The problem was that the update_display_{visual,mode}() functions
(which send a control msg to the actual device) checks to make sure
mydev->shadow_power == 1, otherwise it returns immediately. The
mydev->shadow_power variable is only set in the resume(),
reset_resume(), and suspend() functions. So, when autosuspend is
disabled, resume() is never called, which means mydev->shadow_power is
still set to 0, and the update_display_{visual,mode}() functions
simply return and do not send a msg to the device.
The (new) mydev->has_interface_pm variable is simply to track if we
have called usb_autopm_get_interface(), instead of relying on
mydev->shadow_power to determine if it was called. We need to track
this because the user could "echo 1 > /sys/$DEVICE/powered" multiple
times, but we only want to increment the usage count once. That way,
when they "echo 0 > /sys/$DEVICE/powered" the device's display turns
off and usb_autopm_put_interface() is called making the reference
count 0 (assuming nothing else is using the interface / device). At
this point, the device can be suspended as its display is off.
Harry Metzger
On Mon, Mar 8, 2010 at 5:01 AM, Oliver Neukum <oliver@xxxxxxxxxx> wrote:
> Am Sonntag, 7. März 2010 22:45:42 schrieb Harrison Metzger:
>> /* sysfs_streq can't replace this completely
>> @@ -68,12 +69,16 @@ static void update_display_powered(struc
>> {
>> int rc;
>>
>> - if (!mydev->shadow_power && mydev->powered) {
>> + if (mydev->powered && !mydev->has_interface_pm) {
>> rc = usb_autopm_get_interface(mydev->intf);
>> if (rc < 0)
>> return;
>> + mydev->has_interface_pm = 1;
>> }
>>
>> + if (mydev->shadow_power != 1)
>> + return;
>> +
>
> This looks wrong, because even if you use autosuspend you
> cannot be sure usb_autopm_get_interface() calls resume(),
> as for example lsusb can have woken the driver before.
>
> What is the problem in the logic you are seeing?
>
> Regards
> Oliver
>
--
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
