On 9/15/2023 10:16 AM, Steven Rostedt wrote:

On Fri, 15 Sep 2023 09:11:06 +0800
Linyu Yuan [off-list ref] wrote:

quoted

quoted

quoted

+		snprintf(__s, 9, "ep%d%s", te.address, \
+			(te.caps.dir_in && te.caps.dir_out) ? "" : \
+			te.caps.dir_in ? "in" : "out");

Note, there's a temp buffer trace_seq 'p' available for use as well. See
both include/trace/events/libata.h and include/trace/events/scsi.h:

    const char *libata_trace_parse_status(struct trace_seq*, unsigned char);
    #define __parse_status(s) libata_trace_parse_status(p, s)

I think that can be used instead of adding this TP_printk_init().

the reason add TP_printk_init() because when i first design some macro
which not

related to tracepoint,  it use too much stack.

Not sure what you mean about 'uses too much stack'. This is called by
the reading code and not some arbitrary location, and the above macros
are done in the same location as your "init" call, so I'm not sure how
that makes a difference on the stack.

quoted

but i think  TP_printk_init()  is good as it following most common way
to print.

I really do not want to add more versions of TRACE_EVENT() that I need
to maintain unless there is a really good reason to do so.

And I really don't want to encourage the use of a "TP_printk_init()"
because that just encourages more use cases that will make it hard for
user space to parse the TP_printk().

that's true, it is difficult to understand, when i add this new, it 
report build issue.


will consider other way for this case without new tracepoint macro.

-- Steve