Il 17/05/2012 04:28, Zhang, Yang Z ha scritto: > Changes in v6: > Rebase to latest QEMU > Fix a bug that fail to pass tests/rtc-test: > In previous version, it uses host time as the base point to calculate > guest RTC. It works when guest uses host based clock. But for vm and rt based > clock, it's wrong. Because guest's clock may not synchronous with host. In > current patch, it use rtc_clock as the reference point and successes to pass > the testing. > Adapt paolo's suggestion to use more reasonable code for migration.
Nice work! I've made my testcase even nastier and done the following changes: * simplified the code to remove the double update timer. The idea is that if you can latch the update-in-progress flag 244us before the timer expires in update_in_progress, instead of using a separate timer. * improved bisectability by squashing some patches and separating others. * changed the code to do all computation in nanoseconds * simplified the handling of the offset (and I think also made it more precise) * added another corner case: the SET bit does not stop interrupts, while divider reset does The result is at git://github.com/bonzini/qemu.git in branch rtc-intel. You're welcome to give it a further shake, but IMO it's good to go for 1.2! Paolo