[lkml]   [2019]   [Mar]   [15]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [PATCH v3] tracing: kdb: Allow ftdump to skip all but the last few lines
On Fri, 15 Mar 2019 12:55:31 -0700
Doug Anderson <> wrote:

> >
> > Actually, you can add a function in trace.c that exposes
> > get_total_entries:
> >
> > unsigned long trace_total_entries(struct trace_array *tr)
> > {
> > unsigned long total, entries;
> >
> > if (!tr)
> > tr = &global_trace;
> >
> > get_total_entries(tr->trace_buffer, &total, &entries);
> >
> > return entries;
> > }
> >
> > and then do:
> > cnt = trace_total_entries(NULL);
> OK. I guess I'll need to figure out how to add an argument to limit
> it to just one CPU too since the kdb command allows you to specify a
> single CPU or all CPUs. I think the best way would mean adding an
> argument to get_total_entries(). ...or I can just change the kdb
> command to not allow specifying a CPU? Which do you prefer? I
> personally haven't ever used the feature to just print the ftrace
> buffer for a certain CPU so I'd be OK removing it.

I would make a get_total_entries_cpu() that contains the guts of
get_total_entries() which then calls get_total_entries_cpu()

static void
get_total_entries(struct trace_buffer *buf,
unsigned long *total, unsigned long *entries)
unsigned long t, e;
int cpu;

*total = 0;
*entries = 0;

get_total_entries_cpu(buf, &t, &e, cpu);
*total += t;
*entries += e;

> > Don't modify ftrace_dump_buf()
> I still kinda prefer modifying ftrace_dump_buf() just because it's
> pretty important that the "count" we come up with match pretty exactly
> the count that ftrace_dump_buf() will come up with. My worry probably
> comes from my lack of experience with the internals of ftrace but, for
> instance, I see things like "tr->trace_flags &=
> ~TRACE_ITER_SYM_USEROBJ" in ftrace_dump_buf() and I worry that it will
> affect the count. ...but if you tell me that I need not worry about
> things like that then I won't. :-)

Hmm, actually your method still wont work, because you are only
counting entries not lines. The stack dumps are considered a single
line but will print multiple lines.

Not only that, perhaps you should break apart ftrace_dump_buf(),
because calling it twice (or doing what I suggested), wont stop tracing
in between, and the size of the buffer might change between the two

You need to move out:

for_each_tracing_cpu(cpu) {
atomic_inc(&per_cpu_ptr(iter.trace_buffer->data, cpu)->disabled);


for_each_tracing_cpu(cpu) {
atomic_dec(&per_cpu_ptr(iter.trace_buffer->data, cpu)->disabled);

to disable tracing while you do this.

The get_total_entries() is the faster approach to get the count, but in
either case, the count should end up the same.

-- Steve

 \ /
  Last update: 2019-03-15 21:13    [W:0.060 / U:10.116 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site