>Li, Xin B wrote: >> I've just got x86_64 Linux 2.6.20.3 SMP guest boots on KVM >with 2 vCPUs >> (I'm using a x86_64 RHEL5 image but the RHEL5 kernel can't boot.), >> > >I found that an issue with x86-64 is that the mmu picks up an nx shadow
>page for a non-nx guest pte. This is unrelated to smp, but only shows >up there. It is easily fixed but meanwhile a workaround is to disable >nx in cpuid. > >I have a patch for userspace which does the following: > >- set up a kvm main loop outside vl.c and cpu-exec.c. >- disable signals on ap threads, so that timer and network signals are >only handled on vcpu 0 I prefer to have a separate thread in qemu to execute asynchronous logic, like the PIC/IOAPIC code. >- set up locking so that qemu code is never executed on more than one thread Each vCPU has a corresponding thread and can do synchronous IOs, like IN/OUT and MMIO to a device in DM, basically we don't need lock or serialize since guest OS device driver should guarantee device access won't happen on more than one CPU if that's not allowed. For a bad OS, it just kills its VM, or we may need to serialize accesses to the same device by using a device only lock. Totally we will have CPU# plus 1 thread. By this, we can get the best parallelism. >- hack qemu's apic so it can sipi and ipi other vcpus I prefer to keep LAPIC logic in KVM only, IOAPIC logic can be in Qemu side. ------------------------------------------------------------------------- This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/ _______________________________________________ kvm-devel mailing list kvm-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/kvm-devel
