RE: [PATCH RFC iproute-next 3/5] rdma: Add CQ resource tracking information

["Steve Wise" <swise@opengridcomputing.com>]

> 
> On Wed, Feb 14, 2018 at 01:07:01PM -0800, Steve Wise wrote:
> > Sample output:
> >
> > # rdma resource show cq
> > link cxgb4_0/- cqe 46 usecnt 2 pid 30503 comm rping
> > link cxgb4_0/- cqe 46 usecnt 2 pid 30498 comm rping
> > link mlx4_0/- cqe 63 usecnt 2 pid 30494 comm rping
> > link mlx4_0/- cqe 63 usecnt 2 pid 30489 comm rping
> > link mlx4_0/- cqe 1023 usecnt 2 poll_ctx WORKQUEUE pid 0 comm [ib_core]
> >
> > # rdma resource show cq pid 30489
> > link mlx4_0/- cqe 63 usecnt 2 pid 30489 comm rping
> >
> > Signed-off-by: Steve Wise <swise@opengridcomputing.com>
> > ---
> >  rdma/res.c   | 123
> +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
> >  rdma/utils.c |   5 +++
> >  2 files changed, 128 insertions(+)
> >
> > diff --git a/rdma/res.c b/rdma/res.c
> > index beae7dc..27c1efd 100644
> > --- a/rdma/res.c
> > +++ b/rdma/res.c
> > @@ -21,6 +21,8 @@ static int res_help(struct rd *rd)
> >  	pr_out("          resource show qp link [DEV/PORT] [FILTER-NAME
> FILTER-VALUE]\n");
> >  	pr_out("          resource show cm_id link [DEV/PORT]\n");
> >  	pr_out("          resource show cm_id link [DEV/PORT] [FILTER-NAME
> FILTER-VALUE]\n");
> > +	pr_out("          resource show cq link [DEV/PORT]\n");
> > +	pr_out("          resource show cq link [DEV/PORT] [FILTER-NAME
> FILTER-VALUE]\n");
> >  	return 0;
> >  }
> >
> > @@ -705,6 +707,118 @@ static int res_cm_id_parse_cb(const struct
> nlmsghdr *nlh, void *data)
> >  	return MNL_CB_OK;
> >  }
> >
> > +static void print_cqe(struct rd *rd, uint32_t val)
> > +{
> > +	if (rd->json_output)
> > +		jsonw_uint_field(rd->jw, "cqe", val);
> > +	else
> > +		pr_out("cqe %u ", val);
> > +}
> > +
> > +static void print_usecnt(struct rd *rd, uint64_t val)
> > +{
> > +	if (rd->json_output)
> > +		jsonw_uint_field(rd->jw, "usecnt", val);
> > +	else
> > +		pr_out("usecnt %" PRIu64 " ", val);
> 
> Interesting, how many users are actually know what the "usecnt" actually
> means?
> Will it be more clear to call it "users" instead of "usecnt"?
> 

Users is good.

> > +}
> > +
> > +static const char *poll_ctx_to_str(uint8_t idx)
> > +{
> > +	static const char * const cm_id_states_str[] = { "DIRECT", "SOFTIRQ",
> > +						      "WORKQUEUE"};
> > +
> > +	if (idx < ARRAY_SIZE(cm_id_states_str))
> > +		return cm_id_states_str[idx];
> > +	return "UNKNOWN";
> > +}
> > +
> > +static void print_poll_ctx(struct rd *rd, uint8_t poll_ctx)
> > +{
> > +	if (rd->json_output) {
> > +		jsonw_string_field(rd->jw, "poll_ctx",
> poll_ctx_to_str(poll_ctx));
> > +		return;
> > +	}
> > +	pr_out("poll_ctx %s ", poll_ctx_to_str(poll_ctx));
> > +}
> > +
> > +static int res_cq_parse_cb(const struct nlmsghdr *nlh, void *data)
> > +{
> > +	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX] = {};
> > +	struct nlattr *nla_table, *nla_entry;
> > +	struct rd *rd = data;
> > +	const char *name;
> > +	uint32_t idx;
> > +
> > +	mnl_attr_parse(nlh, 0, rd_attr_cb, tb);
> > +	if (!tb[RDMA_NLDEV_ATTR_DEV_INDEX] ||
> > +	    !tb[RDMA_NLDEV_ATTR_DEV_NAME] ||
> > +	    !tb[RDMA_NLDEV_ATTR_RES_CQ])
> > +		return MNL_CB_ERROR;
> > +
> > +	name = mnl_attr_get_str(tb[RDMA_NLDEV_ATTR_DEV_NAME]);
> > +	idx =  mnl_attr_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
> > +	nla_table = tb[RDMA_NLDEV_ATTR_RES_CQ];
> > +
> > +	mnl_attr_for_each_nested(nla_entry, nla_table) {
> > +		struct nlattr *nla_line[RDMA_NLDEV_ATTR_MAX] = {};
> > +		char *comm = NULL;
> > +		uint32_t pid = 0;
> > +		uint8_t poll_ctx = 0;
> > +		uint64_t usecnt;
> > +		uint32_t cqe;
> > +		int err;
> > +
> > +		err = mnl_attr_parse_nested(nla_entry, rd_attr_cb, nla_line);
> > +		if (err != MNL_CB_OK)
> > +			return MNL_CB_ERROR;
> > +
> > +		if (!nla_line[RDMA_NLDEV_ATTR_RES_CQE] ||
> > +		    !nla_line[RDMA_NLDEV_ATTR_RES_USECNT] ||
> 
> I'm not sure that we will have USECNT in the future, let's not put
> requirement for RDMA_NLDEV_ATTR_RES_USECNT here.

Agreed.

> 
> > +		    (!nla_line[RDMA_NLDEV_ATTR_RES_PID] &&
> > +		     !nla_line[RDMA_NLDEV_ATTR_RES_KERN_NAME])) {
> > +			return MNL_CB_ERROR;
> > +		}
> > +
> > +		cqe =
> mnl_attr_get_u32(nla_line[RDMA_NLDEV_ATTR_RES_CQE]);
> > +		usecnt =
> mnl_attr_get_u64(nla_line[RDMA_NLDEV_ATTR_RES_USECNT]);
> > +		if (nla_line[RDMA_NLDEV_ATTR_RES_POLL_CTX])
> > +			poll_ctx =
> mnl_attr_get_u8(nla_line[RDMA_NLDEV_ATTR_RES_POLL_CTX]);
> > +
> > +		if (nla_line[RDMA_NLDEV_ATTR_RES_PID]) {
> > +			pid =
> mnl_attr_get_u32(nla_line[RDMA_NLDEV_ATTR_RES_PID]);
> > +			comm = get_task_name(pid);
> > +		}
> > +
> > +		if (rd_check_is_filtered(rd, "pid", pid))
> 
> free(comm);
> 

😊

> > +			continue;
> > +
> > +		if (nla_line[RDMA_NLDEV_ATTR_RES_KERN_NAME])
> > +			/* discard const from mnl_attr_get_str */
> > +			comm = (char
> *)mnl_attr_get_str(nla_line[RDMA_NLDEV_ATTR_RES_KERN_NAME]);
> > +
> > +		if (rd->json_output)
> > +			jsonw_start_array(rd->jw);
> > +
> > +		print_link(rd, idx, name, 0, nla_line);
> > +		print_cqe(rd, cqe);
> > +		print_usecnt(rd, usecnt);
> > +		if (nla_line[RDMA_NLDEV_ATTR_RES_POLL_CTX])
> > +			print_poll_ctx(rd, poll_ctx);
> > +		print_pid(rd, pid);
> > +		print_comm(rd, comm, nla_line);
> > +
> > +		if (nla_line[RDMA_NLDEV_ATTR_RES_PID])
> > +			free(comm);
> > +
> > +		if (rd->json_output)
> > +			jsonw_end_array(rd->jw);
> > +		else
> > +			pr_out("\n");
> > +	}
> > +	return MNL_CB_OK;
> > +}
> > +
> >  RES_FUNC(res_no_args,	RDMA_NLDEV_CMD_RES_GET,	NULL, true);
> >
> >  static const struct
> > @@ -758,12 +872,21 @@ filters
> cm_id_valid_filters[MAX_NUMBER_OF_FILTERS] = {{ .name = "link",
> >  RES_FUNC(res_cm_id,	RDMA_NLDEV_CMD_RES_CM_ID_GET,
> cm_id_valid_filters,
> >  	 false);
> >
> > +static const struct
> > +filters cq_valid_filters[MAX_NUMBER_OF_FILTERS] = {{ .name = "link",
> > +						   .is_number = false },
> > +						   { .name = "pid",
> > +						   .is_number = true }};
> 
> Can you please add filter of usecnt too? It will give us easy view on
> "over crowded" CQs.

Good idea.

> 
> > +
> > +RES_FUNC(res_cq,	RDMA_NLDEV_CMD_RES_CQ_GET, cq_valid_filters,
> true);
> > +
> >  static int res_show(struct rd *rd)
> >  {
> >  	const struct rd_cmd cmds[] = {
> >  		{ NULL,		res_no_args	},
> >  		{ "qp",		res_qp		},
> >  		{ "cm_id",	res_cm_id	},
> > +		{ "cq",		res_cq		},
> >  		{ 0 }
> >  	};
> >
> > diff --git a/rdma/utils.c b/rdma/utils.c
> > index 906ca73..11b34fe 100644
> > --- a/rdma/utils.c
> > +++ b/rdma/utils.c
> > @@ -387,6 +387,11 @@ static const enum mnl_attr_data_type
> nldev_policy[RDMA_NLDEV_ATTR_MAX] = {
> >  	[RDMA_NLDEV_ATTR_RES_DEV_TYPE]		= MNL_TYPE_U8,
> >  	[RDMA_NLDEV_ATTR_RES_TRANSPORT_TYPE]	= MNL_TYPE_U8,
> >  	[RDMA_NLDEV_ATTR_RES_NETWORK_TYPE]	= MNL_TYPE_U8,
> > +	[RDMA_NLDEV_ATTR_RES_CQ] = MNL_TYPE_NESTED,
> > +	[RDMA_NLDEV_ATTR_RES_CQ_ENTRY] = MNL_TYPE_NESTED,
> > +	[RDMA_NLDEV_ATTR_RES_CQE] = MNL_TYPE_U32,
> > +	[RDMA_NLDEV_ATTR_RES_USECNT] = MNL_TYPE_U64,
> > +	[RDMA_NLDEV_ATTR_RES_POLL_CTX] = MNL_TYPE_U8,
> >  };
> >
> >  int rd_attr_cb(const struct nlattr *attr, void *data)
> > --
> > 1.8.3.1
> >