Re: [PATCH v2 4/8] vringh: support VA with iotlb

[Eugenio Perez Martin]

On Thu, Mar 16, 2023 at 5:07 PM Stefano Garzarella <sgarzare@xxxxxxxxxx> wrote:
>
> On Fri, Mar 3, 2023 at 3:39 PM Eugenio Perez Martin <eperezma@xxxxxxxxxx> wrote:
> >
> > On Thu, Mar 2, 2023 at 12:35 PM Stefano Garzarella <sgarzare@xxxxxxxxxx> wrote:
> > >
> > > vDPA supports the possibility to use user VA in the iotlb messages.
> > > So, let's add support for user VA in vringh to use it in the vDPA
> > > simulators.
> > >
> > > Signed-off-by: Stefano Garzarella <sgarzare@xxxxxxxxxx>
> > > ---
> > >
> > > Notes:
> > >     v2:
> > >     - replace kmap_atomic() with kmap_local_page() [see previous patch]
> > >     - fix cast warnings when build with W=1 C=1
> > >
> > >  include/linux/vringh.h            |   5 +-
> > >  drivers/vdpa/mlx5/net/mlx5_vnet.c |   2 +-
> > >  drivers/vdpa/vdpa_sim/vdpa_sim.c  |   4 +-
> > >  drivers/vhost/vringh.c            | 247 ++++++++++++++++++++++++------
> > >  4 files changed, 205 insertions(+), 53 deletions(-)
> > >
>
> [...]
>
> >
> > It seems to me iotlb_translate_va and iotlb_translate_pa are very
> > similar, their only difference is that the argument is that iov is
> > iovec instead of bio_vec. And how to fill it, obviously.
> >
> > It would be great to merge both functions, only differing with a
> > conditional on vrh->use_va, or generics, or similar. Or, if following
> > the style of the rest of vringh code, to provide a callback to fill
> > iovec (although I like conditional more).
> >
> > However I cannot think of an easy way to perform that without long
> > macros or type erasure.
>
> Thank you for pushing me :-)
> I finally managed to avoid code duplication (partial patch attached,
> but not yet fully tested).
>
> @Jason: with this refactoring I removed copy_to_va/copy_to_pa, so I
> also avoided getu16_iotlb_va/pa.
>
> I will send the full patch in v3, but I would like to get your opinion
> first ;-)
>
>
>
> diff --git a/drivers/vhost/vringh.c b/drivers/vhost/vringh.c
> index 0ba3ef809e48..71dd67700e36 100644
> --- a/drivers/vhost/vringh.c
> +++ b/drivers/vhost/vringh.c
> @@ -1096,8 +1096,7 @@ EXPORT_SYMBOL(vringh_need_notify_kern);
>
>  static int iotlb_translate(const struct vringh *vrh,
>                            u64 addr, u64 len, u64 *translated,
> -                          struct bio_vec iov[],
> -                          int iov_size, u32 perm)
> +                          void *iov, int iov_size, bool iovec, u32 perm)

I think this is an improvement, but we're doing type erasure here. I
don't think it is a big deal since the function is not exported, it's
pretty contained in this file, so I'd ack this version too. I'm just
throwing ideas here:

a) typedef the union {iovec, bio_vec} and use that type in the parameter.

As a drawback, that union feels out of place in this file. Is this the
only place where it is needed? I don't see other similar uses in the
kernel.

b) To convert from iov to bio_iov at return
The drawback is the extra processing if the compiler is not smart
enough to inline it. I prefer the previous one but I didn't want to
omit it, just in case.

Thanks!

>  {
>         struct vhost_iotlb_map *map;
>         struct vhost_iotlb *iotlb = vrh->iotlb;
> @@ -1107,7 +1106,7 @@ static int iotlb_translate(const struct vringh *vrh,
>         spin_lock(vrh->iotlb_lock);
>
>         while (len > s) {
> -               u64 size, pa, pfn;
> +               u64 size;
>
>                 if (unlikely(ret >= iov_size)) {
>                         ret = -ENOBUFS;
> @@ -1124,10 +1123,22 @@ static int iotlb_translate(const struct vringh *vrh,
>                 }
>
>                 size = map->size - addr + map->start;
> -               pa = map->addr + addr - map->start;
> -               pfn = pa >> PAGE_SHIFT;
> -               bvec_set_page(&iov[ret], pfn_to_page(pfn), min(len - s, size),
> -                             pa & (PAGE_SIZE - 1));
> +               if (iovec) {
> +                       struct iovec *iovec = iov;
> +
> +                       iovec[ret].iov_len = min(len - s, size);
> +                       iovec[ret].iov_base = (void __user *)(unsigned long)
> +                                             (map->addr + addr - map->start);
> +               } else {
> +                       u64 pa = map->addr + addr - map->start;
> +                       u64 pfn = pa >> PAGE_SHIFT;
> +                       struct bio_vec *bvec = iov;
> +
> +                       bvec_set_page(&bvec[ret], pfn_to_page(pfn),
> +                                     min(len - s, size),
> +                                     pa & (PAGE_SIZE - 1));
> +               }
> +
>                 s += size;
>                 addr += size;
>                 ++ret;
> @@ -1141,26 +1152,38 @@ static int iotlb_translate(const struct vringh *vrh,
>         return ret;
>  }
>
> +#define IOTLB_IOV_SIZE 16
> +
>  static inline int copy_from_iotlb(const struct vringh *vrh, void *dst,
>                                   void *src, size_t len)
>  {
>         u64 total_translated = 0;
>
>         while (total_translated < len) {
> -               struct bio_vec iov[16];
> +               union {
> +                       struct iovec iovec[IOTLB_IOV_SIZE];
> +                       struct bio_vec bvec[IOTLB_IOV_SIZE];
> +               } iov;
>                 struct iov_iter iter;
>                 u64 translated;
>                 int ret;
>
>                 ret = iotlb_translate(vrh, (u64)(uintptr_t)src,
>                                       len - total_translated, &translated,
> -                                     iov, ARRAY_SIZE(iov), VHOST_MAP_RO);
> +                                     &iov, IOTLB_IOV_SIZE, vrh->use_va,
> +                                     VHOST_MAP_RO);
>                 if (ret == -ENOBUFS)
> -                       ret = ARRAY_SIZE(iov);
> +                       ret = IOTLB_IOV_SIZE;
>                 else if (ret < 0)
>                         return ret;
>
> -               iov_iter_bvec(&iter, ITER_SOURCE, iov, ret, translated);
> +               if (vrh->use_va) {
> +                       iov_iter_init(&iter, ITER_SOURCE, iov.iovec, ret,
> +                                     translated);
> +               } else {
> +                       iov_iter_bvec(&iter, ITER_SOURCE, iov.bvec, ret,
> +                                     translated);
> +               }
>
>                 ret = copy_from_iter(dst, translated, &iter);
>                 if (ret < 0)
> @@ -1180,20 +1203,30 @@ static inline int copy_to_iotlb(const struct vringh *vrh, void *dst,
>         u64 total_translated = 0;
>
>         while (total_translated < len) {
> -               struct bio_vec iov[16];
> +               union {
> +                       struct iovec iovec[IOTLB_IOV_SIZE];
> +                       struct bio_vec bvec[IOTLB_IOV_SIZE];
> +               } iov;
>                 struct iov_iter iter;
>                 u64 translated;
>                 int ret;
>
>                 ret = iotlb_translate(vrh, (u64)(uintptr_t)dst,
>                                       len - total_translated, &translated,
> -                                     iov, ARRAY_SIZE(iov), VHOST_MAP_WO);
> +                                     &iov, IOTLB_IOV_SIZE, vrh->use_va,
> +                                     VHOST_MAP_WO);
>                 if (ret == -ENOBUFS)
> -                       ret = ARRAY_SIZE(iov);
> +                       ret = IOTLB_IOV_SIZE;
>                 else if (ret < 0)
>                         return ret;
>
> -               iov_iter_bvec(&iter, ITER_DEST, iov, ret, translated);
> +               if (vrh->use_va) {
> +                       iov_iter_init(&iter, ITER_DEST, iov.iovec, ret,
> +                                     translated);
> +               } else {
> +                       iov_iter_bvec(&iter, ITER_DEST, iov.bvec, ret,
> +                                     translated);
> +               }
>
>                 ret = copy_to_iter(src, translated, &iter);
>                 if (ret < 0)
> @@ -1210,20 +1243,32 @@ static inline int copy_to_iotlb(const struct vringh *vrh, void *dst,
>  static inline int getu16_iotlb(const struct vringh *vrh,
>                                u16 *val, const __virtio16 *p)
>  {
> -       struct bio_vec iov;
> -       void *kaddr, *from;
> +       union {
> +               struct iovec iovec;
> +               struct bio_vec bvec;
> +       } iov;
> +       __virtio16 tmp;
>         int ret;
>
>         /* Atomic read is needed for getu16 */
> -       ret = iotlb_translate(vrh, (u64)(uintptr_t)p, sizeof(*p), NULL,
> -                             &iov, 1, VHOST_MAP_RO);
> +       ret = iotlb_translate(vrh, (u64)(uintptr_t)p, sizeof(*p),
> +                             NULL, &iov, 1, vrh->use_va, VHOST_MAP_RO);
>         if (ret < 0)
>                 return ret;
>
> -       kaddr = kmap_local_page(iov.bv_page);
> -       from = kaddr + iov.bv_offset;
> -       *val = vringh16_to_cpu(vrh, READ_ONCE(*(__virtio16 *)from));
> -       kunmap_local(kaddr);
> +       if (vrh->use_va) {
> +               ret = __get_user(tmp, (__virtio16 __user *)iov.iovec.iov_base);
> +               if (ret)
> +                       return ret;
> +       } else {
> +               void *kaddr = kmap_local_page(iov.bvec.bv_page);
> +               void *from = kaddr + iov.bvec.bv_offset;
> +
> +               tmp = READ_ONCE(*(__virtio16 *)from);
> +               kunmap_local(kaddr);
> +       }
> +
> +       *val = vringh16_to_cpu(vrh, tmp);
>
>         return 0;
>  }
> @@ -1231,20 +1276,32 @@ static inline int getu16_iotlb(const struct vringh *vrh,
>  static inline int putu16_iotlb(const struct vringh *vrh,
>                                __virtio16 *p, u16 val)
>  {
> -       struct bio_vec iov;
> -       void *kaddr, *to;
> +       union {
> +               struct iovec iovec;
> +               struct bio_vec bvec;
> +       } iov;
> +       __virtio16 tmp;
>         int ret;
>
>         /* Atomic write is needed for putu16 */
> -       ret = iotlb_translate(vrh, (u64)(uintptr_t)p, sizeof(*p), NULL,
> -                             &iov, 1, VHOST_MAP_WO);
> +       ret = iotlb_translate(vrh, (u64)(uintptr_t)p, sizeof(*p),
> +                             NULL, &iov, 1, vrh->use_va, VHOST_MAP_RO);
>         if (ret < 0)
>                 return ret;
>
> -       kaddr = kmap_local_page(iov.bv_page);
> -       to = kaddr + iov.bv_offset;
> -       WRITE_ONCE(*(__virtio16 *)to, cpu_to_vringh16(vrh, val));
> -       kunmap_local(kaddr);
> +       tmp = cpu_to_vringh16(vrh, val);
> +
> +       if (vrh->use_va) {
> +               ret = __put_user(tmp, (__virtio16 __user *)iov.iovec.iov_base);
> +               if (ret)
> +                       return ret;
> +       } else {
> +               void *kaddr = kmap_local_page(iov.bvec.bv_page);
> +               void *to = kaddr + iov.bvec.bv_offset;
> +
> +               WRITE_ONCE(*(__virtio16 *)to, tmp);
> +               kunmap_local(kaddr);
> +       }
>
>         return 0;
>  }
>