[PATCH] trace: take live traces via telemetry

[Bruce Richardson]

On Thu, Oct 13, 2022 at 09:49:28AM +0200, David Marchand wrote:
> Register telemetry commands to list and configure trace points and later
> save traces for a running DPDK application.
> 
> Note: trace point names contain a '.', so the list of valid characters
> used in telemetry commands and dictionary keys has been extended.
> 
> Example with testpmd running with two net/null ports (startup command
> from devtools/test-null.sh):
> 
> --> /trace/list,lib.ethdev.*
> {"/trace/list": {"lib.ethdev.configure": "Disabled",
>     "lib.ethdev.rxq.setup": "Disabled",
>     "lib.ethdev.txq.setup": "Disabled",
>     "lib.ethdev.start": "Disabled",
>     "lib.ethdev.stop": "Disabled",
>     "lib.ethdev.close": "Disabled",
>     "lib.ethdev.rx.burst": "Disabled",
>     "lib.ethdev.tx.burst": "Disabled"}}
> 
> --> /trace/enable,lib.ethdev.st*
> {"/trace/enable": {"Count": 2}}
> --> /trace/enable,lib.ethdev.st*
> {"/trace/enable": {"Count": 0}}
> 
> --> /trace/list,lib.ethdev.*
> {"/trace/list": {"lib.ethdev.configure": "Disabled",
>     "lib.ethdev.rxq.setup": "Disabled",
>     "lib.ethdev.txq.setup": "Disabled",
>     "lib.ethdev.start": "Enabled",
>     "lib.ethdev.stop": "Enabled",
>     "lib.ethdev.close": "Disabled",
>     "lib.ethdev.rx.burst": "Disabled",
>     "lib.ethdev.tx.burst": "Disabled"}}
> 
> testpmd> stop
> ...
> testpmd> port stop all
> ...
> testpmd> port start all
> ...
> testpmd> start
> ...
> 
> --> /trace/save
> {"/trace/save": {"Status": "OK",
>     "Path": ".../dpdk-traces/rte-2022-10-12-AM-10-51-48"}}
> 
> $ babeltrace .../dpdk-traces/rte-2022-10-12-AM-10-51-48
> [10:51:36.229878723] (+?.?????????) lib.ethdev.stop:
>     { cpu_id = 0x0, name = "dpdk-testpmd" }, { port_id = 0x0, ret = 0 }
> [10:51:36.229880251] (+0.000001528) lib.ethdev.stop:
>     { cpu_id = 0x0, name = "dpdk-testpmd" }, { port_id = 0x1, ret = 0 }
> [10:51:40.449359774] (+4.219479523) lib.ethdev.start:
>     { cpu_id = 0x0, name = "dpdk-testpmd" }, { port_id = 0x0 }
> [10:51:40.449377877] (+0.000018103) lib.ethdev.start:
>     { cpu_id = 0x0, name = "dpdk-testpmd" }, { port_id = 0x1 }
> 
> --> /trace/disable,*
> {"/trace/disable": {"Count": 2}}
> 
> Signed-off-by: David Marchand <david.marchand at redhat.com>
> ---
> For runtime testing, please use this patch in addition with series 25183.
> Depends-on: series-25183 ("Trace subsystem fixes")
> 
> ---
>  lib/eal/common/eal_common_trace.c | 78 +++++++++++++++++++++++++++++++
>  lib/telemetry/telemetry_data.c    |  1 +
>  2 files changed, 79 insertions(+)
> 
> diff --git a/lib/eal/common/eal_common_trace.c b/lib/eal/common/eal_common_trace.c
> index f9b187d15f..9a54987b42 100644
> --- a/lib/eal/common/eal_common_trace.c
> +++ b/lib/eal/common/eal_common_trace.c
> @@ -12,6 +12,7 @@
>  #include <rte_lcore.h>
>  #include <rte_per_lcore.h>
>  #include <rte_string_fns.h>
> +#include <rte_telemetry.h>
>  
>  #include "eal_trace.h"
>  
> @@ -530,3 +531,80 @@ __rte_trace_point_register(rte_trace_point_t *handle, const char *name,
>  
>  	return -rte_errno;
>  }
> +
> +static int
> +trace_telemetry_enable(const char *cmd __rte_unused,
> +	const char *params, struct rte_tel_data *d)
> +{
> +	unsigned int count;
> +
> +	if (params == NULL || strlen(params) == 0)
> +		return -1;
> +	rte_tel_data_start_dict(d);
> +	count = __atomic_load_n(&trace.status, __ATOMIC_RELAXED);
> +	rte_trace_pattern(params, true);
> +	rte_tel_data_add_dict_int(d, "Count",
> +		__atomic_load_n(&trace.status, __ATOMIC_RELAXED) - count);
> +	return 0;
> +
> +}
> +
> +static int
> +trace_telemetry_disable(const char *cmd __rte_unused,
> +	const char *params, struct rte_tel_data *d)
> +{
> +	unsigned int count;
> +
> +	if (params == NULL || strlen(params) == 0)
> +		return -1;
> +	rte_tel_data_start_dict(d);
> +	count = __atomic_load_n(&trace.status, __ATOMIC_RELAXED);
> +	rte_trace_pattern(params, false);
> +	rte_tel_data_add_dict_int(d, "Count",
> +		count - __atomic_load_n(&trace.status, __ATOMIC_RELAXED));
> +	return 0;
> +
> +}
> +
> +static int
> +trace_telemetry_list(const char *cmd __rte_unused,
> +	const char *params, struct rte_tel_data *d)
> +{
> +	struct trace_point *tp;
> +
> +	rte_tel_data_start_dict(d);
> +	STAILQ_FOREACH(tp, &tp_list, next) {
> +		if (params != NULL && fnmatch(params, tp->name, 0) != 0)
> +			continue;
> +
> +		rte_tel_data_add_dict_string(d, tp->name,
> +			rte_trace_point_is_enabled(tp->handle) ?  "Enabled" : "Disabled");
> +	}
> +
> +	return 0;
> +}
> +
> +static int
> +trace_telemetry_save(const char *cmd __rte_unused,
> +	const char *params __rte_unused, struct rte_tel_data *d)
> +{
> +	struct trace *trace = trace_obj_get();
> +
> +	rte_tel_data_start_dict(d);
> +	rte_tel_data_add_dict_string(d, "Status", rte_trace_save() == 0 ? "OK" : "KO");
> +	rte_tel_data_add_dict_string(d, "Path", trace->dir);
> +
> +	return 0;
> +}
> +
> +RTE_INIT(trace_telemetry)
> +{
> +	rte_telemetry_register_cmd("/trace/enable", trace_telemetry_enable,
> +		"Enable trace points matching the provided pattern. Parameters: string pattern.");
> +	rte_telemetry_register_cmd("/trace/disable", trace_telemetry_disable,
> +		"Disable trace points matching the provided pattern. Parameters: string pattern.");
> +	rte_telemetry_register_cmd("/trace/list", trace_telemetry_list,
> +		"List trace points. Parameters: string pattern.");
> +	rte_telemetry_register_cmd("/trace/save", trace_telemetry_save,
> +		"Save current traces. Takes no parameter.");
> +}
> diff --git a/lib/telemetry/telemetry_data.c b/lib/telemetry/telemetry_data.c
> index 34366ecee3..5b319c18fb 100644
> --- a/lib/telemetry/telemetry_data.c
> +++ b/lib/telemetry/telemetry_data.c
> @@ -106,6 +106,7 @@ valid_name(const char *name)
>  			['a' ... 'z'] = 1,
>  			['_'] = 1,
>  			['/'] = 1,
> +			['.'] = 1,
>  	};
>  	while (*name != '\0') {
>  		if ((size_t)*name >= RTE_DIM(allowed) || allowed[(int)*name] == 0)

I don't see an issue with allowing "." characters in dictionary names, so
for this part:

Acked-by: Bruce Richardson <bruce.richardson at intel.com>