|
From: Hollis B. <ho...@us...> - 2007-08-10 17:52:45
|
On Fri, 2007-08-10 at 08:28 -0700, Yoder Stuart-B08248 wrote: > > It's not mode related, but in BookE, TBU/TBL are User read / Supervisor > write. So we won't be able to trap/emulate reads to the timebase on > real Book E hardware if we think the timebase needs to be virtualized > for some reason. Don't think this is a big deal. Agreed. Practically speaking, it actually is pretty important that guests be able to tell how much real time has elapsed. In Linux's case this is where we replay ticks that would have triggered when the VCPU was de-scheduled. Of course, I have no idea how an RTOS would react to missing its decrementer/FIT interrupts. Hopefully we can mitigate that problem by setting appropriate host scheduling parameters in the first place... If we want to consider live migration in the future, the user-accessible timebase will raise some issues... especially if the two systems' timebase frequencies are not equal. -- Hollis Blanchard IBM Linux Technology Center |