Re: [PATCH v13 2/2] vhost-vdpa: add support for vIOMMU

[Jason Wang <jasowang@redhat.com>]

在 2023/2/8 10:57, Cindy Lu 写道:
> 1.Add support for vIOMMU.
> Add the new function to deal with iommu MR.
> - during iommu_region_add register a specific IOMMU notifier,
>    and store all notifiers in a list.
> - during iommu_region_del, compare and delete the IOMMU notifier from the list
> - since the SVQ not support iommu yet, add the check for IOMMU
>    in vhost_vdpa_dev_start, if the SVQ and IOMMU enable at the same time
>    function will return fail.
>
> 2.Skip the check in vhost_vdpa_listener_skipped_section() while
> MR is IOMMU, Move this check to  vhost_vdpa_iommu_map_notify()


This need some tweak as well, it's better not repeat what is done in the 
code but why do you need this change. More could be found at:

https://docs.kernel.org/process/submitting-patches.html#describe-your-changes


>
> Verified in vp_vdpa and vdpa_sim_net driver
>
> Signed-off-by: Cindy Lu <lulu@redhat.com>
> ---
>   hw/virtio/vhost-vdpa.c         | 173 ++++++++++++++++++++++++++++++---
>   include/hw/virtio/vhost-vdpa.h |  11 +++
>   2 files changed, 173 insertions(+), 11 deletions(-)
>
> diff --git a/hw/virtio/vhost-vdpa.c b/hw/virtio/vhost-vdpa.c
> index 542e003101..46f676ab71 100644
> --- a/hw/virtio/vhost-vdpa.c
> +++ b/hw/virtio/vhost-vdpa.c
> @@ -26,6 +26,7 @@
>   #include "cpu.h"
>   #include "trace.h"
>   #include "qapi/error.h"
> +#include "hw/virtio/virtio-access.h"
>   
>   /*
>    * Return one past the end of the end of section. Be careful with uint64_t
> @@ -60,15 +61,22 @@ static bool vhost_vdpa_listener_skipped_section(MemoryRegionSection *section,
>                        iova_min, section->offset_within_address_space);
>           return true;
>       }
> +    /*
> +     * While using vIOMMU, Sometimes the section will be larger than iova_max
> +     * but the memory that  actually mapping is smaller, So skip the check
> +     * here. Will add the check in vhost_vdpa_iommu_map_notify,
> +     *There is the real size that maps to the kernel


Please tweak the comment, it has issues of whitespace, capitalization,  
punctuation marks.


> +     */
>   
> -    llend = vhost_vdpa_section_end(section);
> -    if (int128_gt(llend, int128_make64(iova_max))) {
> -        error_report("RAM section out of device range (max=0x%" PRIx64
> -                     ", end addr=0x%" PRIx64 ")",
> -                     iova_max, int128_get64(llend));
> -        return true;
> +    if (!memory_region_is_iommu(section->mr)) {


Note related to this patch but should we exclude non ram region here as 
well?


> +        llend = vhost_vdpa_section_end(section);
> +        if (int128_gt(llend, int128_make64(iova_max))) {
> +            error_report("RAM section out of device range (max=0x%" PRIx64
> +                         ", end addr=0x%" PRIx64 ")",
> +                         iova_max, int128_get64(llend));
> +            return true;
> +        }
>       }
> -
>       return false;
>   }
>   
> @@ -185,6 +193,118 @@ static void vhost_vdpa_listener_commit(MemoryListener *listener)
>       v->iotlb_batch_begin_sent = false;
>   }
>   
> +static void vhost_vdpa_iommu_map_notify(IOMMUNotifier *n, IOMMUTLBEntry *iotlb)
> +{
> +    struct vdpa_iommu *iommu = container_of(n, struct vdpa_iommu, n);
> +
> +    hwaddr iova = iotlb->iova + iommu->iommu_offset;
> +    struct vhost_vdpa *v = iommu->dev;
> +    void *vaddr;
> +    int ret;
> +    Int128 llend;
> +
> +    if (iotlb->target_as != &address_space_memory) {
> +        error_report("Wrong target AS \"%s\", only system memory is allowed",
> +                     iotlb->target_as->name ? iotlb->target_as->name : "none");
> +        return;
> +    }
> +    RCU_READ_LOCK_GUARD();
> +    /* check if RAM section out of device range */
> +    llend = int128_add(int128_makes64(iotlb->addr_mask), int128_makes64(iova));
> +    if (int128_gt(llend, int128_make64(v->iova_range.last))) {
> +        error_report("RAM section out of device range (max=0x%" PRIx64
> +                     ", end addr=0x%" PRIx64 ")",
> +                     v->iova_range.last, int128_get64(llend));
> +        return;


Can you meet this condition? If yes, should we crop instead of fail here?


> +    }
> +
> +    vhost_vdpa_iotlb_batch_begin_once(v);


Where do we send the VHOST_IOTLB_BATCH_END message, or do we even need 
any batching here?


> +
> +    if ((iotlb->perm & IOMMU_RW) != IOMMU_NONE) {
> +        bool read_only;
> +
> +        if (!memory_get_xlat_addr(iotlb, &vaddr, NULL, &read_only, NULL)) {
> +            return;
> +        }
> +
> +        ret = vhost_vdpa_dma_map(v, VHOST_VDPA_GUEST_PA_ASID, iova,
> +                                 iotlb->addr_mask + 1, vaddr, read_only);
> +        if (ret) {
> +            error_report("vhost_vdpa_dma_map(%p, 0x%" HWADDR_PRIx ", "
> +                         "0x%" HWADDR_PRIx ", %p) = %d (%m)",
> +                         v, iova, iotlb->addr_mask + 1, vaddr, ret);
> +        }
> +    } else {
> +        ret = vhost_vdpa_dma_unmap(v, VHOST_VDPA_GUEST_PA_ASID, iova,
> +                                   iotlb->addr_mask + 1);
> +        if (ret) {
> +            error_report("vhost_vdpa_dma_unmap(%p, 0x%" HWADDR_PRIx ", "
> +                         "0x%" HWADDR_PRIx ") = %d (%m)",
> +                         v, iova, iotlb->addr_mask + 1, ret);
> +        }
> +    }
> +}
> +
> +static void vhost_vdpa_iommu_region_add(MemoryListener *listener,
> +                                        MemoryRegionSection *section)
> +{
> +    struct vhost_vdpa *v = container_of(listener, struct vhost_vdpa, listener);
> +
> +    struct vdpa_iommu *iommu;
> +    Int128 end;
> +    int iommu_idx;
> +    IOMMUMemoryRegion *iommu_mr;
> +    int ret;
> +
> +    iommu_mr = IOMMU_MEMORY_REGION(section->mr);
> +
> +    iommu = g_malloc0(sizeof(*iommu));
> +    end = int128_add(int128_make64(section->offset_within_region),
> +                     section->size);
> +    end = int128_sub(end, int128_one());
> +    iommu_idx = memory_region_iommu_attrs_to_index(iommu_mr,
> +                                                   MEMTXATTRS_UNSPECIFIED);
> +    iommu->iommu_mr = iommu_mr;
> +    iommu_notifier_init(&iommu->n, vhost_vdpa_iommu_map_notify,
> +                        IOMMU_NOTIFIER_IOTLB_EVENTS,
> +                        section->offset_within_region,
> +                        int128_get64(end),
> +                        iommu_idx);
> +    iommu->iommu_offset = section->offset_within_address_space -
> +                          section->offset_within_region;
> +    iommu->dev = v;
> +
> +    ret = memory_region_register_iommu_notifier(section->mr, &iommu->n, NULL);
> +    if (ret) {
> +        g_free(iommu);
> +        return;
> +    }
> +
> +    QLIST_INSERT_HEAD(&v->iommu_list, iommu, iommu_next);
> +    memory_region_iommu_replay(iommu->iommu_mr, &iommu->n);
> +
> +    return;
> +}
> +
> +static void vhost_vdpa_iommu_region_del(MemoryListener *listener,
> +                                        MemoryRegionSection *section)
> +{
> +    struct vhost_vdpa *v = container_of(listener, struct vhost_vdpa, listener);
> +
> +    struct vdpa_iommu *iommu;
> +
> +    QLIST_FOREACH(iommu, &v->iommu_list, iommu_next)
> +    {
> +        if (MEMORY_REGION(iommu->iommu_mr) == section->mr &&
> +            iommu->n.start == section->offset_within_region) {
> +            memory_region_unregister_iommu_notifier(section->mr, &iommu->n);
> +            QLIST_REMOVE(iommu, iommu_next);
> +            g_free(iommu);
> +            break;
> +        }
> +    }
> +}
> +
>   static void vhost_vdpa_listener_region_add(MemoryListener *listener,
>                                              MemoryRegionSection *section)
>   {
> @@ -199,6 +319,10 @@ static void vhost_vdpa_listener_region_add(MemoryListener *listener,
>                                               v->iova_range.last)) {
>           return;
>       }
> +    if (memory_region_is_iommu(section->mr)) {
> +        vhost_vdpa_iommu_region_add(listener, section);
> +        return;
> +    }
>   
>       if (unlikely((section->offset_within_address_space & ~TARGET_PAGE_MASK) !=
>                    (section->offset_within_region & ~TARGET_PAGE_MASK))) {
> @@ -278,6 +402,9 @@ static void vhost_vdpa_listener_region_del(MemoryListener *listener,
>                                               v->iova_range.last)) {
>           return;
>       }
> +    if (memory_region_is_iommu(section->mr)) {
> +        vhost_vdpa_iommu_region_del(listener, section);
> +    }
>   
>       if (unlikely((section->offset_within_address_space & ~TARGET_PAGE_MASK) !=
>                    (section->offset_within_region & ~TARGET_PAGE_MASK))) {
> @@ -288,7 +415,8 @@ static void vhost_vdpa_listener_region_del(MemoryListener *listener,
>       iova = TARGET_PAGE_ALIGN(section->offset_within_address_space);
>       llend = vhost_vdpa_section_end(section);
>   
> -    trace_vhost_vdpa_listener_region_del(v, iova, int128_get64(llend));
> +    trace_vhost_vdpa_listener_region_del(v, iova,
> +        int128_get64(int128_sub(llend, int128_one())));


Seems like an independent fix?


>   
>       if (int128_ge(int128_make64(iova), llend)) {
>           return;
> @@ -315,10 +443,28 @@ static void vhost_vdpa_listener_region_del(MemoryListener *listener,
>           vhost_iova_tree_remove(v->iova_tree, *result);
>       }
>       vhost_vdpa_iotlb_batch_begin_once(v);
> +    /*
> +     * The unmap ioctl doesn't accept a full 64-bit. need to check it
> +     */
> +    if (int128_eq(llsize, int128_2_64())) {
> +        llsize = int128_rshift(llsize, 1);
> +        ret = vhost_vdpa_dma_unmap(v, VHOST_VDPA_GUEST_PA_ASID, iova,
> +                                   int128_get64(llsize));
> +
> +        if (ret) {
> +            error_report("vhost_vdpa_dma_unmap(%p, 0x%" HWADDR_PRIx ", "
> +                         "0x%" HWADDR_PRIx ") = %d (%m)",
> +                         v, iova, int128_get64(llsize), ret);
> +        }
> +        iova += int128_get64(llsize);
> +    }
>       ret = vhost_vdpa_dma_unmap(v, VHOST_VDPA_GUEST_PA_ASID, iova,
>                                  int128_get64(llsize));
> +
>       if (ret) {
> -        error_report("vhost_vdpa dma unmap error!");
> +        error_report("vhost_vdpa_dma_unmap(%p, 0x%" HWADDR_PRIx ", "
> +                     "0x%" HWADDR_PRIx ") = %d (%m)",
> +                     v, iova, int128_get64(llsize), ret);
>       }


Let's use a separate patch to fix this full 64-bit issue.


>   
>       memory_region_unref(section->mr);
> @@ -578,7 +724,6 @@ static int vhost_vdpa_cleanup(struct vhost_dev *dev)
>       v = dev->opaque;
>       trace_vhost_vdpa_cleanup(dev, v);
>       vhost_vdpa_host_notifiers_uninit(dev, dev->nvqs);
> -    memory_listener_unregister(&v->listener);


Any reason for not unregistering listener here?

Thanks


>       vhost_vdpa_svq_cleanup(dev);
>   
>       dev->opaque = NULL;
> @@ -1130,7 +1275,13 @@ static int vhost_vdpa_dev_start(struct vhost_dev *dev, bool started)
>       }
>   
>       if (started) {
> -        memory_listener_register(&v->listener, &address_space_memory);
> +        if (vhost_dev_has_iommu(dev) && (v->shadow_vqs_enabled)) {
> +            error_report("the SVQ can not wortk in with IOMMU enable, please "
> +                         "disable IOMMU and try again");
> +            return -1;
> +        }
> +        memory_listener_register(&v->listener, dev->vdev->dma_as);
> +
>           return vhost_vdpa_add_status(dev, VIRTIO_CONFIG_S_DRIVER_OK);
>       } else {
>           vhost_vdpa_reset_device(dev);
> diff --git a/include/hw/virtio/vhost-vdpa.h b/include/hw/virtio/vhost-vdpa.h
> index 7997f09a8d..54cede86dc 100644
> --- a/include/hw/virtio/vhost-vdpa.h
> +++ b/include/hw/virtio/vhost-vdpa.h
> @@ -49,6 +49,8 @@ typedef struct vhost_vdpa {
>       void *shadow_vq_ops_opaque;
>       struct vhost_dev *dev;
>       VhostVDPAHostNotifier notifier[VIRTIO_QUEUE_MAX];
> +    QLIST_HEAD(, vdpa_iommu) iommu_list;
> +    IOMMUNotifier n;
>   } VhostVDPA;
>   
>   int vhost_vdpa_get_iova_range(int fd, struct vhost_vdpa_iova_range *iova_range);
> @@ -58,4 +60,13 @@ int vhost_vdpa_dma_map(struct vhost_vdpa *v, uint32_t asid, hwaddr iova,
>   int vhost_vdpa_dma_unmap(struct vhost_vdpa *v, uint32_t asid, hwaddr iova,
>                            hwaddr size);
>   
> +typedef struct vdpa_iommu {
> +    struct vhost_vdpa *dev;
> +    IOMMUMemoryRegion *iommu_mr;
> +    hwaddr iommu_offset;
> +    IOMMUNotifier n;
> +    QLIST_ENTRY(vdpa_iommu) iommu_next;
> +} VDPAIOMMUState;
> +
> +
>   #endif