Re: [PATCH v2 2/2] mm/ksm: fix advisor_min_pages_to_scan description

在 2026/8/13 18:56, Lorenzo Stoakes (ARM) 写道:
> On Thu, Aug 13, 2026 at 11:17:22AM +0800, Jinjiang Tu wrote:
>> Both Documentation/admin-guide/mm/ksm.rst and the comment next to the
>> variable definition in mm/ksm.c describe advisor_min_pages_to_scan as a
>> lower limit of the pages_to_scan parameter, but that is not how the
>> scan-time advisor actually uses it. commit 4e5fa4f5eff6 ("mm/ksm: add ksm
>> advisor") only uses it to initialize ksm_thread_pages_to_scan when the
>> scan-time advisor is enabled. This will mislead the users.
>>
>> The semantics of advisor_min_pages_to_scan was updated in the v2 patchset
>> [1], but the documentation wasn't updated.
>>
>> Update the documentation and comment to match the semantics of
>> advisor_min_pages_to_scan.
> Umm, firstly if this were wrong you'd need to change the name of the variable
> too instad of documenting it as something completely distinct.
>
> But secondly AFAICT the logic in scan_time_advisor() suggests that it truly does
> act as a minimum bound as well as being the initial value for
> ksm_thread_pages_to_scan.

As the cover letter[1] said, 'The initial value and the max value for the pages_to_scan parameter can
be limited with:', ksm_advisor_min_pages is used as the initial value, not the min value of pages_to_scan.

The v1 patchset[2] uses ksm_advisor_min_pages as the min value of pages_to_scan, but as David suggested, the
author changed the semantics of advisor_min_pages_to_scan in v2, but forgot to update the comments and variable names.

[2] https://lore.kernel.org/all/[email protected]/

>
> So this patch just looks wrong to me.
>
> But in any case, you are making a claim here yet have provided no evidence for
> it. 'It is just the initial assignment' means nothing - if the logic can then
> only increase it up to the maximum value then it still acts as a minimum.
>
> A brief look at the logic suggests so.
>
> But in any case - the onus is on _you_ to prove that that's not happening.
>
> And even if you could do that, since it is _intended to be a lower bound_, the
> fix wouldn't be alterting comments or documentation, it would be to re-establish
> the minimum as a minimum.
>
>> Link: https://lore.kernel.org/linux-mm/[email protected]/ [1]
> This patch repeatedly states that it is the minimum time. You have failed to
> provide any analysis to suggest otherwise.
>
>> Signed-off-by: Jinjiang Tu <[email protected]>
> Assisted-by: ?
>
>> ---
>>   Documentation/admin-guide/mm/ksm.rst | 4 ++--
>>   mm/ksm.c                             | 2 +-
>>   2 files changed, 3 insertions(+), 3 deletions(-)
>>
>> diff --git a/Documentation/admin-guide/mm/ksm.rst b/Documentation/admin-guide/mm/ksm.rst
>> index c9f533b10f6f..c329ca747b8c 100644
>> --- a/Documentation/admin-guide/mm/ksm.rst
>> +++ b/Documentation/admin-guide/mm/ksm.rst
>> @@ -183,8 +183,8 @@ advisor_target_scan_time
>>           pages. The default value is 200 seconds.
>>
>>   advisor_min_pages_to_scan
>> -        specifies the lower limit of the ``pages_to_scan`` parameter of the
>> -        scan time advisor. The default is 500.
>> +        specifies the initial value of the ``pages_to_scan`` parameter of
>> +        the scan time advisor. The default is 500.
>>
>>   advisor_max_pages_to_scan
>>           specifies the upper limit of the ``pages_to_scan`` parameter of the
>> diff --git a/mm/ksm.c b/mm/ksm.c
>> index 7d5b76478f0b..4a6cf8cf5d60 100644
>> --- a/mm/ksm.c
>> +++ b/mm/ksm.c
>> @@ -342,7 +342,7 @@ static enum ksm_advisor_type ksm_advisor;
>>    * Only called through the sysfs control interface:
>>    */
>>
>> -/* At least scan this many pages per batch. */
>> +/* Initial number of pages to scan per batch. */
>>   static unsigned long ksm_advisor_min_pages_to_scan = 500;
>>
>>   static void set_advisor_defaults(void)
>> --
>> 2.43.0
>>
> --
> Cheers, Lorenzo