Skip to contentSkip to conversation

selftests/bpf: trampoline_count: use dedicated target function

PATCH Superseded Sun Jian 3 participants 5 messages Mar 15 – Mar 19

Subsystems (5)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Sun Jian

author
#0 lore raw

PATCH 1/1 selftests/bpf: trampoline_count: use dedicated target function

trampoline_count fills all trampoline attachment slots for a single target function and verifies that one extra attach fails with -E2BIG.

It currently targets bpf_modify_return_test, which is also used by modify_return and get_func_ip_test. When those tests run in parallel, they can contend for the same per-function trampoline quota and cause unexpected attach failures. This issue is currently masked by harness serialization.

Add a dedicated bpf_trampoline_count_test target and switch trampoline_count to use it. This keeps the test semantics unchanged while isolating it from other modify_return-based selftests, so it no longer needs to run in serial mode. Remove the TODO comment as well.

Tested:
./test_progs -t trampoline_count -vv ./test_progs -t modify_return -vv ./test_progs -t get_func_ip_test -vv ./test_progs -j$(nproc) -t trampoline_count -vv ./test_progs -j$(nproc) -t
trampoline_count,modify_return,get_func_ip_test -vv

---
 net/bpf/test_run.c                                        | 7 +++++++
 tools/testing/selftests/bpf/prog_tests/trampoline_count.c | 3 +--
 tools/testing/selftests/bpf/progs/test_trampoline_count.c | 6 +++---
 3 files changed, 11 insertions(+), 5 deletions(-)
net/bpf/test_run.c +7 -0
index 178c4738e63b..b41147a29f16 100644
@@ -580,6 +580,12 @@ __bpf_kfunc int bpf_modify_return_test2(int a, int *b, short c, int d,
 	return a + *b + c + d + (long)e + f + g;
 }
 
+__bpf_kfunc int bpf_trampoline_count_test(int a, int *b)
+{
+	*b += 1;
+	return a + *b;
+}
+
 __bpf_kfunc int bpf_modify_return_test_tp(int nonce)
 {
 	trace_bpf_trigger_tp(nonce);
@@ -635,6 +641,7 @@ BTF_KFUNCS_START(bpf_test_modify_return_ids)
 BTF_ID_FLAGS(func, bpf_modify_return_test)
 BTF_ID_FLAGS(func, bpf_modify_return_test2)
 BTF_ID_FLAGS(func, bpf_modify_return_test_tp)
+BTF_ID_FLAGS(func, bpf_trampoline_count_test)
 BTF_ID_FLAGS(func, bpf_fentry_test1, KF_SLEEPABLE)
 BTF_KFUNCS_END(bpf_test_modify_return_ids)
 
tools/testing/selftests/bpf/prog_tests/trampoline_count.c +1 -2
index 6cd7349d4a2b..b19358aed126 100644
@@ -30,8 +30,7 @@ static struct bpf_program *load_prog(char *file, char *name, struct inst *inst)
 	return prog;
 }
 
-/* TODO: use different target function to run in concurrent mode */
-void serial_test_trampoline_count(void)
+void test_trampoline_count(void)
 {
 	char *file = "test_trampoline_count.bpf.o";
 	char *const progs[] = { "fentry_test", "fmod_ret_test", "fexit_test" };
tools/testing/selftests/bpf/progs/test_trampoline_count.c +3 -3
index 7765720da7d5..911d282fd423 100644
@@ -3,19 +3,19 @@
 #include <bpf/bpf_helpers.h>
 #include <bpf/bpf_tracing.h>
 
-SEC("fentry/bpf_modify_return_test")
+SEC("fentry/bpf_trampoline_count_test")
 int BPF_PROG(fentry_test, int a, int *b)
 {
 	return 0;
 }
 
-SEC("fmod_ret/bpf_modify_return_test")
+SEC("fmod_ret/bpf_trampoline_count_test")
 int BPF_PROG(fmod_ret_test, int a, int *b, int ret)
 {
 	return 0;
 }
 
-SEC("fexit/bpf_modify_return_test")
+SEC("fexit/bpf_trampoline_count_test")
 int BPF_PROG(fexit_test, int a, int *b, int ret)
 {
 	return 0;

base-commit: 267594792a71018788af69e836c52e34bb8054af

Signature · 1 line
2.43.0
Tested-by on v1 1/1
#1 lore raw
Quoted 15 lines · Sun Jian

trampoline_count fills all trampoline attachment slots for a single target function and verifies that one extra attach fails with -E2BIG.

It currently targets bpf_modify_return_test, which is also used by modify_return and get_func_ip_test. When those tests run in parallel, they can contend for the same per-function trampoline quota and cause unexpected attach failures. This issue is currently masked by harness serialization.

Add a dedicated bpf_trampoline_count_test target and switch trampoline_count to use it. This keeps the test semantics unchanged while isolating it from other modify_return-based selftests, so it no longer needs to run in serial mode. Remove the TODO comment as well.

Quoted 7 lines

Tested:
./test_progs -t trampoline_count -vv ./test_progs -t modify_return -vv ./test_progs -t get_func_ip_test -vv ./test_progs -j$(nproc) -t trampoline_count -vv ./test_progs -j$(nproc) -t
trampoline_count,modify_return,get_func_ip_test -vv

I was able to reproduce the issue here then validate the fix. I however had to include several more tests that use bpf_modify_return_test (get_func_args_test, kprobe_multi_test, missed, and bpf_cookie) before I could reliably reproduce (in a loop that runs until it fails with an unexpected E2BIG).

[...]

author
#2 lore raw

PATCH v2 1/1 selftests/bpf: trampoline_count: use dedicated target function

In reply to Sun Jian

trampoline_count fills all trampoline attachment slots for a single target function and verifies that one extra attach fails with -E2BIG.

It currently targets bpf_modify_return_test, which is also used by modify_return and get_func_ip_test. When those tests run in parallel, they can contend for the same per-function trampoline quota and cause unexpected attach failures. This issue is currently masked by harness serialization.

Add a dedicated bpf_trampoline_count_test target and switch trampoline_count to use it. This keeps the test semantics unchanged while isolating it from other modify_return-based selftests, so it no longer needs to run in serial mode. Remove the TODO comment as well.

Tested:
./test_progs -t trampoline_count -vv ./test_progs -t modify_return -vv ./test_progs -t get_func_ip_test -vv ./test_progs -j$(nproc) -t trampoline_count -vv ./test_progs -j$(nproc) -t
trampoline_count,modify_return,get_func_ip_test -vv

---

v2:
- add Paul's Tested-by

 net/bpf/test_run.c                                        | 7 +++++++
 tools/testing/selftests/bpf/prog_tests/trampoline_count.c | 3 +--
 tools/testing/selftests/bpf/progs/test_trampoline_count.c | 6 +++---
 3 files changed, 11 insertions(+), 5 deletions(-)
net/bpf/test_run.c +7 -0
index 178c4738e63b..b41147a29f16 100644
@@ -580,6 +580,12 @@ __bpf_kfunc int bpf_modify_return_test2(int a, int *b, short c, int d,
 	return a + *b + c + d + (long)e + f + g;
 }
 
+__bpf_kfunc int bpf_trampoline_count_test(int a, int *b)
+{
+	*b += 1;
+	return a + *b;
+}
+
 __bpf_kfunc int bpf_modify_return_test_tp(int nonce)
 {
 	trace_bpf_trigger_tp(nonce);
@@ -635,6 +641,7 @@ BTF_KFUNCS_START(bpf_test_modify_return_ids)
 BTF_ID_FLAGS(func, bpf_modify_return_test)
 BTF_ID_FLAGS(func, bpf_modify_return_test2)
 BTF_ID_FLAGS(func, bpf_modify_return_test_tp)
+BTF_ID_FLAGS(func, bpf_trampoline_count_test)
 BTF_ID_FLAGS(func, bpf_fentry_test1, KF_SLEEPABLE)
 BTF_KFUNCS_END(bpf_test_modify_return_ids)
 
tools/testing/selftests/bpf/prog_tests/trampoline_count.c +1 -2
index 6cd7349d4a2b..b19358aed126 100644
@@ -30,8 +30,7 @@ static struct bpf_program *load_prog(char *file, char *name, struct inst *inst)
 	return prog;
 }
 
-/* TODO: use different target function to run in concurrent mode */
-void serial_test_trampoline_count(void)
+void test_trampoline_count(void)
 {
 	char *file = "test_trampoline_count.bpf.o";
 	char *const progs[] = { "fentry_test", "fmod_ret_test", "fexit_test" };
tools/testing/selftests/bpf/progs/test_trampoline_count.c +3 -3
index 7765720da7d5..911d282fd423 100644
@@ -3,19 +3,19 @@
 #include <bpf/bpf_helpers.h>
 #include <bpf/bpf_tracing.h>
 
-SEC("fentry/bpf_modify_return_test")
+SEC("fentry/bpf_trampoline_count_test")
 int BPF_PROG(fentry_test, int a, int *b)
 {
 	return 0;
 }
 
-SEC("fmod_ret/bpf_modify_return_test")
+SEC("fmod_ret/bpf_trampoline_count_test")
 int BPF_PROG(fmod_ret_test, int a, int *b, int ret)
 {
 	return 0;
 }
 
-SEC("fexit/bpf_modify_return_test")
+SEC("fexit/bpf_trampoline_count_test")
 int BPF_PROG(fexit_test, int a, int *b, int ret)
 {
 	return 0;

base-commit: 267594792a71018788af69e836c52e34bb8054af

Signature · 1 line
2.43.0
#3 lore raw
Quoted 23 lines · Sun Jian

trampoline_count fills all trampoline attachment slots for a single target function and verifies that one extra attach fails with -E2BIG.

It currently targets bpf_modify_return_test, which is also used by modify_return and get_func_ip_test. When those tests run in parallel, they can contend for the same per-function trampoline quota and cause unexpected attach failures. This issue is currently masked by harness serialization.

Add a dedicated bpf_trampoline_count_test target and switch trampoline_count to use it. This keeps the test semantics unchanged while isolating it from other modify_return-based selftests, so it no longer needs to run in serial mode. Remove the TODO comment as well.

Tested:
./test_progs -t trampoline_count -vv ./test_progs -t modify_return -vv ./test_progs -t get_func_ip_test -vv ./test_progs -j$(nproc) -t trampoline_count -vv ./test_progs -j$(nproc) -t
trampoline_count,modify_return,get_func_ip_test -vv

Quoted 6 lines

---

v2:
- add Paul's Tested-by

FYI, in general, unless there are merge conflicts, you don't need to resend just for the tags. Maintainers can apply those when they merge.

[...]

#4 lore raw
In reply to Sun Jian
Quoted 36 lines · Sun Jian

trampoline_count fills all trampoline attachment slots for a single target function and verifies that one extra attach fails with -E2BIG.

It currently targets bpf_modify_return_test, which is also used by modify_return and get_func_ip_test. When those tests run in parallel, they can contend for the same per-function trampoline quota and cause unexpected attach failures. This issue is currently masked by harness serialization.

Add a dedicated bpf_trampoline_count_test target and switch trampoline_count to use it. This keeps the test semantics unchanged while isolating it from other modify_return-based selftests, so it no longer needs to run in serial mode. Remove the TODO comment as well.

Tested:
./test_progs -t trampoline_count -vv ./test_progs -t modify_return -vv ./test_progs -t get_func_ip_test -vv ./test_progs -j$(nproc) -t trampoline_count -vv ./test_progs -j$(nproc) -t
trampoline_count,modify_return,get_func_ip_test -vv

---
 net/bpf/test_run.c                                        | 7 +++++++
 tools/testing/selftests/bpf/prog_tests/trampoline_count.c | 3 +--
 tools/testing/selftests/bpf/progs/test_trampoline_count.c | 6 +++---
 3 files changed, 11 insertions(+), 5 deletions(-)
net/bpf/test_run.c +0 -0
index 178c4738e63b..b41147a29f16 100644

net/bpf/test_run.c __bpf_kfunc int bpf_modify_return_test2(int a, int *b, short c, int d, · line 587

Quoted 8 lines
net/bpf/test_run.c +5 -0
@@ -580,6 +580,12 @@ __bpf_kfunc int bpf_modify_return_test2(int a, int *b, short c, int d,
        return a + *b + c + d + (long)e + f + g;
 }
 
+__bpf_kfunc int bpf_trampoline_count_test(int a, int *b)
+{
+       *b += 1;
+       return a + *b;
+}

No. We're not adding test code to the kernel anymore. That's what test_kmod is for.

pw-bot: cr