Re: [PATCH v3 0/9] btrfs: remove ASSERT()s for folio_order() and folio_test_large()

[Qu Wenruo <wqu@suse.com>]

在 2025/4/4 06:24, David Sterba 写道:
> On Mon, Mar 31, 2025 at 03:17:35PM +1030, Qu Wenruo wrote:
>>
>>
>> 在 2025/3/18 01:43, David Sterba 写道:
>>> On Mon, Mar 17, 2025 at 02:55:02PM +0100, David Sterba wrote:
>>>> On Mon, Mar 17, 2025 at 05:40:45PM +1030, Qu Wenruo wrote:
>>>>> [CHANGELOG]
>>>>> v3:
>>>>> - Prepare btrfs_end_repair_bio() to support larger folios
>>>>>     Unfortunately folio_iter structure is way too large compared to
>>>>>     bvec_iter, thus it's not a good idea to convert bbio::saved_iter to
>>>>>     folio_iter.
>>>>>
>>>>>     Thankfully it's not that complex to grab the folio from a bio_vec.
>>>>>
>>>>> - Add a new patch to prepare defrag for larger data folios
>>>>
>>>> I was not expecting v3 as the series was in for-next so I did some edits
>>> [...]
>>>
>>> Scratch that, this is not the same series.
>>>
>>
>> BTW, any extra commends needs to be addressed? (E.g. should I merge them
>> all into a single patch?)
> 
> I think the patch separation is good, please leave it as it is.
> 
>> I notice several small comment conflicts ("larger folio -> large
>> folio"), but is very easy to resolve (local branch updated).
>>
>> There is a newer series that is already mostly reviewed:
>>
>> https://lore.kernel.org/linux-btrfs/cover.1743239672.git.wqu@suse.com/
>>
>> That relies on this series, and I'm already doing some basic (fsstress
>> runs) large folio tests now.
>>
>> So I'm wondering can I push the series now, or should I continue waiting
>> for extra reviews?
> 
> Please add it to for-next. I did only a light review, we'll find more
> things during testing.

Thanks, but it looks like the read repair and defrag part (the last two 
patches) should be delayed a little.

I'll push the first 7 safe ASSERT() removals into for-next first.


As I finally fixed the last bug exposed by fsstress runs, I can continue 
with fstests runs.

Instead of the untested defrag and read-repair patches, I can do proper 
test and fix (since the read-repair one seems to cause bugs during my 
local tests) bugs in them.

Thanks,
Qu