Scaling Data Analytics Via Confidential Computing on RISC-V
summary
Defines Supervisor Domains as a new privileged context with SDI cache tagging and MPT-based physical memory donation that prevents host access; maps accelerator DMA via IMU using a T-bit and an MPD checker so devices use trusted or untrusted page tables; relies on SPDM, IDE and TDIS for device authentication and link protection; provides a QEMU prototype and a partly implemented TSM, pauses then resumes the COVE ABI work, and leverages OpenTitan root of trust with plans to upstream patches to Linux KVM.
AI-written summary · about
chapters (9)
- 0:00Introduction to talk
- 1:22What is confidential computing
- 4:51Why confidential AI matters
- 8:19RISC-V ISA building blocks
- 12:57Accelerator security requirements
- 16:26Software APIs and standards
- 17:50Root of trust with OpenTitan
- 18:32Current status and next steps
- 20:39Future standardization efforts
related talks
52mEnabling New Security Frontiers: Deep-Dive Into Implementing Confidential Computing on RISC-V
23mTowards Scalable Confidential Computing on RISC-V
01h 05mConfidential Computing for RISC-V-based Platforms
31mConfidential Computing for RISC-V-based Platforms
18mIntroducing RISC-V Confidential Computing for IoT Devices
10mUpdate on RISC-V Confidential VM Extension (CoVE)
29mSecure and Dynamic Hardware Partitioning Management on Heterogeneous SoC
36mEverybody is talking about Confidential Computing - the minimum Devs should know