Skip to contentSkip to conversation

riscv/ftrace: fix ftrace_modify_call bug

PATCH In review Song Shuai 3 participants 6 messages Nov 22, 2022 1 message not in the archive

Subsystems (2)
lore
Share & tools
lore

Song Shuai

author
#0 lore raw

PATCH 1/1 riscv/ftrace: fix ftrace_modify_call bug

With this commit (riscv: ftrace: Reduce the detour code size to half) patched, ftrace bug occurred When hosting kprobe and function tracer at the same function.

Obviously, the variable caller in ftrace_modify_call was assigned by rec->ip with 4 offset failing the code replacing at function entry. And the caller should be assigned by rec->ip directly to indicate the function entry.

The following is the ftrace bug log.

```
[  419.632855] 00000000f8776803: expected (ffe00297 1a4282e7) but got (1a8282e7 f0227179)
[  419.633390] ------------[ ftrace bug ]------------
[  419.633553] ftrace failed to modify
[  419.633569] [<ffffffff802091cc>] kernel_read+0x0/0x52
[  419.633863]  actual:   97:02:e0:ff:e7:82:82:1a
[  419.634087] Updating ftrace call site to call a different ftrace function
[  419.634279] ftrace record flags: e0000002
[  419.634487]  (2) R
[  419.634487]  expected tramp: ffffffff800093cc
[  419.634935] ------------[ cut here ]------------
```
---
 arch/riscv/kernel/ftrace.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/riscv/kernel/ftrace.c +1 -1
index 8c77f236fc71..61b24d767e2e 100644
@@ -132,7 +132,7 @@ int ftrace_modify_call(struct dyn_ftrace *rec, unsigned long old_addr,
 		       unsigned long addr)
 {
 	unsigned int call[2];
-	unsigned long caller = rec->ip + 4;
+	unsigned long caller = rec->ip;
 	int ret;
 
 	make_call_t0(caller, old_addr, call);
Signature · 1 line
2.20.1
#1 lore raw
Quoted 9 lines · Song Shuai

With this commit (riscv: ftrace: Reduce the detour code size to half) patched, ftrace bug occurred When hosting kprobe and function tracer at the same function.

Obviously, the variable caller in ftrace_modify_call was assigned by rec->ip with 4 offset failing the code replacing at function entry. And the caller should be assigned by rec->ip directly to indicate the function entry.

Thank you, it's my fault, but I think the problem is:

Before (riscv: ftrace: Reduce the detour code size to half)

     0: REG_S  ra, -SZREG(sp)
     4: auipc  ra, ?          <- We need "rec->ip + 4" here
     8: jalr   ?(ra)
    12: REG_L  ra, -SZREG(sp)

After (riscv: ftrace: Reduce the detour code size to half)

     0: auipc  t0, ?  <- We needn't "rec->ip + 4" anymore
     4: jalr   t0, ?(t0)

I copied rec->ip + 4 blindly, then caused the bug. Right?

Quoted 37 lines

The following is the ftrace bug log.

```
[  419.632855] 00000000f8776803: expected (ffe00297 1a4282e7) but got (1a8282e7 f0227179)
[  419.633390] ------------[ ftrace bug ]------------
[  419.633553] ftrace failed to modify
[  419.633569] [<ffffffff802091cc>] kernel_read+0x0/0x52
[  419.633863]  actual:   97:02:e0:ff:e7:82:82:1a
[  419.634087] Updating ftrace call site to call a different ftrace function
[  419.634279] ftrace record flags: e0000002
[  419.634487]  (2) R
[  419.634487]  expected tramp: ffffffff800093cc
[  419.634935] ------------[ cut here ]------------
```
---
 arch/riscv/kernel/ftrace.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/riscv/kernel/ftrace.c +1 -2
index 8c77f236fc71..61b24d767e2e 100644
@@ -132,7 +132,7 @@ int ftrace_modify_call(struct dyn_ftrace *rec, unsigned long old_addr,
                       unsigned long addr)
 {
        unsigned int call[2];
-       unsigned long caller = rec->ip + 4;
+       unsigned long caller = rec->ip;
        int ret;
 
        make_call_t0(caller, old_addr, call);
--

2.20.1

Signature · 2 lines
Best Regards
 Guo Ren
author
#2 lore raw
Quoted 19 lines · Guo Ren <[email protected]> 于2022年11月22日周二 08:57写道:
Quoted 9 lines · Song Shuai

With this commit (riscv: ftrace: Reduce the detour code size to half) patched, ftrace bug occurred When hosting kprobe and function tracer at the same function.

Obviously, the variable caller in ftrace_modify_call was assigned by rec->ip with 4 offset failing the code replacing at function entry. And the caller should be assigned by rec->ip directly to indicate the function entry.

Thank you, it's my fault, but I think the problem is:

Before (riscv: ftrace: Reduce the detour code size to half)

     0: REG_S  ra, -SZREG(sp)
     4: auipc  ra, ?          <- We need "rec->ip + 4" here
     8: jalr   ?(ra)
    12: REG_L  ra, -SZREG(sp)
Quoted 6 lines

After (riscv: ftrace: Reduce the detour code size to half)

     0: auipc  t0, ?  <- We needn't "rec->ip + 4" anymore
     4: jalr   t0, ?(t0)

I copied rec->ip + 4 blindly, then caused the bug. Right?

Yes, you're right.

Here's my simple test case for your convenience.

```
echo kernel_read > set_ftrace_filter
echo function > current_tracer
echo 'p:myp kernel_read' > kprobe_events
echo 1 > events/kprobes/myp/enable # ftrace bug
```
Quoted 33 lines
Quoted 33 lines

The following is the ftrace bug log.

```
[  419.632855] 00000000f8776803: expected (ffe00297 1a4282e7) but got (1a8282e7 f0227179)
[  419.633390] ------------[ ftrace bug ]------------
[  419.633553] ftrace failed to modify
[  419.633569] [<ffffffff802091cc>] kernel_read+0x0/0x52
[  419.633863]  actual:   97:02:e0:ff:e7:82:82:1a
[  419.634087] Updating ftrace call site to call a different ftrace function
[  419.634279] ftrace record flags: e0000002
[  419.634487]  (2) R
[  419.634487]  expected tramp: ffffffff800093cc
[  419.634935] ------------[ cut here ]------------
```
---
 arch/riscv/kernel/ftrace.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/riscv/kernel/ftrace.c +1 -1
index 8c77f236fc71..61b24d767e2e 100644
@@ -132,7 +132,7 @@ int ftrace_modify_call(struct dyn_ftrace *rec, unsigned long old_addr,
                       unsigned long addr)
 {
        unsigned int call[2];
-       unsigned long caller = rec->ip + 4;
+       unsigned long caller = rec->ip;
        int ret;
 
Quoted 9 lines
Quoted 4 lines
arch/riscv/kernel/ftrace.c +0 -1
        make_call_t0(caller, old_addr, call);
--

2.20.1

--
Best Regards
Guo Ren

Thanks,
Song

_______________________________________________
linux-riscv mailing list [email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv

A message that is not in the archive (sent off-list or not yet received): CAJF2gTT=r4VeWqpe+Tw2JDdTDU+r2mJiPmHq_i8eV5YRmrrLRA@mail.gmail.com

author
#4 lore raw
Quoted 41 lines · Guo Ren <[email protected]> 于2022年11月22日周二 09:29写道:
Quoted 27 lines · Song Shuai
Quoted 19 lines · Guo Ren <[email protected]> 于2022年11月22日周二 08:57写道:
Quoted 9 lines · Song Shuai

With this commit (riscv: ftrace: Reduce the detour code size to half) patched, ftrace bug occurred When hosting kprobe and function tracer at the same function.

Obviously, the variable caller in ftrace_modify_call was assigned by rec->ip with 4 offset failing the code replacing at function entry. And the caller should be assigned by rec->ip directly to indicate the function entry.

Thank you, it's my fault, but I think the problem is:

Before (riscv: ftrace: Reduce the detour code size to half)

     0: REG_S  ra, -SZREG(sp)
     4: auipc  ra, ?          <- We need "rec->ip + 4" here
     8: jalr   ?(ra)
    12: REG_L  ra, -SZREG(sp)
Quoted 6 lines

After (riscv: ftrace: Reduce the detour code size to half)

     0: auipc  t0, ?  <- We needn't "rec->ip + 4" anymore
     4: jalr   t0, ?(t0)

I copied rec->ip + 4 blindly, then caused the bug. Right?

Quoted 8 lines

Yes, you're right.

Here's my simple test case for your convenience. ```
echo kernel_read > set_ftrace_filter echo function > current_tracer
echo 'p:myp kernel_read' > kprobe_events echo 1 > events/kprobes/myp/enable # ftrace bug

Quoted 4 lines

Thx, I would fold the patch into (riscv: ftrace: Reduce the detour code size to half). And add Co-developed-by: Song Shuai <[email protected]> tag on that.

I would send the v3 series involving your patches.

That's an honor for me.

Thanks,
Song

Quoted 52 lines
Quoted 44 lines

```

Quoted 33 lines
Quoted 33 lines

The following is the ftrace bug log.

```
[  419.632855] 00000000f8776803: expected (ffe00297 1a4282e7) but got (1a8282e7 f0227179)
[  419.633390] ------------[ ftrace bug ]------------
[  419.633553] ftrace failed to modify
[  419.633569] [<ffffffff802091cc>] kernel_read+0x0/0x52
[  419.633863]  actual:   97:02:e0:ff:e7:82:82:1a
[  419.634087] Updating ftrace call site to call a different ftrace function
[  419.634279] ftrace record flags: e0000002
[  419.634487]  (2) R
[  419.634487]  expected tramp: ffffffff800093cc
[  419.634935] ------------[ cut here ]------------
```
---
 arch/riscv/kernel/ftrace.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/riscv/kernel/ftrace.c +1 -1
index 8c77f236fc71..61b24d767e2e 100644
@@ -132,7 +132,7 @@ int ftrace_modify_call(struct dyn_ftrace *rec, unsigned long old_addr,
                       unsigned long addr)
 {
        unsigned int call[2];
-       unsigned long caller = rec->ip + 4;
+       unsigned long caller = rec->ip;
        int ret;
 
Quoted 9 lines
Quoted 4 lines
arch/riscv/kernel/ftrace.c +0 -1
        make_call_t0(caller, old_addr, call);
--

2.20.1

--
Best Regards
Guo Ren

Quoted 2 lines

Thanks,
Song

--
Best Regards
Guo Ren

#5 lore raw
In reply to Song Shuai
Quoted 1 line · Song Shuai

With this commit (riscv: ftrace: Reduce the detour code size to half)

AFAICT the above patch has not been applied & this patch here should be folded into the offending patch?
I've marked this one as "Not Applicable" in patchwork as a result, but let me know if that is an incorrect assumption.

Thanks,
Conor.

Quoted 44 lines

patched, ftrace bug occurred When hosting kprobe and function tracer at the same function.

Obviously, the variable caller in ftrace_modify_call was assigned by rec->ip with 4 offset failing the code replacing at function entry. And the caller should be assigned by rec->ip directly to indicate the function entry.

The following is the ftrace bug log.

```
[  419.632855] 00000000f8776803: expected (ffe00297 1a4282e7) but got (1a8282e7 f0227179)
[  419.633390] ------------[ ftrace bug ]------------
[  419.633553] ftrace failed to modify
[  419.633569] [<ffffffff802091cc>] kernel_read+0x0/0x52
[  419.633863]  actual:   97:02:e0:ff:e7:82:82:1a
[  419.634087] Updating ftrace call site to call a different ftrace function
[  419.634279] ftrace record flags: e0000002
[  419.634487]  (2) R
[  419.634487]  expected tramp: ffffffff800093cc
[  419.634935] ------------[ cut here ]------------
```
---
 arch/riscv/kernel/ftrace.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/riscv/kernel/ftrace.c +1 -1
index 8c77f236fc71..61b24d767e2e 100644
@@ -132,7 +132,7 @@ int ftrace_modify_call(struct dyn_ftrace *rec, unsigned long old_addr,
 		       unsigned long addr)
 {
 	unsigned int call[2];
-	unsigned long caller = rec->ip + 4;
+	unsigned long caller = rec->ip;
 	int ret;
 
 	make_call_t0(caller, old_addr, call);
Signature · 1 line
2.20.1

_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv

author
#6 lore raw
Quoted 4 lines · Conor Dooley <[email protected]> 于2022年11月22日周二 08:57写道:
Quoted 1 line · Song Shuai

With this commit (riscv: ftrace: Reduce the detour code size to half)

Quoted 8 lines

AFAICT the above patch has not been applied & this patch here should be folded into the offending patch?
I've marked this one as "Not Applicable" in patchwork as a result, but let me know if that is an incorrect assumption.

Thanks,
Conor.

Hi, Conor:

Sorry to disturb you with this patch without against the merged commit list,

This patch actually is created for fixing (riscv: ftrace: Reduce the detour code size to half)
which has not been merged yet.

As Guo replied, he will fold it in the target patch. You can ignore this one. Sorry again.

BTW, for dispelling your confusion about my email name. Actually, it was misspelled when registering, but I keep it for daily use. So you can send it without concern. :P

-- Song

Sorry for bothering you with the

Quoted 40 lines
Quoted 40 lines

patched, ftrace bug occurred When hosting kprobe and function tracer at the same function.

Obviously, the variable caller in ftrace_modify_call was assigned by rec->ip with 4 offset failing the code replacing at function entry. And the caller should be assigned by rec->ip directly to indicate the function entry.

The following is the ftrace bug log.

```
[  419.632855] 00000000f8776803: expected (ffe00297 1a4282e7) but got (1a8282e7 f0227179)
[  419.633390] ------------[ ftrace bug ]------------
[  419.633553] ftrace failed to modify
[  419.633569] [<ffffffff802091cc>] kernel_read+0x0/0x52
[  419.633863]  actual:   97:02:e0:ff:e7:82:82:1a
[  419.634087] Updating ftrace call site to call a different ftrace function
[  419.634279] ftrace record flags: e0000002
[  419.634487]  (2) R
[  419.634487]  expected tramp: ffffffff800093cc
[  419.634935] ------------[ cut here ]------------
```
---
 arch/riscv/kernel/ftrace.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/riscv/kernel/ftrace.c +1 -1
index 8c77f236fc71..61b24d767e2e 100644
@@ -132,7 +132,7 @@ int ftrace_modify_call(struct dyn_ftrace *rec, unsigned long old_addr,
                     unsigned long addr)
 {
      unsigned int call[2];
-     unsigned long caller = rec->ip + 4;
+     unsigned long caller = rec->ip;
      int ret;
 
Quoted 4 lines
Quoted 4 lines
arch/riscv/kernel/ftrace.c +0 -1
      make_call_t0(caller, old_addr, call);
--

2.20.1

_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv