hi Eric, Thanks for your reply. On Wed, Jun 19, 2019 at 09:17:41PM +0800, Auger Eric wrote: > Hi Yan, > > [+ Peter] > > On 6/19/19 10:49 AM, Yan Zhao wrote: > > even if an entry overlaps with notifier's range, should not map/unmap > > out of bound part in the entry. > > I don't think the patch was based on the master as the trace at the very > end if not part of the upstream code. > > It's indeed based on the latest master branch. but I added a debug log and forgot to remove that before sending out the patch. sorry for that :)
> > This would cause problem in below case: > > 1. initially there are two notifiers with ranges > > 0-0xfedfffff, 0xfef00000-0xffffffffffffffff, > > IOVAs from 0x3c000000 - 0x3c1fffff is in shadow page table. > > > > 2. in vfio, memory_region_register_iommu_notifier() is followed by > > memory_region_iommu_replay(), which will first call address space unmap, > > and walk and add back all entries in vtd shadow page table. e.g. > > (1) for notifier 0-0xfedfffff, > > IOVAs from 0 - 0xffffffff get unmapped, > > and IOVAs from 0x3c000000 - 0x3c1fffff get mapped > > While the patch looks sensible, the issue is the notifier scope used in > vtd_address_space_unmap is not a valid mask (ctpop64(size) != 1). Then > the size is recomputed (either using n = 64 - clz64(size) for the 1st > notifier or n = s->aw_bits for the 2d) and also the entry (especially > for the 2d notifier where it becomes 0) to get a proper alignment. > maybe the size is calculated right, but 0 for the 2d notifier is because this line below ? entry.iova = n->start & ~(size - 1); > vtd_page_walk sends notifications per block or page (with valid > addr_mask) so stays within the notifier. > > Modifying the entry->iova/addr_mask again in memory_region_notify_one > leads to unaligned start address / addr_mask. I don't think we want that. > if the notifier's start and end is aligned, and entry->iova/addr_mask is aligned before modification, then after modification, the start addr /addr_mask are still aligned ? > Can't we modity the vtd_address_space_unmap() implementation to split > the invalidation in smaller chunks instead? > as this is an API, it cannot reply on the caller to ensure the entry is within its address range. Do you think it's reasonable? Thanks Yan > Thanks > > Eric > > > > (2) for notifier 0xfef00000-0xffffffffffffffff > > IOVAs from 0 - 0x7fffffffff get unmapped,> but IOVAs from > > 0x3c000000 - 0x3c1fffff cannot get mapped back. > > > > Signed-off-by: Yan Zhao <yan.y.z...@intel.com> > > --- > > memory.c | 8 ++++++++ > > 1 file changed, 8 insertions(+) > > > > diff --git a/memory.c b/memory.c > > index 07c8315..a6b9da6 100644 > > --- a/memory.c > > +++ b/memory.c > > @@ -1948,6 +1948,14 @@ void memory_region_notify_one(IOMMUNotifier > > *notifier, > > return; > > } > > > > + if (entry->iova < notifier->start) { > > + entry->iova = notifier->start; > > + } > > + > > + if (entry->iova + entry->addr_mask > notifier->end) { > > + entry->addr_mask = notifier->end - entry->iova;> + } > > + > > if (entry->perm & IOMMU_RW) { > > printf("map %lx %lx\n", entry->iova, entry->iova + > > entry->addr_mask); > > request_flags = IOMMU_NOTIFIER_MAP; > > >