Re: next-20130627 breaks i.MX6 sabre sd UART console

From: Mark Rutland
Date: Mon Jul 01 2013 - 06:05:19 EST


On Sat, Jun 29, 2013 at 03:07:38AM +0100, Stephen Boyd wrote:
> On 06/28, Stephen Boyd wrote:
> > On 06/28/13 09:58, Stehle Vincent-B46079 wrote:
> > > From: linux-next-owner@xxxxxxxxxxxxxxx [mailto:linux-next-owner@xxxxxxxxxxxxxxx] On Behalf Of Stephen Boyd
> > > (..)
> > >> Do you have debug_ll support to get some serial logs?
> > > Hi,
> > >
> > > Thanks for your concern on the matter. You are right, there is debug_ll support for i.MX6. If I activate it for next-20130628 imx_v6_v7_defconfig, that only "buys" me one additional line, though:
> > >
> > > Uncompressing Linux... done, booting the kernel.
> >
> > Oh and also add earlyprintk to the command line and enable the
> > earlyprintk config. It should buy you some more lines.
> >
> > >
> > >> Also could you try booting with maxcpus=1 on the kernel commandline?
> > > Good idea; it goes much further with this one (until it crashes in wm8962_probe(), which is as far as I go when I revert 064706514ec3fea740c2656e03c4f01f6a551ac4).
> >
> > Interesting. Thanks for the test.
> >
> > >
> > >> And it would be useful to have the output of /proc/timer_list if you have it from a working device.
> > > Sure; I pasted it below.
> >
> > Is this an SMP capable device? What devicetree blob are you using?
> >
> > >
> > >> I believe you have a per-cpu TWD timer and a global imx-gpt?
> > > I think you are right from what I see in /proc/interrupts.
> >
> > Well oddly I don't see any twd in the timer list but that may be because
> > you're not running more than one CPU? It would be good to get the output
> > when you run SMP if it has that support.
> >
>
> I tried to recreate this probelm an an smp vexpress qemu and I
> couldn't do it. This is with next-20130628. If you can get
> debug_ll and earlyprinkt working it would be useful if you can
> apply the attached patch and give back the logs. I'm confused how
> the dummy timers are causing problems.

Similarly I couldn't reproduce the issue on a physical A9x4 coretile
with next-20130627. In case there's some weird interaction with the
imx-gpt registration, I tried dropping the rating on the sp804 down to
200 to match, and also removed CLOCKEVT_FEAT_PERIODIC from its features,
but was still able to get to userspace.

If I've followed the code correctly, I see mxc_timer_init is called from
the imx6q_clocks_init, which is marked with CLK_OF_DECLARE. I had a
suspicion that there was an ordering issue that we were somehow
irritating, but given booting with one CPU worked, that doesn't seem to
be the case. On me A9x4. I can't move the sp804 to CLK_OF_INIT to check
as I cannot guarantee I'll have the requisite clocks.

I'm also left scratching my head on this one. A dmesg and timer list
from a working SMP boot would be useful.

Thanks,
Mark.

>
> diff --git a/kernel/time/tick-broadcast.c b/kernel/time/tick-broadcast.c
> index 9d96a54..3795c9c 100644
> --- a/kernel/time/tick-broadcast.c
> +++ b/kernel/time/tick-broadcast.c
> @@ -92,6 +92,8 @@ void tick_install_broadcast_device(struct clock_event_device *dev)
> if (!try_module_get(dev->owner))
> return;
>
> + pr_info("%s is now the broadcast\n", dev->name);
> +
> clockevents_exchange_device(cur, dev);
> if (cur)
> cur->event_handler = clockevents_handle_noop;
> diff --git a/kernel/time/tick-common.c b/kernel/time/tick-common.c
> index edd45f6..24548aa 100644
> --- a/kernel/time/tick-common.c
> +++ b/kernel/time/tick-common.c
> @@ -275,6 +275,8 @@ void tick_check_new_device(struct clock_event_device *newdev)
> struct tick_device *td;
> int cpu;
>
> + pr_info("Checking %s\n", newdev->name);
> +
> cpu = smp_processor_id();
> if (!cpumask_test_cpu(cpu, newdev->cpumask))
> goto out_bc;
> @@ -286,10 +288,14 @@ void tick_check_new_device(struct clock_event_device *newdev)
> if (!tick_check_percpu(curdev, newdev, cpu))
> goto out_bc;
>
> + pr_info("%d: %s is per-cpu\n", cpu, newdev->name);
> +
> /* Preference decision */
> if (!tick_check_preferred(curdev, newdev))
> goto out_bc;
>
> + pr_info("%d: %s is preferred\n", cpu, newdev->name);
> +
> if (!try_module_get(newdev->owner))
> return;
>
> @@ -312,6 +318,7 @@ out_bc:
> /*
> * Can the new device be used as a broadcast device ?
> */
> + pr_info("%d: %s is trying broadcast\n", cpu, newdev->name);
> tick_install_broadcast_device(newdev);
> }
>
>
> --
> Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
> hosted by The Linux Foundation
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/