xhci might run with MSI/MSI-X only, with no support for legacy
interrupts. On these devices the request_irq() call in usb_add_hcd()
will fail, causing the entire device to fail.
For xhci this is especially painful as the driver will enable
interrupts during xhci_run(), so the initial call to request_irq()
is not required anyway.
This patch adds a flag 'msix_supported' to struct usb_hcd, which
will cause usb_add_hcd() to skip interrupt setup. This flag is
set in xhci-pci, so the registration will skip the first request_irq()
and can proceed.
Cc: Bjorn Helgaas <[email protected]>
Cc: Oliver Neukum <[email protected]>
Cc: Thomas Renninger <[email protected]>
Cc: Yinghai Lu <[email protected]>
Cc: Frederik Himpe <[email protected]>
Cc: David Haerdeman <[email protected]>
Signed-off-by: Hannes Reinecke <[email protected]>
diff --git a/drivers/usb/core/hcd.c b/drivers/usb/core/hcd.c
index 8e64adf..e583444 100644
--- a/drivers/usb/core/hcd.c
+++ b/drivers/usb/core/hcd.c
@@ -2536,7 +2536,7 @@ int usb_add_hcd(struct usb_hcd *hcd,
/* enable irqs just before we start the controller,
* if the BIOS provides legacy PCI irqs.
*/
- if (usb_hcd_is_primary_hcd(hcd) && irqnum) {
+ if (usb_hcd_is_primary_hcd(hcd) && irqnum && !hcd->msix_supported) {
retval = usb_hcd_request_irqs(hcd, irqnum, irqflags);
if (retval)
goto err_request_irq;
diff --git a/drivers/usb/host/xhci-pci.c b/drivers/usb/host/xhci-pci.c
index af259e0..a08c0ee 100644
--- a/drivers/usb/host/xhci-pci.c
+++ b/drivers/usb/host/xhci-pci.c
@@ -183,6 +183,12 @@ static int xhci_pci_probe(struct pci_dev *dev, const struct pci_device_id *id)
*/
*((struct xhci_hcd **) xhci->shared_hcd->hcd_priv) = xhci;
+ /*
+ * Interrupts are setup during xhci_run(), so do not try
+ * to request an interrupt during hcd init.
+ */
+ hcd->msix_supported = 1;
+
retval = usb_add_hcd(xhci->shared_hcd, dev->irq,
IRQF_SHARED);
if (retval)
diff --git a/include/linux/usb/hcd.h b/include/linux/usb/hcd.h
index 0a78df5..eb6bf10 100644
--- a/include/linux/usb/hcd.h
+++ b/include/linux/usb/hcd.h
@@ -124,6 +124,7 @@ struct usb_hcd {
/* Flags that get set only during HCD registration or removal. */
unsigned rh_registered:1;/* is root hub registered? */
unsigned rh_pollable:1; /* may we poll the root hub? */
+ unsigned msix_supported:1; /* driver supports MSI-X */
unsigned msix_enabled:1; /* driver has MSI-X enabled? */
/* The next flag is a stopgap, to be removed when all the HCDs
On Fri, Mar 01, 2013 at 09:52:54AM +0100, Hannes Reinecke wrote:
>xhci might run with MSI/MSI-X only, with no support for legacy
>interrupts. On these devices the request_irq() call in usb_add_hcd()
>will fail, causing the entire device to fail.
>For xhci this is especially painful as the driver will enable
>interrupts during xhci_run(), so the initial call to request_irq()
>is not required anyway.
>
>This patch adds a flag 'msix_supported' to struct usb_hcd, which
>will cause usb_add_hcd() to skip interrupt setup. This flag is
>set in xhci-pci, so the registration will skip the first request_irq()
>and can proceed.
>
>Cc: Bjorn Helgaas <[email protected]>
>Cc: Oliver Neukum <[email protected]>
>Cc: Thomas Renninger <[email protected]>
>Cc: Yinghai Lu <[email protected]>
>Cc: Frederik Himpe <[email protected]>
>Cc: David Haerdeman <[email protected]>
>Signed-off-by: Hannes Reinecke <[email protected]>
It doesn't seem to work. I just tried applying the patch to the 3.8
kernel in Debian experimental and this was the result during boot:
[ 1.203390] xhci_hcd 0000:00:14.0: can't derive routing for PCI INT A
[ 1.203393] xhci_hcd 0000:00:14.0: PCI INT A: no GSI
[ 1.203419] xhci_hcd 0000:00:14.0: setting latency timer to 64
[ 1.203423] xhci_hcd 0000:00:14.0: xHCI Host Controller
[ 1.203429] xhci_hcd 0000:00:14.0: new USB bus registered, assigned bus number 1
[ 1.203533] xhci_hcd 0000:00:14.0: cache line size of 64 is not supported
[ 1.203535] xhci_hcd 0000:00:14.0: request interrupt 255 failed
[ 1.203580] xhci_hcd 0000:00:14.0: USB bus 1 deregistered
[ 1.203598] xhci_hcd 0000:00:14.0: can't derive routing for PCI INT A
[ 1.203600] xhci_hcd 0000:00:14.0: init 0000:00:14.0 fail, -22
[ 1.203643] xhci_hcd: probe of 0000:00:14.0 failed with error -22
Regards,
David
On Fri, 1 Mar 2013, Hannes Reinecke wrote:
> xhci might run with MSI/MSI-X only, with no support for legacy
> interrupts. On these devices the request_irq() call in usb_add_hcd()
> will fail, causing the entire device to fail.
> For xhci this is especially painful as the driver will enable
> interrupts during xhci_run(), so the initial call to request_irq()
> is not required anyway.
>
> This patch adds a flag 'msix_supported' to struct usb_hcd, which
> will cause usb_add_hcd() to skip interrupt setup. This flag is
> set in xhci-pci, so the registration will skip the first request_irq()
> and can proceed.
I think it might be better to fix this in hcd-pci.c:usb_hcd_pci_probe.
At that point if we see that MSI/MSI-X is available, the call to
usb_add_hcd can pass 0 instead of dev->irq.
It may even be possible to do this without adding an extra flag
anywhere. It also has the advantage of keeping special knowledge about
MSI encapsulated in the PCI-specific parts of the code.
Alan Stern