Re: [bug report] net/mlx5e: E-Switch, Maintain vhca_id to vport_num mapping

Re: [bug report] net/mlx5e: E-Switch, Maintain vhca_id to vport_num mapping

[Leon Romanovsky]

On Wed, Feb 10, 2021 at 11:51:51AM +0300, Dan Carpenter wrote:
> Hello Vlad Buslov,
>
> The patch 84ae9c1f29c0: "net/mlx5e: E-Switch, Maintain vhca_id to
> vport_num mapping" from Sep 23, 2020, leads to the following static
> checker warning:
>
> 	drivers/net/ethernet/mellanox/mlx5/core/vport.c:1170 mlx5_vport_get_other_func_cap()
> 	warn: odd binop '0x0 & 0x1'
>
> drivers/net/ethernet/mellanox/mlx5/core/vport.c
>   1168  int mlx5_vport_get_other_func_cap(struct mlx5_core_dev *dev, u16 function_id, void *out)
>   1169  {
>   1170          u16 opmod = (MLX5_CAP_GENERAL << 1) | (HCA_CAP_OPMOD_GET_MAX & 0x01);
>
> HCA_CAP_OPMOD_GET_MAX is zero.  The 0x01 is a magical number.
>
>   1171          u8 in[MLX5_ST_SZ_BYTES(query_hca_cap_in)] = {};
>   1172
>   1173          MLX5_SET(query_hca_cap_in, in, opcode, MLX5_CMD_OP_QUERY_HCA_CAP);
>   1174          MLX5_SET(query_hca_cap_in, in, op_mod, opmod);
>   1175          MLX5_SET(query_hca_cap_in, in, function_id, function_id);
>   1176          MLX5_SET(query_hca_cap_in, in, other_function, true);
>   1177          return mlx5_cmd_exec_inout(dev, query_hca_cap, in, out);
>   1178  }

Dan,

I'm running smatch which is based on 6193b3b71beb ("extra: fix some error pointer handling")
and I don't see this error. Should I run something special?

Thanks

>
> regards,
> dan carpenter

Re: [bug report] net/mlx5e: E-Switch, Maintain vhca_id to vport_num mapping

[Dan Carpenter]

On Wed, Feb 10, 2021 at 01:48:20PM +0200, Leon Romanovsky wrote:
> On Wed, Feb 10, 2021 at 11:51:51AM +0300, Dan Carpenter wrote:
> > Hello Vlad Buslov,
> >
> > The patch 84ae9c1f29c0: "net/mlx5e: E-Switch, Maintain vhca_id to
> > vport_num mapping" from Sep 23, 2020, leads to the following static
> > checker warning:
> >
> > 	drivers/net/ethernet/mellanox/mlx5/core/vport.c:1170 mlx5_vport_get_other_func_cap()
> > 	warn: odd binop '0x0 & 0x1'
> >
> > drivers/net/ethernet/mellanox/mlx5/core/vport.c
> >   1168  int mlx5_vport_get_other_func_cap(struct mlx5_core_dev *dev, u16 function_id, void *out)
> >   1169  {
> >   1170          u16 opmod = (MLX5_CAP_GENERAL << 1) | (HCA_CAP_OPMOD_GET_MAX & 0x01);
> >
> > HCA_CAP_OPMOD_GET_MAX is zero.  The 0x01 is a magical number.
> >
> >   1171          u8 in[MLX5_ST_SZ_BYTES(query_hca_cap_in)] = {};
> >   1172
> >   1173          MLX5_SET(query_hca_cap_in, in, opcode, MLX5_CMD_OP_QUERY_HCA_CAP);
> >   1174          MLX5_SET(query_hca_cap_in, in, op_mod, opmod);
> >   1175          MLX5_SET(query_hca_cap_in, in, function_id, function_id);
> >   1176          MLX5_SET(query_hca_cap_in, in, other_function, true);
> >   1177          return mlx5_cmd_exec_inout(dev, query_hca_cap, in, out);
> >   1178  }
> 
> Dan,
> 
> I'm running smatch which is based on 6193b3b71beb ("extra: fix some error pointer handling")
> and I don't see this error. Should I run something special?
> 

This check is too crap to publish.

The heuristic was "a bitwise AND which always results in zero" but a lot
of code does stuff like:  "data = 0x00 << 0 | 0x04 << 8 | 0x12 << 16;"
I could never figure out a way to make the check useful enough to
publish.

regards,
dan carpenter

Re: [bug report] net/mlx5e: E-Switch, Maintain vhca_id to vport_num mapping

[Leon Romanovsky]

On Wed, Feb 10, 2021 at 03:28:01PM +0300, Dan Carpenter wrote:
> On Wed, Feb 10, 2021 at 01:48:20PM +0200, Leon Romanovsky wrote:
> > On Wed, Feb 10, 2021 at 11:51:51AM +0300, Dan Carpenter wrote:
> > > Hello Vlad Buslov,
> > >
> > > The patch 84ae9c1f29c0: "net/mlx5e: E-Switch, Maintain vhca_id to
> > > vport_num mapping" from Sep 23, 2020, leads to the following static
> > > checker warning:
> > >
> > > 	drivers/net/ethernet/mellanox/mlx5/core/vport.c:1170 mlx5_vport_get_other_func_cap()
> > > 	warn: odd binop '0x0 & 0x1'
> > >
> > > drivers/net/ethernet/mellanox/mlx5/core/vport.c
> > >   1168  int mlx5_vport_get_other_func_cap(struct mlx5_core_dev *dev, u16 function_id, void *out)
> > >   1169  {
> > >   1170          u16 opmod = (MLX5_CAP_GENERAL << 1) | (HCA_CAP_OPMOD_GET_MAX & 0x01);
> > >
> > > HCA_CAP_OPMOD_GET_MAX is zero.  The 0x01 is a magical number.
> > >
> > >   1171          u8 in[MLX5_ST_SZ_BYTES(query_hca_cap_in)] = {};
> > >   1172
> > >   1173          MLX5_SET(query_hca_cap_in, in, opcode, MLX5_CMD_OP_QUERY_HCA_CAP);
> > >   1174          MLX5_SET(query_hca_cap_in, in, op_mod, opmod);
> > >   1175          MLX5_SET(query_hca_cap_in, in, function_id, function_id);
> > >   1176          MLX5_SET(query_hca_cap_in, in, other_function, true);
> > >   1177          return mlx5_cmd_exec_inout(dev, query_hca_cap, in, out);
> > >   1178  }
> >
> > Dan,
> >
> > I'm running smatch which is based on 6193b3b71beb ("extra: fix some error pointer handling")
> > and I don't see this error. Should I run something special?
> >
>
> This check is too crap to publish.
>
> The heuristic was "a bitwise AND which always results in zero" but a lot
> of code does stuff like:  "data = 0x00 << 0 | 0x04 << 8 | 0x12 << 16;"
> I could never figure out a way to make the check useful enough to
> publish.

But you can warn about simple cases like above, which is constant zero
AND something.

Thanks

>
> regards,
> dan carpenter
>