Re: [PATCH v7 2/4] PCI: dwc: rockchip: add legacy interrupt support

[Marc Zyngier <maz@kernel.org>]

On Sat, 16 Apr 2022 14:24:26 +0100,
Peter Geis <pgwipeout@gmail.com> wrote:
> 
> On Sat, Apr 16, 2022 at 8:54 AM Marc Zyngier <maz@kernel.org> wrote:
> >
> > Peter,
> >
> > May I suggest that you slow down on the number of versions you send?
> > This is the 7th in 5 days, the 3rd today.
> >
> > At this stage, this is entirely counterproductive.
> 
> Apologies, I'll be sure to be at least one cup of coffee in before
> doing early morning code.

Even with a steady intake of coffee, there is a pretty clear policy
around the frequency of patch submission, see [1].

[1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/Documentation/process/submitting-patches.rst#n337

There is no hard enforcement of this process, but that should give you
an idea of how to deal with it. In any case, 7 series in less than a
week is a clear sign that this series should be *ignored*, as the
author is likely to post yet another one in the next few hours.

> 
> >
> > On 2022-04-16 12:05, Peter Geis wrote:
> > > The legacy interrupts on the rk356x pcie controller are handled by a
> > > single muxed interrupt. Add irq domain support to the pcie-dw-rockchip
> > > driver to support the virtual domain.
> > >
> > > Signed-off-by: Peter Geis <pgwipeout@gmail.com>
> > > ---
> > >  drivers/pci/controller/dwc/pcie-dw-rockchip.c | 112 +++++++++++++++++-
> > >  1 file changed, 110 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/drivers/pci/controller/dwc/pcie-dw-rockchip.c
> > > b/drivers/pci/controller/dwc/pcie-dw-rockchip.c
> > > index c9b341e55cbb..863374604fb1 100644
> > > --- a/drivers/pci/controller/dwc/pcie-dw-rockchip.c
> > > +++ b/drivers/pci/controller/dwc/pcie-dw-rockchip.c
> > > @@ -10,9 +10,12 @@
> > >
> > >  #include <linux/clk.h>
> > >  #include <linux/gpio/consumer.h>
> > > +#include <linux/irqchip/chained_irq.h>
> > > +#include <linux/irqdomain.h>
> > >  #include <linux/mfd/syscon.h>
> > >  #include <linux/module.h>
> > >  #include <linux/of_device.h>
> > > +#include <linux/of_irq.h>
> > >  #include <linux/phy/phy.h>
> > >  #include <linux/platform_device.h>
> > >  #include <linux/regmap.h>
> > > @@ -36,10 +39,13 @@
> > >  #define PCIE_LINKUP                  (PCIE_SMLH_LINKUP | PCIE_RDLH_LINKUP)
> > >  #define PCIE_L0S_ENTRY                       0x11
> > >  #define PCIE_CLIENT_GENERAL_CONTROL  0x0
> > > +#define PCIE_CLIENT_INTR_STATUS_LEGACY       0x8
> > > +#define PCIE_CLIENT_INTR_MASK_LEGACY 0x1c
> > >  #define PCIE_CLIENT_GENERAL_DEBUG    0x104
> > > -#define PCIE_CLIENT_HOT_RESET_CTRL      0x180
> > > +#define PCIE_CLIENT_HOT_RESET_CTRL   0x180
> > >  #define PCIE_CLIENT_LTSSM_STATUS     0x300
> > > -#define PCIE_LTSSM_ENABLE_ENHANCE       BIT(4)
> > > +#define PCIE_LEGACY_INT_ENABLE               GENMASK(3, 0)
> > > +#define PCIE_LTSSM_ENABLE_ENHANCE    BIT(4)
> > >  #define PCIE_LTSSM_STATUS_MASK               GENMASK(5, 0)
> > >
> > >  struct rockchip_pcie {
> > > @@ -51,6 +57,8 @@ struct rockchip_pcie {
> > >       struct reset_control            *rst;
> > >       struct gpio_desc                *rst_gpio;
> > >       struct regulator                *vpcie3v3;
> > > +     struct irq_domain               *irq_domain;
> > > +     raw_spinlock_t                  irq_lock;
> > >  };
> > >
> > >  static int rockchip_pcie_readl_apb(struct rockchip_pcie *rockchip,
> > > @@ -65,6 +73,94 @@ static void rockchip_pcie_writel_apb(struct
> > > rockchip_pcie *rockchip,
> > >       writel_relaxed(val, rockchip->apb_base + reg);
> > >  }
> > >
> > > +static void rockchip_pcie_legacy_int_handler(struct irq_desc *desc)
> > > +{
> > > +     struct irq_chip *chip = irq_desc_get_chip(desc);
> > > +     struct rockchip_pcie *rockchip = irq_desc_get_handler_data(desc);
> > > +     unsigned long reg, hwirq;
> > > +
> > > +     chained_irq_enter(chip, desc);
> > > +
> > > +     reg = rockchip_pcie_readl_apb(rockchip,
> > > PCIE_CLIENT_INTR_STATUS_LEGACY);
> > > +
> > > +     for_each_set_bit(hwirq, &reg, 8)
> >
> > 8? And yet:
> >
> > #define PCI_NUM_INTX        4
> >
> > So whatever bits are set above bit 3, you are feeding garbage
> > to the irqdomain code.
> 
> There are 8 bits in total, the top four are for the TX interrupts, for
> which EP mode is not yet supported by the driver.

So why aren't they excluded from the set of bits that you look at?

> I can constrain this further and let it be expanded when that support
> is added, if that works for you?

Well, you can't have INTx interrupts in EP mode (that's a TLP going
out of the device, and not something that is signalled *to* the
CPU). So the two should be mutually exclusive.

> 
> >
> > > +             generic_handle_domain_irq(rockchip->irq_domain, hwirq);
> > > +
> > > +     chained_irq_exit(chip, desc);
> > > +}
> > > +
> > > +static void rockchip_intx_mask(struct irq_data *data)
> > > +{
> > > +     struct rockchip_pcie *rockchip = irq_data_get_irq_chip_data(data);
> > > +     unsigned long flags;
> > > +     u32 val;
> > > +
> > > +     /* disable legacy interrupts */
> > > +     raw_spin_lock_irqsave(&rockchip->irq_lock, flags);
> > > +     val = HIWORD_UPDATE_BIT(PCIE_LEGACY_INT_ENABLE);
> > > +     val |= PCIE_LEGACY_INT_ENABLE;
> > > +     rockchip_pcie_writel_apb(rockchip, val,
> > > PCIE_CLIENT_INTR_MASK_LEGACY);
> > > +     raw_spin_unlock_irqrestore(&rockchip->irq_lock, flags);
> >
> > This is completely busted. INTx lines must be controlled individually.
> > If I disable one device's INTx output, I don't want to see the
> > interrupt firing because another one has had its own enabled.
> 
> Okay, that makes sense. I'm hitting the entire block when it should be
> the individual IRQ.
> I also notice some drivers protect this with a spinlock while others
> do not, how should this be handled?

It obviously depends on how the HW. works. If this is a shared
register using a RMW sequence, then you need some form of mutual
exclusion in order to preserve the atomicity of the update.

If the HW supports updating the masks using a set of hot bits (with
separate clear/set registers), than there is no need for locking.  In
your case PCIE_CLIENT_INTR_MASK_LEGACY seems to support this odd
"write-enable" feature which can probably be used to implement a
lockless access, something like:

	void mask(struct irq_data *d)
	{
		u32 val = BIT(d->hwirq + 16) | BIT(d->hwirq);
		writel_relaxed(val, ...);
	}

	void mask(struct irq_data *d)
	{
		u32 val = BIT(d->hwirq + 16);
		writel_relaxed(val, ...);
	}

Another thing is that it is completely unclear to me what initialises
these interrupts the first place (INTR_MASK_LEGACY, INTR_EN_LEGACY).
Are you relying on the firmware to do that for you?

Thanks,

	M.

-- 
Without deviation from the norm, progress is not possible.