A Smart Port Card Tutorial --- Software John DeHart Washington University jdd@arl.wustl

A Smart Port Card Tutorial --- Software John DeHart Washington University jdd@arl.wustl.edu http://www.arl.wustl.edu/~jdd

Software Overview • No MSR details included in this talk • That will come in the MSR Tutorial • SPC Specific Kernel Modifications • Building a Kernel • config • compile • Loading Kernel onto SPC to execute it • download via APIC • Initiating Kernel execution • boot loader • Configuring Devices • APIC

SPC Kernel Modifications • Kernel Configuration File: • Memory Disk • Serial Console • Most device drivers unneeded (e.g no Ethernet…) • Source Code Changes: • PCI Device Configuration: ./dev/pci/if_apic_pci.c • APIC Specific configuration handled by driver • Change some APIC defaults: ./dev/ic/apic.c • made switch port the default • No HLT instruction: locore.s • Kernel Memory Page Allocation: • ./arch/i386/include/pmap.new.h • RTC: ./arch/i386/isa/clock.c, ./ufs/ffs/ffs_vfsops.c • We still had to remove some RTC code...

Steps to an Executing Kernel on an SPC • Configure Kernel • Start with a typical NetBSD kernel config file • Add SPC specific option: • options SPC • add memory disk (filesystem will reside in kernel!) • options MEMORY_DISK* • options MINIROOTSIZE • serial console port • options CONSDEVNAME • options CONS_OVERRIDE • remove “extra” devices (Ethernet, mouse, …)

Steps to an Executing Kernel on an SPC • Filesystem • we use a vnode disk device: vnd(4) • disk device in a regular file • pseudo-device vnd • must be included in your kernel where you are building the FS • Does NOT need to be included in your kernel for the SPC • build filesystem • dd(1): create the file • vnconfig(8) - configure vnode pseudo disk as a regular file • disklabel(8) - label it • newfs(8) - construct a new file system • mount(8)/umount(8): mount/unmount filesystem • mknod(8) - make device files • populate regular files from an existing system

… Steps to an Executing Kernel on an SPC • Compile Kernel • Make a symbol only version of kernel for /netbsd • ps, netstat, … use symbols from /netbsd • real kernel with file system CANNOT reside in itself! • Download kernel and File System separately • see following slides on AAL5download • Or, Download as one file: • Add File System to Kernel • mdsetimage(8) • Create an image version of Kernel • sets up bss section and symbol table • Kernel bss file is ready to download!

… Steps to an Executing Kernel on an SPC • Reset SPC • Download Boot Loader • via APIC control cells • loaded at physical memory location: 0x0 • Flip bit at location 0xFFE00 to start boot loader • configure NorthBridge (before kernel download!!) • cache, memory size, memory type, timing, etc • wait for kernel download to complete • looks for location 0x3FC to change from 0 to 1 • set up parameters on stack and jump to kernel • kernel located at 0x100000 • Download kernel bss file • Flip bit at location 0x3FC - kernel starts booting... This is the OLD WAY Please stop using download_APIC It is slow and painful

AAL5_Download • Replaces old download_APIC utility • Can “shotgun” download to <= 8 SPCs at once • Puts exactly same kernel on all 8 • This works fine for MSR where we want same kernel • May not be good for all uses • “Shotgun” number controlled from command line • Can download kernel and file system separately • no need to load file system into kernel before download • still need to load symbol version of kernel into file system before downloading • Takes about 1 minute for the whole download and boot process.

AAL5_Download: For SPC • Implemented and runs in Stages: • Stage1 uses APIC Control cells to: • reset the SPC(s) • download Stage2 to run on the SPC(s) • Stage2 (boot.out) runs on the SPC • runs SPC boot code to configure CPU and Northbridge • configures APIC on SPC to use AAL5 descriptors • sends pings back to CP, waiting for Stage3 • receives AAL5 frames from Stage3 and sends ACK/NACK • Stage3 runs on PC • syncs with Stage2 • downloads the kernel or FS via 36KB AAL5 frames • retransmits if any errors

AAL5_Download: For SPC-II • We are still working out the final details • We are starting to test the first batch of SPC-IIs • Implemented and runs in Stages: • Stage2 (boot.out) runs on the SPC-II • Already loaded on the IDE Disk on Module. • no need to download it • runs SPC boot code to configure CPU and Northbridge • configures APIC on SPC to use AAL5 descriptors • sends pings back to CP, waiting for Stage3 • receives AAL5 frames from Stage3 and sends ACK/NACK • Stage3 runs on PC • syncs with Stage2 • downloads the kernel or FS via 36KB AAL5 frames • retransmits if any errors

Simple View Stage1 PC SPC-I Stage1 VCI=0x321 APIC VCI=21 VCI=21 VCI=25 Stage3 VCI=23 Stage2 VCI=24

Control cells Reset cell AAL5 Frames Sequential View SPC Memory PC SPC 0x0 Boot and Stage2 Stage1 APIC RESET 0x0FFE00 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 Stage3 Stage2 Kernel and Filesystem 0x0100000 0xFFFFFFF0 Boot ROM

Source and Executables • source location: wu_arl/utilities/SPC/AAL5Download • binary location: • wu_arl/utilities/bin/$OSTYPE/Stage1 • wu_arl/utilities/bin/$OSTYPE/Stage3 • wu_arl/utilities/bin/$OSTYPE/AAL5_download.sh • wu_arl/utilities/bin/$OSTYPE/boot.out • wu_arl/utilities/bin/$OSTYPE/FindNM

Usage • Individual command line commands • ./Stage1 –shotgun 8 –svci 21 –rvci 21 –dvci 23 –uvci 24 –cvci 25 • ./Stage3 –shotgun 8 –native netbsd.MSR –both 24 • ./Stage3 –shotgun 8 –bss SPC24MB.fs –address 0x0196434 –both 24 -go • AAL5download.sh script • ./AAL5download 8 netbsd.MSR SCP24MB.fs 21 21 23 24 25 • Use from MSR Configuration Module • system(msrDownload.sh);

Special Usage Note on Linux • AAL5download.sh script: • needs to be able to determine the start location for the File System • NetBSD uses nm(1) • Linux nm(1) won’t work on a NetBSD binary • We have written a version of nm specifically for AAL5_download.sh: FindNM • FindNM is used automatically when AAL5_download.sh is run on Linux

AAL5_download.sh Usage • > ./AAL5_download.sh • Usage: ./AAL5_download.sh <netbsd_file> <filesystem_file> <start_vci> [<second_vci>] [<third_vci>][<fourth_vci>] [<fifth_vci>]

Stage1 Usage • > ./Stage1 • ./Stage1: [vpi/vci data] [load info] • ./Stage1: [-svpi #] (VPI for local sending of AAL5 control cells (0)) • ./Stage1: [-svci #] (VCI for local sending of AAL5 control cells) • ./Stage1: [-rvpi #] (VPI for local reading of AAL5 control cells (0)) • ./Stage1: [-rvci #] (VCI for local reading of AAL5 control cells) • ./Stage1: [-cvpi #] (VPI for remote CC writes of AAL5 control cells (0)) • ./Stage1: [-cvci #] (VCI for remote CC writes of AAL5 control cells) • ./Stage1: [-dvpi #] (VPI for local write of AAL5 data frames (0)) • ./Stage1: [-dvci #] (VCI for local write of AAL5 data frames) • ./Stage1: [-uvpi #] (VPI for local read of AAL5 data frames (0)) • ./Stage1: [-uvci #] (VCI for local read of AAL5 data frames) • ./Stage1: [-resp #] (Response port of AAL5 control cells) • ./Stage1: [-start #] (Initial VCI sequence (41, 42, 43, 44)) • ./Stage1: [-watch 0x#] (Monitor location for changes) • ./Stage1: [-apic #] (Linux: Apic Id (0 <= X <= 3)) • ./Stage1: [-abit #] (Alternating bit setting) • ./Stage1: [-file <path>] (Name of boot loader (boot.out)) • ./Stage1: [-swap] (use htonl() on output) • ./Stage1: [-shotgun #] (multicast loading 8X) • ./Stage1: [-msr #] (VCI for MSR ONLY AAL5 data frames) • ./Stage1: [-address #] (Physical data write address (0)) • ./Stage1: [-noreset] (do not write reset register)

Stage3 Usage • > ./Stage3 -help • [-switches] (-help) args • -bss <file> Load .bss format file • -native <file> Load native uncompressed file • -strip Remove Symbols from kernel • -debug print lots of stuff • -verbose Dump Elf headers • -elf Elf format execution • -rvpi <VPI> Load VPI number • -rvci <VCI> Load VCI number • -svpi <VPI> Load VCI number • -svci <VCI> Load VCI number • -both <VCI> Set both VCI numbers • -address <PhysAddr> Load file at <addr> • -start <PhysAddr> Start file at <addr> • -go Start kernel • -shotgun <count> Multicast loading 8X

WUGS Switch 41 0x0321 Port 1 Port 0 Port 7 Port 2 Port 6 PC Port 5 APIC Port 3 SPC/ Port 4 APIC ID Host APIC to SPC Control Connections Example 0/41 0/42 0/33 (0x21) 0/0x0321

APIC Driver at SPC Boot Time • PCI Configuration Space (sys/dev/pci/if_apic_pci.c): • Base memory address: • APIC Physical memory in SPC starts at 0xF0000000 • Memory mapping options: • Memory Enable • Master Enable • Write Invalidate Enable • IRQ • Driver normally would read this from PCI config space • APIC statically assigned IRQ 5 by System FPGA • For consistency, driver writes 5 to Interrupt Line Register

Device ID Vendor ID Status Command Class Code Revision ID BIST Header Type Latency Timer Cache Line Size Base Address Registers Cardbus CIS Pointer Subsystem ID Subsystem Vendor ID Expansion ROM Base Address Reserved Reserved Max_Lat Min_Gnt Interrupt Pin Interrupt Line PCI Configuration Space Configuration Space Header

Defining a Route on APIC Network Interface • NOTE: With the MSR you do not use ifconfig and atm_ifconfig • the MSR uses a different APIC driver then “plain” NetBSD for the SPC • Configure the APIC • ifconfig apic0 inet 192.168.10.1 netmask 0xffffff00 • network interface: apic0 • our IP address: 192.168.10.1 • Configure a VCI • atm_ifconfig apic0 0xc5 open -aal5 -llc -besteffort -maxqueuebufs 100 -inport FIBER -outports FIBER • network interface: apic0 • vci: 0xc5 • besteffort (vs. paced or lowdelay) • Tx buffer queue limit for this VCI: ‘maxqueuebufs 100’ • Configure the route • route add -iface 192.168.10.5 -link apic0:0.0.0.c5 • destination: 192.168.10.5 • network interface: apic0 • vci: 0xc5

atm_ifconfig • atm_ifconfig Usage: atm_ifconfig <interface> <vpivci> [open|close|modify|getall [options...]] Options are: -aal0 | -aal5 -llc | -nollc <protocol>: use LLC/SNAP? protocol is inet | ratm | value of ethernet type -loopback -inport <input_port_number> -outports <bitvector_of_output_ports> bitvector: 1 => port 0, 2 => port 1, 3 => both -lowdelay | -paced <rate> | -besteffort -maxqueuebufs <max_descriptors_in_tx_queue> <rate> is in Kbits/sec Changes HERE -inport and -outport values have changed.

6 6 5 5 6 5 IPP IPP IPP IPP OPP OPP OPP OPP IPP OPP Pentium Cache SPC SPC SPC SPC TI TI TI TI 3 3 3 6 6 6 APIC Configuration ControlProcessor Switch Fabric Sys.FPGA 32-64 MB APIC NorthBridge . . . SPC TI

IPP OPP Pentium Cache Defining APIC Ports for Routing atm_ifconfig apic0 0xca open -aal5 -llc -besteffort -inport $INPORT -outports $OUTPORTs Switch Sys.FPGA 32-64 MB INPORT OUTPORTs LINK LINK APIC NorthBridge TI Link

IPP OPP Pentium Cache Defining APIC Ports for Routing Switch Sys.FPGA 32-64 MB INPORT OUTPORTs LINK SWITCH APIC NorthBridge TI Link atm_ifconfig apic0 0xca open -aal5 -llc -besteffort -inport $INPORT -outports $OUTPORTs

IPP OPP Pentium Cache Defining APIC Ports for Routing Switch INPORT OUTPORTs SWITCH LINK Sys.FPGA 32-64 MB APIC NorthBridge TI Link atm_ifconfig apic0 0xca open -aal5 -llc -besteffort -inport $INPORT -outports $OUTPORTs

IPP OPP Pentium Cache Defining APIC Ports for Routing Switch INPORT OUTPORTs SWITCH SWITCH Sys.FPGA 32-64 MB APIC NorthBridge TI Link atm_ifconfig apic0 0xca open -aal5 -llc -besteffort -inport $INPORT -outports $OUTPORTs

IPP OPP Pentium Cache Defining APIC Ports for Routing Switch INPORT OUTPORTs LINK BOTH Sys.FPGA 32-64 MB APIC NorthBridge TI Link atm_ifconfig apic0 0xca open -aal5 -llc -besteffort -inport $INPORT -outports $OUTPORTs

IPP OPP Pentium Cache Defining APIC Ports for Routing Switch INPORT OUTPORTs SWITCH BOTH Sys.FPGA 32-64 MB APIC NorthBridge TI Link atm_ifconfig apic0 0xca open -aal5 -llc -besteffort -inport $INPORT -outports $OUTPORTs

Caveats • You cannot change the SPC-I IRQ assignments: • They are fixed in the System FPGA • And there should be no reason to change them • HLT instruction in locore.s commented out • remnant of when we were having INTEL problems • we can leave it this way • gives a slight performance improvement • SPC needs to be reset before EACH download

A Smart Port Card Tutorial --- Software John DeHart Washington University jdd@arl.wustl