Re: [PATCH 2/3] ftrace: Add debug_print trace to print data from kernel to userspace

["Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>]

On Fri, Nov 14, 2008 at 12:21:20PM +0100, Frédéric Weisbecker wrote:
> Hi Aneesh!
> 
> 2008/11/14 Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>:
> > The trace add a new interface debug_print() which can be used
> > to dump data from kernel to user space using ftrace framework.
> 
> 
> The actual "nop tracer" which is the default selected (if trace_boot
> is not configured) let
> the tracing engine able to receive and handle TRACE_PRINT events.
> Even if nop tracer doesn't handle its output itself, the TRACE_PRINT
> event output is relayed by
> print_trace_fmt() in trace.c
> 
> Your output does almost the same but it is already implemented.

We also want to make sure dp_printk doesn't do anything when tracer
is disabled. We do

int do_dp_printk(const char *fmt, ...)
{
	int ret;
	va_list args;

	if (!tracer_enabled)
		return 0;

.........
.......

> 
> 
> 
> > +static void dp_trace_ctrl_update(struct trace_array *tr)
> > +{
> > +       /* When starting a new trace, reset the buffers */
> > +       if (tr->ctrl)
> > +               start_dp_trace(tr);
> > +       else
> > +               stop_dp_trace(tr);
> > +}
> 
> 
> BTW, ctrl_update() have been removed very recently.
> Perhaps are you implementing this against the mainline? Its a better idea to
> submit a new tracer against latest -tip tree.

Yes the patches are against mainline.

> 
> > +
> > +#ifdef CONFIG_NOP_TRACER
> > +int
> > +trace_selftest_startup_dp(struct tracer *trace, struct trace_array *tr)
> > +{
> > +       /* What could possibly go wrong? */
> > +       return 0;
> > +}
> > +#endif
> 
> 
> CONFIG_NOP_TRACER ?
> Wouldn't it rather depend on CONFIG_FTRACE_SELFTEST? :-)
> 
> That would perhaps have been better to raise a ftrace_printk to make a test.
> If you don't do anything in your selftest, then it is unnecessary to
> implement one for
> your tracer.

I dropped the selftest callback.

> 
> 
> > +static enum print_line_t debug_print_line(struct trace_iterator *iter)
> > +{
> > +       struct trace_seq *s = &iter->seq;
> > +       struct trace_entry *entry;
> > +
> > +       entry = iter->ent;
> > +       switch (entry->type) {
> > +       case TRACE_PRINT: {
> > +               struct print_entry *field;
> > +               trace_assign_type(field, entry);
> > +
> > +               trace_seq_printf(s, "%s", field->buf);
> > +               if (entry->flags & TRACE_FLAG_CONT)
> > +                       trace_seq_print_cont(s, iter);
> > +               break;
> 
> 
> You should test if trace_seq_printf successed to print the whole line.
> If not, that's better to return TRACE_TYPE_PARTIAL_LINE, then the
> output will be retried later.

I am looking at this. I don't see TRACE_TYPE_PARTIAL_LINE in other
places in the code. Will look more.

> 
> 
> 
> > +       }
> > +       default:
> > +               printk(KERN_INFO "Unsupported type in debug_print\n");
> > +               return TRACE_TYPE_UNHANDLED;
> > +       }
> > +
> > +       return TRACE_TYPE_HANDLED;
> > +}
> > +
> > +struct tracer dp_trace __read_mostly =
> > +{
> > +       .name           = "debug_print",
> > +       .init           = dp_trace_init,
> > +       .reset          = dp_trace_reset,
> > +       .ctrl_update    = dp_trace_ctrl_update,
> > +#ifdef CONFIG_FTRACE_SELFTEST
> > +       .selftest       = trace_selftest_startup_dp,
> > +#endif
> > +       .print_line     = debug_print_line,
> > +};
> > +
> > +__init static int init_dp_trace(void)
> > +{
> > +       return register_tracer(&dp_trace);
> > +}
> > +device_initcall(init_dp_trace);
> 
> 
> Primarily, such a debug tracer is not a bad idea, IMHO.
> And note that all of I just wrote in this answer in only my opinion.
> Perhaps other people
> would find other uses of this tracer that actual default output of the
> tracing engine doesn't handle well, or trace.c is
> would not be the right place for further enhancements that could
> happen on debug entries if you need to.
> 
> But the actual simple output that you are submitting along this tracer
> is already handled by the default output of the tracing internals.

What I wanted to get was a dmesg style output. The default output will
add pid and other information. That is why i did a print_line callback.
I also wanted to drop the header in the trace file.  I didn't find a way
to do that.

-aneesh