[PATCH 2/2] irqchip: nvic: Use GENERIC_IRQ_MULTI_HANDLER

[Ard Biesheuvel]

On Thu, 2 Dec 2021 at 15:12, Vladimir Murzin <vladimir.murzin at arm.com> wrote:
>
> On 12/2/21 11:05 AM, Ard Biesheuvel wrote:
> > On Thu, 2 Dec 2021 at 10:45, Arnd Bergmann <arnd at arndb.de> wrote:
> >>
> >> On Thu, Dec 2, 2021 at 10:33 AM Vladimir Murzin <vladimir.murzin at arm.com> wrote:
> >>>
> >>> On 12/2/21 9:23 AM, Arnd Bergmann wrote:
> >>>> On Thu, Dec 2, 2021 at 9:35 AM Ard Biesheuvel <ardb at kernel.org> wrote:
> >>>>>
> >>>>> It seems trivial enough to add that for v7m as well, so I will include
> >>>>> a patch for it in my v3. But I won't be able to test it, though.
> >>>>
> >>>> Ok, perfect. Adding Jesse to Cc, they are in the process of adding another
> >>>> Cortex-M based platform and should be able to test this by merging in your
> >>>> changes from [1] after you have uploaded the latest version.
> >>>
> >>> I'm happy to test it as well. I'm wondering if there are specific scenarios
> >>> (or maybe some tests) to consider?
> >>
> >> If the IRQ stacks fail in some way, it most likely won't boot at all, so as long
> >> as you can get into user space, I'd be 99% confident that it's good.
> >>
> >> For the MMU based platforms, the series lets us detect stack overflows
> >> when either the per-thread stack or the IRQ stack runs into an unmapped
> >> vmalloc page, and there are some unit tests that exercise this.
> >> I don't think you can do this with the MPU though, and certainly not when
> >> the MPU is disabled.
> >>
> >
> > Latest version of the code is here, if anyone wants to test it:
> > https://git.kernel.org/pub/scm/linux/kernel/git/ardb/linux.git/log/?h=arm-up-ti-in-task-v3
> >
>
> I'm afraid it is broken (even w/o last patch). I'm getting
>
> [    0.071350] Mount-cache hash table entries: 1024 (order: 0, 4096 bytes, linear)
> [    0.071747] Mountpoint-cache hash table entries: 1024 (order: 0, 4096 bytes, linear)
> [    0.084497] bad: scheduling from the idle thread!
> [    0.084744] CPU: 0 PID: 0 Comm: kworker/0:0H Not tainted 5.16.0-rc1-0f57a2aad139 #2428
> [    0.085357] Hardware name: Generic DT based system
>
> $ git bisect log
> git bisect start
> # good: [fa55b7dcdc43c1aa1ba12bca9d2dd4318c2a0dbf] Linux 5.16-rc1
> git bisect good fa55b7dcdc43c1aa1ba12bca9d2dd4318c2a0dbf
> # bad: [e42c59caa4d8f37444388cc388bfc55653bf00dc] ARM: v7m: enable support for IRQ stacks
> git bisect bad e42c59caa4d8f37444388cc388bfc55653bf00dc
> # good: [cbf7b30e70b7e2eacebd885674675746e93583f6] ARM: entry: rework stack realignment code in svc_entry
> git bisect good cbf7b30e70b7e2eacebd885674675746e93583f6
> # good: [02c0315277721c4fc2082850da0714127594473d] irqchip: nvic: Use GENERIC_IRQ_MULTI_HANDLER
> git bisect good 02c0315277721c4fc2082850da0714127594473d
> # good: [9cc24032abc2847a8e61e55e2ec9bd34e1eff1c4] ARM: percpu: add SMP_ON_UP support
> git bisect good 9cc24032abc2847a8e61e55e2ec9bd34e1eff1c4
> # good: [8adfea395669037ee79620b8b534e101d5ce65e6] ARM: smp: defer TPIDRURO update for SMP v6 configurations too
> git bisect good 8adfea395669037ee79620b8b534e101d5ce65e6
> # bad: [0f57a2aad13912dde5746ffc5f43bd46aa3d8b2d] ARM: implement THREAD_INFO_IN_TASK for uniprocessor systems
> git bisect bad 0f57a2aad13912dde5746ffc5f43bd46aa3d8b2d
> # first bad commit: [0f57a2aad13912dde5746ffc5f43bd46aa3d8b2d] ARM: implement THREAD_INFO_IN_TASK for uniprocessor systems
>
> At the first glance it seems that __switch_to in entry-v7m.S missing set_current
>

Thanks for testing, and thanks for bisecting. Unfortunately, I can't
easily test these changes myself.

Could you please check whether the following change would be
sufficient to fix this?

https://git.kernel.org/pub/scm/linux/kernel/git/ardb/linux.git/commit/?h=tmp

Thanks,
Ard.