Re: [PATCH rdma-rc] RDMA/mlx5: Add dummy umem to IB_MR_TYPE_DM

[Leon Romanovsky]

On Wed, Sep 15, 2021 at 02:08:25AM +0300, Leon Romanovsky wrote:
> From: Alaa Hleihel <alaa@xxxxxxxxxx>
> 
> After the cited patch, and for the case of IB_MR_TYPE_DM that doesn't
> have a umem (even though it is a user MR), function mlx5_free_priv_descs()
> will think that it's a kernel MR, leading to wrongly accessing mr->descs
> that will get wrong values in the union which leads to attempting to
> release resources that were not allocated in the first place.
> 
> For example:
>  DMA-API: mlx5_core 0000:08:00.1: device driver tries to free DMA memory it has not allocated [device address=0x0000000000000000] [size=0 bytes]
>  WARNING: CPU: 8 PID: 1021 at kernel/dma/debug.c:961 check_unmap+0x54f/0x8b0
>  RIP: 0010:check_unmap+0x54f/0x8b0
>  Call Trace:
>   debug_dma_unmap_page+0x57/0x60
>   mlx5_free_priv_descs+0x57/0x70 [mlx5_ib]
>   mlx5_ib_dereg_mr+0x1fb/0x3d0 [mlx5_ib]
>   ib_dereg_mr_user+0x60/0x140 [ib_core]
>   uverbs_destroy_uobject+0x59/0x210 [ib_uverbs]
>   uobj_destroy+0x3f/0x80 [ib_uverbs]
>   ib_uverbs_cmd_verbs+0x435/0xd10 [ib_uverbs]
>   ? uverbs_finalize_object+0x50/0x50 [ib_uverbs]
>   ? lock_acquire+0xc4/0x2e0
>   ? lock_acquired+0x12/0x380
>   ? lock_acquire+0xc4/0x2e0
>   ? lock_acquire+0xc4/0x2e0
>   ? ib_uverbs_ioctl+0x7c/0x140 [ib_uverbs]
>   ? lock_release+0x28a/0x400
>   ib_uverbs_ioctl+0xc0/0x140 [ib_uverbs]
>   ? ib_uverbs_ioctl+0x7c/0x140 [ib_uverbs]
>   __x64_sys_ioctl+0x7f/0xb0
>   do_syscall_64+0x38/0x90
> 
> Fix it by adding a dummy umem to IB_MR_TYPE_DM MRs.
> 
> Fixes: f18ec4223117 ("RDMA/mlx5: Use a union inside mlx5_ib_mr")
> Signed-off-by: Alaa Hleihel <alaa@xxxxxxxxxx>
> Signed-off-by: Leon Romanovsky <leonro@xxxxxxxxxx>
> ---
>  drivers/infiniband/core/umem.c  | 21 +++++++++++++++++++++
>  drivers/infiniband/hw/mlx5/mr.c |  5 +++++
>  include/rdma/ib_umem.h          |  5 +++++
>  3 files changed, 31 insertions(+)

Please drop this patch, it seems that the proposed solution is too naive.

Thanks