Re: [PATCH v2] softmmu/physmem: Fix address of FlatView access in address_space_(read|write)_cached_slow()

[Stefan Hajnoczi]

On Fri, 2 Dec 2022 at 07:31, Alberto Faria <afaria@redhat.com> wrote:
>
> On Mon, Sep 5, 2022 at 8:45 AM Paolo Bonzini <pbonzini@redhat.com> wrote:
> > Thanks, I will queue it.
> >
> > Paolo
> >
> > Il lun 5 set 2022, 02:00 Alberto Faria <afaria@redhat.com> ha scritto:
> >>
> >> flatview_(read|write)_continue() must be called with an address in the
> >> FlatView's address space, but `addr` is relative to the
> >> MemoryRegionCache.
> >>
> >> Convert `addr` from the MemoryRegionCache's address space to the
> >> FlatView's before passing it to flatview_(read|write)_continue().
> >>
> >> This problem was uncovered while attempting to perform unaligned writes,
> >> which caused flatview_write_continue() to reach the call to
> >> flatview_translate(), which then translated the erroneous address and
> >> caused the subsequent write to fail.
> >>
> >> Fixes: 48564041a7 ("exec: reintroduce MemoryRegion caching")
> >> Co-Developed-by: Stefan Hajnoczi <stefanha@redhat.com>
> >> Signed-off-by: Alberto Faria <afaria@redhat.com>
> >> ---
> >>  softmmu/physmem.c | 18 ++++++++++++++----
> >>  1 file changed, 14 insertions(+), 4 deletions(-)
> >>
> >> diff --git a/softmmu/physmem.c b/softmmu/physmem.c
> >> index 50231bab30..bdde4eb927 100644
> >> --- a/softmmu/physmem.c
> >> +++ b/softmmu/physmem.c
> >> @@ -3420,6 +3420,14 @@ static inline MemoryRegion 
> >> *address_space_translate_cached(
> >>      return section.mr;
> >>  }
> >>
> >> +/* Converts `addr` from the address space of `cache` to that of 
> >> `cache->fv`. */
> >> +static inline hwaddr addr_in_cache_to_in_flat_view(MemoryRegionCache 
> >> *cache,
> >> +                                                   hwaddr addr)
> >> +{
> >> +    hwaddr addr_in_mrs = addr + cache->xlat - 
> >> cache->mrs.offset_within_region;
> >> +    return addr_in_mrs + cache->mrs.offset_within_address_space;
> >> +}
> >> +
> >>  /* Called from RCU critical section. address_space_read_cached uses this
> >>   * out of line function when the target is an MMIO or IOMMU region.
> >>   */
> >> @@ -3434,8 +3442,9 @@ address_space_read_cached_slow(MemoryRegionCache 
> >> *cache, hwaddr addr,
> >>      mr = address_space_translate_cached(cache, addr, &addr1, &l, false,
> >>                                          MEMTXATTRS_UNSPECIFIED);
> >>      return flatview_read_continue(cache->fv,
> >> -                                  addr, MEMTXATTRS_UNSPECIFIED, buf, len,
> >> -                                  addr1, l, mr);
> >> +                                  addr_in_cache_to_in_flat_view(cache, 
> >> addr),
> >> +                                  MEMTXATTRS_UNSPECIFIED, buf, len, 
> >> addr1, l,
> >> +                                  mr);
> >>  }
> >>
> >>  /* Called from RCU critical section. address_space_write_cached uses this
> >> @@ -3452,8 +3461,9 @@ address_space_write_cached_slow(MemoryRegionCache 
> >> *cache, hwaddr addr,
> >>      mr = address_space_translate_cached(cache, addr, &addr1, &l, true,
> >>                                          MEMTXATTRS_UNSPECIFIED);
> >>      return flatview_write_continue(cache->fv,
> >> -                                   addr, MEMTXATTRS_UNSPECIFIED, buf, len,
> >> -                                   addr1, l, mr);
> >> +                                   addr_in_cache_to_in_flat_view(cache, 
> >> addr),
> >> +                                   MEMTXATTRS_UNSPECIFIED, buf, len, 
> >> addr1, l,
> >> +                                   mr);
> >>  }
> >>
> >>  #define ARG1_DECL                MemoryRegionCache *cache
> >> --
> >> 2.37.2
>
> This doesn't seem to be in master yet. Maybe it fell through the cracks?

Ping.

Stefan