[dpdk-dev] [PATCH v2 4/5] net/mlx5: add mlx5 APIs for single flow dump feature
Slava Ovsiienko
viacheslavo at nvidia.com
Tue Apr 13 09:39:54 CEST 2021
> -----Original Message-----
> From: Haifei Luo
> Sent: Tuesday, April 13, 2021 9:29 AM
> To: Slava Ovsiienko <viacheslavo at nvidia.com>; dev at dpdk.org
> Cc: Ori Kam <orika at nvidia.com>; Raslan Darawsheh <rasland at nvidia.com>;
> Xueming(Steven) Li <xuemingl at nvidia.com>; Matan Azrad
> <matan at nvidia.com>; Shahaf Shuler <shahafs at nvidia.com>
> Subject: RE: [PATCH v2 4/5] net/mlx5: add mlx5 APIs for single flow dump
> feature
>
> Hi Slava,
> For #1, The steering tool send messages to DPDK to request dump.
> Server/Client use data structure "struct msghdr"
> to communicate. It has " msg_iov " ," msg_iovlen" and etc.
> In the tool side, Msg_iov is constructed as 1 byte for port_id, 8 bytes for
> flowptr. In DPDK, then we parse the message this way.
Yes, it is clear. In my opinion we should not use byte array and parse
we should present the some structure instead:
struct mlx5_flow_dump_req {
uint8_t port_id;
void *flow_id;
} __rte_packed;
BTW, why port_id is 1 byte? port_id in DPDK is 16-bit value.
If the request dump tool uses somr structure - it should be defined in some
common file. IMO, it is not a good practice to rely on raw byte array layout.
With best regards, Slava
> For #2, I will move them to the beginning.
>
> -----Original Message-----
> From: Slava Ovsiienko <viacheslavo at nvidia.com>
> Sent: Monday, April 12, 2021 3:38 PM
> To: Haifei Luo <haifeil at nvidia.com>; dev at dpdk.org
> Cc: Ori Kam <orika at nvidia.com>; Raslan Darawsheh <rasland at nvidia.com>;
> Xueming(Steven) Li <xuemingl at nvidia.com>; Haifei Luo
> <haifeil at nvidia.com>; Matan Azrad <matan at nvidia.com>; Shahaf Shuler
> <shahafs at nvidia.com>
> Subject: RE: [PATCH v2 4/5] net/mlx5: add mlx5 APIs for single flow dump
> feature
>
> > -----Original Message-----
> > From: Haifei Luo <haifeil at nvidia.com>
> > Sent: Wednesday, April 7, 2021 9:09
> > To: dev at dpdk.org
> > Cc: Ori Kam <orika at nvidia.com>; Slava Ovsiienko
> > <viacheslavo at nvidia.com>; Raslan Darawsheh <rasland at nvidia.com>;
> > Xueming(Steven) Li <xuemingl at nvidia.com>; Haifei Luo
> > <haifeil at nvidia.com>; Matan Azrad <matan at nvidia.com>; Shahaf Shuler
> > <shahafs at nvidia.com>
> > Subject: [PATCH v2 4/5] net/mlx5: add mlx5 APIs for single flow dump
> > feature
> >
> > Modify API mlx5_flow_dev_dump to support the feature.
> > Modify mlx5_socket since one extra arg flow_ptr is added.
> >
> > Signed-off-by: Haifei Luo <haifeil at nvidia.com>
>
> Sorry, this patch is errorneously acked instead of the
> "common/mlx5: add mlx5 APIs for single flow dump feature"
>
> I have comment for this one.
>
> > +#ifndef _GNU_SOURCE
> > +#define _GNU_SOURCE
> > +#endif
> > +
> > #include <sys/types.h>
> > #include <sys/socket.h>
> > #include <sys/un.h>
> > @@ -29,11 +33,15 @@
> > {
> > int conn_sock;
> > int ret;
> > + int j;
> > struct cmsghdr *cmsg = NULL;
> > - int data;
> > + #define LENGTH 9
> > + /* The first byte for port_id and the rest for flowptr. */
> > + int data[LENGTH];
>
> So, we define 36/72 bytes array? And then use each int as byte to save
> flow_idx value?
> I suppose the correct way would be to define the structure of message in
> stead of using ints array, something likle this:
>
> struct mlx5_ipc_msg {
> int status;
> void* flow_idx;
> }
>
> > + /* The first byte in data for port_id and the following 8 for flowptr */
> > + for (j = 1; j < LENGTH; j++)
> > + flow_ptr = (flow_ptr << 8) + data[j];
> If structure is define, there should be:
> flow_ptr = msg->flow_idx
>
> > + if (flow_ptr == 0)
> > + ret = mlx5_flow_dev_dump(dev, NULL, file, NULL);
> > + else
> > + ret = mlx5_flow_dev_dump(dev,
> > + (struct rte_flow *)((uintptr_t)flow_ptr), file, &err);
> > +
>
> > + /*dump one*/
> > + uint32_t handle_idx;
> > + int ret;
> > + struct mlx5_flow_handle *dh;
> > + struct rte_flow *flow = mlx5_ipool_get(priv->sh->ipool
> > + [MLX5_IPOOL_RTE_FLOW], (uintptr_t)(void *)flow_idx);
> > +
> Please, move variable declarations to the routine beginning, to others
>
> With best regards, Slava
More information about the dev
mailing list