The current EHCI code sleeps a flat 110ms in the resume path if there
was a USB 1.1 device connected to its companion controller during
suspend, waiting for the device to reappear and reset so that it can be
handed back to the companion. This is necessary if the device uses
persist, so that the companion controller can actually see it during its
own resume path.
However, if the device doesn't use persist, this is entirely
unnecessary. We might just as well ignore it and have the normal device
detection/reset/handoff code handle it asynchronously when it eventually
shows up. As USB 1.1 devices are almost exclusively HIDs these days (for
which persist has no value), this can allow distros to shave another
tenth of a second off their resume time.
Signed-off-by: Julius Werner <[email protected]>
---
drivers/usb/host/ehci-hub.c | 21 +++++++++++++++++++++
1 file changed, 21 insertions(+)
diff --git a/drivers/usb/host/ehci-hub.c b/drivers/usb/host/ehci-hub.c
index 7d06e77..2507afb 100644
--- a/drivers/usb/host/ehci-hub.c
+++ b/drivers/usb/host/ehci-hub.c
@@ -29,6 +29,8 @@
/*-------------------------------------------------------------------------*/
#include <linux/usb/otg.h>
+#include "../core/usb.h"
+
#define PORT_WAKE_BITS (PORT_WKOC_E|PORT_WKDISC_E|PORT_WKCONN_E)
#ifdef CONFIG_PM
@@ -42,6 +44,18 @@ static int ehci_hub_control(
u16 wLength
);
+
+static int companion_persist(struct device *dev, void *unused)
+{
+ struct usb_device *udev = container_of(dev, struct usb_device, dev);
+
+ if (!is_usb_device(dev))
+ return 0; /* Filter out struct usb_interface on the same bus */
+
+ return !udev->maxchild && udev->persist_enabled &&
+ udev->bus->root_hub->speed < USB_SPEED_HIGH;
+}
+
/* After a power loss, ports that were owned by the companion must be
* reset so that the companion can still own them.
*/
@@ -56,6 +70,13 @@ static void ehci_handover_companion_ports(struct ehci_hcd *ehci)
if (!ehci->owned_ports)
return;
+ /* Check if any USB 1.1 controller in the system uses a non-hub device
+ * with persist enabled. If they don't, the device will get reenumerated
+ * anyway and there is no need to slow resume down to wait for it.
+ */
+ if (!bus_for_each_dev(&usb_bus_type, NULL, NULL, companion_persist))
+ return;
+
/* Make sure the ports are powered */
port = HCS_N_PORTS(ehci->hcs_params);
while (port--) {
--
1.8.2.1
On Tue, 23 Apr 2013, Julius Werner wrote:
> The current EHCI code sleeps a flat 110ms in the resume path if there
> was a USB 1.1 device connected to its companion controller during
> suspend, waiting for the device to reappear and reset so that it can be
> handed back to the companion. This is necessary if the device uses
> persist, so that the companion controller can actually see it during its
> own resume path.
>
> However, if the device doesn't use persist, this is entirely
> unnecessary. We might just as well ignore it and have the normal device
> detection/reset/handoff code handle it asynchronously when it eventually
> shows up. As USB 1.1 devices are almost exclusively HIDs these days (for
> which persist has no value), this can allow distros to shave another
> tenth of a second off their resume time.
This is a good idea. The patch needs some fixes, though.
> Signed-off-by: Julius Werner <[email protected]>
> ---
> drivers/usb/host/ehci-hub.c | 21 +++++++++++++++++++++
> 1 file changed, 21 insertions(+)
>
> diff --git a/drivers/usb/host/ehci-hub.c b/drivers/usb/host/ehci-hub.c
> index 7d06e77..2507afb 100644
> --- a/drivers/usb/host/ehci-hub.c
> +++ b/drivers/usb/host/ehci-hub.c
> @@ -29,6 +29,8 @@
> /*-------------------------------------------------------------------------*/
> #include <linux/usb/otg.h>
>
> +#include "../core/usb.h"
Unfortunately, this won't work if ehci-hcd is built as a module.
usbcore does not export usb_device_type, which is used by
is_usb_device().
Also, including a core header into a non-core driver is ugly. Perhaps
a better approach would be to have usbcore export an iterator that runs
over devices only -- a for_each_usb_device() routine.
> +
> #define PORT_WAKE_BITS (PORT_WKOC_E|PORT_WKDISC_E|PORT_WKCONN_E)
>
> #ifdef CONFIG_PM
> @@ -42,6 +44,18 @@ static int ehci_hub_control(
> u16 wLength
> );
>
> +
> +static int companion_persist(struct device *dev, void *unused)
This name isn't so great. Something like
"persistent_device_on_usb11_bus" would be more accurate, if not more
readable. Maybe you can think of something better.
> +{
> + struct usb_device *udev = container_of(dev, struct usb_device, dev);
> +
> + if (!is_usb_device(dev))
> + return 0; /* Filter out struct usb_interface on the same bus */
> +
> + return !udev->maxchild && udev->persist_enabled &&
> + udev->bus->root_hub->speed < USB_SPEED_HIGH;
> +}
> +
> /* After a power loss, ports that were owned by the companion must be
> * reset so that the companion can still own them.
> */
> @@ -56,6 +70,13 @@ static void ehci_handover_companion_ports(struct ehci_hcd *ehci)
> if (!ehci->owned_ports)
> return;
>
> + /* Check if any USB 1.1 controller in the system uses a non-hub device
> + * with persist enabled. If they don't, the device will get reenumerated
> + * anyway and there is no need to slow resume down to wait for it.
> + */
/*
* New comments should be formatted like this.
*/
Also, the comment is grammatically unsound: It talks about non-existent
devices getting reenumerated. What you really mean is more like this:
If all devices (aside from hubs) on USB-1.1 buses have persist
disabled then we don't need to perform the handover, with its
> 100 ms delay. Any such devices will get reenumerated
normally; typically they are only HID anyway.
> + if (!bus_for_each_dev(&usb_bus_type, NULL, NULL, companion_persist))
> + return;
> +
> /* Make sure the ports are powered */
> port = HCS_N_PORTS(ehci->hcs_params);
> while (port--) {
Alan Stern