> Yes, but you're forgetting that this bug fix really needs to go into the
> stable kernels. You need one small patch (that doesn't change the
> current USB PCI interrupt allocation for other xHCI hosts) to fix the
> problem with the Panther Point xHCI host issue, and a second bigger
> patch to just skip the legacy interrupt PCI allocation all together.
>
> > > Let's figure out internally which BIOS versions actually have this
> > > problem, and how to differentiate between the two.
> >
> > Do you mean we can ignore such BIOS issue?
> > If you can skip such problem in BIOS and make HCD with MSI work like
> > Windows, why not? And guess there always have BIOS engineers forgetting
> > do test for Linux.
>
> Again, for the stable kernel reasons. I'm not really comfortable with
> changing the PCI interrupt allocation in the stable kernel trees for
> other xHCI hosts.
Thanks, Sarah. I understand your concern now.
Maybe checking MSI first in usb_hcd_request_irqs is your favorite too,
right?
I take a quick glance on the code. May we have 2 ways to implement it.
A, move current xhci code and a few MSI related objects into usb_hcd
core code and structure. the skeleton like below.
@@ -2340,12 +2340,25 @@ int usb_hcd_is_primary_hcd(struct usb_hcd *hcd)
return hcd == hcd->primary_hcd;
}
EXPORT_SYMBOL_GPL(usb_hcd_is_primary_hcd);
+static int hcd_setup_msix();
+static int hcd_setup_msi();
+static int hcd_try_enable_msi(struct usb_hdcd *hcd);
+
static int usb_hcd_request_irqs(struct usb_hcd *hcd,
unsigned int irqnum, unsigned long irqflags)
{
int retval;
+ /* only PCI HCD enabled HCD_MSI_FIRST */
+ if (hcd->driver->flags & | HCD_MSI_FIRST) {
+ /* checking driver quriks, like xhci */
+ ret = hcd_try_enable_msi(hcd);
+ if (ret)
+ return ret;
+ }
if (hcd->driver->irq) {
/* IRQF_DISABLED doesn't work as advertised when used together
B, left xhci code into it's place, and let different driver own self msi
enabling function, like this:
@@ -2346,6 +2346,17 @@ static int usb_hcd_request_irqs(struct usb_hcd *hcd,
{
int retval;
+ /* only PCI HCD enabled HCD_MSI_FIRST */
+ if (hcd->driver->flags & | HCD_MSI_FIRST) {
+ switch (driver->mode) {
+ xhci:
+ ret = xhci_try_enable_msi(hcd);
+ ehci:
+ ret = xhci_try_enable_msi(hcd);
+
+ if (ret)
+ return ret;
+ }
if (hcd->driver->irq) {
I believe the plan A is more common and clear. Any comments for this?
--
To unsubscribe from this list: send the line "unsubscribe linux-usb" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
