ACRN Project Technical Community Meeting Minutes - 4/17/2019
ACRN Project TCM - 17th April 2019
One technical ACRN paper named “ACRN: A Big Littler Hypervisor for IoT Development” published in ACM/VEE’19 conference. Feel free to download from: https://conf.researchr.org/event/vee-2019/vee-2019-papers-parrot-a-big-little-hypervisor-for-iot-development .
Download foil from ACRN Presentation->ACRN_TCM->WW17’19
Q1: Is this CAT feature pre-configured before VM launch by ACRN?
A1: Yes. The CAT masks and which mask the UOS want to use, are already written in VM configuration source code; The HV use these info to set up the MSRs before launch any VM.
Q2: CAT feature is used by real time VM (RTVM), must the RTVM be bundled to a designated physical CPU?
A2: Yes. A RTVM is running on the same PCPU, during its entire lifecycle. RTVMs would not switch between different physical CPUs.
Q3: For CPU Sharing feature, does it support RTVM?
A3: No. The CPU sharing feature will not used by RTVM, will be used between SOS and UOS. We’ll have a PoC feature for CPU sharing by end of 2019.
Q4: Does CAT improve the RT performance by improving CPU cache’s instructions shooting accuracy?
A4: The CAT support allocating cache only for code or data, but we do not use this feature. IF we allocate a cache dedicating to a VM’s instructions, it should have some advantages. But that needs another cache mask, we may not have ample mask MSRs.
Q5: Does CAT feature apply only to Intel Atom platform ? In the case, the CAT is for L2 instead of L3. So what about the L3 ?
A5: CAT is an IA-32/IA-32e features. The APL platform support L2 CAT and Xoen support L3 CAT. The ACRN will choose the right CAT type. ACRN can use both of them.
Tel: +86-21-6116 7445
MP: +86-1364 1793 689