selftests/bpf: fix flaky send_signal test
Subsystems (3)
cover selftests/bpf: fix flaky send_signal test
The bpf selftest send_signal() is flaky for its subtests trying to send signals in softirq/nmi context. To reduce flakiness, the signal-targetted process priority is boosted, which should minimize preemption of that process and improve the possibility that the underlying task in softirq/nmi context is the bpf_send_signal() wanted task.
Patch #1 did a refactoring to use ASSERT_* instead of old CHECK macros. Patch #2 did actual change of boosting priority.
Yonghong Song (2):
selftests/bpf: replace CHECK with ASSERT_* macros in send_signal.c selftests/bpf: fix flaky send_signal test
.../selftests/bpf/prog_tests/send_signal.c | 68 ++++++++++++------- .../bpf/progs/test_send_signal_kern.c | 3 +- 2 files changed, 43 insertions(+), 28 deletions(-)
Signature · 1 line
2.30.2
PATCH 1/2 selftests/bpf: replace CHECK with ASSERT_* macros in send_signal.c
Replace CHECK in send_signal.c with ASSERT_* macros as ASSERT_* macros are generally preferred. There is no funcitonality change.
--- .../selftests/bpf/prog_tests/send_signal.c | 45 +++++++++---------- 1 file changed, 20 insertions(+), 25 deletions(-)
tools/testing/selftests/bpf/prog_tests/send_signal.c +20 -25
@@ -10,29 +10,25 @@ static void sigusr1_handler(int signum) }static void test_send_signal_common (struct perf_event_attr *attr, -bool signal_thread, -const char *test_name) +bool signal_thread) {struct test_send_signal_kern *skel;int pipe_c2p[2 ], pipe_p2c[2 ];int err =-1 , pmu_fd =-1 ; -__u32 duration =0 ;char buf[256 ];pid_t pid; -if (CHECK (pipe (pipe_c2p), test_name, -"pipe pipe_c2p error: %s\n" ,strerror (errno))) +if (!ASSERT_OK (pipe (pipe_c2p),"pipe_c2p" ))return ; -if (CHECK (pipe (pipe_p2c), test_name, -"pipe pipe_p2c error: %s\n" ,strerror (errno))) { +if (!ASSERT_OK (pipe (pipe_p2c),"pipe_p2c" )) {close (pipe_c2p[0 ]);close (pipe_c2p[1 ]);return ; } pid =fork (); -if (CHECK (pid <0 , test_name,"fork error: %s\n" ,strerror (errno))) { +if (!ASSERT_GE (pid,0 ,"fork" )) {close (pipe_c2p[0 ]);close (pipe_c2p[1 ]);close (pipe_p2c[0 ]);
@@ -48,19 +44,19 @@ static void test_send_signal_common(struct perf_event_attr *attr,close (pipe_p2c[1 ]);/* close write */ /* notify parent signal handler is installed */ -CHECK (write (pipe_c2p[1 ], buf,1 ) !=1 ,"pipe_write" ,"err %d\n" , -errno); +ASSERT_EQ (write (pipe_c2p[1 ], buf,1 ),1 ,"pipe_write" );/* make sure parent enabled bpf program to send_signal */ -CHECK (read (pipe_p2c[0 ], buf,1 ) !=1 ,"pipe_read" ,"err %d\n" , -errno); +ASSERT_EQ (read (pipe_p2c[0 ], buf,1 ),1 ,"pipe_read" );/* wait a little for signal handler */ sleep (1 ); buf[0 ] = sigusr1_received ?'2' :'0' ; -CHECK (write (pipe_c2p[1 ], buf,1 ) !=1 ,"pipe_write" ,"err %d\n" , -errno); +ASSERT_EQ (write (pipe_c2p[1 ], buf,1 ),1 ,"pipe_write" );/* wait for parent notification and exit */ -CHECK (read (pipe_p2c[0 ], buf,1 ) !=1 ,"pipe_read" ,"err %d\n" , -errno); +ASSERT_EQ (read (pipe_p2c[0 ], buf,1 ),1 ,"pipe_read" );close (pipe_c2p[1 ]);close (pipe_p2c[0 ]);
@@ -71,20 +67,19 @@ static void test_send_signal_common(struct perf_event_attr *attr,close (pipe_p2c[0 ]);/* close read */ skel =test_send_signal_kern__open_and_load (); -if (CHECK (!skel,"skel_open_and_load" ,"skeleton open_and_load failed\n" )) +if (!ASSERT_OK_PTR (skel,"skel_open_and_load" )) goto skel_open_load_failure;if (!attr) { err =test_send_signal_kern__attach (skel); -if (CHECK (err,"skel_attach" ,"skeleton attach failed\n" )) { +if (!ASSERT_OK (err,"skel_attach" )) { err =-1 ; goto destroy_skel; } }else { pmu_fd =syscall (__NR_perf_event_open, attr, pid,-1 ,-1 /* group id */ ,0 /* flags */ ); -if (CHECK (pmu_fd <0 , test_name,"perf_event_open error: %s\n" , -strerror (errno))) { +if (!ASSERT_GE (pmu_fd,0 ,"perf_event_open" )) { err =-1 ; goto destroy_skel; }
@@ -96,7 +91,7 @@ static void test_send_signal_common(struct perf_event_attr *attr, }/* wait until child signal handler installed */ -CHECK (read (pipe_c2p[0 ], buf,1 ) !=1 ,"pipe_read" ,"err %d\n" , -errno); +ASSERT_EQ (read (pipe_c2p[0 ], buf,1 ),1 ,"pipe_read" );/* trigger the bpf send_signal */ skel->bss->pid = pid;
@@ -104,21 +99,21 @@ static void test_send_signal_common(struct perf_event_attr *attr, skel->bss->signal_thread = signal_thread;/* notify child that bpf program can send_signal now */ -CHECK (write (pipe_p2c[1 ], buf,1 ) !=1 ,"pipe_write" ,"err %d\n" , -errno); +ASSERT_EQ (write (pipe_p2c[1 ], buf,1 ),1 ,"pipe_write" );/* wait for result */ err =read (pipe_c2p[0 ], buf,1 ); -if (CHECK (err <0 , test_name,"reading pipe error: %s\n" ,strerror (errno))) +if (!ASSERT_GE (err,0 ,"reading pipe" )) goto disable_pmu; -if (CHECK (err ==0 , test_name,"reading pipe error: size 0\n" )) { +if (!ASSERT_GT (err,0 ,"reading pipe error: size 0" )) { err =-1 ; goto disable_pmu; } -CHECK (buf[0 ] !='2' , test_name,"incorrect result\n" ); +ASSERT_EQ (buf[0 ],'2' ,"incorrect result" );/* notify child safe to exit */ -CHECK (write (pipe_p2c[1 ], buf,1 ) !=1 ,"pipe_write" ,"err %d\n" , -errno); +ASSERT_EQ (write (pipe_p2c[1 ], buf,1 ),1 ,"pipe_write" ); disable_pmu:close (pmu_fd);
@@ -132,7 +127,7 @@ static void test_send_signal_common(struct perf_event_attr *attr,static void test_send_signal_tracepoint (bool signal_thread) { -test_send_signal_common (NULL , signal_thread,"tracepoint" ); +test_send_signal_common (NULL , signal_thread); }static void test_send_signal_perf (bool signal_thread)
@@ -143,7 +138,7 @@ static void test_send_signal_perf(bool signal_thread) .config = PERF_COUNT_SW_CPU_CLOCK, }; -test_send_signal_common (&attr, signal_thread,"perf_sw_event" ); +test_send_signal_common (&attr, signal_thread); }static void test_send_signal_nmi (bool signal_thread)
@@ -172,7 +167,7 @@ static void test_send_signal_nmi(bool signal_thread)close (pmu_fd); } -test_send_signal_common (&attr, signal_thread,"perf_hw_event" ); +test_send_signal_common (&attr, signal_thread); }void test_send_signal (void )
Signature · 1 line
2.30.2
PATCH 2/2 selftests/bpf: fix flaky send_signal test
In reply to Yonghong Songlibbpf CI has reported send_signal test is flaky although I am not able to reproduce it in my local environment. But I am able to reproduce with on-demand libbpf CI ([1]).
Through code analysis, the following is possible reason. The failed subtest runs bpf program in softirq environment. Since bpf_send_signal() only sends to a fork of "test_progs" process. If the underlying current task is
not "test_progs", bpf_send_signal() will not be triggered and the subtest will fail.
To reduce the chances where the underlying process is not the intended one, this patch boosted scheduling priority to -20 (highest allowed by setpriority() call). And I did 10 runs with on-demand libbpf CI with this patch and I didn't observe any failures.
[1] https://github.com/libbpf/libbpf/actions/workflows/ondemand.yml
--- .../selftests/bpf/prog_tests/send_signal.c | 33 +++++++++++++++---- .../bpf/progs/test_send_signal_kern.c | 3 +- 2 files changed, 28 insertions(+), 8 deletions(-)
tools/testing/selftests/bpf/prog_tests/send_signal.c +27 -6
@@ -1,5 +1,7 @@// SPDX-License-Identifier: GPL-2.0 #include <test_progs.h> +#include <sys/time.h> +#include <sys/resource.h> #include "test_send_signal_kern.skel.h" int sigusr1_received =0 ;
@@ -10,7 +12,7 @@ static void sigusr1_handler(int signum) }static void test_send_signal_common (struct perf_event_attr *attr, -bool signal_thread) +bool signal_thread,bool allow_skip) {struct test_send_signal_kern *skel;int pipe_c2p[2 ], pipe_p2c[2 ];
@@ -37,12 +39,23 @@ static void test_send_signal_common(struct perf_event_attr *attr, }if (pid ==0 ) { +int old_prio; +/* install signal handler and notify parent */ signal (SIGUSR1 , sigusr1_handler);close (pipe_c2p[0 ]);/* close read */ close (pipe_p2c[1 ]);/* close write */ +/* boost with a high priority so we got a higher chance +* that if an interrupt happens, the underlying task +* is this process. +*/ + errno =0 ; + old_prio =getpriority (PRIO_PROCESS ,0 ); +ASSERT_OK (errno,"getpriority" ); +ASSERT_OK (setpriority (PRIO_PROCESS ,0 ,-20 ),"setpriority" ); +/* notify parent signal handler is installed */ ASSERT_EQ (write (pipe_c2p[1 ], buf,1 ),1 ,"pipe_write" );
@@ -58,6 +71,9 @@ static void test_send_signal_common(struct perf_event_attr *attr,/* wait for parent notification and exit */ ASSERT_EQ (read (pipe_p2c[0 ], buf,1 ),1 ,"pipe_read" ); +/* restore the old priority */ +ASSERT_OK (setpriority (PRIO_PROCESS ,0 , old_prio),"setpriority" ); +close (pipe_c2p[1 ]);close (pipe_p2c[0 ]);exit (0 );
@@ -110,11 +126,16 @@ static void test_send_signal_common(struct perf_event_attr *attr, goto disable_pmu; } -ASSERT_EQ (buf[0 ],'2' ,"incorrect result" ); -/* notify child safe to exit */ ASSERT_EQ (write (pipe_p2c[1 ], buf,1 ),1 ,"pipe_write" ); +if (skel->bss->status ==0 && allow_skip) { +printf ("%s:SKIP\n" , __func__); +test__skip (); + }else if (skel->bss->status !=1 ) { +ASSERT_EQ (buf[0 ],'2' ,"incorrect result" ); + } + disable_pmu:close (pmu_fd); destroy_skel:
@@ -127,7 +148,7 @@ static void test_send_signal_common(struct perf_event_attr *attr,static void test_send_signal_tracepoint (bool signal_thread) { -test_send_signal_common (NULL , signal_thread); +test_send_signal_common (NULL , signal_thread, false); }static void test_send_signal_perf (bool signal_thread)
@@ -138,7 +159,7 @@ static void test_send_signal_perf(bool signal_thread) .config = PERF_COUNT_SW_CPU_CLOCK, }; -test_send_signal_common (&attr, signal_thread); +test_send_signal_common (&attr, signal_thread, true); }static void test_send_signal_nmi (bool signal_thread)
@@ -167,7 +188,7 @@ static void test_send_signal_nmi(bool signal_thread)close (pmu_fd); } -test_send_signal_common (&attr, signal_thread); +test_send_signal_common (&attr, signal_thread, true); }void test_send_signal (void )
tools/testing/selftests/bpf/progs/test_send_signal_kern.c +1 -2
@@ -18,8 +18,7 @@ static __always_inline int bpf_send_signal_test(void *ctx) ret =bpf_send_signal_thread (sig); else ret =bpf_send_signal (sig); -if (ret ==0 ) - status =1 ; + status = (ret ==0 ) ?1 :2 ; }return 0 ;
Signature · 1 line
2.30.2
Quoted 132 lines · Yonghong Song
libbpf CI has reported
send_signaltest is flaky although I am not able to reproduce it in my local environment. But I am able to reproduce with on-demand libbpf CI ([1]).Through code analysis, the following is possible reason. The failed subtest runs bpf program in softirq environment. Since
bpf_send_signal()only sends to a fork of "test_progs" process. If the underlying current task is
not "test_progs",bpf_send_signal()will not be triggered and the subtest will fail.To reduce the chances where the underlying process is not the intended one, this patch boosted scheduling priority to -20 (highest allowed by
setpriority()call). And I did 10 runs with on-demand libbpf CI with this patch and I didn't observe any failures.[1] https://github.com/libbpf/libbpf/actions/workflows/ondemand.yml
Signed-off-by Yonghong Song <[email protected]>--- .../selftests/bpf/prog_tests/send_signal.c | 33 +++++++++++++++---- .../bpf/progs/test_send_signal_kern.c | 3 +- 2 files changed, 28 insertions(+), 8 deletions(-)tools/testing/selftests/bpf/prog_tests/send_signal.c +27
-6@@ -1,5 +1,7 @@// SPDX-License-Identifier: GPL-2.0 #include <test_progs.h> +#include <sys/time.h> +#include <sys/resource.h> #include "test_send_signal_kern.skel.h" int sigusr1_received =0 ;@@ -10,7 +12,7 @@ static void sigusr1_handler(int signum) }static void test_send_signal_common (struct perf_event_attr *attr, -bool signal_thread) +bool signal_thread,bool allow_skip) {struct test_send_signal_kern *skel;int pipe_c2p[2 ], pipe_p2c[2 ];@@ -37,12 +39,23 @@ static void test_send_signal_common(struct perf_event_attr *attr, }if (pid ==0 ) { +int old_prio; +/* install signal handler and notify parent */ signal (SIGUSR1 , sigusr1_handler);close (pipe_c2p[0 ]);/* close read */ close (pipe_p2c[1 ]);/* close write */ +/* boost with a high priority so we got a higher chance +* that if an interrupt happens, the underlying task +* is this process. +*/ + errno =0 ; + old_prio =getpriority (PRIO_PROCESS ,0 ); +ASSERT_OK (errno,"getpriority" ); +ASSERT_OK (setpriority (PRIO_PROCESS ,0 ,-20 ),"setpriority" ); +/* notify parent signal handler is installed */ ASSERT_EQ (write (pipe_c2p[1 ], buf,1 ),1 ,"pipe_write" );@@ -58,6 +71,9 @@ static void test_send_signal_common(struct perf_event_attr *attr,/* wait for parent notification and exit */ ASSERT_EQ (read (pipe_p2c[0 ], buf,1 ),1 ,"pipe_read" ); +/* restore the old priority */ +ASSERT_OK (setpriority (PRIO_PROCESS ,0 , old_prio),"setpriority" ); +close (pipe_c2p[1 ]);close (pipe_p2c[0 ]);exit (0 );@@ -110,11 +126,16 @@ static void test_send_signal_common(struct perf_event_attr *attr, goto disable_pmu; } -ASSERT_EQ (buf[0 ],'2' ,"incorrect result" ); -/* notify child safe to exit */ ASSERT_EQ (write (pipe_p2c[1 ], buf,1 ),1 ,"pipe_write" ); +if (skel->bss->status ==0 && allow_skip) { +printf ("%s:SKIP\n" , __func__); +test__skip (); + }else if (skel->bss->status !=1 ) { +ASSERT_EQ (buf[0 ],'2' ,"incorrect result" ); + } + disable_pmu:close (pmu_fd); destroy_skel:@@ -127,7 +148,7 @@ static void test_send_signal_common(struct perf_event_attr *attr,static void test_send_signal_tracepoint (bool signal_thread) { -test_send_signal_common (NULL , signal_thread); +test_send_signal_common (NULL , signal_thread, false); }static void test_send_signal_perf (bool signal_thread)@@ -138,7 +159,7 @@ static void test_send_signal_perf(bool signal_thread) .config = PERF_COUNT_SW_CPU_CLOCK, }; -test_send_signal_common (&attr, signal_thread); +test_send_signal_common (&attr, signal_thread, true); }static void test_send_signal_nmi (bool signal_thread)@@ -167,7 +188,7 @@ static void test_send_signal_nmi(bool signal_thread)close (pmu_fd); } -test_send_signal_common (&attr, signal_thread); +test_send_signal_common (&attr, signal_thread, true); }void test_send_signal (void )tools/testing/selftests/bpf/progs/test_send_signal_kern.c +0
-0
tools/testing/selftests/bpf/progs/test_send_signal_kern.c static __always_inline int bpf_send_signal_test(void *ctx) · line 21
Quoted 6 lines
tools/testing/selftests/bpf/progs/test_send_signal_kern.c +1
-2@@ -18,8 +18,7 @@ static __always_inline int bpf_send_signal_test(void *ctx) ret =bpf_send_signal_thread (sig); else ret =bpf_send_signal (sig); -if (ret ==0 ) - status =1 ; + status = (ret ==0 ) ?1 :2 ;
This doesn't make sense to me. status == 0 is the default value, it will stay 0 even if nothing is triggered, no BPF program is called, etc.
If we are doing the skipping of the test logic (which I'd honestly just not do right now to see if we actually fixed the test), then I'd set status = 3 for the case when signal was triggered, but the current task is not test_progs. And only skip test if we get status 3. That is, status 0 and status 2 are bad (either not triggered, or some error when sending signal), 1 is OK, 3 is SKIP.
But really, skipping a test that we couldn't randomly run doesn't feel good. Can you please leave the priority boosting part and drop the skipping part for now?
Quoted 6 lines
tools/testing/selftests/bpf/progs/test_send_signal_kern.c +0
-1}return 0 ; --2.30.2
Quoted 127 lines · Andrii Nakryiko
Quoted 126 lines · Yonghong Song
libbpf CI has reported
send_signaltest is flaky although I am not able to reproduce it in my local environment. But I am able to reproduce with on-demand libbpf CI ([1]).Through code analysis, the following is possible reason. The failed subtest runs bpf program in softirq environment. Since
bpf_send_signal()only sends to a fork of "test_progs" process. If the underlying current task is
not "test_progs",bpf_send_signal()will not be triggered and the subtest will fail.To reduce the chances where the underlying process is not the intended one, this patch boosted scheduling priority to -20 (highest allowed by
setpriority()call). And I did 10 runs with on-demand libbpf CI with this patch and I didn't observe any failures.[1] https://github.com/libbpf/libbpf/actions/workflows/ondemand.yml
Signed-off-by Yonghong Song <[email protected]>--- .../selftests/bpf/prog_tests/send_signal.c | 33 +++++++++++++++---- .../bpf/progs/test_send_signal_kern.c | 3 +- 2 files changed, 28 insertions(+), 8 deletions(-)tools/testing/selftests/bpf/prog_tests/send_signal.c +27
-6@@ -1,5 +1,7 @@// SPDX-License-Identifier: GPL-2.0 #include <test_progs.h> +#include <sys/time.h> +#include <sys/resource.h> #include "test_send_signal_kern.skel.h" int sigusr1_received =0 ;@@ -10,7 +12,7 @@ static void sigusr1_handler(int signum) }static void test_send_signal_common (struct perf_event_attr *attr, -bool signal_thread) +bool signal_thread,bool allow_skip) {struct test_send_signal_kern *skel;int pipe_c2p[2 ], pipe_p2c[2 ];@@ -37,12 +39,23 @@ static void test_send_signal_common(struct perf_event_attr *attr, }if (pid ==0 ) { +int old_prio; +/* install signal handler and notify parent */ signal (SIGUSR1 , sigusr1_handler);close (pipe_c2p[0 ]);/* close read */ close (pipe_p2c[1 ]);/* close write */ +/* boost with a high priority so we got a higher chance +* that if an interrupt happens, the underlying task +* is this process. +*/ + errno =0 ; + old_prio =getpriority (PRIO_PROCESS ,0 ); +ASSERT_OK (errno,"getpriority" ); +ASSERT_OK (setpriority (PRIO_PROCESS ,0 ,-20 ),"setpriority" ); +/* notify parent signal handler is installed */ ASSERT_EQ (write (pipe_c2p[1 ], buf,1 ),1 ,"pipe_write" );@@ -58,6 +71,9 @@ static void test_send_signal_common(struct perf_event_attr *attr,/* wait for parent notification and exit */ ASSERT_EQ (read (pipe_p2c[0 ], buf,1 ),1 ,"pipe_read" ); +/* restore the old priority */ +ASSERT_OK (setpriority (PRIO_PROCESS ,0 , old_prio),"setpriority" ); +close (pipe_c2p[1 ]);close (pipe_p2c[0 ]);exit (0 );@@ -110,11 +126,16 @@ static void test_send_signal_common(struct perf_event_attr *attr, goto disable_pmu; } -ASSERT_EQ (buf[0 ],'2' ,"incorrect result" ); -/* notify child safe to exit */ ASSERT_EQ (write (pipe_p2c[1 ], buf,1 ),1 ,"pipe_write" ); +if (skel->bss->status ==0 && allow_skip) { +printf ("%s:SKIP\n" , __func__); +test__skip (); + }else if (skel->bss->status !=1 ) { +ASSERT_EQ (buf[0 ],'2' ,"incorrect result" ); + } + disable_pmu:close (pmu_fd); destroy_skel:@@ -127,7 +148,7 @@ static void test_send_signal_common(struct perf_event_attr *attr,static void test_send_signal_tracepoint (bool signal_thread) { -test_send_signal_common (NULL , signal_thread); +test_send_signal_common (NULL , signal_thread, false); }static void test_send_signal_perf (bool signal_thread)@@ -138,7 +159,7 @@ static void test_send_signal_perf(bool signal_thread) .config = PERF_COUNT_SW_CPU_CLOCK, }; -test_send_signal_common (&attr, signal_thread); +test_send_signal_common (&attr, signal_thread, true); }static void test_send_signal_nmi (bool signal_thread)@@ -167,7 +188,7 @@ static void test_send_signal_nmi(bool signal_thread)close (pmu_fd); } -test_send_signal_common (&attr, signal_thread); +test_send_signal_common (&attr, signal_thread, true); }
Quoted 16 lines
tools/testing/selftests/bpf/progs/test_send_signal_kern.c static __always_inline int bpf_send_signal_test(void *ctx) · line 21
Quoted 12 lines
tools/testing/selftests/bpf/prog_tests/send_signal.c +0
-0void test_send_signal (void )tools/testing/selftests/bpf/progs/test_send_signal_kern.c +1
-2@@ -18,8 +18,7 @@ static __always_inline int bpf_send_signal_test(void *ctx) ret =bpf_send_signal_thread (sig); else ret =bpf_send_signal (sig); -if (ret ==0 ) - status =1 ; + status = (ret ==0 ) ?1 :2 ;This doesn't make sense to me. status == 0 is the default value, it will stay 0 even if nothing is triggered, no BPF program is called, etc.
that is true.
Quoted 7 lines
If we are doing the skipping of the test logic (which I'd honestly just not do right now to see if we actually fixed the test), then I'd set status = 3 for the case when signal was triggered, but the current task is not
test_progs. And only skip test if we get status 3. That is, status 0 and status 2 are bad (either not triggered, or some error when sending signal), 1 is OK, 3 is SKIP.
Here, we *assume* bpf program always got called which should be the case unless softirq/nmi logic goes wrong. so status = 0 means pid doesn't match, and status = 1 means good bpf_send_signal happens,
status = 2 means bpf_send_signal helper fails.
Quoted 4 lines
But really, skipping a test that we couldn't randomly run doesn't feel good. Can you please leave the priority boosting part and drop the skipping part for now?
Sure. Let me drop skipping part. With the patch, I am expecting in *most* cases, we should not observe flakiness.
Quoted 7 lines
Quoted 7 lines
tools/testing/selftests/bpf/progs/test_send_signal_kern.c +0
-1}return 0 ; --2.30.2
Quoted 158 lines · Yonghong Song
Quoted 127 lines · Andrii Nakryiko
Quoted 126 lines · Yonghong Song
libbpf CI has reported
send_signaltest is flaky although I am not able to reproduce it in my local environment. But I am able to reproduce with on-demand libbpf CI ([1]).Through code analysis, the following is possible reason. The failed subtest runs bpf program in softirq environment. Since
bpf_send_signal()only sends to a fork of "test_progs" process. If the underlying current task is
not "test_progs",bpf_send_signal()will not be triggered and the subtest will fail.To reduce the chances where the underlying process is not the intended one, this patch boosted scheduling priority to -20 (highest allowed by
setpriority()call). And I did 10 runs with on-demand libbpf CI with this patch and I didn't observe any failures.[1] https://github.com/libbpf/libbpf/actions/workflows/ondemand.yml
Signed-off-by Yonghong Song <[email protected]>--- .../selftests/bpf/prog_tests/send_signal.c | 33 +++++++++++++++---- .../bpf/progs/test_send_signal_kern.c | 3 +- 2 files changed, 28 insertions(+), 8 deletions(-)tools/testing/selftests/bpf/prog_tests/send_signal.c +27
-6@@ -1,5 +1,7 @@// SPDX-License-Identifier: GPL-2.0 #include <test_progs.h> +#include <sys/time.h> +#include <sys/resource.h> #include "test_send_signal_kern.skel.h" int sigusr1_received =0 ;@@ -10,7 +12,7 @@ static void sigusr1_handler(int signum) }static void test_send_signal_common (struct perf_event_attr *attr, -bool signal_thread) +bool signal_thread,bool allow_skip) {struct test_send_signal_kern *skel;int pipe_c2p[2 ], pipe_p2c[2 ];@@ -37,12 +39,23 @@ static void test_send_signal_common(struct perf_event_attr *attr, }if (pid ==0 ) { +int old_prio; +/* install signal handler and notify parent */ signal (SIGUSR1 , sigusr1_handler);close (pipe_c2p[0 ]);/* close read */ close (pipe_p2c[1 ]);/* close write */ +/* boost with a high priority so we got a higher chance +* that if an interrupt happens, the underlying task +* is this process. +*/ + errno =0 ; + old_prio =getpriority (PRIO_PROCESS ,0 ); +ASSERT_OK (errno,"getpriority" ); +ASSERT_OK (setpriority (PRIO_PROCESS ,0 ,-20 ),"setpriority" ); +/* notify parent signal handler is installed */ ASSERT_EQ (write (pipe_c2p[1 ], buf,1 ),1 ,"pipe_write" );@@ -58,6 +71,9 @@ static void test_send_signal_common(struct perf_event_attr *attr,/* wait for parent notification and exit */ ASSERT_EQ (read (pipe_p2c[0 ], buf,1 ),1 ,"pipe_read" ); +/* restore the old priority */ +ASSERT_OK (setpriority (PRIO_PROCESS ,0 , old_prio),"setpriority" ); +close (pipe_c2p[1 ]);close (pipe_p2c[0 ]);exit (0 );@@ -110,11 +126,16 @@ static void test_send_signal_common(struct perf_event_attr *attr, goto disable_pmu; } -ASSERT_EQ (buf[0 ],'2' ,"incorrect result" ); -/* notify child safe to exit */ ASSERT_EQ (write (pipe_p2c[1 ], buf,1 ),1 ,"pipe_write" ); +if (skel->bss->status ==0 && allow_skip) { +printf ("%s:SKIP\n" , __func__); +test__skip (); + }else if (skel->bss->status !=1 ) { +ASSERT_EQ (buf[0 ],'2' ,"incorrect result" ); + } + disable_pmu:close (pmu_fd); destroy_skel:@@ -127,7 +148,7 @@ static void test_send_signal_common(struct perf_event_attr *attr,static void test_send_signal_tracepoint (bool signal_thread) { -test_send_signal_common (NULL , signal_thread); +test_send_signal_common (NULL , signal_thread, false); }static void test_send_signal_perf (bool signal_thread)@@ -138,7 +159,7 @@ static void test_send_signal_perf(bool signal_thread) .config = PERF_COUNT_SW_CPU_CLOCK, }; -test_send_signal_common (&attr, signal_thread); +test_send_signal_common (&attr, signal_thread, true); }static void test_send_signal_nmi (bool signal_thread)@@ -167,7 +188,7 @@ static void test_send_signal_nmi(bool signal_thread)close (pmu_fd); } -test_send_signal_common (&attr, signal_thread); +test_send_signal_common (&attr, signal_thread, true); }Quoted 16 lines
tools/testing/selftests/bpf/progs/test_send_signal_kern.c static __always_inline int bpf_send_signal_test(void *ctx) · line 21
Quoted 12 lines
tools/testing/selftests/bpf/prog_tests/send_signal.c +0
-0void test_send_signal (void )tools/testing/selftests/bpf/progs/test_send_signal_kern.c +1
-2@@ -18,8 +18,7 @@ static __always_inline int bpf_send_signal_test(void *ctx) ret =bpf_send_signal_thread (sig); else ret =bpf_send_signal (sig); -if (ret ==0 ) - status =1 ; + status = (ret ==0 ) ?1 :2 ;This doesn't make sense to me. status == 0 is the default value, it will stay 0 even if nothing is triggered, no BPF program is called, etc.
that is true.
Quoted 7 lines
If we are doing the skipping of the test logic (which I'd honestly just not do right now to see if we actually fixed the test), then I'd set status = 3 for the case when signal was triggered, but the current task is not
test_progs. And only skip test if we get status 3. That is, status 0 and status 2 are bad (either not triggered, or some error when sending signal), 1 is OK, 3 is SKIP.
Quoted 4 lines
Here, we *assume* bpf program always got called which should be the case unless softirq/nmi logic goes wrong. so status = 0 means
pid doesn't match, and status = 1 means goodbpf_send_signalhappens, status = 2 meansbpf_send_signalhelper fails.
Sorry, I didn't make my point clear. I meant that test shouldn't just assume that BPF program ran, so I'd add
if ((bpf_get_current_pid_tgid() >> 32) == pid) {
...
} else {
status = 3;
}Just to capture that we did get bpf_send_signal_test() called, but we didn't have correct current.
But it doesn't matter for now, I'd like to see if prio games get us to stable tests with no skipping first.
Quoted 8 lines
Quoted 5 lines
But really, skipping a test that we couldn't randomly run doesn't feel good. Can you please leave the priority boosting part and drop the skipping part for now?
Sure. Let me drop skipping part. With the patch, I am expecting in *most* cases, we should not observe flakiness.
Yep, thanks!
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
tools/testing/selftests/bpf/progs/test_send_signal_kern.c +0
-1}return 0 ; --2.30.2