[sane-devel] [RFC] Locking devices via sanei_access lib...

Oliver Rauch Oliver.Rauch@Rauch-Domain.DE
07 Feb 2005 22:07:55 +0100


Am Mon, 2005-02-07 um 20.10 schrieb Gerhard Jaeger:
> On Monday 07 February 2005 16:50, Oliver Rauch wrote:
> > Am Mon, 2005-02-07 um 14.10 schrieb Gerhard Jaeger:
> > > Hi,
> > >
> > > On Monday 07 February 2005 12:41, Johannes Meixner wrote:
> > > > Hello,
> > > >
> > > > On Feb 7 12:11 Gerhard Jaeger wrote (shortened):
> > > > > The idea is to provide a simple locking mechanism for the backends,
> > > > > to have exclusive access to one scanner during an operation:
> > > >
> > > > ...
> > > >
> > > > > What do you guys think of this lib - is it useful?
> > > >
> > > > I even think exclusive access should be used by default
> > > > because I think normally it doesn't make sense when there
> > > > is more than one process which talks to the same scanner
> > > > (i.e. "the same scanner" but not "the same backend" because
> > > > one backend can drive several scanners).
> > >
> > > I agree! Each device a backend talks to needs some exclusive
> > > locking!
> >
> > We already had a discussion about this some time ago (2-3 years).
> > My opinion is a bit different.
> 
> Really ;)
> 
> > On my work we have one scanner and about 30 people who have access
> > to the scanner. And I know that some people keep open the
> > scanning frontend all the time. With exclusive locking the device
> > would almost be unusable.
> 
> I totally agree. The button-daemon also keeps the backend always open.
> 
> >
> > The umax (scsi) backend locks the scanner when sane_start is called
> > and releases the scanner when sane_cancel is called. This way there
> > is no problem to keep the frontend opened all the time.
> 
> Oh, I didn't know this - how is this done in the umax backend?

The umax scsi scanners have a command RESERVE_UNIT and a command
RELEASE_UNIT that is used for device internal locking.

Oliver
> 
> In my test-implementation also only the "real" device access
> is locked exclusive. This real access covers - detection (during sane_init),
> button status check (sane_control_option) and calibration/scanning 
> (sane_open/sane_cancel).
> 
> >
> > I think it makes sense that opening the scanning frontend does not
> > lock the device. But in this case we need at least a button locking.
> 
> Absolutely - and for this locking, I'd like to have one common lib, like
> the sanei_thread lib. 
> 
> Gerhard
> 
> >
> > Oliver
> >
> > > > Once it happened while I did a nice (you may say stupid) stress test
> > > >
> > > > for i in $( seq 100 )
> > > > do sane-find-scanner & scanimage -L &
> > > > done
> > > >
> > > > that one of my USB scanners made funny noise: It seems it tried
> > > > to move its scanning unit beyond its physical limits.
> > > > I have no idea how this could have happened but I guess the scanner
> > > > was simply totally confused by the multiple concurrent accesses.
> > >
> > > It should not happen, but even such stupid tests show, that the locking
> > > needs to be done in another way, as it's currently done, if it's done.
> > > Rather the same "confusion" will happen when you have something
> > > like Rene Rebes' button daemon which does nothing else than checking the
> > > button status of a scanner - depending on the scanner, the backend needs
> > > to check some or at least one register and when another frontend
> > > currently is scanning - the scanner will not work properly afterwards
> > > (highly depends on that device)
> > >
> > > > By the way:
> > > > The reason for the above stress-test was that this way I could
> > > > stop the kernel (no single error message in the logs  - just a
> > > > sudden stop) - meanwhile I do no longer use the crap SCSI
> > > > host adapter ;-)
> > >
> > > This test should become another stress-test for a new backend ;)