Re: [Qemu-devel] [PATCH] hw/char/cmsdk-apb-timer: Correctly identify and set one-shot mode

[Guenter Roeck]

On Mon, Jul 02, 2018 at 07:06:17PM +0100, Peter Maydell wrote:
> On 2 July 2018 at 18:57, Guenter Roeck <address@hidden> wrote:
> > On Mon, Jul 02, 2018 at 06:10:36PM +0100, Peter Maydell wrote:
> >> On 2 July 2018 at 18:05, Guenter Roeck <address@hidden> wrote:
> >> > I don't think I'll have more time available to spend on this
> >> > anytime soon. If it works for you, just drop the patch. I can
> >> > keep it local in my branch. My version of qemu is different anyway,
> >> > after all.
> >>
> >> I'd kind of like to fix the bug, though. I'd just like you
> >> to provide the image you're testing with...
> >>
> >
> > Ah, but that won't help. We are dealing with a heisenbug, and me using
> > a different host than you (or even a different gcc version) may be
> > sufficient to make the problem disappear on your system.
> 
> True, but we can at least rule out some differences between
> your setup and mine. (For instance, your kernel config and
> version and rootfs are all going to be different from what
> I've been testing.)
> 
> > The mps2 images use high resolution timers, which is likely what is
> > triggering here. My working assumption is that those timers trigger at
> > the wrong time because the timer is set up in periodic instead of one-time
> > mode. With that, behavior is more or less random. On some systems it will
> > work, on some systems it won't. Since we can not agree that setting the
> > timer in periodic mode instead of one-shot mode is a problem, I don't
> > really see how we can proceed. I would somehow have to prove to you that
> > there is an observable problem in the first place, well beyond of what
> > I already did. That might take a lot of time - much more than I would
> > be able to spend on it.
> 
> I certainly believe you that there's an observable problem.
> I just disagree about your proposed fix, because I don't see
> from the code why it would make any difference. So I want to
> get to a point where I can run exactly the same thing you're
> running on my system to have a look at what's happening.
> 

Please try http://server.roeck-us.net/qemu/an385/

vmlinux is plain v4.18-rc3 compiled with the provided defconfig and
the provided toolchain (position independent, created with buildroot).
defconfig is mps2_defconfig+CONFIG_BLK_DEV_INITRD=y. Older kernels work
as well (I tested as far back as 4.9.y), with the same configuration.

qemu is from the master-local branch in my repository, built with
https://github.com/groeck/linux-build-test/blob/master/qemu/buildall.sh.

The boot wrapper was generated from
https://github.com/groeck/linux-build-test/tree/master/rootfs/arm/arm-m3-bootwrapper/.

run.sh should boot the system into a shell (at least it does for me).

Guenter