On Thu, Jan 14, 2010 at 4:05 PM, Spencer Oliver <s...@spen-soft.co.uk> wrote: > Øyvind Harboe wrote: >>> >>> I am a bit bemused by this. >>> GDB will have a good copy of the registers when it connects to openocd >>> anyway. >> >> Not necessarily true: >> >> target remote >> load >> monitor reg xxxx >> => GDB is out of sync >> stepi >> => gdb is in sync here. >> >> > > but surely the regs should be saved/restored by openocd during the gdb load?
The problem is that openocd or the target state can change behind GDB's back. E.g. if you power cycle the target while connected to GDB, or hotplug. There are *raft* of issues and I don't recall them all. Basically the assumption that OpenOCD, GDB and the target are in sync is seriously flawed on all levels. Especially for hardware type debugging like openocd. E.g. now connecting with to GDB is a no-operation. It does not affect the target state. This means that it is possible to connect to GDB and issue GDB monitor commands to bring the target into the "correct" state before GDB starts talking to it. The target could be running or halted during GDB connect. The stepi synchronizes GDB to the target stage, i.e. if the target was running while GDB was connected, stepi will not return until the target halts. I posted several messages on the gdb_sync issues, here's one: http://lists.berlios.de/pipermail/openocd-development/2009-September/010779.html -- Øyvind Harboe US toll free 1-866-980-3434 / International +47 51 63 25 00 http://www.zylin.com/zy1000.html ARM7 ARM9 ARM11 XScale Cortex JTAG debugger and flash programmer _______________________________________________ Openocd-development mailing list Openocd-development@lists.berlios.de https://lists.berlios.de/mailman/listinfo/openocd-development
