Re: [PATCH v2 2/6] perf test: Handle metric reuse in pmu-events parsing test

[John Garry]

On 01/04/2021 14:49, Jiri Olsa wrote:
> On Thu, Mar 25, 2021 at 06:33:14PM +0800, John Garry wrote:
>
> SNIP
>
> > +struct metric {
> > +	struct list_head list;
> > +	struct metric_ref metric_ref;
> > +};
> > +
> > +static int resolve_metric_simple(struct expr_parse_ctx *pctx,
> > +				 struct list_head *compound_list,
> > +				 struct pmu_events_map *map,
> > +				 const char *metric_name)
> > +{
> > +	struct hashmap_entry *cur, *cur_tmp;
> > +	struct metric *metric, *tmp;
> > +	size_t bkt;
> > +	bool all;
> > +	int rc;
> > +
> > +	do {
> > +		all = true;
> > +		hashmap__for_each_entry_safe((&pctx->ids), cur, cur_tmp, bkt) {
> > +			struct metric_ref *ref;
> > +			struct pmu_event *pe;
> > +
> > +			pe = metrcgroup_find_metric(cur->key, map);

*

> > +			if (!pe)
> > +				continue;
> > +
> > +			if (!strcmp(metric_name, (char *)cur->key)) {
> > +				pr_warning("Recursion detected for metric %s\n", metric_name);
> > +				rc = -1;
> > +				goto out_err;
> > +			}
> > +
> > +			all = false;
> > +
> > +			/* The metric key itself needs to go out.. */
> > +			expr__del_id(pctx, cur->key);
> > +
> > +			metric = malloc(sizeof(*metric));
> > +			if (!metric) {
> > +				rc = -ENOMEM;
> > +				goto out_err;
> > +			}
> > +
> > +			ref = &metric->metric_ref;
> > +			ref->metric_name = pe->metric_name;
> > +			ref->metric_expr = pe->metric_expr;
> > +			list_add_tail(&metric->list, compound_list);
> > +
> > +			rc = expr__find_other(pe->metric_expr, NULL, pctx, 0);

Hi Jirka,

> so this might add new items to pctx->ids, I think you need
> to restart the iteration as we do it in __resolve_metric
> otherwise you could miss some new keys

I thought that I was doing this. Indeed, this code is very much like 
__resolve_metric() ;)

So expr__find_other() may add a new item to pctx->ids, and we always 
iterate again, and try to lookup any pmu_events, *, above. If none 
exist, then we have broken down pctx into primitive events aliases and 
unresolvable metrics, and stop iterating. And then unresolvable metrics 
would be found in check_parse_cpu().

As an example, we can deal with metric test1, below, which references 2x 
other metrics:

    {
        "MetricExpr": "IDQ_UOPS_NOT_DELIVERED.CORE / (4 * (( ( 
CPU_CLK_UNHALTED.THREAD / 2 ) * ( 1 + CPU_CLK_UNHALTED.ONE_THREAD_ACTIVE 
/ CPU_CLK_UNHALTED.REF_XCLK ) )))",
      "MetricName": "Frontend_Bound",
    },
    {
        "MetricExpr": "( UOPS_ISSUED.ANY - UOPS_RETIRED.RETIRE_SLOTS + 
4 * INT_MISC.RECOVERY_CYCLES ) / (4 * cycles)",
        "MetricName": "Bad_Speculation",
    },
    {
        "MetricExpr": "Bad_Speculation + Frontend_Bound",
        "MetricName": "test1",
    },

Does that satisfy your concern, or have I missed something?

Thanks,
John

> jirka
>
> > +			if (rc)
> > +				goto out_err;
> > +		}
> > +	} while (!all);
> > +
> > +	return 0;
> > +
> > +out_err:
> > +	list_for_each_entry_safe(metric, tmp, compound_list, list)
> > +		free(metric);
> > +
> > +	return rc;
> > +
> > +}
>
> SNIP
>
> .