On 16 April 2013 13:09, François Legal <de...@thom.fr.eu.org> wrote:
Ugh. Your mail client has completely mangled things (it's run all the lines of code into each other and it's still posting as multipart text+HTML). Please could you look at fixing its configuration -- it makes it hard to reply in response to individual things you've said. Anyway, you may be right about needing multiple qemutimers, I think I misunderstood that part. We set up one timer for each comparator, right? (ie it fires when the comparator would fire). In that case I think that is correct. You might like to try having each gTimerBlock have an ARMMPGTimerState* field, so you get at the non-banked parts of the control register with 'gtb->gtimer.gtimer_control' rather than via an anonymous uint32_t*. I think that would be clearer. Regarding gdb access to memory mapped registers causing a crash due to NULL cpu_single_env -- I think this is a general issue with the gdb support code. I suggest you drop those parts of your patch for now; we should look at fixing it in a coherent way separately and later (eg by having gdb memory accesses always look as if they are from CPU 0, or something). PS: I didn't mention it, but the struct names and so on should also change in line with my suggested new device name/filename. thanks -- PMM
