[PATCH v4 0/5] Write-placement hints and FDP

[Kanchan Joshi]

On 9/3/2024 8:05 PM, Christian Brauner wrote:
> On Tue, Sep 03, 2024 at 07:58:46PM GMT, Kanchan Joshi wrote:
>> Hi Amir,
>>
>>
>> On 8/26/2024 10:36 PM, Kanchan Joshi wrote:
>>> Current write-hint infrastructure supports 6 temperature-based data life
>>> hints.
>>> The series extends the infrastructure with a new temperature-agnostic
>>> placement-type hint. New fcntl codes F_{SET/GET}_RW_HINT_EX allow to
>>> send the hint type/value on file. See patch #3 commit description for
>>> the details.
>>>
>>> Overall this creates 128 placement hint values [*] that users can pass.
>>> Patch #5 adds the ability to map these new hint values to nvme-specific
>>> placement-identifiers.
>>> Patch #4 restricts SCSI to use only life hint values.
>>> Patch #1 and #2 are simple prep patches.
>>>
>>> [*] While the user-interface can support more, this limit is due to the
>>> in-kernel plumbing consideration of the inode size. Pahole showed 32-bit
>>> hole in the inode, but the code had this comment too:
>>>
>>> /* 32-bit hole reserved for expanding i_fsnotify_mask */
>>>
>>> Not must, but it will be good to know if a byte (or two) can be used
>>> here.
>>
>> Since having one extra byte will simplify things, I can't help but ask -
>> do you still have the plans to use this space (in entirety) within inode?
> 
> I just freed up 8 bytes in struct inode with what's currently in -next.
> There will be no using up those 8 bytes unless it's for a good reason
> and something that is very widely useful.

I see, so now there are two holes. Seems the plan is to co-locate these 
and reduce the size by 8 bytes. Thanks for the pointer. Primary reason 
is a bit cleaner plumbing, but I'll manage without extra space.