Re: [PATCH v14 0/5] mm/vmalloc: free unused pages on vrealloc() shrink

[Shivam Kalra]

On 20/05/26 03:01, Shivam Kalra wrote:
> On 19/05/26 23:37, Andrew Morton wrote:
>> On Tue, 19 May 2026 17:42:13 +0530 Shivam Kalra via B4 Relay <devnull+shivamkalra98.zohomail.in@xxxxxxxxxx> wrote:
>>
>>> This series implements the TODO in vrealloc() to unmap and free unused
>>> pages when shrinking across a page boundary.
>>
>> Thanks, I'll add this to mm.git's mm-new branch for testing.  A few
>> days hence I'll hopefully move it into the mm-unstable branch for
>> linux-next exposure.
>>
>> Sashiko is still saying things.  Please double-check that it's all been
>> dealt with and there's nothing new in there?
>>
>> 	https://sashiko.dev/#/patchset/20260519-vmalloc-shrink-v14-0-70b96ee3e9c9@xxxxxxxxxxx
> I've reviewed all five Sashiko findings:
> 
> - Issues 1+2 (32-bit shift overflow in nr_pages << PAGE_SHIFT):
>   nr_pages is unsigned int; a pre-existing type choice in struct
>   vm_struct. Uladzislau reviewed this and explicitly asked to
>   drop the (size_t) cast in v11.
> 
> - Issue 3 (VM_ALLOC areas with nr_pages==0): Valid point.
>   pcpu_get_vm_areas() creates VM_ALLOC areas but manages pages through
>   its own path and never sets nr_pages. With patch 3, vread_iter would
>   skip these areas instead of reading them; a minor /proc/kcore
>   observability regression. I'll send a follow-up fix to fall back to
>   get_vm_area_size() when nr_pages is 0.
> 
> - Issue 4 (kmemleak scanner race): __delete_object() clears
>   OBJECT_ALLOCATED under object->lock, and
>   scan_object() re-checks that flag under the same lock after
>   cond_resched(). Additionally, kmemleak_free_part()
>   is called before vunmap_range(), so pages are still mapped during
>   any scan window.
> 
> - Issue 5 (__GFP_ZERO on grow-after-shrink): Not a regression per
>   Sashiko's own assessment. The API doc already requires callers to
>   consistently pass __GFP_ZERO for every call.
> 

Andrew, could you please fold this into
"mm/vmalloc: use physical page count in vread_iter() for VM_ALLOC areas"?

pcpu_get_vm_areas() creates VM_ALLOC areas without setting nr_pages.
Add an nr_pages check to fall back to get_vm_area_size() for these
areas.

-			size = (vm->flags & VM_ALLOC) ?
+			size = (vm->flags & VM_ALLOC && vm->nr_pages) ?