net: xsk: update tx queue consumer
Subsystems (6)
v3 cover net: xsk: update tx queue consumer
From: Jason Xing <[email protected]>
Patch 1 makes sure the consumer is updated at the end of xsk xmit. Patch 2 adds corresponding test.
Jason Xing (2):
net: xsk: update tx queue consumer immediately after transmission selftests/bpf: check if the global consumer of tx queue updates after
send call
net/xdp/xsk.c | 3 +++ tools/testing/selftests/bpf/xskxceiver.c | 30 ++++++++++++++++++++++-- 2 files changed, 31 insertions(+), 2 deletions(-)
Signature · 1 line
2.41.3
PATCH v3 1/2 net: xsk: update tx queue consumer immediately after transmission
From: Jason Xing <[email protected]>
For afxdp, the return value of sendto() syscall doesn't reflect how many descs handled in the kernel. One of use cases is that when user-space application tries to know the number of transmitted skbs and then decides if it continues to send, say, is it stopped due to max tx budget?
The following formular can be used after sending to learn how many skbs/descs the kernel takes care of:
tx_queue.consumers_before - tx_queue.consumers_after
Prior to the current patch, in non-zc mode, the consumer of tx queue is not immediately updated at the end of each sendto syscall when error occurs, which leads to the consumer value out-of-dated from the perspective of user space. So this patch requires store operation to pass the cached value to the shared value to handle the problem.
More than those explicit errors appearing in the while() loop in __xsk_generic_xmit(), there are a few possible error cases that might be neglected in the following call trace:__xsk_generic_xmit()
xskq_cons_peek_desc()
xskq_cons_read_desc()
xskq_cons_is_valid_desc()
It will also cause the premature exit in the while() loop even if not all the descs are consumed.
Based on the above analysis, using 'cached_prod != cached_cons' could cover all the possible cases because it represents there are remaining descs that are not handled and cached_cons are not updated to the global state of consumer at this time.
---
v3
1. use xskq_has_descs helper.
2. add selftest
V2
1. filter out those good cases because only those that return error need updates.
Side note:
1. in non-batched zero copy mode, at the end of every caller of xsk_tx_peek_desc(), there is always a xsk_tx_release() function that used to update the local consumer to the global state of consumer. So for the zero copy mode, no need to change at all.
2. Actually I have no strong preference between v1 (see the above link) and v2 because smp_store_release() shouldn't cause side effect. Considering the exactitude of writing code, v2 is a more preferable one.
--- net/xdp/xsk.c | 3 +++ 1 file changed, 3 insertions(+)
net/xdp/xsk.c +3 -0
@@ -856,6 +856,9 @@ static int __xsk_generic_xmit(struct sock *sk) } out: +if (xskq_has_descs (xs->tx)) +__xskq_cons_release (xs->tx); +if (sent_frame)if (xsk_tx_writeable (xs)) sk->sk_write_space (sk);
Signature · 1 line
2.41.3
Maciej Fijalkowski
maintainerRe: [PATCH net-next v3 1/2] net: xsk: update tx queue consumer immediately after transmission
Quoted 62 lines · Jason Xing
From: Jason Xing <[email protected]>
For afxdp, the return value of
sendto()syscall doesn't reflect how many descs handled in the kernel. One of use cases is that when user-space application tries to know the number of transmitted skbs and then decides if it continues to send, say, is it stopped due to max tx budget?The following formular can be used after sending to learn how many skbs/descs the kernel takes care of:
tx_queue.consumers_before - tx_queue.consumers_after
Prior to the current patch, in non-zc mode, the consumer of tx queue is not immediately updated at the end of each sendto syscall when error occurs, which leads to the consumer value out-of-dated from the perspective of user space. So this patch requires store operation to pass the cached value to the shared value to handle the problem.
More than those explicit errors appearing in the
while()loop in__xsk_generic_xmit(), there are a few possible error cases that might be neglected in the following call trace:__xsk_generic_xmit()
xskq_cons_peek_desc()
xskq_cons_read_desc()
xskq_cons_is_valid_desc()
It will also cause the premature exit in thewhile()loop even if not all the descs are consumed.Based on the above analysis, using '
cached_prod!=cached_cons' could cover all the possible cases because it represents there are remaining descs that are not handled andcached_consare not updated to the global state of consumer at this time.Signed-off-by Jason Xing <[email protected]>---
v31. use
xskq_has_descshelper.
2. add selftestV2
1. filter out those good cases because only those that return error need updates.
Side note:
1. in non-batched zero copy mode, at the end of every caller ofxsk_tx_peek_desc(), there is always axsk_tx_release()function that used to update the local consumer to the global state of consumer. So for the zero copy mode, no need to change at all.
2. Actually I have no strong preference between v1 (see the above link) and v2 becausesmp_store_release()shouldn't cause side effect. Considering the exactitude of writing code, v2 is a more preferable one.--- net/xdp/xsk.c | 3 +++ 1 file changed, 3 insertions(+)net/xdp/xsk.c +0
-0
net/xdp/xsk.c static int __xsk_generic_xmit(struct sock *sk) · line 864
Quoted 9 lines
net/xdp/xsk.c +3
-0@@ -856,6 +856,9 @@ static int __xsk_generic_xmit(struct sock *sk) } out: +if (xskq_has_descs (xs->tx)) +__xskq_cons_release (xs->tx); +if (sent_frame)if (xsk_tx_writeable (xs)) sk->sk_write_space (sk);
Hi Jason,
IMHO below should be enough to address the issue:
if (sent_frame) {
__xskq_cons_release(xs->tx);
if (xsk_tx_writeable(xs))
sk->sk_write_space(sk);
}which basically is what xsk_tx_release() does for each tx socket in list. zc drivers call it whenever there was a single descriptor produced to HW ring. So should we on generic xmit side, based on @sent_frame.
We could even wrap these 3 lines onto internal function, say __xsk_tx_release() and use it in xsk_tx_release() as well.
Quoted 3 lines
Signature · 1 line
2.41.3
Re: [PATCH net-next v3 1/2] net: xsk: update tx queue consumer immediately after transmission
Hi Maciej,
Quoted 66 lines · Maciej Fijalkowski
Quoted 64 lines · Jason Xing
From: Jason Xing <[email protected]>
For afxdp, the return value of
sendto()syscall doesn't reflect how many descs handled in the kernel. One of use cases is that when user-space application tries to know the number of transmitted skbs and then decides if it continues to send, say, is it stopped due to max tx budget?The following formular can be used after sending to learn how many skbs/descs the kernel takes care of:
tx_queue.consumers_before - tx_queue.consumers_after
Prior to the current patch, in non-zc mode, the consumer of tx queue is not immediately updated at the end of each sendto syscall when error occurs, which leads to the consumer value out-of-dated from the perspective of user space. So this patch requires store operation to pass the cached value to the shared value to handle the problem.
More than those explicit errors appearing in the
while()loop in__xsk_generic_xmit(), there are a few possible error cases that might be neglected in the following call trace:__xsk_generic_xmit()
xskq_cons_peek_desc()
xskq_cons_read_desc()
xskq_cons_is_valid_desc()
It will also cause the premature exit in thewhile()loop even if not all the descs are consumed.Based on the above analysis, using '
cached_prod!=cached_cons' could cover all the possible cases because it represents there are remaining descs that are not handled andcached_consare not updated to the global state of consumer at this time.Signed-off-by Jason Xing <[email protected]>---
v31. use
xskq_has_descshelper.
2. add selftestV2
1. filter out those good cases because only those that return error need updates.
Side note:
1. in non-batched zero copy mode, at the end of every caller ofxsk_tx_peek_desc(), there is always axsk_tx_release()function that used to update the local consumer to the global state of consumer. So for the zero copy mode, no need to change at all.
2. Actually I have no strong preference between v1 (see the above link) and v2 becausesmp_store_release()shouldn't cause side effect. Considering the exactitude of writing code, v2 is a more preferable one.--- net/xdp/xsk.c | 3 +++ 1 file changed, 3 insertions(+)net/xdp/xsk.c +0
-0@@ -856,6 +856,9 @@ static int __xsk_generic_xmit(struct sock *sk) }
Quoted 10 lines
net/xdp/xsk.c static int __xsk_generic_xmit(struct sock *sk) · line 864
Quoted 7 lines
net/xdp/xsk.c +3
-0out: +if (xskq_has_descs (xs->tx)) +__xskq_cons_release (xs->tx); +if (sent_frame)if (xsk_tx_writeable (xs)) sk->sk_write_space (sk);Hi Jason,
IMHO below should be enough to address the issue:
Sure, it can.
Can I ask one more thing? Technically it's not considered a bug, right? I'm not sure if it's worth telling the stable team to backport in older versions.
Quoted 2 lines
if (sent_frame) {Using this condition means the consumer is updated in majority cases including those good cases [1]. The intention of the current patch is to update the consumer only when the error occurs because in other cases xskq_cons_peek_desc() does it.
[1]: https://lore.kernel.org/all/aFVr60tw3QJopcOo@mini-arch/
Quoted 8 lines
__xskq_cons_release(xs->tx); if (xsk_tx_writeable(xs)) sk->sk_write_space(sk); }which basically is what
xsk_tx_release()does for each tx socket in list. zc drivers call it whenever there was a single descriptor produced to HW ring. So should we on generic xmit side, based on @sent_frame.
As you said, they would be the same :)
Quoted 3 lines
We could even wrap these 3 lines onto internal function, say
__xsk_tx_release()and use it inxsk_tx_release()as well.
I can do it in the next respin.
But I have no obvious opinion on how to write it. If no one is opposed to the taste of patch, I will follow your advice. Thanks.
Thanks,
Jason
Quoted 4 lines
Quoted 4 lines
--
2.41.3
PATCH v3 2/2 selftests/bpf: check if the global consumer of tx queue updates after send call
In reply to Jason XingFrom: Jason Xing <[email protected]>
The subtest sends 33 packets at one time on purpose to see if xsk exitting __xsk_generic_xmit() updates the global consumer of tx queue when reaching the max loop (max_tx_budget, 32 by default). The number 33 can avoid xskq_cons_peek_desc() updates the consumer, to accurately check if the issue that the first patch resolves remains.
Speaking of the selftest implementation, it's not possible to use the normal validation_func to check if the issue happens because the whole send packets logic will call the sendto multiple times such that we're unable to detect in time.
--- tools/testing/selftests/bpf/xskxceiver.c | 30 ++++++++++++++++++++++-- 1 file changed, 28 insertions(+), 2 deletions(-)
tools/testing/selftests/bpf/xskxceiver.c +28 -2
@@ -109,6 +109,8 @@#include <network_helpers.h> +#define MAX_TX_BUDGET_DEFAULT 32 +static bool opt_verbose;static bool opt_print_tests;static enum test_mode opt_mode =TEST_MODE_ALL ;
@@ -1323,7 +1325,8 @@ static int receive_pkts(struct test_spec *test)return TEST_PASS ; } -static int __send_pkts (struct ifobject *ifobject,struct xsk_socket_info *xsk,bool timeout) +static int __send_pkts (struct test_spec *test,struct ifobject *ifobject, +struct xsk_socket_info *xsk,bool timeout) {u32 i, idx =0 , valid_pkts =0 , valid_frags =0 , buffer_len;struct pkt_stream *pkt_stream = xsk->pkt_stream;
@@ -1437,9 +1440,21 @@ static int __send_pkts(struct ifobject *ifobject, struct xsk_socket_info *xsk, b }if (!timeout) { +int prev_tx_consumer; + +if (!strncmp ("TX_QUEUE_CONSUMER" , test->name,MAX_TEST_NAME_SIZE )) + prev_tx_consumer = *xsk->tx.consumer; +if (complete_pkts (xsk, i))return TEST_FAILURE ; +if (!strncmp ("TX_QUEUE_CONSUMER" , test->name,MAX_TEST_NAME_SIZE )) { +int delta = *xsk->tx.consumer - prev_tx_consumer; + +if (delta !=MAX_TX_BUDGET_DEFAULT ) +return TEST_FAILURE ; + } +usleep (10 );return TEST_PASS ; }
@@ -1492,7 +1507,7 @@ static int send_pkts(struct test_spec *test, struct ifobject *ifobject)__set_bit (i, bitmap);continue ; } - ret =__send_pkts (ifobject, &ifobject->xsk_arr[i], timeout); + ret =__send_pkts (test, ifobject, &ifobject->xsk_arr[i], timeout);if (ret ==TEST_CONTINUE && !test->fail)continue ;
@@ -2613,6 +2628,16 @@ static int testapp_adjust_tail_grow_mb(struct test_spec *test) XSK_UMEM__LARGE_FRAME_SIZE * 2); } +static int testapp_tx_queue_consumer (struct test_spec *test) +{ +int nr_packets =MAX_TX_BUDGET_DEFAULT +1 ; + +pkt_stream_replace (test, nr_packets,MIN_PKT_SIZE ); + test->ifobj_tx->xsk->batch_size = nr_packets; + +return testapp_validate_traffic (test); +} +static void run_pkt_test (struct test_spec *test) {int ret;
@@ -2723,6 +2748,7 @@ static const struct test_spec tests[] = { {.name = "XDP_ADJUST_TAIL_SHRINK_MULTI_BUFF", .test_func = testapp_adjust_tail_shrink_mb}, {.name = "XDP_ADJUST_TAIL_GROW", .test_func = testapp_adjust_tail_grow}, {.name = "XDP_ADJUST_TAIL_GROW_MULTI_BUFF", .test_func = testapp_adjust_tail_grow_mb}, + {.name = "TX_QUEUE_CONSUMER", .test_func = testapp_tx_queue_consumer}, };static void print_tests (void )
Signature · 1 line
2.41.3
Maciej Fijalkowski
maintainerRe: [PATCH net-next v3 2/2] selftests/bpf: check if the global consumer of tx queue updates after send call
Quoted 23 lines · Jason Xing
From: Jason Xing <[email protected]>
The subtest sends 33 packets at one time on purpose to see if xsk exitting
__xsk_generic_xmit()updates the global consumer of tx queue when reaching the max loop (max_tx_budget, 32 by default). The number 33 can avoidxskq_cons_peek_desc()updates the consumer, to accurately check if the issue that the first patch resolves remains.Speaking of the selftest implementation, it's not possible to use the normal
validation_functo check if the issue happens because the whole send packets logic will call the sendto multiple times such that we're unable to detect in time.Signed-off-by Jason Xing <[email protected]>--- tools/testing/selftests/bpf/xskxceiver.c | 30 ++++++++++++++++++++++-- 1 file changed, 28 insertions(+), 2 deletions(-)tools/testing/selftests/bpf/xskxceiver.c +0
-0
tools/testing/selftests/bpf/xskxceiver.c line 112
Quoted 4 lines
tools/testing/selftests/bpf/xskxceiver.c +1
-0@@ -109,6 +109,8 @@#include <network_helpers.h> +#define MAX_TX_BUDGET_DEFAULT 32
and what if in the future you would increase the generic xmit budget on the system? it would be better to wait with test addition when you introduce the setsockopt patch.
plus keep in mind that xskxceiver tests ZC drivers as well. so either we should have a test that serves all modes or keep it for skb mode only.
Quoted 18 lines
tools/testing/selftests/bpf/xskxceiver.c +3
-1+static bool opt_verbose;static bool opt_print_tests;static enum test_mode opt_mode =TEST_MODE_ALL ;@@ -1323,7 +1325,8 @@ static int receive_pkts(struct test_spec *test)return TEST_PASS ; } -static int __send_pkts (struct ifobject *ifobject,struct xsk_socket_info *xsk,bool timeout) +static int __send_pkts (struct test_spec *test,struct ifobject *ifobject, +struct xsk_socket_info *xsk,bool timeout) {u32 i, idx =0 , valid_pkts =0 , valid_frags =0 , buffer_len;struct pkt_stream *pkt_stream = xsk->pkt_stream;@@ -1437,9 +1440,21 @@ static int __send_pkts(struct ifobject *ifobject, struct xsk_socket_info *xsk, b }if (!timeout) {
tools/testing/selftests/bpf/xskxceiver.c static int __send_pkts(struct ifobject *ifobject, struct xsk_socket_info *xsk, b · line 1452
Quoted 10 lines
tools/testing/selftests/bpf/xskxceiver.c +7
-0+int prev_tx_consumer; + +if (!strncmp ("TX_QUEUE_CONSUMER" , test->name,MAX_TEST_NAME_SIZE )) + prev_tx_consumer = *xsk->tx.consumer; +if (complete_pkts (xsk, i))return TEST_FAILURE ; +if (!strncmp ("TX_QUEUE_CONSUMER" , test->name,MAX_TEST_NAME_SIZE )) { +int delta = *xsk->tx.consumer - prev_tx_consumer;
hacking the data path logic for single test purpose is rather not good. I am also not really sure if this deserves a standalone test case or could we just introduce a check in data path in appropriate place.
Quoted 45 lines
tools/testing/selftests/bpf/xskxceiver.c +17
-1+ +if (delta !=MAX_TX_BUDGET_DEFAULT ) +return TEST_FAILURE ; + } +usleep (10 );return TEST_PASS ; }@@ -1492,7 +1507,7 @@ static int send_pkts(struct test_spec *test, struct ifobject *ifobject)__set_bit (i, bitmap);continue ; } - ret =__send_pkts (ifobject, &ifobject->xsk_arr[i], timeout); + ret =__send_pkts (test, ifobject, &ifobject->xsk_arr[i], timeout);if (ret ==TEST_CONTINUE && !test->fail)continue ;@@ -2613,6 +2628,16 @@ static int testapp_adjust_tail_grow_mb(struct test_spec *test) XSK_UMEM__LARGE_FRAME_SIZE * 2); } +static int testapp_tx_queue_consumer (struct test_spec *test) +{ +int nr_packets =MAX_TX_BUDGET_DEFAULT +1 ; + +pkt_stream_replace (test, nr_packets,MIN_PKT_SIZE ); + test->ifobj_tx->xsk->batch_size = nr_packets; + +return testapp_validate_traffic (test); +} +static void run_pkt_test (struct test_spec *test) {int ret;@@ -2723,6 +2748,7 @@ static const struct test_spec tests[] = { {.name = "XDP_ADJUST_TAIL_SHRINK_MULTI_BUFF", .test_func = testapp_adjust_tail_shrink_mb}, {.name = "XDP_ADJUST_TAIL_GROW", .test_func = testapp_adjust_tail_grow}, {.name = "XDP_ADJUST_TAIL_GROW_MULTI_BUFF", .test_func = testapp_adjust_tail_grow_mb}, + {.name = "TX_QUEUE_CONSUMER", .test_func = testapp_tx_queue_consumer}, };static void print_tests (void )Signature · 1 line
2.41.3
Re: [PATCH net-next v3 2/2] selftests/bpf: check if the global consumer of tx queue updates after send call
Quoted 63 lines · Maciej Fijalkowski
Quoted 23 lines · Jason Xing
From: Jason Xing <[email protected]>
The subtest sends 33 packets at one time on purpose to see if xsk exitting
__xsk_generic_xmit()updates the global consumer of tx queue when reaching the max loop (max_tx_budget, 32 by default). The number 33 can avoidxskq_cons_peek_desc()updates the consumer, to accurately check if the issue that the first patch resolves remains.Speaking of the selftest implementation, it's not possible to use the normal
validation_functo check if the issue happens because the whole send packets logic will call the sendto multiple times such that we're unable to detect in time.Signed-off-by Jason Xing <[email protected]>--- tools/testing/selftests/bpf/xskxceiver.c | 30 ++++++++++++++++++++++-- 1 file changed, 28 insertions(+), 2 deletions(-)tools/testing/selftests/bpf/xskxceiver.c +0
-0tools/testing/selftests/bpf/xskxceiver.c line 112
Quoted 4 lines
tools/testing/selftests/bpf/xskxceiver.c +1
-0@@ -109,6 +109,8 @@#include <network_helpers.h> +#define MAX_TX_BUDGET_DEFAULT 32and what if in the future you would increase the generic xmit budget on the system? it would be better to wait with test addition when you introduce the setsockopt patch.
plus keep in mind that xskxceiver tests ZC drivers as well. so either we should have a test that serves all modes or keep it for skb mode only.
Quoted 26 lines
tools/testing/selftests/bpf/xskxceiver.c +8
-1+static bool opt_verbose;static bool opt_print_tests;static enum test_mode opt_mode =TEST_MODE_ALL ;@@ -1323,7 +1325,8 @@ static int receive_pkts(struct test_spec *test)return TEST_PASS ; } -static int __send_pkts (struct ifobject *ifobject,struct xsk_socket_info *xsk,bool timeout) +static int __send_pkts (struct test_spec *test,struct ifobject *ifobject, +struct xsk_socket_info *xsk,bool timeout) {u32 i, idx =0 , valid_pkts =0 , valid_frags =0 , buffer_len;struct pkt_stream *pkt_stream = xsk->pkt_stream;@@ -1437,9 +1440,21 @@ static int __send_pkts(struct ifobject *ifobject, struct xsk_socket_info *xsk, b }if (!timeout) { +int prev_tx_consumer; + +if (!strncmp ("TX_QUEUE_CONSUMER" , test->name,MAX_TEST_NAME_SIZE )) + prev_tx_consumer = *xsk->tx.consumer; +if (complete_pkts (xsk, i))return TEST_FAILURE ;
Quoted 6 lines
tools/testing/selftests/bpf/xskxceiver.c static int __send_pkts(struct ifobject *ifobject, struct xsk_socket_info *xsk, b · line 1452
Quoted 2 lines
tools/testing/selftests/bpf/xskxceiver.c +2
-0+if (!strncmp ("TX_QUEUE_CONSUMER" , test->name,MAX_TEST_NAME_SIZE )) { +int delta = *xsk->tx.consumer - prev_tx_consumer;hacking the data path logic for single test purpose is rather not good. I am also not really sure if this deserves a standalone test case or could we just introduce a check in data path in appropriate place.
The big headache is that if we expect to detect such a case, we have to re-invent a similar send packet logic or hack the data path (a bit like this patch). I admit it's ugly as I mentioned yesterday.
Sorry, Stanislav, no offense here. If you read this, please don't blame me. I know you wish me to add one related test case. So here we are. Since Maciej brought up the similar thought, I keep wondering if we should give up such a standalone test patch? Honestly it already involved more time than expected. The primary reason for me is that the issue doesn't cause much trouble to the application.
Thanks,
Jason
Quoted 46 lines
Quoted 46 lines
tools/testing/selftests/bpf/xskxceiver.c +17
-2+ +if (delta !=MAX_TX_BUDGET_DEFAULT ) +return TEST_FAILURE ; + } +usleep (10 );return TEST_PASS ; }@@ -1492,7 +1507,7 @@ static int send_pkts(struct test_spec *test, struct ifobject *ifobject)__set_bit (i, bitmap);continue ; } - ret =__send_pkts (ifobject, &ifobject->xsk_arr[i], timeout); + ret =__send_pkts (test, ifobject, &ifobject->xsk_arr[i], timeout);if (ret ==TEST_CONTINUE && !test->fail)continue ;@@ -2613,6 +2628,16 @@ static int testapp_adjust_tail_grow_mb(struct test_spec *test) XSK_UMEM__LARGE_FRAME_SIZE * 2); } +static int testapp_tx_queue_consumer (struct test_spec *test) +{ +int nr_packets =MAX_TX_BUDGET_DEFAULT +1 ; + +pkt_stream_replace (test, nr_packets,MIN_PKT_SIZE ); + test->ifobj_tx->xsk->batch_size = nr_packets; + +return testapp_validate_traffic (test); +} +static void run_pkt_test (struct test_spec *test) {int ret;@@ -2723,6 +2748,7 @@ static const struct test_spec tests[] = { {.name = "XDP_ADJUST_TAIL_SHRINK_MULTI_BUFF", .test_func = testapp_adjust_tail_shrink_mb}, {.name = "XDP_ADJUST_TAIL_GROW", .test_func = testapp_adjust_tail_grow}, {.name = "XDP_ADJUST_TAIL_GROW_MULTI_BUFF", .test_func = testapp_adjust_tail_grow_mb}, + {.name = "TX_QUEUE_CONSUMER", .test_func = testapp_tx_queue_consumer}, };static void print_tests (void ) --2.41.3
Re: [PATCH net-next v3 2/2] selftests/bpf: check if the global consumer of tx queue updates after send call
Quoted 28 lines · Jason Xing
Quoted 26 lines · Maciej Fijalkowski
Quoted 24 lines · Jason Xing
From: Jason Xing <[email protected]>
The subtest sends 33 packets at one time on purpose to see if xsk exitting
__xsk_generic_xmit()updates the global consumer of tx queue when reaching the max loop (max_tx_budget, 32 by default). The number 33 can avoidxskq_cons_peek_desc()updates the consumer, to accurately check if the issue that the first patch resolves remains.Speaking of the selftest implementation, it's not possible to use the normal
validation_functo check if the issue happens because the whole send packets logic will call the sendto multiple times such that we're unable to detect in time.Signed-off-by Jason Xing <[email protected]>--- tools/testing/selftests/bpf/xskxceiver.c | 30 ++++++++++++++++++++++-- 1 file changed, 28 insertions(+), 2 deletions(-)tools/testing/selftests/bpf/xskxceiver.c +0
-0@@ -109,6 +109,8 @@
Quoted 7 lines
Quoted 7 lines
tools/testing/selftests/bpf/xskxceiver.c line 112
Quoted 3 lines
tools/testing/selftests/bpf/xskxceiver.c +1
-0#include <network_helpers.h> +#define MAX_TX_BUDGET_DEFAULT 32and what if in the future you would increase the generic xmit budget on the system? it would be better to wait with test addition when you introduce the setsockopt patch.
We can always update it to follow new budget. The purpose of the test is to document/verify userspace expectations. Sincle even with the setsockopt we are still gonna have the default budget.
Quoted 40 lines
Quoted 29 lines
plus keep in mind that xskxceiver tests ZC drivers as well. so either we should have a test that serves all modes or keep it for skb mode only.
Quoted 26 lines
tools/testing/selftests/bpf/xskxceiver.c +8
-1+static bool opt_verbose;static bool opt_print_tests;static enum test_mode opt_mode =TEST_MODE_ALL ;@@ -1323,7 +1325,8 @@ static int receive_pkts(struct test_spec *test)return TEST_PASS ; } -static int __send_pkts (struct ifobject *ifobject,struct xsk_socket_info *xsk,bool timeout) +static int __send_pkts (struct test_spec *test,struct ifobject *ifobject, +struct xsk_socket_info *xsk,bool timeout) {u32 i, idx =0 , valid_pkts =0 , valid_frags =0 , buffer_len;struct pkt_stream *pkt_stream = xsk->pkt_stream;@@ -1437,9 +1440,21 @@ static int __send_pkts(struct ifobject *ifobject, struct xsk_socket_info *xsk, b }if (!timeout) { +int prev_tx_consumer; + +if (!strncmp ("TX_QUEUE_CONSUMER" , test->name,MAX_TEST_NAME_SIZE )) + prev_tx_consumer = *xsk->tx.consumer; +if (complete_pkts (xsk, i))return TEST_FAILURE ;Quoted 6 lines
tools/testing/selftests/bpf/xskxceiver.c static int __send_pkts(struct ifobject *ifobject, struct xsk_socket_info *xsk, b · line 1452
Quoted 2 lines
tools/testing/selftests/bpf/xskxceiver.c +2
-0+if (!strncmp ("TX_QUEUE_CONSUMER" , test->name,MAX_TEST_NAME_SIZE )) { +int delta = *xsk->tx.consumer - prev_tx_consumer;hacking the data path logic for single test purpose is rather not good. I am also not really sure if this deserves a standalone test case or could we just introduce a check in data path in appropriate place.
The big headache is that if we expect to detect such a case, we have to re-invent a similar send packet logic or hack the data path (a bit like this patch). I admit it's ugly as I mentioned yesterday.
Quoted 6 lines
Sorry, Stanislav, no offense here. If you read this, please don't blame me. I know you wish me to add one related test case. So here we are. Since Maciej brought up the similar thought, I keep wondering if we should give up such a standalone test patch? Honestly it already involved more time than expected. The primary reason for me is that the issue doesn't cause much trouble to the application.
IIUC, Maciej does not suggest to completely drop the test but rather to move this check (unconditionally and only for skb mode) somewhere into __send_pkts/complete_pkts to make sure the number of completed packets is always <= budget. Maciej correct me if I misread..
Re: [PATCH net-next v3 2/2] selftests/bpf: check if the global consumer of tx queue updates after send call
Quoted 89 lines · Stanislav Fomichev
Quoted 28 lines · Jason Xing
Quoted 26 lines · Maciej Fijalkowski
Quoted 24 lines · Jason Xing
From: Jason Xing <[email protected]>
The subtest sends 33 packets at one time on purpose to see if xsk exitting
__xsk_generic_xmit()updates the global consumer of tx queue when reaching the max loop (max_tx_budget, 32 by default). The number 33 can avoidxskq_cons_peek_desc()updates the consumer, to accurately check if the issue that the first patch resolves remains.Speaking of the selftest implementation, it's not possible to use the normal
validation_functo check if the issue happens because the whole send packets logic will call the sendto multiple times such that we're unable to detect in time.Signed-off-by Jason Xing <[email protected]>--- tools/testing/selftests/bpf/xskxceiver.c | 30 ++++++++++++++++++++++-- 1 file changed, 28 insertions(+), 2 deletions(-)tools/testing/selftests/bpf/xskxceiver.c +0
-0@@ -109,6 +109,8 @@Quoted 7 lines
Quoted 7 lines
tools/testing/selftests/bpf/xskxceiver.c line 112
Quoted 3 lines
tools/testing/selftests/bpf/xskxceiver.c +1
-0#include <network_helpers.h> +#define MAX_TX_BUDGET_DEFAULT 32and what if in the future you would increase the generic xmit budget on the system? it would be better to wait with test addition when you introduce the setsockopt patch.
We can always update it to follow new budget. The purpose of the test is to document/verify userspace expectations. Sincle even with the setsockopt we are still gonna have the default budget.
Quoted 40 lines
Quoted 29 lines
plus keep in mind that xskxceiver tests ZC drivers as well. so either we should have a test that serves all modes or keep it for skb mode only.
Quoted 26 lines
tools/testing/selftests/bpf/xskxceiver.c +8
-1+static bool opt_verbose;static bool opt_print_tests;static enum test_mode opt_mode =TEST_MODE_ALL ;@@ -1323,7 +1325,8 @@ static int receive_pkts(struct test_spec *test)return TEST_PASS ; } -static int __send_pkts (struct ifobject *ifobject,struct xsk_socket_info *xsk,bool timeout) +static int __send_pkts (struct test_spec *test,struct ifobject *ifobject, +struct xsk_socket_info *xsk,bool timeout) {u32 i, idx =0 , valid_pkts =0 , valid_frags =0 , buffer_len;struct pkt_stream *pkt_stream = xsk->pkt_stream;@@ -1437,9 +1440,21 @@ static int __send_pkts(struct ifobject *ifobject, struct xsk_socket_info *xsk, b }if (!timeout) { +int prev_tx_consumer; + +if (!strncmp ("TX_QUEUE_CONSUMER" , test->name,MAX_TEST_NAME_SIZE )) + prev_tx_consumer = *xsk->tx.consumer; +if (complete_pkts (xsk, i))return TEST_FAILURE ;Quoted 6 lines
tools/testing/selftests/bpf/xskxceiver.c static int __send_pkts(struct ifobject *ifobject, struct xsk_socket_info *xsk, b · line 1452
Quoted 2 lines
tools/testing/selftests/bpf/xskxceiver.c +2
-0+if (!strncmp ("TX_QUEUE_CONSUMER" , test->name,MAX_TEST_NAME_SIZE )) { +int delta = *xsk->tx.consumer - prev_tx_consumer;hacking the data path logic for single test purpose is rather not good. I am also not really sure if this deserves a standalone test case or could we just introduce a check in data path in appropriate place.
The big headache is that if we expect to detect such a case, we have to re-invent a similar send packet logic or hack the data path (a bit like this patch). I admit it's ugly as I mentioned yesterday.
Quoted 6 lines
Sorry, Stanislav, no offense here. If you read this, please don't blame me. I know you wish me to add one related test case. So here we are. Since Maciej brought up the similar thought, I keep wondering if we should give up such a standalone test patch? Honestly it already involved more time than expected. The primary reason for me is that the issue doesn't cause much trouble to the application.
Quoted 2 lines
IIUC, Maciej does not suggest to completely drop the test but rather to move this check (unconditionally and only for skb mode) somewhere
I prefer the former: make it suitable for all the cases. Whether it's zero copy mode or non-zc one, the behaviour of the consumer should be the same: update when finishing sending packets. I will give it a try again tomorrow since it's too late for me right now. Sorry.
Thanks,
Jason
Re: [PATCH net-next v3 2/2] selftests/bpf: check if the global consumer of tx queue updates after send call
Quoted 91 lines · Jason Xing
Quoted 89 lines · Stanislav Fomichev
Quoted 28 lines · Jason Xing
Quoted 26 lines · Maciej Fijalkowski
Quoted 24 lines · Jason Xing
From: Jason Xing <[email protected]>
The subtest sends 33 packets at one time on purpose to see if xsk exitting
__xsk_generic_xmit()updates the global consumer of tx queue when reaching the max loop (max_tx_budget, 32 by default). The number 33 can avoidxskq_cons_peek_desc()updates the consumer, to accurately check if the issue that the first patch resolves remains.Speaking of the selftest implementation, it's not possible to use the normal
validation_functo check if the issue happens because the whole send packets logic will call the sendto multiple times such that we're unable to detect in time.Signed-off-by Jason Xing <[email protected]>--- tools/testing/selftests/bpf/xskxceiver.c | 30 ++++++++++++++++++++++-- 1 file changed, 28 insertions(+), 2 deletions(-)tools/testing/selftests/bpf/xskxceiver.c +0
-0@@ -109,6 +109,8 @@Quoted 7 lines
Quoted 7 lines
tools/testing/selftests/bpf/xskxceiver.c line 112
Quoted 3 lines
tools/testing/selftests/bpf/xskxceiver.c +1
-0#include <network_helpers.h> +#define MAX_TX_BUDGET_DEFAULT 32and what if in the future you would increase the generic xmit budget on the system? it would be better to wait with test addition when you introduce the setsockopt patch.
We can always update it to follow new budget. The purpose of the test is to document/verify userspace expectations. Sincle even with the setsockopt we are still gonna have the default budget.
Quoted 46 lines
Quoted 29 lines
plus keep in mind that xskxceiver tests ZC drivers as well. so either we should have a test that serves all modes or keep it for skb mode only.
Quoted 26 lines
tools/testing/selftests/bpf/xskxceiver.c +8
-1+static bool opt_verbose;static bool opt_print_tests;static enum test_mode opt_mode =TEST_MODE_ALL ;@@ -1323,7 +1325,8 @@ static int receive_pkts(struct test_spec *test)return TEST_PASS ; } -static int __send_pkts (struct ifobject *ifobject,struct xsk_socket_info *xsk,bool timeout) +static int __send_pkts (struct test_spec *test,struct ifobject *ifobject, +struct xsk_socket_info *xsk,bool timeout) {u32 i, idx =0 , valid_pkts =0 , valid_frags =0 , buffer_len;struct pkt_stream *pkt_stream = xsk->pkt_stream;@@ -1437,9 +1440,21 @@ static int __send_pkts(struct ifobject *ifobject, struct xsk_socket_info *xsk, b }if (!timeout) { +int prev_tx_consumer; + +if (!strncmp ("TX_QUEUE_CONSUMER" , test->name,MAX_TEST_NAME_SIZE )) + prev_tx_consumer = *xsk->tx.consumer; +if (complete_pkts (xsk, i))return TEST_FAILURE ;Quoted 6 lines
tools/testing/selftests/bpf/xskxceiver.c static int __send_pkts(struct ifobject *ifobject, struct xsk_socket_info *xsk, b · line 1452
Quoted 2 lines
tools/testing/selftests/bpf/xskxceiver.c +2
-0+if (!strncmp ("TX_QUEUE_CONSUMER" , test->name,MAX_TEST_NAME_SIZE )) { +int delta = *xsk->tx.consumer - prev_tx_consumer;hacking the data path logic for single test purpose is rather not good. I am also not really sure if this deserves a standalone test case or could we just introduce a check in data path in appropriate place.
The big headache is that if we expect to detect such a case, we have to re-invent a similar send packet logic or hack the data path (a bit like this patch). I admit it's ugly as I mentioned yesterday.
Sorry, Stanislav, no offense here. If you read this, please don't blame me. I know you wish me to add one related test case. So here we are. Since Maciej brought up the similar thought, I keep wondering if we should give up such a standalone test patch? Honestly it already involved more time than expected. The primary reason for me is that the issue doesn't cause much trouble to the application.
Quoted 5 lines
Quoted 2 lines
IIUC, Maciej does not suggest to completely drop the test but rather to move this check (unconditionally and only for skb mode) somewhere
I prefer the former: make it suitable for all the cases. Whether it's zero copy mode or non-zc one, the behaviour of the consumer should be
But it does not work the same in zc and non-zc modes :-( There are a bunch of (uncodumented) quirks here and there. With a test case we can at least highlight/document them.
Re: [PATCH net-next v3 2/2] selftests/bpf: check if the global consumer of tx queue updates after send call
Quoted 96 lines · Stanislav Fomichev
Quoted 94 lines · Jason Xing
Quoted 91 lines · Stanislav Fomichev
Quoted 28 lines · Jason Xing
Quoted 26 lines · Maciej Fijalkowski
Quoted 24 lines · Jason Xing
From: Jason Xing <[email protected]>
The subtest sends 33 packets at one time on purpose to see if xsk exitting
__xsk_generic_xmit()updates the global consumer of tx queue when reaching the max loop (max_tx_budget, 32 by default). The number 33 can avoidxskq_cons_peek_desc()updates the consumer, to accurately check if the issue that the first patch resolves remains.Speaking of the selftest implementation, it's not possible to use the normal
validation_functo check if the issue happens because the whole send packets logic will call the sendto multiple times such that we're unable to detect in time.Signed-off-by Jason Xing <[email protected]>--- tools/testing/selftests/bpf/xskxceiver.c | 30 ++++++++++++++++++++++-- 1 file changed, 28 insertions(+), 2 deletions(-)tools/testing/selftests/bpf/xskxceiver.c +0
-0@@ -109,6 +109,8 @@Quoted 7 lines
Quoted 7 lines
tools/testing/selftests/bpf/xskxceiver.c line 112
Quoted 3 lines
tools/testing/selftests/bpf/xskxceiver.c +1
-0#include <network_helpers.h> +#define MAX_TX_BUDGET_DEFAULT 32and what if in the future you would increase the generic xmit budget on the system? it would be better to wait with test addition when you introduce the setsockopt patch.
We can always update it to follow new budget. The purpose of the test is to document/verify userspace expectations. Sincle even with the setsockopt we are still gonna have the default budget.
Quoted 40 lines
Quoted 29 lines
plus keep in mind that xskxceiver tests ZC drivers as well. so either we should have a test that serves all modes or keep it for skb mode only.
Quoted 26 lines
tools/testing/selftests/bpf/xskxceiver.c +8
-1+static bool opt_verbose;static bool opt_print_tests;static enum test_mode opt_mode =TEST_MODE_ALL ;@@ -1323,7 +1325,8 @@ static int receive_pkts(struct test_spec *test)return TEST_PASS ; } -static int __send_pkts (struct ifobject *ifobject,struct xsk_socket_info *xsk,bool timeout) +static int __send_pkts (struct test_spec *test,struct ifobject *ifobject, +struct xsk_socket_info *xsk,bool timeout) {u32 i, idx =0 , valid_pkts =0 , valid_frags =0 , buffer_len;struct pkt_stream *pkt_stream = xsk->pkt_stream;@@ -1437,9 +1440,21 @@ static int __send_pkts(struct ifobject *ifobject, struct xsk_socket_info *xsk, b }if (!timeout) { +int prev_tx_consumer; + +if (!strncmp ("TX_QUEUE_CONSUMER" , test->name,MAX_TEST_NAME_SIZE )) + prev_tx_consumer = *xsk->tx.consumer; +if (complete_pkts (xsk, i))return TEST_FAILURE ;Quoted 6 lines
tools/testing/selftests/bpf/xskxceiver.c static int __send_pkts(struct ifobject *ifobject, struct xsk_socket_info *xsk, b · line 1452
Quoted 2 lines
tools/testing/selftests/bpf/xskxceiver.c +2
-0+if (!strncmp ("TX_QUEUE_CONSUMER" , test->name,MAX_TEST_NAME_SIZE )) { +int delta = *xsk->tx.consumer - prev_tx_consumer;hacking the data path logic for single test purpose is rather not good. I am also not really sure if this deserves a standalone test case or could we just introduce a check in data path in appropriate place.
The big headache is that if we expect to detect such a case, we have to re-invent a similar send packet logic or hack the data path (a bit like this patch). I admit it's ugly as I mentioned yesterday.
Quoted 6 lines
Sorry, Stanislav, no offense here. If you read this, please don't blame me. I know you wish me to add one related test case. So here we are. Since Maciej brought up the similar thought, I keep wondering if we should give up such a standalone test patch? Honestly it already involved more time than expected. The primary reason for me is that the issue doesn't cause much trouble to the application.
IIUC, Maciej does not suggest to completely drop the test but rather to move this check (unconditionally and only for skb mode) somewhere
Quoted 6 lines
Quoted 2 lines
I prefer the former: make it suitable for all the cases. Whether it's zero copy mode or non-zc one, the behaviour of the consumer should be
But it does not work the same in zc and non-zc modes :-( There are a bunch of (uncodumented) quirks here and there. With a test case we can at least highlight/document them.
Oh, sorry, I didn't explain myself well. I was trying to say I would add in the data path so that the new test patch can easily recognize the problem for generic xmit path (in either mode just for compatibility).
Do you have any good advice to continue to do such a thing you mentioned?
Thanks,
Jason