Re: [PATCH 6.16 000/627] 6.16.1-rc1 review

[Zhang Yi <yi.zhang@huaweicloud.com>]

On 2025/8/13 22:53, Greg Kroah-Hartman wrote:
> On Wed, Aug 13, 2025 at 08:01:51PM +0530, Naresh Kamboju wrote:
>> Hi Greg,
>>
>>>> 2)
>>>>
>>>> The following list of LTP syscalls failure noticed on qemu-arm64 with
>>>> stable-rc 6.16.1-rc1 with CONFIG_ARM64_64K_PAGES=y build configuration.
>>>>
>>>> Most failures report ENOSPC (28) or mkswap errors, which may be related
>>>> to disk space handling in the 64K page configuration on qemu-arm64.
>>>>
>>>> The issue is reproducible on multiple runs.
>>>>
>>>> * qemu-arm64, ltp-syscalls - 64K page size test failures list,
>>>>
>>>>   - fallocate04
>>>>   - fallocate05
>>>>   - fdatasync03
>>>>   - fsync01
>>>>   - fsync04
>>>>   - ioctl_fiemap01
>>>>   - swapoff01
>>>>   - swapoff02
>>>>   - swapon01
>>>>   - swapon02
>>>>   - swapon03
>>>>   - sync01
>>>>   - sync_file_range02
>>>>   - syncfs01
>>>>
>>>> Reproducibility:
>>>>  - 64K config above listed test fails
>>>>  - 4K config above listed test pass.
>>>>
>>>> Regression Analysis:
>>>> - New regression? yes
>>>
>>> Regression from 6.16?  Or just from 6.15.y?
>>
>> Based on available data, the issue is not present in v6.16 or v6.15.
>>
>> Anders, bisected this regression and found,
>>
>>   ext4: correct the reserved credits for extent conversion
>>     [ Upstream commit 95ad8ee45cdbc321c135a2db895d48b374ef0f87 ]
>>
>> Report lore link,
>>
>> https://lore.kernel.org/stable/CA+G9fYtBnCSa2zkaCn-oZKYz8jz5FZj0HS7DjSfMeamq3AXqNg@mail.gmail.com/
> 
> Great, and that's also affecting 6.17-rc1 so we are "bug compatible"?
> :)
> 

Hi,

This issue has already fixed in 6.17-rc1 through this series:

https://lore.kernel.org/linux-ext4/20250707140814.542883-1-yi.zhang@huaweicloud.com/


To fix this issue in 6.16, it's necessary to backport the whole series
instead of just pick 5137d6c8906b ("ext4: fix insufficient credits
calculation in ext4_meta_trans_blocks()") and 95ad8ee45cdb {"ext4: correct
the reserved credits for extent conversion").  Otherwise, this will make
the problem more likely to occur.

Thanks,
Yi.