Skip to contentSkip to conversation

selftests: drv-net: driver tests for HW GRO

PATCH Applied Jakub Kicinski 2 participants 8 messages Mar 18 – Mar 20

Subsystems (3)
lore
Share & tools
lore

Jakub Kicinski

author maintainer
#0 lore raw

v4 cover selftests: drv-net: driver tests for HW GRO

Add tests for HW GRO stats, packet ordering and depth.

The ynltool and bnxt patches from v2 were applied separately.

v4:
- [patch 6] rename [FAIL] -> [MISS]; NIPA thinks every [FAIL] means test failed, 'cause bash test don't ktap :/
v3: https://lore.kernel.org/[email protected]
- [patch 3] add a note about fq being needed
v2: https://lore.kernel.org/[email protected]
- [patch 3] do not enable SO_TXTIME for multi-flow tests
- [patch 6] move the test to cover SW GRO
v1: https://lore.kernel.org/[email protected]

Jakub Kicinski (6):
selftests: net: move gro to lib for HW vs SW reuse
selftests: drv-net: give HW stats sync time extra 25% of margin
selftests: drv-net: gro: use SO_TXTIME to schedule packets together
selftests: drv-net: gro: test GRO stats
selftests: drv-net: gro: add test for packet ordering
selftests: drv-net: gro: add a test for GRO depth

 tools/testing/selftests/drivers/net/Makefile  |   1 -
 .../testing/selftests/drivers/net/hw/Makefile |   1 +
 tools/testing/selftests/net/lib/Makefile      |   1 +
 .../selftests/{drivers/net => net/lib}/gro.c  | 252 ++++++++++++++-
 .../testing/selftests/drivers/net/.gitignore  |   1 -
 tools/testing/selftests/drivers/net/gro.py    | 203 ++++++++++--
 .../selftests/drivers/net/hw/gro_hw.py        | 294 ++++++++++++++++++
 .../selftests/drivers/net/lib/py/env.py       |   4 +-
 tools/testing/selftests/net/lib/.gitignore    |   1 +
 9 files changed, 728 insertions(+), 30 deletions(-)
 rename tools/testing/selftests/{drivers/net => net/lib}/gro.c (86%)
 create mode 100755 tools/testing/selftests/drivers/net/hw/gro_hw.py
Signature · 1 line
2.53.0
author maintainer
#1 lore raw

PATCH v4 1/6 selftests: net: move gro to lib for HW vs SW reuse

The gro.c packet sender is used for SW testing but bulk of incoming new tests will be HW-specific. So it's better to put them under drivers/net/hw/, to avoid tip-toeing around netdevsim. Move gro.c to lib so we can reuse it.

---
 tools/testing/selftests/drivers/net/Makefile           | 1 -
 tools/testing/selftests/net/lib/Makefile               | 1 +
 tools/testing/selftests/{drivers/net => net/lib}/gro.c | 2 +-
 tools/testing/selftests/drivers/net/.gitignore         | 1 -
 tools/testing/selftests/drivers/net/gro.py             | 2 +-
 tools/testing/selftests/net/lib/.gitignore             | 1 +
 6 files changed, 4 insertions(+), 4 deletions(-)
 rename tools/testing/selftests/{drivers/net => net/lib}/gro.c (99%)
tools/testing/selftests/drivers/net/Makefile +0 -1
index 8154d6d429d3..7c7fa75b80c2 100644
@@ -6,7 +6,6 @@ TEST_INCLUDES := $(wildcard lib/py/*.py) \
 		 ../../net/lib.sh \
 
 TEST_GEN_FILES := \
-	gro \
 	napi_id_helper \
 # end of TEST_GEN_FILES
 
tools/testing/selftests/net/lib/Makefile +1 -0
index 5339f56329e1..ff83603397d0 100644
@@ -14,6 +14,7 @@ TEST_FILES := \
 TEST_GEN_FILES := \
 	$(patsubst %.c,%.o,$(wildcard *.bpf.c)) \
 	csum \
+	gro \
 	xdp_helper \
 # end of TEST_GEN_FILES
 
tools/testing/selftests/net/lib/gro.c renamed from tools/testing/selftests/drivers/net/gro.c +1 -1
similarity index 99%
rename from tools/testing/selftests/drivers/net/gro.c
rename to tools/testing/selftests/net/lib/gro.c
index 3c0745b68bfa..02e29509fbea 100644
@@ -77,7 +77,7 @@
 #include <unistd.h>
 
 #include "kselftest.h"
-#include "../../net/lib/ksft.h"
+#include "ksft.h"
 
 #define DPORT 8000
 #define SPORT 1500
tools/testing/selftests/drivers/net/.gitignore +0 -1
index 3633c7a3ed65..585ecb4d5dc4 100644
@@ -1,4 +1,3 @@
 # SPDX-License-Identifier: GPL-2.0-only
-gro
 napi_id_helper
 psp_responder
tools/testing/selftests/drivers/net/gro.py +1 -1
index cbc1b19dbc91..2da53686354f 100755
@@ -117,7 +117,7 @@ from lib.py import ksft_variants
     """ Setup hardware loopback mode for GRO testing. """
 
     if not hasattr(cfg, "bin_remote"):
-        cfg.bin_local = cfg.test_dir / "gro"
+        cfg.bin_local = cfg.net_lib_dir / "gro"
         cfg.bin_remote = cfg.remote.deploy(cfg.bin_local)
 
     if not hasattr(cfg, "feat"):
tools/testing/selftests/net/lib/.gitignore +1 -0
index bbc97d6bf556..6cd2b762af5d 100644
@@ -1,3 +1,4 @@
 # SPDX-License-Identifier: GPL-2.0-only
 csum
+gro
 xdp_helper
Signature · 1 line
2.53.0
author maintainer
#2 lore raw

PATCH v4 2/6 selftests: drv-net: give HW stats sync time extra 25% of margin

In reply to Jakub Kicinski

There are transient failures for devices which update stats periodically, especially if it's the FW DMA'ing the stats rather than host periodic work querying the FW. Wait 25% longer than strictly necessary.

For devices which don't report stats-block-usecs we retain 25 msec as the default wait time (0.025sec == 20,000usec * 1.25).

---
 tools/testing/selftests/drivers/net/lib/py/env.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
tools/testing/selftests/drivers/net/lib/py/env.py +2 -2
index ccff345fe1c1..6a71c7e7f136 100644
@@ -288,8 +288,8 @@ from . import bpftool, RtnlFamily, Netlink
                 if "Operation not supported" not in e.cmd.stderr:
                     raise
 
-            self._stats_settle_time = 0.025 + \
-                data.get('stats-block-usecs', 0) / 1000 / 1000
+            self._stats_settle_time = \
+                1.25 * data.get('stats-block-usecs', 20000) / 1000 / 1000
 
         time.sleep(self._stats_settle_time)
 
Signature · 1 line
2.53.0
author maintainer
#3 lore raw

PATCH v4 3/6 selftests: drv-net: gro: use SO_TXTIME to schedule packets together

In reply to Jakub Kicinski

Longer packet sequence tests are quite flaky when the test is run over a real network. Try to avoid at least the jitter on the sender side by scheduling all the packets to be sent at once using SO_TXTIME. Use hardcoded tx time of 5msec in the future. In my test increasing this time past 2msec makes no difference so 5msec is plenty of margin. Since we now expect more output buffering make sure to raise SNDBUF.

Note that this is an opportunistic reliability improvement which will only work if the qdisc can schedule Tx time for us (fq). Fiddling with qdisc config was deemed too complex, so it's not part of the patch.

---
v3:
- update the commit msg
v2: https://lore.kernel.org/[email protected]
- only enable TXTIME when we are sending a single flow
v1: https://lore.kernel.org/[email protected]

---
 tools/testing/selftests/net/lib/gro.c | 57 +++++++++++++++++++++++++--
 1 file changed, 54 insertions(+), 3 deletions(-)
tools/testing/selftests/net/lib/gro.c +54 -3
index 02e29509fbea..5b3f8fca08e6 100644
@@ -63,6 +63,7 @@
 #include <linux/filter.h>
 #include <linux/if_packet.h>
 #include <linux/ipv6.h>
+#include <linux/net_tstamp.h>
 #include <net/ethernet.h>
 #include <net/if.h>
 #include <netinet/in.h>
@@ -74,6 +75,7 @@
 #include <stdio.h>
 #include <stdarg.h>
 #include <string.h>
+#include <time.h>
 #include <unistd.h>
 
 #include "kselftest.h"
@@ -123,6 +125,9 @@ static int tcp_offset = -1;
 static int total_hdr_len = -1;
 static int ethhdr_proto = -1;
 static bool ipip;
+static uint64_t txtime_ns;
+
+#define TXTIME_DELAY_MS 5
 
 static void vlog(const char *fmt, ...)
 {
@@ -330,13 +335,37 @@ static void fill_transportlayer(void *buf, int seq_offset, int ack_offset,
 
 static void write_packet(int fd, char *buf, int len, struct sockaddr_ll *daddr)
 {
+	char control[CMSG_SPACE(sizeof(uint64_t))];
+	struct msghdr msg = {};
+	struct iovec iov = {};
+	struct cmsghdr *cm;
 	int ret = -1;
 
-	ret = sendto(fd, buf, len, 0, (struct sockaddr *)daddr, sizeof(*daddr));
+	iov.iov_base = buf;
+	iov.iov_len = len;
+
+	msg.msg_iov = &iov;
+	msg.msg_iovlen = 1;
+	msg.msg_name = daddr;
+	msg.msg_namelen = sizeof(*daddr);
+
+	if (txtime_ns) {
+		memset(control, 0, sizeof(control));
+		msg.msg_control = control;
+		msg.msg_controllen = sizeof(control);
+
+		cm = CMSG_FIRSTHDR(&msg);
+		cm->cmsg_level = SOL_SOCKET;
+		cm->cmsg_type = SCM_TXTIME;
+		cm->cmsg_len = CMSG_LEN(sizeof(uint64_t));
+		memcpy(CMSG_DATA(cm), &txtime_ns, sizeof(txtime_ns));
+	}
+
+	ret = sendmsg(fd, &msg, 0);
 	if (ret == -1)
-		error(1, errno, "sendto failure");
+		error(1, errno, "sendmsg failure");
 	if (ret != len)
-		error(1, errno, "sendto wrong length");
+		error(1, 0, "sendmsg wrong length: %d vs %d", ret, len);
 }
 
 static void create_packet(void *buf, int seq_offset, int ack_offset,
@@ -1058,6 +1087,7 @@ static void check_recv_pkts(int fd, int *correct_payload,
 
 static void gro_sender(void)
 {
+	int bufsize = 4 * 1024 * 1024; /* 4 MB */
 	const int fin_delay_us = 100 * 1000;
 	static char fin_pkt[MAX_HDR_LEN];
 	struct sockaddr_ll daddr = {};
@@ -1067,6 +1097,27 @@ static void gro_sender(void)
 	if (txfd < 0)
 		error(1, errno, "socket creation");
 
+	if (setsockopt(txfd, SOL_SOCKET, SO_SNDBUF, &bufsize, sizeof(bufsize)))
+		error(1, errno, "cannot set sndbuf size, setsockopt failed");
+
+	/* Enable SO_TXTIME unless test case generates more than one flow
+	 * SO_TXTIME could result in qdisc layer sorting the packets at sender.
+	 */
+	if (true) {
+		struct sock_txtime so_txtime = { .clockid = CLOCK_MONOTONIC, };
+		struct timespec ts;
+
+		if (setsockopt(txfd, SOL_SOCKET, SO_TXTIME,
+			       &so_txtime, sizeof(so_txtime)))
+			error(1, errno, "setsockopt SO_TXTIME");
+
+		if (clock_gettime(CLOCK_MONOTONIC, &ts))
+			error(1, errno, "clock_gettime");
+
+		txtime_ns = ts.tv_sec * 1000000000ULL + ts.tv_nsec;
+		txtime_ns += TXTIME_DELAY_MS * 1000000ULL;
+	}
+
 	memset(&daddr, 0, sizeof(daddr));
 	daddr.sll_ifindex = if_nametoindex(ifname);
 	if (daddr.sll_ifindex == 0)
Signature · 1 line
2.53.0
author maintainer
#4 lore raw

PATCH v4 4/6 selftests: drv-net: gro: test GRO stats

In reply to Jakub Kicinski

Test accuracy of GRO stats. We want to cover two potentially tricky cases:
- single segment GRO
- packets which were eligible but didn't get GRO'd

The first case is trivial, teach gro.c to send one packet, and check GRO stats didn't move.

Second case requires gro.c to send a lot of flows expecting the NIC to run out of GRO flow capacity.

To avoid system traffic noise we steer the packets to a dedicated queue and operate on qstat.

---
 .../testing/selftests/drivers/net/hw/Makefile |   1 +
 tools/testing/selftests/net/lib/gro.c         | 163 ++++++++++-
 .../selftests/drivers/net/hw/gro_hw.py        | 271 ++++++++++++++++++
 3 files changed, 433 insertions(+), 2 deletions(-)
 create mode 100755 tools/testing/selftests/drivers/net/hw/gro_hw.py
tools/testing/selftests/drivers/net/hw/Makefile +1 -0
index 91df028abfc0..3c97dac9baaa 100644
@@ -26,6 +26,7 @@ TEST_PROGS = \
 	ethtool_extended_state.sh \
 	ethtool_mm.sh \
 	ethtool_rmon.sh \
+	gro_hw.py \
 	hw_stats_l3.sh \
 	hw_stats_l3_gre.sh \
 	iou-zcrx.py \
tools/testing/selftests/net/lib/gro.c +161 -2

223 lines not on this page · raw message

tools/testing/selftests/drivers/net/hw/gro_hw.py new file +271 -0

271 lines not on this page · raw message

Signature · 1 line
2.53.0
author maintainer
#5 lore raw

PATCH v4 5/6 selftests: drv-net: gro: add test for packet ordering

In reply to Jakub Kicinski

Add a test to check if the NIC reorders packets if the hit GRO.

---
 tools/testing/selftests/net/lib/gro.c         | 38 +++++++++++++++++--
 .../selftests/drivers/net/hw/gro_hw.py        | 29 ++++++++++++--
 2 files changed, 61 insertions(+), 6 deletions(-)
tools/testing/selftests/net/lib/gro.c +35 -3
index 41794b9f6f8a..3e611ae25f61 100644
@@ -131,6 +131,7 @@ static int ethhdr_proto = -1;
 static bool ipip;
 static uint64_t txtime_ns;
 static int num_flows = 4;
+static bool order_check;
 
 #define CAPACITY_PAYLOAD_LEN 200
 
@@ -1136,6 +1137,7 @@ static void check_capacity_pkts(int fd)
 	static char buffer[IP_MAXPACKET + ETH_HLEN + 1];
 	struct iphdr *iph = (struct iphdr *)(buffer + ETH_HLEN);
 	struct ipv6hdr *ip6h = (struct ipv6hdr *)(buffer + ETH_HLEN);
+	int num_pkt = 0, num_coal = 0, pkt_idx;
 	const char *fail_reason = NULL;
 	int flow_order[num_flows * 2];
 	int coalesced[num_flows];
@@ -1144,8 +1146,6 @@ static void check_capacity_pkts(int fd)
 	int total_data = 0;
 	int pkt_size = -1;
 	int data_len = 0;
-	int num_pkt = 0;
-	int num_coal = 0;
 	int flow_id;
 	int sport;
 
@@ -1203,6 +1203,34 @@ static void check_capacity_pkts(int fd)
 		total_data += data_len;
 	}
 
+	/* Check flow ordering. We expect to see all non-coalesced first segs
+	 * then interleaved coalesced and non-coalesced second frames.
+	 */
+	pkt_idx = 0;
+	for (flow_id = 0; order_check && flow_id < num_flows; flow_id++) {
+		bool coaled = coalesced[flow_id] > CAPACITY_PAYLOAD_LEN;
+
+		if (coaled)
+			continue;
+
+		if (flow_order[pkt_idx] != flow_id) {
+			vlog("Flow order mismatch (non-coalesced) at position %d: expected flow %d, got flow %d\n",
+			     pkt_idx, flow_id, flow_order[pkt_idx]);
+			fail_reason = fail_reason ?: "bad packet order (1)";
+		}
+		pkt_idx++;
+	}
+	for (flow_id = 0; order_check && flow_id < num_flows; flow_id++) {
+		bool coaled = coalesced[flow_id] > CAPACITY_PAYLOAD_LEN;
+
+		if (flow_order[pkt_idx] != flow_id) {
+			vlog("Flow order mismatch at position %d: expected flow %d, got flow %d, coalesced: %d\n",
+			     pkt_idx, flow_id, flow_order[pkt_idx], coaled);
+			fail_reason = fail_reason ?: "bad packet order (2)";
+		}
+		pkt_idx++;
+	}
+
 	if (!fail_reason) {
 		vlog("All %d flows coalesced correctly\n", num_flows);
 		printf("Test succeeded\n\n");
@@ -1622,12 +1650,13 @@ static void parse_args(int argc, char **argv)
 		{ "saddr", required_argument, NULL, 's' },
 		{ "smac", required_argument, NULL, 'S' },
 		{ "test", required_argument, NULL, 't' },
+		{ "order-check", no_argument, NULL, 'o' },
 		{ "verbose", no_argument, NULL, 'v' },
 		{ 0, 0, 0, 0 }
 	};
 	int c;
 
-	while ((c = getopt_long(argc, argv, "46d:D:ei:n:rs:S:t:v", opts, NULL)) != -1) {
+	while ((c = getopt_long(argc, argv, "46d:D:ei:n:rs:S:t:ov", opts, NULL)) != -1) {
 		switch (c) {
 		case '4':
 			proto = PF_INET;
@@ -1666,6 +1695,9 @@ static void parse_args(int argc, char **argv)
 		case 't':
 			testname = optarg;
 			break;
+		case 'o':
+			order_check = true;
+			break;
 		case 'v':
 			verbose = true;
 			break;
tools/testing/selftests/drivers/net/hw/gro_hw.py +26 -3
index 3bca19e8f339..10e08b22ee0e 100755
@@ -10,7 +10,7 @@ import glob
 import re
 
 from lib.py import ksft_run, ksft_exit, ksft_pr
-from lib.py import ksft_eq, ksft_ge
+from lib.py import ksft_eq, ksft_ge, ksft_variants
 from lib.py import NetDrvEpEnv, NetdevFamily
 from lib.py import KsftSkipEx
 from lib.py import bkg, cmd, defer, ethtool, ip
@@ -78,7 +78,8 @@ GRO_DPORT = 8000
     return test_queue
 
 
-def _run_gro_test(cfg, test_name, num_flows=None, ignore_fail=False):
+def _run_gro_test(cfg, test_name, num_flows=None, ignore_fail=False,
+                  order_check=False):
     """Run gro binary with given test and return output."""
     if not hasattr(cfg, "bin_remote"):
         cfg.bin_local = cfg.net_lib_dir / "gro"
@@ -98,6 +99,8 @@ GRO_DPORT = 8000
     ]
     if num_flows:
         base_args.append(f"--num-flows {num_flows}")
+    if order_check:
+        base_args.append("--order-check")
 
     args = " ".join(base_args)
 
@@ -257,13 +260,33 @@ def _check_gro_stats(cfg, test_queue, stats_before,
                      expect_wire=gro_coalesced * 2)
 
 
+@ksft_variants([4, 32, 512])
+def test_gro_order(cfg, num_flows):
+    """
+    Test that HW GRO preserves packet ordering between flows.
+
+    Packets may get delayed until the aggregate is released,
+    but reordering between aggregates and packet terminating
+    the aggregate and normal packets should not happen.
+
+    Note that this test is stricter than truly required.
+    Reordering packets between flows should not cause issues.
+    This test will also fail if traffic is run over an ECMP fabric.
+    """
+    _setup_hw_gro(cfg)
+    _setup_isolated_queue(cfg)
+
+    _run_gro_test(cfg, "capacity", num_flows=num_flows, order_check=True)
+
+
 def main() -> None:
     """ Ksft boiler plate main """
 
     with NetDrvEpEnv(__file__, nsim_test=False) as cfg:
         cfg.netnl = NetdevFamily()
         ksft_run([test_gro_stats_single,
-                  test_gro_stats_full], args=(cfg,))
+                  test_gro_stats_full,
+                  test_gro_order], args=(cfg,))
     ksft_exit()
 
 
Signature · 1 line
2.53.0
author maintainer
#6 lore raw

PATCH v4 6/6 selftests: drv-net: gro: add a test for GRO depth

In reply to Jakub Kicinski

Reuse the long sequence test to max out the GRO contexts. Repeat for a single queue, 8 queues, and default number of queues but flow steering to just one.

The SW GRO's capacity should be around 64 per queue (8 buckets, up to 8 skbs in a chain).

---
v3:
- rename [FAIL] to [MISS] to avoid confusing NIPA
v2: https://lore.kernel.org/[email protected]
- move this "test" to SW, we can check SW GRO capacity too
v1: https://lore.kernel.org/[email protected]

---
 tools/testing/selftests/drivers/net/gro.py | 201 +++++++++++++++++++--
 1 file changed, 181 insertions(+), 20 deletions(-)
tools/testing/selftests/drivers/net/gro.py +181 -20
index 2da53686354f..70709bf670c7 100755
@@ -35,11 +35,18 @@ coalescing behavior.
   - large_rem: Large packet remainder handling
 """
 
+import glob
 import os
+import re
 from lib.py import ksft_run, ksft_exit, ksft_pr
 from lib.py import NetDrvEpEnv, KsftXfailEx
+from lib.py import NetdevFamily, EthtoolFamily
 from lib.py import bkg, cmd, defer, ethtool, ip
-from lib.py import ksft_variants
+from lib.py import ksft_variants, KsftNamedVariant
+
+
+# gro.c uses hardcoded DPORT=8000
+GRO_DPORT = 8000
 
 
 def _resolve_dmac(cfg, ipver):
@@ -113,6 +120,98 @@ from lib.py import ksft_variants
         ksft_pr(eth_cmd)
 
 
+def _get_queue_stats(cfg, queue_id):
+    """Get stats for a specific Rx queue."""
+    cfg.wait_hw_stats_settle()
+    data = cfg.netnl.qstats_get({"ifindex": cfg.ifindex, "scope": ["queue"]},
+                                dump=True)
+    for q in data:
+        if q.get('queue-type') == 'rx' and q.get('queue-id') == queue_id:
+            return q
+    return {}
+
+
+def _setup_isolated_queue(cfg):
+    """Set up an isolated queue for testing using ntuple filter.
+
+    Remove queue 1 from the default RSS context and steer test traffic to it.
+    """
+    test_queue = 1
+
+    qcnt = len(glob.glob(f"/sys/class/net/{cfg.ifname}/queues/rx-*"))
+    if qcnt < 2:
+        raise KsftXfailEx(f"Need at least 2 queues, have {qcnt}")
+
+    # Remove queue 1 from default RSS context by setting its weight to 0
+    weights = ["1"] * qcnt
+    weights[test_queue] = "0"
+    ethtool(f"-X {cfg.ifname} weight " + " ".join(weights))
+    defer(ethtool, f"-X {cfg.ifname} default")
+
+    # Set up ntuple filter to steer our test traffic to the isolated queue
+    flow  = f"flow-type tcp{cfg.addr_ipver} "
+    flow += f"dst-ip {cfg.addr} dst-port {GRO_DPORT} action {test_queue}"
+    output = ethtool(f"-N {cfg.ifname} {flow}").stdout
+    ntuple_id = int(output.split()[-1])
+    defer(ethtool, f"-N {cfg.ifname} delete {ntuple_id}")
+
+    return test_queue
+
+
+def _setup_queue_count(cfg, num_queues):
+    """Configure the NIC to use a specific number of queues."""
+    channels = cfg.ethnl.channels_get({'header': {'dev-index': cfg.ifindex}})
+    ch_max = channels.get('combined-max', 0)
+    qcnt = channels['combined-count']
+
+    if ch_max < num_queues:
+        raise KsftXfailEx(f"Need at least {num_queues} queues, max={ch_max}")
+
+    defer(ethtool, f"-L {cfg.ifname} combined {qcnt}")
+    ethtool(f"-L {cfg.ifname} combined {num_queues}")
+
+
+def _run_gro_bin(cfg, test_name, protocol=None, num_flows=None,
+                 order_check=False, verbose=False, fail=False):
+    """Run gro binary with given test and return the process result."""
+    if not hasattr(cfg, "bin_remote"):
+        cfg.bin_local = cfg.net_lib_dir / "gro"
+        cfg.bin_remote = cfg.remote.deploy(cfg.bin_local)
+
+    if protocol is None:
+        ipver = cfg.addr_ipver
+        protocol = f"ipv{ipver}"
+    else:
+        ipver = "6" if protocol[-1] == "6" else "4"
+
+    dmac = _resolve_dmac(cfg, ipver)
+
+    base_args = [
+        f"--{protocol}",
+        f"--dmac {dmac}",
+        f"--smac {cfg.remote_dev['address']}",
+        f"--daddr {cfg.addr_v[ipver]}",
+        f"--saddr {cfg.remote_addr_v[ipver]}",
+        f"--test {test_name}",
+    ]
+    if num_flows:
+        base_args.append(f"--num-flows {num_flows}")
+    if order_check:
+        base_args.append("--order-check")
+    if verbose:
+        base_args.append("--verbose")
+
+    args = " ".join(base_args)
+
+    rx_cmd = f"{cfg.bin_local} {args} --rx --iface {cfg.ifname}"
+    tx_cmd = f"{cfg.bin_remote} {args} --iface {cfg.remote_ifname}"
+
+    with bkg(rx_cmd, ksft_ready=True, exit_wait=True, fail=fail) as rx_proc:
+        cmd(tx_cmd, host=cfg.remote)
+
+    return rx_proc
+
+
 def _setup(cfg, mode, test_name):
     """ Setup hardware loopback mode for GRO testing. """
 
@@ -233,30 +332,14 @@ from lib.py import ksft_variants
 
     _setup(cfg, mode, test_name)
 
-    base_cmd_args = [
-        f"--{protocol}",
-        f"--dmac {_resolve_dmac(cfg, ipver)}",
-        f"--smac {cfg.remote_dev['address']}",
-        f"--daddr {cfg.addr_v[ipver]}",
-        f"--saddr {cfg.remote_addr_v[ipver]}",
-        f"--test {test_name}",
-        "--verbose"
-    ]
-    base_args = " ".join(base_cmd_args)
-
     # Each test is run 6 times to deflake, because given the receive timing,
     # not all packets that should coalesce will be considered in the same flow
     # on every try.
     max_retries = 6
     for attempt in range(max_retries):
-        rx_cmd = f"{cfg.bin_local} {base_args} --rx --iface {cfg.ifname}"
-        tx_cmd = f"{cfg.bin_remote} {base_args} --iface {cfg.remote_ifname}"
-
         fail_now = attempt >= max_retries - 1
-
-        with bkg(rx_cmd, ksft_ready=True, exit_wait=True,
-                 fail=fail_now) as rx_proc:
-            cmd(tx_cmd, host=cfg.remote)
+        rx_proc = _run_gro_bin(cfg, test_name, protocol=protocol,
+                               verbose=True, fail=fail_now)
 
         if rx_proc.ret == 0:
             return
@@ -270,11 +353,89 @@ from lib.py import ksft_variants
         ksft_pr(f"Attempt {attempt + 1}/{max_retries} failed, retrying...")
 
 
+def _capacity_variants():
+    """Generate variants for capacity test: mode x queue setup."""
+    setups = [
+        ("isolated", _setup_isolated_queue),
+        ("1q", lambda cfg: _setup_queue_count(cfg, 1)),
+        ("8q", lambda cfg: _setup_queue_count(cfg, 8)),
+    ]
+    for mode in ["sw", "hw", "lro"]:
+        for name, func in setups:
+            yield KsftNamedVariant(f"{mode}_{name}", mode, func)
+
+
+@ksft_variants(_capacity_variants())
+def test_gro_capacity(cfg, mode, setup_func):
+    """
+    Probe GRO capacity.
+
+    Start with 8 flows and increase by 2x on each successful run.
+    Retry up to 3 times on failure.
+
+    Variants combine mode (sw, hw, lro) with queue setup:
+      - isolated: Use a single queue isolated from RSS
+      - 1q: Configure NIC to use 1 queue
+      - 8q: Configure NIC to use 8 queues
+    """
+    max_retries = 3
+
+    _setup(cfg, mode, "capacity")
+    queue_id = setup_func(cfg)
+
+    num_flows = 8
+    while True:
+        success = False
+        for attempt in range(max_retries):
+            if queue_id is not None:
+                stats_before = _get_queue_stats(cfg, queue_id)
+
+            rx_proc = _run_gro_bin(cfg, "capacity", num_flows=num_flows)
+            output = rx_proc.stdout
+
+            if queue_id is not None:
+                stats_after = _get_queue_stats(cfg, queue_id)
+                qstat_pkts = (stats_after.get('rx-packets', 0) -
+                              stats_before.get('rx-packets', 0))
+                gro_pkts = (stats_after.get('rx-hw-gro-packets', 0) -
+                            stats_before.get('rx-hw-gro-packets', 0))
+                qstat_str = f" qstat={qstat_pkts} hw-gro={gro_pkts}"
+            else:
+                qstat_str = ""
+
+            # Parse and print STATS line
+            match = re.search(
+                r'STATS: received=(\d+) wire=(\d+) coalesced=(\d+)', output)
+            if match:
+                received = int(match.group(1))
+                wire = int(match.group(2))
+                coalesced = int(match.group(3))
+                status = "PASS" if received == num_flows else "MISS"
+                ksft_pr(f"flows={num_flows} attempt={attempt + 1} "
+                        f"received={received} wire={wire} "
+                        f"coalesced={coalesced}{qstat_str} [{status}]")
+                if received == num_flows:
+                    success = True
+                    break
+            else:
+                ksft_pr(rx_proc)
+                ksft_pr(f"flows={num_flows} attempt={attempt + 1}"
+                        f"{qstat_str} [FAIL - can't parse stats]")
+
+        if not success:
+            ksft_pr(f"Stopped at {num_flows} flows")
+            break
+
+        num_flows *= 2
+
+
 def main() -> None:
     """ Ksft boiler plate main """
 
     with NetDrvEpEnv(__file__) as cfg:
-        ksft_run(cases=[test], args=(cfg,))
+        cfg.ethnl = EthtoolFamily()
+        cfg.netnl = NetdevFamily()
+        ksft_run(cases=[test, test_gro_capacity], args=(cfg,))
     ksft_exit()
 
 
Signature · 1 line
2.53.0
#7 lore raw
In reply to Jakub Kicinski
This series was applied to netdev/net-next.git (main) by Jakub Kicinski <[email protected]>:

Hello:

This series was applied to netdev/net-next.git (main) by Jakub Kicinski <[email protected]>:

Quoted 14 lines · On Tue, 17 Mar 2026 20:38:13 -0700 you wrote:

Add tests for HW GRO stats, packet ordering and depth.

The ynltool and bnxt patches from v2 were applied separately.

v4:
- [patch 6] rename [FAIL] -> [MISS]; NIPA thinks every [FAIL] means test failed, 'cause bash test don't ktap :/
v3: https://lore.kernel.org/[email protected]
- [patch 3] add a note about fq being needed
v2: https://lore.kernel.org/[email protected]
- [patch 3] do not enable SO_TXTIME for multi-flow tests
- [patch 6] move the test to cover SW GRO
v1: https://lore.kernel.org/[email protected]

Quoted 1 line

[...]

Here is the summary with links:
- [net-next,v4,1/6] selftests: net: move gro to lib for HW vs SW reuse

    https://git.kernel.org/netdev/net-next/c/8888bf4fb980
  - [net-next,v4,2/6] selftests: drv-net: give HW stats sync time extra 25% of margin
    https://git.kernel.org/netdev/net-next/c/9b29afa11660
  - [net-next,v4,3/6] selftests: drv-net: gro: use SO_TXTIME to schedule packets together
    https://git.kernel.org/netdev/net-next/c/f26d43acf12f
  - [net-next,v4,4/6] selftests: drv-net: gro: test GRO stats
    https://git.kernel.org/netdev/net-next/c/ba5d4128fca8
  - [net-next,v4,5/6] selftests: drv-net: gro: add test for packet ordering
    https://git.kernel.org/netdev/net-next/c/ff1cb3ad2abc
  - [net-next,v4,6/6] selftests: drv-net: gro: add a test for GRO depth
    https://git.kernel.org/netdev/net-next/c/7dae8ffb0987

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html