Skip to contentSkip to conversation

selftests/ftrace: Account for fprobe attachment at creation

PATCH Superseded Cao Ruichuang 2 participants 8 messages Apr 7 – Apr 13

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 selftests/ftrace: Account for fprobe attachment at creation

add_remove_fprobe.tc assumes that enabling an fprobe event is what adds its target function to enabled_functions.

On the current kernel, the fprobe target already appears in enabled_functions as soon as the event is created, and enabling the event does not change that count again. That makes the test fail even though the event lifecycle itself works.

Record the attachment baseline after creating the probe events and only check that enabling them keeps the expected functions attached. The cleanup checks still verify that removing the events returns enabled_functions to its original state.

---
 .../test.d/dynevent/add_remove_fprobe.tc      | 28 +++++++++++++------
 1 file changed, 19 insertions(+), 9 deletions(-)
tools/testing/selftests/ftrace/test.d/dynevent/add_remove_fprobe.tc +19 -9
index 47067a5e3..ff08bd1ac 100644
@@ -26,23 +26,29 @@ grep -q myevent2 dynamic_events
 grep -q myevent3 dynamic_events
 test -d events/fprobes/myevent1
 test -d events/fprobes/myevent2
-
-echo 1 > events/fprobes/myevent1/enable
-# Make sure the event is attached.
 grep -q $PLACE enabled_functions
+grep -q $PLACE2 enabled_functions
 cnt=`cat enabled_functions | wc -l`
-if [ $cnt -eq $ocnt ]; then
+if [ $cnt -le $ocnt ]; then
+	exit_fail
+fi
+
+echo 1 > events/fprobes/myevent1/enable
+cnt1=`cat enabled_functions | wc -l`
+if [ $cnt1 -ne $cnt ]; then
 	exit_fail
 fi
 
 echo 1 > events/fprobes/myevent2/enable
 cnt2=`cat enabled_functions | wc -l`
+if [ $cnt2 -ne $cnt1 ]; then
+	exit_fail
+fi
 
 echo 1 > events/fprobes/myevent3/enable
-# If the function is different, the attached function should be increased
 grep -q $PLACE2 enabled_functions
 cnt=`cat enabled_functions | wc -l`
-if [ $cnt -eq $cnt2 ]; then
+if [ $cnt -ne $cnt2 ]; then
 	exit_fail
 fi
 
@@ -62,11 +68,15 @@ if [ $cnt -ne $ocnt ]; then
 fi
 
 echo "f:myevent4 $PLACE" >> dynamic_events
+grep -q $PLACE enabled_functions
+cnt=`cat enabled_functions | wc -l`
+if [ $cnt -le $ocnt ]; then
+	exit_fail
+fi
 
 echo 1 > events/fprobes/myevent4/enable
-# Should only have one enabled
-cnt=`cat enabled_functions | wc -l`
-if [ $cnt -ne $((ocnt + 1)) ]; then
+cnt2=`cat enabled_functions | wc -l`
+if [ $cnt2 -ne $cnt ]; then
 	exit_fail
 fi
 
Signature · 1 line
2.39.5 (Apple Git-154)
#1 lore raw
Quoted 8 lines · On Tue, 7 Apr 2026 19:57:51 +0800 Cao Ruichuang <[email protected]> wrote:

add_remove_fprobe.tc assumes that enabling an fprobe event is what adds its target function to enabled_functions.

On the current kernel, the fprobe target already appears in enabled_functions as soon as the event is created, and enabling the event does not change that count again. That makes the test fail even though the event lifecycle itself works.

Quoted 5 lines

Record the attachment baseline after creating the probe events and only check that enabling them keeps the expected functions attached. The cleanup checks still verify that removing the events returns enabled_functions to its original state.

Hmm, did you run this test with not clean environment? It is not designed to do so.
(I think we should not check ocnt, or if there are any entries,
we should stop this test as unresolved.)

Thank you,

Quoted 67 lines
---
 .../test.d/dynevent/add_remove_fprobe.tc      | 28 +++++++++++++------
 1 file changed, 19 insertions(+), 9 deletions(-)
tools/testing/selftests/ftrace/test.d/dynevent/add_remove_fprobe.tc +19 -9
index 47067a5e3..ff08bd1ac 100644
@@ -26,23 +26,29 @@ grep -q myevent2 dynamic_events
 grep -q myevent3 dynamic_events
 test -d events/fprobes/myevent1
 test -d events/fprobes/myevent2
-
-echo 1 > events/fprobes/myevent1/enable
-# Make sure the event is attached.
 grep -q $PLACE enabled_functions
+grep -q $PLACE2 enabled_functions
 cnt=`cat enabled_functions | wc -l`
-if [ $cnt -eq $ocnt ]; then
+if [ $cnt -le $ocnt ]; then
+	exit_fail
+fi
+
+echo 1 > events/fprobes/myevent1/enable
+cnt1=`cat enabled_functions | wc -l`
+if [ $cnt1 -ne $cnt ]; then
 	exit_fail
 fi
 
 echo 1 > events/fprobes/myevent2/enable
 cnt2=`cat enabled_functions | wc -l`
+if [ $cnt2 -ne $cnt1 ]; then
+	exit_fail
+fi
 
 echo 1 > events/fprobes/myevent3/enable
-# If the function is different, the attached function should be increased
 grep -q $PLACE2 enabled_functions
 cnt=`cat enabled_functions | wc -l`
-if [ $cnt -eq $cnt2 ]; then
+if [ $cnt -ne $cnt2 ]; then
 	exit_fail
 fi
 
@@ -62,11 +68,15 @@ if [ $cnt -ne $ocnt ]; then
 fi
 
 echo "f:myevent4 $PLACE" >> dynamic_events
+grep -q $PLACE enabled_functions
+cnt=`cat enabled_functions | wc -l`
+if [ $cnt -le $ocnt ]; then
+	exit_fail
+fi
 
 echo 1 > events/fprobes/myevent4/enable
-# Should only have one enabled
-cnt=`cat enabled_functions | wc -l`
-if [ $cnt -ne $((ocnt + 1)) ]; then
+cnt2=`cat enabled_functions | wc -l`
+if [ $cnt2 -ne $cnt ]; then
 	exit_fail
 fi
 
Signature · 1 line
2.39.5 (Apple Git-154)
Signature · 1 line
Masami Hiramatsu (Google) <[email protected]>
#2 lore raw

Thanks, that is a good point.

I need to separate a clean source tree from a clean tracing runtime environment. I will rerun this in a fresh tracing environment and re-evaluate the testcase assumption before pushing anything further.

If this testcase is only valid for a clean environment, then an unresolved check or a different precondition will make more sense than the current patch.

Thanks,
Cao Ruichuang

#3 lore raw
In reply to Masami Hiramatsu (Google)

Hi Masami,

I reran this in a clean QEMU boot environment with only the minimal tracefs setup and the add_remove_fprobe sequence.

I saw the same behavior there: after creating myevent1/2/3, enabled_functions went from 2 to 4 before any event was enabled, and enabling myevent1/2/3 did not increase the count further. After cleanup, it returned to the original baseline again.

So this does not look like a dirty tracing environment issue. The failing testcase assumption appears to be that the attachment happens on enable, while on the current kernel it is already visible after create.

Thanks,
Cao Ruichuang

#4 lore raw
Quoted 9 lines · On Thu, 09 Apr 2026 13:18:19 +0800 Cao Ruichuang <[email protected]> wrote:

Hi Masami,

I reran this in a clean QEMU boot environment with only the minimal tracefs setup and the add_remove_fprobe sequence.

I saw the same behavior there: after creating myevent1/2/3, enabled_functions went from 2 to 4 before any event was enabled, and enabling myevent1/2/3 did not increase the count further. After cleanup, it returned to the original baseline again.

Hmm, that does not mean test failure but could be a bug. Can you dump the enabled_functions and share it when the test failed?

Quoted 3 lines

So this does not look like a dirty tracing environment issue. The failing testcase assumption appears to be that the attachment happens on enable, while on the current kernel it is already visible after create.

What kernel are you using? I need to check why this happens. I guess there is a change which introduces this issue.

Thank you,

Quoted 4 lines

Thanks,
Cao Ruichuang

Signature · 1 line
Masami Hiramatsu (Google) <[email protected]>
#5 lore raw

Hi Masami,

I reran this in clean QEMU on two kernels and got different results.

1. Ubuntu distro kernel:
Linux 6.8.0-100-generic #100-Ubuntu SMP PREEMPT_DYNAMIC Tue Jan 13 16:40:06 UTC 2026

baseline count=2
after_create123 count=4
after_enable1/2/3 count=4

baseline enabled_functions:

     __hid_bpf_tail_call ...

after_create123 enabled_functions:

     kernel_clone (2) R ->arch_ftrace_ops_list_func+0x0/0x280
     kmem_cache_free (1) R tramp: ... ->fprobe_handler+0x0/0x40
     __hid_bpf_tail_call ...

2. Current source-tree kernel built from the clean snapshot of my patch branch:
Linux 7.0.0-rc6 #2 SMP PREEMPT_DYNAMIC Fri Apr 10 12:19:39 CST 2026

baseline count=0
after_create123 count=0
after_enable1 count=1
after_enable2 count=1
after_enable3 count=2

after_create123 enabled_functions:

     <empty>

after_enable3 enabled_functions:

     kernel_clone (2) ->arch_ftrace_ops_list_func+0x0/0x200
     kmem_cache_free (1) tramp: ... ->fprobe_ftrace_entry+0x0/0x220

So the behavior I reported earlier reproduces on that Ubuntu 6.8 kernel, but not on the current source-tree kernel. I think my earlier conclusion was too broad.

I will stop pushing this testcase change for now unless I can narrow down which kernel change caused the difference.

Thanks,
Cao Ruichuang

#6 lore raw
Quoted 37 lines · On Fri, 10 Apr 2026 12:32:43 +0800 Cao Ruichuang <[email protected]> wrote:

Hi Masami,

I reran this in clean QEMU on two kernels and got different results.

1. Ubuntu distro kernel:

   Linux 6.8.0-100-generic #100-Ubuntu SMP PREEMPT_DYNAMIC
   Tue Jan 13 16:40:06 UTC 2026

   baseline count=2
   after_create123 count=4
   after_enable1/2/3 count=4

   baseline enabled_functions:
     __hid_bpf_tail_call ...

   after_create123 enabled_functions:
     kernel_clone (2) R ->arch_ftrace_ops_list_func+0x0/0x280
     kmem_cache_free (1) R tramp: ... ->fprobe_handler+0x0/0x40
     __hid_bpf_tail_call ...

2. Current source-tree kernel built from the clean snapshot of my patch branch:

   Linux 7.0.0-rc6 #2 SMP PREEMPT_DYNAMIC Fri Apr 10 12:19:39 CST 2026

   baseline count=0
   after_create123 count=0
   after_enable1 count=1
   after_enable2 count=1
   after_enable3 count=2

   after_create123 enabled_functions:
     <empty>

   after_enable3 enabled_functions:
     kernel_clone (2) ->arch_ftrace_ops_list_func+0x0/0x200
     kmem_cache_free (1) tramp: ... ->fprobe_ftrace_entry+0x0/0x220
Quoted 3 lines

So the behavior I reported earlier reproduces on that Ubuntu 6.8 kernel, but not on the current source-tree kernel. I think my earlier conclusion was too broad.

Thanks for reporting the difference of behaviors.

Quoted 3 lines

I will stop pushing this testcase change for now unless I can narrow down which kernel change caused the difference.

OK.

Thanks!

Quoted 4 lines

Thanks,
Cao Ruichuang

Signature · 1 line
Masami Hiramatsu (Google) <[email protected]>
#7 lore raw

PATCH v2 1/1 selftests/ftrace: Account for fprobe attachment before enable

In reply to Cao Ruichuang

add_remove_fprobe.tc assumes that enabling an fprobe event is what adds its target function to enabled_functions.

After resetting tracing state with initialize_system(), the current kernel still attaches the target functions as soon as the probe events are created. Enabling the three events does not increase the enabled_functions count further, and the original testcase fails on the check after enabling myevent3.

Record the baseline after creating the fprobe events and verify that subsequent enables keep that attachment set unchanged.

---
v2:
- rerun in a reset tracing environment after initialize_system()
- keep the fix focused on the failing myevent3 assumption and drop the unnecessary myevent4 changes

 .../ftrace/test.d/dynevent/add_remove_fprobe.tc | 17 ++++++++++++-----
 1 file changed, 12 insertions(+), 5 deletions(-)
tools/testing/selftests/ftrace/test.d/dynevent/add_remove_fprobe.tc +12 -5
index 47067a5e3..a55179f5d 100644
@@ -27,22 +27,29 @@ grep -q myevent3 dynamic_events
 test -d events/fprobes/myevent1
 test -d events/fprobes/myevent2
 
-echo 1 > events/fprobes/myevent1/enable
-# Make sure the event is attached.
 grep -q $PLACE enabled_functions
+grep -q $PLACE2 enabled_functions
 cnt=`cat enabled_functions | wc -l`
-if [ $cnt -eq $ocnt ]; then
+if [ $cnt -le $ocnt ]; then
+	exit_fail
+fi
+
+echo 1 > events/fprobes/myevent1/enable
+cnt1=`cat enabled_functions | wc -l`
+if [ $cnt1 -ne $cnt ]; then
 	exit_fail
 fi
 
 echo 1 > events/fprobes/myevent2/enable
 cnt2=`cat enabled_functions | wc -l`
+if [ $cnt2 -ne $cnt1 ]; then
+	exit_fail
+fi
 
 echo 1 > events/fprobes/myevent3/enable
-# If the function is different, the attached function should be increased
 grep -q $PLACE2 enabled_functions
 cnt=`cat enabled_functions | wc -l`
-if [ $cnt -eq $cnt2 ]; then
+if [ $cnt -ne $cnt2 ]; then
 	exit_fail
 fi
 
Signature · 1 line
2.39.5 (Apple Git-154)