Re: [PATCH qemu] vfio/spapr: Fix page size calculation

[Philippe Mathieu-Daudé]

On 3/24/20 3:30 PM, Greg Kurz wrote:
> On Tue, 24 Mar 2020 14:27:35 +0100
> Philippe Mathieu-Daudé <address@hidden> wrote:
>
> > On 3/24/20 7:39 AM, Alexey Kardashevskiy wrote:
> > > Coverity detected an issue (CID 1421903) with potential call of clz64(0)
> > > which returns 64 which make it do "<<" with a negative number.
> > >
> > > This checks the mask and avoids undefined behaviour.
> > >
> > > In practice pgsizes and memory_region_iommu_get_min_page_size() always
> > > have some common page sizes and even if they did not, the resulting page
> > > size would be 0x8000.0000.0000.0000 (gcc 9.2) and
> > > ioctl(VFIO_IOMMU_SPAPR_TCE_CREATE) would fail anyway.
> > >
> > > Signed-off-by: Alexey Kardashevskiy <address@hidden>
> > > ---
> > >    hw/vfio/spapr.c | 6 +++---
> > >    1 file changed, 3 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/hw/vfio/spapr.c b/hw/vfio/spapr.c
> > > index 33692fc86fd6..2900bd19417a 100644
> > > --- a/hw/vfio/spapr.c
> > > +++ b/hw/vfio/spapr.c
> > > @@ -147,7 +147,7 @@ int vfio_spapr_create_window(VFIOContainer *container,
> > >    {
> > >        int ret = 0;
> > >        IOMMUMemoryRegion *iommu_mr = IOMMU_MEMORY_REGION(section->mr);
> > > -    uint64_t pagesize = memory_region_iommu_get_min_page_size(iommu_mr);
> > > +    uint64_t pagesize = memory_region_iommu_get_min_page_size(iommu_mr), 
> > > pgmask;
> > >        unsigned entries, bits_total, bits_per_level, max_levels;
> > >        struct vfio_iommu_spapr_tce_create create = { .argsz = sizeof(create) };
> > >        long rampagesize = qemu_minrampagesize();
> > > @@ -159,8 +159,8 @@ int vfio_spapr_create_window(VFIOContainer *container,
> > >        if (pagesize > rampagesize) {
> > >            pagesize = rampagesize;
> > >        }
> > > -    pagesize = 1ULL << (63 - clz64(container->pgsizes &
> > > -                                   (pagesize | (pagesize - 1))));
> > > +    pgmask = container->pgsizes & (pagesize | (pagesize - 1));
> >
> > Is that ROUND_UP(container->pgsizes, pagesize)?
>
> This means we clip all page sizes greater than pagesize from
> container->pgsizes... It doesn't look like ROUND_UP() semantics
> to me.

Ah. Extracting it as a new function with meaningful name would help code 
review :)

> > > +    pagesize = pgmask ? (1ULL << (63 - clz64(pgmask))) : 0;
> > >        if (!pagesize) {
> > >            error_report("Host doesn't support page size 0x%"PRIx64
> > >                         ", the supported mask is 0x%lx",