Re: [PATCH v2 3/3] btrfs: file-item: refactor btrfs_lookup_bio_sums() to handle out-of-order bvecs

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

On 2020/10/29 下午9:54, David Sterba wrote:
> On Thu, Oct 29, 2020 at 08:43:12PM +0800, Qu Wenruo wrote:
>> On 2020/10/29 下午7:50, Nikolay Borisov wrote:
>>> On 29.10.20 г. 9:12 ч., Qu Wenruo wrote:
>>>> +{
>>>> +	struct btrfs_csum_item *item = NULL;
>>>> +	struct btrfs_key key;
>>>> +	u32 csum_size = btrfs_super_csum_size(fs_info->super_copy);
>>>
>>> nit: Why u32, btrfs_super_csum_size is defined as returning 'int',
>>> however this function is really returning u16 since "struct btrfs_csums"
>>> is defined as u16.
>>
>> It was misguided by u32 from btrfs super block, where sectorsize,
>> nodesize are all u32.
>>
>> Any recommendation on this? Just u16 for csum_size or follow
>> nodesize/sectorsize to use u32?
>
> u32 is ok, this generates a bit better assembly than u16. I'm about to
> send the patchset lifting it to fs_info and it uses u32.
>
BTW, if you're going to change that, what about adding some thing like
nodesize_bits/sectorsize_bits/csumsize_bits?

There are a lot of locations that u64/u32 leads to 32bit compiling
error, and a bit shifting would be a much better solution than div_u64()
macro.

Thanks,
Qu