On 30/04/11 10:12, Sylvain Munaut wrote:

Hi,

On Sat, Apr 30, 2011 at 9:53 AM, Wolfram Sang<wolfram@the-dreams.de> wrote:

...

It seems just to be a debugging aid, but brings in an unwanted calypso-dependency.

Well it is :)

I suggest you just replace it by a 'board_panic' call or something and leave it to the board specific code how to signal panic conditions.

I had another look and still have the impression that backlight usage should just go. Look at the implementation when the file was introduced: +panic: + while (1) { + bl_level(++i % 50); + delay_ms(50); + } + return NULL; This, I understand, it blinks. But it was changed three days later to the current state (c917fd43797c30385a1ba860fef95be97c84d51d) which broke it ('i' is constant now). So it is broken for over a year and nobody noticed :) So, instead of introducing a board_panic which nobody seems to need, I'd just suggest using plain osmo_panic as in other places. Much cleaner, too, IMO. Regards, Wolfram

Hi Wolfgang, On Sun, May 01, 2011 at 10:18:04AM +0200, Wolfram Sang wrote:

This, I understand, it blinks. But it was changed three days later to the current state (c917fd43797c30385a1ba860fef95be97c84d51d) which broke it ('i' is constant now). So it is broken for over a year and nobody noticed :)

Well, let's say we noticed that we get layer1 panics without any indication.

So, instead of introducing a board_panic which nobody seems to need, I'd just suggest using plain osmo_panic as in other places. Much cleaner, too, IMO.

I disagree. We desperately need board_panic, (with much more than just backlight blinking in the future), to be able to solve the various mysterious layer1/firmware crashes that we're seeing more or less all the time. I had recently discussed this with Andreas and Sylvain at the Easterhegg, and they both agreed that we need a way to debug crashes better. We need to tap into the various exception handlers, we need to use the debug unit to get backtraces, and we need ways like backlight blinking or buzzer sounds that can be done from a few lines of stack-independent assembly code to reliably indicate what has happened to the system. Regards, Harald -- - Harald Welte <laforge@gnumonks.org> http://laforge.gnumonks.org/ ============================================================================ "Privacy in residential applications is a desirable marketing option." (ETSI EN 300 175-7 Ch. A6)

On 02/05/11 14:02, Sylvain Munaut wrote:

...

I disagree. We desperately need board_panic, (with much more than just backlight blinking in the future), to be able to solve the various mysterious layer1/firmware crashes that we're seeing more or less all the time.

Yes, it definitely needed.

And calling osmo_panic might be the right thing, and then have the board register their osmo_panic handler in its init (the buzzer would be a good choice :)

Okay, understood that it is needed. Still, the discussion also showed that I seem to not have the proper knowledge of what is desired instead by those people hacking at layer 1 (what I don't do) :) Thus, may I suggest that I just remove the (static value) backlight setting in msgb.c to remove the calypso-dependency and leave a proper board_panic for someone else with a seperate patchset? To me, Harald's post indicates that it is a seperate topic? Regards, Wolfram