Re: [PATCH v2] irqchip: add TS7800v1 fpga based controller driver

From: Marc Zyngier
Date: Fri Oct 21 2022 - 13:18:41 EST


On Fri, 21 Oct 2022 17:19:02 +0100,
firas ashkar <firas.ashkar@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> On Fri, 2022-10-21 at 10:22 +0100, Marc Zyngier wrote:
> > On Thu, 20 Oct 2022 15:13:51 +0100,
> > Firas Ashkar <firas.ashkar@xxxxxxxxxxxxxxxxxxxx> wrote:
> > >
> > > 1. add TS-7800v1 fpga based irq controller driver, and
> > > 2. add related memory and irq resources
> > >
> > > By default only mapped FPGA interrupts will be chained/multiplexed,
> > > these
> > > chained interrupts will then be available to other device drivers
> > > to
> > > request via the corresponding Linux IRQs.
> > >
> > > $ cat /proc/cpuinfo
> > > processor       : 0
> > > model name      : Feroceon rev 0 (v5l)
> > > BogoMIPS        : 333.33
> > > Features        : swp half thumb fastmult edsp
> > > CPU implementer : 0x41
> > > CPU architecture: 5TEJ
> > > CPU variant     : 0x0
> > > CPU part        : 0x926
> > > CPU revision    : 0
> > >
> > > Hardware        : Technologic Systems TS-78xx SBC
> > > Revision        : 0000
> > > Serial          : 0000000000000000
> > > $
> > >
> > > $ cat /proc/interrupts
> > >            CPU0
> > >   1:        902  orion_irq     Level     orion_tick
> > >   4:        795  orion_irq     Level     ttyS0
> > >  13:          0  orion_irq     Level     ehci_hcd:usb2
> > >  18:          0  orion_irq     Level     ehci_hcd:usb1
> > >  22:         69  orion_irq     Level     eth0
> > >  23:        171  orion_irq     Level     orion-mdio
> > >  29:          0  orion_irq     Level     mv_crypto
> > >  31:          2  orion_irq     Level     mv_xor.0
> > >  65:          1  ts7800-irqc   0 Edge      ts-dmac-cpupci
> > > Err:          0
> > > $
> > >
> > > $ uname -a
> > > Linux ts-7800 6.1.0-rc1 #2 PREEMPT Mon Oct 17 11:19:12 EDT 2022
> > > armv5tel
> > >  GNU/Linux
> > > $
> > >
> > > Signed-off-by: Firas Ashkar <firas.ashkar@xxxxxxxxxxxxxxxxxxxx>
> > > ---
> > >
> > > Notes:
> > >     Changes in V2
> > >     * limit the commit message
> >
> > Well, there is still a lot of work to do. Most of this commit message
> > could be reduced to a single paragraph:
> >
> > "Add support for FooBar IRQ controller usually found on Zorglub
> > platform."
> >
> > The rest is plain obvious.
> nack, commit message is fine as is!

Allow me to be the judge of that.

> >
> > >     * format comments in source code
> > >     * use raw spin locks to protect mask/unmask ops
> > >     * use 'handle_edge_irq' and 'irq_ack' logic
> > >     * remove 'irq_domain_xlate_onecell'
> > >     * remove unnecessary status flags
> > >     * use 'builtin_platform_driver' helper routine
> > >
> > > :100644 100644 2f4fe3ca5c1a ed8378893208 M      arch/arm/mach-
> > > orion5x/ts78xx-fpga.h
> > > :100644 100644 af810e7ccd79 d319a68b7160 M      arch/arm/mach-
> > > orion5x/ts78xx-setup.c
> > > :100644 100644 7ef9f5e696d3 d184fb435c5d
> > > M      drivers/irqchip/Kconfig
> > > :100644 100644 87b49a10962c b022eece2042
> > > M      drivers/irqchip/Makefile
> > > :000000 100644 000000000000 e975607fa4d5
> > > A      drivers/irqchip/irq-ts7800.c
> > >  arch/arm/mach-orion5x/ts78xx-fpga.h  |   1 +
> > >  arch/arm/mach-orion5x/ts78xx-setup.c |  55 +++++
> > >  drivers/irqchip/Kconfig              |  12 +
> > >  drivers/irqchip/Makefile             |   1 +
> > >  drivers/irqchip/irq-ts7800.c         | 347
> > > +++++++++++++++++++++++++++
> > >  5 files changed, 416 insertions(+)
> > >
> > > diff --git a/arch/arm/mach-orion5x/ts78xx-fpga.h b/arch/arm/mach-
> > > orion5x/ts78xx-fpga.h
> > > index 2f4fe3ca5c1a..ed8378893208 100644
> > > --- a/arch/arm/mach-orion5x/ts78xx-fpga.h
> > > +++ b/arch/arm/mach-orion5x/ts78xx-fpga.h
> > > @@ -32,6 +32,7 @@ struct fpga_devices {
> > >         struct fpga_device      ts_rtc;
> > >         struct fpga_device      ts_nand;
> > >         struct fpga_device      ts_rng;
> > > +       struct fpga_device      ts_irqc;
> > >  };
> > >  
> > >  struct ts78xx_fpga_data {
> > > diff --git a/arch/arm/mach-orion5x/ts78xx-setup.c b/arch/arm/mach-
> > > orion5x/ts78xx-setup.c
> > > index af810e7ccd79..d319a68b7160 100644
> > > --- a/arch/arm/mach-orion5x/ts78xx-setup.c
> > > +++ b/arch/arm/mach-orion5x/ts78xx-setup.c
> > > @@ -322,6 +322,49 @@ static void ts78xx_ts_rng_unload(void)
> > >         platform_device_del(&ts78xx_ts_rng_device);
> > >  }
> > >  
> > > +/*****************************************************************
> > > ************
> > > + * fpga irq controller
> > > +
> > > *******************************************************************
> > > *********/
> >
> > [...]
> >
> > Sorry, but I have to ask: what part of "we're not taking any
> > additional non-DT changes to these obsolete setups" did I fail to
> > accurately communicate?
> >
> > Until this board is entirely converted to DT, I'm not taking any
> > irqchip changes other than the most obvious bug fixes.
> as long as this board is present in the kernel in its current legacy
> form, this is a valid patch!

No. There is a cut of point. Code that we would taken 10 years ago
isn't necessarily valid anymore. We want to improve the kernel as a
whole, and not keep it in the past.

>
> >
> > [...]
> >
> > > +static void ts7800_irq_mask(struct irq_data *d)
> > > +{
> > > +       struct ts7800_irq_data *data =
> > > irq_data_get_irq_chip_data(d);
> > > +       u32 fpga_mask_reg = readl(data->base + IRQ_MASK_REG);
> > > +       u32 mask_bits = 1 << d->hwirq;
> > > +       unsigned long flags;
> > > +
> > > +       raw_spin_lock_irqsave(&data->lock, flags);
> > > +       writel(fpga_mask_reg & ~mask_bits, data->base +
> > > IRQ_MASK_REG);
> > > +       writel(fpga_mask_reg & ~mask_bits, data->bridge +
> > > IRQ_MASK_REG);
> > > +       raw_spin_unlock_irqrestore(&data->lock, flags);
> >
> > OMFG. What do you expect this protects? Same question applies to all
> > the instances where you take this pointless lock.
> don't jump now
> the locks added as per your previous comment, quoting:
> "I know your HW is UP, but seeing these RMW sequences without a lock
> makes me jump."
> On this single CPU based arch TS78xx, locks are waste of cpu cycles,
> also note that IRQs/preemption are/is already off in this context
>
> maybe you meant adding locks as to promote general correct coding ?

Let me spell it out for you: RMW means Read-Modify-Write. Putting a
lock around a *write only* serves zero purpose. It is just overhead,
and it is incorrect.

>
> or maybe it was like this previous nonsense comment, quoting :
> "We don't remove interrupt controllers. What happens if someone still
> had a mapping?"

And I stand by it.

>
>
> >
> > [...]
> >
> > > +static void ts7800_ic_chained_handle_irq(struct irq_desc *desc)
> > > +{
> > > +       struct ts7800_irq_data *data =
> > > irq_desc_get_handler_data(desc);
> > > +       struct irq_chip *chip = irq_desc_get_chip(desc);
> > > +       u32 mask_bits = readl(data->base + IRQ_MASK_REG);
> > > +       u32 status = readl(data->bridge);
> > > +
> > > +       chained_irq_enter(chip, desc);
> > > +
> > > +       if (unlikely(status == 0)) {
> > > +               handle_bad_irq(desc);
> > > +               goto out;
> > > +       }
> > > +
> > > +       do {
> > > +               unsigned int bit = __ffs(status);
> > > +               int irq = irq_find_mapping(data->domain, bit);
> > > +
> > > +               if (irq && (mask_bits & BIT(bit)))
> > > +                       generic_handle_irq(irq);
> >
> > Again, this is not appropriate. I've pointed you to the right API in
> > my previous review of this patch.
> 'generic_handle_domain_irq' causing some issues

Which issue?

> >
> > [...]
> >
> > > +static struct platform_driver ts7800_ic_driver = {
> > > +       .probe  = ts7800_ic_probe,
> > > +       .remove = ts7800_ic_remove,
> > > +       .id_table       = ts7800v1_ic_ids,
> > > +       .driver = {
> > > +               .name = DRIVER_NAME,
> > > +               .probe_type = PROBE_PREFER_ASYNCHRONOUS,
> > > +       },
> > > +};
> > > +builtin_platform_driver(ts7800_ic_driver);
> >
> > Again, this isn't appropriate either, and I pointed it out last
> > time. We have specific helpers for irqchip, and using them isn't
> > optional. But of course, you'll need to move to DT for that.
> >
> > Anyway, this is the last time I review this patch until you convert
> > the corresponding platform to DT.
> no problems, though have to note this is not constructive!

I've pointed out a bunch of issues, and provided advise on how you can
fix them. That's constructive. A non constructive approach would be to
just ignore your patch. If that's what you prefer, please say so.

Thanks,

M.

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