[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <316f5042-b47d-2cee-48de-514467817e7a@dev.mellanox.co.il>
Date: Wed, 30 May 2018 15:31:34 +0300
From: Yishai Hadas <yishaih@....mellanox.co.il>
To: Jason Gunthorpe <jgg@...pe.ca>
Cc: Leon Romanovsky <leon@...nel.org>,
Doug Ledford <dledford@...hat.com>,
Leon Romanovsky <leonro@...lanox.com>,
RDMA mailing list <linux-rdma@...r.kernel.org>,
Boris Pismenny <borisp@...lanox.com>,
Matan Barak <matanb@...lanox.com>,
Raed Salem <raeds@...lanox.com>,
Yishai Hadas <yishaih@...lanox.com>,
Saeed Mahameed <saeedm@...lanox.com>,
linux-netdev <netdev@...r.kernel.org>,
Alex Rosenbaum <alexr@...lanox.com>
Subject: Re: [PATCH mlx5-next v2 11/13] IB/mlx5: Add flow counters binding
support
On 5/29/2018 10:56 PM, Jason Gunthorpe wrote:
> On Tue, May 29, 2018 at 04:09:15PM +0300, Leon Romanovsky wrote:
>> diff --git a/include/uapi/rdma/mlx5-abi.h b/include/uapi/rdma/mlx5-abi.h
>> index 508ea8c82da7..ef3f430a7050 100644
>> +++ b/include/uapi/rdma/mlx5-abi.h
>> @@ -443,4 +443,18 @@ enum {
>> enum {
>> MLX5_IB_CLOCK_INFO_V1 = 0,
>> };
>> +
>> +struct mlx5_ib_flow_counters_data {
>> + __aligned_u64 counters_data;
>> + __u32 ncounters;
>> + __u32 reserved;
>> +};
>> +
>> +struct mlx5_ib_create_flow {
>> + __u32 ncounters_data;
>> + __u32 reserved;
>> + /* Following are counters data based on ncounters_data */
>> + struct mlx5_ib_flow_counters_data data[];
>> +};
>> +
>> #endif /* MLX5_ABI_USER_H */
>
> This uapi thing still needs to be fixed as I pointed out before.
In V3 we can go with below, no change in memory layout but it can
clarify the code/usage.
struct mlx5_ib_flow_counters_desc {
__u32 description;
__u32 index;
};
struct mlx5_ib_flow_counters_data {
RDMA_UAPI_PTR(struct mlx5_ib_flow_counters_desc *, counters_data);
__u32 ncounters;
__u32 reserved;
};
struct mlx5_ib_create_flow {
__u32 ncounters_data;
__u32 reserved;
/* Following are counters data based on ncounters_data */
struct mlx5_ib_flow_counters_data data[];
> I still can't figure out why this should be a 2d array.
This comes to support the future case of multiple counters objects/specs
passed with the same flow. There is a need to differentiate mapping data
for each counters object and that is done via the 'ncounters_data' field
and the 2d array.
I think it
> should be written simply as:
>
> struct mlx5_ib_flow_counter_desc {
> __u32 description;
> __u32 index;
> };
>
> struct mlx5_ib_create_flow {
> RDMA_UAPI_PTR(struct mlx5_ib_flow_counter_desc, counters_data);
> __u32 ncounters;
> __u32 reserved;
> };
>
> With the corresponding changes elsewhere.
>
This doesn't support the above use case.
> A flex array at the end of a struct means that the struct can never be
> extended again which seems like a terrible idea,
The header [1] has a fixed size and will always exist even if there will
be no counters. Future extensions [2] will be added in the memory post
the flex array which its size depends on 'ncounters_data'. This pattern
is used also in other extended APIs. [3]
struct mlx5_ib_create_flow {
__u32 ncounters_data;
__u32 reserved;
[1] /* Header is above ********
/* Following are counters data based on ncounters_data */
struct mlx5_ib_flow_counters_data data[];
[2] Future fields.
[3]
https://elixir.bootlin.com/linux/latest/source/include/uapi/rdma/ib_user_verbs.h#L1145
Powered by blists - more mailing lists