stackcount: add -c option for CPU only events (iovisor#2307)

Add the -c option to trace events happened only the requested CPU.
Update command line, man and example.

Signed-off-by: Dima Stepanov <[email protected]>
Signed-off-by: Yury Kotov <[email protected]>

man/man8/stackcount.8

@@ -2,8 +2,8 @@
 .SH NAME
 stackcount \- Count function calls and their stack traces. Uses Linux eBPF/bcc.
 .SH SYNOPSIS
-.B stackcount [\-h] [\-p PID] [\-i INTERVAL] [\-D DURATION] [\-T] [\-r] [\-s]
- [\-P] [\-K] [\-U] [\-v] [\-d] [\-f] pattern
+.B stackcount [\-h] [\-p PID] [\-c CPU] [\-i INTERVAL] [\-D DURATION] [\-T] [\-r]
+ [\-s] [\-P] [\-K] [\-U] [\-v] [\-d] [\-f] pattern
 .SH DESCRIPTION
 stackcount traces functions and frequency counts them with their entire
 stack trace, kernel stack and user stack, summarized in-kernel for efficiency.
@@ -54,6 +54,9 @@ Folded output format.
 \-p PID
 Trace this process ID only (filtered in-kernel).
 .TP
+\-c CPU
+Trace this CPU only (filtered in-kernel).
+.TP
 .TP
 pattern
 A function name, or a search pattern. Can include wildcards ("*"). If the
@@ -104,6 +107,10 @@ Only count stacks when PID 185 is on-CPU:
 #
 .B stackcount \-p 185 ip_output
 .TP
+Only count stacks for CPU 1:
+#
+.B stackcount \-c 1 put_prev_entity
+.TP
 Count user stacks for dynamic heap allocations with malloc in PID 185:
 #
 .B stackcount \-p 185 c:malloc

tools/stackcount.py

@@ -3,8 +3,8 @@
 # stackcount Count events and their stack traces.
 # For Linux, uses BCC, eBPF.
 #
-# USAGE: stackcount.py [-h] [-p PID] [-i INTERVAL] [-D DURATION] [-T] [-r] [-s]
-# [-P] [-K] [-U] [-v] [-d] [-f] [--debug]
+# USAGE: stackcount.py [-h] [-p PID] [-c CPU] [-i INTERVAL] [-D DURATION] [-T]
+# [-r] [-s] [-P] [-K] [-U] [-v] [-d] [-f] [--debug]
 #
 # The pattern is a string with optional '*' wildcards, similar to file
 # globbing. If you'd prefer to use regular expressions, use the -r option.
@@ -28,7 +28,7 @@
 
 class Probe(object):
  def __init__(self, pattern, kernel_stack, user_stack, use_regex=False,
- pid=None, per_pid=False):
+ pid=None, per_pid=False, cpu=None):
  """Init a new probe.
 
  Init the probe from the pattern provided by the user. The supported
@@ -75,6 +75,7 @@ def __init__(self, pattern, kernel_stack, user_stack, use_regex=False,
 
  self.pid = pid
  self.per_pid = per_pid
+ self.cpu = cpu
  self.matched = 0
 
  def is_kernel_probe(self):
@@ -149,14 +150,18 @@ def load(self):
 BPF_STACK_TRACE(stack_traces, 1024);
  """
 
+ filter_text = []
  # We really mean the tgid from the kernel's perspective, which is in
  # the top 32 bits of bpf_get_current_pid_tgid().
  if self.is_kernel_probe() and self.pid:
- trace_count_text = trace_count_text.replace('FILTER',
- ('u32 pid; pid = bpf_get_current_pid_tgid() >> 32; ' +
- 'if (pid != %d) { return 0; }') % (self.pid))
- else:
- trace_count_text = trace_count_text.replace('FILTER', '')
+ filter_text.append('u32 pid; pid = bpf_get_current_pid_tgid() >> 32; ' +
+ 'if (pid != %d) { return 0; }' % self.pid)
+
+ if self.is_kernel_probe() and self.cpu:
+ filter_text.append('struct task_struct *task; task = (struct task_struct*)bpf_get_current_task(); ' +
+ 'if (task->cpu != %d) { return 0; }' % self.cpu)
+
+ trace_count_text = trace_count_text.replace('FILTER', '\n '.join(filter_text))
 
  # We need per-pid statistics when tracing a user-space process, because
  # the meaning of the symbols depends on the pid. We also need them if
@@ -211,6 +216,7 @@ def __init__(self):
  ./stackcount -r '^tcp_send.*' # same as above, using regular expressions
  ./stackcount -Ti 5 ip_output # output every 5 seconds, with timestamps
  ./stackcount -p 185 ip_output # count ip_output stacks for PID 185 only
+ ./stackcount -c 1 put_prev_entity # count put_prev_entity stacks for CPU 1 only
  ./stackcount -p 185 c:malloc # count stacks for malloc in PID 185
  ./stackcount t:sched:sched_fork # count stacks for sched_fork tracepoint
  ./stackcount -p 185 u:node:* # count stacks for all USDT probes in node
@@ -223,6 +229,8 @@ def __init__(self):
  epilog=examples)
  parser.add_argument("-p", "--pid", type=int,
  help="trace this PID only")
+ parser.add_argument("-c", "--cpu", type=int,
+ help="trace this CPU only")
  parser.add_argument("-i", "--interval",
  help="summary interval, seconds")
  parser.add_argument("-D", "--duration",
@@ -271,7 +279,7 @@ def __init__(self):
 
  self.probe = Probe(self.args.pattern,
  self.kernel_stack, self.user_stack,
- self.args.regexp, self.args.pid, self.args.perpid)
+ self.args.regexp, self.args.pid, self.args.perpid, self.args.cpu)
  self.need_delimiter = self.args.delimited and not (
  self.args.kernel_stacks_only or self.args.user_stacks_only)
 

tools/stackcount_example.txt

@@ -843,8 +843,8 @@ This folded output can be piped directly into flamegraph.pl (the Perl version).
 USAGE message:
 
 # ./stackcount -h
-usage: stackcount [-h] [-p PID] [-i INTERVAL] [-D DURATION] [-T] [-r] [-s]
- [-P] [-K] [-U] [-v] [-d] [-f] [--debug]
+usage: stackcount [-h] [-p PID] [-c CPU] [-i INTERVAL] [-D DURATION] [-T] [-r]
+ [-s] [-P] [-K] [-U] [-v] [-d] [-f] [--debug]
  pattern
 
 Count events and their stack traces
@@ -855,6 +855,7 @@ positional arguments:
 optional arguments:
  -h, --help show this help message and exit
  -p PID, --pid PID trace this PID only
+ -c CPU, --cpu CPU trace this CPU only
  -i INTERVAL, --interval INTERVAL
  summary interval, seconds
  -D DURATION, --duration DURATION
@@ -886,5 +887,6 @@ examples:
  ./stackcount -p 185 c:malloc # count stacks for malloc in PID 185
  ./stackcount t:sched:sched_fork # count stacks for sched_fork tracepoint
  ./stackcount -p 185 u:node:* # count stacks for all USDT probes in node
+ ./stackcount -c 1 put_prev_entity # count put_prev_entity stacks for CPU 1 only
  ./stackcount -K t:sched:sched_switch # kernel stacks only
  ./stackcount -U t:sched:sched_switch # user stacks only