GDB also had some stuff integrated for "Reverse Debugging". It started around the April '06 time frame. I don't have any links, this is just from memory. It supposedly allowed one to step a target "backward" in time. That's kinda neat and a useful use of trace information. I know there have been times when I have stepped and thought "Crap, what just happened" and wished I could unwind the state to a previous point and step forward again. I know this isn't always possible if you have messed with hardware, but still.
David Brownell wrote > On Wednesday 23 September 2009, Øyvind Harboe wrote: > >> I don't know much about the hardware trace capabilities, >> but I have understood that it's effectively a trace log of >> what happened. >> > > Various kinds of events yes, like branches, data access, > CPU state transitions and so on. The exact events vary > between chips; and there's a *LOT* of attention spent on > how to limit the events that get recorded. Without such > pre-filtering, there'd be too much data to use... > > > >> Wouldn't it be neat(even useful??) if it was possible to step >> through the last N instructions? >> > > Should be pretty trivial to configure: save *everything* into > a circular buffer, then dump it and ignore all but the last N > instruction events. > > > >> Here is what I have in mind: >> >> - Issue a "monitor trace 100" => we want to step through the >> last 100 instructions. >> >> - You can now issue stepi to step through each of the last >> 100 instructions. >> > > The way it's set up now is that you set things up, then > collect the data, and finally dump it (or save it). > There's no GDB integration. And I don't actually know > for sure that it all works, anyway. Yet! :) > > There are quite a few commercial products which support > tracing; there's stuff to be learned by seeing what they > do. I don't know that any integrate with their QEMU > analogues though ... the ones I know about seem to focus > on simulation for early development, or for providing > cycle-accurate timings in advance of tracing cycle counts > on the actual hardware. For performance work, that is. > > - Dave > > _______________________________________________ > Openocd-development mailing list > Openocd-development@lists.berlios.de > https://lists.berlios.de/mailman/listinfo/openocd-development > > _______________________________________________ Openocd-development mailing list Openocd-development@lists.berlios.de https://lists.berlios.de/mailman/listinfo/openocd-development
