On Wed, 2011-08-17 at 13:10 +0200, Sebastian Andrzej Siewior wrote:
> * Andiry Xu | 2011-08-17 18:01:42 [+0800]:
>
> >This patch tests USB2 software LPM for a USB2 LPM-capable device.
> >
> >When a lpm-capable device is addressed, if the host also supports software
> >LPM, apply a test by putting the device into L1 state and resume it to see
> >if the device can do L1 suspend/resume successfully.
> >
> >If the device fails to enter L1 or resume from L1 state, it may not
> >function normally and usbcore may disconnect and re-enumerate it. In this
> >case, store the device's Vid and Pid information, make sure the host will
> >not test LPM for it twice.
> >
> >The test result is per device/host. Some devices claim to be lpm-capable,
> >but fail to enter L1 or resume. The behavior is also different on different
> >IP-vendor hosts: one device passes the test on a host but fails to resume
> >on another with different IP vendor.
>
> The one IPcore that fails on one device, does it also fail on other
> devices or does it pass on other devices? So pattern is something like:
>
> IP Core A B C
> device A ok ok fail
> device B fail fail ok
> device C fail ok fail
>
I only observe one device shows different behaviors on different IP
hosts. Like:
IP A B
DevA pass pass
DevB fail pass
DevC fail fail
I only have two hosts supporting LPM, so I'm not sure what behavior
other IP hosts will have.
> i.e. totally random but reproducible mapping between IP core and device?
> Do you see any difference with a USB sniffer?
>
> >diff --git a/drivers/usb/host/xhci.c b/drivers/usb/host/xhci.c
> >index 3a0f695..fce413a 100644
> >--- a/drivers/usb/host/xhci.c
> >+++ b/drivers/usb/host/xhci.c
> >@@ -37,6 +37,10 @@ static int link_quirk;
> > module_param(link_quirk, int, S_IRUGO | S_IWUSR);
> > MODULE_PARM_DESC(link_quirk, "Don't clear the chain bit on a link TRB");
> >
> >+/* BESL to HIRD Encoding array for USB2 LPM */
> >+static int xhci_besl_encoding[16] = {125, 150, 200, 300, 400, 500, 1000, 2000,
> >+ 3000, 4000, 5000, 6000, 7000, 8000, 9000, 10000};
> >+
> > /* TODO: copied from ehci-hcd.c - can this be refactored? */
> > /*
> > * handshake - spin reading hc until handshake completes or fails
> >@@ -2957,6 +2961,166 @@ int xhci_address_device(struct usb_hcd *hcd, struct usb_device *udev)
> > return 0;
> > }
> >
> >+/* Calculate HIRD/BESL for USB2 PORTPMSC*/
> >+static int xhci_calculate_hird_besl(int u2del, bool use_besl)
> >+{
> >+ int hird;
> >+
> >+ if (use_besl) {
> >+ for (hird = 0; hird < 16; hird++) {
>
> ARRAY_SIZE_OF
>From the spec, the besl encoding array is fixed with 16 elements.
>
> >+ if (xhci_besl_encoding[hird] >= u2del)
> >+ break;
> >+ }
> >+ } else {
> >+ if (u2del <= 50)
> >+ hird = 0;
> >+ else
> >+ hird = (u2del - 51) / 75 + 1;
> >+
> >+ if (hird > 15)
> >+ hird = 15;
> >+ }
> >+
> >+ return hird;
> >+}
> >+
> >+static int xhci_usb2_software_lpm_test(struct xhci_hcd *xhci,
> >+ struct usb_device *udev, unsigned long flags)
> >+{
> >+ struct usb_hcd *hcd = xhci_to_hcd(xhci);
> >+ struct dev_info *dev_info;
> >+ __le32 __iomem **port_array;
> >+ __le32 __iomem *addr, *pm_addr;
> >+ u32 temp, dev_id;
> >+ unsigned int port_num;
> >+ int u2del, hird;
> >+ int ret;
> >+
> >+ if (hcd->speed == HCD_USB3 || !xhci->sw_lpm_support ||
> >+ !udev->lpm_capable)
> >+ return -EINVAL;
> >+
> >+ /* we only support lpm for non-hub device connected to root hub yet */
> >+ if (!udev->parent || udev->parent->parent ||
> >+ udev->descriptor.bDeviceClass == USB_CLASS_HUB)
> >+ return -EINVAL;
> >+
> >+ /* Look for devices in lpm_failed_devs list */
> >+ dev_id = le16_to_cpu(udev->descriptor.idVendor) << 16 |
> >+ le16_to_cpu(udev->descriptor.idProduct);
> >+ list_for_each_entry(dev_info, &xhci->lpm_failed_devs, list) {
> >+ if (dev_info->dev_id == dev_id)
> >+ return -EINVAL;
> >+ }
>
> I'm still not really convinced about the list for each device. If the
> device passes the test you re-test it on the next plug-in. So why not
> re-test it in the fail case if your static array is full?
> Can you quote some numbers about the time a successfull and a failed
> test needs? According to the code would say about 20ms in both cases
> (unless it is already in the list).
>From my observation, the test result is constant: a device always pass
or always fail on the same host. Re-test a passed device does not
involve much effort, but re-test a failed device may involve transfer
error, disconnection and re-initialization.
Actually LPM suspend and resume does not need 20ms, it's designed to
suspend/resume very quickly. Here I just use msleep to make sure it
enters the right link state.
Thanks,
Andiry
--
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
