Re: [PATCH] gpio: gpio-mockup: Fix usage of new GPIO_LINE_DIRECTION

[Kent Gibson]

On Tue, Dec 10, 2019 at 04:34:21PM +0100, Bartosz Golaszewski wrote:
> wt., 10 gru 2019 o 15:55 Kent Gibson <warthog618@xxxxxxxxx> napisaÅ(a):
> >
> > On Tue, Dec 10, 2019 at 03:11:12PM +0100, Bartosz Golaszewski wrote:
> > > wt., 10 gru 2019 o 03:15 Kent Gibson <warthog618@xxxxxxxxx> napisaÅ(a):
> > > >
> > > > Restore the external behavior of gpio-mockup to what it was prior to the
> > > > change to using GPIO_LINE_DIRECTION.
> > > >
> > > > Signed-off-by: Kent Gibson <warthog618@xxxxxxxxx>
> > > > ---
> > > >
> > > > Fix a regression introduced in v5.5-rc1.
> > > >
> > > > The change to GPIO_LINE_DIRECTION reversed the polarity of the
> > > > dir field within gpio-mockup.c, but overlooked inverting the value on
> > > > initialization and when returned by gpio_mockup_get_direction.
> > > > The latter is a bug.
> > > > The former is a problem for tests which assume initial conditions,
> > > > specifically the mockup used to initialize chips with all lines as inputs.
> > > > That superficially appeared to be the case after the previous patch due
> > > > to the bug in gpio_mockup_get_direction.
> > > >
> > > >  drivers/gpio/gpio-mockup.c | 7 +++++--
> > > >  1 file changed, 5 insertions(+), 2 deletions(-)
> > > >
> > > > diff --git a/drivers/gpio/gpio-mockup.c b/drivers/gpio/gpio-mockup.c
> > > > index 56d647a30e3e..c4fdc192ea4e 100644
> > > > --- a/drivers/gpio/gpio-mockup.c
> > > > +++ b/drivers/gpio/gpio-mockup.c
> > > > @@ -226,7 +226,7 @@ static int gpio_mockup_get_direction(struct gpio_chip *gc, unsigned int offset)
> > > >         int direction;
> > > >
> > > >         mutex_lock(&chip->lock);
> > > > -       direction = !chip->lines[offset].dir;
> > > > +       direction = chip->lines[offset].dir;
> > > >         mutex_unlock(&chip->lock);
> > > >
> > > >         return direction;
> > > > @@ -395,7 +395,7 @@ static int gpio_mockup_probe(struct platform_device *pdev)
> > > >         struct gpio_chip *gc;
> > > >         struct device *dev;
> > > >         const char *name;
> > > > -       int rv, base;
> > > > +       int rv, base, i;
> > > >         u16 ngpio;
> > > >
> > > >         dev = &pdev->dev;
> > > > @@ -447,6 +447,9 @@ static int gpio_mockup_probe(struct platform_device *pdev)
> > > >         if (!chip->lines)
> > > >                 return -ENOMEM;
> > > >
> > > > +       for (i = 0; i < gc->ngpio; i++)
> > > > +               chip->lines[i].dir = GPIO_LINE_DIRECTION_IN;
> > > > +
> > > >         if (device_property_read_bool(dev, "named-gpio-lines")) {
> > > >                 rv = gpio_mockup_name_lines(dev, chip);
> > > >                 if (rv)
> > > > --
> > > > 2.24.0
> > > >
> > >
> > > Hi Kent,
> > >
> > > I was applying and testing your libgpiod series and noticed that the
> > > gpio-tools tests fail after applying patches 16 & 17 (with linux
> > > v5.5-rc1). Is this fix related to this?
> > >
> >
> > I don't think so.  I've only been able to trip this problem with a
> > couple of corner cases in my Go uapi test suite.
> > I have been unable to reproduce it with the tools as it requires
> > multiple requests with the same chip fd, including an as-is, to trip.
> >
> > And running the libgpiod tests against v5.5-rc1 works for me.
> > Can you provide more details as to the errors you are seeing?
> >
> 
> Hmm whatever that was, it's gone now. Must have been some leftovers
> from previous builds. All works now.
> 
> > Btw, I was writing tests for your LINEINFO_WATCH patch v2, which I was
> > applying to v5.5-rc1, when I ran across this.  That works ok if I
> > __packed the changed struct.
> 
> These things can still change, so don't spend too much time on it yet. :)
> 

Absolutely.  But as this is an ABI addition I wanted to have something
to give it a decent workout before it gets applied.
So far the only problems I've found are the alignment and isolation
issues already mentioned.

> Since the lineinfo struct is not __packed, I'd prefer to not use it
> for any struct embedding it. I'll just add appropriate padding for
> 64-bit alignment.
> 

Explicit padding for 64-bit alignment makes sense to me.

> > And I can confirm that patch v2 doesn't isolate watches on different
> > chip fds.
> >
> 
> Yeah, I'll fix this.

One more thing - since it is possible to lose change events due to fifo
overflow, how about adding a seqnum?  And if you do end up doing a v2 of
the event ABI to fix its alignment, adding one there as well.

Kent.