Re: [PATCH v5 8/8] mm: huge_memory: enable debugfs to split huge pages to any order.

[Zi Yan <ziy@nvidia.com>]

[-- Attachment #1: Type: text/plain, Size: 7645 bytes --]

On 4 Mar 2024, at 13:25, Aishwarya TCV wrote:

> On 04/03/2024 15:57, Zi Yan wrote:
>> On 4 Mar 2024, at 10:44, Aishwarya TCV wrote:
>>
>>> On 04/03/2024 14:58, Zi Yan wrote:
>>>> On 4 Mar 2024, at 4:50, Aishwarya TCV wrote:
>>>>
>>>>> On 01/03/2024 21:10, Zi Yan wrote:
>>>>>> On 1 Mar 2024, at 15:02, Zi Yan wrote:
>>>>>>
>>>>>>> On 1 Mar 2024, at 14:37, Zi Yan wrote:
>>>>>>>
>>>>>>>> On 1 Mar 2024, at 4:51, Aishwarya TCV wrote:
>>>>>>>>
>>>>>>>>> On 26/02/2024 20:55, Zi Yan wrote:
>>>>>>>>>> From: Zi Yan <ziy@nvidia.com>
>>>>>>>>>>
>>>>>>>>>> It is used to test split_huge_page_to_list_to_order for pagecache THPs.
>>>>>>>>>> Also add test cases for split_huge_page_to_list_to_order via both
>>>>>>>>>> debugfs.
>>>>>>>>>>
>>>>>>>>>> Signed-off-by: Zi Yan <ziy@nvidia.com>
>>>>>>>>>> ---
>>>>>>>>>>  mm/huge_memory.c                              |  34 ++++--
>>>>>>>>>>  .../selftests/mm/split_huge_page_test.c       | 115 +++++++++++++++++-
>>>>>>>>>>  2 files changed, 131 insertions(+), 18 deletions(-)
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>> Hi Zi,
>>>>>>>>>
>>>>>>>>> When booting the kernel against next-master(20240228)with Arm64 on
>>>>>>>>> Marvell Thunder X2 (TX2), the kselftest-mm test 'split_huge_page_test'
>>>>>>>>> is failing in our CI (with rootfs over NFS). I can send the full logs if
>>>>>>>>> required.
>>>>>>>>>
>>>>>>>>> A bisect (full log below) identified this patch as introducing the
>>>>>>>>> failure. Bisected it on the tag "next-20240228" at repo
>>>>>>>>> "https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git".
>>>>>>>>>
>>>>>>>>> This works fine on  Linux version 6.8.0-rc6
>>>>>>>>
>>>>>>>> Hi Aishwarya,
>>>>>>>>
>>>>>>>> Can you try the attached patch and see if it fixes the failure? I changed
>>>>>>>> the test to accept XFS dev as input, mount XFS on a temp folder under /tmp,
>>>>>>>> and skip if no XFS is mounted.
>>>>>>>
>>>>>>> Please try this updated one. It allows you to specify a XFS device path
>>>>>>> in SPLIT_HUGE_PAGE_TEST_XFS_PATH env variable, which is passed to
>>>>>>> split_huge_page_test in run_vmtests.sh. It at least allow CI/CD to run
>>>>>>> the test without too much change.
>>>>>>
>>>>>> OK. This hopefully will be my last churn. Now split_huge_page_test accepts
>>>>>> a path that is backed by XFS and run_vmtest.sh creates a XFS image in /tmp,
>>>>>> mounts it in /tmp, and gives the path to split_huge_page_test. I tested
>>>>>> it locally and it works. Let me know if you have any issue. Thanks.
>>>>>>
>>>>>> --
>>>>>> Best Regards,
>>>>>> Yan, Zi
>>>>>
>>>>> Hi Zi,
>>>>>
>>>>> Tested the patch by applying it on next-20240304. Logs from our CI with
>>>>> rootfs over nfs is attached below. "Bail out! cannot remove tmp dir:
>>>>> Directory not empty" is still observed.
>>>>
>>>> Hi Aishwarya,
>>>>
>>>> Do you have the config file for the CI kernel? And /tmp is also on nfs?
>>>> Any detailed information about CI machine environment? I cannot reproduce
>>>> the error locally, either on bare metal or VM. Maybe because my /tmp is
>>>> not NFS mounted?
>>>>
>>>
>>> Hi Zi,
>>>
>>> Please find the details below. Hope it helps.
>>>
>>> Do you have the config file for the CI kernel?
>>> - We are using:
>>> defconfig+https://github.com/torvalds/linux/blob/master/tools/testing/selftests/mm/config
>>>
>>> And /tmp is also on nfs?
>>> - Yes
>>>
>>> Any detailed information about CI machine environment?
>>> - We are running the test using LAVA device Cavium Thunder X2 (TX2),
>>> - We have very similar rootfs as - nfsrootfs:
>>> https://storage.kernelci.org/images/rootfs/debian/bullseye-kselftest/20240129.0/arm64/full.rootfs.tar.xz
>>> - We are using grub boot method over nfs
>>> - Additionally Ryan mentioned "Looks like it is failing because he is
>>> trying to delete the temp dir with rmdir() but rmdir() requires the
>>> directory to be empty, which it is not."
>>
>> Hi Aishwarya,
>>
>> Thank you for the information and I am able to reproduce it on a NFS folder.
>> The error comes from that the opened test files are not munmapped and their
>> file descriptors are not closed in the skip path. NFS creates .nfsXXX files
>> for them, making the temp folder not empty.
>>
>> The attached patch cleans up properly and works on a NFS folder. Let me know
>> if it works on your side. Thanks.
>>
>> --
>> Best Regards,
>> Yan, Zi
>
> Hi Zi,
>
> Tested the attached patch on next-20240304. Confirming that the test is
> running fine. Test run log is attached below.
>
> Test run log:
> # # ------------------------------
> # # running ./split_huge_page_test
> # # ------------------------------
> # # TAP version 13
> # # 1..12
> # # ok 1 Split huge pages successful
> # # ok 2 Split PTE-mapped huge pages successful
> # # # Please enable pr_debug in split_huge_pages_in_file() for more info.
> # # # Please check dmesg for more information
> # # ok 3 File-backed THP split test done
> <6>[ 1769.710429] split_huge_page (111119): drop_caches: 3
> <6>[ 1769.710429] split_huge_page (111119): drop_caches: 3
> # # # No large pagecache folio generated, please provide a filesystem
> supporting large folio
> # # ok 4 # SKIP Pagecache folio split skipped
> <6>[ 1775.302315] split_huge_page (111119): drop_caches: 3
> <6>[ 1775.302315] split_huge_page (111119): drop_caches: 3
> # # # No large pagecache folio generated, please provide a filesystem
> supporting large folio
> # # ok 5 # SKIP Pagecache folio split skipped
> <6>[ 1780.924147] split_huge_page (111119): drop_caches: 3
> <6>[ 1780.924147] split_huge_page (111119): drop_caches: 3
> # # # No large pagecache folio generated, please provide a filesystem
> supporting large folio
> # # ok 6 # SKIP Pagecache folio split skipped
> <6>[ 1786.524931] split_huge_page (111119): drop_caches: 3
> <6>[ 1786.524931] split_huge_page (111119): drop_caches: 3
> # # # No large pagecache folio generated, please provide a filesystem
> supporting large folio
> # # ok 7 # SKIP Pagecache folio split skipped
> <6>[ 1792.112869] split_huge_page (111119): drop_caches: 3
> <6>[ 1792.112869] split_huge_page (111119): drop_caches: 3
> # # # No large pagecache folio generated, please provide a filesystem
> supporting large folio
> # # ok 8 # SKIP Pagecache folio split skipped
> <6>[ 1797.718863] split_huge_page (111119): drop_caches: 3
> <6>[ 1797.718863] split_huge_page (111119): drop_caches: 3
> # # # No large pagecache folio generated, please provide a filesystem
> supporting large folio
> # # ok 9 # SKIP Pagecache folio split skipped
> <6>[ 1803.332343] split_huge_page (111119): drop_caches: 3
> <6>[ 1803.332343] split_huge_page (111119): drop_caches: 3
> # # # No large pagecache folio generated, please provide a filesystem
> supporting large folio
> # # ok 10 # SKIP Pagecache folio split skipped
> <6>[ 1808.947913] split_huge_page (111119): drop_caches: 3
> <6>[ 1808.947913] split_huge_page (111119): drop_caches: 3
> # # # No large pagecache folio generated, please provide a filesystem
> supporting large folio
> # # ok 11 # SKIP Pagecache folio split skipped
> <6>[ 1814.537995] split_huge_page (111119): drop_caches: 3
> <6>[ 1814.537995] split_huge_page (111119): drop_caches: 3
> # # # No large pagecache folio generated, please provide a filesystem
> supporting large folio
> # # ok 12 # SKIP Pagecache folio split skipped
> # # # Totals: pass:3 fail:0 xfail:0 xpass:0 skip:9 error:0
> # # [PASS]
> # ok 51 split_huge_page_test
> # # -------------------

Thanks a lot for testing.

--
Best Regards,
Yan, Zi

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 854 bytes --]