[PATCH 05/23] m68k: allow pte_offset_map[_lock]() to fail

[Geert Uytterhoeven]

Hi Hugh,

On Thu, May 11, 2023 at 4:58 AM Hugh Dickins <hughd at google.com> wrote:
> On Wed, 10 May 2023, Geert Uytterhoeven wrote:
> > On Wed, May 10, 2023 at 6:48 AM Hugh Dickins <hughd at google.com> wrote:
> > > In rare transient cases, not yet made possible, pte_offset_map() and
> > > pte_offset_map_lock() may not find a page table: handle appropriately.
> > >
> > > Restructure cf_tlb_miss() with a pte_unmap() (previously omitted)
> > > at label out, followed by one local_irq_restore() for all.
> >
> > That's a bug fix, which should be a separate patch?
>
> No, that's not a bug fix for the current tree, since m68k does not
> offer CONFIG_HIGHPTE, so pte_unmap() is never anything but a no-op
> for m68k (see include/linux/pgtable.h).
>
> But I want to change pte_unmap() to do something even without
> CONFIG_HIGHPTE, so have to fix up any such previously harmless
> omissions in this series first.

OK.

> > > --- a/arch/m68k/include/asm/mmu_context.h
> > > +++ b/arch/m68k/include/asm/mmu_context.h
> > > @@ -99,7 +99,7 @@ static inline void load_ksp_mmu(struct task_struct *task)
> > >         p4d_t *p4d;
> > >         pud_t *pud;
> > >         pmd_t *pmd;
> > > -       pte_t *pte;
> > > +       pte_t *pte = NULL;
> > >         unsigned long mmuar;
> > >
> > >         local_irq_save(flags);
> > > @@ -139,7 +139,7 @@ static inline void load_ksp_mmu(struct task_struct *task)
> > >
> > >         pte = (mmuar >= PAGE_OFFSET) ? pte_offset_kernel(pmd, mmuar)
> > >                                      : pte_offset_map(pmd, mmuar);
> > > -       if (pte_none(*pte) || !pte_present(*pte))
> > > +       if (!pte || pte_none(*pte) || !pte_present(*pte))
> > >                 goto bug;
> >
> > If the absence of a pte is to become a non-abnormal case, it should
> > probably jump to "end" instead, to avoid spamming the kernel log.
>
> I don't think so (but of course it's hard for you to tell, without
> seeing all completed series of series).  If pmd_none(*pmd) can safely
> goto bug just above, and pte_none(*pte) goto bug here, well, the !pte
> case is going to be stranger than either of those.
>
> My understanding of this function, load_ksp_mmu(), is that it's dealing
> at context switch with a part of userspace which very much needs to be
> present: whatever keeps that from being swapped out or migrated at
> present, will be sure to keep the !pte case away - we cannot steal its
> page table just at random (and a THP on m68k would be surprising too).
>
> Though there is one case I can think of which will cause !pte here,
> and so goto bug: if the pmd entry has got corrupted, and counts as
> pmd_bad(), which will be tested (and cleared) in pte_offset_map().
> But it is okay to report a bug in that case.
>
> I can certainly change this to goto end instead if you still prefer,
> no problem; but I'd rather keep it as is, if only for me to be proved
> wrong by you actually seeing spam there.

OK, makes sense.

> > > @@ -161,6 +161,8 @@ static inline void load_ksp_mmu(struct task_struct *task)
> > >  bug:
> > >         pr_info("ksp load failed: mm=0x%p ksp=0x08%lx\n", mm, mmuar);
> > >  end:
> > > +       if (pte && mmuar < PAGE_OFFSET)
> > > +               pte_unmap(pte);
> >
> > Is this also a bugfix, not mentioned in the patch description?
>
> I'm not sure whether you're referring to the pte_unmap() which we
> already discussed above, or you're seeing something else in addition;
> but I don't think there's a bugfix here, just a rearrangement because
> we now want lots of cases to do the pte_unmap() and local_irq_restore().

I was referring to the addition of pte_unmap().
As per your explanation above, this is not a bugfix.

Gr{oetje,eeting}s,

                        Geert

-- 
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert at linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds