Hi all,
The v3 of the extending histogram exprssions series. The previous versions
were posted at:
v2: https://lore.kernel.org/r/[email protected]/
v1: https://lore.kernel.org/r/[email protected]/
Patches 4 through 6 are new and adds some optimizations/improvements
suggested by Steven Rostedt.
The cover letter is copied below for convenience.
Thanks,
Kalesh
---
The frequency of the rss_stat trace event is known to be of the same
magnitude as that of the sched_switch event on Android devices. This can
cause flooding of the trace buffer with rss_stat traces leading to a
decreased trace buffer capacity and loss of data.
If it is not necessary to monitor very small changes in rss (as is the
case in Android) then the rss_stat tracepoint can be throttled to only
emit the event once there is a large enough change in the rss size.
The original patch that introduced the rss_stat tracepoint also proposed
a fixed throttling mechanism that only emits the rss_stat event
when the rss size crosses a 512KB boundary. It was concluded that more
generic support for this type of filtering/throttling was need, so that
it can be applied to any trace event. [1]
From the discussion in [1], histogram triggers seemed the most likely
candidate to support this type of throttling. For instance to achieve the
same throttling as was proposed in [1]:
(1) Create a histogram variable to save the 512KB bucket of the rss size
(2) Use the onchange handler to generate a synthetic event when the
rss size bucket changes.
The only missing pieces to support such a hist trigger are:
(1) Support for setting a hist variable to a specific value -- to set
the bucket size / granularity.
(2) Support for division arithmetic operation -- to determine the
corresponding bucket for an rss size.
This series extends histogram trigger expressions to:
(1) Allow assigning numeric literals to hist variable (eg. x=1234)
and using literals directly in expressions (eg. x=size/1234)
(2) Support division and multiplication in hist expressions.
(eg. a=$x/$y*z); and
(3) Fixes expression parsing for non-associative operators: subtraction
and division. (eg. 8-4-2 should be 2 not 6)
The rss_stat event can then be throttled using histogram triggers as
below:
# Create a synthetic event to monitor instead of the high frequency
# rss_stat event
echo 'rss_stat_throttled unsigned int mm_id; unsigned int curr;
int member; long size' >> tracing/synthetic_events
# Create a hist trigger that emits the synthetic rss_stat_throttled
# event only when the rss size crosses a 512KB boundary.
echo 'hist:keys=mm_id,member:bucket=size/0x80000:onchange($bucket)
.rss_stat_throttled(mm_id,curr,member,size)'
>> events/kmem/rss_stat/trigger
------ Test Results ------
Histograms can also be used to evaluate the effectiveness of this
throttling by noting the Total Hits on each trigger:
echo 'hist:keys=common_pid' >> events/sched/sched_switch/trigger
echo 'hist:keys=common_pid' >> events/kmem/rss_stat/trigger
echo 'hist:keys=common_pid'
>> events/synthetic/rss_stat_throttled/trigger
Allowing the above example (512KB granularity) run for 5 minutes on
an arm64 device with 5.10 kernel:
sched_switch : total hits = 147153
rss_stat : total hits = 38863
rss_stat_throttled: total hits = 2409
The synthetic rss_stat_throttled event is ~16x less frequent than the
rss_stat event when using a 512KB granularity.
The results are more pronounced when rss size is changing at a higher
rate in small increments. For instance the following results were obtained
by recording the hits on the above events for a run of Android's
lmkd_unit_test [2], which continually forks processes that map anonymous
memory until there is an oom kill:
sched_switch : total hits = 148832
rss_stat : total hits = 4754802
rss_stat_throttled: total hits = 96214
In this stress test, the synthetic rss_stat_throttled event is ~50x less
frequent than the rss_stat event when using a 512KB granularity.
[1] https://lore.kernel.org/lkml/[email protected]/
[2] https://cs.android.com/android/platform/superproject/+/master:system/memory/lmkd/tests/lmkd_test.cpp
Kalesh Singh (8):
tracing: Add support for creating hist trigger variables from literal
tracing: Add division and multiplication support for hist triggers
tracing: Fix operator precedence for hist triggers expression
tracing/histogram: Simplify handling of .sym-offset in expressions
tracing/histogram: Covert expr to const if both operands are constants
tracing/histogram: Optimize division by a power of 2
tracing/selftests: Add tests for hist trigger expression parsing
tracing/histogram: Document expression arithmetic and constants
Documentation/trace/histogram.rst | 14 +
kernel/trace/trace_events_hist.c | 400 ++++++++++++++----
.../testing/selftests/ftrace/test.d/functions | 4 +-
.../trigger/trigger-hist-expressions.tc | 72 ++++
4 files changed, 412 insertions(+), 78 deletions(-)
create mode 100644 tools/testing/selftests/ftrace/test.d/trigger/trigger-hist-expressions.tc
base-commit: ac8a6eba2a117e0fdc04da62ab568d1b7ca4c8f6
--
2.33.0.1079.g6e70778dc9-goog
Currently hist trigger expressions don't support the use of numeric
literals:
e.g. echo 'hist:keys=common_pid:x=$y-1234'
--> is not valid expression syntax
Having the ability to use numeric constants in hist triggers supports
a wider range of expressions for creating variables.
Add support for creating trace event histogram variables from numeric
literals.
e.g. echo 'hist:keys=common_pid:x=1234,y=size-1024' >> event/trigger
A negative numeric constant is created, using unary minus operator
(parentheses are required).
e.g. echo 'hist:keys=common_pid:z=-(2)' >> event/trigger
Constants can be used with division/multiplication (added in the
next patch in this series) to implement granularity filters for frequent
trace events. For instance we can limit emitting the rss_stat
trace event to when there is a 512KB cross over in the rss size:
# Create a synthetic event to monitor instead of the high frequency
# rss_stat event
echo 'rss_stat_throttled unsigned int mm_id; unsigned int curr;
int member; long size' >> tracing/synthetic_events
# Create a hist trigger that emits the synthetic rss_stat_throttled
# event only when the rss size crosses a 512KB boundary.
echo 'hist:keys=keys=mm_id,member:bucket=size/0x80000:onchange($bucket)
.rss_stat_throttled(mm_id,curr,member,size)'
>> events/kmem/rss_stat/trigger
A use case for using constants with addition/subtraction is not yet
known, but for completeness the use of constants are supported for all
operators.
Signed-off-by: Kalesh Singh <[email protected]>
Change-Id: I142121d28dc3475dfbc3a882e7b2368d833474eb
---
Changes in v3:
- Remove the limit on the number of constants that can be created,
per Steven Rostedt
Changes in v2:
- Add description of use case for constants in arithmetic
operations in commit message, per Steven Rostedt
- Add Namhyung's Reviewed-by
kernel/trace/trace_events_hist.c | 71 +++++++++++++++++++++++++++++++-
1 file changed, 70 insertions(+), 1 deletion(-)
diff --git a/kernel/trace/trace_events_hist.c b/kernel/trace/trace_events_hist.c
index f01e442716e2..28f711224944 100644
--- a/kernel/trace/trace_events_hist.c
+++ b/kernel/trace/trace_events_hist.c
@@ -66,7 +66,8 @@
C(EMPTY_SORT_FIELD, "Empty sort field"), \
C(TOO_MANY_SORT_FIELDS, "Too many sort fields (Max = 2)"), \
C(INVALID_SORT_FIELD, "Sort field must be a key or a val"), \
- C(INVALID_STR_OPERAND, "String type can not be an operand in expression"),
+ C(INVALID_STR_OPERAND, "String type can not be an operand in expression"), \
+ C(EXPECT_NUMBER, "Expecting numeric literal"),
#undef C
#define C(a, b) HIST_ERR_##a
@@ -89,6 +90,7 @@ typedef u64 (*hist_field_fn_t) (struct hist_field *field,
#define HIST_FIELD_OPERANDS_MAX 2
#define HIST_FIELDS_MAX (TRACING_MAP_FIELDS_MAX + TRACING_MAP_VARS_MAX)
#define HIST_ACTIONS_MAX 8
+#define HIST_CONST_DIGITS_MAX 21
enum field_op_id {
FIELD_OP_NONE,
@@ -152,6 +154,9 @@ struct hist_field {
bool read_once;
unsigned int var_str_idx;
+
+ /* Numeric literals are represented as u64 */
+ u64 constant;
};
static u64 hist_field_none(struct hist_field *field,
@@ -163,6 +168,15 @@ static u64 hist_field_none(struct hist_field *field,
return 0;
}
+static u64 hist_field_const(struct hist_field *field,
+ struct tracing_map_elt *elt,
+ struct trace_buffer *buffer,
+ struct ring_buffer_event *rbe,
+ void *event)
+{
+ return field->constant;
+}
+
static u64 hist_field_counter(struct hist_field *field,
struct tracing_map_elt *elt,
struct trace_buffer *buffer,
@@ -341,6 +355,7 @@ enum hist_field_flags {
HIST_FIELD_FL_CPU = 1 << 15,
HIST_FIELD_FL_ALIAS = 1 << 16,
HIST_FIELD_FL_BUCKET = 1 << 17,
+ HIST_FIELD_FL_CONST = 1 << 18,
};
struct var_defs {
@@ -1516,6 +1531,12 @@ static void expr_field_str(struct hist_field *field, char *expr)
{
if (field->flags & HIST_FIELD_FL_VAR_REF)
strcat(expr, "$");
+ else if (field->flags & HIST_FIELD_FL_CONST) {
+ char str[HIST_CONST_DIGITS_MAX];
+
+ snprintf(str, HIST_CONST_DIGITS_MAX, "%llu", field->constant);
+ strcat(expr, str);
+ }
strcat(expr, hist_field_name(field, 0));
@@ -1689,6 +1710,15 @@ static struct hist_field *create_hist_field(struct hist_trigger_data *hist_data,
goto out;
}
+ if (flags & HIST_FIELD_FL_CONST) {
+ hist_field->fn = hist_field_const;
+ hist_field->size = sizeof(u64);
+ hist_field->type = kstrdup("u64", GFP_KERNEL);
+ if (!hist_field->type)
+ goto free;
+ goto out;
+ }
+
if (flags & HIST_FIELD_FL_STACKTRACE) {
hist_field->fn = hist_field_none;
goto out;
@@ -2090,6 +2120,29 @@ static struct hist_field *create_alias(struct hist_trigger_data *hist_data,
return alias;
}
+static struct hist_field *parse_const(struct hist_trigger_data *hist_data,
+ char *str, char *var_name,
+ unsigned long *flags)
+{
+ struct trace_array *tr = hist_data->event_file->tr;
+ struct hist_field *field = NULL;
+ u64 constant;
+
+ if (kstrtoull(str, 0, &constant)) {
+ hist_err(tr, HIST_ERR_EXPECT_NUMBER, errpos(str));
+ return NULL;
+ }
+
+ *flags |= HIST_FIELD_FL_CONST;
+ field = create_hist_field(hist_data, NULL, *flags, var_name);
+ if (!field)
+ return NULL;
+
+ field->constant = constant;
+
+ return field;
+}
+
static struct hist_field *parse_atom(struct hist_trigger_data *hist_data,
struct trace_event_file *file, char *str,
unsigned long *flags, char *var_name)
@@ -2100,6 +2153,15 @@ static struct hist_field *parse_atom(struct hist_trigger_data *hist_data,
unsigned long buckets = 0;
int ret = 0;
+ if (isdigit(str[0])) {
+ hist_field = parse_const(hist_data, str, var_name, flags);
+ if (!hist_field) {
+ ret = -EINVAL;
+ goto out;
+ }
+ return hist_field;
+ }
+
s = strchr(str, '.');
if (s) {
s = strchr(++s, '.');
@@ -4950,6 +5012,8 @@ static void hist_field_debug_show_flags(struct seq_file *m,
if (flags & HIST_FIELD_FL_ALIAS)
seq_puts(m, " HIST_FIELD_FL_ALIAS\n");
+ else if (flags & HIST_FIELD_FL_CONST)
+ seq_puts(m, " HIST_FIELD_FL_CONST\n");
}
static int hist_field_debug_show(struct seq_file *m,
@@ -4971,6 +5035,9 @@ static int hist_field_debug_show(struct seq_file *m,
field->var.idx);
}
+ if (field->flags & HIST_FIELD_FL_CONST)
+ seq_printf(m, " constant: %llu\n", field->constant);
+
if (field->flags & HIST_FIELD_FL_ALIAS)
seq_printf(m, " var_ref_idx (into hist_data->var_refs[]): %u\n",
field->var_ref_idx);
@@ -5213,6 +5280,8 @@ static void hist_field_print(struct seq_file *m, struct hist_field *hist_field)
if (hist_field->flags & HIST_FIELD_FL_CPU)
seq_puts(m, "common_cpu");
+ else if (hist_field->flags & HIST_FIELD_FL_CONST)
+ seq_printf(m, "%llu", hist_field->constant);
else if (field_name) {
if (hist_field->flags & HIST_FIELD_FL_VAR_REF ||
hist_field->flags & HIST_FIELD_FL_ALIAS)
--
2.33.0.1079.g6e70778dc9-goog
The '-' in .sym-offset can confuse the hist trigger arithmetic
expression parsing. Simplify the handling of this by replacing the
'sym-offset' with 'symXoffset'. This allows us to correctly evaluate
expressions where the user may have inadvertently added a .sym-offset
modifier to one of the operands in an expression, instead of bailing
out. In this case the .sym-offset has no effect on the evaluation of the
expression. The only valid use of the .sym-offset is as a hist key
modifier.
Change-Id: Icef416c67138576718e25394d85f8e991d8850d6
Signed-off-by: Kalesh Singh <[email protected]>
Suggested-by: Steven Rostedt <[email protected]>
---
kernel/trace/trace_events_hist.c | 43 +++++++++++++-------------------
1 file changed, 17 insertions(+), 26 deletions(-)
diff --git a/kernel/trace/trace_events_hist.c b/kernel/trace/trace_events_hist.c
index e10c7d9611e5..34aba07d23f8 100644
--- a/kernel/trace/trace_events_hist.c
+++ b/kernel/trace/trace_events_hist.c
@@ -68,8 +68,7 @@
C(INVALID_SORT_FIELD, "Sort field must be a key or a val"), \
C(INVALID_STR_OPERAND, "String type can not be an operand in expression"), \
C(EXPECT_NUMBER, "Expecting numeric literal"), \
- C(UNARY_MINUS_SUBEXPR, "Unary minus not supported in sub-expressions"), \
- C(SYM_OFFSET_SUBEXPR, ".sym-offset not supported in sub-expressions"),
+ C(UNARY_MINUS_SUBEXPR, "Unary minus not supported in sub-expressions"),
#undef C
#define C(a, b) HIST_ERR_##a
@@ -1672,10 +1671,6 @@ static int contains_operator(char *str, char **sep)
*/
minus_op = strrchr(str, '-');
if (minus_op) {
- /* Unfortunately, the modifier ".sym-offset" can confuse things. */
- if (minus_op - str >= 4 && !strncmp(minus_op - 4, ".sym-offset", 11))
- goto out;
-
/*
* Unary minus is not supported in sub-expressions. If
* present, it is always the next root operator.
@@ -2138,7 +2133,11 @@ parse_field(struct hist_trigger_data *hist_data, struct trace_event_file *file,
*flags |= HIST_FIELD_FL_HEX;
else if (strcmp(modifier, "sym") == 0)
*flags |= HIST_FIELD_FL_SYM;
- else if (strcmp(modifier, "sym-offset") == 0)
+ /*
+ * 'sym-offset' occurrences in the trigger string are modified
+ * to 'symXoffset' to simplify arithmetic expression parsing.
+ */
+ else if (strcmp(modifier, "symXoffset") == 0)
*flags |= HIST_FIELD_FL_SYM_OFFSET;
else if ((strcmp(modifier, "execname") == 0) &&
(strcmp(field_name, "common_pid") == 0))
@@ -2463,24 +2462,6 @@ static struct hist_field *parse_expr(struct hist_trigger_data *hist_data,
return ERR_PTR(-EINVAL);
}
- /*
- * ".sym-offset" in expressions has no effect on their evaluation,
- * but can confuse operator parsing.
- */
- if (*n_subexprs == 0) {
- sep = strstr(str, ".sym-offset");
- if (sep) {
- *sep = '\0';
- if (strpbrk(str, "+-/*") || strpbrk(sep + 11, "+-/*")) {
- *sep = '.';
- hist_err(file->tr, HIST_ERR_SYM_OFFSET_SUBEXPR,
- errpos(sep));
- return ERR_PTR(-EINVAL);
- }
- *sep = '.';
- }
- }
-
field_op = contains_operator(str, &sep);
if (field_op == FIELD_OP_NONE)
@@ -6004,7 +5985,7 @@ static int event_hist_trigger_func(struct event_command *cmd_ops,
struct synth_event *se;
const char *se_name;
bool remove = false;
- char *trigger, *p;
+ char *trigger, *p, *start;
int ret = 0;
lockdep_assert_held(&event_mutex);
@@ -6052,6 +6033,16 @@ static int event_hist_trigger_func(struct event_command *cmd_ops,
trigger = strstrip(trigger);
}
+ /*
+ * To simplify arithmetic expression parsing, replace occurrences of
+ * '.sym-offset' modifier with '.symXoffset'
+ */
+ start = strstr(trigger, ".sym-offset");
+ while (start) {
+ *(start + 4) = 'X';
+ start = strstr(start + 11, ".sym-offset");
+ };
+
attrs = parse_hist_trigger_attrs(file->tr, trigger);
if (IS_ERR(attrs))
return PTR_ERR(attrs);
--
2.33.0.1079.g6e70778dc9-goog
The current histogram expression evaluation logic evaluates the
expression from right to left. This can lead to incorrect results
if the operations are not associative (as is the case for subtraction
and, the now added, division operators).
e.g. 16-8-4-2 should be 2 not 10 --> 16-8-4-2 = ((16-8)-4)-2
64/8/4/2 should be 1 not 16 --> 64/8/4/2 = ((64/8)/4)/2
Division and multiplication are currently limited to single operation
expression due to operator precedence support not yet implemented.
Rework the expression parsing to support the correct evaluation of
expressions containing operators of different precedences; and fix
the associativity error by evaluating expressions with operators of
the same precedence from left to right.
Examples:
(1) echo 'hist:keys=common_pid:a=8,b=4,c=2,d=1,w=$a-$b-$c-$d' \
>> event/trigger
(2) echo 'hist:keys=common_pid:x=$a/$b/3/2' >> event/trigger
(3) echo 'hist:keys=common_pid:y=$a+10/$c*1024' >> event/trigger
(4) echo 'hist:keys=common_pid:z=$a/$b+$c*$d' >> event/trigger
Signed-off-by: Kalesh Singh <[email protected]>
Reviewed-by: Namhyung Kim <[email protected]>
---
Changed in v2:
- Add Namhyung's Reviewed-by
kernel/trace/trace_events_hist.c | 210 ++++++++++++++++++++-----------
1 file changed, 140 insertions(+), 70 deletions(-)
diff --git a/kernel/trace/trace_events_hist.c b/kernel/trace/trace_events_hist.c
index 522355a06f58..e10c7d9611e5 100644
--- a/kernel/trace/trace_events_hist.c
+++ b/kernel/trace/trace_events_hist.c
@@ -67,7 +67,9 @@
C(TOO_MANY_SORT_FIELDS, "Too many sort fields (Max = 2)"), \
C(INVALID_SORT_FIELD, "Sort field must be a key or a val"), \
C(INVALID_STR_OPERAND, "String type can not be an operand in expression"), \
- C(EXPECT_NUMBER, "Expecting numeric literal"),
+ C(EXPECT_NUMBER, "Expecting numeric literal"), \
+ C(UNARY_MINUS_SUBEXPR, "Unary minus not supported in sub-expressions"), \
+ C(SYM_OFFSET_SUBEXPR, ".sym-offset not supported in sub-expressions"),
#undef C
#define C(a, b) HIST_ERR_##a
@@ -1644,40 +1646,96 @@ static char *expr_str(struct hist_field *field, unsigned int level)
return expr;
}
-static int contains_operator(char *str)
+/*
+ * If field_op != FIELD_OP_NONE, *sep points to the root operator
+ * of the expression tree to be evaluated.
+ */
+static int contains_operator(char *str, char **sep)
{
enum field_op_id field_op = FIELD_OP_NONE;
- char *op;
+ char *minus_op, *plus_op, *div_op, *mult_op;
+
+
+ /*
+ * Report the last occurrence of the operators first, so that the
+ * expression is evaluated left to right. This is important since
+ * subtraction and division are not associative.
+ *
+ * e.g
+ * 64/8/4/2 is 1, i.e 64/8/4/2 = ((64/8)/4)/2
+ * 14-7-5-2 is 0, i.e 14-7-5-2 = ((14-7)-5)-2
+ */
- op = strpbrk(str, "+-/*");
- if (!op)
- return FIELD_OP_NONE;
+ /*
+ * First, find lower precedence addition and subtraction
+ * since the expression will be evaluated recursively.
+ */
+ minus_op = strrchr(str, '-');
+ if (minus_op) {
+ /* Unfortunately, the modifier ".sym-offset" can confuse things. */
+ if (minus_op - str >= 4 && !strncmp(minus_op - 4, ".sym-offset", 11))
+ goto out;
- switch (*op) {
- case '-':
/*
- * Unfortunately, the modifier ".sym-offset"
- * can confuse things.
+ * Unary minus is not supported in sub-expressions. If
+ * present, it is always the next root operator.
*/
- if (op - str >= 4 && !strncmp(op - 4, ".sym-offset", 11))
- return FIELD_OP_NONE;
-
- if (*str == '-')
+ if (minus_op == str) {
field_op = FIELD_OP_UNARY_MINUS;
- else
- field_op = FIELD_OP_MINUS;
- break;
- case '+':
- field_op = FIELD_OP_PLUS;
- break;
- case '/':
+ goto out;
+ }
+
+ field_op = FIELD_OP_MINUS;
+ }
+
+ plus_op = strrchr(str, '+');
+ if (plus_op || minus_op) {
+ /*
+ * For operators of the same precedence use to rightmost as the
+ * root, so that the expression is evaluated left to right.
+ */
+ if (plus_op > minus_op)
+ field_op = FIELD_OP_PLUS;
+ goto out;
+ }
+
+ /*
+ * Multiplication and division have higher precedence than addition and
+ * subtraction.
+ */
+ div_op = strrchr(str, '/');
+ if (div_op)
field_op = FIELD_OP_DIV;
- break;
- case '*':
+
+ mult_op = strrchr(str, '*');
+ /*
+ * For operators of the same precedence use to rightmost as the
+ * root, so that the expression is evaluated left to right.
+ */
+ if (mult_op > div_op)
field_op = FIELD_OP_MULT;
- break;
- default:
- break;
+
+out:
+ if (sep) {
+ switch (field_op) {
+ case FIELD_OP_UNARY_MINUS:
+ case FIELD_OP_MINUS:
+ *sep = minus_op;
+ break;
+ case FIELD_OP_PLUS:
+ *sep = plus_op;
+ break;
+ case FIELD_OP_DIV:
+ *sep = div_op;
+ break;
+ case FIELD_OP_MULT:
+ *sep = mult_op;
+ break;
+ case FIELD_OP_NONE:
+ default:
+ *sep = NULL;
+ break;
+ }
}
return field_op;
@@ -2003,7 +2061,7 @@ static char *field_name_from_var(struct hist_trigger_data *hist_data,
if (strcmp(var_name, name) == 0) {
field = hist_data->attrs->var_defs.expr[i];
- if (contains_operator(field) || is_var_ref(field))
+ if (contains_operator(field, NULL) || is_var_ref(field))
continue;
return field;
}
@@ -2266,21 +2324,24 @@ static struct hist_field *parse_atom(struct hist_trigger_data *hist_data,
static struct hist_field *parse_expr(struct hist_trigger_data *hist_data,
struct trace_event_file *file,
char *str, unsigned long flags,
- char *var_name, unsigned int level);
+ char *var_name, unsigned int *n_subexprs);
static struct hist_field *parse_unary(struct hist_trigger_data *hist_data,
struct trace_event_file *file,
char *str, unsigned long flags,
- char *var_name, unsigned int level)
+ char *var_name, unsigned int *n_subexprs)
{
struct hist_field *operand1, *expr = NULL;
unsigned long operand_flags;
int ret = 0;
char *s;
+ /* Unary minus operator, increment n_subexprs */
+ ++*n_subexprs;
+
/* we support only -(xxx) i.e. explicit parens required */
- if (level > 3) {
+ if (*n_subexprs > 3) {
hist_err(file->tr, HIST_ERR_TOO_MANY_SUBEXPR, errpos(str));
ret = -EINVAL;
goto free;
@@ -2297,8 +2358,16 @@ static struct hist_field *parse_unary(struct hist_trigger_data *hist_data,
}
s = strrchr(str, ')');
- if (s)
+ if (s) {
+ /* unary minus not supported in sub-expressions */
+ if (*(s+1) != '\0') {
+ hist_err(file->tr, HIST_ERR_UNARY_MINUS_SUBEXPR,
+ errpos(str));
+ ret = -EINVAL;
+ goto free;
+ }
*s = '\0';
+ }
else {
ret = -EINVAL; /* no closing ')' */
goto free;
@@ -2312,7 +2381,7 @@ static struct hist_field *parse_unary(struct hist_trigger_data *hist_data,
}
operand_flags = 0;
- operand1 = parse_expr(hist_data, file, str, operand_flags, NULL, ++level);
+ operand1 = parse_expr(hist_data, file, str, operand_flags, NULL, n_subexprs);
if (IS_ERR(operand1)) {
ret = PTR_ERR(operand1);
goto free;
@@ -2382,60 +2451,61 @@ static int check_expr_operands(struct trace_array *tr,
static struct hist_field *parse_expr(struct hist_trigger_data *hist_data,
struct trace_event_file *file,
char *str, unsigned long flags,
- char *var_name, unsigned int level)
+ char *var_name, unsigned int *n_subexprs)
{
struct hist_field *operand1 = NULL, *operand2 = NULL, *expr = NULL;
unsigned long operand_flags;
int field_op, ret = -EINVAL;
char *sep, *operand1_str;
- if (level > 3) {
+ if (*n_subexprs > 3) {
hist_err(file->tr, HIST_ERR_TOO_MANY_SUBEXPR, errpos(str));
return ERR_PTR(-EINVAL);
}
- field_op = contains_operator(str);
+ /*
+ * ".sym-offset" in expressions has no effect on their evaluation,
+ * but can confuse operator parsing.
+ */
+ if (*n_subexprs == 0) {
+ sep = strstr(str, ".sym-offset");
+ if (sep) {
+ *sep = '\0';
+ if (strpbrk(str, "+-/*") || strpbrk(sep + 11, "+-/*")) {
+ *sep = '.';
+ hist_err(file->tr, HIST_ERR_SYM_OFFSET_SUBEXPR,
+ errpos(sep));
+ return ERR_PTR(-EINVAL);
+ }
+ *sep = '.';
+ }
+ }
+
+ field_op = contains_operator(str, &sep);
if (field_op == FIELD_OP_NONE)
return parse_atom(hist_data, file, str, &flags, var_name);
if (field_op == FIELD_OP_UNARY_MINUS)
- return parse_unary(hist_data, file, str, flags, var_name, ++level);
+ return parse_unary(hist_data, file, str, flags, var_name, n_subexprs);
- switch (field_op) {
- case FIELD_OP_MINUS:
- sep = "-";
- break;
- case FIELD_OP_PLUS:
- sep = "+";
- break;
- case FIELD_OP_DIV:
- sep = "/";
- break;
- case FIELD_OP_MULT:
- sep = "*";
- break;
- default:
- goto free;
- }
+ /* Binary operator found, increment n_subexprs */
+ ++*n_subexprs;
- /*
- * Multiplication and division are only supported in single operator
- * expressions, since the expression is always evaluated from right
- * to left.
- */
- if ((field_op == FIELD_OP_DIV || field_op == FIELD_OP_MULT) && level > 0) {
- hist_err(file->tr, HIST_ERR_TOO_MANY_SUBEXPR, errpos(str));
- return ERR_PTR(-EINVAL);
- }
+ /* Split the expression string at the root operator */
+ if (!sep)
+ goto free;
+ *sep = '\0';
+ operand1_str = str;
+ str = sep+1;
- operand1_str = strsep(&str, sep);
if (!operand1_str || !str)
goto free;
operand_flags = 0;
- operand1 = parse_atom(hist_data, file, operand1_str,
- &operand_flags, NULL);
+
+ /* LHS of string is an expression e.g. a+b in a+b+c */
+ operand1 = parse_expr(hist_data, file, operand1_str, operand_flags, NULL, n_subexprs);
if (IS_ERR(operand1)) {
ret = PTR_ERR(operand1);
operand1 = NULL;
@@ -2447,9 +2517,9 @@ static struct hist_field *parse_expr(struct hist_trigger_data *hist_data,
goto free;
}
- /* rest of string could be another expression e.g. b+c in a+b+c */
+ /* RHS of string is another expression e.g. c in a+b+c */
operand_flags = 0;
- operand2 = parse_expr(hist_data, file, str, operand_flags, NULL, ++level);
+ operand2 = parse_expr(hist_data, file, str, operand_flags, NULL, n_subexprs);
if (IS_ERR(operand2)) {
ret = PTR_ERR(operand2);
operand2 = NULL;
@@ -3883,9 +3953,9 @@ static int __create_val_field(struct hist_trigger_data *hist_data,
unsigned long flags)
{
struct hist_field *hist_field;
- int ret = 0;
+ int ret = 0, n_subexprs = 0;
- hist_field = parse_expr(hist_data, file, field_str, flags, var_name, 0);
+ hist_field = parse_expr(hist_data, file, field_str, flags, var_name, &n_subexprs);
if (IS_ERR(hist_field)) {
ret = PTR_ERR(hist_field);
goto out;
@@ -4026,7 +4096,7 @@ static int create_key_field(struct hist_trigger_data *hist_data,
struct hist_field *hist_field = NULL;
unsigned long flags = 0;
unsigned int key_size;
- int ret = 0;
+ int ret = 0, n_subexprs = 0;
if (WARN_ON(key_idx >= HIST_FIELDS_MAX))
return -EINVAL;
@@ -4039,7 +4109,7 @@ static int create_key_field(struct hist_trigger_data *hist_data,
hist_field = create_hist_field(hist_data, NULL, flags, NULL);
} else {
hist_field = parse_expr(hist_data, file, field_str, flags,
- NULL, 0);
+ NULL, &n_subexprs);
if (IS_ERR(hist_field)) {
ret = PTR_ERR(hist_field);
goto out;
--
2.33.0.1079.g6e70778dc9-goog
The division is a slow operation. If the divisor is a power of 2, use a
shift instead.
Results were obtained using Android's version of perf (simpleperf[1]) as
described below:
1. hist_field_div() is modified to call 2 test functions:
test_hist_field_div_[not]_optimized(); passing them the
same args. Use noinline and volatile to ensure these are
not optimized out by the compiler.
2. Create a hist event trigger that uses division:
events/kmem/rss_stat$ echo 'hist:keys=common_pid:x=size/<divisor>'
>> trigger
events/kmem/rss_stat$ echo 'hist:keys=common_pid:vals=$x'
>> trigger
3. Run Android's lmkd_test[2] to generate rss_stat events, and
record CPU samples with Android's simpleperf:
simpleperf record -a --exclude-perf --post-unwind=yes -m 16384 -g
-f 2000 -o perf.data
== Results ==
Divisor is a power of 2 (divisor == 32):
test_hist_field_div_not_optimized | 8,717,091 cpu-cycles
test_hist_field_div_optimized | 1,643,137 cpu-cycles
If the divisor is a power of 2, the optimized version is ~5.3x faster.
Divisor is not a power of 2 (divisor == 33):
test_hist_field_div_not_optimized | 4,444,324 cpu-cycles
test_hist_field_div_optimized | 5,497,958 cpu-cycles
If the divisor is not a power of 2, as expected, the optimized version is
slightly slower (~24% slower).
[1] https://android.googlesource.com/platform/system/extras/+/master/simpleperf/doc/README.md
[2] https://cs.android.com/android/platform/superproject/+/master:system/memory/lmkd/tests/lmkd_test.cpp
Signed-off-by: Kalesh Singh <[email protected]>
Suggested-by: Steven Rostedt <[email protected]>
---
kernel/trace/trace_events_hist.c | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/kernel/trace/trace_events_hist.c b/kernel/trace/trace_events_hist.c
index db28bcf976f4..364cb3091789 100644
--- a/kernel/trace/trace_events_hist.c
+++ b/kernel/trace/trace_events_hist.c
@@ -304,6 +304,10 @@ static u64 hist_field_div(struct hist_field *hist_field,
if (!val2)
return -1;
+ /* Use shift if the divisor is a power of 2 */
+ if (!(val2 & (val2 - 1)))
+ return val1 >> __ffs64(val2);
+
return div64_u64(val1, val2);
}
--
2.33.0.1079.g6e70778dc9-goog
If both operands of a hist trigger expression are constants, convert the
expression to a constant. This optimization avoids having to perform the
same calculation multiple times and also saves on memory since the
merged constants are represented by a single struct hist_field instead
or multiple.
Change-Id: I83388bbbc6e109980f08eac437172238576055d5
Signed-off-by: Kalesh Singh <[email protected]>
Suggested-by: Steven Rostedt <[email protected]>
---
kernel/trace/trace_events_hist.c | 104 ++++++++++++++++++++++---------
1 file changed, 74 insertions(+), 30 deletions(-)
diff --git a/kernel/trace/trace_events_hist.c b/kernel/trace/trace_events_hist.c
index 34aba07d23f8..db28bcf976f4 100644
--- a/kernel/trace/trace_events_hist.c
+++ b/kernel/trace/trace_events_hist.c
@@ -2411,9 +2411,15 @@ static struct hist_field *parse_unary(struct hist_trigger_data *hist_data,
return ERR_PTR(ret);
}
+/*
+ * If the operands are var refs, return pointers the
+ * variable(s) referenced in var1 and var2, else NULL.
+ */
static int check_expr_operands(struct trace_array *tr,
struct hist_field *operand1,
- struct hist_field *operand2)
+ struct hist_field *operand2,
+ struct hist_field **var1,
+ struct hist_field **var2)
{
unsigned long operand1_flags = operand1->flags;
unsigned long operand2_flags = operand2->flags;
@@ -2426,6 +2432,7 @@ static int check_expr_operands(struct trace_array *tr,
if (!var)
return -EINVAL;
operand1_flags = var->flags;
+ *var1 = var;
}
if ((operand2_flags & HIST_FIELD_FL_VAR_REF) ||
@@ -2436,6 +2443,7 @@ static int check_expr_operands(struct trace_array *tr,
if (!var)
return -EINVAL;
operand2_flags = var->flags;
+ *var2 = var;
}
if ((operand1_flags & HIST_FIELD_FL_TIMESTAMP_USECS) !=
@@ -2453,9 +2461,12 @@ static struct hist_field *parse_expr(struct hist_trigger_data *hist_data,
char *var_name, unsigned int *n_subexprs)
{
struct hist_field *operand1 = NULL, *operand2 = NULL, *expr = NULL;
- unsigned long operand_flags;
+ struct hist_field *var1 = NULL, *var2 = NULL;
+ unsigned long operand_flags, operand2_flags;
int field_op, ret = -EINVAL;
char *sep, *operand1_str;
+ hist_field_fn_t op_fn;
+ bool combine_consts;
if (*n_subexprs > 3) {
hist_err(file->tr, HIST_ERR_TOO_MANY_SUBEXPR, errpos(str));
@@ -2512,11 +2523,38 @@ static struct hist_field *parse_expr(struct hist_trigger_data *hist_data,
goto free;
}
- ret = check_expr_operands(file->tr, operand1, operand2);
+ switch (field_op) {
+ case FIELD_OP_MINUS:
+ op_fn = hist_field_minus;
+ break;
+ case FIELD_OP_PLUS:
+ op_fn = hist_field_plus;
+ break;
+ case FIELD_OP_DIV:
+ op_fn = hist_field_div;
+ break;
+ case FIELD_OP_MULT:
+ op_fn = hist_field_mult;
+ break;
+ default:
+ ret = -EINVAL;
+ goto free;
+ }
+
+ ret = check_expr_operands(file->tr, operand1, operand2, &var1, &var2);
if (ret)
goto free;
- flags |= HIST_FIELD_FL_EXPR;
+ operand_flags = var1 ? var1->flags : operand1->flags;
+ operand2_flags = var2 ? var2->flags : operand2->flags;
+
+ /*
+ * If both operands are constant, the expression can be
+ * collapsed to a single constant.
+ */
+ combine_consts = operand_flags & operand2_flags & HIST_FIELD_FL_CONST;
+
+ flags |= combine_consts ? HIST_FIELD_FL_CONST : HIST_FIELD_FL_EXPR;
flags |= operand1->flags &
(HIST_FIELD_FL_TIMESTAMP | HIST_FIELD_FL_TIMESTAMP_USECS);
@@ -2533,37 +2571,43 @@ static struct hist_field *parse_expr(struct hist_trigger_data *hist_data,
expr->operands[0] = operand1;
expr->operands[1] = operand2;
- /* The operand sizes should be the same, so just pick one */
- expr->size = operand1->size;
+ if (combine_consts) {
+ if (var1)
+ expr->operands[0] = var1;
+ if (var2)
+ expr->operands[1] = var2;
- expr->operator = field_op;
- expr->name = expr_str(expr, 0);
- expr->type = kstrdup_const(operand1->type, GFP_KERNEL);
- if (!expr->type) {
- ret = -ENOMEM;
- goto free;
- }
+ expr->constant = op_fn(expr, NULL, NULL, NULL, NULL);
- switch (field_op) {
- case FIELD_OP_MINUS:
- expr->fn = hist_field_minus;
- break;
- case FIELD_OP_PLUS:
- expr->fn = hist_field_plus;
- break;
- case FIELD_OP_DIV:
- expr->fn = hist_field_div;
- break;
- case FIELD_OP_MULT:
- expr->fn = hist_field_mult;
- break;
- default:
- ret = -EINVAL;
- goto free;
+ expr->operands[0] = NULL;
+ expr->operands[1] = NULL;
+
+ /*
+ * var refs won't be destroyed immediately
+ * See: destroy_hist_field()
+ */
+ destroy_hist_field(operand2, 0);
+ destroy_hist_field(operand1, 0);
+
+ expr->name = expr_str(expr, 0);
+ } else {
+ expr->fn = op_fn;
+
+ /* The operand sizes should be the same, so just pick one */
+ expr->size = operand1->size;
+
+ expr->operator = field_op;
+ expr->type = kstrdup_const(operand1->type, GFP_KERNEL);
+ if (!expr->type) {
+ ret = -ENOMEM;
+ goto free;
+ }
+
+ expr->name = expr_str(expr, 0);
}
return expr;
- free:
+free:
destroy_hist_field(operand1, 0);
destroy_hist_field(operand2, 0);
destroy_hist_field(expr, 0);
--
2.33.0.1079.g6e70778dc9-goog
Add tests for the parsing of hist trigger expressions; and to
validate expression evaluation.
Signed-off-by: Kalesh Singh <[email protected]>
Reviewed-by: Namhyung Kim <[email protected]>
Change-Id: Id3871697eb2d7537c8185494f5c1d347077798bc
---
Changes in v3:
- Remove .sym-offset error check tests
Changes in v2:
- Add Namhyung's Reviewed-by
- Update comment to clarify err_pos in "Too many subexpressions" test
.../testing/selftests/ftrace/test.d/functions | 4 +-
.../trigger/trigger-hist-expressions.tc | 72 +++++++++++++++++++
2 files changed, 74 insertions(+), 2 deletions(-)
create mode 100644 tools/testing/selftests/ftrace/test.d/trigger/trigger-hist-expressions.tc
diff --git a/tools/testing/selftests/ftrace/test.d/functions b/tools/testing/selftests/ftrace/test.d/functions
index 000fd05e84b1..1855a63559ad 100644
--- a/tools/testing/selftests/ftrace/test.d/functions
+++ b/tools/testing/selftests/ftrace/test.d/functions
@@ -16,13 +16,13 @@ reset_tracer() { # reset the current tracer
reset_trigger_file() {
# remove action triggers first
- grep -H ':on[^:]*(' $@ |
+ grep -H ':on[^:]*(' $@ | tac |
while read line; do
cmd=`echo $line | cut -f2- -d: | cut -f1 -d"["`
file=`echo $line | cut -f1 -d:`
echo "!$cmd" >> $file
done
- grep -Hv ^# $@ |
+ grep -Hv ^# $@ | tac |
while read line; do
cmd=`echo $line | cut -f2- -d: | cut -f1 -d"["`
file=`echo $line | cut -f1 -d:`
diff --git a/tools/testing/selftests/ftrace/test.d/trigger/trigger-hist-expressions.tc b/tools/testing/selftests/ftrace/test.d/trigger/trigger-hist-expressions.tc
new file mode 100644
index 000000000000..e715641c54d3
--- /dev/null
+++ b/tools/testing/selftests/ftrace/test.d/trigger/trigger-hist-expressions.tc
@@ -0,0 +1,72 @@
+#!/bin/sh
+# SPDX-License-Identifier: GPL-2.0
+# description: event trigger - test histogram expression parsing
+# requires: set_event events/sched/sched_process_fork/trigger events/sched/sched_process_fork/hist error_log
+
+
+fail() { #msg
+ echo $1
+ exit_fail
+}
+
+get_hist_var() { #var_name hist_path
+ hist_output=`grep -m1 "$1: " $2`
+ hitcount=`echo $hist_output | awk '{ for (i=1; i<=NF; ++i) { if ($i ~ "hitcount:") print $(i+1)} }'`
+ var_sum=`echo $hist_output | awk '{ for (i=1; i<=NF; ++i) { if ($i ~ "'$1':") print $(i+1)} }'`
+ var_val=$(( var_sum / hitcount ))
+ echo $var_val
+}
+
+test_hist_expr() { # test_name expression expected_val
+ reset_trigger
+
+ echo "Test hist trigger expressions - $1"
+
+ echo "hist:keys=common_pid:x=$2" > events/sched/sched_process_fork/trigger
+ echo 'hist:keys=common_pid:vals=$x' >> events/sched/sched_process_fork/trigger
+ for i in `seq 1 10` ; do ( echo "forked" > /dev/null); done
+
+ actual=`get_hist_var x events/sched/sched_process_fork/hist`
+
+ if [ $actual != $3 ]; then
+ fail "Failed hist trigger expression evaluation: Expression: $2 Expected: $3, Actual: $actual"
+ fi
+
+ reset_trigger
+}
+
+check_error() { # test_name command-with-error-pos-by-^
+ reset_trigger
+
+ echo "Test hist trigger expressions - $1"
+ ftrace_errlog_check 'hist:sched:sched_process_fork' "$2" 'events/sched/sched_process_fork/trigger'
+
+ reset_trigger
+}
+
+test_hist_expr "Variable assignment" "123" "123"
+
+test_hist_expr "Subtraction not associative" "16-8-4-2" "2"
+
+test_hist_expr "Division not associative" "64/8/4/2" "1"
+
+test_hist_expr "Same precedence operators (+,-) evaluated left to right" "16-8+4+2" "14"
+
+test_hist_expr "Same precedence operators (*,/) evaluated left to right" "4*3/2*2" "12"
+
+test_hist_expr "Multiplication evaluated before addition/subtraction" "4+3*2-2" "8"
+
+test_hist_expr "Division evaluated before addition/subtraction" "4+6/2-2" "5"
+
+# Division by zero returns -1
+test_hist_expr "Handles division by zero" "3/0" "-1"
+
+# err pos for "too many subexpressions" is dependent on where
+# the last subexpression was detected. This can vary depending
+# on how the expression tree was generated.
+check_error "Too many subexpressions" 'hist:keys=common_pid:x=32+^10*3/20-4'
+check_error "Too many subexpressions" 'hist:keys=common_pid:x=^1+2+3+4+5'
+
+check_error "Unary minus not supported in subexpression" 'hist:keys=common_pid:x=-(^1)+2'
+
+exit 0
--
2.33.0.1079.g6e70778dc9-goog
Histogram expressions now support division, and multiplication in
addition to the already supported subtraction and addition operators.
Numeric constants can also be used in a hist trigger expressions
or assigned to a variable and used by refernce in an expression.
Signed-off-by: Kalesh Singh <[email protected]>
Reviewed-by: Namhyung Kim <[email protected]>
---
Changes in v2:
- Add Namhyung's Reviewed-by
Documentation/trace/histogram.rst | 14 ++++++++++++++
1 file changed, 14 insertions(+)
diff --git a/Documentation/trace/histogram.rst b/Documentation/trace/histogram.rst
index 533415644c54..e12699abaee8 100644
--- a/Documentation/trace/histogram.rst
+++ b/Documentation/trace/histogram.rst
@@ -1763,6 +1763,20 @@ using the same key and variable from yet another event::
# echo 'hist:key=pid:wakeupswitch_lat=$wakeup_lat+$switchtime_lat ...' >> event3/trigger
+Expressions support the use of addition, subtraction, multiplication and
+division operators (+-*/).
+
+Note that division by zero always returns -1.
+
+Numeric constants can also be used directly in an expression::
+
+ # echo 'hist:keys=next_pid:timestamp_secs=common_timestamp/1000000 ...' >> event/trigger
+
+or assigned to a variable and referenced in a subsequent expression::
+
+ # echo 'hist:keys=next_pid:us_per_sec=1000000 ...' >> event/trigger
+ # echo 'hist:keys=next_pid:timestamp_secs=common_timestamp/$us_per_sec ...' >> event/trigger
+
2.2.2 Synthetic Events
----------------------
--
2.33.0.1079.g6e70778dc9-goog