2021-05-31 17:05:05

by Eli Cohen

Subject: [PATCH v1 1/2] vdpa/mlx5: Support creating resources with uid == 0

Currently all resources must be created with uid != 0 which is essential
when userspace processes are allocating virtquueue resources. Since this
is a kernel implementation, it is perfectly legal to open resources with
uid == 0.

In case firmware supports, avoid allocating user context.

Signed-off-by: Eli Cohen <[email protected]>
---
v0 --> v1:
Fix typo and modify phrasing

drivers/vdpa/mlx5/core/resources.c | 6 ++++++
include/linux/mlx5/mlx5_ifc.h | 4 +++-
2 files changed, 9 insertions(+), 1 deletion(-)

diff --git a/drivers/vdpa/mlx5/core/resources.c b/drivers/vdpa/mlx5/core/resources.c
index 6521cbd0f5c2..836ab9ef0fa6 100644
--- a/drivers/vdpa/mlx5/core/resources.c
+++ b/drivers/vdpa/mlx5/core/resources.c
@@ -54,6 +54,9 @@ static int create_uctx(struct mlx5_vdpa_dev *mvdev, u16 *uid)
void *in;
int err;

+ if (MLX5_CAP_GEN(mvdev->mdev, umem_uid_0))
+ return 0;
+
/* 0 means not supported */
if (!MLX5_CAP_GEN(mvdev->mdev, log_max_uctx))
return -EOPNOTSUPP;
@@ -79,6 +82,9 @@ static void destroy_uctx(struct mlx5_vdpa_dev *mvdev, u32 uid)
u32 out[MLX5_ST_SZ_DW(destroy_uctx_out)] = {};
u32 in[MLX5_ST_SZ_DW(destroy_uctx_in)] = {};

+ if (!uid)
+ return;
+
MLX5_SET(destroy_uctx_in, in, opcode, MLX5_CMD_OP_DESTROY_UCTX);
MLX5_SET(destroy_uctx_in, in, uid, uid);

diff --git a/include/linux/mlx5/mlx5_ifc.h b/include/linux/mlx5/mlx5_ifc.h
index 9c68b2da14c6..606d2aeacad4 100644
--- a/include/linux/mlx5/mlx5_ifc.h
+++ b/include/linux/mlx5/mlx5_ifc.h
@@ -1487,7 +1487,9 @@ struct mlx5_ifc_cmd_hca_cap_bits {
u8 uar_4k[0x1];
u8 reserved_at_241[0x9];
u8 uar_sz[0x6];
- u8 reserved_at_250[0x8];
+ u8 reserved_at_248[0x2];
+ u8 umem_uid_0[0x1];
+ u8 reserved_at_250[0x5];
u8 log_pg_sz[0x8];

u8 bf[0x1];
--
2.31.1

2021-05-31 22:18:17

by Michael S. Tsirkin

Subject: Re: [PATCH v1 1/2] vdpa/mlx5: Support creating resources with uid == 0

On Mon, May 31, 2021 at 07:04:04PM +0300, Eli Cohen wrote:
> Currently all resources must be created with uid != 0 which is essential
> when userspace processes are allocating virtquueue resources. Since this
> is a kernel implementation, it is perfectly legal to open resources with
> uid == 0.
>
> In case firmware supports, avoid allocating user context.
>
> Signed-off-by: Eli Cohen <[email protected]>
> ---
> v0 --> v1:
> Fix typo and modify phrasing

threading's broken now

> drivers/vdpa/mlx5/core/resources.c | 6 ++++++
> include/linux/mlx5/mlx5_ifc.h | 4 +++-
> 2 files changed, 9 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/vdpa/mlx5/core/resources.c b/drivers/vdpa/mlx5/core/resources.c
> index 6521cbd0f5c2..836ab9ef0fa6 100644
> --- a/drivers/vdpa/mlx5/core/resources.c
> +++ b/drivers/vdpa/mlx5/core/resources.c
> @@ -54,6 +54,9 @@ static int create_uctx(struct mlx5_vdpa_dev *mvdev, u16 *uid)
> void *in;
> int err;
>
> + if (MLX5_CAP_GEN(mvdev->mdev, umem_uid_0))
> + return 0;
> +
> /* 0 means not supported */
> if (!MLX5_CAP_GEN(mvdev->mdev, log_max_uctx))
> return -EOPNOTSUPP;
> @@ -79,6 +82,9 @@ static void destroy_uctx(struct mlx5_vdpa_dev *mvdev, u32 uid)
> u32 out[MLX5_ST_SZ_DW(destroy_uctx_out)] = {};
> u32 in[MLX5_ST_SZ_DW(destroy_uctx_in)] = {};
>
> + if (!uid)
> + return;
> +
> MLX5_SET(destroy_uctx_in, in, opcode, MLX5_CMD_OP_DESTROY_UCTX);
> MLX5_SET(destroy_uctx_in, in, uid, uid);
>
> diff --git a/include/linux/mlx5/mlx5_ifc.h b/include/linux/mlx5/mlx5_ifc.h
> index 9c68b2da14c6..606d2aeacad4 100644
> --- a/include/linux/mlx5/mlx5_ifc.h
> +++ b/include/linux/mlx5/mlx5_ifc.h
> @@ -1487,7 +1487,9 @@ struct mlx5_ifc_cmd_hca_cap_bits {
> u8 uar_4k[0x1];
> u8 reserved_at_241[0x9];
> u8 uar_sz[0x6];
> - u8 reserved_at_250[0x8];
> + u8 reserved_at_248[0x2];
> + u8 umem_uid_0[0x1];
> + u8 reserved_at_250[0x5];
> u8 log_pg_sz[0x8];
>
> u8 bf[0x1];
> --
> 2.31.1

2021-06-01 01:57:27

by Jason Wang

Subject: Re: [PATCH v1 1/2] vdpa/mlx5: Support creating resources with uid == 0

?? 2021/6/1 ????12:04, Eli Cohen д??:
> Currently all resources must be created with uid != 0 which is essential
> when userspace processes are allocating virtquueue resources. Since this
> is a kernel implementation, it is perfectly legal to open resources with
> uid == 0.
>
> In case firmware supports, avoid allocating user context.
>
> Signed-off-by: Eli Cohen <[email protected]>


Acked-by: Jason Wang <[email protected]>


> ---
> v0 --> v1:
> Fix typo and modify phrasing
>
> drivers/vdpa/mlx5/core/resources.c | 6 ++++++
> include/linux/mlx5/mlx5_ifc.h | 4 +++-
> 2 files changed, 9 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/vdpa/mlx5/core/resources.c b/drivers/vdpa/mlx5/core/resources.c
> index 6521cbd0f5c2..836ab9ef0fa6 100644
> --- a/drivers/vdpa/mlx5/core/resources.c
> +++ b/drivers/vdpa/mlx5/core/resources.c
> @@ -54,6 +54,9 @@ static int create_uctx(struct mlx5_vdpa_dev *mvdev, u16 *uid)
> void *in;
> int err;
>
> + if (MLX5_CAP_GEN(mvdev->mdev, umem_uid_0))
> + return 0;
> +
> /* 0 means not supported */
> if (!MLX5_CAP_GEN(mvdev->mdev, log_max_uctx))
> return -EOPNOTSUPP;
> @@ -79,6 +82,9 @@ static void destroy_uctx(struct mlx5_vdpa_dev *mvdev, u32 uid)
> u32 out[MLX5_ST_SZ_DW(destroy_uctx_out)] = {};
> u32 in[MLX5_ST_SZ_DW(destroy_uctx_in)] = {};
>
> + if (!uid)
> + return;
> +
> MLX5_SET(destroy_uctx_in, in, opcode, MLX5_CMD_OP_DESTROY_UCTX);
> MLX5_SET(destroy_uctx_in, in, uid, uid);
>
> diff --git a/include/linux/mlx5/mlx5_ifc.h b/include/linux/mlx5/mlx5_ifc.h
> index 9c68b2da14c6..606d2aeacad4 100644
> --- a/include/linux/mlx5/mlx5_ifc.h
> +++ b/include/linux/mlx5/mlx5_ifc.h
> @@ -1487,7 +1487,9 @@ struct mlx5_ifc_cmd_hca_cap_bits {
> u8 uar_4k[0x1];
> u8 reserved_at_241[0x9];
> u8 uar_sz[0x6];
> - u8 reserved_at_250[0x8];
> + u8 reserved_at_248[0x2];
> + u8 umem_uid_0[0x1];
> + u8 reserved_at_250[0x5];
> u8 log_pg_sz[0x8];
>
> u8 bf[0x1];

2021-06-01 04:36:18

by Eli Cohen

Subject: Re: [PATCH v1 1/2] vdpa/mlx5: Support creating resources with uid == 0

On Mon, May 31, 2021 at 06:17:12PM -0400, Michael S. Tsirkin wrote:
> On Mon, May 31, 2021 at 07:04:04PM +0300, Eli Cohen wrote:
> > Currently all resources must be created with uid != 0 which is essential
> > when userspace processes are allocating virtquueue resources. Since this
> > is a kernel implementation, it is perfectly legal to open resources with
> > uid == 0.
> >
> > In case firmware supports, avoid allocating user context.
> >
> > Signed-off-by: Eli Cohen <[email protected]>
> > ---
> > v0 --> v1:
> > Fix typo and modify phrasing
>
> threading's broken now

Sorry about this, will repost again.
>
> > drivers/vdpa/mlx5/core/resources.c | 6 ++++++
> > include/linux/mlx5/mlx5_ifc.h | 4 +++-
> > 2 files changed, 9 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/vdpa/mlx5/core/resources.c b/drivers/vdpa/mlx5/core/resources.c
> > index 6521cbd0f5c2..836ab9ef0fa6 100644
> > --- a/drivers/vdpa/mlx5/core/resources.c
> > +++ b/drivers/vdpa/mlx5/core/resources.c
> > @@ -54,6 +54,9 @@ static int create_uctx(struct mlx5_vdpa_dev *mvdev, u16 *uid)
> > void *in;
> > int err;
> >
> > + if (MLX5_CAP_GEN(mvdev->mdev, umem_uid_0))
> > + return 0;
> > +
> > /* 0 means not supported */
> > if (!MLX5_CAP_GEN(mvdev->mdev, log_max_uctx))
> > return -EOPNOTSUPP;
> > @@ -79,6 +82,9 @@ static void destroy_uctx(struct mlx5_vdpa_dev *mvdev, u32 uid)
> > u32 out[MLX5_ST_SZ_DW(destroy_uctx_out)] = {};
> > u32 in[MLX5_ST_SZ_DW(destroy_uctx_in)] = {};
> >
> > + if (!uid)
> > + return;
> > +
> > MLX5_SET(destroy_uctx_in, in, opcode, MLX5_CMD_OP_DESTROY_UCTX);
> > MLX5_SET(destroy_uctx_in, in, uid, uid);
> >
> > diff --git a/include/linux/mlx5/mlx5_ifc.h b/include/linux/mlx5/mlx5_ifc.h
> > index 9c68b2da14c6..606d2aeacad4 100644
> > --- a/include/linux/mlx5/mlx5_ifc.h
> > +++ b/include/linux/mlx5/mlx5_ifc.h
> > @@ -1487,7 +1487,9 @@ struct mlx5_ifc_cmd_hca_cap_bits {
> > u8 uar_4k[0x1];
> > u8 reserved_at_241[0x9];
> > u8 uar_sz[0x6];
> > - u8 reserved_at_250[0x8];
> > + u8 reserved_at_248[0x2];
> > + u8 umem_uid_0[0x1];
> > + u8 reserved_at_250[0x5];
> > u8 log_pg_sz[0x8];
> >
> > u8 bf[0x1];
> > --
> > 2.31.1
>