Re: [External] Re: [PATCH v2 1/3] riscv: mm: Use ASID in update_mmu_cache()

[Klara Modin]

On 2026-07-18 00:57:06 +0800, Xu Lu wrote:
> Hi Paul,
> 
> A gentle ping.
> 
> I have no idea what I can do regarding linux-next. Could you please
> help revert the patch from version 1 and cherry-pick all the patches
> from version 2 instead? Or what can I do for this?

Paul has updated the for-next branch at
git.kernel.org/pub/scm/linux/kernel/git/riscv/linux.git so it should
sort itself out to either today's or monday's next version (depending on
if the tree has already been pulled or not). Even if it is not pulled
until monday I think it will be fine, this is what linux-next is for
after all.

Regards,
Klara Modin

> 
> Best regards,
> Xu Lu
> 
> On Fri, Jul 17, 2026 at 10:11 AM Xu Lu <luxu.kernel@xxxxxxxxxxxxx> wrote:
> >
> > Hi Paul,
> >
> > Klara found that linux-next cherry-picks the new version of patch 2
> > while picking the old version of patch 1, which leads to an
> > uninitialized asid. Could you please help correct this?
> >
> > Best regards,
> > Xu Lu
> >
> > On Fri, Jul 17, 2026 at 12:25 AM Klara Modin <klarasmodin@xxxxxxxxx> wrote:
> > >
> > > Hi,
> > >
> > > On 2026-07-15 21:20:07 +0800, Xu Lu wrote:
> > > > Only flush TLB entries for the specified mm in update_mmu_cache_range().
> > > >
> > > > Signed-off-by: Xu Lu <luxu.kernel@xxxxxxxxxxxxx>
> > > > ---
> > > >  arch/riscv/include/asm/pgtable.h  | 4 +++-
> > > >  arch/riscv/include/asm/tlbflush.h | 5 +++++
> > > >  arch/riscv/mm/tlbflush.c          | 5 -----
> > > >  3 files changed, 8 insertions(+), 6 deletions(-)
> > > >
> > > > diff --git a/arch/riscv/include/asm/pgtable.h b/arch/riscv/include/asm/pgtable.h
> > > > index 5d5756bda82e..9926556099ae 100644
> > > > --- a/arch/riscv/include/asm/pgtable.h
> > > > +++ b/arch/riscv/include/asm/pgtable.h
> > > > @@ -568,6 +568,8 @@ static inline void update_mmu_cache_range(struct vm_fault *vmf,
> > > >               struct vm_area_struct *vma, unsigned long address,
> > > >               pte_t *ptep, unsigned int nr)
> > > >  {
> > > > +     unsigned long asid = get_mm_asid(vma->vm_mm);
> > > > +
> > >
> > > It seems the old version of this patch is still used (at least in
> > > next-202650716, 15ce1d7c4ddfe0dff00fcaee4ccaeef3efbc62c6), which means
> > > asid is not initialised here. We instead have:
> > >
> > > diff --git a/arch/riscv/include/asm/pgtable.h b/arch/riscv/include/asm/pgtable.h
> > > index 5d5756bda82e..755495a542cc 100644
> > > --- a/arch/riscv/include/asm/pgtable.h
> > > +++ b/arch/riscv/include/asm/pgtable.h
> > > @@ -568,6 +568,8 @@ static inline void update_mmu_cache_range(struct vm_fault *vmf,
> > >                 struct vm_area_struct *vma, unsigned long address,
> > >                 pte_t *ptep, unsigned int nr)
> > >  {
> > > +       unsigned long asid;
> > > +
> > >         /*
> > >          * Svvptc guarantees that the new valid pte will be visible within
> > >          * a bounded timeframe, so when the uarch does not cache invalid
> > > @@ -583,10 +585,11 @@ static inline void update_mmu_cache_range(struct vm_fault *vmf,
> > >          * Relying on flush_tlb_fix_spurious_fault would suffice, but
> > >          * the extra traps reduce performance.  So, eagerly SFENCE.VMA.
> > >          */
> > > +       asid = get_mm_asid(vma->vm_mm);
> > >         while (nr--)
> > > -               local_flush_tlb_page(address + nr * PAGE_SIZE);
> > > -
> > > +               local_flush_tlb_page_asid(address + nr * PAGE_SIZE, asid);
> > >  }
> > > +
> > >  #define update_mmu_cache(vma, addr, ptep) \
> > >         update_mmu_cache_range(NULL, vma, addr, ptep, 1)
> > >
> > >
> > > This means that patch 2 of this series will use the uninitialised asid
> > > for local_svinval_vma() and I get the same behaviour as before (hang on
> > > initramfs). If I revert the two patches of this series which are in next
> > > and apply this again it's fine.
> > >
> > > Regards,
> > > Klara Modin

...