tugouxp <13824125...@163.com> writes: > Thank you! > > BTW, TCG also has two backends implemention, 1. Native host arch tranlation. > and 2. TCI byte code interpreter. The > difference is that the first one will tranlation the target arch code to host > arch instrtions and then executed in host cpu. but > the latter is a high level interperation. compare these two backends, Are > there any capability difference on the system > emulation about the two backends? thank you.
No - TCI is just a slower fallback to interpret TCG ops directly compared to executing JITed code. > > At 2024-06-24 16:57:40, "Alex Bennée" <alex.ben...@linaro.org> wrote: >>tugouxp <13824125...@163.com> writes: >> >>> Hello folks: >>> I have a puzzle on qemu major two emuation implentions on whole system >>> emulation, that is ,except the emuation speed, >>> did the TCG work mode has any weakness than "KVM" work mode on whole system >>> emulations(including kernek and >>> user-space)? >> >>It depends on the guest architecture. Some are more complete than >>others. The x86 emulation for example doesn't cover all the modern x86 >>extensions although there have been some improvements to its vector >>handling recently. >> >>One thing TCG can do that KVM can't is emulate code running at higher >>priority than kernel-mode. For example for ARM we can emulate the >>EL3/Root domain and secure and confidential realms. In KVM you can only >>run a guest EL1 kernel and its user space. >> >>> is there any work that kvm can do but TCG cant? >>> >>> of course kvm is much faster than TCG, but my question just about the >>> funtion, not care about speed. >>> >>> thanks for your kindly help! >>> BRs >>> zilong. >> >>-- >>Alex Bennée >>Virtualisation Tech Lead @ Linaro -- Alex Bennée Virtualisation Tech Lead @ Linaro
