VPP/AArch64

From fd.io
< VPP
Revision as of 08:17, 15 February 2018 by GabrielGanne (Talk | contribs)

Jump to: navigation, search

Meeting Details

Weekly on Wednesdays, 6AM PT / 3PM CET / 7:30PM IST / 10PM CST. FD.io Zoom Meeting room

Join from PC, Mac, Linux, iOS or Android: https://zoom.us/j/5301185804 or https://zoom.us/my/fastdata

Or iPhone one-tap (US Toll):  +14086380968,,5301185804# or +16465588656,,5301185804#

Or Telephone:
    Dial: +1 408 638 0968 (US Toll) or +1 646 558 8656 (US Toll)
    +1 855 880 1246 (US Toll Free)
    +1 877 369 0926 (US Toll Free)
    Meeting ID: 530 118 5804
    International numbers available: https://zoom.us/zoomconference?m=ppBOQMQTVxGYmbxNsVemC6KNo8eX2ptF

IRC Channel

#fdio-arm on freenode.net

FD.io Lab

Platform Status
SoftIron OverDrive 1000 Arrived - One unit RMA'd. LF RT ticket open to add these machines as Jenkins slaves for CI build jobs.
Huawei TaiShan 2280 Shipping
Marvell MACCHIATObin Ordering
Cavium ThunderX Arriving

Build, unit test, packaging

The following is tracked manually until hardware is integrated into upstream FD.io CI
Cmd Status timing
make bootstrap OK 0m45
make build OK 11m45
make build-release OK 14m56
make test OK 33m40
make test-all KO (kubeproxy) 46m30
make test-debug OK 22m32
make test-all-debug KO (kubeproxy) 33m29

Status on commit: a332c46a51f9b4e13963340dfee1318e7513c124 (Mon Feb 5 21:45:28 2018 +0000)

kubeproxy tests are broken on purpose: corresponding features are not fully implemented

Timing consideration on platform: Hierofalcon with Cortex-A57 & Fedora 26

Distro Cmd Status
Fedora 26 (Server Edition) make pkg-rpm OK
Ubuntu 17.10 make pkg-deb OK
Ubuntu 16.04.3 LTS make pkg-deb OK

CSIT

https://wiki.fd.io/view/CSIT/AArch64

AArch64 Porting & Tuning

All: JIRA issues with ARM64 label

Assigned and New:

CSIT-922 aarch64 VM crash at startup
VPP-1130 Test with 64K pages Brian to check with Jeremy Linton
VPP-1129 Investigate enabling CLASSIFY_USE_SSE on Arm Gabriel Ganne
VPP-1126 Benchmark and optimize clib_memcpy64_x4() Khemendra Kumar
VPP-1114 Ensure correctness of atomics and memory ordering Kevin Wang
VPP-1103 Use correct CPU freq on ARM platforms Kevin Wang
VPP-1166 Add ILP32 support in VPP

Known Issues

Compilation may fail on systems with less than 1GB memory per core. One workaround is to search for -j in build-root/Makefile and multiply by 1 instead of 2.

GCC 5.3.x ICEs during FP register allocation. Please use GCC 5.4+.

Try disabling ASLR if experiencing random crashes: sysctl -w kernel.randomize_va_space=0

Recent Patches

replace CLASSIFY_USE_SSE with CLIB_HAVE_VEC128 VPP-1129 Gabriel Ganne
add CLIB_HAVE_VEC128 with NEON intrinsics Merged 02/08 VPP-1127 Gabriel Ganne
add NEON versions for ethernet_frame_is_tagged() Gabriel Ganne
vhost: Added ARMV8 NEON version of function map_guest_mem() VPP-1085 Nitin Saxena
vppinfra: use __atomic_fetch_add instead of __sync_fetch_and_add builtins VPP-1114 Kevin Wang
Arm system counter cleanup Merged 1/30 VPP-1125 Brian Brooks
Add top level make argument for custom cache line size VPP-1064 Nitin Saxena
svm: ... on autodetected VA space size (fixup again) Merged 01/10 Gabriel Ganne
svm: calc base address on AArch64 based on autodetected VA space size (fixup) Merged 01/10 Gabriel Ganne
svm: calc base address on AArch64 based on autodetected VA space size Merged 01/09 Damjan Marion
show cpu microarchitecture Merged 01/06 Gabriel Ganne
Fix Debian Packaging on AARCH64 Merged 01/06 Nitin Saxena
more extended tests fixes Merged 12/16 Gabriel Ganne
Use crc32 wrapper Merged 12/16 VPP-1086 Gabriel Ganne
implement clib_smp_pause() for arm and aarch64 platform Merged 12/15 VPP-1066 Kevin Wang
make "test-all" target pass again (for all platforms) Merged 12/13 Gabriel Ganne
fill "show cpu" Flag list on aarch64 platforms Merged 12/06 VPP-1065 Gabriel Ganne
remove smp dead code Merged 12/06 VPP-1066 Gabriel Ganne
net/virtio: support modern device id Merged 11/28 Gabriel Ganne
use REV on aarch64 for endianness swapping Merged 11/21 VPP-1067 Gabriel Ganne
armv8 crc32 - fix macro name Merged 11/15 Gabriel Ganne
bier - fix node table declaration Merged 11/14 Gabriel Ganne
Map SVM regions at a sane offset on arm64 Merged 11/10 Brian Brooks
bfd tests fix Merged 11/07 Gabriel Ganne
debian packaging fix Merged 11/06 Gabriel Ganne
lb test fix Merged 10/31 Gabriel Ganne
conditional x86intrin.h inclusion Merged 10/25 Gabriel Ganne
fix test_lb_ip4_gre6() cleanup Merged 10/24 Gabriel Ganne
null-terminate some formatted string Merged 10/20 Gabriel Ganne
lb plugin - fix format() type mismatches Merged 10/16 Gabriel Ganne
Use AESNI=y only on x86_64 machines Merged 10/14 Brian Brooks
Improved arm64 chip detection Merged 09/11 Brian Brooks
Native arm64 build: dpdk/Makefile change Merged 08/31 Brian Brooks

Meeting Minutes

2/14/2018

  • FD.io lab
    • Working on getting access to LF lab in order to setup OD1000 environment
    • Check with tykeal & zxiiro on trust policy for getting others access (Brian)
    • VEXXHOST
      • Mohammed says they do not have extra rack shelf - we need to send one for 3x MACCHIATObin
      • LF RT tickets: #52434 (ThunderX), #52435 (TaiShan2280), #52436 (MACCHIATObin)
  • VPP
    • Build, unit test, deb/rpm
      • 64B/128B cache line size - working on passing this configuration to rest of build system i.e. DPDK (Nitin)
      • RPi3 32-bit
        • Some parts of patch are 32-bit related, some RPi3 related
        • If there is justification, look into maintaining a 32-bit build on ARM
    • Porting & Tuning
      • If patches need to be tested on multiple Arm chips, please use DO_NOT_MERGE and Code Review -2
      • Two NEON related patches merged, working in progress on others, Nitin testing CLASSIFY_USE_SSE
  • CSIT
    • Please open JIRA ticket with details on VM crashing on startup. DONE: CSIT-922
    • Khem working on running VPP func tests on internal setup

2/7/2018

  • LF lab
    • OD1000 - last machine was racked; Vanessa needs credentials
    • Taishan2280 - machines arrived at Vexxhost; confirm with Rudy/Mohammed
    • ThunderX - machines arrived at Vexxhost; send board details to Mohammed
    • MACCHIATObin - boards arrived in Arm SJC waiting for enclosures (Andy)
  • Build, unit test, packaging
  • VPP
    • NEON usage in vhost - sent first patch for review (Nitin)
      • Need to verify how it performs on other Arm-based machines (Brian)
      • VPP maintainers prefer to use SIMD wrappers, but it might not always be possible
        • Cavium/Arm had to rewrite algorithm for AArch64 instead of use SIMD wrappers in DPDK
    • CLIB_HAVE_VEC128 - working on it (Gabriel)
    • Discussed compiler builtins for atomics in VPP call; need to spin another patch with wrappers based on architecture (Kevin)
    • Seeing prefetch hostspots on TX2+MlnxCX4en (similar to Armada8040) (Nitin)
  • CSIT
    • libvirt crashing on VM startup (Hierofalcon) (Gabriel)
      • Need someone who can reproduce this issue (Arm TBD)
    • Huawei also seeing VM issues (Khem)
    • buildroot doesn't work on Arm (Nitin)
      • Root issue: no support in GRUB for AArch64 in buildroot (?)
        • Need someone who can reproduce this issue (Arm TBD)
      • Peter Mikus replied to Nitin on csit-dev mail list
      • Using a temporary workaround: use a different VM image (Ubuntu Cloud) instead of one produced by buildroot
        • Working on patching DPDK in VM image (Ubuntu Cloud) just like done in buildroot
  • Misc
    • OpenFlow (Nitin, Damjan)
      • Is there an OpenFlow agent for VPP, and can VPP implement OpenFlow rules/tables?
      • VPP is not flow-based like OVS is; they are different
      • Can ODL/Honeycomb be used?

1/31/2018

  • LF lab
    • OD1000 - 1 replacement being installed this week
    • Huawei & Cavium boards should arrive at colo this week; confirm with Rudy
  • Build, unit test, packaging
    • Kubeproxy/NAT failures
      • Not arch related
      • Part of extended unit tests, so does not block CI
    • `make test` passes on D03 & D05 (Ubuntu)
  • MACCHIATObin
    • Seeing hotspots in VPP graph nodes
      • L3 forwarding - ip4 rewrite node
      • L2 cross-connect
      • Try reducing quad loop to a dual loop
      • dpdk-input node highly opt for x86 (could contribute to low perf) but hotspots still in rte_mbuf_t conversion(?)
    • Some examples of runtime code selection based on uarch exist in the codebase
  • CSIT
    • Adrian Oanca join from Enea
    • Gabriel seeing VM crashing during boot; related to # interfaces assigned (6)
    • Nitin ran into issue with buildroot on arm64; see thread on csit-dev

1/24/2018

  • VPP
    • DPDK issue with non-pci network cards
    • build & test status updated
    • VPP-1127 (VEC_128 enable) under discussion. Should we enable this by default ?
    • add Nitin to review Neon commits
    • VPP-1114 currently internal review
    • VPP-1064 under rework after review by Damjan
  • CSIT
    • first 3-nodes functional tests status list
    • TODO Gabriel: share CSIT VM setup env
    • nested VM: build-root package support for ARM. Create Jira ticket for Brian.

1/17/2018

  • Tina to send calendar invite for meeting
  • FD.io lab
    • Cavium shipping
  • VPP
    • Kubeproxy tests failing
    • Khem trying to find out the PCIe address for a given netdev interface
  • CSIT
    • Gabriel setting up 3 node topo with VMs
    • Gabriel working on PASS/FAIL status
  • CSIT 17.10 report

1/10/2018

  • Meeting moved 2 hours earlier - 6AM PT / 3PM CET / 7:30PM IST / 10PM CST
  • FD.io lab
    • Cavium ThunderX shipping soon
  • VPP
  • CSIT
    • Gabriel's patch for aarch64 support in CSIT merged
    • VirtualBox not supported on Arm / Vagrant unknown
      • This is OK for upstream since automation expects VMs to already exist
  • Performance
    • Need plan for 1T; use TaiShans that were sent to lab
  • AIs
    • Brian: Follow up with Vanessa and EdW regarding 'resource issue'
    • Gabriel: Update CSIT wiki page; which tests are passing/failing?
    • Brian: Check with Vanessa how to split machines between CI jobs and CSIT jobs

1/3/2018

  • FD.io lab
    • One OD1000 sent for RMA
    • Huawei PO sent out
    • Cavium PO sent out (?)
  • VPP
    • Gabriel working on patch for "show cpu" to display MIDR as human readable
    • Nitin sent preliminary patch for vhost-user NEON impl
      • Seeing perf differences on different cores; tradeoff is single-threaded perf vs. NEON
    • Kumar built and unit test successfully on D03
    • Nitin to resume patch for supporting different cache line sizes for the same arch
  • CSIT
    • Gabriel cleaned up WIP patch; ready for review
    • Kumar starting CSIT func tests with Ubuntu VMs
      • Scripts for running on dedicated hardware need to be modified, e.g. PCIe resources
    • Kumar to send doc on testing
  • Performance
    • Kumar to start thread on performance testing
  • AIs
    • Brian: Check with Tina on shipping and open LF RT ticket once they have arrived
    • Brian: Need a way to choose either SW or NEON impl based on chip
    • Gabriel: Create list of broken CSIT tests for 2-node topology

12/20/2017

No meeting next week - Dec 27

  • FD.io lab
    • OD1000s - build only
      • 1 of 3 needs to be RMAd
      • Can these be up in time to show 'make test' passes on ARM for 18.01 release report?
    • TaiShan
      • PO in progress
    • ThunderX - build only
      • PO went out
  • VPP
    • Patches / JIRAs
      • Patch for extended test failure, but still more (new) extended test failures - Gabriel
      • Nitin to post vhost-user.c changes for NEON
        • Nitin will finish Gabriel's original NEON patch to add CLIB_HAVE_VEC_128
    • Can we share code on Github e.g. NEON perf tests?
  • CSIT
    • Leading question: How many CSIT test cases are passing/failing?
    • Environment issues preventing running through all CSIT test cases; Gabriel needs dedicated machines or more RAM
    • Cavium & Huawei will join Gabriel in CSIT replication on ARM hardware next week
      • Cavium previously ran vhost test cases manually, now moving to CSIT

12/13/2017

  • VPP
    • Quick overview of work items
    • Waiting to hear back from LF about OD1000 connectivity
      • Changes needed to ci-mgmt
  • CSIT
    • Starting to reproduce CSIT on x86 and ARM (with Gabriel's WIP patch)
      • Some issues with environment variables (perf tests on 2-node)
    • Need Nexus to support aarch64 packages
      • Need a contact for Nexus
  • Share known issues on wiki!
  • Request CSIT 'deep dive'

12/06/2017

11/29/2017

  • VPP
    • vhost-user.c - SSE4.2 only. Implement range search using NEON. (nitin)
    • OD1000 status ?
      • build only
      • can we access them ?
      • what wan we do to help in general ?
    • x86 intrinsic review
    • build VPP on ARM VM on x86
  • CSIT
    • what platforms wil lbe made available

11/22/2017

  • VPP CI
    • 3 ThunderX for Chrismas
  • CSIT
  • Next steps
    • VPP
    • CSIT
      • structure work & send email (Gabriel)
      • is xxhash vs crc32 finished ? (Gabriel)
      • ask Maciek & setup a presentation meeting with someone from CSIT (Tina)
      • find a time to reschedule this meeting before the CSIT weekly call (Brian)

11/15/2017

  • VPP upstream status
    • build && build-release OK
    • "make test" && "make test-debug" OK
    • packaging:
      • Ubuntu 16.04 OK
      • Ubuntu 17.10 ? (TBC)
      • fedora-26 OK
  • vpp continuous test
    • all task required for jenkin's "verify" job are ready
    • TODO: request gerrit hook to Dave Barachs / vpp-dev (NB & GG)
    • set up ci in fdio lab
  • CSIT
    • setting up env
    • ThunderX platforms should arrive this week
    • csit work sharing

11/8/2017

  • Unit tests
    • Tests pass except for random initialization failures
    • Need to hear back from upstream about Extended unit tests
  • Should we run plugins such as NSH SFC?
  • Hardware to lab
    • Huawei h/w stalled
    • 3x ThunderX shipping to FD.io lab
  • CSIT replication
    • Cavium replicating on ThunderX2; getting started
  • Let's track our work in Jira; Brian to migrate tasks to Jira

10/25/2017

  • Gabriel working on vpp init failure in linux_pci_init()
  • Kumar to check with GeorgeZ on Huawei boards shipped to CSIT; need to verify tests also on this environment (package versions from distro)
  • Brian to check whether anything else needs to be done besides 'make test' for upstream enablement