[RFC net-next 6/6] selftests: drv-net: measure devmem performance with kperf

[Stanislav Fomichev <sdf.kernel@gmail.com>]

Add devmem performance test using kperf and SystemMonitor.

Tested on two mlx5 hosts. Benchmark metric output:

  # # ktap-metric-policy-json: {"name":"throughput","regression":{"better":"higher","compare":"p50","relative_tolerance":0.05},"summarize":"distribution"}
  # # ktap-metric-policy-json: {"name":"tcp.retransmits","summarize":"max"}
  # # ktap-metric-json: {"host":"remote","name":"throughput","shape":"scalar","time":20.457972,"value":321.225}
  # # ktap-metric-json: {"host":"remote","name":"tcp.retransmits","shape":"scalar","time":20.457989,"value":903.0}
  # # ktap-metric-json: {"host":"local","name":"throughput","shape":"scalar","time":20.457993,"value":321.16499999999996}
  # # ktap-metric-json: {"host":"local","name":"tcp.retransmits","shape":"scalar","time":20.457998,"value":0.0}
  # ok 6 devmem.check_kperf
  # # Totals: pass:5 fail:0 xfail:0 xpass:0 skip:1 error:0

Note this is still CUDA (because 4K udmabuf needs my posted genpool
changes), will move to udmabuf once the genpoll stuff lands.

Signed-off-by: Stanislav Fomichev <sdf@fomichev.me>
---
 .../selftests/drivers/net/hw/devmem.py        | 70 ++++++++++++++++++-
 1 file changed, 67 insertions(+), 3 deletions(-)

diff --git a/tools/testing/selftests/drivers/net/hw/devmem.py b/tools/testing/selftests/drivers/net/hw/devmem.py
index 82c11ffc4add..4795b33e5efc 100755
--- a/tools/testing/selftests/drivers/net/hw/devmem.py
+++ b/tools/testing/selftests/drivers/net/hw/devmem.py
@@ -3,8 +3,9 @@
 
 from os import path
 from devmem_lib import (setup_test, run_rx, run_tx, run_tx_chunks, run_rx_hds,
-                        run_rx_large_niov)
-from lib.py import ksft_run, ksft_exit, ksft_disruptive
+                        run_rx_large_niov, require_devmem)
+from lib.py import (defer, EthtoolFamily, KperfRunner, ksft_run, SystemMonitor,
+                    ksft_exit, ksft_disruptive)
 from lib.py import NetDrvEpEnv
 
 
@@ -36,12 +37,75 @@ from lib.py import NetDrvEpEnv
     run_rx_large_niov(cfg)
 
 
+@ksft_disruptive
+def check_kperf(cfg) -> None:
+    """Measure CUDA-backed devmem RX performance with kperf."""
+    cfg.require_nsim(False)
+
+    kperf = KperfRunner(cfg)
+
+    dmabuf_size = 2048
+    # once udmabuf works well with 4K:
+    # dmabuf_size = 64
+    # if cmd("test -c /dev/udmabuf", fail=False).ret:
+    #     raise KsftSkipEx("Test requires /dev/udmabuf")
+    # size_limit = cmd("cat /sys/module/udmabuf/parameters/size_limit_mb",
+    #                  fail=False)
+    # if size_limit.ret == 0 and int(size_limit.stdout) < dmabuf_size:
+    #     raise KsftSkipEx(
+    #         f"Test requires a {dmabuf_size} MiB udmabuf size limit")
+
+    require_devmem(cfg)
+
+    ethnl = EthtoolFamily()
+    rings = ethnl.rings_get({"header": {"dev-index": cfg.ifindex}})
+    defer(ethnl.rings_set, {
+        "header": {"dev-index": cfg.ifindex},
+        "tcp-data-split": rings.get("tcp-data-split", "unknown"),
+        "rx": rings["rx"],
+    })
+    ethnl.rings_set({
+        "header": {"dev-index": cfg.ifindex},
+        "tcp-data-split": "enabled",
+        "rx": rings["rx-max"],
+    })
+
+    rss = ethnl.rss_get({"header": {"dev-index": cfg.ifindex}})
+    defer(ethnl.rss_set, {
+        "header": {"dev-index": cfg.ifindex},
+        "indir": rss["indir"],
+    })
+
+    devmem_args = [
+        "--time", "15",
+        "--num-connections", "4",
+        "--read-size", str(64 << 20),
+        "--write-size", str(64 << 20),
+        "--pin-off", "4",
+        "--num-rx-queues", "4",
+        "--dmabuf-rx-size-mb", str(dmabuf_size),
+        "--dmabuf-tx-size-mb", str(dmabuf_size),
+        "--validate", "no",
+        "--devmem-rx",
+        "--devmem-tx",
+        "--devmem-rx-memory", "cuda",
+        "--devmem-tx-memory", "cuda",
+        "--devmem-src-dev", "0000:4e:00",
+        "--devmem-dst-dev", "0000:4e:00",
+    ]
+
+    # The remote source uses CUDA TX and the local destination uses CUDA RX.
+    with kperf:
+        with SystemMonitor(cfg):
+            kperf.run(client_args=devmem_args)
+
+
 def main() -> None:
     """Run the devmem test cases."""
     with NetDrvEpEnv(__file__) as cfg:
         setup_test(cfg, path.abspath(path.dirname(__file__) + "/ncdevmem"))
         ksft_run([check_rx, check_tx, check_tx_chunks, check_rx_hds,
-                  check_rx_large_niov],
+                  check_rx_large_niov, check_kperf],
                  args=(cfg,))
     ksft_exit()
 
-- 
2.53.0-Meta