Re: [RFC v2 09/11] block/bdev: lift block size restrictions and use common definition

[John Garry <john.g.garry@oracle.com>]

On 16/12/2024 09:19, Ming Lei wrote:
> On Mon, Dec 16, 2024 at 4:58 PM John Garry <john.g.garry@oracle.com> wrote:
>>
>> On 14/12/2024 03:10, Luis Chamberlain wrote:
>>> index 167d82b46781..b57dc4bff81b 100644
>>> --- a/block/bdev.c
>>> +++ b/block/bdev.c
>>> @@ -157,8 +157,7 @@ int set_blocksize(struct file *file, int size)
>>>        struct inode *inode = file->f_mapping->host;
>>>        struct block_device *bdev = I_BDEV(inode);
>>>
>>> -     /* Size must be a power of two, and between 512 and PAGE_SIZE */
>>> -     if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
>>> +     if (blk_validate_block_size(size))
>>>                return -EINVAL;
>>
>> I suppose that this can be sent as a separate patch to be merged now.
> 
> There have been some bugs found in case that PAGE_SIZE == 64K, and I
> think it is bad to use PAGE_SIZE for validating many hw/queue limits, we might
> have to fix them first.

I am just suggesting to remove duplicated code, as these checks are same 
as blk_validate_block_size()

> 
> Such as:

Aren't the below list just enforcing block layer requirements? And so 
only block drivers need fixing for PAGE_SIZE > 4K (or cannot be used for 
PAGE_SIZE > 4K), right?

> 
> 1) blk_validate_limits()
> 
> - failure if max_segment_size is less than 64K
 > > - max_user_sectors
> 
> if (lim->max_user_sectors < PAGE_SIZE / SECTOR_SIZE)
>         return -EINVAL;
> 
> 2) bio_may_need_split()
> 
> - max_segment_size may be less than 64K
> 
> Thanks,



>