Am Donnerstag, 1. Februar 2007 15:51 schrieb Alan Stern:
> On Wed, 31 Jan 2007, Oliver Neukum wrote:
>
> > This would call mod_timer() for every completed in-URB. Are you
> > sure this is better than an approach maintaining a periodical timer
> > and check for IO in the meantime?
>
> I think we can have it both ways. Add an atomic bitflag to the usb_device
> structure and have the autosuspend work routine check whether the flag is
> set. If it is, just restart the timer. The HID driver can simply set the
> flag whenever an URB is received.
This is exactly what I am doing in the HID driver, just with a private
timer. It is true that the timer could be switched off if other interfaces
preclude suspending the device anyhow, but apart from that I just
don't see the point in putting a specialised mechanism into generic code.
> There are other changes we should make that would go along with this.
>
> The existing bitfields in usb_device should be changed over
> to atomic bitflags, to make them SMP-safe.
Is this needed?
> The default autosuspend delay should become a usbcore module
> parameter rather than a constant. 0 can mean "don't autosuspend".
Good idea.
> Your new autosuspend attribute file should be renamed to
> autosuspend_delay and it should expose an unsigned short instead
> of a boolean: the number of seconds for the device's autosuspend
> timer. 0 can mean "don't autosuspend".
Roger.
> We should add a similar "suspend" attribute file. Writing
> "1" would immediately suspend and "0" would immediately resume
> the device. This will allow userspace to micro-manage the
> device state and will replace the deprecated power/state file.
In which way will this avoid that interface's principal problems?
Currently, we know that a driver has either requested autosuspend
and thus should be prepared to suspend or the guarantees for
a system wide freeze are valid. We'd need to audit all drivers.
> We should write sysfs_add_to_group(), so that these two
> attributes can be added to the existing device/power
> subdirectory. (sysfs_remove_from_group() doesn't seem to
> be needed, but it could be written as well.)
I will not comment on sysfs stuff without prior discussion with the
sysfs people. I am simply not competent to do so.
> We could add an unsigned short value (autosuspend_time) to
> struct usb_interface as well as to struct usb_device, so that
> individual interface drivers can specify a timeout value. Or
> we could just let them modify the value in usb_device, which
> would be a lot simpler.
>
> This leaves open the question of what timeout value we should actually
> use. If we have interface-specific values then clearly we want to use
> their maximum. But should a value set via sysfs always take precedence?
To the last question: yes.
> How does this sound?
A bit too complex with too much change in very important generic code.
Regards
Oliver
-------------------------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier.
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642
_______________________________________________
[email protected]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
