On Wed, 2021-02-10 at 10:49 -0800, Florian Fainelli wrote:
> On 2/10/21 7:49 AM, Dave Stevenson wrote:
> > Hi Marc.
> > 
> > On Wed, 10 Feb 2021 at 15:30, Marc Zyngier <maz@kernel.org> wrote:
> > > 
> > > Hi Maxime,
> > > 
> > > On 2021-02-10 14:40, Maxime Ripard wrote:
> > > > Hi Dave,
> > > > 
> > > > On Tue, Feb 09, 2021 at 09:49:05AM +0000, Dave Stevenson wrote:
> > > > > On Mon, 11 Jan 2021 at 14:23, Maxime Ripard <maxime@cerno.tech> wrote:
> > > > > > 
> > > > > > The BSC controllers used for the HDMI DDC have an interrupt controller
> > > > > > shared between both instances. Let's add it to avoid polling.
> > > > > 
> > > > > This seems to have unintended side effects.
> > > > > GIC interrupt 117 is shared between the standard I2C controllers
> > > > > (i2c-bcm2835) and the l2-intc block handling the HDMI I2C interrupts.
> > > > > 
> > > > > Whilst i2c-bcm2835 requests the interrupt with IRQF_SHARED, that
> > > > > doesn't appear to be an option for l2-intc registering as an interrupt
> > > > > controller. i2c-bcm2835 therefore loses out and fails to register for
> > > > > the interrupt.
> > > > > 
> > > > > Is there an equivalent flag that an interrupt controller can add to
> > > > > say that the parent interrupt is shared? Is that even supported?
> > > > 
> > > > Indeed, it looks like setting an equivalent to IRQF_SHARED would be the
> > > > solution, but I couldn't find anything that would allow us to in the
> > > > irqchip code.
> > > > 
> > > > Marc, Thomas, is it something that is allowed?
> > > 
> > > No, not really. That's because the chained handler is actually an
> > > interrupt flow, and not a normal handler. IRQF_SHARED acts at the wrong
> > > level for that.
> > > 
> > > I can see two possibilities:
> > > 
> > > - the l2-intc gets turned into a normal handler, and does the demux
> > >    from there. Horrible stuff.
> > > 
> > > - the i2c controller gets parented to the l2c-int as a fake interrupt,
> > >    and gets called from there. Horrible stuff.
> > > 
> > > Pick your poison... :-/
> > 
> > Thanks for the info.
> > 
> > Option 3 - remove l2-intc and drop back to polling the i2c-brcmstb
> > blocks (which the driver supports anyway).
> > HDMI I2C generally isn't heavily used once displays are connected, so
> > I'd be OK with that.
> > 
> > (We can keep the l2-intc that handles CEC and HPD as that is on a
> > unique GIC interrupt).
> 
> Agreed, Maxime or Nicolas do you want me to send a revert of this patch?

Reverting seems the safe move, but I'll defer to whatever Maxime says.

Regards,
Nicolas