Message-ID: <20210921185222.246b15bb.pasic@linux.ibm.com>
Date:   Tue, 21 Sep 2021 18:52:22 +0200
From:   Halil Pasic <pasic@...ux.ibm.com>
To:     Vineeth Vijayan <vneethv@...ux.ibm.com>
Cc:     Cornelia Huck <cohuck@...hat.com>,
        Heiko Carstens <hca@...ux.ibm.com>,
        Vasily Gorbik <gor@...ux.ibm.com>,
        Christian Borntraeger <borntraeger@...ibm.com>,
        Pierre Morel <pmorel@...ux.ibm.com>,
        Michael Mueller <mimu@...ux.ibm.com>,
        linux-s390@...r.kernel.org,
        virtualization@...ts.linux-foundation.org, kvm@...r.kernel.org,
        linux-kernel@...r.kernel.org, bfu@...hat.com,
        Halil Pasic <pasic@...ux.ibm.com>,
        Peter Oberparleiter <oberpar@...ux.ibm.com>
Subject: Re: [PATCH 1/1] virtio/s390: fix vritio-ccw device teardown

On Tue, 21 Sep 2021 15:31:03 +0200
Vineeth Vijayan <vneethv@...ux.ibm.com> wrote:

> On Tue, 2021-09-21 at 05:25 +0200, Halil Pasic wrote:
> > On Mon, 20 Sep 2021 12:07:23 +0200
> > Cornelia Huck <cohuck@...hat.com> wrote:
> >   
> > > On Mon, Sep 20 2021, Vineeth Vijayan <vneethv@...ux.ibm.com> wrote:
> > >   
> > > > On Mon, 2021-09-20 at 00:39 +0200, Halil Pasic wrote:    
> > > > > On Fri, 17 Sep 2021 10:40:20 +0200
> > > > > Cornelia Huck <cohuck@...hat.com> wrote:
> > > > >     
> > > > ...snip...    
> > > > > > > Thanks, if I find time for it, I will try to understand
> > > > > > > this
> > > > > > > better and
> > > > > > > come back with my findings.
> > > > > > >      
> > > > > > > > > * Can virtio_ccw_remove() get called while !cdev-  
> > > > > > > > > >online and   
> > > > > > > > >   virtio_ccw_online() is running on a different cpu? If
> > > > > > > > > yes,
> > > > > > > > > what would
> > > > > > > > >   happen then?        
> > > > > > > > 
> > > > > > > > All of the remove/online/... etc. callbacks are invoked
> > > > > > > > via the
> > > > > > > > ccw bus
> > > > > > > > code. We have to trust that it gets it correct :) (Or
> > > > > > > > have the
> > > > > > > > common
> > > > > > > > I/O layer maintainers double-check it.)
> > > > > > > >       
> > > > > > > 
> > > > > > > Vineeth, what is your take on this? Are the struct
> > > > > > > ccw_driver
> > > > > > > virtio_ccw_remove and the virtio_ccw_online callbacks
> > > > > > > mutually
> > > > > > > exclusive. Please notice that we may initiate the onlining
> > > > > > > by
> > > > > > > calling ccw_device_set_online() from a workqueue.
> > > > > > > 
> > > > > > > @Conny: I'm not sure what is your definition of 'it gets it
> > > > > > > correct'...
> > > > > > > I doubt CIO can make things 100% foolproof in this
> > > > > > > area.      
> > > > > > 
> > > > > > Not 100% foolproof, but "don't online a device that is in the
> > > > > > progress
> > > > > > of going away" seems pretty basic to me.
> > > > > >     
> > > > > 
> > > > > I hope Vineeth will chime in on this.    
> > > > Considering the online/offline processing, 
> > > > The ccw_device_set_offline function or the online/offline is
> > > > handled
> > > > inside device_lock. Also, the online_store function takes care of
> > > > avoiding multiple online/offline processing. 
> > > > 
> > > > Now, when we consider the unconditional remove of the device,
> > > > I am not familiar with the virtio_ccw driver. My assumptions are
> > > > based
> > > > on how CIO/dasd drivers works. If i understand correctly, the
> > > > dasd
> > > > driver sets different flags to make sure that a device_open is
> > > > getting
> > > > prevented while the the device is in progress of offline-ing.     
> > > 
> > > Hm, if we are invoking the online/offline callbacks under the
> > > device
> > > lock already,   
> > 
> > I believe we have a misunderstanding here. I believe that Vineeth is
> > trying to tell us, that online_store_handle_offline() and
> > online_store_handle_offline() are called under the a device lock of
> > the ccw device. Right, Vineeth?  
> Yes. I wanted to bring-out both the scenario.The set_offline/_online()
> calls and the unconditional-remove call.

I don't understand the paragraph above. I can't map the terms
set_offline/_online() and unconditional-remove call to chunks of code.
:( 

> For the set_online The virtio_ccw_online() also invoked with ccwlock
> held. (ref: ccw_device_set_online)

I don't think virtio_ccw_online() is invoked with the ccwlock held. I
think we call virtio_ccw_online() in this line:
https://elixir.bootlin.com/linux/v5.15-rc2/source/drivers/s390/cio/device.c#L394
and we have released the cdev->ccwlock literally 2 lines above.


> > 
> > Conny, I believe, by online/offline callbacks, you mean
> > virtio_ccw_online() and virtio_ccw_offline(), right?
> > 
> > But the thing is that virtio_ccw_online() may get called (and is
> > typically called, AFAICT) with no locks held via:
> > virtio_ccw_probe() --> async_schedule(virtio_ccw_auto_online, cdev)
> > -*-> virtio_ccw_auto_online(cdev) --> ccw_device_set_online(cdev) -->
> > virtio_ccw_online()
> > 
> > Furthermore after a closer look, I believe because we don't take
> > a reference to the cdev in probe, we may get virtio_ccw_auto_online()
> > called with an invalid pointer (the pointer is guaranteed to be valid
> > in probe, but because of async we have no guarantee that it will be
> > called in the context of probe).
> > 
> > Shouldn't we take a reference to the cdev in probe?  
> We just had a quick look at the virtio_ccw_probe() function.
> Did you mean to have a get_device() during the probe() and put_device()
> just after the virtio_ccw_auto_online() ?

Yes, that would ensure that cdev pointer is still valid when
virtio_ccw_auto_online() is executed, and that things are cleaned up
properly, I guess. But I'm not 100% sure about all the interactions.
AFAIR ccw_device_set_online(cdev) would bail out if !drv. But then
we have the case where we already assigned it to a new driver (e.g.
vfio for dasd).

BTW I believe if we have a problem here, the dasd driver has the same
problem as well. The code looks very, very similar.

And shouldn't this auto-online be common CIO functionality? What is the
reason the char devices don't seem to have it?

Regards,
Halil

Powered by blists - more mailing lists