Re: [PATCH] s390/vfio-ap: do not use open locks during VFIO_GROUP_NOTIFY_SET_KVM notification

[Tony Krowiak]

On 7/6/21 9:49 AM, Jason Gunthorpe wrote:
> On Tue, Jul 06, 2021 at 09:39:29AM -0400, Tony Krowiak wrote:
> > On 7/5/21 10:13 AM, Jason Gunthorpe wrote:
> > > On Thu, Jul 01, 2021 at 10:28:52AM -0400, Tony Krowiak wrote:
> > >
> > > > > I think Jason was talking about open coding locks in general.
> > > > That may be so, but his comments were in support of his
> > > > statement that the  mutex + wait_queue did not resolve
> > > > the issue reported vai the lockdep splat because it turned
> > > > off lockdep.
> > > Rgiht, if this used to be proper locks and lockdep complained then
> > > whatever potential deadlock it found is not magically removed by going
> > > to a wait_queue. It just removes the lockdep annotations that would
> > > identify the issue early.
> > >
> > > This is why people should not open code locks, it completely defeats
> > > lockdep. That alone is merit enough for this patch.
> > When you use the phrase "open code locks", to what are you
> > specifically referring? I am confused by the use of the phrase
> > "open code" in this context because open coding, at least as
> > I understand it, has to do with data analysis.
> "open code" here means you write the algorithm of a standard lock in
> your own functions instead of calling the standard library.
>
> Testing/setting the busy and sleeping on a wait_event is exactly a
> standard lock.
>
> Ie if I write
>
>     for (len = 0; str[len] != 0; len++)
>             ;
>
> Then I have open coded strlen()
>
> Jason

Thanks for the explanation.