Re: [PATCH v2 6/6] xfs/260: Move xfs/260 to generic

[Xiao Yang <yangx.jy@cn.fujitsu.com>]

On 2020/6/30 23:50, Ira Weiny wrote:
> On Tue, Jun 30, 2020 at 03:08:50PM +0800, Xiao Yang wrote:
>> On 2020/6/24 2:08, Ira Weiny wrote:
>>> On Tue, Jun 23, 2020 at 04:51:31PM +0800, Xiao Yang wrote:
>>>> On 2020/6/19 23:15, Ira Weiny wrote:
>>>>> On Thu, Jun 18, 2020 at 11:35:56PM +0800, Xiao Yang wrote:
>>>>>> On 6/18/20 5:48 AM, Ira Weiny wrote:
>>>>>>> On Wed, Jun 17, 2020 at 05:32:04PM +0800, Xiao Yang wrote:
>>>>>>>> From: Xiao Yang<yangx.jy@cn.fujitsu.com>
>>> [snip]
>>>
>>>>>>>      }
>>>>>>> @@ -124,17 +151,11 @@ do_tests()
>>>>>>>      # make xfs aligned for PMD fault testing
>>>>>>>      _scratch_mkfs_geom $(_get_hugepagesize) 1>>    $seqres.full 2>&1
>>>>>>> -# mount with dax option
>>>>>>> -_scratch_mount "-o dax"
>>>>>>> -
>>>> Hi Ira,
>>>>
>>>> Why do you want to remove this combination(i.e. test per-inode DAX flag
>>>> under mounting with dax option) ?
>>>> Is it because mounting with dax option ignore FS_XFLAG_DAX flag?
>>>> I think it is a reasonable combination. :-)
>>> Yes running with the DAX mount option really does not test anything IMO.
>> Hi Ira,
>>
>> Sorry for the late reply because I was busy with other tasks last week. :-)
>>
>> After reading related code, setting/clearing FS_XFLAG_DAX have no chance to
>> change S_DAX
>> flag if mount with dax option, so I will remove this combination in v3
>> patch.
>>
>>>>>>>      tsize=$((128 * 1024 * 1024))
>>>>>>> -do_tests
>>>>>>> -_scratch_unmount
>>>>>>> -
>>>>>>>      # mount again without dax option
>>>>>>>      export MOUNT_OPTIONS=""
>>>>>>> -_scratch_mount
>>>>>>> +_scratch_mount "-o dax=inode"
>>>>>>>      do_tests
>>>>>>>      # success, all done
>>>> Could we keep _scratch_mount without dax so that this test can run on both
>>>> old and new kernel?
>>>> See the following reasons:
>>>> 1) FS_XFLAG_DAX was introduced by commit 58f88ca("xfs: introduce per-inode
>>>> DAX enablement") since 2017.
>>>> 2) _scratch_mount with dax=inode is equal to _scratch_mount without dax.
>>> I suppose that would be ok.  But being generic what happens when this runs on
>>> FS's which don't have the FS_XFLAG_DAX flag?  is it ignored?
>> Test will report notrun by _require_scratch_dax_iflag() If fs doesn't
>> support FS_XFLAG_DAX flag.
>>
>>> FWIW I believe that any FS (which includes older kernels) which do not support
>>> dax=inode have no need for this test to run.  The use of FS_XFLAG_DAX on older
>>> xfs does not do anything and simply does not exist elsewhere.  Only FS's which
>>> support dax=inode have behavior which needs to be tested IMO.
>> I just want to test FS_XFLAG_DAX on older xfs by keeping _scratch_mount
>> without dax because
>> the original test(i.e. xfs/260) is designed to test this point. :-)
> Sure you can test swapping the flag itself but it really does nothing.
>
>> BTW:  It seems that older xfs can do dax mmap by setting/clearing
>> FS_XFLAG_DAX, or do I miss something?
> No it can only use DAX with '-o dax' mount option.  Changing FS_XFLAG_DAX on
> xfs prior to 5.8 is effectively a no-op as the S_DAX flag would not be set.
>
> See "742d84290739 xfs: disable per-inode DAX flag"
Hi Ira,

Ah, I actually missed the marco disabling per-inode DAX flag.
OK, I will use dax=inode directly in v4 patch.

Best Regards,
Xiao Yang
> Ira
>
>> Best Regards,
>> Xiao Yang
>>> Ira
>>>
>>>> Best Regards,
>>>> Xiao Yang
>>>>> .
>>>>>
>>> .
>>>
>>
>>
>
> .
>