Hi, Gabe. Thank you for your reply. In my first implementation I put the handling function in the write() function which receives the data from the CPU, and wait for the handling to finish before returning the write() function. But It seems that it is not the right way in real hardware as the handling would take far more than 1 cycle thus the write() and thus the tick would be prolonged. That is not the true case in real hardware.
As you said, I can implement an event to get scheduled in the write() function. Do you mean that I should put the event in the mainEventQueue and just return? then the event would be serviced in the cycle I designated? Thank you very much! 在 2010年3月13日 上午2:37，Gabe Black <[email protected]>写道： > You could make your device actually process the data in the function > that receives it. If you want that to have latency or to process the > data gradually (although it sounds like you don't) you could use that > function to schedule an event or events for later. Conceptually you'd be > implementing an interrupt (or event, or callback, or ...) driven > architecture instead of a polling architecture, but external to the > device it would look the same. On top of being easier to implement, your > simulation should run faster. > > Gabe > > Veydan Wu wrote: > > I just want my I/O device to run in parallel with CPU. Does the CPUs > > run over and over? More specifically speaking, CPU may send something > > to the I/O device to handle, but the time is not fixed. So it is > > better to let the device keep running and when it receive data ( by > > the write() function ), it can handle it. I do want the handling > > invoked within the write() function, I want it to act like a real > > hardware device. Is there any way I can do this? Thank you very much! > > > > Weidan >
_______________________________________________ m5-users mailing list [email protected] http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
