[offcputime] allow for the speficication of a threshold for collection (#697)

There are situations, specially when using non-folded mode, where we are
interested only in very high latencies that happen due to blocking.

While we can certainly filter out the very small ones out of the output, it is a
lot more convenient to do this from the tool itself, as it would be difficult
from an external filter to do this in one pass.

But if we are to discard unused measurements, we can do even better: we can
change the bpf code itself not to grab those traces, and gain a bit of
efficiency in scenarios in which we are only concerned about peak latencies.

This scheme can be easily extended to also allow a maximum cap in the latencies
we are interested in.

After this patch is applied, the options -m and -M can be used to set those
limits respectively.

Fixes: #588

diff --git a/tools/offcputime.py b/tools/offcputime.py
index 89f5d70..95da546 100755 (executable)
--- a/tools/offcputime.py
+++ b/tools/offcputime.py
@@ -40,6 +40,8 @@ examples = """examples:
     ./offcputime             # trace off-CPU stack time until Ctrl-C
     ./offcputime 5           # trace for 5 seconds only
     ./offcputime -f 5        # 5 seconds, and output in folded format
+    ./offcputime -m 1000     # trace only events that last more than 1000 usec.
+    ./offcputime -M 10000    # trace only events that last less than 10000 usec.
     ./offcputime -p 185      # only trace threads for PID 185
     ./offcputime -t 188      # only trace thread 188
     ./offcputime -u          # only trace user threads (no kernel)
@@ -78,6 +80,12 @@ parser.add_argument("--stack-storage-size", default=1024,
 parser.add_argument("duration", nargs="?", default=99999999,
     type=positive_nonzero_int,
     help="duration of trace, in seconds")
+parser.add_argument("-m", "--min-block-time", default=1,
+    type=positive_nonzero_int,
+    help="the amount of time in microseconds over which we store traces (default 1)")
+parser.add_argument("-M", "--max-block-time", default=(1<<64)-1,
+    type=positive_nonzero_int,
+    help="the amount of time in microseconds under which we store traces (default U64_MAX)")
 args = parser.parse_args()
 if args.pid and args.tgid:
     parser.error("specify only one of -p and -t")
@@ -93,7 +101,8 @@ bpf_text = """
 #include <uapi/linux/ptrace.h>
 #include <linux/sched.h>
 
-#define MINBLOCK_US    1
+#define MINBLOCK_US    MINBLOCK_US_VALUEULL
+#define MAXBLOCK_US    MAXBLOCK_US_VALUEULL
 
 struct key_t {
     u32 pid;
@@ -129,7 +138,7 @@ int oncpu(struct pt_regs *ctx, struct task_struct *prev) {
     u64 delta = bpf_ktime_get_ns() - *tsp;
     start.delete(&pid);
     delta = delta / 1000;
-    if (delta < MINBLOCK_US) {
+    if ((delta < MINBLOCK_US) || (delta > MAXBLOCK_US)) {
         return 0;
     }
 
@@ -170,6 +179,8 @@ bpf_text = bpf_text.replace('THREAD_FILTER', thread_filter)
 
 # set stack storage size
 bpf_text = bpf_text.replace('STACK_STORAGE_SIZE', str(args.stack_storage_size))
+bpf_text = bpf_text.replace('MINBLOCK_US_VALUE', str(args.min_block_time))
+bpf_text = bpf_text.replace('MAXBLOCK_US_VALUE', str(args.max_block_time))
 
 # handle stack args
 kernel_stack_get = "stack_traces.get_stackid(ctx, BPF_F_REUSE_STACKID)"