Re: [Qemu-devel] [PATCH qemu v15 16/17] vfio/spapr: Use VFIO_SPAPR_TCE_v2_IOMMU

[Alexey Kardashevskiy]

On 04/07/2016 11:10 AM, David Gibson wrote:
> Subject doesn't seem quite right, since you added at least minimal
> support for the SPAPRv2 IOMMU in the prereg patch.
>
> On Mon, Apr 04, 2016 at 07:33:45PM +1000, Alexey Kardashevskiy wrote:
> > New VFIO_SPAPR_TCE_v2_IOMMU type supports dynamic DMA window management.
> > This adds ability to VFIO common code to dynamically allocate/remove
> > DMA windows in the host kernel when new VFIO container is added/removed.
> >
> > This adds VFIO_IOMMU_SPAPR_TCE_CREATE ioctl to vfio_listener_region_add
> > and adds just created IOMMU into the host IOMMU list; the opposite
> > action is taken in vfio_listener_region_del.
> >
> > When creating a new window, this uses euristic to decide on the TCE table
> > levels number.
>
> "heuristic" has an 'h' (yes, English spelling is stupid[0]).
>
> [0] The historical reasons for that are kind of fascinating, though.

Tried googling, could not spot quickly the reasoning, any hints what to 
google for? Or just a link with an explanation? :)


> > This should cause no guest visible change in behavior.
> >
> > Signed-off-by: Alexey Kardashevskiy <address@hidden>
> > ---
> > Changes:
> > v14:
> > * new to the series
> >
> > ---
> > TODO:
> > * export levels to PHB
> > ---
> >   hw/vfio/common.c | 115 ++++++++++++++++++++++++++++++++++++++++++++++++++-----
> >   trace-events     |   2 +
> >   2 files changed, 107 insertions(+), 10 deletions(-)
> >
> > diff --git a/hw/vfio/common.c b/hw/vfio/common.c
> > index 5e5b77c..57a51df 100644
> > --- a/hw/vfio/common.c
> > +++ b/hw/vfio/common.c
> > @@ -279,6 +279,14 @@ static int vfio_host_win_add(VFIOContainer *container,
> >       return 0;
> >   }
> >
> > +static void vfio_host_win_del(VFIOContainer *container, hwaddr min_iova)
> > +{
> > +    VFIOHostDMAWindow *hostwin = vfio_host_win_lookup(container, min_iova, 1);
> > +
> > +    g_assert(hostwin);
> > +    QLIST_REMOVE(hostwin, hiommu_next);
> > +}
> > +
> >   static bool vfio_listener_skipped_section(MemoryRegionSection *section)
> >   {
> >       return (!memory_region_is_ram(section->mr) &&
> > @@ -392,6 +400,63 @@ static void vfio_listener_region_add(MemoryListener 
> > *listener,
> >       }
> >       end = int128_get64(llend);
> >
> > +    if (container->iommu_type == VFIO_SPAPR_TCE_v2_IOMMU) {
>
> I think the "add region" path could do with being split out into a
> different function - vfio_listener_region_add() is getting pretty
> huge.

It is big but not huge and I am trying to avoid having functions with 
"spapr" in their names in common.c as once they appear, we will start 
having a discussion if they should move to a separate file and if they do, 
then may be some other code should too, etc...


> > +        unsigned entries, pages, pagesize = qemu_real_host_page_size;
> > +        struct vfio_iommu_spapr_tce_create create = { .argsz = sizeof(create) 
> > };
> > +
> > +        trace_vfio_listener_region_add_iommu(iova, end - 1);
> > +        if (section->mr->iommu_ops) {
> > +            pagesize = section->mr->iommu_ops->get_page_sizes(section->mr);
>
> Since you're querying the guest IOMMU here, I assume pagesize is
> supposed to represent *guest* IOMMU pagesizes, in which case it should
> default to TARGET_PAGE_SIZE, instead of qemu_real_host_page_size.
> (didn't you already have a function which implemented that fallback?)

Yes, will use the helper.


> > +        }
> > +        /*
> > +         * FIXME: For VFIO iommu types which have KVM acceleration to
> > +         * avoid bouncing all map/unmaps through qemu this way, this
> > +         * would be the right place to wire that up (tell the KVM
> > +         * device emulation the VFIO iommu handles to use).
> > +         */
> > +        create.window_size = int128_get64(section->size);
> > +        create.page_shift = ctz64(pagesize);
> > +        /*
> > +         * SPAPR host supports multilevel TCE tables, there is some
> > +         * euristic to decide how many levels we want for our table:
>
> s/some euristic/a heuristic/
>
> > +         * 0..64 = 1; 65..4096 = 2; 4097..262144 = 3; 262145.. = 4
> > +         */
> > +        entries = create.window_size >> create.page_shift;
> > +        pages = (entries * sizeof(uint64_t)) / getpagesize();
> > +        create.levels = ctz64(pow2ceil(pages) - 1) / 6 + 1;
> > +
> > +        if (vfio_host_win_lookup(container, create.start_addr,
> > +                                 create.start_addr + create.window_size - 1)) {
> > +            goto fail;
>
> Hmm.. if you successfully look up a host window, it seems to me you
> shouldn't fail, but in fact don't even need to create a new window
> (the removal path gets harder though, because you need to check if any
> guest window requires that host window).


At the moment if the window is there, it is failure in the environment I am 
testing it in. And, having a host kernel which cannot allocate and map 
windows randomly, it is unlikely that I'll have a setup where this spot 
won't mean that something went wrong. x86 case needs lot more than this anyway.


> Requiring that the host windows exactly match the guest windows is
> probably ok for a first version - except that in that case any overlap
> should cause a failure, not just a complete inclusion.

Yes, for now this should fail too, I'll fix it.


> > +        }
> > +
> > +        ret = ioctl(container->fd, VFIO_IOMMU_SPAPR_TCE_CREATE, &create);
> > +        if (ret) {
> > +            error_report("Failed to create a window, ret = %d (%m)", ret);
> > +            goto fail;
> > +        }
> > +
> > +        if (create.start_addr != section->offset_within_address_space) {
> > +            struct vfio_iommu_spapr_tce_remove remove = {
> > +                .argsz = sizeof(remove),
> > +                .start_addr = create.start_addr
> > +            };
> > +            error_report("Host doesn't support DMA window at %"HWADDR_PRIx", must 
> > be %"PRIx64,
> > +                         section->offset_within_address_space,
> > +                         create.start_addr);
> > +            ioctl(container->fd, VFIO_IOMMU_SPAPR_TCE_REMOVE, &remove);
> > +            ret = -EINVAL;
> > +            goto fail;
> > +        }
> > +        trace_vfio_spapr_create_window(create.page_shift,
> > +                                       create.window_size,
> > +                                       create.start_addr);
> > +
> > +        vfio_host_win_add(container, create.start_addr,
> > +                          create.start_addr + create.window_size - 1,
> > +                          1ULL << create.page_shift);
> > +    }
> > +
> >       if (!vfio_host_win_lookup(container, iova, end - 1)) {
> >           error_report("vfio: IOMMU container %p can't map guest IOVA region"
> >                        " 0x%"HWADDR_PRIx"..0x%"HWADDR_PRIx,
> > @@ -525,6 +590,22 @@ static void vfio_listener_region_del(MemoryListener 
> > *listener,
> >                        container, iova, end - iova, ret);
> >       }
> >
> > +    if (container->iommu_type == VFIO_SPAPR_TCE_v2_IOMMU) {
> > +        struct vfio_iommu_spapr_tce_remove remove = {
> > +            .argsz = sizeof(remove),
> > +            .start_addr = section->offset_within_address_space,
> > +        };
> > +        ret = ioctl(container->fd, VFIO_IOMMU_SPAPR_TCE_REMOVE, &remove);
> > +        if (ret) {
> > +            error_report("Failed to remove window at %"PRIx64,
> > +                         remove.start_addr);
> > +        }
> > +
> > +        vfio_host_win_del(container, section->offset_within_address_space);
> > +
> > +        trace_vfio_spapr_remove_window(remove.start_addr);
> > +    }
> > +
> >       if (iommu && iommu->iommu_ops && iommu->iommu_ops->vfio_stop) {
> >           iommu->iommu_ops->vfio_stop(section->mr);
> >       }
> > @@ -915,11 +996,6 @@ static int vfio_connect_container(VFIOGroup *group, 
> > AddressSpace *as)
> >               }
> >           }
> >
> > -        /*
> > -         * This only considers the host IOMMU's 32-bit window.  At
> > -         * some point we need to add support for the optional 64-bit
> > -         * window and dynamic windows
> > -         */
> >           info.argsz = sizeof(info);
> >           ret = ioctl(fd, VFIO_IOMMU_SPAPR_TCE_GET_INFO, &info);
> >           if (ret) {
> > @@ -928,11 +1004,30 @@ static int vfio_connect_container(VFIOGroup *group, 
> > AddressSpace *as)
> >               goto listener_release_exit;
> >           }
> >
> > -        /* The default table uses 4K pages */
> > -        vfio_host_win_add(container, info.dma32_window_start,
> > -                          info.dma32_window_start +
> > -                          info.dma32_window_size - 1,
> > -                          0x1000);
> > +        if (v2) {
> > +            /*
> > +             * There is a default window in just created container.
> > +             * To make region_add/del simpler, we better remove this
> > +             * window now and let those iommu_listener callbacks
> > +             * create/remove them when needed.
> > +             */
> > +            struct vfio_iommu_spapr_tce_remove remove = {
> > +                .argsz = sizeof(remove),
> > +                .start_addr = info.dma32_window_start,
> > +            };
> > +            ret = ioctl(fd, VFIO_IOMMU_SPAPR_TCE_REMOVE, &remove);
> > +            if (ret) {
> > +                error_report("vfio: VFIO_IOMMU_SPAPR_TCE_REMOVE failed: %m");
> > +                ret = -errno;
> > +                goto free_container_exit;
> > +            }
> > +        } else {
> > +            /* The default table uses 4K pages */
> > +            vfio_host_win_add(container, info.dma32_window_start,
> > +                              info.dma32_window_start +
> > +                              info.dma32_window_size - 1,
> > +                              0x1000);
> > +        }
> >       } else {
> >           error_report("vfio: No available IOMMU models");
> >           ret = -EINVAL;
> > diff --git a/trace-events b/trace-events
> > index 23ca0b9..5c651fa 100644
> > --- a/trace-events
> > +++ b/trace-events
> > @@ -1738,6 +1738,8 @@ vfio_region_finalize(const char *name, int index) "Device %s, 
> > region %d"
> >   vfio_region_mmaps_set_enabled(const char *name, bool enabled) "Region %s mmaps 
> > enabled: %d"
> >   vfio_ram_register(uint64_t va, uint64_t size, int ret) "va=%"PRIx64" 
> > size=%"PRIx64" ret=%d"
> >   vfio_ram_unregister(uint64_t va, uint64_t size, int ret) "va=%"PRIx64" 
> > size=%"PRIx64" ret=%d"
> > +vfio_spapr_create_window(int ps, uint64_t ws, uint64_t off) "pageshift=0x%x 
> > winsize=0x%"PRIx64" offset=0x%"PRIx64
> > +vfio_spapr_remove_window(uint64_t off) "offset=%"PRIx64
> >
> >   # hw/vfio/platform.c
> >   vfio_platform_base_device_init(char *name, int groupid) "%s belongs to group 
> > #%d"


--
Alexey