Re: [PATCH mlx5-next 4/4] net/mlx5: Remove spinlock support from mlx5_write64

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Fri, Jan 18, 2019 at 11:43 PM Leon Romanovsky <leonro@xxxxxxxxxxxx> wrote:
>
> On Fri, Jan 18, 2019 at 04:33:13PM -0800, Saeed Mahameed wrote:
> > From: Maxim Mikityanskiy <maximmi@xxxxxxxxxxxx>
> >
> > As there is no user of mlx5_write64 that passes a spinlock to
> > mlx5_write64, remove this functionality and simplify the function.
> >
> > Signed-off-by: Maxim Mikityanskiy <maximmi@xxxxxxxxxxxx>
> > Reviewed-by: Eran Ben Elisha <eranbe@xxxxxxxxxxxx>
> > Signed-off-by: Saeed Mahameed <saeedm@xxxxxxxxxxxx>
> > ---
> >  drivers/infiniband/hw/mlx5/qp.c               |  2 +-
> >  drivers/net/ethernet/mellanox/mlx5/core/en.h  |  2 +-
> >  .../ethernet/mellanox/mlx5/core/fpga/conn.c   |  2 +-
> >  include/linux/mlx5/cq.h                       |  2 +-
> >  include/linux/mlx5/doorbell.h                 | 28 ++++---------------
> >  5 files changed, 9 insertions(+), 27 deletions(-)
> >
> > diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
> > index dd2ae640bc84..816c34ee91cf 100644
> > --- a/drivers/infiniband/hw/mlx5/qp.c
> > +++ b/drivers/infiniband/hw/mlx5/qp.c
> > @@ -5015,7 +5015,7 @@ static int _mlx5_ib_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr,
> >               wmb();
> >
> >               /* currently we support only regular doorbells */
> > -             mlx5_write64((__be32 *)ctrl, bf->bfreg->map + bf->offset, NULL);
> > +             mlx5_write64((__be32 *)ctrl, bf->bfreg->map + bf->offset);
> >               /* Make sure doorbells don't leak out of SQ spinlock
> >                * and reach the HCA out of order.
> >                */
> > diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
> > index 8fa8fdd30b85..2623d3fb6b96 100644
> > --- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
> > +++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h
> > @@ -916,7 +916,7 @@ void mlx5e_notify_hw(struct mlx5_wq_cyc *wq, u16 pc,
> >        */
> >       wmb();
> >
> > -     mlx5_write64((__be32 *)ctrl, uar_map, NULL);
> > +     mlx5_write64((__be32 *)ctrl, uar_map);
> >  }
> >
> >  static inline void mlx5e_cq_arm(struct mlx5e_cq *cq)
> > diff --git a/drivers/net/ethernet/mellanox/mlx5/core/fpga/conn.c b/drivers/net/ethernet/mellanox/mlx5/core/fpga/conn.c
> > index 873541ef4c1b..ca2296a2f9ee 100644
> > --- a/drivers/net/ethernet/mellanox/mlx5/core/fpga/conn.c
> > +++ b/drivers/net/ethernet/mellanox/mlx5/core/fpga/conn.c
> > @@ -135,7 +135,7 @@ static void mlx5_fpga_conn_notify_hw(struct mlx5_fpga_conn *conn, void *wqe)
> >       *conn->qp.wq.sq.db = cpu_to_be32(conn->qp.sq.pc);
> >       /* Make sure that doorbell record is visible before ringing */
> >       wmb();
> > -     mlx5_write64(wqe, conn->fdev->conn_res.uar->map + MLX5_BF_OFFSET, NULL);
> > +     mlx5_write64(wqe, conn->fdev->conn_res.uar->map + MLX5_BF_OFFSET);
> >  }
> >
> >  static void mlx5_fpga_conn_post_send(struct mlx5_fpga_conn *conn,
> > diff --git a/include/linux/mlx5/cq.h b/include/linux/mlx5/cq.h
> > index 612c8c2f2466..769326ea1d9b 100644
> > --- a/include/linux/mlx5/cq.h
> > +++ b/include/linux/mlx5/cq.h
> > @@ -170,7 +170,7 @@ static inline void mlx5_cq_arm(struct mlx5_core_cq *cq, u32 cmd,
> >       doorbell[0] = cpu_to_be32(sn << 28 | cmd | ci);
> >       doorbell[1] = cpu_to_be32(cq->cqn);
> >
> > -     mlx5_write64(doorbell, uar_page + MLX5_CQ_DOORBELL, NULL);
> > +     mlx5_write64(doorbell, uar_page + MLX5_CQ_DOORBELL);
> >  }
> >
> >  static inline void mlx5_cq_hold(struct mlx5_core_cq *cq)
> > diff --git a/include/linux/mlx5/doorbell.h b/include/linux/mlx5/doorbell.h
> > index 9ef3f9d00154..c12523cc8102 100644
> > --- a/include/linux/mlx5/doorbell.h
> > +++ b/include/linux/mlx5/doorbell.h
> > @@ -36,38 +36,20 @@
> >  #define MLX5_BF_OFFSET             0x800
> >  #define MLX5_CQ_DOORBELL      0x20
> >
> > -#if BITS_PER_LONG == 64
> >  /* Assume that we can just write a 64-bit doorbell atomically.  s390
> >   * actually doesn't have writeq() but S/390 systems don't even have
> > - * PCI so we won't worry about it.
> > + * PCI so we won't worry about it. Note that the write is not atomic
> > + * on 32-bit systems.
> >   */
> >
> > -static inline void mlx5_write64(__be32 val[2], void __iomem *dest,
> > -                             spinlock_t *doorbell_lock)
> > +static inline void mlx5_write64(__be32 val[2], void __iomem *dest)
> >  {
> > +#if BITS_PER_LONG == 64
> >       __raw_writeq(*(u64 *)val, dest);
> > -}
> > -
> >  #else
> > -
> > -/* Just fall back to a spinlock to protect the doorbell if
> > - * BITS_PER_LONG is 32 -- there's no portable way to do atomic 64-bit
> > - * MMIO writes.
> > - */
> > -
> > -static inline void mlx5_write64(__be32 val[2], void __iomem *dest,
> > -                             spinlock_t *doorbell_lock)
> > -{
> > -     unsigned long flags;
> > -
> > -     if (doorbell_lock)
> > -             spin_lock_irqsave(doorbell_lock, flags);
>
> Saeed, Maxim
>
> It is incomplete change. In your patch 3, you removed declaration
> of spinlock from 32bits compiles and made this write completely
> unsafe.
>
> You need to do one of two things:
> 1. Require CONFIG_64BIT and delete this 32bit code.

Not required, if something did compile for 32bit, why stop it now ?
it compiles today and it doesn't function, there is a big warning in
the code and no one is using a spinlock in 32bit arch,
so this is just an API cleanup for unused code path.

> 2. Declare global mlx5 DB spinlock and use on 32bit systems, something
> like this:
> #if BITS_PER_LONG == 64
>  __raw_writeq(*(u64 *)val, dest);
> #else
>   spin_lock_irqsave(doorbell_lock, flags);
>   __raw_writel((__force u32) val[0], dest);
>   __raw_writel((__force u32) val[1], dest + 4);
>    spin_unlock_irqrestore(doorbell_lock, flags);
> #endif
>

I would rather keep it broken than adding some global spinlock.

Every time we touch this code someone gets sensitive about it, this is
Just a cleanup patch !!
Bottom line, 32bit doesn't work, we are cleaning up API, don't make a
big deal out of it,
Please don't mix two different things, 32bit support is not related to
this patch at all.
if you want to fix 32bit, you can do it on top of this patch.

> Thanks
>
> >       __raw_writel((__force u32) val[0], dest);
> >       __raw_writel((__force u32) val[1], dest + 4);
> > -     if (doorbell_lock)
> > -             spin_unlock_irqrestore(doorbell_lock, flags);
> > -}
> > -
> >  #endif
> > +}
> >
> >  #endif /* MLX5_DOORBELL_H */
> > --
> > 2.20.1
> >



[Index of Archives]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Photo]     [Yosemite News]     [Yosemite Photos]     [Linux Kernel]     [Linux SCSI]     [XFree86]

  Powered by Linux