Going Linux on Massive Multicore
summary
Port of Linux targets a Kalray MPPA256 256-core chip running at 400 MHz and about 200–230 GFLOPS; developers deliver a GCC port, GDB port, Buildroot userland and an instruction-set simulator and run Linux on I/O clusters while a custom runtime handles the compute matrix. Work includes adding MMU and tracing in mainline 3.10, implementing PCI boot and a shared-memory DMA protocol, debugging atomic operations and spinlocks (one-line swap bug fixed), and building Network-on-Chip device support.
AI-written summary · about
chapters (24)
- 0:00Start
- 2:26First MPPA-256 Chips with TSMC 28nm CMOS
- 3:29The MPPA-256 Processor (1)
- 4:04The MPPA-256 Processor (2)
- 4:43The MPPA-256 Processor Core ISA
- 6:16Kalray Software Development Kit for MPPA-256 (1)
- 7:39The MPPA POSIX Programming Model
- 8:34The MPPA & Linux
- 10:22Features
- 11:15Agenda
- 12:40Lessons Learned 1: Read the Specs Carefully
- 15:34Optimization Case: Atomic Ops
- 17:08Lessons Learned 2: Use the GCC
- 17:39Interrupt Controllers
- 19:00Memory Areas
- 20:07The PCI Express Driver (1)
- 21:47The PCI Express Driver (2)
- 22:43Boot
- 26:18The Ethernet Device Drivers
- 27:25Traces
- 28:41The Simulator
- 29:22What was Difficult
- 31:58What was Easy
- 33:09Future Plans
related talks
44mReal-Time Linux on Embedded Multicore Processors
47mLinux on POWER Driver Porting, Optimization, and Test
28mLinux on the Arm architecture and beyond
29mBuilding the Linux kernel with Clang
27mAccelerated Mainline Linux Development Ahead of SoC Availability
24mOpenRISC Update
54m3D Printing with Linux and Xenomai
44mLinux in the Cloud, on Prem, or... on a Mainframe?