linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Eugenio Perez Martin <eperezma@redhat.com>
To: Dragos Tatulea <dtatulea@nvidia.com>
Cc: "xuanzhuo@linux.alibaba.com" <xuanzhuo@linux.alibaba.com>,
	Parav Pandit <parav@nvidia.com>,
	"virtualization@lists.linux-foundation.org" 
	<virtualization@lists.linux-foundation.org>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
	"si-wei.liu@oracle.com" <si-wei.liu@oracle.com>,
	"kvm@vger.kernel.org" <kvm@vger.kernel.org>,
	"jasowang@redhat.com" <jasowang@redhat.com>,
	"mst@redhat.com" <mst@redhat.com>
Subject: Re: [PATCH vhost v4 12/16] vdpa/mlx5: Improve mr update flow
Date: Tue, 24 Oct 2023 16:50:21 +0200	[thread overview]
Message-ID: <CAJaqyWfa1=OSypE7L=F-U1EXFSm2HeNKamOo--WsGtmRAwegyw@mail.gmail.com> (raw)
In-Reply-To: <63f678675497539fc695f9de38550170aa45382a.camel@nvidia.com>

On Mon, Oct 23, 2023 at 10:07 AM Dragos Tatulea <dtatulea@nvidia.com> wrote:
>
> On Fri, 2023-10-20 at 18:01 +0200, Eugenio Perez Martin wrote:
> > On Wed, Oct 18, 2023 at 7:21 PM Dragos Tatulea <dtatulea@nvidia.com> wrote:
> > >
> > > On Wed, 2023-10-18 at 20:14 +0300, Dragos Tatulea wrote:
> > > > The current flow for updating an mr works directly on mvdev->mr which
> > > > makes it cumbersome to handle multiple new mr structs.
> > > >
> > > > This patch makes the flow more straightforward by having
> > > > mlx5_vdpa_create_mr return a new mr which will update the old mr (if
> > > > any). The old mr will be deleted and unlinked from mvdev. For the case
> > > > when the iotlb is empty (not NULL), the old mr will be cleared.
> > > >
> > > > This change paves the way for adding mrs for different ASIDs.
> > > >
> > > > The initialized bool is no longer needed as mr is now a pointer in the
> > > > mlx5_vdpa_dev struct which will be NULL when not initialized.
> > > >
> > > > Acked-by: Eugenio Pérez <eperezma@redhat.com>
> > > > Acked-by: Jason Wang <jasowang@redhat.com>
> > > > Signed-off-by: Dragos Tatulea <dtatulea@nvidia.com>
> > > > ---
> > > >  drivers/vdpa/mlx5/core/mlx5_vdpa.h | 14 +++--
> > > >  drivers/vdpa/mlx5/core/mr.c        | 87 ++++++++++++++++--------------
> > > >  drivers/vdpa/mlx5/net/mlx5_vnet.c  | 53 +++++++++---------
> > > >  3 files changed, 82 insertions(+), 72 deletions(-)
> > > >
> > > > diff --git a/drivers/vdpa/mlx5/core/mlx5_vdpa.h
> > > > b/drivers/vdpa/mlx5/core/mlx5_vdpa.h
> > > > index 9c6ac42c21e1..bbe4335106bd 100644
> > > > --- a/drivers/vdpa/mlx5/core/mlx5_vdpa.h
> > > > +++ b/drivers/vdpa/mlx5/core/mlx5_vdpa.h
> > > > @@ -31,8 +31,6 @@ struct mlx5_vdpa_mr {
> > > >         struct list_head head;
> > > >         unsigned long num_directs;
> > > >         unsigned long num_klms;
> > > > -       /* state of dvq mr */
> > > > -       bool initialized;
> > > >
> > > >         bool user_mr;
> > > >  };
> > > > @@ -91,7 +89,7 @@ struct mlx5_vdpa_dev {
> > > >         u16 max_idx;
> > > >         u32 generation;
> > > >
> > > > -       struct mlx5_vdpa_mr mr;
> > > > +       struct mlx5_vdpa_mr *mr;
> > > >         /* serialize mr access */
> > > >         struct mutex mr_mtx;
> > > >         struct mlx5_control_vq cvq;
> > > > @@ -114,14 +112,14 @@ void mlx5_vdpa_free_resources(struct mlx5_vdpa_dev
> > > > *mvdev);
> > > >  int mlx5_vdpa_create_mkey(struct mlx5_vdpa_dev *mvdev, u32 *mkey, u32
> > > > *in,
> > > >                           int inlen);
> > > >  int mlx5_vdpa_destroy_mkey(struct mlx5_vdpa_dev *mvdev, u32 mkey);
> > > > -int mlx5_vdpa_handle_set_map(struct mlx5_vdpa_dev *mvdev, struct
> > > > vhost_iotlb
> > > > *iotlb,
> > > > -                            bool *change_map, unsigned int asid);
> > > > -int mlx5_vdpa_create_mr(struct mlx5_vdpa_dev *mvdev,
> > > > -                       struct mlx5_vdpa_mr *mr,
> > > > -                       struct vhost_iotlb *iotlb);
> > > > +struct mlx5_vdpa_mr *mlx5_vdpa_create_mr(struct mlx5_vdpa_dev *mvdev,
> > > > +                                        struct vhost_iotlb *iotlb);
> > > >  void mlx5_vdpa_destroy_mr_resources(struct mlx5_vdpa_dev *mvdev);
> > > >  void mlx5_vdpa_destroy_mr(struct mlx5_vdpa_dev *mvdev,
> > > >                           struct mlx5_vdpa_mr *mr);
> > > > +void mlx5_vdpa_update_mr(struct mlx5_vdpa_dev *mvdev,
> > > > +                        struct mlx5_vdpa_mr *mr,
> > > > +                        unsigned int asid);
> > > >  int mlx5_vdpa_update_cvq_iotlb(struct mlx5_vdpa_dev *mvdev,
> > > >                                 struct vhost_iotlb *iotlb,
> > > >                                 unsigned int asid);
> > > > diff --git a/drivers/vdpa/mlx5/core/mr.c b/drivers/vdpa/mlx5/core/mr.c
> > > > index abd6a6fb122f..00eff5a07152 100644
> > > > --- a/drivers/vdpa/mlx5/core/mr.c
> > > > +++ b/drivers/vdpa/mlx5/core/mr.c
> > > > @@ -495,30 +495,51 @@ static void destroy_user_mr(struct mlx5_vdpa_dev
> > > > *mvdev,
> > > > struct mlx5_vdpa_mr *mr
> > > >
> > > >  static void _mlx5_vdpa_destroy_mr(struct mlx5_vdpa_dev *mvdev, struct
> > > > mlx5_vdpa_mr *mr)
> > > >  {
> > > > -       if (!mr->initialized)
> > > > -               return;
> > > > -
> > > >         if (mr->user_mr)
> > > >                 destroy_user_mr(mvdev, mr);
> > > >         else
> > > >                 destroy_dma_mr(mvdev, mr);
> > > > -
> > > > -       mr->initialized = false;
> > > >  }
> > > >
> > > >  void mlx5_vdpa_destroy_mr(struct mlx5_vdpa_dev *mvdev,
> > > >                           struct mlx5_vdpa_mr *mr)
> > > >  {
> > > > +       if (!mr)
> > > > +               return;
> > > > +
> > > >         mutex_lock(&mvdev->mr_mtx);
> > > >
> > > >         _mlx5_vdpa_destroy_mr(mvdev, mr);
> > > >
> > > > +       if (mvdev->mr == mr)
> > > > +               mvdev->mr = NULL;
> > > > +
> > > > +       mutex_unlock(&mvdev->mr_mtx);
> > > > +
> > > > +       kfree(mr);
> > > > +}
> > > > +
> > > > +void mlx5_vdpa_update_mr(struct mlx5_vdpa_dev *mvdev,
> > > > +                        struct mlx5_vdpa_mr *new_mr,
> > > > +                        unsigned int asid)
> > > > +{
> > > > +       struct mlx5_vdpa_mr *old_mr = mvdev->mr;
> > > > +
> > > > +       mutex_lock(&mvdev->mr_mtx);
> > > > +
> > > > +       mvdev->mr = new_mr;
> > > > +       if (old_mr) {
> > > > +               _mlx5_vdpa_destroy_mr(mvdev, old_mr);
> > > > +               kfree(old_mr);
> > > > +       }
> > > > +
> > > >         mutex_unlock(&mvdev->mr_mtx);
> > > > +
> > > >  }
> > > >
> > > >  void mlx5_vdpa_destroy_mr_resources(struct mlx5_vdpa_dev *mvdev)
> > > >  {
> > > > -       mlx5_vdpa_destroy_mr(mvdev, &mvdev->mr);
> > > > +       mlx5_vdpa_destroy_mr(mvdev, mvdev->mr);
> > > >         prune_iotlb(mvdev);
> > > >  }
> > > >
> > > > @@ -528,52 +549,36 @@ static int _mlx5_vdpa_create_mr(struct mlx5_vdpa_dev
> > > > *mvdev,
> > > >  {
> > > >         int err;
> > > >
> > > > -       if (mr->initialized)
> > > > -               return 0;
> > > > -
> > > >         if (iotlb)
> > > >                 err = create_user_mr(mvdev, mr, iotlb);
> > > >         else
> > > >                 err = create_dma_mr(mvdev, mr);
> > > >
> > > > -       if (err)
> > > > -               return err;
> > > > -
> > > > -       mr->initialized = true;
> > > > -
> > > > -       return 0;
> > > > +       return err;
> > > >  }
> > > >
> > > > -int mlx5_vdpa_create_mr(struct mlx5_vdpa_dev *mvdev,
> > > > -                       struct mlx5_vdpa_mr *mr,
> > > > -                       struct vhost_iotlb *iotlb)
> > > > +struct mlx5_vdpa_mr *mlx5_vdpa_create_mr(struct mlx5_vdpa_dev *mvdev,
> > > > +                                        struct vhost_iotlb *iotlb)
> > > >  {
> > > > +       struct mlx5_vdpa_mr *mr;
> > > >         int err;
> > > >
> > > > +       mr = kzalloc(sizeof(*mr), GFP_KERNEL);
> > > > +       if (!mr)
> > > > +               return ERR_PTR(-ENOMEM);
> > > > +
> > > >         mutex_lock(&mvdev->mr_mtx);
> > > >         err = _mlx5_vdpa_create_mr(mvdev, mr, iotlb);
> > > >         mutex_unlock(&mvdev->mr_mtx);
> > > >
> > > > -       return err;
> > > > -}
> > > > -
> > > > -int mlx5_vdpa_handle_set_map(struct mlx5_vdpa_dev *mvdev, struct
> > > > vhost_iotlb
> > > > *iotlb,
> > > > -                            bool *change_map, unsigned int asid)
> > > > -{
> > > > -       struct mlx5_vdpa_mr *mr = &mvdev->mr;
> > > > -       int err = 0;
> > > > +       if (err)
> > > > +               goto out_err;
> > > >
> > > > -       *change_map = false;
> > > > -       mutex_lock(&mvdev->mr_mtx);
> > > > -       if (mr->initialized) {
> > > > -               mlx5_vdpa_info(mvdev, "memory map update\n");
> > > > -               *change_map = true;
> > > > -       }
> > > > -       if (!*change_map)
> > > > -               err = _mlx5_vdpa_create_mr(mvdev, mr, iotlb);
> > > > -       mutex_unlock(&mvdev->mr_mtx);
> > > > +       return mr;
> > > >
> > > > -       return err;
> > > > +out_err:
> > > > +       kfree(mr);
> > > > +       return ERR_PTR(err);
> > > >  }
> > > >
> > > >  int mlx5_vdpa_update_cvq_iotlb(struct mlx5_vdpa_dev *mvdev,
> > > > @@ -597,11 +602,13 @@ int mlx5_vdpa_update_cvq_iotlb(struct mlx5_vdpa_dev
> > > > *mvdev,
> > > >
> > > >  int mlx5_vdpa_create_dma_mr(struct mlx5_vdpa_dev *mvdev)
> > > >  {
> > > > -       int err;
> > > > +       struct mlx5_vdpa_mr *mr;
> > > >
> > > > -       err = mlx5_vdpa_create_mr(mvdev, &mvdev->mr, NULL);
> > > > -       if (err)
> > > > -               return err;
> > > > +       mr = mlx5_vdpa_create_mr(mvdev, NULL);
> > > > +       if (IS_ERR(mr))
> > > > +               return PTR_ERR(mr);
> > > > +
> > > > +       mlx5_vdpa_update_mr(mvdev, mr, 0);
> > > >
> > > >         return mlx5_vdpa_update_cvq_iotlb(mvdev, NULL, 0);
> > > >  }
> > > > diff --git a/drivers/vdpa/mlx5/net/mlx5_vnet.c
> > > > b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> > > > index 256fdd80c321..7b878995b6aa 100644
> > > > --- a/drivers/vdpa/mlx5/net/mlx5_vnet.c
> > > > +++ b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> > > > @@ -873,7 +873,7 @@ static int create_virtqueue(struct mlx5_vdpa_net
> > > > *ndev,
> > > > struct mlx5_vdpa_virtque
> > > >         MLX5_SET64(virtio_q, vq_ctx, desc_addr, mvq->desc_addr);
> > > >         MLX5_SET64(virtio_q, vq_ctx, used_addr, mvq->device_addr);
> > > >         MLX5_SET64(virtio_q, vq_ctx, available_addr, mvq->driver_addr);
> > > > -       MLX5_SET(virtio_q, vq_ctx, virtio_q_mkey, ndev->mvdev.mr.mkey);
> > > > +       MLX5_SET(virtio_q, vq_ctx, virtio_q_mkey, ndev->mvdev.mr->mkey);
> > > >         MLX5_SET(virtio_q, vq_ctx, umem_1_id, mvq->umem1.id);
> > > >         MLX5_SET(virtio_q, vq_ctx, umem_1_size, mvq->umem1.size);
> > > >         MLX5_SET(virtio_q, vq_ctx, umem_2_id, mvq->umem2.id);
> > > > @@ -2633,7 +2633,7 @@ static void restore_channels_info(struct
> > > > mlx5_vdpa_net
> > > > *ndev)
> > > >  }
> > > >
> > > >  static int mlx5_vdpa_change_map(struct mlx5_vdpa_dev *mvdev,
> > > > -                               struct vhost_iotlb *iotlb, unsigned int
> > > > asid)
> > > > +                               struct mlx5_vdpa_mr *new_mr, unsigned int
> > > > asid)
> > > >  {
> > > >         struct mlx5_vdpa_net *ndev = to_mlx5_vdpa_ndev(mvdev);
> > > >         int err;
> > > > @@ -2641,27 +2641,18 @@ static int mlx5_vdpa_change_map(struct
> > > > mlx5_vdpa_dev
> > > > *mvdev,
> > > >         suspend_vqs(ndev);
> > > >         err = save_channels_info(ndev);
> > > >         if (err)
> > > > -               goto err_mr;
> > > > +               return err;
> > > >
> > > >         teardown_driver(ndev);
> > > > -       mlx5_vdpa_destroy_mr(mvdev, &mvdev->mr);
> > > > -       err = mlx5_vdpa_create_mr(mvdev, &mvdev->mr, iotlb);
> > > > -       if (err)
> > > > -               goto err_mr;
> > > > +
> > > > +       mlx5_vdpa_update_mr(mvdev, new_mr, asid);
> > > >
> > > >         if (!(mvdev->status & VIRTIO_CONFIG_S_DRIVER_OK) || mvdev-
> > > > >suspended)
> > > > -               goto err_mr;
> > > > +               return 0;
> > > >
> > > >         restore_channels_info(ndev);
> > > >         err = setup_driver(mvdev);
> > > > -       if (err)
> > > > -               goto err_setup;
> > > > -
> > > > -       return 0;
> > > >
> > > > -err_setup:
> > > > -       mlx5_vdpa_destroy_mr(mvdev, &mvdev->mr);
> > > > -err_mr:
> > > >         return err;
> > > >  }
> > > >
> > > > @@ -2875,26 +2866,40 @@ static u32 mlx5_vdpa_get_generation(struct
> > > > vdpa_device
> > > > *vdev)
> > > >  static int set_map_data(struct mlx5_vdpa_dev *mvdev, struct vhost_iotlb
> > > > *iotlb,
> > > >                         unsigned int asid)
> > > >  {
> > > > -       bool change_map;
> > > > +       struct mlx5_vdpa_mr *new_mr;
> > > >         int err;
> > > >
> > > >         if (mvdev->group2asid[MLX5_VDPA_DATAVQ_GROUP] != asid)
> > > >                 goto end;
> > > >
> > > > -       err = mlx5_vdpa_handle_set_map(mvdev, iotlb, &change_map, asid);
> > > > -       if (err) {
> > > > -               mlx5_vdpa_warn(mvdev, "set map failed(%d)\n", err);
> > > > -               return err;
> > > > +       if (vhost_iotlb_itree_first(iotlb, 0, U64_MAX)) {
> > > > +               new_mr = mlx5_vdpa_create_mr(mvdev, iotlb);
> > > > +               if (IS_ERR(new_mr)) {
> > > > +                       err = PTR_ERR(new_mr);
> > > > +                       mlx5_vdpa_warn(mvdev, "create map failed(%d)\n",
> > > > err);
> > > > +                       return err;
> > > > +               }
> > > > +       } else {
> > > > +               /* Empty iotlbs don't have an mr but will clear the
> > > > previous
> > > > mr. */
> > > > +               new_mr = NULL;
> > > >         }
> > > Hi Jason and/or Eugenio, could you have a quick look at this part of the
> > > patch
> > > that changed please?
> > >
> > > Thanks,
> > > Dragos
> > > >
> > > > -       if (change_map) {
> > > > -               err = mlx5_vdpa_change_map(mvdev, iotlb, asid);
> > > > -               if (err)
> > > > -                       return err;
> > > > +       if (!mvdev->mr) {
> > > > +               mlx5_vdpa_update_mr(mvdev, new_mr, asid);
> > > > +       } else {
> > > > +               err = mlx5_vdpa_change_map(mvdev, new_mr, asid);
> > > > +               if (err) {
> > > > +                       mlx5_vdpa_warn(mvdev, "change map failed(%d)\n",
> > > > err);
> > > > +                       goto out_err;
> > > > +               }
> > > >         }
> > > >
> > > >  end:
> > > >         return mlx5_vdpa_update_cvq_iotlb(mvdev, iotlb, asid);
> > > > +
> > > > +out_err:
> > > > +       mlx5_vdpa_destroy_mr(mvdev, new_mr);
> >
> > Is it possible to reach this mlx5_vdpa_destroy_mr call with new_mr ==
> > NULL? Like:
> > * iotlb does not have any entries
> > * mdev already has a mr
> > * mlx5_vdpa_change_map fails
> >
> It could happen.
>
> > If I'm not wrong, mlx5_vdpa_destroy_mr may dereference new_mr through
> > _mlx5_vdpa_destroy_mr -> vhost_iotlb_free(mr->iotlb).
> >
> mlx5_vdpa_destroy_mr checks for mr being NULL first.
>
> The other place where _mlx5_vdpa_destroy_mr gets called is from
> mlx5_vdpa_update_mr on the old mr IF it exists (it is not NULL).
>
> This looks safe to me.
>

Right, I don't know how I missed that, sorry for the noise! Already
acked in the patch message, but just in case:

Acked-by: Eugenio Pérez <eperezma@redhat.com>

Thanks!

> Thanks,
> Dragos
>
> > Am I missing something?
> >
> > Thanks!
> >
> >
> >
> >
> > > > +       return err;
> > > >  }
> > > >
> > > >  static int mlx5_vdpa_set_map(struct vdpa_device *vdev, unsigned int asid,
> > >
> >
>


  reply	other threads:[~2023-10-24 14:51 UTC|newest]

Thread overview: 24+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2023-10-18 17:14 [PATCH vhost v4 00/16] vdpa: Add support for vq descriptor mappings Dragos Tatulea
2023-10-18 17:14 ` [PATCH vhost v4 01/16] vdpa/mlx5: Expose descriptor group mkey hw capability Dragos Tatulea
2023-10-18 17:14 ` [PATCH vhost v4 02/16] vdpa: introduce dedicated descriptor group for virtqueue Dragos Tatulea
2023-10-18 17:14 ` [PATCH vhost v4 03/16] vhost-vdpa: introduce descriptor group backend feature Dragos Tatulea
2023-10-18 17:14 ` [PATCH vhost v4 04/16] vhost-vdpa: uAPI to get dedicated descriptor group id Dragos Tatulea
2023-10-18 17:14 ` [PATCH vhost v4 05/16] vdpa/mlx5: Create helper function for dma mappings Dragos Tatulea
2023-10-18 17:14 ` [PATCH vhost v4 06/16] vdpa/mlx5: Decouple cvq iotlb handling from hw mapping code Dragos Tatulea
2023-10-18 17:14 ` [PATCH vhost v4 07/16] vdpa/mlx5: Take cvq iotlb lock during refresh Dragos Tatulea
2023-10-18 17:14 ` [PATCH vhost v4 08/16] vdpa/mlx5: Collapse "dvq" mr add/delete functions Dragos Tatulea
2023-10-18 17:14 ` [PATCH vhost v4 09/16] vdpa/mlx5: Rename mr destroy functions Dragos Tatulea
2023-10-18 17:14 ` [PATCH vhost v4 10/16] vdpa/mlx5: Allow creation/deletion of any given mr struct Dragos Tatulea
2023-10-18 17:14 ` [PATCH vhost v4 11/16] vdpa/mlx5: Move mr mutex out of " Dragos Tatulea
2023-10-18 17:14 ` [PATCH vhost v4 12/16] vdpa/mlx5: Improve mr update flow Dragos Tatulea
2023-10-18 17:21   ` Dragos Tatulea
2023-10-20 16:01     ` Eugenio Perez Martin
2023-10-23  8:07       ` Dragos Tatulea
2023-10-24 14:50         ` Eugenio Perez Martin [this message]
2023-10-18 17:14 ` [PATCH vhost v4 13/16] vdpa/mlx5: Introduce mr for vq descriptor Dragos Tatulea
2023-10-18 17:14 ` [PATCH vhost v4 14/16] vdpa/mlx5: Enable hw support for vq descriptor mapping Dragos Tatulea
2023-10-18 17:14 ` [PATCH vhost v4 15/16] vdpa/mlx5: Make iotlb helper functions more generic Dragos Tatulea
2023-10-18 17:14 ` [PATCH vhost v4 16/16] vdpa/mlx5: Update cvq iotlb mapping on ASID change Dragos Tatulea
2023-10-18 17:24 ` [PATCH vhost v4 00/16] vdpa: Add support for vq descriptor mappings Michael S. Tsirkin
2023-10-19 23:47 ` Si-Wei Liu
2023-10-24  6:45   ` Lei Yang

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to='CAJaqyWfa1=OSypE7L=F-U1EXFSm2HeNKamOo--WsGtmRAwegyw@mail.gmail.com' \
    --to=eperezma@redhat.com \
    --cc=dtatulea@nvidia.com \
    --cc=jasowang@redhat.com \
    --cc=kvm@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=mst@redhat.com \
    --cc=parav@nvidia.com \
    --cc=si-wei.liu@oracle.com \
    --cc=virtualization@lists.linux-foundation.org \
    --cc=xuanzhuo@linux.alibaba.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).