usb: ehci: Only sleep for post-resume handover if devices use persist
Subsystems (2)
PATCH v2 1/1 usb: ehci: Only sleep for post-resume handover if devices use persist
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.
--- drivers/usb/core/usb.c | 33 +++++++++++++++++++++++++++++++++ drivers/usb/host/ehci-hub.c | 17 +++++++++++++++++ include/linux/usb.h | 1 + 3 files changed, 51 insertions(+)
drivers/usb/core/usb.c +33 -0
@@ -209,6 +209,39 @@ struct usb_interface *usb_find_interface(struct usb_driver *drv, int minor) }EXPORT_SYMBOL_GPL (usb_find_interface); +struct each_dev_arg { +void *data; +int (*fn)(struct usb_device *,void *); +}; + +static int __each_dev (struct device *dev,void *data) +{ +struct each_dev_arg *arg = (struct each_dev_arg *)data; + +/* There are struct usb_interface on the same bus, filter them out */ +if (!is_usb_device (dev)) +return 0 ; + + return arg->fn(container_of(dev, struct usb_device, dev), arg->data); +} + +/** +* usb_for_each_dev - iterate over all USB devices in the system +* @data: data pointer that will be handed to the callback function +* @fn: callback function to be called for each USB device +* +* Iterate over all USB devices and call @fn for each, passing it @data. If it +* returns anything other than 0, we break the iteration prematurely and return +* that value. +*/ +int usb_for_each_dev (void *data,int (*fn)(struct usb_device *,void *)) +{ +struct each_dev_arg arg = {data, fn}; + +return bus_for_each_dev (&usb_bus_type,NULL , &arg, __each_dev); +} +EXPORT_SYMBOL_GPL (usb_for_each_dev); +/** * usb_release_dev - free a usb device structure when all users of it are finished. * @dev: device that's been disconnected
drivers/usb/host/ehci-hub.c +17 -0
@@ -42,6 +42,13 @@ static int ehci_hub_control(u16 wLength ); + +static int persist_enabled_on_companion (struct usb_device *udev,void *unused) +{ +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 +63,16 @@ static void ehci_handover_companion_ports(struct ehci_hcd *ehci)if (!ehci->owned_ports)return ; +/* +* USB 1.1 devices are mostly HIDs, which don't need to persist across +* suspends. If we ensure that none of our companion's devices have +* persist_enabled (by looking through all USB 1.1 buses in the system), +* we can skip this and avoid slowing resume down. Devices without +* persist will just get reenumerated shortly after resume anyway. +*/ +if (!usb_for_each_dev (NULL , persist_enabled_on_companion)) +return ; +/* Make sure the ports are powered */ port =HCS_N_PORTS (ehci->hcs_params);while (port--) {
include/linux/usb.h +1 -0
@@ -719,6 +719,7 @@ const struct usb_device_id *usb_match_id(struct usb_interface *interface,extern int usb_match_one_id (struct usb_interface *interface,const struct usb_device_id *id); +extern int usb_for_each_dev (void *data,int (*fn)(struct usb_device *,void *));extern struct usb_interface *usb_find_interface (struct usb_driver *drv,int minor); extern struct usb_interface *usb_ifnum_to_if(const struct usb_device *dev,
Signature · 1 line
1.7.12.4
Quoted 7 lines · Julius Werner
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.
Quoted 6 lines
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.
The patch description needs to mention the new for_each_usb_device() core routine.
drivers/usb/host/ehci-hub.c static int ehci_hub_control( · line 45
Quoted 7 lines
drivers/usb/host/ehci-hub.c +1
-0@@ -42,6 +42,13 @@ static int ehci_hub_control(u16 wLength ); +
Unnecessary blank line added.
include/linux/usb.h const struct usb_device_id *usb_match_id(struct usb_interface *interface, · line 722
Quoted 11 lines
drivers/usb/host/ehci-hub.c +2
-0+static int persist_enabled_on_companion (struct usb_device *udev,void *unused) +{include/linux/usb.h +1
-0@@ -719,6 +719,7 @@ const struct usb_device_id *usb_match_id(struct usb_interface *interface,extern int usb_match_one_id (struct usb_interface *interface,const struct usb_device_id *id); +extern int usb_for_each_dev (void *data,int (*fn)(struct usb_device *,void *));
Line is too long.
Aside from these minor issues, the patch is fine. Fix them and you can add
Quoted 25 lines · Alan Stern
Quoted 7 lines · Julius Werner
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.
Quoted 6 lines
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.
The patch description needs to mention the new
for_each_usb_device()core routine.Quoted 6 lines
drivers/usb/host/ehci-hub.c +0
-0@@ -42,6 +42,13 @@ static int ehci_hub_control(u16 wLength );
Quoted 3 lines
drivers/usb/host/ehci-hub.c static int ehci_hub_control( · line 45
Quoted 1 line
drivers/usb/host/ehci-hub.c +1
-0+Unnecessary blank line added.
Sorry, will fix that.
Quoted 10 lines
Quoted 10 lines
drivers/usb/host/ehci-hub.c +2
-0+static int persist_enabled_on_companion (struct usb_device *udev,void *unused) +{include/linux/usb.h +0
-0@@ -719,6 +719,7 @@ const struct usb_device_id *usb_match_id(struct usb_interface *interface,extern int usb_match_one_id (struct usb_interface *interface,const struct usb_device_id *id);
Quoted 3 lines
include/linux/usb.h const struct usb_device_id *usb_match_id(struct usb_interface *interface, · line 722
Quoted 1 line
include/linux/usb.h +1
-0+extern int usb_for_each_dev (void *data,int (*fn)(struct usb_device *,void *));Line is too long.
The limit is 80 (inclusive), right? Without the diff's leading '+' I count exactly that...
Quoted 5 lines
Aside from these minor issues, the patch is fine. Fix them and you can add
Acked-by Alan Stern <[email protected]>
Alan Stern
maintainerQuoted 6 lines · Julius Werner
Quoted 3 lines
Quoted 1 line
+extern int
usb_for_each_dev(void *data, int (*fn)(struct usb_device *, void *));Line is too long.
The limit is 80 (inclusive), right? Without the diff's leading '+' I count exactly that...
Oh, okay, sorry. I was fooled by the email client.
Alan Stern