Re: [PATCH] arm64/mm: Remove [PUD|PMD]_TABLE_BIT from [pud|pmd]_bad()

[Catalin Marinas]

On Thu, May 13, 2021 at 10:44:04AM +0530, Anshuman Khandual wrote:
> On 5/11/21 7:37 PM, Mark Rutland wrote:
> > On Tue, May 11, 2021 at 09:21:46AM +0530, Anshuman Khandual wrote:
> >> On 5/10/21 8:13 PM, Mark Rutland wrote:
> >>> On Mon, May 10, 2021 at 04:37:51PM +0530, Anshuman Khandual wrote:
> >>>> Semantics wise, [pud|pmd]_bad() have always implied that a given [PUD|PMD]
> >>>> entry does not have a pointer to the next level page table. This had been
> >>>> made clear in the commit a1c76574f345 ("arm64: mm: use *_sect to check for
> >>>> section maps"). Hence explicitly check for a table entry rather than just
> >>>> testing a single bit. This basically redefines [pud|pmd]_bad() in terms of
> >>>> [pud|pmd]_table() making the semantics clear.
> >>>>
> >>>> Cc: Catalin Marinas <catalin.marinas@xxxxxxx>
> >>>> Cc: Will Deacon <will@xxxxxxxxxx>
> >>>> Cc: Mark Rutland <mark.rutland@xxxxxxx>
> >>>> Cc: linux-arm-kernel@xxxxxxxxxxxxxxxxxxx
> >>>> Cc: linux-kernel@xxxxxxxxxxxxxxx
> >>>> Signed-off-by: Anshuman Khandual <anshuman.khandual@xxxxxxx>
> >>>
> >>> I have no strong feelings either way, so: 
> >>>
> >>> Acked-by: Mark Rutland <mark.rutland@xxxxxxx>
> >>>
> >>> ... that said, I think that the "bad" naming is unclear and misleading,
> >>> and it'd be really nice if we could clean that up treewide with
> >>> something clearer than "bad".
> >>
> >> Agreed, the name is misleading.
> >>
> >>> It does seem that would roughly fit p??_leaf() if we had
> >>
> >> But what if the platform does not support huge page aka leaf mapping
> >> at the given level ? Also a non table i.e bad entry might not always
> >> mean a leaf/section/huge page mapping, it could simply imply that the
> >> entry is not just pointing to next level and might be just in an bad
> >> intermediate or invalid state.
> > 
> > Ah, so that's also covering swap entries, too? It's not entirely clear
> > to me what "bad intermediate or invalid state" means, because I assume
> > it's not arbitrary junk or this wouldn't be sound genrally.
> 
> Intermediate states like swap, migration or probably even splitting THP.
> Though I am not really sure whether pxx_bad() only gets used for valid
> table or leaf entries i.e things which are mapped. Hence checking just
> for non table entry is better and even safer, than looking out for what
> other states the entry could be in.

I had a quick look through some of the uses and it seems the expectation
is that after a !pmd_bad(), the pmd is a table. The checks for
migration, huge page etc. are prior to the pmd_bad() check.

For this patch:

Acked-by: Catalin Marinas <catalin.marinas@xxxxxxx>