Re: [PATCHv2 1/8] khugepaged: Add self test

From: Zi Yan <ziy@nvidia.com>
To: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: <akpm@linux-foundation.org>,
	Andrea Arcangeli <aarcange@redhat.com>,
	"Yang Shi" <yang.shi@linux.alibaba.com>, <linux-mm@kvack.org>,
	<linux-kernel@vger.kernel.org>,
	"Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Subject: Re: [PATCHv2 1/8] khugepaged: Add self test
Date: Fri, 10 Apr 2020 11:03:09 -0400	[thread overview]
Message-ID: <511724CE-3FF3-4535-BDB1-D4B9ACE9F3DA@nvidia.com> (raw)
In-Reply-To: <20200410145804.fczyvec4pngpep6t@box>

[-- Attachment #1: Type: text/plain, Size: 2796 bytes --]

On 10 Apr 2020, at 10:58, Kirill A. Shutemov wrote:

> External email: Use caution opening links or attachments
>
>
> On Fri, Apr 10, 2020 at 10:36:58AM -0400, Zi Yan wrote:
>> On 10 Apr 2020, at 7:47, Kirill A. Shutemov wrote:
>>
>>> External email: Use caution opening links or attachments
>>>
>>>
>>> On Mon, Apr 06, 2020 at 10:59:52AM -0400, Zi Yan wrote:
>>>> I ran this test with all patches from this series applied to Linus’s tree, but still see several failures. Is it expected?
>>>> The config file is attached. Let me know if I miss anything. BTW, I am running in a VM.
>>>>
>>>> Thanks.
>>>>
>>>> The output:
>>>>
>>>> ➜  ~ sudo ./khugepaged
>>>> Save THP and khugepaged settings... OK
>>>> Adjust settings... OK
>>>> Allocate huge page on fault... OK
>>>> Split huge PMD on MADV_DONTNEED... OK
>>>> Collapse fully populated PTE table.... Fail
>>>
>>> I was able to reproduce the issue. And it's fun failure mode.
>>>
>>> How did you get the test case inside the VM? Copy-paste source using 'cat'
>>> or something similar inside the VM?
>>
>> First of all, the failure above was from a bare metal and was the only
>> failure I saw, whereas I saw more failures in my VM. The test program
>> was not messed up in either environment.
>
> Hm. In the quote you are saying "BTW, I am running in a VM".

Sorry, misread the email thread. I was referring to another my email on running
tests on a bare metal, where only “Collapse with max_ptes_swap pages swapped out”
failed.

Here is the link to the email:
https://lore.kernel.org/linux-mm/C66E1309-2069-495B-BACD-7F3282C6EC7D@nvidia.com/

>
>>
>> For VM failures I mentioned before, I used scp to copy the source code
>> into the VM. My VM has its port 22 forwarded to host’s port 11022. “-net
>> user,hostfwd=tcp::11022-:22”. I also copied a binary into my VM and saw
>> the same failures.
>>
>> I kinda think the failures are not related to your patches but something else.
>>
>>>
>>> It screwed up CHECK_HUGE_FMT and CHECK_SWAP_FMT for me. Double back slash
>>> was converted to single. As result check_huge() and check_swap() gave the
>>> false-negative result all the time.
>>
>> It was not my case, since CHECK_HUGE_FMT and CHECK_SWAP_FMT from my
>> khugepaged.c match your patch code.
>>
>>>
>>> Could you check that the source of the test-case is not mangled and
>>> re-test if it is.
>>
>> I can confirm that the test-case is not mangled. I think it must be my
>> VM setup or kernel configuration.
>>
>> Do you mind sharing your .config file with me and which kernel commit
>> you apply the patches on top of in your setup? I can look into it and
>> check what the problem is.
>
> My config is attached.

Thanks.

—
Best Regards,
Yan Zi

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