On Tue, Apr 29, 2014 at 8:14 AM, Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> wrote:
> On Tue, 29 Apr 2014, David Laight wrote:
>
>> From: Dan Williams
>> > Before we add another 2 bitfields to the port state, convert it to an
>> > unsigned long and use bitops helpers to manipulate it. This later
>> > enables setting bits to request action without worrying about colliding
>> > updates. In particular, a flag to request that the child device be
>> > woken up when the port restores power is added in a later patch.
>> ...
>> > struct usb_port {
>> > struct usb_device *child;
>> > @@ -96,8 +95,9 @@ struct usb_port {
>> > enum usb_port_connect_type connect_type;
>> > usb_port_location_t location;
>> > u8 portnum;
>> > - unsigned power_is_on:1;
>> > - unsigned did_runtime_put:1;
>> > + #define USB_PORTDEV_POWER 0
>> > + #define USB_PORTDEV_DID_RUNTIME_PUT 1
>> > + unsigned long flags;
>> > };
>> ...
>>
>>
>> An alternative is to 'waste' 7 bits and use byte-sized fields
>> for the separate flags.
>> I doubt the size of the data structure matters.
>> (Or do we have to worry about cpus that use RMW cycles for byte writes?)
>
> We do.
>
Right, last I recall archs don't guarantee atomic access of anything
less than a long.
> Another way to avoid the waste is to move the flags into the usb_hub
> structure.
>
I don't see how that avoids waste? We need a long per flag. In the
struct usb_port we can combine status and requests in the same
storage.
--
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
