On Tue, Jan 31, 2023 at 8:06 AM Andrei Gherzan andrei.gherzan@canonical.com wrote:

"udpgro_bench.sh" invokes udpgso_bench_rx/udpgso_bench_tx programs subsequently and while doing so, there is a chance that the rx one is not ready to accept socket connections. This racing bug could fail the test with at least one of the following:

./udpgso_bench_tx: connect: Connection refused ./udpgso_bench_tx: sendmsg: Connection refused ./udpgso_bench_tx: write: Connection refused

This change addresses this by making udpgro_bench.sh wait for the rx program to be ready before firing off the tx one - with an exponential back off algorithm from 1s to 10s.

Signed-off-by: Andrei Gherzan andrei.gherzan@canonical.com

please CC: reviewers of previous revisions on new revisions

also for upcoming patches: please clearly mark net or net-next.

---

tools/testing/selftests/net/udpgso_bench.sh | 18 ++++++++++++++++++ 1 file changed, 18 insertions(+)

diff --git a/tools/testing/selftests/net/udpgso_bench.sh b/tools/testing/selftests/net/udpgso_bench.sh index dc932fd65363..20b5db8fcbde 100755 --- a/tools/testing/selftests/net/udpgso_bench.sh +++ b/tools/testing/selftests/net/udpgso_bench.sh @@ -7,6 +7,7 @@ readonly GREEN='\033[0;92m' readonly YELLOW='\033[0;33m' readonly RED='\033[0;31m' readonly NC='\033[0m' # No Color +readonly TESTPORT=8000 # Keep this in sync with udpgso_bench_rx/tx

then also pass explicit -p argument to the processes to keep all three consistent

readonly KSFT_PASS=0 readonly KSFT_FAIL=1 @@ -56,10 +57,27 @@ trap wake_children EXIT

run_one() { local -r args=$@

•   local -r init_delay_s=1
  

•   local -r max_delay_s=10
  

•   local delay_s=0
  

•   local nr_socks=0
  
    ./udpgso_bench_rx &
    ./udpgso_bench_rx -t &
  
  

•   # Wait for the above test program to get ready to receive connections.
  

•   delay_s="${init_delay_s}"
  

•   while [ "$delay_s" -lt "$max_delay_s" ]; do
  

•           nr_socks="$(ss -lnHi | grep -c "\*:${TESTPORT}")"
  

•           [ "$nr_socks" -eq 2 ] && break
  

•           sleep "$delay_s"
  

•           delay="$((delay*2))"
  

I don't think we need exponential back-off for something this simple

•   done
  

•   if [ "$nr_socks" -ne 2 ]; then
  

•           echo "timed out while waiting for udpgso_bench_rx"
  

•           exit 1
  

•   fi
  

•   ./udpgso_bench_tx ${args}
  

}

-- 2.34.1

On 23/01/31 08:33AM, Willem de Bruijn wrote:

On Tue, Jan 31, 2023 at 8:06 AM Andrei Gherzan andrei.gherzan@canonical.com wrote:

...

"udpgro_bench.sh" invokes udpgso_bench_rx/udpgso_bench_tx programs subsequently and while doing so, there is a chance that the rx one is not ready to accept socket connections. This racing bug could fail the test with at least one of the following:

./udpgso_bench_tx: connect: Connection refused ./udpgso_bench_tx: sendmsg: Connection refused ./udpgso_bench_tx: write: Connection refused

This change addresses this by making udpgro_bench.sh wait for the rx program to be ready before firing off the tx one - with an exponential back off algorithm from 1s to 10s.

Signed-off-by: Andrei Gherzan andrei.gherzan@canonical.com

please CC: reviewers of previous revisions on new revisions

also for upcoming patches: please clearly mark net or net-next.

Ack. I'll do all that in next v.

...

---

tools/testing/selftests/net/udpgso_bench.sh | 18 ++++++++++++++++++ 1 file changed, 18 insertions(+)

diff --git a/tools/testing/selftests/net/udpgso_bench.sh b/tools/testing/selftests/net/udpgso_bench.sh index dc932fd65363..20b5db8fcbde 100755 --- a/tools/testing/selftests/net/udpgso_bench.sh +++ b/tools/testing/selftests/net/udpgso_bench.sh @@ -7,6 +7,7 @@ readonly GREEN='\033[0;92m' readonly YELLOW='\033[0;33m' readonly RED='\033[0;31m' readonly NC='\033[0m' # No Color +readonly TESTPORT=8000 # Keep this in sync with udpgso_bench_rx/tx

then also pass explicit -p argument to the processes to keep all three consistent

I have no idea how I've missed it. It will be in v4.

...

readonly KSFT_PASS=0 readonly KSFT_FAIL=1 @@ -56,10 +57,27 @@ trap wake_children EXIT

run_one() { local -r args=$@

•   local -r init_delay_s=1
  

•   local -r max_delay_s=10
  

•   local delay_s=0
  

•   local nr_socks=0
  
    ./udpgso_bench_rx &
    ./udpgso_bench_rx -t &
  
  

•   # Wait for the above test program to get ready to receive connections.
  

•   delay_s="${init_delay_s}"
  

•   while [ "$delay_s" -lt "$max_delay_s" ]; do
  

•           nr_socks="$(ss -lnHi | grep -c "\*:${TESTPORT}")"
  

•           [ "$nr_socks" -eq 2 ] && break
  

•           sleep "$delay_s"
  

•           delay="$((delay*2))"
  

I don't think we need exponential back-off for something this simple

I'm happy to drop/simplify. I'll go for a simple sleep 1 in a timeout loop.

...

•   done
  

•   if [ "$nr_socks" -ne 2 ]; then
  

•           echo "timed out while waiting for udpgso_bench_rx"
  

•           exit 1
  

•   fi
  

•   ./udpgso_bench_tx ${args}
  

}

-- 2.34.1

On Tue, Jan 31, 2023 at 8:06 AM Andrei Gherzan andrei.gherzan@canonical.com wrote:

The test tool can check that the zerocopy number of completions value is valid taking into consideration the number of datagram send calls. This can catch the system into a state where the datagrams are still in the system (for example in a qdisk, waiting for the network interface to return a completion notification, etc).

This change adds a retry logic of computing the number of completions up to a configurable (via CLI) timeout (default: 2 seconds).

Signed-off-by: Andrei Gherzan andrei.gherzan@canonical.com

tools/testing/selftests/net/udpgso_bench_tx.c | 38 +++++++++++++++---- 1 file changed, 30 insertions(+), 8 deletions(-)

diff --git a/tools/testing/selftests/net/udpgso_bench_tx.c b/tools/testing/selftests/net/udpgso_bench_tx.c index b47b5c32039f..5a29b5f24023 100644 --- a/tools/testing/selftests/net/udpgso_bench_tx.c +++ b/tools/testing/selftests/net/udpgso_bench_tx.c @@ -62,6 +62,7 @@ static int cfg_payload_len = (1472 * 42); static int cfg_port = 8000; static int cfg_runtime_ms = -1; static bool cfg_poll; +static int cfg_poll_loop_timeout_ms = 2000; static bool cfg_segment; static bool cfg_sendmmsg; static bool cfg_tcp; @@ -235,16 +236,17 @@ static void flush_errqueue_recv(int fd) } }

-static void flush_errqueue(int fd, const bool do_poll) +static void flush_errqueue(int fd, const bool do_poll,

•           unsigned long poll_timeout, const bool poll_err)
  

{ if (do_poll) { struct pollfd fds = {0}; int ret;

            fds.fd = fd;

•           ret = poll(&fds, 1, 500);
  

•           ret = poll(&fds, 1, poll_timeout);
            if (ret == 0) {
  

•                   if (cfg_verbose)
  

•                   if ((cfg_verbose) && (poll_err))
                            fprintf(stderr, "poll timeout\n");
            } else if (ret < 0) {
                    error(1, errno, "poll");
  

@@ -254,6 +256,22 @@ static void flush_errqueue(int fd, const bool do_poll) flush_errqueue_recv(fd); }

+static void flush_errqueue_retry(int fd, const bool do_poll, unsigned long num_sends) +{

•   unsigned long tnow, tstop;
  

•   bool first_try = true;
  

•   tnow = gettimeofday_ms();
  

•   tstop = tnow + cfg_poll_loop_timeout_ms;
  

•   do {
  

•           flush_errqueue(fd, do_poll, tstop - tnow, first_try);
  

•           first_try = false;
  

•           if (!do_poll)
  

•                   usleep(1000);  // a throttling delay if polling is enabled
  

never reached

On 23/01/31 03:51PM, Paolo Abeni wrote:

On Tue, 2023-01-31 at 13:04 +0000, Andrei Gherzan wrote:

...

The test tool can check that the zerocopy number of completions value is valid taking into consideration the number of datagram send calls. This can catch the system into a state where the datagrams are still in the system (for example in a qdisk, waiting for the network interface to return a completion notification, etc).

This change adds a retry logic of computing the number of completions up to a configurable (via CLI) timeout (default: 2 seconds).

Signed-off-by: Andrei Gherzan andrei.gherzan@canonical.com

tools/testing/selftests/net/udpgso_bench_tx.c | 38 +++++++++++++++---- 1 file changed, 30 insertions(+), 8 deletions(-)

diff --git a/tools/testing/selftests/net/udpgso_bench_tx.c b/tools/testing/selftests/net/udpgso_bench_tx.c index b47b5c32039f..5a29b5f24023 100644 --- a/tools/testing/selftests/net/udpgso_bench_tx.c +++ b/tools/testing/selftests/net/udpgso_bench_tx.c @@ -62,6 +62,7 @@ static int cfg_payload_len = (1472 * 42); static int cfg_port = 8000; static int cfg_runtime_ms = -1; static bool cfg_poll; +static int cfg_poll_loop_timeout_ms = 2000; static bool cfg_segment; static bool cfg_sendmmsg; static bool cfg_tcp; @@ -235,16 +236,17 @@ static void flush_errqueue_recv(int fd) } } -static void flush_errqueue(int fd, const bool do_poll) +static void flush_errqueue(int fd, const bool do_poll,

• unsigned long poll_timeout, const bool poll_err)
  

{ if (do_poll) { struct pollfd fds = {0}; int ret; fds.fd = fd;

• ret = poll(&fds, 1, 500);
  

• ret = poll(&fds, 1, poll_timeout);
  

  if (ret == 0) {

• 	if (cfg_verbose)
  

• 	if ((cfg_verbose) && (poll_err))
  	fprintf(stderr, "poll timeout\n");
  

  } else if (ret < 0) { error(1, errno, "poll");

@@ -254,6 +256,22 @@ static void flush_errqueue(int fd, const bool do_poll) flush_errqueue_recv(fd); } +static void flush_errqueue_retry(int fd, const bool do_poll, unsigned long num_sends) +{

• unsigned long tnow, tstop;
• bool first_try = true;
• tnow = gettimeofday_ms();
• tstop = tnow + cfg_poll_loop_timeout_ms;
• do {

• flush_errqueue(fd, do_poll, tstop - tnow, first_try);
  

• first_try = false;
  

• if (!do_poll)
  

• 	usleep(1000);  // a throttling delay if polling is enabled
  

Even if the kernel codying style is not very strictly enforced for self-tests, please avoid c++ style comments.

More importantly, as Willem noded, this function is always called with do_poll == true. You should drop such argument and the related branch above.

Agreed. I will drop.

...

• tnow = gettimeofday_ms();
  

• } while ((stat_zcopies != num_sends) && (tnow < tstop));

+}

static int send_tcp(int fd, char *data) { int ret, done = 0, count = 0; @@ -413,8 +431,9 @@ static int send_udp_segment(int fd, char *data) static void usage(const char *filepath) {

• error(1, 0, "Usage: %s [-46acmHPtTuvz] [-C cpu] [-D dst ip] [-l secs] [-M messagenr] [-p port] [-s sendsize] [-S gsosize]",

•     filepath);
  

• error(1, 0,

• 	"Usage: %s [-46acmHPtTuvz] [-C cpu] [-D dst ip] [-l secs] [-L secs] [-M messagenr] [-p port] [-s sendsize] [-S gsosize]",
  

• 	filepath);
  

Please avoid introducing unnecessary white-space changes (no reason to move the usage text on a new line)

The only reason why I've done this was to make scripts/checkpatch.pl happy:

WARNING: line length of 141 exceeds 100 columns #83: FILE: tools/testing/selftests/net/udpgso_bench_tx.c:432:

I can drop and ignore the warning, or maybe it would have been better to just mention this in git message. What do you prefer?

On 23/01/31 05:22PM, Paolo Abeni wrote:

On Tue, 2023-01-31 at 15:08 +0000, Andrei Gherzan wrote:

...

On 23/01/31 03:51PM, Paolo Abeni wrote:

...

On Tue, 2023-01-31 at 13:04 +0000, Andrei Gherzan wrote:

...

The test tool can check that the zerocopy number of completions value is valid taking into consideration the number of datagram send calls. This can catch the system into a state where the datagrams are still in the system (for example in a qdisk, waiting for the network interface to return a completion notification, etc).

This change adds a retry logic of computing the number of completions up to a configurable (via CLI) timeout (default: 2 seconds).

Signed-off-by: Andrei Gherzan andrei.gherzan@canonical.com

tools/testing/selftests/net/udpgso_bench_tx.c | 38 +++++++++++++++---- 1 file changed, 30 insertions(+), 8 deletions(-)

diff --git a/tools/testing/selftests/net/udpgso_bench_tx.c b/tools/testing/selftests/net/udpgso_bench_tx.c index b47b5c32039f..5a29b5f24023 100644 --- a/tools/testing/selftests/net/udpgso_bench_tx.c +++ b/tools/testing/selftests/net/udpgso_bench_tx.c @@ -62,6 +62,7 @@ static int cfg_payload_len = (1472 * 42); static int cfg_port = 8000; static int cfg_runtime_ms = -1; static bool cfg_poll; +static int cfg_poll_loop_timeout_ms = 2000; static bool cfg_segment; static bool cfg_sendmmsg; static bool cfg_tcp; @@ -235,16 +236,17 @@ static void flush_errqueue_recv(int fd) } } -static void flush_errqueue(int fd, const bool do_poll) +static void flush_errqueue(int fd, const bool do_poll,

• unsigned long poll_timeout, const bool poll_err)
  

{ if (do_poll) { struct pollfd fds = {0}; int ret; fds.fd = fd;

• ret = poll(&fds, 1, 500);
  

• ret = poll(&fds, 1, poll_timeout);
  

  if (ret == 0) {

• 	if (cfg_verbose)
  

• 	if ((cfg_verbose) && (poll_err))
  	fprintf(stderr, "poll timeout\n");
  

  } else if (ret < 0) { error(1, errno, "poll");

@@ -254,6 +256,22 @@ static void flush_errqueue(int fd, const bool do_poll) flush_errqueue_recv(fd); } +static void flush_errqueue_retry(int fd, const bool do_poll, unsigned long num_sends) +{

• unsigned long tnow, tstop;
• bool first_try = true;
• tnow = gettimeofday_ms();
• tstop = tnow + cfg_poll_loop_timeout_ms;
• do {

• flush_errqueue(fd, do_poll, tstop - tnow, first_try);
  

• first_try = false;
  

• if (!do_poll)
  

• 	usleep(1000);  // a throttling delay if polling is enabled
  

Even if the kernel codying style is not very strictly enforced for self-tests, please avoid c++ style comments.

More importantly, as Willem noded, this function is always called with do_poll == true. You should drop such argument and the related branch above.

Agreed. I will drop.

...

...

• tnow = gettimeofday_ms();
  

• } while ((stat_zcopies != num_sends) && (tnow < tstop));

+}

static int send_tcp(int fd, char *data) { int ret, done = 0, count = 0; @@ -413,8 +431,9 @@ static int send_udp_segment(int fd, char *data) static void usage(const char *filepath) {

• error(1, 0, "Usage: %s [-46acmHPtTuvz] [-C cpu] [-D dst ip] [-l secs] [-M messagenr] [-p port] [-s sendsize] [-S gsosize]",

•     filepath);
  

• error(1, 0,

• 	"Usage: %s [-46acmHPtTuvz] [-C cpu] [-D dst ip] [-l secs] [-L secs] [-M messagenr] [-p port] [-s sendsize] [-S gsosize]",
  

• 	filepath);
  

Please avoid introducing unnecessary white-space changes (no reason to move the usage text on a new line)

The only reason why I've done this was to make scripts/checkpatch.pl happy:

WARNING: line length of 141 exceeds 100 columns #83: FILE: tools/testing/selftests/net/udpgso_bench_tx.c:432:

I can drop and ignore the warning, or maybe it would have been better to just mention this in git message. What do you prefer?

Long lines are allowed for (kernel) messages, to make them easily grep- able.

In this specific case you can either append the new text to the message without introducing that strange indentation or even better break the usage string alike:

"Usage: %s [-46acmHPtTuvz] [-C cpu] [-D dst ip] [-l secs] [-L secs]" " [-L secs] [-M messagenr] [-p port] [-s sendsize] [-S gsosize]"

Funny I went through this too but it also fails with:

WARNING: quoted string split across lines #84: FILE: tools/testing/selftests/net/udpgso_bench_tx.c:433

This is how I usually do it but it seems like it's flagged too.

On 23/01/31 07:26PM, Paolo Abeni wrote:

On Tue, 2023-01-31 at 16:31 +0000, Andrei Gherzan wrote:

...

On 23/01/31 05:22PM, Paolo Abeni wrote:

...

On Tue, 2023-01-31 at 15:08 +0000, Andrei Gherzan wrote:

...

On 23/01/31 03:51PM, Paolo Abeni wrote:

...

On Tue, 2023-01-31 at 13:04 +0000, Andrei Gherzan wrote:

...

The test tool can check that the zerocopy number of completions value is valid taking into consideration the number of datagram send calls. This can catch the system into a state where the datagrams are still in the system (for example in a qdisk, waiting for the network interface to return a completion notification, etc).

This change adds a retry logic of computing the number of completions up to a configurable (via CLI) timeout (default: 2 seconds).

Signed-off-by: Andrei Gherzan andrei.gherzan@canonical.com

tools/testing/selftests/net/udpgso_bench_tx.c | 38 +++++++++++++++---- 1 file changed, 30 insertions(+), 8 deletions(-)

diff --git a/tools/testing/selftests/net/udpgso_bench_tx.c b/tools/testing/selftests/net/udpgso_bench_tx.c index b47b5c32039f..5a29b5f24023 100644 --- a/tools/testing/selftests/net/udpgso_bench_tx.c +++ b/tools/testing/selftests/net/udpgso_bench_tx.c @@ -62,6 +62,7 @@ static int cfg_payload_len = (1472 * 42); static int cfg_port = 8000; static int cfg_runtime_ms = -1; static bool cfg_poll; +static int cfg_poll_loop_timeout_ms = 2000; static bool cfg_segment; static bool cfg_sendmmsg; static bool cfg_tcp; @@ -235,16 +236,17 @@ static void flush_errqueue_recv(int fd) } } -static void flush_errqueue(int fd, const bool do_poll) +static void flush_errqueue(int fd, const bool do_poll,

• unsigned long poll_timeout, const bool poll_err)
  

{ if (do_poll) { struct pollfd fds = {0}; int ret; fds.fd = fd;

• ret = poll(&fds, 1, 500);
  

• ret = poll(&fds, 1, poll_timeout);
  

  if (ret == 0) {

• 	if (cfg_verbose)
  

• 	if ((cfg_verbose) && (poll_err))
  	fprintf(stderr, "poll timeout\n");
  

  } else if (ret < 0) { error(1, errno, "poll");

@@ -254,6 +256,22 @@ static void flush_errqueue(int fd, const bool do_poll) flush_errqueue_recv(fd); } +static void flush_errqueue_retry(int fd, const bool do_poll, unsigned long num_sends) +{

• unsigned long tnow, tstop;
• bool first_try = true;
• tnow = gettimeofday_ms();
• tstop = tnow + cfg_poll_loop_timeout_ms;
• do {

• flush_errqueue(fd, do_poll, tstop - tnow, first_try);
  

• first_try = false;
  

• if (!do_poll)
  

• 	usleep(1000);  // a throttling delay if polling is enabled
  

Even if the kernel codying style is not very strictly enforced for self-tests, please avoid c++ style comments.

More importantly, as Willem noded, this function is always called with do_poll == true. You should drop such argument and the related branch above.

Agreed. I will drop.

...

...

• tnow = gettimeofday_ms();
  

• } while ((stat_zcopies != num_sends) && (tnow < tstop));

+}

static int send_tcp(int fd, char *data) { int ret, done = 0, count = 0; @@ -413,8 +431,9 @@ static int send_udp_segment(int fd, char *data) static void usage(const char *filepath) {

• error(1, 0, "Usage: %s [-46acmHPtTuvz] [-C cpu] [-D dst ip] [-l secs] [-M messagenr] [-p port] [-s sendsize] [-S gsosize]",

•     filepath);
  

• error(1, 0,

• 	"Usage: %s [-46acmHPtTuvz] [-C cpu] [-D dst ip] [-l secs] [-L secs] [-M messagenr] [-p port] [-s sendsize] [-S gsosize]",
  

• 	filepath);
  

Please avoid introducing unnecessary white-space changes (no reason to move the usage text on a new line)

The only reason why I've done this was to make scripts/checkpatch.pl happy:

WARNING: line length of 141 exceeds 100 columns #83: FILE: tools/testing/selftests/net/udpgso_bench_tx.c:432:

I can drop and ignore the warning, or maybe it would have been better to just mention this in git message. What do you prefer?

Long lines are allowed for (kernel) messages, to make them easily grep- able.

In this specific case you can either append the new text to the message without introducing that strange indentation or even better break the usage string alike:

"Usage: %s [-46acmHPtTuvz] [-C cpu] [-D dst ip] [-l secs] [-L secs]" " [-L secs] [-M messagenr] [-p port] [-s sendsize] [-S gsosize]"

Funny I went through this too but it also fails with:

WARNING: quoted string split across lines #84: FILE: tools/testing/selftests/net/udpgso_bench_tx.c:433

This is how I usually do it but it seems like it's flagged too.

I'm all for ignoring this warning in this specific context. Among other things it will be consistent with other existing self-tests.

Eventually the checkpatch script could be tuned (with an unrelated patch) to discriminate between kernel and self-tests code.

In that case I will use quoted strings split across lines in the next version