Re: [PATCH] page_ext: move up page_ext_init() to catch early page allocation if DEFERRED_STRUCT_PAGE_INIT is n

Re: [PATCH] page_ext: move up page_ext_init() to catch early page allocation if DEFERRED_STRUCT_PAGE_INIT is n

[lizhe.67]

On 2022-08-18 7:36 UTC, mhocko@suse.com wrote:
>> From: Li Zhe <lizhe.67@bytedance.com>
>> 
>> In 'commit 2f1ee0913ce5 ("Revert "mm: use early_pfn_to_nid in page_ext_init"")',
>> we call page_ext_init() after page_alloc_init_late() to avoid some panic
>> problem. It seems that we cannot track early page allocations in current
>> kernel even if page structure has been initialized early.
>> 
>> This patch move up page_ext_init() to catch early page allocations when
>> DEFERRED_STRUCT_PAGE_INIT is n. After this patch, we only need to turn
>> DEFERRED_STRUCT_PAGE_INIT to n then we are able to analyze the early page
>> allocations. This is useful especially when we find that the free memory
>> value is not the same right after different kernel booting.
>
>is this actually useful in practice? I mean who is going to disable
>DEFERRED_STRUCT_PAGE_INIT and recompile the kernel for debugging early
>allocations?

Yes it is useful. We use this method to catch the difference of early
page allocations between two kernel.

> I do see how debugging those early allocations might be useful but that
> would require a boot time option to be practical IMHO. Would it make
> sense to add a early_page_ext parameter which would essentially disable
> the deferred ipage initialization. That should be quite trivial to
> achieve (just hook into defer_init AFAICS).

It is a good idea. A cmdline parameter is a flexible and dynamic method for
us to decide whether to defer page's and page_ext's initilization. For
comparison, this patch provides a static method to decide whether to defer
page's and page_ext's initilization. They are not conflicting. My next
work is trying to achieve your idea.
--
Li Zhe

Re: [PATCH] page_ext: move up page_ext_init() to catch early page allocation if DEFERRED_STRUCT_PAGE_INIT is n

[Vlastimil Babka]

On 8/20/22 03:02, lizhe.67@bytedance.com wrote:
> On 2022-08-18 7:36 UTC, mhocko@suse.com wrote:
>>> From: Li Zhe <lizhe.67@bytedance.com>
>>> 
>>> In 'commit 2f1ee0913ce5 ("Revert "mm: use early_pfn_to_nid in page_ext_init"")',
>>> we call page_ext_init() after page_alloc_init_late() to avoid some panic
>>> problem. It seems that we cannot track early page allocations in current
>>> kernel even if page structure has been initialized early.
>>> 
>>> This patch move up page_ext_init() to catch early page allocations when
>>> DEFERRED_STRUCT_PAGE_INIT is n. After this patch, we only need to turn
>>> DEFERRED_STRUCT_PAGE_INIT to n then we are able to analyze the early page
>>> allocations. This is useful especially when we find that the free memory
>>> value is not the same right after different kernel booting.
>>
>>is this actually useful in practice? I mean who is going to disable
>>DEFERRED_STRUCT_PAGE_INIT and recompile the kernel for debugging early
>>allocations?
> 
> Yes it is useful. We use this method to catch the difference of early
> page allocations between two kernel.
> 
>> I do see how debugging those early allocations might be useful but that
>> would require a boot time option to be practical IMHO. Would it make
>> sense to add a early_page_ext parameter which would essentially disable
>> the deferred ipage initialization. That should be quite trivial to
>> achieve (just hook into defer_init AFAICS).
> 
> It is a good idea. A cmdline parameter is a flexible and dynamic method for
> us to decide whether to defer page's and page_ext's initilization. For
> comparison, this patch provides a static method to decide whether to defer
> page's and page_ext's initilization. They are not conflicting. My next
> work is trying to achieve your idea.

As we already have to pass page_owner=on parameter to enable the page
allocation tracking in the first place, maybe that alone could also disable
deffered init, and no need for another parameter?

> --
> Li Zhe

Re: [PATCH] page_ext: move up page_ext_init() to catch early page allocation if DEFERRED_STRUCT_PAGE_INIT is n'

[lizhe.67]

On 2022-08-22  7:00 UTC, vbabka@suse.cz wrote:
>> On 2022-08-18 7:36 UTC, mhocko@suse.com wrote:
>>>> From: Li Zhe <lizhe.67@bytedance.com>
>>>> 
>>>> In 'commit 2f1ee0913ce5 ("Revert "mm: use early_pfn_to_nid in page_ext_init"")',
>>>> we call page_ext_init() after page_alloc_init_late() to avoid some panic
>>>> problem. It seems that we cannot track early page allocations in current
>>>> kernel even if page structure has been initialized early.
>>>> 
>>>> This patch move up page_ext_init() to catch early page allocations when
>>>> DEFERRED_STRUCT_PAGE_INIT is n. After this patch, we only need to turn
>>>> DEFERRED_STRUCT_PAGE_INIT to n then we are able to analyze the early page
>>>> allocations. This is useful especially when we find that the free memory
>>>> value is not the same right after different kernel booting.
>>>
>>>is this actually useful in practice? I mean who is going to disable
>>>DEFERRED_STRUCT_PAGE_INIT and recompile the kernel for debugging early
>>>allocations?
>> 
>> Yes it is useful. We use this method to catch the difference of early
>> page allocations between two kernel.
>> 
>>> I do see how debugging those early allocations might be useful but that
>>> would require a boot time option to be practical IMHO. Would it make
>>> sense to add a early_page_ext parameter which would essentially disable
>>> the deferred ipage initialization. That should be quite trivial to
>>> achieve (just hook into defer_init AFAICS).
>> 
>> It is a good idea. A cmdline parameter is a flexible and dynamic method for
>> us to decide whether to defer page's and page_ext's initilization. For
>> comparison, this patch provides a static method to decide whether to defer
>> page's and page_ext's initilization. They are not conflicting. My next
>> work is trying to achieve your idea.
>
>As we already have to pass page_owner=on parameter to enable the page
>allocation tracking in the first place, maybe that alone could also disable
>deffered init, and no need for another parameter?

In my opinion, adding a new parameter is better. Page owner is not the only
feature attached to page_ext. For scalability reasons, adding a new parameter
is a more flexible method. Thanks for your advice.