chiang wrote: > > * Kenji Kaneshige <kaneshige.kenji@jp.fujitsu.com>: > > > > > > I have a concern about this change. > > > > > > The acpi_pci_irq_del_prt() against dev->bus removes not only > > > the _PRT entries for PCI function corresponding to specified > > > acpi_device, but also other _PRT entries for working PCI > > > devices/functions on the same bus. As a result, interrupt > > > initialization for those PCI functions would no longer work > > > properly after that. > > > > > > So I think we should not call acpi_pci_irq_del_prt() against > > > dev->bus. > > > > Thanks for the review. I agree with you. > > I agree that this respun version makes things more the way they were, > so in that sense, it should do no harm. But I still have the niggling > concern that .bind() adds _PRT info for non-bridges, and there's no > corresponding removal. There should be some path that makes this > more symmetric. Hm, in another forum, you suggested that dynamic PRT lookups might be a solution, which I kinda like. So, the plan that I would prefer is: a) get this patchset in [and we 'do no harm' here so _hopefully_ aren't introducing regressions] b) work on dynamic PRT lookups in a future patchset. Thanks. /ac -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/drivers/acpi/pci_bind.c b/drivers/acpi/pci_bind.c index 62cb383..c9cc650 100644 --- a/drivers/acpi/pci_bind.c +++ b/drivers/acpi/pci_bind.c @@ -109,11 +109,13 @@ static int acpi_pci_unbind(struct acpi_device *device) struct pci_dev *dev; dev = acpi_get_pci_dev(device->handle); - if (!dev) + if (!dev || !dev->subordinate) return 0; - if (dev->subordinate) - acpi_pci_irq_del_prt(dev->subordinate); + acpi_pci_irq_del_prt(dev->subordinate); + + device->ops.bind = NULL; + device->ops.unbind = NULL; pci_dev_put(dev); return 0;