Re: linux-next: Tree for Aug 31 (new arm, arm64, s390 failures)

[Guenter Roeck <linux@roeck-us.net>]

Hi Marc,

On 08/31/2015 09:18 AM, Marc Zyngier wrote:
> On Mon, 31 Aug 2015 08:47:07 -0700
> Guenter Roeck <linux@roeck-us.net> wrote:
>
>> Hi Marc,
>>
>> On 08/31/2015 08:31 AM, Marc Zyngier wrote:
>>> On Mon, 31 Aug 2015 07:17:36 -0700
>>> Guenter Roeck <linux@roeck-us.net> wrote:
>>>
>>> Hi Guenter,
>>>
>>>> Qemu test results:
>>>> 	total: 85 pass: 74 fail: 11
>>>> Failed tests:
>>>> 	arm:vexpress-a9:arm_vexpress_defconfig:vexpress-v2p-ca9
>>>> 	arm:vexpress-a15:arm_vexpress_defconfig:vexpress-v2p-ca15-tc1
>>>> 	arm:vexpress-a9:multi_v7_defconfig:vexpress-v2p-ca9
>>>> 	arm:vexpress-a15:multi_v7_defconfig:vexpress-v2p-ca15-tc1
>>>> 	arm:realview-pb-a8:arm_realview_pb_defconfig
>>>> 	arm:realview-eb:arm_realview_eb_defconfig
>>>> 	mips:fuloong2e_defconfig
>>>> 	xtensa:dc232b:lx60:xtensa_defconfig
>>>> 	xtensa:dc232b:kc705:xtensa_defconfig
>>>> 	xtensa:dc233c:ml605:generic_kc705_defconfig
>>>> 	xtensa:dc233c:kc705:generic_kc705_defconfi
>>>>
>>>> Notable new failures (since next-20150828) are the s390 build failures,
>>>> the arm64 build failure, and the arm qemu test failures.
>>>>
>>>
>>> [...]
>>>
>>>> The qemu arm tests all fail silently, meaning there is no console
>>>> output. Bisect points to 'irqchip/GIC: Convert to EOImode == 1'.
>>>> Bisect log attached.
>>>
>>> Could you give me a qemu command-line I can use to track this down?
>>> Real HW seems happy enough, from what I can see...
>>>
>>
>> That is what I was most concerned about :-(. Unfortunately, it
>> affects many of the most widely used arm qemu emulations, so it
>> would be very desirable to get this fixed, either in the kernel
>> or in qemu.
>>
>> See https://github.com/groeck/linux-build-test, specifically
>> https://github.com/groeck/linux-build-test/tree/master/rootfs/arm/.
>> run-qemu-arm.sh includes the various command lines and configurations.
>>
>> Note that some of the tests require a patched version of qemu.
>> The tests failing above should all work with the latest published
>> version of qemu (2.4), though.
>>
>> Please let me know if there is anything I can do to help tracking
>> this down.
>
> I give it a quick go with qemu 2.1.2 as installed on my laptop, and the
> results are interesting:
>
> - With -next as of today, qemu segfaults. Humpffff.
>
> - If I use my branch that contains the EOImode==1 patch, the system
>    boots normally.
>
> So there is an interaction between this patch and whatever is in -next
> at the moment, but that patch on its own is not what triggers the issue.
>
Looks like it.

I did a couple of tests.
- Revert 'irqchip/GIC: Don't deactivate interrupts forwarded to a guest'.
   Same problem.
- Revert both 'irqchip/GIC: Don't deactivate interrupts forwarded to a guest'
   and 'irqchip/GIC: Convert to EOImode == 1'.
   Problem is no longer seen.

There are several other patches in drivers/irqchip/irq-gic.c since 4.2.

4c2880b31c70 irqchip/gic: Ensure gic_cpu_if_up/down() programs correct GIC instance
567e5a014848 irqchip/gic: Only allow the primary GIC to set the CPU map
4b979e4c611c Merge branch 'linus' into irq/core
0d3f2c92e004 irqchip/gic: Remove redundant gic_set_irqchip_flags
aec89ef72ba6 irqchip/gic: Enable SKIP_SET_WAKE and MASK_ON_SUSPEND
5b29264c659c irqchip: Use irq_desc_get_xxx() to avoid redundant lookup of irq_desc
4d83fcf8d615 irqchip/gic: Consolidate chained IRQ handler install/remove
41a83e06e2bb irqchip: Prepare for local stub header removal

Maybe there is an interaction between those and your patch ?

> I need to build a more recent version of qemu, but the above doesn't
> fill be with confidence...
>
My patched version of qemu 2.4 doesn't crash for me, it simply hangs.
Not that this is much better.

Thanks,
Guenter