Hi Tom, On Mon, 13 Oct 2025 at 15:15, Tom Rini <[email protected]> wrote: > > On Sat, Oct 11, 2025 at 08:19:53AM +0100, Simon Glass wrote: > > Hi Tom, > > > > On Fri, 10 Oct 2025 at 15:35, Tom Rini <[email protected]> wrote: > > > > > > On Fri, Oct 10, 2025 at 11:36:10AM +0100, Simon Glass wrote: > > > > Hi Tom, > > > > > > > > On Thu, 9 Oct 2025 at 18:30, Tom Rini <[email protected]> wrote: > > > > > > > > > > On Thu, Oct 09, 2025 at 03:29:54AM -0600, Simon Glass wrote: > > > > > > > > > > > This function is the core of the bootstd iteration. Add some > > > > > > debugging > > > > > > for the decisions it makes along the way, to make it easier to track > > > > > > what is going on. > > > > > > > > > > > > Signed-off-by: Simon Glass <[email protected]> > > > > > > --- > > > > > > > > > > > > (no changes since v1) > > > > > > > > > > > > boot/bootflow.c | 23 +++++++++++++++++++---- > > > > > > 1 file changed, 19 insertions(+), 4 deletions(-) > > > > > > > > > > > > diff --git a/boot/bootflow.c b/boot/bootflow.c > > > > > > index 78df09f369d..de69e27bec7 100644 > > > > > > --- a/boot/bootflow.c > > > > > > +++ b/boot/bootflow.c > > > > > > @@ -193,8 +193,10 @@ static int iter_incr(struct bootflow_iter > > > > > > *iter) > > > > > > log_debug("entry: err=%d\n", iter->err); > > > > > > global = iter->doing_global; > > > > > > > > > > > > - if (iter->err == BF_NO_MORE_DEVICES) > > > > > > + if (iter->err == BF_NO_MORE_DEVICES) { > > > > > > + log_debug("-> err: no more devices1\n"); > > > > > > return BF_NO_MORE_DEVICES; > > > > > > + } > > > > > > > > > > Thinking more about what I said in the previous iteration about git > > > > > blame history, ones like this should be log_msg_ret (the history on > > > > > "when did the test for == BF_NO_MORE_DEVICES come from is unchanged, > > > > > but > > > > > now you can have debug statements when enabled). > > > > > > > > Yes we can add that as well, but I still want to have the log_debug() > > > > as this doesn't require enabling CONFIG_LOG_ERROR_RETURN. That feature > > > > produces a lot of output even in normal operation since it shows > > > > errors dealt with by higher level code. It is really only designed to > > > > find the source of a particular error when you are stuck. > > > > > > > > > > > > > > [snip] > > > > > > @@ -228,11 +234,15 @@ static int iter_incr(struct bootflow_iter > > > > > > *iter) > > > > > > > > > > > > if (iter->err != BF_NO_MORE_PARTS) { > > > > > > /* ...select next partition */ > > > > > > - if (++iter->part <= iter->max_part) > > > > > > + if (++iter->part <= iter->max_part) { > > > > > > + log_debug("-> next partition %d max %d\n", > > > > > > iter->part, > > > > > > + iter->max_part); > > > > > > return 0; > > > > > > + } > > > > > > > > > > Shouldn't this be a debug message instead in the caller? > > > > > > > > I am trying to log_debug() every exit from this function...so you can > > > > see the entry and then which path it took. > > > > > > > > > > > > > > [snip] > > > > > > @@ -326,8 +336,13 @@ static int iter_incr(struct bootflow_iter > > > > > > *iter) > > > > > > } > > > > > > > > > > > > /* if there are no more bootdevs, give up */ > > > > > > - if (ret) > > > > > > + if (ret) { > > > > > > + log_debug("-> no more bootdevs\n"); > > > > > > return log_msg_ret("incr", BF_NO_MORE_DEVICES); > > > > > > + } > > > > > > > > > > Then do we actually need both a log_debug and a log_msg_ret? > > > > > > > > Please see above. > > > > > > I guess my question is, but why? Is this something that's going to be > > > debugged frequently? Why doesn't every function have meaningful text > > > string log_debug messages, just in case? And then why bother with > > > log_msg_ret at all? > > > > I have found myself in this function quite a few times, trying to work > > out what it is up to, so I decided to add more debugging. > > > > Anyway, would you like to just drop this patch, or something else? > > I'd like to fix the underlying problem so that we don't have a similar > discussion on your next series. As yes, I think in general all of these > patches to add more detailed logging on top of log_msg_ret logging are > too much. And I gather you're debugging these problems on real hardware > where using some source level debugger with sandbox isn't an option?
What is the underlying problem? The debugging was useful with the EFI app where it is sort-of possible to use a debugger, but is a bit painful. I ended up comparing logs to figure it out. Debuggers are not easy when the function being debugged is called dozens of times. I can imagine it being useful when someone reports a problem with the iteration...we could just turn on the option and the person could post a dump. Anyway, I'm happy to drop this patch. Regards, Simon
