Re: [PATCH RFC 05/12] mm/slab: abstract slabobj_ext.objcg access

[Ye Liu]

在 2026/7/21 03:58, Omar Sandoval 写道:
> On Fri, Jul 17, 2026 at 12:00:29PM +0200, Vlastimil Babka (SUSE) wrote:
>> On 7/15/26 12:10, Vlastimil Babka (SUSE) wrote:
>>> In preparation for changes to the structure, abstract access to the
>>> objcg field with a slab_obj_ext_objcgp() function.
>>> Rename the field to _objcg to make an unexpected direct access a compile
>>> error.
>>>
>>> No functional change intended.
>>>
>>> Signed-off-by: Vlastimil Babka (SUSE) <vbabka@xxxxxxxxxx>
>> Hm sashiko finds [1] that this is breaks tools/mm/show_page_info.py
>>
>> But I think it was already buggy. get_memcg_info() is supposed to return
>> memcg info for a page but in case of MEMCG_DATA_OBJEXTS there's no such info
>> for the whole page, there is only for individual slab objects and it doesn't
>> know which object so it effectively gets the first one. I think it should
>> just stop handling MEMCG_DATA_OBJEXTS, but that's a fix that should be sent
>> unrelated to this series.
>>
>> There's also tools/cgroup/memcg_slabinfo.py and that I think has been
>> already broken with memalloc profiling as it assuems slab.memcg_data is
>> still a raw struct obj_cgroup * and not struct slabobj_ext.
>>
>> I think it will be easier to fix it at once to handle the layout after this
>> series than trying to fix it first to handle the pre-series layout and then
>> update it with every relevant change.
>>
>> [1]
>> https://sashiko.dev/#/patchset/20260715-b4-objext_split-v1-0-9a49c4ccf4c3@xxxxxxxxxx?part=5
> FWIW, if these scripts were in the drgn repo with test cases, these
> breakages would be caught by our CI and someone (usually me) would fix
> them up promptly :)

When I first used the drgn tool, I noticed that some scripts were directly      
merged into the kernel repository, so I merged show_page_info.py into the kernel
repository as well. Later, I found that merging into the drgn repository was    
also a good option, so I merged it into the drgn repository as well,            
and maintain it there. Therefore, I currently use show_page_info.py regularly   
and update it promptly if I find any issues.                                    

-- 
Thanks,
Ye Liu