Re: [RFC] mm: Drop pxd_ERROR()

On Thu, Aug 06, 2026 at 01:47:23PM +0200, David Hildenbrand (Arm) wrote:
> On 8/6/26 10:06, Anshuman Khandual wrote:
> > pxd_ERROR() has been used in generic mm just to print the page table entry
> > in pxd_clear_bad() before clearing those out with pxd_clear() later. These
> > pxd_ERROR() macros have been provided by all platforms which basically did
> > the same thing.
> > 
> > Make pxd_clear_bad() use recently added ptval_to_str() instead for printing
> > page table entries thus completely dropping dependency on platform provided
> > pxd_ERROR() macros which can then be dropped off. First move all required
> > helpers in core MM into a header file which could then be used else where.
> > 
> > Although some platforms still use those macros internally as well. In which
> > case just move these macros inside the platform for now.
> 
> No, just get rid of them first.
> 
> The ones in arch/sh/mm/init.c are just supid:
> 
> if (pgd_none(*pgd)) {
> 	pgd_ERROR(*pgd);
> 
> printing a none PGD? Stupid. :)

Will drop them from arch/sh in a pre-requsite patch.

> 
> Then there is arch/parisc/kernel/pci-dma.c, which just open-codes
> pmd_clear_bad() ?
> 
> 	pmd_ERROR(*pmd);
> 	pmd_clear(pmd);

Will replace these instances in arch/parisc/ with corresponding generic
[pmd|pgd]_clear_bad() helpers as necessary in a pre-requsite patch.

> 
> 
> Anything else?

Just one more instance inside arch/loongarch/mm/init.c which could be
dropped as well.

        if (!pte_none(ptep_get(ptep))) {
                pte_ERROR(*ptep);
                return;
        }
> 
> > 
> > Cc: David Hildenbrand (Arm) <[email protected]>
> > Cc: Andrew Morton <[email protected]> 
> > Cc: [email protected]
> > Cc: [email protected]
> > Cc: [email protected]
> > Signed-off-by: Anshuman Khandual <[email protected]>
> > ---
> > This applies on v7.2-rc6 but after the following patch
> > 
> > https://lore.kernel.org/all/[email protected]/
> > 
> > Would it also make sense to just drop __FILE__ and __LINE__ from the output
> > in pxd_clear_bad() helpers as they always print the same details regardless
> > the error path and does not add much value.
> 
> Yes, drop that, it's nonsensical given that all relevant calls are from
> mm/pgtable-generic.c

Will drop them.

> 
> Thanks for looking into this!
> 
> -- 
> Cheers,
> 
> David