Re: [PATCH 2/3] iommu/vt-d: Use try_cmpxchg64() in intel_pasid_get_entry()

[Uros Bizjak]

On Thu, May 23, 2024 at 3:24 PM Baolu Lu <baolu.lu@xxxxxxxxxxxxxxx> wrote:
>
> On 2024/5/22 16:26, Uros Bizjak wrote:
> > Use try_cmpxchg64() instead of cmpxchg64 (*ptr, old, new) != old in
> > intel_pasid_get_entry().  cmpxchg returns success in ZF flag, so
> > this change saves a compare after cmpxchg (and related move
> > instruction in front of cmpxchg).
> >
> > Signed-off-by: Uros Bizjak <ubizjak@xxxxxxxxx>
> > Cc: David Woodhouse <dwmw2@xxxxxxxxxxxxx>
> > Cc: Lu Baolu <baolu.lu@xxxxxxxxxxxxxxx>
> > Cc: Joerg Roedel <joro@xxxxxxxxxx>
> > Cc: Will Deacon <will@xxxxxxxxxx>
> > Cc: Robin Murphy <robin.murphy@xxxxxxx>
> > ---
> >   drivers/iommu/intel/pasid.c | 7 +++++--
> >   1 file changed, 5 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/iommu/intel/pasid.c b/drivers/iommu/intel/pasid.c
> > index abce19e2ad6f..9bf45bc4b967 100644
> > --- a/drivers/iommu/intel/pasid.c
> > +++ b/drivers/iommu/intel/pasid.c
> > @@ -146,6 +146,8 @@ static struct pasid_entry *intel_pasid_get_entry(struct device *dev, u32 pasid)
> >   retry:
> >       entries = get_pasid_table_from_pde(&dir[dir_index]);
> >       if (!entries) {
> > +             u64 tmp;
> > +
> >               entries = iommu_alloc_page_node(info->iommu->node, GFP_ATOMIC);
> >               if (!entries)
> >                       return NULL;
> > @@ -156,8 +158,9 @@ static struct pasid_entry *intel_pasid_get_entry(struct device *dev, u32 pasid)
> >                * clear. However, this entry might be populated by others
> >                * while we are preparing it. Use theirs with a retry.
> >                */
> > -             if (cmpxchg64(&dir[dir_index].val, 0ULL,
> > -                           (u64)virt_to_phys(entries) | PASID_PTE_PRESENT)) {
> > +             tmp = 0ULL;
>
> nit: can this simply be
>                 tmp = 0;
> ?

I just took the same suffix as it was in the original code, but for
zero literal, it does not make any change. I can change it, it
preferred.

> > +             if (!try_cmpxchg64(&dir[dir_index].val, &tmp,
> > +                                (u64)virt_to_phys(entries) | PASID_PTE_PRESENT)) {
>
> Above change will cause a dead loop during boot. It should be

No, it is correct as written:

if (cmpxchg64(*ptr, 0, new))

can be written as:

if (cmpxchg64(*ptr, 0, new) != 0)

this is equivalent to:

tmp = 0ULL;
if (!try_cmpxchg64(*ptr, &tmp, new))

Thanks,
Uros.