Re: [PATCH v3 4/9] mm: keep writable entries writable on protection changes

["Carlos López" <clopez@suse.de>]

On 11/13/26 7:56 PM, David Hildenbrand (Arm) wrote:
> On 11/12/26 15:20, Carlos López wrote:
>> change_protection() applies vma->vm_page_prot, which lacks the write bit
>> for CoW and write-notified mappings. MM_CP_TRY_CHANGE_WRITABLE makes
>> entries writable again if a write fault would only set the write bit.
>> In shared mappings, it relies on the dirty bit for this, as a clean
>> entry might still need a write fault to notify the FS.
>>
>> However, this misses clean entries that were already writable. For
>> example, a uffd-wp unprotect makes shmem PTEs read-only if they were
>> mapped by a read fault before uffd-wp was registered, so the next write
>> to each of them faults.
> 
> See my other reply, adding such optimizations without any motivation why we
> should even care is not really what we want.
> 
> Drop it from your series and keep the MM changes to a bare minimum please.
> 
> Or is there a real dependency for your work?
> 

Yes there is, but I clearly did not explain it properly.

The whole point of the series is to avoid zaps when no permission bits
are removed by a protection change, e.g. mprotect(PROT_READ |
PROT_WRITE) on a read-only VMA. My proposal is for change_protection()
to signal this is the case to MMU notifiers with a new flag. However,
change_protection() sometimes write-protects entries that were writable,
which makes the flag wrong, so this patch fixes that.

I'll be sure to explain this properly.