k_pipe: fix trace point for blocking writes

[eriktamlin]

Fix the trace point in k_pipe wait_for.

Fixes #84486

It checks if "waitq" points to pipe->space, in which case it is a write operation.
A read will have "waitq" point to pipe->data instead.

I'm not entirely satisfied with this fix since it relies on comparing the address passed to wait_for in order to determine which trace point to use. Maybe add another "is_read" parameter to wait_for instead?

If anyone has a better way of doing this please let me know.

Signed-off-by: Erik Tamlin [email protected]

[eriktamlin]

@Mattemagikern suggested an alternative approach where we put the NO_WAIT check outside wait_for together with the trace points in the calling functions instead.

Example for k_pipe_write would be (note the changed parameter list for wait_for that now expects timeout instead of end):
timeout = sys_timepoint_timeout(end);
if (K_TIMEOUT_EQ(timeout, K_NO_WAIT)) {
rc = written ? written : -EAGAIN;
break;
}
SYS_PORT_TRACING_OBJ_FUNC_BLOCKING(k_pipe, write, pipe, timeout);
rc = wait_for(&pipe->space, pipe, &key, timeout, &need_resched);

Same changes in k_pipe_read.
Is this a preferred implementation?

[npitre]

What you did is fine. The alternative would add potential runtime overhead
to the code even when tracing is disabled.

Please merge both commits together though.