Re: Fwd: Kernel memory management bug at mm/migrate.c:662 when flushing caches

[Bagas Sanjaya]

On 02/08/2023 18:25, Linux regression tracking (Thorsten Leemhuis) wrote:
> On 02.08.23 13:08, Bagas Sanjaya wrote:
>>
>> I notice a regression report on Bugzilla [1]. Quoting from it:
>>
>>> I hit this kernel bug on the latest 6.3.9 kernel after executing this script to cleanup hugepages from the kernel before booting up a Windows 11 VM with QEMU (otherwise I don't have enough contiguous memory to allocate the pages to the VM)
>>>
>>> snip
>>> if [[ $VM_ACTION == 'prepare' ]];
>>> then
>>>     sync
>>>     echo 3 > /proc/sys/vm/drop_caches
>>>     echo 1 > /proc/sys/vm/compact_memory
>>> endsnip
>>>
>>> Attached is the full QEMU script that I used. I do use ZFS as a root filesystem, as you can see from the loaded modules.
> 
> Bagas, FWIW, I'd totally understand if developers will ignore this
> (remains to be seen if that is the case, maybe we are lucky and somebody
> will take a look), as I think you for now shouldn't have forwarded this
> for two reasons:
> 
> * 6.3.y is old and EOL; testing mainline or at least a fresh 6.4.y
> kernel) would have been a must here.
> * with out-of-tree modules like ZFS anything can happen, the user is own
> its own.
> 

I have already asked the reporter to try reproducing this with
"officially-supported" other filesystems (e.g. ext4, xfs, or btrfs).
The reporter also stated that he was now trying the mainline (but not
rc release, just latest stable).

> As I can see from the bug both things will likely clear up soon, hence
> waiting would have been wise here.
> 
> Please in the future do not forward such bugs, as developers might
> otherwise start to ignore mails wrt to regression tracking -- which we
> really need to avoid, as that will make things a lot harder.
> 

Thanks for another tip!

-- 
An old man doll... just what I always wanted! - Clara