Re: [PATCH v6] arm pl011 serial: support multi-irq request

[Robin Murphy]

On 2021-08-16 08:42, Bing Fan wrote:
> At present, i think a focus of our discussion is whether this patch is 
> necessary.
>
> As for the other points you mentioned, I think they can be used as code 
> review comments.
>
>
> Yes, as you described below, most dts files have only one interrupt, but 
> not all platforms are like this.
>
> The scene I'm encountering now is the latter: the interrupt lines of the 
> uart is connected to the gic separately
>
> so the dts should be define like this:
>
>                 duart1: serial@5E139000 {
>                          compatible = "arm,pl011", "arm,primecell";
>                          reg = <0x00 0x5E139000 0x0 0x1000>;
>                          interrupts = <GIC_SPI 178 IRQ_TYPE_LEVEL_HIGH>,
>                                  <GIC_SPI 179 IRQ_TYPE_LEVEL_HIGH>,
>                                  <GIC_SPI 180 IRQ_TYPE_LEVEL_HIGH>,
>                                  <GIC_SPI 181 IRQ_TYPE_LEVEL_HIGH>;
>                          clocks = <&sysclk>;
>                          clock-names = "apb_pclk";
>                  };

Apologies for being unclear - the point I was implying is that of course 
you can do that in practice, but if you run that DTS through `make 
dtbs_check` it will fail. The binding needs extending to make it valid 
to specify more than one interrupt, and that's a separate patch and 
discussion in itself (simply increasing "maxitems" for the "interrupts" 
property is not enough to be robust).

Robin.

> The current tty-master code cannot meet this scenario, so I submitted 
> this patch.
>
>
>
>
>
> 在 2021/8/13 下午10:37, Robin Murphy 写道:
> > [ +Russell as the listed PL011 maintainer ]
> >
> > On 2021-08-13 04:31, Bing Fan wrote:
> > > From: Bing Fan <tombinfan@xxxxxxxxxxx>
> > >
> > > In order to make pl011 work better, multiple interrupts are
> > > required, such as TXIM, RXIM, RTIM, error interrupt(FE/PE/BE/OE);
> > > at the same time, pl011 to GIC does not merge the interrupt
> > > lines(each serial-interrupt corresponding to different GIC hardware
> > > interrupt), so need to enable and request multiple gic interrupt
> > > numbers in the driver.
> > >
> > > Signed-off-by: Bing Fan <tombinfan@xxxxxxxxxxx>
> > > ---
> > >   drivers/tty/serial/amba-pl011.c | 39 +++++++++++++++++++++++++++++++--
> > >   1 file changed, 37 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/drivers/tty/serial/amba-pl011.c 
> > > b/drivers/tty/serial/amba-pl011.c
> > > index e14f3378b8a0..eaac3431459c 100644
> > > --- a/drivers/tty/serial/amba-pl011.c
> > > +++ b/drivers/tty/serial/amba-pl011.c
> > > @@ -1701,6 +1701,41 @@ static void pl011_write_lcr_h(struct 
> > > uart_amba_port *uap, unsigned int lcr_h)
> > >       }
> > >   }
> > >   +static void pl011_release_multi_irqs(struct uart_amba_port *uap, 
> > > unsigned int max_cnt)
> > > +{
> > > +    struct amba_device *amba_dev = container_of(uap->port.dev, 
> > > struct amba_device, dev);
> > > +    int i;
> > > +
> > > +    for (i = 0; i < max_cnt; i++)
> > > +        if (amba_dev->irq[i])
> > > +            free_irq(amba_dev->irq[i], uap);
> >
> > When you request the IRQs you break at the first zero, so this could 
> > potentially try to free IRQs that you haven't requested, if there 
> > happen to be any nonzero values beyond that. Maybe that can never 
> > happen, but there seems little need for deliberate inconsistency here.
> >
> > > +}
> > > +
> > > +static int pl011_allocate_multi_irqs(struct uart_amba_port *uap)
> > > +{
> > > +    int ret = 0;
> > > +    int i;
> > > +    unsigned int virq;
> > > +    struct amba_device *amba_dev = container_of(uap->port.dev, 
> > > struct amba_device, dev);
> > > +
> > > +    pl011_write(uap->im, uap, REG_IMSC);
> > > +
> > > +    for (i = 0; i < AMBA_NR_IRQS; i++) {
> >
> > It's not clear where these extra IRQs are expected to come from given 
> > that the DT binding explicitly defines only one :/
> >
> > > +        virq = amba_dev->irq[i];
> > > +        if (virq == 0)
> > > +            break;
> > > +
> > > +        ret = request_irq(virq, pl011_int, IRQF_SHARED, 
> > > dev_name(&amba_dev->dev), uap);
> >
> > Note that using dev_name() here technically breaks user ABI - scripts 
> > looking in /proc for an irq named "uart-pl011" will no longer find it.
> >
> > Furthermore, the "dev" cookie passed to request_irq is supposed to be 
> > globally unique, which "uap" isn't once you start registering it 
> > multiple times. If firmware did describe all the individual PL011 IRQ 
> > outputs on a system where they are muxed to the same physical IRQ 
> > anyway, you'd end up registering ambiguous IRQ actions here. Of course 
> > in practice you might still get away with that, but it is technically 
> > wrong.
> >
> > Robin.
> >
> > > +        if (ret) {
> > > +            dev_err(uap->port.dev, "request %u interrupt failed\n", 
> > > virq);
> > > +            pl011_release_multi_irqs(uap, i - 1);
> > > +            break;
> > > +        }
> > > +    }
> > > +
> > > +    return ret;
> > > +}
> > > +
> > >   static int pl011_allocate_irq(struct uart_amba_port *uap)
> > >   {
> > >       pl011_write(uap->im, uap, REG_IMSC);
> > > @@ -1753,7 +1788,7 @@ static int pl011_startup(struct uart_port *port)
> > >       if (retval)
> > >           goto clk_dis;
> > >   -    retval = pl011_allocate_irq(uap);
> > > +    retval = pl011_allocate_multi_irqs(uap);
> > >       if (retval)
> > >           goto clk_dis;
> > >   @@ -1864,7 +1899,7 @@ static void pl011_shutdown(struct uart_port 
> > > *port)
> > >         pl011_dma_shutdown(uap);
> > >   -    free_irq(uap->port.irq, uap);
> > > +    pl011_release_multi_irqs(uap, AMBA_NR_IRQS);
> > >         pl011_disable_uart(uap);