Hi Andiry,
Now that I have a bit of time on my hands (before the next crisis
comes), I finally took at look at your multiple event rings patchset.
The biggest issue is that when an interrupt comes in for any MSI-X
interrupter, you try to service the event rings for all interrupters.
I think you would have fewer cache misses if you only serviced the
interrupter that is tied to that particular CPU.
Matthew has a couple patches to make it easier to only service the MSI-X
vector that caused the xHCI driver interrupt. He'll be sending those
shortly. Can you please review them and look at rebasing this patchset
on top of those and the usb-next tree? I'd like to try to get this
patchset into 2.6.41/3.1.
Sarah Sharp
On Thu, Feb 17, 2011 at 06:07:16PM +0800, Andiry Xu wrote:
> @@ -2088,9 +2092,9 @@ irqreturn_t xhci_irq(struct usb_hcd *hcd)]
...
> @@ -2126,53 +2121,75 @@ hw_died:
> */
> status |= STS_EINT;
> xhci_writel(xhci, status, &xhci->op_regs->status);
> - /* FIXME when MSI-X is supported and there are multiple vectors */
> - /* Clear the MSI-X event interrupt status */
>
> - if (hcd->irq != -1) {
> - u32 irq_pending;
> - /* Acknowledge the PCI interrupt */
> - irq_pending = xhci_readl(xhci, &xhci->ir_set->irq_pending);
> - irq_pending |= 0x3;
> - xhci_writel(xhci, irq_pending, &xhci->ir_set->irq_pending);
> - }
> + /* Check all the interrupters */
> + for (i = 0; i < xhci->intr_num; i++) {
> + xhci->ir_set = (void *) &xhci->run_regs->ir_set[i];
> + if (hcd->irq != -1) {
> + u32 irq_pending;
> + /* Acknowledge the PCI interrupt */
> + irq_pending =
> + xhci_readl(xhci, &xhci->ir_set->irq_pending);
> + /* If the interrupt is not from this interrupter,
> + * continue immediately
> + */
> + if (!(irq_pending & 0x1))
> + continue;
> + irq_pending |= 0x3;
> + xhci_writel(xhci, irq_pending,
> + &xhci->ir_set->irq_pending);
> + }
>
--
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
