> -----Original Message-----
> From: Ferruh Yigit <ferruh.yi...@amd.com>
> Sent: Tuesday, October 10, 2023 6:49 PM
> To: Zhang, Qi Z <qi.z.zh...@intel.com>; Singh, Aman Deep
> <aman.deep.si...@intel.com>; Zhang, Yuying <yuying.zh...@intel.com>
> Cc: dev@dpdk.org; Dumitrescu, Cristian <cristian.dumitre...@intel.com>;
> or...@nvidia.com
> Subject: Re: [PATCH v4] app/testpmd: enable cli for programmable action
>
> On 10/7/2023 11:47 AM, Qi Zhang wrote:
> > Parsing command line for rte_flow_action_prog.
> >
> > Syntax:
> >
> > "prog name <name> [arguments <arg_name_0> <arg_value_0> \
> > <arg_name_1> <arg_value1> ... end]"
> >
>
> Can you please put full rte flow command in the commit log? Like what is the
> 'pattern' for above command?
The pattern part should be independent of the action part,
though for our P4 device, we will prefer use rte_flow_flex_item, something like:
flow create 0 pattern flex item is xxx pattern is xxx / flex item is xxx
pattern is / actions prog name ......
but it does not limit PMD to support flow like below
flow create 0 pattern eth / ipv4 src is 1.1.1.1 / actions prog name ......
So I think it may not be necessary to highlight the pattern format here.
>
>
> > Use parse_string0 to parse name string.
> > Use parse_hex to parse hex string.
> > Use struct action_prog_data to store parsed result.
> >
> > Example:
> >
> > Action with 2 arguments:
> >
> > "prog name action0 arguments field0 03FF field1 55AA end"
> >
> > Action without argument:
> >
> > "prog name action1"
> >
> > Signed-off-by: Qi Zhang <qi.z.zh...@intel.com>
> >
>
> Is there an existing driver implementation, checking it helps to understand
> feature implementation?
This work is still ongoing, currently we target to upstream on DPDK 24.03
>
>
> > ---
> >
> > v4:
> > - be more generous on the max size of name and value.
> >
> > v3:
> > - refine struct action_prog_data
> > - enlarge the max size
> >
> > v2:
> > - fix title
> > - minor coding style refine.
> >
> > app/test-pmd/cmdline_flow.c | 232
> > ++++++++++++++++++++++++++++++++++++
> > 1 file changed, 232 insertions(+)
> >
>
> Hi Qi,
>
> Can you please update documentation too,
> `doc/guides/testpmd_app_ug/testpmd_funcs.rst`, `Flow rules management`
> section.
Sure.
>
>
> > diff --git a/app/test-pmd/cmdline_flow.c b/app/test-pmd/cmdline_flow.c
> > index 21828c144c..ae5556e704 100644
> > --- a/app/test-pmd/cmdline_flow.c
> > +++ b/app/test-pmd/cmdline_flow.c
> > @@ -719,6 +719,13 @@ enum index {
> > ACTION_IPV6_EXT_PUSH,
> > ACTION_IPV6_EXT_PUSH_INDEX,
> > ACTION_IPV6_EXT_PUSH_INDEX_VALUE,
> > + ACTION_PROG,
> > + ACTION_PROG_NAME,
> > + ACTION_PROG_NAME_STRING,
> > + ACTION_PROG_ARGUMENTS,
> > + ACTION_PROG_ARG_NAME,
> > + ACTION_PROG_ARG_VALUE,
> > + ACTION_PROG_ARG_END,
> > };
> >
> > /** Maximum size for pattern in struct rte_flow_item_raw. */ @@
> > -749,6 +756,23 @@ struct action_rss_data {
> > uint16_t queue[ACTION_RSS_QUEUE_NUM]; };
> >
> > +#define ACTION_PROG_NAME_SIZE_MAX 256 #define
> ACTION_PROG_ARG_NUM_MAX
> > +16 #define ACTION_PROG_ARG_VALUE_SIZE_MAX 64
> > +
> > +/** Storage for struct rte_flow_action_prog including external data.
> > +*/ struct action_prog_data {
> > + struct rte_flow_action_prog conf;
> > + struct {
> > + char name[ACTION_PROG_NAME_SIZE_MAX];
> > + struct rte_flow_action_prog_argument
> args[ACTION_PROG_ARG_NUM_MAX];
> > + struct {
> > + char names[ACTION_PROG_NAME_SIZE_MAX];
> > + uint8_t
> value[ACTION_PROG_ARG_VALUE_SIZE_MAX];
> > + } arg_data[ACTION_PROG_ARG_NUM_MAX];
> > + } data;
> > +};
> > +
> > /** Maximum data size in struct rte_flow_action_raw_encap. */
> > #define ACTION_RAW_ENCAP_MAX_DATA 512 #define
> RAW_ENCAP_CONFS_MAX_NUM
> > 8 @@ -2169,6 +2193,7 @@ static const enum index next_action[] = {
> > ACTION_QUOTA_QU,
> > ACTION_IPV6_EXT_REMOVE,
> > ACTION_IPV6_EXT_PUSH,
> > + ACTION_PROG,
> > ZERO,
> > };
> >
> > @@ -2510,6 +2535,13 @@ static const enum index
> action_represented_port[] = {
> > ZERO,
> > };
> >
> > +static const enum index action_prog[] = {
> > + ACTION_PROG_NAME,
> > + ACTION_PROG_ARGUMENTS,
> > + ACTION_NEXT,
> > + ZERO,
> > +};
> > +
> > static int parse_set_raw_encap_decap(struct context *, const struct token
> > *,
> > const char *, unsigned int,
> > void *, unsigned int);
> > @@ -2786,6 +2818,18 @@ static int
> > parse_qu_mode_name(struct context *ctx, const struct token *token,
> > const char *str, unsigned int len, void *buf,
> > unsigned int size);
> > +static int
> > +parse_vc_action_prog(struct context *, const struct token *,
> > + const char *, unsigned int, void *,
> > + unsigned int);
> > +static int
> > +parse_vc_action_prog_arg_name(struct context *, const struct token *,
> > + const char *, unsigned int, void *,
> > + unsigned int);
> > +static int
> > +parse_vc_action_prog_arg_value(struct context *, const struct token *,
> > + const char *, unsigned int, void *,
> > + unsigned int);
> > static int comp_none(struct context *, const struct token *,
> > unsigned int, char *, unsigned int); static int
> > comp_boolean(struct context *, const struct token *, @@ -7518,6
> > +7562,48 @@ static const struct token token_list[] = {
> > .args = ARGS(ARGS_ENTRY(struct rte_flow_item_tx_queue,
> > tx_queue)),
> > },
> > + [ACTION_PROG] = {
> > + .name = "prog",
> > + .help = "match a programmable action",
> > + .priv = PRIV_ACTION(PROG, sizeof(struct action_prog_data)),
> > + .next = NEXT(action_prog),
> > + .call = parse_vc_action_prog,
> > + },
> > + [ACTION_PROG_NAME] = {
> > + .name = "name",
> > + .help = "programble action name",
> >
>
> Can you please remind me again what was the 'name' filed of "struct
> rte_flow_action_prog" was for?
The 'name' field serves as a means for the driver to identify an action schema,
enabling it to verify if the number of parameters and the size of each
parameter value align with the P4 definition.
Subsequently, the driver translates these values into hardware-specific
configurations. If there is a misalignment, the PMD will return a failure.
>
>
> > + .next = NEXT(action_prog,
> NEXT_ENTRY(ACTION_PROG_NAME_STRING)),
> > + .args = ARGS(ARGS_ENTRY(struct action_prog_data,
> data.name)),
> > + },
> > + [ACTION_PROG_NAME_STRING] = {
> > + .name = "{string}",
> > + .type = "STRING",
> > + .help = "programmable action name string",
> > + .call = parse_string0,
> > + },
> > + [ACTION_PROG_ARGUMENTS] = {
> > + .name = "arguments",
> > + .help = "programmable action name",
> > + .next = NEXT(action_prog,
> NEXT_ENTRY(ACTION_PROG_ARG_NAME)),
> > + .call = parse_vc_conf,
> > + },
> > + [ACTION_PROG_ARG_NAME] = {
> > + .name = "{string}",
> > + .help = "programmable action argument name",
> > + .next = NEXT(NEXT_ENTRY(ACTION_PROG_ARG_VALUE)),
> > + .call = parse_vc_action_prog_arg_name,
> > + },
> > + [ACTION_PROG_ARG_VALUE] = {
> > + .name = "{hex}",
> > + .help = "programmable action argument value",
> > + .next = NEXT(NEXT_ENTRY(ACTION_PROG_ARG_END,
> ACTION_PROG_ARG_NAME)),
> > + .call = parse_vc_action_prog_arg_value,
> > + },
> > + [ACTION_PROG_ARG_END] = {
> > + .name = "end",
> > + .help = "end of the programmable action arguments",
> > + },
> > +
> >
>
> Does this means two 'end' required if multiple args provided, like:
> prog name "name" arguments field0 03FF field1 1 end / end
> I am aware there is variable length of key/value, and need a marker to stop,
> but this end specific for action is not used,
Actually I borrowed the idea from the queue group in the RSS action:
'actions rss queues 0 1 2 3 end / end ..."
The difference is that the 'end' check was previously hidden within the
'parse_vc_action_rss_queue' function, while in my implementation, I've defined
it as a distinct state.
Thanks
Qi
>I wonder if that is because there is
> a better way to do it, @Ori may comment better.
>
>
