lkml.org 
[lkml]   [2003]   [Aug]   [28]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
SubjectRe: [RFC] /proc/ide/hdx/settings with ide-default pseudo-driver is a 2.6/2.7 show-stopper
Date
On Thursday 28 of August 2003 20:06, Alan Cox wrote:
> On Iau, 2003-08-28 at 16:47, Bartlomiej Zolnierkiewicz wrote:
> > > Its also used to avoid special cases elsewhere.
> >
> > /proc/ide/hdX/settings is the source (directly/indirectly)
> > for 95% of these special cases.
>
> /proc/ide/foo/identify is another (and tools rely on that working
> without ide-cd/ide-disk etc loaded btw) [HP monitoring tools for one]

Eeeh... easy to workaround: no settings -> drive->id doesnt change.
There is minor problem here: no driver -> driver -> no driver.

> > Why for hotplug? We should move all code physically touching
> > devices to use request queue (REQ_SPECIAL), then you can simply
> > mark device as unplugged when its gone and check this flag inside
> > ide_do_request().
>
> Because you need to manipulate drives not attached to a driver
> currently. I guess you could go through hoops to avoid it, but the
> old IDE driver was just full of bugs that ide_default removed,
> and it removed rather more code than it added.

Heh.. bugs yes, code not :).

> You also need to be able to open the device to talk to the empty

You can't open devices owned by ide-default in 2.6 because ide-default,
probably because it doesnt call add_disk() for them.

> slot about reprobing it, changing the bus state or even figuring
> what driver is missing.

I think scsi does it without default driver. procfs or sysfs

> > With this scheme you also shouldn't need hwif->unplugged_ops hack.
> > You mark all drives belonging to hwif as unplugged and you are happy.
>
> You need the unplugged_ops for controller unplug, I'm more worried about
> disk unplug (which I have working now). For controller unplug you either

So you can unplug disk in the middle of the transfer,
replug and transfer is continued?

> have to know that ide_unregister does not ever return until the device
> is truely idle - including interrupts, proc access currently running
> etc. If so then notionally you can handle it without the unplugged ops
> although they certainly make it safer.
>
> I don't know what the constraints on PCI hotplug are but I don't believe
> anything will run off and reuse the resources after an unplug event,
> although your accesses might generate machine checks (non fatal) on IA64
> etc.
>
> > Problem is when integrating ide with driverfs.
> > Then you need to register/unregister ide-default as driverfs driver
> > and now it can "steal" devices, ie. you have cd drive owned by
> > ide-default, later you load ide-cdrom driver and your cd drive needs to
> > be unregistered from ide-default driver first before it can be registered
> > with ide-cdrom driver - you need to add code to do this or device will be
> > "stealed". Its not very hard to do but it adds complexity.
>
> My own feeling is that its a lot easier than trying to catch all the
> corner cases ide_default fixed, especially when you do drive level
> hotplug. If the IDE code was glorious and elegant, refcounted and with
> a clear life cycle of objects I'd agree with you 8)

Eeehh... okay I will try "plan b". :-\

--bartlomiej

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 13:48    [from the cache]
©2003-2011 Jasper Spaans. Advertise on this site