Thanks, Jan. I am pretty clear now. On 2021/10/18 19:36, Jan Kiszka wrote: > On 18.10.21 09:00, Zheng Chuan wrote: >> >> >> On 2021/10/18 13:41, Jan Kiszka wrote: >>> On 18.10.21 06:24, Zheng Chuan wrote: >>>> Hi, Jan. >>>> >>>> On 2021/10/14 17:34, Jan Kiszka wrote: >>>>> On 12.10.21 13:57, Zheng Chuan wrote: >>>>>> Hi all, >>>>>> >>>>>> After reading the code of jailhouse, I have a few questions. >>>>>> 1.Once Jailhouse is activated, the boot Linux will run as a VM. If it is >>>>>> possible to run VM by KVM(nested mode) in the root cell? >>>>>> >>>>> >>>>> Nope, but you can find traces of an attempt in >>>>> https://github.com/siemens/jailhouse/commits/wip/kvm >>>>> >>>> OK. I'll look into it. >>>> >>>>>> 2.I wonder why the root cell need to runs in guest mode? Whether >>>>>> jailhouse can run root cell in host mode, create KVM VM inside >>>>>> root cell. So jailhouse can support both partition and VM. >>>>> >>>>> Two main reasons: >>>>> >>>>> - security/safety: If you leave the root Linux running in host mode, it >>>>> becomes part of the trusted code base, increasing it by a "few" >>>>> orders of magnitude >>>>> - functionality: only by intercepting certain I/O accesses, Jailhouse >>>>> is able to emulate the ivshmem devices >>>>> >>>> OK, I got it >>>>>> >>>>>> 3.When create a non-root cell, the jailhouse driver executes cpu_down() >>>>>> to offline CPUs, but the offline CPU seems to be running >>>>>> and respond to interruptions. And the hypervisor seems to have done >>>>>> nothing to wake up the CPUs(assigned to non-root cell), >>>>>> or did I miss the important part of the code? >>>>> >>>>> The offlined and then "stolen" CPUs of the non-root cell are first >>>>> parked and then finally started again at (amost) architectural reset >>>>> state when doing "jailhouse cell start". On x86, you can see the SIPIs >>>>> being injected for that in the Jailhouse log. >>>>> >>>> Yes. However, I am still a little confused about the status of "offline" >>>> cpu. >>>> i. On x86, is it at real mode or protected mode after we do cpu_down() >>>> within vmlanuch? >>>> ii. is that different from cpu_down() of host? >>>> iii. on x86，does it conflict with the bootstrap of linux guest since the >>>> kernel will do transform from real mode to protected mode? >>> >>> We hand over the CPUs in real-mode, as architecturally defined. The only >>> differences are: >>> >>> - reset address can be configured to be different to 0xFFFFFFF0 >>> - all CPUs are started at once in a new cell, not just a single boot- >>> strap processor >>> >>> You can study that by looking at inmates/lib/x86/header-{32,64}.S. This >>> is the code run first after cell start when using our own inmates. >>> >>> Jan >>> >> OK, Let me list the initialization of non-root cell in my mind. >> 1. When we do cell_create, the cpus we wanted are "offline" by the root cell >> module. now the "offline" cpus are in real-mode with vmlanuch. > > More exactly, jailhouse module takes CPUs of upcoming non-root cell > offline (just cli; hlt, not yet real mode), but Linux could still bring > them up again at this point. > >> 2. The jailhouse sends nmi, and the wanted cpus do vm-exits into jailhouse >> (well, i am not sure can cpu response the nmi in real-mode?) > > NMIs are handled only in hypervisor mode (root mode). They serve as > hypervisor IPIs, take the guest into hypervisor mode, and let the > hypervisor do what needs to be done. > >> 3. The jailhouse will park them and point the rip into parking_code, and >> then the cpus will do hlt code after vm-entry > > Yep. From that point on, Linux will no longer be able to bring the CPUs > online again - INIT/SIPI for those cores will be rejected. > >> 4. When we do cell_start, the jailhouse sends nmi again, the wanted cpus do >> vm-exits into jailhouse. >> 5. The jailhouse will reset the rip into 0xFFFFFFF0, and the cpus will do >> vm-entry after arch_reset_cpu(). >> 6. ok, the /lib/x86/header-{32,64}.S turns on, and also the jailhouse will >> work while the inmates sends the sint and sipi. > > Those INIT/SIPI calls are actually emulated. Allowing the guest to issue > them in reality would bring the CPUs in the wrong state. > >> >> Is that right? >> > > Generally, yes! > > Jan >
-- Regards. Chuan -- You received this message because you are subscribed to the Google Groups "Jailhouse" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/jailhouse-dev/df8c5de1-1062-cf73-2060-601934a30e9f%40huawei.com.
