Re: [PATCH 4/4] xfs/260: Move xfs/260 to generic

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

On 2020/6/10 23:59, Darrick J. Wong wrote:
> On Wed, Jun 10, 2020 at 02:16:45PM +0800, Xiao Yang wrote:
>> On 2020/6/10 0:42, Darrick J. Wong wrote:
>>> On Wed, Jun 03, 2020 at 10:01:15PM +0800, Xiao Yang wrote:
>>>> Both ext4 and xfs support per-inode DAX flag now so move it to generic.
>>>>
>>>> Signed-off-by: Xiao Yang<yangx.jy@cn.fujitsu.com>
>>>> ---
>>>>    tests/{xfs/260 =>   generic/602} | 12 ++++++++----
>>>>    tests/generic/602.out          |  2 ++
>>>>    tests/generic/group            |  1 +
>>>>    tests/xfs/260.out              |  2 --
>>>>    tests/xfs/group                |  1 -
>>>>    5 files changed, 11 insertions(+), 7 deletions(-)
>>>>    rename tests/{xfs/260 =>   generic/602} (91%)
>>>>    create mode 100644 tests/generic/602.out
>>>>    delete mode 100644 tests/xfs/260.out
>>>>
>>>> diff --git a/tests/xfs/260 b/tests/generic/602
>>>> similarity index 91%
>>>> rename from tests/xfs/260
>>>> rename to tests/generic/602
>>>> index fbdc4cd8..0cd20f31 100755
>>>> --- a/tests/xfs/260
>>>> +++ b/tests/generic/602
>>>> @@ -2,7 +2,7 @@
>>>>    # SPDX-License-Identifier: GPL-2.0
>>>>    # Copyright (c) 2017 Red Hat Inc.  All Rights Reserved.
>>>>    #
>>>> -# FS QA Test 260
>>>> +# FS QA Test 602
>>>>    #
>>>>    # Test per-inode DAX flag by mmap direct/buffered IO.
>>>>    #
>>>> @@ -28,7 +28,7 @@ _cleanup()
>>>>    # remove previous $seqres.full before test
>>>>    rm -f $seqres.full
>>>>
>>>> -_supported_fs xfs
>>>> +_supported_fs generic
>>>>    _supported_os Linux
>>>>    _require_scratch_dax_mountopt "dax"
>>>>    _require_test_program "feature"
>>>> @@ -121,8 +121,12 @@ do_tests()
>>>>    	t_dax_flag_mmap_dio $((64 * 1024 * 1024))
>>>>    }
>>>>
>>>> -# make xfs 2Mb aligned for PMD fault testing
>>>> -_scratch_mkfs "-d su=2m,sw=1">   /dev/null 2>&1
>>>> +# make 2Mb aligned for PMD fault testing
>>>> +mkfs_opts=""
>>>> +[ "$FSTYP" = "ext4" ]&&   mkfs_opts="-E stride=512,stripe_width=1"
>>>> +[ "$FSTYP" = "xfs" ]&&   mkfs_opts="-d su=2m,sw=1"
>>> /me wonders if this ought to be smart enough to sniff out the hugepage
>>> size on the test machine...
>> Hi Darrick,
>>
>> PMD page is always 2M, I am not sure why we need to sniff out the hugepage
>> size?  Could you give me an example?
> arm64, which can have 512M hugepages.
> ppc64, which can have 16M huge pages (I think?)
Hi Darrick,

After reading kernel xfs code, do you mean that PMD page fault is only 
used by hugepage?
----------------------------------------------------------------
xfs/xfs_file.c:
...
1223 static vm_fault_t
1224 xfs_filemap_fault(
1225         struct vm_fault         *vmf)
1226 {
1227         /* DAX can shortcut the normal fault path on write faults! */
1228         return __xfs_filemap_fault(vmf, PE_SIZE_PTE,
1229                         IS_DAX(file_inode(vmf->vma->vm_file)) &&
1230                         (vmf->flags & FAULT_FLAG_WRITE));
1231 }
1232
1233 static vm_fault_t
1234 xfs_filemap_huge_fault(
1235         struct vm_fault         *vmf,
1236         enum page_entry_size    pe_size)
1237 {
1238         if (!IS_DAX(file_inode(vmf->vma->vm_file)))
1239                 return VM_FAULT_FALLBACK;
1240
1241         /* DAX can shortcut the normal fault path on write faults! */
1242         return __xfs_filemap_fault(vmf, pe_size,
1243                         (vmf->flags & FAULT_FLAG_WRITE));
1244 }
...
----------------------------------------------------------------

Another qeustion: ext4 seems not use PMD page fault,does I miss something?
----------------------------------------------------------------
ext4/file.c:
...
705 static vm_fault_t ext4_dax_fault(struct vm_fault *vmf)
706 {
707         return ext4_dax_huge_fault(vmf, PE_SIZE_PTE);
708 }
709
710 static const struct vm_operations_struct ext4_dax_vm_ops = {
711         .fault          = ext4_dax_fault,
712         .huge_fault     = ext4_dax_huge_fault,
...
----------------------------------------------------------------

Best Regards,
Xiao Yang
> --D
>
>> Best Regards,
>> Xiao Yang
>>> --D
>>>
>>>> +
>>>> +_scratch_mkfs "$mkfs_opts">   /dev/null 2>&1
>>>>
>>>>    # mount with dax option
>>>>    _scratch_mount "-o dax"
>>>> diff --git a/tests/generic/602.out b/tests/generic/602.out
>>>> new file mode 100644
>>>> index 00000000..61976e5e
>>>> --- /dev/null
>>>> +++ b/tests/generic/602.out
>>>> @@ -0,0 +1,2 @@
>>>> +QA output created by 602
>>>> +Silence is golden
>>>> diff --git a/tests/generic/group b/tests/generic/group
>>>> index c6ce029c..4bda3772 100644
>>>> --- a/tests/generic/group
>>>> +++ b/tests/generic/group
>>>> @@ -604,3 +604,4 @@
>>>>    599 auto quick remount shutdown
>>>>    600 auto quick quota
>>>>    601 auto quick quota
>>>> +602 auto attr quick dax
>>>> diff --git a/tests/xfs/260.out b/tests/xfs/260.out
>>>> deleted file mode 100644
>>>> index 18ca517c..00000000
>>>> --- a/tests/xfs/260.out
>>>> +++ /dev/null
>>>> @@ -1,2 +0,0 @@
>>>> -QA output created by 260
>>>> -Silence is golden
>>>> diff --git a/tests/xfs/group b/tests/xfs/group
>>>> index daf54add..71c30898 100644
>>>> --- a/tests/xfs/group
>>>> +++ b/tests/xfs/group
>>>> @@ -257,7 +257,6 @@
>>>>    257 auto quick clone
>>>>    258 auto quick clone
>>>>    259 auto quick
>>>> -260 auto attr quick dax
>>>>    261 auto quick quota
>>>>    262 dangerous_fuzzers dangerous_scrub dangerous_online_repair
>>>>    263 auto quick quota
>>>> --
>>>> 2.21.0
>>>>
>>>>
>>>>
>>>
>>> .
>>>
>>
>>
>
> .
>