2021-12-23 18:56:42

by Ian Rogers

Subject: [PATCH v3] perf test: Make metric testing more robust.

When testing metric expressions we fake counter values from 1 going
upward. For some metrics this can yield negative values that are clipped
to zero, and then cause divide by zero failures. Such clipping is
questionable but may be a result of tools automatically generating
metrics. A workaround for this case is to try a second time with counter
values going in the opposite direction.

This case was seen in a metric like:
event1 / max(event2 - event3, 0)
But it may also happen in more sensible metrics like:
event1 / (event2 + event3 - 1 - event4)

v2. Rebase and more detail in commit message.
v3. Is a rebase.

Signed-off-by: Ian Rogers <[email protected]>
---
tools/perf/tests/pmu-events.c | 32 ++++++++++++++++++++++++++------
1 file changed, 26 insertions(+), 6 deletions(-)

diff --git a/tools/perf/tests/pmu-events.c b/tools/perf/tests/pmu-events.c
index df1c9a3cc05b..b2ddf928d32a 100644
--- a/tools/perf/tests/pmu-events.c
+++ b/tools/perf/tests/pmu-events.c
@@ -962,8 +962,18 @@ static int test__parsing(struct test_suite *test __maybe_unused,
}

if (expr__parse(&result, ctx, pe->metric_expr)) {
- expr_failure("Parse failed", map, pe);
- ret++;
+ /*
+ * Parsing failed, make numbers go from large to
+ * small which can resolve divide by zero
+ * issues.
+ */
+ k = 1024;
+ hashmap__for_each_entry(ctx->ids, cur, bkt)
+ expr__add_id_val(ctx, strdup(cur->key), k--);
+ if (expr__parse(&result, ctx, pe->metric_expr)) {
+ expr_failure("Parse failed", map, pe);
+ ret++;
+ }
}
}
}
@@ -1022,10 +1032,20 @@ static int metric_parse_fake(const char *str)
}
}

- if (expr__parse(&result, ctx, str))
- pr_err("expr__parse failed\n");
- else
- ret = 0;
+ ret = 0;
+ if (expr__parse(&result, ctx, str)) {
+ /*
+ * Parsing failed, make numbers go from large to small which can
+ * resolve divide by zero issues.
+ */
+ i = 1024;
+ hashmap__for_each_entry(ctx->ids, cur, bkt)
+ expr__add_id_val(ctx, strdup(cur->key), i--);
+ if (expr__parse(&result, ctx, str)) {
+ pr_err("expr__parse failed\n");
+ ret = -1;
+ }
+ }

out:
expr__ctx_free(ctx);
--
2.34.1.307.g9b7440fafd-goog

2021-12-24 08:49:48

by John Garry

Subject: Re: [PATCH v3] perf test: Make metric testing more robust.

On 23/12/2021 18:56, Ian Rogers wrote:
> When testing metric expressions we fake counter values from 1 going
> upward. For some metrics this can yield negative values that are clipped
> to zero, and then cause divide by zero failures. Such clipping is
> questionable but may be a result of tools automatically generating
> metrics. A workaround for this case is to try a second time with counter
> values going in the opposite direction.
>
> This case was seen in a metric like:
> event1 / max(event2 - event3, 0)
> But it may also happen in more sensible metrics like:
> event1 / (event2 + event3 - 1 - event4)
>
> v2. Rebase and more detail in commit message.
> v3. Is a rebase.

Incorrect location for this info

>
> Signed-off-by: Ian Rogers <[email protected]>

Reviewed-by: John Garry <[email protected]>

> ---
> tools/perf/tests/pmu-events.c | 32 ++++++++++++++++++++++++++------
> 1 file changed, 26 insertions(+), 6 deletions(-)
>
> diff --git a/tools/perf/tests/pmu-events.c b/tools/perf/tests/pmu-events.c
> index df1c9a3cc05b..b2ddf928d32a 100644
> --- a/tools/perf/tests/pmu-events.c
> +++ b/tools/perf/tests/pmu-events.c
> @@ -962,8 +962,18 @@ static int test__parsing(struct test_suite *test __maybe_unused,
> }
>
> if (expr__parse(&result, ctx, pe->metric_expr)) {
> - expr_failure("Parse failed", map, pe);
> - ret++;
> + /*
> + * Parsing failed, make numbers go from large to
> + * small which can resolve divide by zero

can or may?

> + * issues.
> + */
> + k = 1024;
> + hashmap__for_each_entry(ctx->ids, cur, bkt)
> + expr__add_id_val(ctx, strdup(cur->key), k--);
> + if (expr__parse(&result, ctx, pe->metric_expr)) {
> + expr_failure("Parse failed", map, pe);
> + ret++;
> + }
> }
> }
> }
> @@ -1022,10 +1032,20 @@ static int metric_parse_fake(const char *str)
> }
> }
>
> - if (expr__parse(&result, ctx, str))
> - pr_err("expr__parse failed\n");
> - else
> - ret = 0;
> + ret = 0;
> + if (expr__parse(&result, ctx, str)) {
> + /*
> + * Parsing failed, make numbers go from large to small which can
> + * resolve divide by zero issues.
> + */
> + i = 1024;
> + hashmap__for_each_entry(ctx->ids, cur, bkt)
> + expr__add_id_val(ctx, strdup(cur->key), i--);
> + if (expr__parse(&result, ctx, str)) {
> + pr_err("expr__parse failed\n");
> + ret = -1;
> + }
> + }
>
> out:
> expr__ctx_free(ctx);
>

2022-02-15 15:46:15

by Arnaldo Carvalho de Melo

Subject: Re: [PATCH v3] perf test: Make metric testing more robust.

Em Fri, Dec 24, 2021 at 08:49:35AM +0000, John Garry escreveu:
> On 23/12/2021 18:56, Ian Rogers wrote:
> > When testing metric expressions we fake counter values from 1 going
> > upward. For some metrics this can yield negative values that are clipped
> > to zero, and then cause divide by zero failures. Such clipping is
> > questionable but may be a result of tools automatically generating
> > metrics. A workaround for this case is to try a second time with counter
> > values going in the opposite direction.
> >
> > This case was seen in a metric like:
> > event1 / max(event2 - event3, 0)
> > But it may also happen in more sensible metrics like:
> > event1 / (event2 + event3 - 1 - event4)
> >
> > v2. Rebase and more detail in commit message.
> > v3. Is a rebase.
>
> Incorrect location for this info
>
> >
> > Signed-off-by: Ian Rogers <[email protected]>
>
> Reviewed-by: John Garry <[email protected]>

Found thru the cracks, applying.

- Arnaldo

> > ---
> > tools/perf/tests/pmu-events.c | 32 ++++++++++++++++++++++++++------
> > 1 file changed, 26 insertions(+), 6 deletions(-)
> >
> > diff --git a/tools/perf/tests/pmu-events.c b/tools/perf/tests/pmu-events.c
> > index df1c9a3cc05b..b2ddf928d32a 100644
> > --- a/tools/perf/tests/pmu-events.c
> > +++ b/tools/perf/tests/pmu-events.c
> > @@ -962,8 +962,18 @@ static int test__parsing(struct test_suite *test __maybe_unused,
> > }
> > if (expr__parse(&result, ctx, pe->metric_expr)) {
> > - expr_failure("Parse failed", map, pe);
> > - ret++;
> > + /*
> > + * Parsing failed, make numbers go from large to
> > + * small which can resolve divide by zero
>
> can or may?
>
> > + * issues.
> > + */
> > + k = 1024;
> > + hashmap__for_each_entry(ctx->ids, cur, bkt)
> > + expr__add_id_val(ctx, strdup(cur->key), k--);
> > + if (expr__parse(&result, ctx, pe->metric_expr)) {
> > + expr_failure("Parse failed", map, pe);
> > + ret++;
> > + }
> > }
> > }
> > }
> > @@ -1022,10 +1032,20 @@ static int metric_parse_fake(const char *str)
> > }
> > }
> > - if (expr__parse(&result, ctx, str))
> > - pr_err("expr__parse failed\n");
> > - else
> > - ret = 0;
> > + ret = 0;
> > + if (expr__parse(&result, ctx, str)) {
> > + /*
> > + * Parsing failed, make numbers go from large to small which can
> > + * resolve divide by zero issues.
> > + */
> > + i = 1024;
> > + hashmap__for_each_entry(ctx->ids, cur, bkt)
> > + expr__add_id_val(ctx, strdup(cur->key), i--);
> > + if (expr__parse(&result, ctx, str)) {
> > + pr_err("expr__parse failed\n");
> > + ret = -1;
> > + }
> > + }
> > out:
> > expr__ctx_free(ctx);
> >

--

- Arnaldo