Hi,

On Wednesday, September 17, 2014 05:39:29 PM Kevin Hilman wrote:

Thomas Abraham [off-list ref] writes:

quoted

On Thu, Sep 11, 2014 at 12:16 AM, Kevin Hilman [off-list ref] wrote:

quoted

Tyler Baker [off-list ref] writes:

quoted

Exynos5420-based Arndale octa boards have recently started failing boot
tests due to imprecise external aborts.  This only appears to happen
when using exynos_defconfig and boots fine with multi_v7_defconfig.  The
issue seems to be intermittent, so is not reliably reproducable and
difficult to bisect.  Here are a few boot logs from recent
mainline/linux-next kernels that are failing:

FYI, I'm seeing the same periodic aborts.  For example, here's my boot
of next-20140910:
http://images.armcloud.us/kernel-ci/next/next-20140910/arm-exynos_defconfig/boot-exynos5420-arndale-octa.html

However, my userspace is much simpler and doesn't seem to cause a panic,
so my boot tests report passing. (I should fixup my scripts so these
imprecise aborts are reported as a FAIL.)

I'm glad you pointed out that it happens only with exynos_defconfig and
not multi_v7_defconfig because I noticed that too.  I haven't had the
time to track it any further than that, so maybe the exynos folks can
help track it down from here.

Thanks for reporting this,

Kevin

Hi Tyler, Kevin,

From the bootlog you have shared,

[    1.060016] CPU4: failed to come online
[    2.070031] CPU5: failed to come online
[    3.080049] CPU6: failed to come online
[    4.090066] CPU7: failed to come online
[    4.090099] Brought up 4 CPUs
[    4.090109] SMP: Total of 4 processors activated.
[    4.090119] CPU: WARNING: CPU(s) started in wrong/inconsistent
modes (primary CPU mode 0x13)
[    4.090128] CPU: This may indicate a broken bootloader or firmware.

Would it be possible to set max cpus to 1, disable switcher and try

Which "max cpus" do you mean?  For NR_CPUS the lower limit is 2.

quoted

again. I don't have a arndale octa board but I have tested mainline
kernel with smdk5420 board. It boots all eight CPUs, switcher works
fine and there are no imprecise aborts seen.

Sorry for the delay, I'm travelling this week.

FWIW, the same CPU boot failures you hilight above are happening on
multi_v7_defconfig[1] which is not getting the imprecise abort.  This is
only happening on exynos_defconfig[2], so I'm curious why you think the
switcher or NR_CPUS might be the issues.

Anyways, I narrowed this down a bit and discovered it's
CONFIG_EXYNOS5420_MCPM=y that's the root cause.  If I use
exynos_defconfig and then disable that option, I don't get any more
imprecise aborts.

I have exactly the same issue with Exynos5420 Arndale Octa.

CONFIG_EXYNOS5420_MCPM was enabled by commit fc3791f3a95d
("ARM: exynos_defconfig: Update exynos_defconfig") which
get merged into v3.17-rc1.  It seems that this part of
the patch should be reverted for the final v3.17 kernel.

Kevin

[1] http://images.armcloud.us/kernel-ci/mainline/v3.17-rc5-25-g8ba4caf/arm-multi_v7_defconfig/boot-exynos5420-arndale-octa.html
[2] http://images.armcloud.us/kernel-ci/mainline/v3.17-rc5-25-g8ba4caf/arm-exynos_defconfig/boot-exynos5420-arndale-octa.html

Best regards,
--
Bartlomiej Zolnierkiewicz
Samsung R&D Institute Poland
Samsung Electronics