Re: [PATCH] ArmPkg: update InvalidateInstructionCacheRange to flush only to PoU

[Laszlo Ersek <lersek@redhat.com>]

On 01/29/19 12:19, xinliang wrote:
> 
> 
> On 2019/1/29 14:47, Tan Xiaojun wrote:
>> +cc Xinliang
>>
>> Sorry, I didn't react to what you discussed at the beginning, because
>> this
>> problem is mainly handled by Xinliang.
>>
>> The host we used for testing is Huawei D06 (AArch64), and its CPU chip is
>> 1620 (self-developed chip that follows the arm v8.2 specification).
>> Its vmid
>> is 16 bit.
>>
>> Thanks.
>> Xiaojun.
>>
>> On 2019/1/28 21:46, Mark Rutland wrote:
>>> On Mon, Jan 28, 2019 at 09:09:26PM +0800, Tan Xiaojun wrote:
>>>> On 2019/1/28 19:54, Laszlo Ersek wrote:
>>>>> On 01/28/19 11:46, Mark Rutland wrote:
>>>>>> On Wed, Jan 23, 2019 at 10:54:56AM +0100, Laszlo Ersek wrote:
>>>>>>> And even on the original (unspecified) hardware, the same binary
>>>>>>> works
>>>>>>> frequently. My understanding is that there are five VMs executing
>>>>>>> reboot
>>>>>>> loops in parallel, on the same host, and 4 out of 5 may hit the
>>>>>>> issue in
>>>>>>> a reasonable time period (300 reboots or so).
>>>>>> Interesting.
>>>>>>
>>>>>> Do you happen to know how many VMID bits the host has? If it has
>>>>>> an 8-bit VMID,
>>>>>> this could be indicative of some problem upon overflow.
>>>>> I'll let Tan Xiaojun (CC'd) answer this questions.
>>>>>
>>>>>> Can you point us at the host kernel?
>>>>> In the report, Tan Xiaojun wrote "4.18.0-48.el8.aarch64"; I guess that
>>>>> information is mostly useless in an upstream discussion.
>>>>> Unfortunately,
>>>>> I couldn't reproduce the symptom at all (I know nothing about the
>>>>> hardware in question), so I can't myself retest with an upstream host
>>>>> kernel.
>>>> I don't understand, what do you want me to do? What is the specific
>>>> problem?
>>> Could you let us know which CPU/system you've seen this issue with?
>>>
>>> ... and what the value of ID_AA64MMFR1_EL1.VMIDBits is?
> 
> Hi, our hardware is D06 board which compatible with armv8.1. So it
> should be 16-bit VMID.
> 
> BTW, we now only reproduce this issue on guest without shim, such as
> suse SLE15-SP1 guest which is easy to reproduce. This founding not sure
> could help for this issue.

Unfortunately, it doesn't help without access to the host -- I used the
exact same guest installer ISO from the original report, and failed to
reproduce the symptom.

... Since we've been discussing this for a while in public too, can
someone from Huawei please authorize Red Hat to open up
<https://bugzilla.redhat.com/show_bug.cgi?id=1666280> to the public? If
you agree, please comment so on the BZ itself.

Thanks!
Laszlo