Re: [Qemu-devel] [PATCH v2 03/13] armv7m: Rewrite NVIC to not use any GIC code

[Peter Maydell <peter.maydell@linaro.org>]

On 16 February 2017 at 16:35, Peter Maydell <peter.maydell@linaro.org> wrote:
> From: Michael Davidsaver <mdavidsaver@gmail.com>
>
> Despite some superficial similarities of register layout, the
> M-profile NVIC is really very different from the A-profile GIC.
> Our current attempt to reuse the GIC code means that we have
> significant bugs in our NVIC.
>
> Implement the NVIC as an entirely separate device, to give
> us somewhere we can get the behaviour correct.
>
> This initial commit does not attempt to implement exception
> priority escalation, since the GIC-based code didn't either.
> It does fix a few bugs in passing:
>  * ICSR.RETTOBASE polarity was wrong and didn't account for
>    internal exceptions
>  * ICSR.VECTPENDING was 16 too high if the pending exception
>    was for an external interrupt
>  * UsageFault, BusFault and MemFault were not disabled on reset
>    as they are supposed to be
>
> Signed-off-by: Michael Davidsaver <mdavidsaver@gmail.com>
> [PMM: reworked, various bugs and stylistic cleanups]
> Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

> +    case 0x400 ... 0x5ef: /* NVIC Priority */
> +        startvec = 8 * (offset - 0x400) + NVIC_FIRST_IRQ; /* vector # */
> +
> +        for (i = 0; i < size; i++) {

Just noticed this line should be
+        for (i = 0; i < size && startvec + i < s->num_irq; i++) {

which brings it into line with the nvic_sysreg_read() code
and prevents an assert() in set_prio() if the guest writes to
registers beyond the end of the implemented IRQ range.

> +            set_prio(s, startvec + i, (value >> (i * 8)) & 0xff);
> +        }
> +        nvic_irq_update(s);
> +        return;

Unless there's some other problem that means I need
to respin anyway I propose to just squash in that fix.

thanks
-- PMM