On Thu, Jun 20, 2019 at 08:59:55PM +0800, Peter Xu wrote:
> On Thu, Jun 20, 2019 at 10:35:29AM +0200, Paolo Bonzini wrote:
> > On 20/06/19 06:02, Peter Xu wrote:
> > > Seems workable, to be explicit - we can even cut it into chunks with
> > > different size to be efficient.
> >
> > Yes, this is not hard (completely untested):
> >
> > diff --git a/hw/i386/intel_iommu.c b/hw/i386/intel_iommu.c
> > index 44b1231157..541538bc6c 100644
> > --- a/hw/i386/intel_iommu.c
> > +++ b/hw/i386/intel_iommu.c
> > @@ -3388,39 +3388,34 @@ static void vtd_address_space_unmap(VTDAddressSpace
> > *as, IOMMUNotifier *n)
> > }
> >
> > assert(start <= end);
> > - size = end - start;
> > + while (end > start) {
> > + size = end - start;
> > + /* Only keep the lowest bit of either size or start. */
> > + size = MIN(size & -size, start & -start);
>
> I feel like this can be problematic. I'm imaging:
>
> start=0x1000_0000, size=0x1000_1000
>
> This will get size=0x1000 but actually we can do size=0x1000_0000 as
> the first.
>
> > + /* Should not happen, but limit to address width too just in case
> > */
> > + size = MIN(size, 1ULL << s->aw_bits);
> >
> > - if (ctpop64(size) != 1) {
> > - /*
> > - * This size cannot format a correct mask. Let's enlarge it to
> > - * suite the minimum available mask.
> > - */
> > - int n = 64 - clz64(size);
> > - if (n > s->aw_bits) {
> > - /* should not happen, but in case it happens, limit it */
> > - n = s->aw_bits;
> > - }
> > - size = 1ULL << n;
> > - }
> > + assert((start & (size - 1)) == 0);
> >
> > - entry.target_as = &address_space_memory;
> > - /* Adjust iova for the size */
> > - entry.iova = n->start & ~(size - 1);
> > - /* This field is meaningless for unmap */
> > - entry.translated_addr = 0;
> > - entry.perm = IOMMU_NONE;
> > - entry.addr_mask = size - 1;
> > + entry.target_as = &address_space_memory;
> > + entry.iova = start;
> > + /* This field is meaningless for unmap */
> > + entry.translated_addr = 0;
> > + entry.perm = IOMMU_NONE;
> > + entry.addr_mask = size - 1;
>
> (some of the fields can be moved out of loop because they are
> constants)
>
> >
> > - trace_vtd_as_unmap_whole(pci_bus_num(as->bus),
> > - VTD_PCI_SLOT(as->devfn),
> > - VTD_PCI_FUNC(as->devfn),
> > - entry.iova, size);
> > + trace_vtd_as_unmap_whole(pci_bus_num(as->bus),
> > + VTD_PCI_SLOT(as->devfn),
> > + VTD_PCI_FUNC(as->devfn),
> > + entry.iova, size);
>
> Can move this out because this is a trace only so we don't have
> restriction on mask?
>
> >
> > - map.iova = entry.iova;
> > - map.size = entry.addr_mask;
> > - iova_tree_remove(as->iova_tree, &map);
> > + map.iova = entry.iova;
> > + map.size = entry.addr_mask;
> > + iova_tree_remove(as->iova_tree, &map);
>
> Same here?
>
> >
> > - memory_region_notify_one(n, &entry);
> > + memory_region_notify_one(n, &entry);
> > + start += size;
> > + }
> > }
> >
> > static void vtd_address_space_unmap_all(IntelIOMMUState *s)
> >
> >
> > Yan,
> >
> > if something like this works for you, let me know and I will submit it
> > as a proper patch.
> >
> > Paolo
>
> Since during review I'm thinking how to generate a correct sequence of
> these masks... here's my try below with above issues fixed... :)
>
> I've tried compile but not tested. Yan can test it, or I can do it
> too tomorrow after I find some machines.
>
> Thanks,
>
hi Peter and Paolo,
I can help test it next week. Didn't manage to find a time for it today.
Thanks
Yan
> ------------------------------------------------------------
> diff --git a/hw/i386/intel_iommu.c b/hw/i386/intel_iommu.c
> index 44b1231157..cfbd225f0a 100644
> --- a/hw/i386/intel_iommu.c
> +++ b/hw/i386/intel_iommu.c
> @@ -3363,11 +3363,32 @@ VTDAddressSpace *vtd_find_add_as(IntelIOMMUState *s,
> PCIBus *bus, int devfn)
> return vtd_dev_as;
> }
>
> +static uint64_t vtd_get_next_mask(uint64_t start, uint64_t size, int gaw)
> +{
> + /* Tries to find smallest mask from start first */
> + uint64_t rmask = start & -start, max_mask = 1ULL << gaw;
> +
> + assert(size && gaw > 0 && gaw < 64);
> +
> + /* Zero start, or too big */
> + if (!rmask || rmask > max_mask) {
> + rmask = max_mask;
> + }
> +
> + /* If the start mask worked, then use it */
> + if (rmask <= size) {
> + return rmask;
> + }
> +
> + /* Find the largest page mask from size */
> + return 1ULL << (63 - clz64(size));
> +}
> +
> /* Unmap the whole range in the notifier's scope. */
> static void vtd_address_space_unmap(VTDAddressSpace *as, IOMMUNotifier *n)
> {
> IOMMUTLBEntry entry;
> - hwaddr size;
> + hwaddr size, remain;
> hwaddr start = n->start;
> hwaddr end = n->end;
> IntelIOMMUState *s = as->iommu_state;
> @@ -3388,39 +3409,28 @@ static void vtd_address_space_unmap(VTDAddressSpace
> *as, IOMMUNotifier *n)
> }
>
> assert(start <= end);
> - size = end - start;
> -
> - if (ctpop64(size) != 1) {
> - /*
> - * This size cannot format a correct mask. Let's enlarge it to
> - * suite the minimum available mask.
> - */
> - int n = 64 - clz64(size);
> - if (n > s->aw_bits) {
> - /* should not happen, but in case it happens, limit it */
> - n = s->aw_bits;
> - }
> - size = 1ULL << n;
> - }
> -
> + size = remain = end - start;
> entry.target_as = &address_space_memory;
> - /* Adjust iova for the size */
> - entry.iova = n->start & ~(size - 1);
> + entry.perm = IOMMU_NONE;
> /* This field is meaningless for unmap */
> entry.translated_addr = 0;
> - entry.perm = IOMMU_NONE;
> - entry.addr_mask = size - 1;
> +
> + while (remain) {
> + uint64_t mask = vtd_get_next_mask(start, remain, s->aw_bits);
> +
> + entry.iova = start;
> + entry.addr_mask = mask - 1;
> + memory_region_notify_one(n, &entry);
> + }
>
> trace_vtd_as_unmap_whole(pci_bus_num(as->bus),
> VTD_PCI_SLOT(as->devfn),
> VTD_PCI_FUNC(as->devfn),
> - entry.iova, size);
> + n->start, size);
>
> - map.iova = entry.iova;
> - map.size = entry.addr_mask;
> + map.iova = n->start;
> + map.size = size;
> iova_tree_remove(as->iova_tree, &map);
> -
> - memory_region_notify_one(n, &entry);
> }
>
> static void vtd_address_space_unmap_all(IntelIOMMUState *s)
> ------------------------------------------------------------
>
> Regards,
>
> --
> Peter Xu
