Re: [aarch64] Kernel crash on v5.1-rc5, __arch_copy_from_user+0x1bc/0x240

[Qu Wenruo <quwenruo.btrfs@gmx.com>]

[-- Attachment #1.1: Type: text/plain, Size: 1642 bytes --]



On 2019/4/24 下午9:50, James Morse wrote:
> Hi Qu,
> 
> On 22/04/2019 02:14, Qu Wenruo wrote:
>> On 2019/4/21 下午9:28, Matthew Wilcox wrote:
>>> On Sun, Apr 21, 2019 at 05:12:50PM +0800, Qu Wenruo wrote:
>>>> On 2019/4/21 下午4:20, Qu Wenruo wrote:
>>>>> Just hit one crash on v5.1-rc5 kernel, on ext4 filesystem.
>>>>
>>>> Well, also hit the same one in v5.0.8 kernel.
>>>>
>>>> Exact the same backtrace.
>>>>
>>>> Really not sure which part is to blame, ARM or ext4?
>>>
>>> You probably have faulty hardware:
> 
>> I tried memtester, and kernel also crashed.
>>
>> Maybe it's really faulty memory or I'm using wrong memory speed.
> 
> As another option: there may be no memory at this physical address.
> If your board only has 1G of memory, but the bootloader/DT is reporting 2G, you could see
> SError like this (assuming this is the first access to that page).

Device tree is correct and it matches.

The reason is the memory controller is not initialized correctly by Uboot.

After reverting Uboot from upstream one (with basic support) to legacy
one, no such problem anymore.

To my surprise, it's UBoot to initialize the memory controller, not the
rockchip idbloader nor the trusted image.

> 
> SError can be a fatal interrupt from the hardware, its also how the CPU tells us about
> 'asynchronous external abort'. In this case it could be an attempt to access a physical
> address where nothing exists.

So it's the equvilent of X86 machine check exception?

Anyway, at least I learned something about the new ARM world.

Thanks,
Qu

> 
> 
> Thanks,
> 
> James
> 


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 488 bytes --]