AMD Broadcast TLB Invalidation Linux Patches Reworked In 4th Spin

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts
  • phoronix
    Administrator
    • Jan 2007
    • 67385

    AMD Broadcast TLB Invalidation Linux Patches Reworked In 4th Spin

    Phoronix: AMD Broadcast TLB Invalidation Linux Patches Reworked In 4th Spin

    A nice Christmas surprise for 2024 was Meta engineer Rik van Riel posting Linux kernel patches for making use of the AMD INVLPGB instruction found since Zen 3 processors for broadcast TLB invalidation...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite
  • fong38
    Senior Member
    • May 2022
    • 146

    #2
    Which branch are the patches based on? They don't cleanly apply on v6.13-rc7.

    EDIT: Nvm seems like the download mbox.gz option shuffles the order of the patches around...
    Last edited by fong38; 13 January 2025, 08:31 AM.

    Comment

    • npwx
      Senior Member
      • Mar 2022
      • 138

      #3
      I tested these patches on top of 6.12.7 with a few 64 core systems running heavily threaded go applications doing numerical optimization using worker threads on dedicated cores, tuned for latency of individual optimization jobs. I was hoping this is the kind of workload that might benefit from INVTLB. Unfortunately, there is no observable improvement. Latencies, throughput, alll remain unchanged.

      EDIT: actually it might make a difference. Another system using different batch sizes show some improvement, and profiling shows +20% in queued_spin_lock_slowpath(), which is the worker threads hitting on a futex once their workqueue is empty, so there seems to be a shift in where time is spent. Needs more thorough investigation.
      Last edited by npwx; 13 January 2025, 09:40 PM.

      Comment

      Working...
      X