Re: [PATCH v7 2/2] irqchip: Add Qualcomm MPM controller driver

[Shawn Guo]

On Wed, Mar 02, 2022 at 01:57:27PM +0000, Marc Zyngier wrote:
> On Wed, 02 Mar 2022 13:34:41 +0000,
> Shawn Guo <shawn.guo@xxxxxxxxxx> wrote:
> > 
> > On Wed, Mar 02, 2022 at 10:25:45AM +0000, Marc Zyngier wrote:
> > > On Wed, 02 Mar 2022 08:40:28 +0000,
> > > Shawn Guo <shawn.guo@xxxxxxxxxx> wrote:
> > > > 
> > > > Hi Marc,
> > > > 
> > > > On Tue, Mar 01, 2022 at 11:13:30AM +0000, Marc Zyngier wrote:
> > > > > Hi Shawn,
> > > 
> > > [...]
> > > 
> > > > > 
> > > > > > +static int qcom_mpm_set_type(struct irq_data *d, unsigned int type)
> > > > > > +{
> > > > > > +	struct qcom_mpm_priv *priv = d->chip_data;
> > > > > > +	int pin = d->hwirq;
> > > > > > +	unsigned int index = pin / 32;
> > > > > > +	unsigned int shift = pin % 32;
> > > > > > +
> > > > > > +	switch (type & IRQ_TYPE_SENSE_MASK) {
> > > > > > +	case IRQ_TYPE_EDGE_RISING:
> > > > > > +		mpm_set_type(priv, !!(type & IRQ_TYPE_EDGE_RISING),
> > > > > > +			     MPM_REG_RISING_EDGE, index, shift);
> > > > > > +		break;
> > > > > > +	case IRQ_TYPE_EDGE_FALLING:
> > > > > > +		mpm_set_type(priv, !!(type & IRQ_TYPE_EDGE_FALLING),
> > > > > > +			     MPM_REG_FALLING_EDGE, index, shift);
> > > > > > +		break;
> > > > > > +	case IRQ_TYPE_LEVEL_HIGH:
> > > > > > +		mpm_set_type(priv, !!(type & IRQ_TYPE_LEVEL_HIGH),
> > > > > > +			     MPM_REG_POLARITY, index, shift);
> > > > > > +		break;
> > > > > > +	}
> > > > > 
> > > > > All these '!!(type & BLAH)' are totally superfluous, as they all expand
> > > > > to 'true' by construction.
> > > > 
> > > > Yes, you are right!
> > > > 
> > > > > And this leads to a few questions:
> > > > > 
> > > > > - Shouldn't a rising interrupt clear the falling detection?
> > > > > - Shouldn't a level-low clear the polarity?
> > > > > - How do you handle IRQ_TYPE_EDGE_BOTH?
> > > > > - How is MPM_REG_POLARITY evaluated for edge interrupts (resp the EDGE
> > > > >   registers for level interrupts), as you never seem to be configuring
> > > > >   a type here?
> > > > 
> > > > Honestly, qcom_mpm_set_type() was mostly taken from downstream without
> > > > too much thinking.
> > 
> > I have to take this statement back.  It seems that the current code has
> > been diverted from the downstream in a wrong way.
> > 
> > > > I trusted it as a "good" reference as I have no
> > > > document to verify the code.  These questions are great and resulted the
> > > > code changes are pretty sensible to me.
> > > 
> > > I don't think these changes are enough. For example, an interrupt
> > > being switched from level to edge is likely to misbehave (how do you
> > > distinguish the two?). If that's what the downstream driver does, then
> > > it is terminally broken.
> > 
> > Could you take a look at downstream code and see if it answers all your
> > questions?
> 
> This code actually makes me ask more questions. Why is it programming
> 2 'pins' for each IRQ?

The mapping between MPM pin and GIC IRQ is not strictly 1-1.  There are
some rare case that up to 2 MPM pins map to a single GIC IRQ, for
example the last two in QC2290 'qcom,mpm-pin-map' below.

	qcom,mpm-pin-map = <2 275>,     /* tsens0_tsens_upper_lower_int */
			   <5 296>,     /* lpass_irq_out_sdc */
			   <12 422>,    /* b3_lfps_rxterm_irq */
			   <24 79>,     /* bi_px_lpi_1_aoss_mx */
			   <86 183>,    /* mpm_wake,spmi_m */
			   <90 260>,    /* eud_p0_dpse_int_mx */
			   <91 260>;    /* eud_p0_dmse_int_mx */


The downstream uses a DT bindings that specifies GIC hwirq number in
client device nodes.  In that case, d->hwirq in the driver is GIC IRQ
number, and the driver will need to query mapping table, find out the
possible 2 MPM pins, and set them up.

The patches I'm posting here use a different bindings that specifies MPM
pin instead in client device nodes.  Thus the driver can simply get the
MPM pin from d->hwirq, so that the whole look-up procedure can be saved.

> 
> > 
> > It seems MPM_REG_POLARITY is only meant for level interrupts, since edge
> > interrupts already have separate registers for rising and falling.
> 
> Then level interrupts must clear both the edge registers at all times.

The downstream logic already covers that, right?  The edge register bits
will be cleared as long as 'flowtype' is not EDGE.

static void msm_mpm_set_type(struct irq_data *d,
                                        unsigned int flowtype, bool is_mpmgic)
{   
        int mpm_pin[MAX_MPM_PIN_PER_IRQ] = {-1, -1};
        unsigned long flags;  
        int i = 0;
        unsigned int index, mask;      
        unsigned int reg = 0; 
    
        msm_get_mpm_pin(d, mpm_pin, is_mpmgic);
        for (i = 0; i < MAX_MPM_PIN_PER_IRQ; i++) {
                if (mpm_pin[i] < 0)            
                        return;                        
    
                index = mpm_pin[i]/32;         
                mask = mpm_pin[i]%32;          
    
                spin_lock_irqsave(&mpm_lock, flags);
                reg = MPM_REG_RISING_EDGE;     
                if (flowtype & IRQ_TYPE_EDGE_RISING)
                        msm_mpm_program_set_type(1, reg, index, mask);
                else          
                        msm_mpm_program_set_type(0, reg, index, mask);

                reg = MPM_REG_FALLING_EDGE;    
                if (flowtype & IRQ_TYPE_EDGE_FALLING)
                        msm_mpm_program_set_type(1, reg, index, mask);
                else          
                        msm_mpm_program_set_type(0, reg, index, mask);
    
                reg = MPM_REG_POLARITY;        
                if (flowtype & IRQ_TYPE_LEVEL_HIGH)
                        msm_mpm_program_set_type(1, reg, index, mask);
                else
                        msm_mpm_program_set_type(0, reg, index, mask);
                spin_unlock_irqrestore(&mpm_lock, flags);
        }
}

> > I will fix my broken code by respecting the downstream logic.
> > 
> > > As I asked before, we need some actual specs, or at least someone to
> > > paraphrase it for us. There are a number of QC folks on Cc, and I
> > > expect them to chime in and explain how MPM works here.
> > > 
> > > > 
> > > > > - What initialises the MPM trigger types at boot time?
> > > > 
> > > > I dumped the vMPM region and it's all zeros.  My understanding is if
> > > > vMPM needs any sort of initialization, it should be done by RPM firmware
> > > > before APSS gets booting.
> > > 
> > > What about kexec? We can't rely on this memory region to always be
> > > 0-initialised, nor do we know what that means.
> > 
> > We are not relying on it being 0-initialised, but being initialised by
> > RPM with initial physical MPM register values.
> 
> Whatever. It simply cannot be trusted. If you kexec another kernel,
> you need to be able to restore a sane state at probe time. This isn't
> optional.

Right, I will add an explicit initialization on vMPM region at probe
time.

Shawn