[RFC PATCH bpf-next v2 0/2] Pass external callchain entry to get_perf_callchain

[RFC PATCH bpf-next v2 0/2] Pass external callchain entry to get_perf_callchain

[Tao Chen]

Background
==========
Alexei noted we should use preempt_disable to protect get_perf_callchain
in bpf stackmap.
https://lore.kernel.org/bpf/CAADnVQ+s8B7-fvR1TNO-bniSyKv57cH_ihRszmZV7pQDyV=VDQ@mail.gmail.com

A previous patch was submitted to attempt fixing this issue. And Andrii
suggested teach get_perf_callchain to let us pass that buffer directly to
avoid that unnecessary copy.
https://lore.kernel.org/bpf/20250926153952.1661146-1-chen.dylane@linux.dev

Proposed Solution
=================
Add external perf_callchain_entry parameter for get_perf_callchain to
allow us to use external buffer from BPF side. The biggest advantage is
that it can reduce unnecessary copies.

Todo
====
If the above changes are reasonable, it seems that get_callchain_entry_for_task
could also use an external perf_callchain_entry.

But I'm not sure if this modification is appropriate. After all, the
implementation of get_callchain_entry in the perf subsystem seems much more
complex than directly using an external buffer.

Comments and suggestions are always welcome.

Change list:
 - v1 -> v2
   From Jiri
   - rebase code, fix confict
 - v1: https://lore.kernel.org/bpf/20251013174721.2681091-1-chen.dylane@linux.dev

Tao Chen (2):
  perf: Use extern perf_callchain_entry for get_perf_callchain
  bpf: Pass external callchain entry to get_perf_callchain

 include/linux/perf_event.h |  4 ++--
 kernel/bpf/stackmap.c      | 19 +++++++++++--------
 kernel/events/callchain.c  | 13 +++++++++----
 kernel/events/core.c       |  2 +-
 4 files changed, 23 insertions(+), 15 deletions(-)

-- 
2.48.1

[RFC PATCH bpf-next v2 1/2] perf: Use extern perf_callchain_entry for get_perf_callchain

[Tao Chen]

From bpf stack map, we want to use our own buffers to avoid unnecessary
copy, so let us pass it directly. BPF will use this in the next patch.

Signed-off-by: Tao Chen <chen.dylane@linux.dev>
---
 include/linux/perf_event.h |  4 ++--
 kernel/bpf/stackmap.c      |  4 ++--
 kernel/events/callchain.c  | 13 +++++++++----
 kernel/events/core.c       |  2 +-
 4 files changed, 14 insertions(+), 9 deletions(-)

diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index fd1d91017b9..b144da7d803 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -1719,8 +1719,8 @@ DECLARE_PER_CPU(struct perf_callchain_entry, perf_callchain_entry);
 extern void perf_callchain_user(struct perf_callchain_entry_ctx *entry, struct pt_regs *regs);
 extern void perf_callchain_kernel(struct perf_callchain_entry_ctx *entry, struct pt_regs *regs);
 extern struct perf_callchain_entry *
-get_perf_callchain(struct pt_regs *regs, bool kernel, bool user,
-		   u32 max_stack, bool crosstask, bool add_mark);
+get_perf_callchain(struct pt_regs *regs, struct perf_callchain_entry *external_entry,
+		   bool kernel, bool user, u32 max_stack, bool crosstask, bool add_mark);
 extern int get_callchain_buffers(int max_stack);
 extern void put_callchain_buffers(void);
 extern struct perf_callchain_entry *get_callchain_entry(int *rctx);
diff --git a/kernel/bpf/stackmap.c b/kernel/bpf/stackmap.c
index 4d53cdd1374..94e46b7f340 100644
--- a/kernel/bpf/stackmap.c
+++ b/kernel/bpf/stackmap.c
@@ -314,7 +314,7 @@ BPF_CALL_3(bpf_get_stackid, struct pt_regs *, regs, struct bpf_map *, map,
 	if (max_depth > sysctl_perf_event_max_stack)
 		max_depth = sysctl_perf_event_max_stack;
 
-	trace = get_perf_callchain(regs, kernel, user, max_depth,
+	trace = get_perf_callchain(regs, NULL, kernel, user, max_depth,
 				   false, false);
 
 	if (unlikely(!trace))
@@ -451,7 +451,7 @@ static long __bpf_get_stack(struct pt_regs *regs, struct task_struct *task,
 	else if (kernel && task)
 		trace = get_callchain_entry_for_task(task, max_depth);
 	else
-		trace = get_perf_callchain(regs, kernel, user, max_depth,
+		trace = get_perf_callchain(regs, NULL, kernel, user, max_depth,
 					   crosstask, false);
 
 	if (unlikely(!trace) || trace->nr < skip) {
diff --git a/kernel/events/callchain.c b/kernel/events/callchain.c
index 808c0d7a31f..851e8f9d026 100644
--- a/kernel/events/callchain.c
+++ b/kernel/events/callchain.c
@@ -217,8 +217,8 @@ static void fixup_uretprobe_trampoline_entries(struct perf_callchain_entry *entr
 }
 
 struct perf_callchain_entry *
-get_perf_callchain(struct pt_regs *regs, bool kernel, bool user,
-		   u32 max_stack, bool crosstask, bool add_mark)
+get_perf_callchain(struct pt_regs *regs, struct perf_callchain_entry *external_entry,
+		   bool kernel, bool user, u32 max_stack, bool crosstask, bool add_mark)
 {
 	struct perf_callchain_entry *entry;
 	struct perf_callchain_entry_ctx ctx;
@@ -228,7 +228,11 @@ get_perf_callchain(struct pt_regs *regs, bool kernel, bool user,
 	if (crosstask && user && !kernel)
 		return NULL;
 
-	entry = get_callchain_entry(&rctx);
+	if (external_entry)
+		entry = external_entry;
+	else
+		entry = get_callchain_entry(&rctx);
+
 	if (!entry)
 		return NULL;
 
@@ -260,7 +264,8 @@ get_perf_callchain(struct pt_regs *regs, bool kernel, bool user,
 	}
 
 exit_put:
-	put_callchain_entry(rctx);
+	if (!external_entry)
+		put_callchain_entry(rctx);
 
 	return entry;
 }
diff --git a/kernel/events/core.c b/kernel/events/core.c
index 7541f6f85fc..5d8e146003a 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -8217,7 +8217,7 @@ perf_callchain(struct perf_event *event, struct pt_regs *regs)
 	if (!kernel && !user)
 		return &__empty_callchain;
 
-	callchain = get_perf_callchain(regs, kernel, user,
+	callchain = get_perf_callchain(regs, NULL, kernel, user,
 				       max_stack, crosstask, true);
 	return callchain ?: &__empty_callchain;
 }
-- 
2.48.1

[RFC PATCH bpf-next v2 2/2] bpf: Pass external callchain entry to get_perf_callchain

[Tao Chen]

As Alexei noted, get_perf_callchain() return values may be reused
if a task is preempted after the BPF program enters migrate disable
mode. Drawing on the per-cpu design of bpf_perf_callchain_entries,
stack-allocated memory of bpf_perf_callchain_entry is used here.

Signed-off-by: Tao Chen <chen.dylane@linux.dev>
---
 kernel/bpf/stackmap.c | 19 +++++++++++--------
 1 file changed, 11 insertions(+), 8 deletions(-)

diff --git a/kernel/bpf/stackmap.c b/kernel/bpf/stackmap.c
index 94e46b7f340..acd72c021c0 100644
--- a/kernel/bpf/stackmap.c
+++ b/kernel/bpf/stackmap.c
@@ -31,6 +31,11 @@ struct bpf_stack_map {
 	struct stack_map_bucket *buckets[] __counted_by(n_buckets);
 };
 
+struct bpf_perf_callchain_entry {
+	u64 nr;
+	u64 ip[PERF_MAX_STACK_DEPTH];
+};
+
 static inline bool stack_map_use_build_id(struct bpf_map *map)
 {
 	return (map->map_flags & BPF_F_STACK_BUILD_ID);
@@ -305,6 +310,7 @@ BPF_CALL_3(bpf_get_stackid, struct pt_regs *, regs, struct bpf_map *, map,
 	bool user = flags & BPF_F_USER_STACK;
 	struct perf_callchain_entry *trace;
 	bool kernel = !user;
+	struct bpf_perf_callchain_entry entry = { 0 };
 
 	if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
 			       BPF_F_FAST_STACK_CMP | BPF_F_REUSE_STACKID)))
@@ -314,12 +320,8 @@ BPF_CALL_3(bpf_get_stackid, struct pt_regs *, regs, struct bpf_map *, map,
 	if (max_depth > sysctl_perf_event_max_stack)
 		max_depth = sysctl_perf_event_max_stack;
 
-	trace = get_perf_callchain(regs, NULL, kernel, user, max_depth,
-				   false, false);
-
-	if (unlikely(!trace))
-		/* couldn't fetch the stack trace */
-		return -EFAULT;
+	trace = get_perf_callchain(regs, (struct perf_callchain_entry *)&entry,
+				   kernel, user, max_depth, false, false);
 
 	return __bpf_get_stackid(map, trace, flags);
 }
@@ -412,6 +414,7 @@ static long __bpf_get_stack(struct pt_regs *regs, struct task_struct *task,
 	u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
 	bool user = flags & BPF_F_USER_STACK;
 	struct perf_callchain_entry *trace;
+	struct bpf_perf_callchain_entry entry = { 0 };
 	bool kernel = !user;
 	int err = -EINVAL;
 	u64 *ips;
@@ -451,8 +454,8 @@ static long __bpf_get_stack(struct pt_regs *regs, struct task_struct *task,
 	else if (kernel && task)
 		trace = get_callchain_entry_for_task(task, max_depth);
 	else
-		trace = get_perf_callchain(regs, NULL, kernel, user, max_depth,
-					   crosstask, false);
+		trace = get_perf_callchain(regs, (struct perf_callchain_entry *)&entry,
+					   kernel, user, max_depth, crosstask, false);
 
 	if (unlikely(!trace) || trace->nr < skip) {
 		if (may_fault)
-- 
2.48.1

Re: [RFC PATCH bpf-next v2 2/2] bpf: Pass external callchain entry to get_perf_callchain

[Jiri Olsa]

On Tue, Oct 14, 2025 at 06:01:28PM +0800, Tao Chen wrote:
> As Alexei noted, get_perf_callchain() return values may be reused
> if a task is preempted after the BPF program enters migrate disable
> mode. Drawing on the per-cpu design of bpf_perf_callchain_entries,
> stack-allocated memory of bpf_perf_callchain_entry is used here.
> 
> Signed-off-by: Tao Chen <chen.dylane@linux.dev>
> ---
>  kernel/bpf/stackmap.c | 19 +++++++++++--------
>  1 file changed, 11 insertions(+), 8 deletions(-)
> 
> diff --git a/kernel/bpf/stackmap.c b/kernel/bpf/stackmap.c
> index 94e46b7f340..acd72c021c0 100644
> --- a/kernel/bpf/stackmap.c
> +++ b/kernel/bpf/stackmap.c
> @@ -31,6 +31,11 @@ struct bpf_stack_map {
>  	struct stack_map_bucket *buckets[] __counted_by(n_buckets);
>  };
>  
> +struct bpf_perf_callchain_entry {
> +	u64 nr;
> +	u64 ip[PERF_MAX_STACK_DEPTH];
> +};
> +
>  static inline bool stack_map_use_build_id(struct bpf_map *map)
>  {
>  	return (map->map_flags & BPF_F_STACK_BUILD_ID);
> @@ -305,6 +310,7 @@ BPF_CALL_3(bpf_get_stackid, struct pt_regs *, regs, struct bpf_map *, map,
>  	bool user = flags & BPF_F_USER_STACK;
>  	struct perf_callchain_entry *trace;
>  	bool kernel = !user;
> +	struct bpf_perf_callchain_entry entry = { 0 };

so IIUC having entries on stack we do not need to do preempt_disable
you had in the previous version, right?

I saw Andrii's justification to have this on the stack, I think it's
fine, but does it have to be initialized? it seems that only used
entries are copied to map

jirka

>  
>  	if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
>  			       BPF_F_FAST_STACK_CMP | BPF_F_REUSE_STACKID)))
> @@ -314,12 +320,8 @@ BPF_CALL_3(bpf_get_stackid, struct pt_regs *, regs, struct bpf_map *, map,
>  	if (max_depth > sysctl_perf_event_max_stack)
>  		max_depth = sysctl_perf_event_max_stack;
>  
> -	trace = get_perf_callchain(regs, NULL, kernel, user, max_depth,
> -				   false, false);
> -
> -	if (unlikely(!trace))
> -		/* couldn't fetch the stack trace */
> -		return -EFAULT;
> +	trace = get_perf_callchain(regs, (struct perf_callchain_entry *)&entry,
> +				   kernel, user, max_depth, false, false);
>  
>  	return __bpf_get_stackid(map, trace, flags);
>  }
> @@ -412,6 +414,7 @@ static long __bpf_get_stack(struct pt_regs *regs, struct task_struct *task,
>  	u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
>  	bool user = flags & BPF_F_USER_STACK;
>  	struct perf_callchain_entry *trace;
> +	struct bpf_perf_callchain_entry entry = { 0 };
>  	bool kernel = !user;
>  	int err = -EINVAL;
>  	u64 *ips;
> @@ -451,8 +454,8 @@ static long __bpf_get_stack(struct pt_regs *regs, struct task_struct *task,
>  	else if (kernel && task)
>  		trace = get_callchain_entry_for_task(task, max_depth);
>  	else
> -		trace = get_perf_callchain(regs, NULL, kernel, user, max_depth,
> -					   crosstask, false);
> +		trace = get_perf_callchain(regs, (struct perf_callchain_entry *)&entry,
> +					   kernel, user, max_depth, crosstask, false);
>  
>  	if (unlikely(!trace) || trace->nr < skip) {
>  		if (may_fault)
> -- 
> 2.48.1
> 

Re: [RFC PATCH bpf-next v2 2/2] bpf: Pass external callchain entry to get_perf_callchain

[Tao Chen]

在 2025/10/14 20:14, Jiri Olsa 写道:
>> +	struct bpf_perf_callchain_entry entry = { 0 };
> so IIUC having entries on stack we do not need to do preempt_disable
> you had in the previous version, right?
> 

Yes, i think so, preempt_disable seems unnecessary.

> I saw Andrii's justification to have this on the stack, I think it's
> fine, but does it have to be initialized? it seems that only used
> entries are copied to map

That makes sense. Removing it definitely looks better.

-- 
Best Regards
Tao Chen

Re: [RFC PATCH bpf-next v2 2/2] bpf: Pass external callchain entry to get_perf_callchain

[Alexei Starovoitov]

On Tue, Oct 14, 2025 at 5:14 AM Jiri Olsa <olsajiri@gmail.com> wrote:
>
> On Tue, Oct 14, 2025 at 06:01:28PM +0800, Tao Chen wrote:
> > As Alexei noted, get_perf_callchain() return values may be reused
> > if a task is preempted after the BPF program enters migrate disable
> > mode. Drawing on the per-cpu design of bpf_perf_callchain_entries,
> > stack-allocated memory of bpf_perf_callchain_entry is used here.
> >
> > Signed-off-by: Tao Chen <chen.dylane@linux.dev>
> > ---
> >  kernel/bpf/stackmap.c | 19 +++++++++++--------
> >  1 file changed, 11 insertions(+), 8 deletions(-)
> >
> > diff --git a/kernel/bpf/stackmap.c b/kernel/bpf/stackmap.c
> > index 94e46b7f340..acd72c021c0 100644
> > --- a/kernel/bpf/stackmap.c
> > +++ b/kernel/bpf/stackmap.c
> > @@ -31,6 +31,11 @@ struct bpf_stack_map {
> >       struct stack_map_bucket *buckets[] __counted_by(n_buckets);
> >  };
> >
> > +struct bpf_perf_callchain_entry {
> > +     u64 nr;
> > +     u64 ip[PERF_MAX_STACK_DEPTH];
> > +};
> > +
> >  static inline bool stack_map_use_build_id(struct bpf_map *map)
> >  {
> >       return (map->map_flags & BPF_F_STACK_BUILD_ID);
> > @@ -305,6 +310,7 @@ BPF_CALL_3(bpf_get_stackid, struct pt_regs *, regs, struct bpf_map *, map,
> >       bool user = flags & BPF_F_USER_STACK;
> >       struct perf_callchain_entry *trace;
> >       bool kernel = !user;
> > +     struct bpf_perf_callchain_entry entry = { 0 };
>
> so IIUC having entries on stack we do not need to do preempt_disable
> you had in the previous version, right?
>
> I saw Andrii's justification to have this on the stack, I think it's
> fine, but does it have to be initialized? it seems that only used
> entries are copied to map

No. We're not adding 1k stack consumption.

pw-bot: cr

Re: [RFC PATCH bpf-next v2 2/2] bpf: Pass external callchain entry to get_perf_callchain

[kernel test robot]

Hi Tao,

[This is a private test report for your RFC patch.]
kernel test robot noticed the following build warnings:

[auto build test WARNING on bpf-next/master]

url:    https://github.com/intel-lab-lkp/linux/commits/Tao-Chen/perf-Use-extern-perf_callchain_entry-for-get_perf_callchain/20251014-181245
base:   https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next.git master
patch link:    https://lore.kernel.org/r/20251014100128.2721104-3-chen.dylane%40linux.dev
patch subject: [RFC PATCH bpf-next v2 2/2] bpf: Pass external callchain entry to get_perf_callchain
config: sh-allyesconfig (https://download.01.org/0day-ci/archive/20251016/202510161955.CqDInkdX-lkp@intel.com/config)
compiler: sh4-linux-gcc (GCC) 15.1.0
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20251016/202510161955.CqDInkdX-lkp@intel.com/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <lkp@intel.com>
| Closes: https://lore.kernel.org/oe-kbuild-all/202510161955.CqDInkdX-lkp@intel.com/

All warnings (new ones prefixed by >>):

   In file included from kernel/bpf/stackmap.c:6:
   kernel/bpf/stackmap.c: In function 'bpf_get_stackid':
>> include/linux/filter.h:618:9: warning: the frame size of 1036 bytes is larger than 1024 bytes [-Wframe-larger-than=]
     618 |         }                                                                      \
         |         ^
   include/linux/filter.h:626:33: note: in expansion of macro 'BPF_CALL_x'
     626 | #define BPF_CALL_3(name, ...)   BPF_CALL_x(3, __NOATTR, name, __VA_ARGS__)
         |                                 ^~~~~~~~~~
   kernel/bpf/stackmap.c:305:1: note: in expansion of macro 'BPF_CALL_3'
     305 | BPF_CALL_3(bpf_get_stackid, struct pt_regs *, regs, struct bpf_map *, map,
         | ^~~~~~~~~~
   kernel/bpf/stackmap.c: In function '__bpf_get_stack':
>> kernel/bpf/stackmap.c:497:1: warning: the frame size of 1080 bytes is larger than 1024 bytes [-Wframe-larger-than=]
     497 | }
         | ^


vim +497 kernel/bpf/stackmap.c

d5a3b1f691865b Alexei Starovoitov 2016-02-17  304  
7b04d6d60fcfb5 Song Liu           2020-07-23 @305  BPF_CALL_3(bpf_get_stackid, struct pt_regs *, regs, struct bpf_map *, map,
7b04d6d60fcfb5 Song Liu           2020-07-23  306  	   u64, flags)
7b04d6d60fcfb5 Song Liu           2020-07-23  307  {
7b04d6d60fcfb5 Song Liu           2020-07-23  308  	u32 max_depth = map->value_size / stack_map_data_size(map);
ee2a098851bfbe Namhyung Kim       2022-03-14  309  	u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
7b04d6d60fcfb5 Song Liu           2020-07-23  310  	bool user = flags & BPF_F_USER_STACK;
7b04d6d60fcfb5 Song Liu           2020-07-23  311  	struct perf_callchain_entry *trace;
7b04d6d60fcfb5 Song Liu           2020-07-23  312  	bool kernel = !user;
8d90c803bbb115 Tao Chen           2025-10-14  313  	struct bpf_perf_callchain_entry entry = { 0 };
7b04d6d60fcfb5 Song Liu           2020-07-23  314  
7b04d6d60fcfb5 Song Liu           2020-07-23  315  	if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
7b04d6d60fcfb5 Song Liu           2020-07-23  316  			       BPF_F_FAST_STACK_CMP | BPF_F_REUSE_STACKID)))
7b04d6d60fcfb5 Song Liu           2020-07-23  317  		return -EINVAL;
7b04d6d60fcfb5 Song Liu           2020-07-23  318  
ee2a098851bfbe Namhyung Kim       2022-03-14  319  	max_depth += skip;
ee2a098851bfbe Namhyung Kim       2022-03-14  320  	if (max_depth > sysctl_perf_event_max_stack)
ee2a098851bfbe Namhyung Kim       2022-03-14  321  		max_depth = sysctl_perf_event_max_stack;
ee2a098851bfbe Namhyung Kim       2022-03-14  322  
8d90c803bbb115 Tao Chen           2025-10-14  323  	trace = get_perf_callchain(regs, (struct perf_callchain_entry *)&entry,
8d90c803bbb115 Tao Chen           2025-10-14  324  				   kernel, user, max_depth, false, false);
7b04d6d60fcfb5 Song Liu           2020-07-23  325  
7b04d6d60fcfb5 Song Liu           2020-07-23  326  	return __bpf_get_stackid(map, trace, flags);
7b04d6d60fcfb5 Song Liu           2020-07-23  327  }
7b04d6d60fcfb5 Song Liu           2020-07-23  328  
d5a3b1f691865b Alexei Starovoitov 2016-02-17  329  const struct bpf_func_proto bpf_get_stackid_proto = {
d5a3b1f691865b Alexei Starovoitov 2016-02-17  330  	.func		= bpf_get_stackid,
d5a3b1f691865b Alexei Starovoitov 2016-02-17  331  	.gpl_only	= true,
d5a3b1f691865b Alexei Starovoitov 2016-02-17  332  	.ret_type	= RET_INTEGER,
d5a3b1f691865b Alexei Starovoitov 2016-02-17  333  	.arg1_type	= ARG_PTR_TO_CTX,
d5a3b1f691865b Alexei Starovoitov 2016-02-17  334  	.arg2_type	= ARG_CONST_MAP_PTR,
d5a3b1f691865b Alexei Starovoitov 2016-02-17  335  	.arg3_type	= ARG_ANYTHING,
d5a3b1f691865b Alexei Starovoitov 2016-02-17  336  };
d5a3b1f691865b Alexei Starovoitov 2016-02-17  337  
7b04d6d60fcfb5 Song Liu           2020-07-23  338  static __u64 count_kernel_ip(struct perf_callchain_entry *trace)
7b04d6d60fcfb5 Song Liu           2020-07-23  339  {
7b04d6d60fcfb5 Song Liu           2020-07-23  340  	__u64 nr_kernel = 0;
7b04d6d60fcfb5 Song Liu           2020-07-23  341  
7b04d6d60fcfb5 Song Liu           2020-07-23  342  	while (nr_kernel < trace->nr) {
7b04d6d60fcfb5 Song Liu           2020-07-23  343  		if (trace->ip[nr_kernel] == PERF_CONTEXT_USER)
7b04d6d60fcfb5 Song Liu           2020-07-23  344  			break;
7b04d6d60fcfb5 Song Liu           2020-07-23  345  		nr_kernel++;
7b04d6d60fcfb5 Song Liu           2020-07-23  346  	}
7b04d6d60fcfb5 Song Liu           2020-07-23  347  	return nr_kernel;
7b04d6d60fcfb5 Song Liu           2020-07-23  348  }
7b04d6d60fcfb5 Song Liu           2020-07-23  349  
7b04d6d60fcfb5 Song Liu           2020-07-23  350  BPF_CALL_3(bpf_get_stackid_pe, struct bpf_perf_event_data_kern *, ctx,
7b04d6d60fcfb5 Song Liu           2020-07-23  351  	   struct bpf_map *, map, u64, flags)
7b04d6d60fcfb5 Song Liu           2020-07-23  352  {
7b04d6d60fcfb5 Song Liu           2020-07-23  353  	struct perf_event *event = ctx->event;
7b04d6d60fcfb5 Song Liu           2020-07-23  354  	struct perf_callchain_entry *trace;
7b04d6d60fcfb5 Song Liu           2020-07-23  355  	bool kernel, user;
7b04d6d60fcfb5 Song Liu           2020-07-23  356  	__u64 nr_kernel;
7b04d6d60fcfb5 Song Liu           2020-07-23  357  	int ret;
7b04d6d60fcfb5 Song Liu           2020-07-23  358  
7b04d6d60fcfb5 Song Liu           2020-07-23  359  	/* perf_sample_data doesn't have callchain, use bpf_get_stackid */
16817ad7e8b317 Namhyung Kim       2022-09-08  360  	if (!(event->attr.sample_type & PERF_SAMPLE_CALLCHAIN))
7b04d6d60fcfb5 Song Liu           2020-07-23  361  		return bpf_get_stackid((unsigned long)(ctx->regs),
7b04d6d60fcfb5 Song Liu           2020-07-23  362  				       (unsigned long) map, flags, 0, 0);
7b04d6d60fcfb5 Song Liu           2020-07-23  363  
7b04d6d60fcfb5 Song Liu           2020-07-23  364  	if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
7b04d6d60fcfb5 Song Liu           2020-07-23  365  			       BPF_F_FAST_STACK_CMP | BPF_F_REUSE_STACKID)))
7b04d6d60fcfb5 Song Liu           2020-07-23  366  		return -EINVAL;
7b04d6d60fcfb5 Song Liu           2020-07-23  367  
7b04d6d60fcfb5 Song Liu           2020-07-23  368  	user = flags & BPF_F_USER_STACK;
7b04d6d60fcfb5 Song Liu           2020-07-23  369  	kernel = !user;
7b04d6d60fcfb5 Song Liu           2020-07-23  370  
7b04d6d60fcfb5 Song Liu           2020-07-23  371  	trace = ctx->data->callchain;
7b04d6d60fcfb5 Song Liu           2020-07-23  372  	if (unlikely(!trace))
7b04d6d60fcfb5 Song Liu           2020-07-23  373  		return -EFAULT;
7b04d6d60fcfb5 Song Liu           2020-07-23  374  
7b04d6d60fcfb5 Song Liu           2020-07-23  375  	nr_kernel = count_kernel_ip(trace);
7b04d6d60fcfb5 Song Liu           2020-07-23  376  
7b04d6d60fcfb5 Song Liu           2020-07-23  377  	if (kernel) {
7b04d6d60fcfb5 Song Liu           2020-07-23  378  		__u64 nr = trace->nr;
7b04d6d60fcfb5 Song Liu           2020-07-23  379  
7b04d6d60fcfb5 Song Liu           2020-07-23  380  		trace->nr = nr_kernel;
7b04d6d60fcfb5 Song Liu           2020-07-23  381  		ret = __bpf_get_stackid(map, trace, flags);
7b04d6d60fcfb5 Song Liu           2020-07-23  382  
7b04d6d60fcfb5 Song Liu           2020-07-23  383  		/* restore nr */
7b04d6d60fcfb5 Song Liu           2020-07-23  384  		trace->nr = nr;
7b04d6d60fcfb5 Song Liu           2020-07-23  385  	} else { /* user */
7b04d6d60fcfb5 Song Liu           2020-07-23  386  		u64 skip = flags & BPF_F_SKIP_FIELD_MASK;
7b04d6d60fcfb5 Song Liu           2020-07-23  387  
7b04d6d60fcfb5 Song Liu           2020-07-23  388  		skip += nr_kernel;
7b04d6d60fcfb5 Song Liu           2020-07-23  389  		if (skip > BPF_F_SKIP_FIELD_MASK)
7b04d6d60fcfb5 Song Liu           2020-07-23  390  			return -EFAULT;
7b04d6d60fcfb5 Song Liu           2020-07-23  391  
7b04d6d60fcfb5 Song Liu           2020-07-23  392  		flags = (flags & ~BPF_F_SKIP_FIELD_MASK) | skip;
7b04d6d60fcfb5 Song Liu           2020-07-23  393  		ret = __bpf_get_stackid(map, trace, flags);
7b04d6d60fcfb5 Song Liu           2020-07-23  394  	}
7b04d6d60fcfb5 Song Liu           2020-07-23  395  	return ret;
7b04d6d60fcfb5 Song Liu           2020-07-23  396  }
7b04d6d60fcfb5 Song Liu           2020-07-23  397  
7b04d6d60fcfb5 Song Liu           2020-07-23  398  const struct bpf_func_proto bpf_get_stackid_proto_pe = {
7b04d6d60fcfb5 Song Liu           2020-07-23  399  	.func		= bpf_get_stackid_pe,
7b04d6d60fcfb5 Song Liu           2020-07-23  400  	.gpl_only	= false,
7b04d6d60fcfb5 Song Liu           2020-07-23  401  	.ret_type	= RET_INTEGER,
7b04d6d60fcfb5 Song Liu           2020-07-23  402  	.arg1_type	= ARG_PTR_TO_CTX,
7b04d6d60fcfb5 Song Liu           2020-07-23  403  	.arg2_type	= ARG_CONST_MAP_PTR,
7b04d6d60fcfb5 Song Liu           2020-07-23  404  	.arg3_type	= ARG_ANYTHING,
7b04d6d60fcfb5 Song Liu           2020-07-23  405  };
7b04d6d60fcfb5 Song Liu           2020-07-23  406  
fa28dcb82a38f8 Song Liu           2020-06-29  407  static long __bpf_get_stack(struct pt_regs *regs, struct task_struct *task,
7b04d6d60fcfb5 Song Liu           2020-07-23  408  			    struct perf_callchain_entry *trace_in,
d4dd9775ec2424 Andrii Nakryiko    2024-08-29  409  			    void *buf, u32 size, u64 flags, bool may_fault)
c195651e565ae7 Yonghong Song      2018-04-28  410  {
ee2a098851bfbe Namhyung Kim       2022-03-14  411  	u32 trace_nr, copy_len, elem_size, num_elem, max_depth;
c195651e565ae7 Yonghong Song      2018-04-28  412  	bool user_build_id = flags & BPF_F_USER_BUILD_ID;
b8e3a87a627b57 Jordan Rome        2023-11-08  413  	bool crosstask = task && task != current;
c195651e565ae7 Yonghong Song      2018-04-28  414  	u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
c195651e565ae7 Yonghong Song      2018-04-28  415  	bool user = flags & BPF_F_USER_STACK;
c195651e565ae7 Yonghong Song      2018-04-28  416  	struct perf_callchain_entry *trace;
8d90c803bbb115 Tao Chen           2025-10-14  417  	struct bpf_perf_callchain_entry entry = { 0 };
c195651e565ae7 Yonghong Song      2018-04-28  418  	bool kernel = !user;
c195651e565ae7 Yonghong Song      2018-04-28  419  	int err = -EINVAL;
c195651e565ae7 Yonghong Song      2018-04-28  420  	u64 *ips;
c195651e565ae7 Yonghong Song      2018-04-28  421  
c195651e565ae7 Yonghong Song      2018-04-28  422  	if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
c195651e565ae7 Yonghong Song      2018-04-28  423  			       BPF_F_USER_BUILD_ID)))
c195651e565ae7 Yonghong Song      2018-04-28  424  		goto clear;
c195651e565ae7 Yonghong Song      2018-04-28  425  	if (kernel && user_build_id)
c195651e565ae7 Yonghong Song      2018-04-28  426  		goto clear;
c195651e565ae7 Yonghong Song      2018-04-28  427  
d4dd9775ec2424 Andrii Nakryiko    2024-08-29  428  	elem_size = user_build_id ? sizeof(struct bpf_stack_build_id) : sizeof(u64);
c195651e565ae7 Yonghong Song      2018-04-28  429  	if (unlikely(size % elem_size))
c195651e565ae7 Yonghong Song      2018-04-28  430  		goto clear;
c195651e565ae7 Yonghong Song      2018-04-28  431  
fa28dcb82a38f8 Song Liu           2020-06-29  432  	/* cannot get valid user stack for task without user_mode regs */
fa28dcb82a38f8 Song Liu           2020-06-29  433  	if (task && user && !user_mode(regs))
fa28dcb82a38f8 Song Liu           2020-06-29  434  		goto err_fault;
fa28dcb82a38f8 Song Liu           2020-06-29  435  
b8e3a87a627b57 Jordan Rome        2023-11-08  436  	/* get_perf_callchain does not support crosstask user stack walking
b8e3a87a627b57 Jordan Rome        2023-11-08  437  	 * but returns an empty stack instead of NULL.
b8e3a87a627b57 Jordan Rome        2023-11-08  438  	 */
b8e3a87a627b57 Jordan Rome        2023-11-08  439  	if (crosstask && user) {
b8e3a87a627b57 Jordan Rome        2023-11-08  440  		err = -EOPNOTSUPP;
b8e3a87a627b57 Jordan Rome        2023-11-08  441  		goto clear;
b8e3a87a627b57 Jordan Rome        2023-11-08  442  	}
b8e3a87a627b57 Jordan Rome        2023-11-08  443  
c195651e565ae7 Yonghong Song      2018-04-28  444  	num_elem = size / elem_size;
ee2a098851bfbe Namhyung Kim       2022-03-14  445  	max_depth = num_elem + skip;
ee2a098851bfbe Namhyung Kim       2022-03-14  446  	if (sysctl_perf_event_max_stack < max_depth)
ee2a098851bfbe Namhyung Kim       2022-03-14  447  		max_depth = sysctl_perf_event_max_stack;
fa28dcb82a38f8 Song Liu           2020-06-29  448  
d4dd9775ec2424 Andrii Nakryiko    2024-08-29  449  	if (may_fault)
d4dd9775ec2424 Andrii Nakryiko    2024-08-29  450  		rcu_read_lock(); /* need RCU for perf's callchain below */
d4dd9775ec2424 Andrii Nakryiko    2024-08-29  451  
7b04d6d60fcfb5 Song Liu           2020-07-23  452  	if (trace_in)
7b04d6d60fcfb5 Song Liu           2020-07-23  453  		trace = trace_in;
7b04d6d60fcfb5 Song Liu           2020-07-23  454  	else if (kernel && task)
ee2a098851bfbe Namhyung Kim       2022-03-14  455  		trace = get_callchain_entry_for_task(task, max_depth);
fa28dcb82a38f8 Song Liu           2020-06-29  456  	else
8d90c803bbb115 Tao Chen           2025-10-14  457  		trace = get_perf_callchain(regs, (struct perf_callchain_entry *)&entry,
8d90c803bbb115 Tao Chen           2025-10-14  458  					   kernel, user, max_depth, crosstask, false);
c195651e565ae7 Yonghong Song      2018-04-28  459  
d4dd9775ec2424 Andrii Nakryiko    2024-08-29  460  	if (unlikely(!trace) || trace->nr < skip) {
d4dd9775ec2424 Andrii Nakryiko    2024-08-29  461  		if (may_fault)
d4dd9775ec2424 Andrii Nakryiko    2024-08-29  462  			rcu_read_unlock();
c195651e565ae7 Yonghong Song      2018-04-28  463  		goto err_fault;
d4dd9775ec2424 Andrii Nakryiko    2024-08-29  464  	}
c195651e565ae7 Yonghong Song      2018-04-28  465  
ee2a098851bfbe Namhyung Kim       2022-03-14  466  	trace_nr = trace->nr - skip;
c195651e565ae7 Yonghong Song      2018-04-28  467  	trace_nr = (trace_nr <= num_elem) ? trace_nr : num_elem;
c195651e565ae7 Yonghong Song      2018-04-28  468  	copy_len = trace_nr * elem_size;
ee2a098851bfbe Namhyung Kim       2022-03-14  469  
ee2a098851bfbe Namhyung Kim       2022-03-14  470  	ips = trace->ip + skip;
d4dd9775ec2424 Andrii Nakryiko    2024-08-29  471  	if (user_build_id) {
4f4c4fc0153fb1 Andrii Nakryiko    2024-08-29  472  		struct bpf_stack_build_id *id_offs = buf;
4f4c4fc0153fb1 Andrii Nakryiko    2024-08-29  473  		u32 i;
4f4c4fc0153fb1 Andrii Nakryiko    2024-08-29  474  
4f4c4fc0153fb1 Andrii Nakryiko    2024-08-29  475  		for (i = 0; i < trace_nr; i++)
4f4c4fc0153fb1 Andrii Nakryiko    2024-08-29  476  			id_offs[i].ip = ips[i];
4f4c4fc0153fb1 Andrii Nakryiko    2024-08-29  477  	} else {
c195651e565ae7 Yonghong Song      2018-04-28  478  		memcpy(buf, ips, copy_len);
4f4c4fc0153fb1 Andrii Nakryiko    2024-08-29  479  	}
c195651e565ae7 Yonghong Song      2018-04-28  480  
d4dd9775ec2424 Andrii Nakryiko    2024-08-29  481  	/* trace/ips should not be dereferenced after this point */
d4dd9775ec2424 Andrii Nakryiko    2024-08-29  482  	if (may_fault)
d4dd9775ec2424 Andrii Nakryiko    2024-08-29  483  		rcu_read_unlock();
d4dd9775ec2424 Andrii Nakryiko    2024-08-29  484  
d4dd9775ec2424 Andrii Nakryiko    2024-08-29  485  	if (user_build_id)
d4dd9775ec2424 Andrii Nakryiko    2024-08-29  486  		stack_map_get_build_id_offset(buf, trace_nr, user, may_fault);
d4dd9775ec2424 Andrii Nakryiko    2024-08-29  487  
c195651e565ae7 Yonghong Song      2018-04-28  488  	if (size > copy_len)
c195651e565ae7 Yonghong Song      2018-04-28  489  		memset(buf + copy_len, 0, size - copy_len);
c195651e565ae7 Yonghong Song      2018-04-28  490  	return copy_len;
c195651e565ae7 Yonghong Song      2018-04-28  491  
c195651e565ae7 Yonghong Song      2018-04-28  492  err_fault:
c195651e565ae7 Yonghong Song      2018-04-28  493  	err = -EFAULT;
c195651e565ae7 Yonghong Song      2018-04-28  494  clear:
c195651e565ae7 Yonghong Song      2018-04-28  495  	memset(buf, 0, size);
c195651e565ae7 Yonghong Song      2018-04-28  496  	return err;
c195651e565ae7 Yonghong Song      2018-04-28 @497  }
c195651e565ae7 Yonghong Song      2018-04-28  498  

-- 
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki

Re: [RFC PATCH bpf-next v2 2/2] bpf: Pass external callchain entry to get_perf_callchain

[kernel test robot]

Hi Tao,

[This is a private test report for your RFC patch.]
kernel test robot noticed the following build warnings:

[auto build test WARNING on bpf-next/master]

url:    https://github.com/intel-lab-lkp/linux/commits/Tao-Chen/perf-Use-extern-perf_callchain_entry-for-get_perf_callchain/20251014-181245
base:   https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next.git master
patch link:    https://lore.kernel.org/r/20251014100128.2721104-3-chen.dylane%40linux.dev
patch subject: [RFC PATCH bpf-next v2 2/2] bpf: Pass external callchain entry to get_perf_callchain
config: hexagon-allmodconfig (https://download.01.org/0day-ci/archive/20251017/202510170004.yEwSO5VB-lkp@intel.com/config)
compiler: clang version 17.0.6 (https://github.com/llvm/llvm-project 6009708b4367171ccdbf4b5905cb6a803753fe18)
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20251017/202510170004.yEwSO5VB-lkp@intel.com/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <lkp@intel.com>
| Closes: https://lore.kernel.org/oe-kbuild-all/202510170004.yEwSO5VB-lkp@intel.com/

All warnings (new ones prefixed by >>):

>> kernel/bpf/stackmap.c:305:12: warning: stack frame size (1064) exceeds limit (1024) in 'bpf_get_stackid' [-Wframe-larger-than]
     305 | BPF_CALL_3(bpf_get_stackid, struct pt_regs *, regs, struct bpf_map *, map,
         |            ^
>> kernel/bpf/stackmap.c:350:12: warning: stack frame size (1064) exceeds limit (1024) in 'bpf_get_stackid_pe' [-Wframe-larger-than]
     350 | BPF_CALL_3(bpf_get_stackid_pe, struct bpf_perf_event_data_kern *, ctx,
         |            ^
>> kernel/bpf/stackmap.c:407:13: warning: stack frame size (1120) exceeds limit (1024) in '__bpf_get_stack' [-Wframe-larger-than]
     407 | static long __bpf_get_stack(struct pt_regs *regs, struct task_struct *task,
         |             ^
   3 warnings generated.


vim +/bpf_get_stackid +305 kernel/bpf/stackmap.c

d5a3b1f691865be Alexei Starovoitov 2016-02-17  304  
7b04d6d60fcfb5b Song Liu           2020-07-23 @305  BPF_CALL_3(bpf_get_stackid, struct pt_regs *, regs, struct bpf_map *, map,
7b04d6d60fcfb5b Song Liu           2020-07-23  306  	   u64, flags)
7b04d6d60fcfb5b Song Liu           2020-07-23  307  {
7b04d6d60fcfb5b Song Liu           2020-07-23  308  	u32 max_depth = map->value_size / stack_map_data_size(map);
ee2a098851bfbe8 Namhyung Kim       2022-03-14  309  	u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
7b04d6d60fcfb5b Song Liu           2020-07-23  310  	bool user = flags & BPF_F_USER_STACK;
7b04d6d60fcfb5b Song Liu           2020-07-23  311  	struct perf_callchain_entry *trace;
7b04d6d60fcfb5b Song Liu           2020-07-23  312  	bool kernel = !user;
8d90c803bbb1151 Tao Chen           2025-10-14  313  	struct bpf_perf_callchain_entry entry = { 0 };
7b04d6d60fcfb5b Song Liu           2020-07-23  314  
7b04d6d60fcfb5b Song Liu           2020-07-23  315  	if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
7b04d6d60fcfb5b Song Liu           2020-07-23  316  			       BPF_F_FAST_STACK_CMP | BPF_F_REUSE_STACKID)))
7b04d6d60fcfb5b Song Liu           2020-07-23  317  		return -EINVAL;
7b04d6d60fcfb5b Song Liu           2020-07-23  318  
ee2a098851bfbe8 Namhyung Kim       2022-03-14  319  	max_depth += skip;
ee2a098851bfbe8 Namhyung Kim       2022-03-14  320  	if (max_depth > sysctl_perf_event_max_stack)
ee2a098851bfbe8 Namhyung Kim       2022-03-14  321  		max_depth = sysctl_perf_event_max_stack;
ee2a098851bfbe8 Namhyung Kim       2022-03-14  322  
8d90c803bbb1151 Tao Chen           2025-10-14  323  	trace = get_perf_callchain(regs, (struct perf_callchain_entry *)&entry,
8d90c803bbb1151 Tao Chen           2025-10-14  324  				   kernel, user, max_depth, false, false);
7b04d6d60fcfb5b Song Liu           2020-07-23  325  
7b04d6d60fcfb5b Song Liu           2020-07-23  326  	return __bpf_get_stackid(map, trace, flags);
7b04d6d60fcfb5b Song Liu           2020-07-23  327  }
7b04d6d60fcfb5b Song Liu           2020-07-23  328  
d5a3b1f691865be Alexei Starovoitov 2016-02-17  329  const struct bpf_func_proto bpf_get_stackid_proto = {
d5a3b1f691865be Alexei Starovoitov 2016-02-17  330  	.func		= bpf_get_stackid,
d5a3b1f691865be Alexei Starovoitov 2016-02-17  331  	.gpl_only	= true,
d5a3b1f691865be Alexei Starovoitov 2016-02-17  332  	.ret_type	= RET_INTEGER,
d5a3b1f691865be Alexei Starovoitov 2016-02-17  333  	.arg1_type	= ARG_PTR_TO_CTX,
d5a3b1f691865be Alexei Starovoitov 2016-02-17  334  	.arg2_type	= ARG_CONST_MAP_PTR,
d5a3b1f691865be Alexei Starovoitov 2016-02-17  335  	.arg3_type	= ARG_ANYTHING,
d5a3b1f691865be Alexei Starovoitov 2016-02-17  336  };
d5a3b1f691865be Alexei Starovoitov 2016-02-17  337  
7b04d6d60fcfb5b Song Liu           2020-07-23  338  static __u64 count_kernel_ip(struct perf_callchain_entry *trace)
7b04d6d60fcfb5b Song Liu           2020-07-23  339  {
7b04d6d60fcfb5b Song Liu           2020-07-23  340  	__u64 nr_kernel = 0;
7b04d6d60fcfb5b Song Liu           2020-07-23  341  
7b04d6d60fcfb5b Song Liu           2020-07-23  342  	while (nr_kernel < trace->nr) {
7b04d6d60fcfb5b Song Liu           2020-07-23  343  		if (trace->ip[nr_kernel] == PERF_CONTEXT_USER)
7b04d6d60fcfb5b Song Liu           2020-07-23  344  			break;
7b04d6d60fcfb5b Song Liu           2020-07-23  345  		nr_kernel++;
7b04d6d60fcfb5b Song Liu           2020-07-23  346  	}
7b04d6d60fcfb5b Song Liu           2020-07-23  347  	return nr_kernel;
7b04d6d60fcfb5b Song Liu           2020-07-23  348  }
7b04d6d60fcfb5b Song Liu           2020-07-23  349  
7b04d6d60fcfb5b Song Liu           2020-07-23 @350  BPF_CALL_3(bpf_get_stackid_pe, struct bpf_perf_event_data_kern *, ctx,
7b04d6d60fcfb5b Song Liu           2020-07-23  351  	   struct bpf_map *, map, u64, flags)
7b04d6d60fcfb5b Song Liu           2020-07-23  352  {
7b04d6d60fcfb5b Song Liu           2020-07-23  353  	struct perf_event *event = ctx->event;
7b04d6d60fcfb5b Song Liu           2020-07-23  354  	struct perf_callchain_entry *trace;
7b04d6d60fcfb5b Song Liu           2020-07-23  355  	bool kernel, user;
7b04d6d60fcfb5b Song Liu           2020-07-23  356  	__u64 nr_kernel;
7b04d6d60fcfb5b Song Liu           2020-07-23  357  	int ret;
7b04d6d60fcfb5b Song Liu           2020-07-23  358  
7b04d6d60fcfb5b Song Liu           2020-07-23  359  	/* perf_sample_data doesn't have callchain, use bpf_get_stackid */
16817ad7e8b3172 Namhyung Kim       2022-09-08  360  	if (!(event->attr.sample_type & PERF_SAMPLE_CALLCHAIN))
7b04d6d60fcfb5b Song Liu           2020-07-23  361  		return bpf_get_stackid((unsigned long)(ctx->regs),
7b04d6d60fcfb5b Song Liu           2020-07-23  362  				       (unsigned long) map, flags, 0, 0);
7b04d6d60fcfb5b Song Liu           2020-07-23  363  
7b04d6d60fcfb5b Song Liu           2020-07-23  364  	if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
7b04d6d60fcfb5b Song Liu           2020-07-23  365  			       BPF_F_FAST_STACK_CMP | BPF_F_REUSE_STACKID)))
7b04d6d60fcfb5b Song Liu           2020-07-23  366  		return -EINVAL;
7b04d6d60fcfb5b Song Liu           2020-07-23  367  
7b04d6d60fcfb5b Song Liu           2020-07-23  368  	user = flags & BPF_F_USER_STACK;
7b04d6d60fcfb5b Song Liu           2020-07-23  369  	kernel = !user;
7b04d6d60fcfb5b Song Liu           2020-07-23  370  
7b04d6d60fcfb5b Song Liu           2020-07-23  371  	trace = ctx->data->callchain;
7b04d6d60fcfb5b Song Liu           2020-07-23  372  	if (unlikely(!trace))
7b04d6d60fcfb5b Song Liu           2020-07-23  373  		return -EFAULT;
7b04d6d60fcfb5b Song Liu           2020-07-23  374  
7b04d6d60fcfb5b Song Liu           2020-07-23  375  	nr_kernel = count_kernel_ip(trace);
7b04d6d60fcfb5b Song Liu           2020-07-23  376  
7b04d6d60fcfb5b Song Liu           2020-07-23  377  	if (kernel) {
7b04d6d60fcfb5b Song Liu           2020-07-23  378  		__u64 nr = trace->nr;
7b04d6d60fcfb5b Song Liu           2020-07-23  379  
7b04d6d60fcfb5b Song Liu           2020-07-23  380  		trace->nr = nr_kernel;
7b04d6d60fcfb5b Song Liu           2020-07-23  381  		ret = __bpf_get_stackid(map, trace, flags);
7b04d6d60fcfb5b Song Liu           2020-07-23  382  
7b04d6d60fcfb5b Song Liu           2020-07-23  383  		/* restore nr */
7b04d6d60fcfb5b Song Liu           2020-07-23  384  		trace->nr = nr;
7b04d6d60fcfb5b Song Liu           2020-07-23  385  	} else { /* user */
7b04d6d60fcfb5b Song Liu           2020-07-23  386  		u64 skip = flags & BPF_F_SKIP_FIELD_MASK;
7b04d6d60fcfb5b Song Liu           2020-07-23  387  
7b04d6d60fcfb5b Song Liu           2020-07-23  388  		skip += nr_kernel;
7b04d6d60fcfb5b Song Liu           2020-07-23  389  		if (skip > BPF_F_SKIP_FIELD_MASK)
7b04d6d60fcfb5b Song Liu           2020-07-23  390  			return -EFAULT;
7b04d6d60fcfb5b Song Liu           2020-07-23  391  
7b04d6d60fcfb5b Song Liu           2020-07-23  392  		flags = (flags & ~BPF_F_SKIP_FIELD_MASK) | skip;
7b04d6d60fcfb5b Song Liu           2020-07-23  393  		ret = __bpf_get_stackid(map, trace, flags);
7b04d6d60fcfb5b Song Liu           2020-07-23  394  	}
7b04d6d60fcfb5b Song Liu           2020-07-23  395  	return ret;
7b04d6d60fcfb5b Song Liu           2020-07-23  396  }
7b04d6d60fcfb5b Song Liu           2020-07-23  397  
7b04d6d60fcfb5b Song Liu           2020-07-23  398  const struct bpf_func_proto bpf_get_stackid_proto_pe = {
7b04d6d60fcfb5b Song Liu           2020-07-23  399  	.func		= bpf_get_stackid_pe,
7b04d6d60fcfb5b Song Liu           2020-07-23  400  	.gpl_only	= false,
7b04d6d60fcfb5b Song Liu           2020-07-23  401  	.ret_type	= RET_INTEGER,
7b04d6d60fcfb5b Song Liu           2020-07-23  402  	.arg1_type	= ARG_PTR_TO_CTX,
7b04d6d60fcfb5b Song Liu           2020-07-23  403  	.arg2_type	= ARG_CONST_MAP_PTR,
7b04d6d60fcfb5b Song Liu           2020-07-23  404  	.arg3_type	= ARG_ANYTHING,
7b04d6d60fcfb5b Song Liu           2020-07-23  405  };
7b04d6d60fcfb5b Song Liu           2020-07-23  406  
fa28dcb82a38f8e Song Liu           2020-06-29 @407  static long __bpf_get_stack(struct pt_regs *regs, struct task_struct *task,
7b04d6d60fcfb5b Song Liu           2020-07-23  408  			    struct perf_callchain_entry *trace_in,
d4dd9775ec24242 Andrii Nakryiko    2024-08-29  409  			    void *buf, u32 size, u64 flags, bool may_fault)
c195651e565ae7f Yonghong Song      2018-04-28  410  {
ee2a098851bfbe8 Namhyung Kim       2022-03-14  411  	u32 trace_nr, copy_len, elem_size, num_elem, max_depth;
c195651e565ae7f Yonghong Song      2018-04-28  412  	bool user_build_id = flags & BPF_F_USER_BUILD_ID;
b8e3a87a627b575 Jordan Rome        2023-11-08  413  	bool crosstask = task && task != current;
c195651e565ae7f Yonghong Song      2018-04-28  414  	u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
c195651e565ae7f Yonghong Song      2018-04-28  415  	bool user = flags & BPF_F_USER_STACK;
c195651e565ae7f Yonghong Song      2018-04-28  416  	struct perf_callchain_entry *trace;
8d90c803bbb1151 Tao Chen           2025-10-14  417  	struct bpf_perf_callchain_entry entry = { 0 };
c195651e565ae7f Yonghong Song      2018-04-28  418  	bool kernel = !user;
c195651e565ae7f Yonghong Song      2018-04-28  419  	int err = -EINVAL;
c195651e565ae7f Yonghong Song      2018-04-28  420  	u64 *ips;
c195651e565ae7f Yonghong Song      2018-04-28  421  
c195651e565ae7f Yonghong Song      2018-04-28  422  	if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
c195651e565ae7f Yonghong Song      2018-04-28  423  			       BPF_F_USER_BUILD_ID)))
c195651e565ae7f Yonghong Song      2018-04-28  424  		goto clear;
c195651e565ae7f Yonghong Song      2018-04-28  425  	if (kernel && user_build_id)
c195651e565ae7f Yonghong Song      2018-04-28  426  		goto clear;
c195651e565ae7f Yonghong Song      2018-04-28  427  
d4dd9775ec24242 Andrii Nakryiko    2024-08-29  428  	elem_size = user_build_id ? sizeof(struct bpf_stack_build_id) : sizeof(u64);
c195651e565ae7f Yonghong Song      2018-04-28  429  	if (unlikely(size % elem_size))
c195651e565ae7f Yonghong Song      2018-04-28  430  		goto clear;
c195651e565ae7f Yonghong Song      2018-04-28  431  
fa28dcb82a38f8e Song Liu           2020-06-29  432  	/* cannot get valid user stack for task without user_mode regs */
fa28dcb82a38f8e Song Liu           2020-06-29  433  	if (task && user && !user_mode(regs))
fa28dcb82a38f8e Song Liu           2020-06-29  434  		goto err_fault;
fa28dcb82a38f8e Song Liu           2020-06-29  435  
b8e3a87a627b575 Jordan Rome        2023-11-08  436  	/* get_perf_callchain does not support crosstask user stack walking
b8e3a87a627b575 Jordan Rome        2023-11-08  437  	 * but returns an empty stack instead of NULL.
b8e3a87a627b575 Jordan Rome        2023-11-08  438  	 */
b8e3a87a627b575 Jordan Rome        2023-11-08  439  	if (crosstask && user) {
b8e3a87a627b575 Jordan Rome        2023-11-08  440  		err = -EOPNOTSUPP;
b8e3a87a627b575 Jordan Rome        2023-11-08  441  		goto clear;
b8e3a87a627b575 Jordan Rome        2023-11-08  442  	}
b8e3a87a627b575 Jordan Rome        2023-11-08  443  
c195651e565ae7f Yonghong Song      2018-04-28  444  	num_elem = size / elem_size;
ee2a098851bfbe8 Namhyung Kim       2022-03-14  445  	max_depth = num_elem + skip;
ee2a098851bfbe8 Namhyung Kim       2022-03-14  446  	if (sysctl_perf_event_max_stack < max_depth)
ee2a098851bfbe8 Namhyung Kim       2022-03-14  447  		max_depth = sysctl_perf_event_max_stack;
fa28dcb82a38f8e Song Liu           2020-06-29  448  
d4dd9775ec24242 Andrii Nakryiko    2024-08-29  449  	if (may_fault)
d4dd9775ec24242 Andrii Nakryiko    2024-08-29  450  		rcu_read_lock(); /* need RCU for perf's callchain below */
d4dd9775ec24242 Andrii Nakryiko    2024-08-29  451  
7b04d6d60fcfb5b Song Liu           2020-07-23  452  	if (trace_in)
7b04d6d60fcfb5b Song Liu           2020-07-23  453  		trace = trace_in;
7b04d6d60fcfb5b Song Liu           2020-07-23  454  	else if (kernel && task)
ee2a098851bfbe8 Namhyung Kim       2022-03-14  455  		trace = get_callchain_entry_for_task(task, max_depth);
fa28dcb82a38f8e Song Liu           2020-06-29  456  	else
8d90c803bbb1151 Tao Chen           2025-10-14  457  		trace = get_perf_callchain(regs, (struct perf_callchain_entry *)&entry,
8d90c803bbb1151 Tao Chen           2025-10-14  458  					   kernel, user, max_depth, crosstask, false);
c195651e565ae7f Yonghong Song      2018-04-28  459  
d4dd9775ec24242 Andrii Nakryiko    2024-08-29  460  	if (unlikely(!trace) || trace->nr < skip) {
d4dd9775ec24242 Andrii Nakryiko    2024-08-29  461  		if (may_fault)
d4dd9775ec24242 Andrii Nakryiko    2024-08-29  462  			rcu_read_unlock();
c195651e565ae7f Yonghong Song      2018-04-28  463  		goto err_fault;
d4dd9775ec24242 Andrii Nakryiko    2024-08-29  464  	}
c195651e565ae7f Yonghong Song      2018-04-28  465  
ee2a098851bfbe8 Namhyung Kim       2022-03-14  466  	trace_nr = trace->nr - skip;
c195651e565ae7f Yonghong Song      2018-04-28  467  	trace_nr = (trace_nr <= num_elem) ? trace_nr : num_elem;
c195651e565ae7f Yonghong Song      2018-04-28  468  	copy_len = trace_nr * elem_size;
ee2a098851bfbe8 Namhyung Kim       2022-03-14  469  
ee2a098851bfbe8 Namhyung Kim       2022-03-14  470  	ips = trace->ip + skip;
d4dd9775ec24242 Andrii Nakryiko    2024-08-29  471  	if (user_build_id) {
4f4c4fc0153fb11 Andrii Nakryiko    2024-08-29  472  		struct bpf_stack_build_id *id_offs = buf;
4f4c4fc0153fb11 Andrii Nakryiko    2024-08-29  473  		u32 i;
4f4c4fc0153fb11 Andrii Nakryiko    2024-08-29  474  
4f4c4fc0153fb11 Andrii Nakryiko    2024-08-29  475  		for (i = 0; i < trace_nr; i++)
4f4c4fc0153fb11 Andrii Nakryiko    2024-08-29  476  			id_offs[i].ip = ips[i];
4f4c4fc0153fb11 Andrii Nakryiko    2024-08-29  477  	} else {
c195651e565ae7f Yonghong Song      2018-04-28  478  		memcpy(buf, ips, copy_len);
4f4c4fc0153fb11 Andrii Nakryiko    2024-08-29  479  	}
c195651e565ae7f Yonghong Song      2018-04-28  480  
d4dd9775ec24242 Andrii Nakryiko    2024-08-29  481  	/* trace/ips should not be dereferenced after this point */
d4dd9775ec24242 Andrii Nakryiko    2024-08-29  482  	if (may_fault)
d4dd9775ec24242 Andrii Nakryiko    2024-08-29  483  		rcu_read_unlock();
d4dd9775ec24242 Andrii Nakryiko    2024-08-29  484  
d4dd9775ec24242 Andrii Nakryiko    2024-08-29  485  	if (user_build_id)
d4dd9775ec24242 Andrii Nakryiko    2024-08-29  486  		stack_map_get_build_id_offset(buf, trace_nr, user, may_fault);
d4dd9775ec24242 Andrii Nakryiko    2024-08-29  487  
c195651e565ae7f Yonghong Song      2018-04-28  488  	if (size > copy_len)
c195651e565ae7f Yonghong Song      2018-04-28  489  		memset(buf + copy_len, 0, size - copy_len);
c195651e565ae7f Yonghong Song      2018-04-28  490  	return copy_len;
c195651e565ae7f Yonghong Song      2018-04-28  491  
c195651e565ae7f Yonghong Song      2018-04-28  492  err_fault:
c195651e565ae7f Yonghong Song      2018-04-28  493  	err = -EFAULT;
c195651e565ae7f Yonghong Song      2018-04-28  494  clear:
c195651e565ae7f Yonghong Song      2018-04-28  495  	memset(buf, 0, size);
c195651e565ae7f Yonghong Song      2018-04-28  496  	return err;
c195651e565ae7f Yonghong Song      2018-04-28  497  }
c195651e565ae7f Yonghong Song      2018-04-28  498  

-- 
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki

Re: [RFC PATCH bpf-next v2 2/2] bpf: Pass external callchain entry to get_perf_callchain

[Andrii Nakryiko]

On Tue, Oct 14, 2025 at 8:02 AM Alexei Starovoitov
<alexei.starovoitov@gmail.com> wrote:
>
> On Tue, Oct 14, 2025 at 5:14 AM Jiri Olsa <olsajiri@gmail.com> wrote:
> >
> > On Tue, Oct 14, 2025 at 06:01:28PM +0800, Tao Chen wrote:
> > > As Alexei noted, get_perf_callchain() return values may be reused
> > > if a task is preempted after the BPF program enters migrate disable
> > > mode. Drawing on the per-cpu design of bpf_perf_callchain_entries,
> > > stack-allocated memory of bpf_perf_callchain_entry is used here.
> > >
> > > Signed-off-by: Tao Chen <chen.dylane@linux.dev>
> > > ---
> > >  kernel/bpf/stackmap.c | 19 +++++++++++--------
> > >  1 file changed, 11 insertions(+), 8 deletions(-)
> > >
> > > diff --git a/kernel/bpf/stackmap.c b/kernel/bpf/stackmap.c
> > > index 94e46b7f340..acd72c021c0 100644
> > > --- a/kernel/bpf/stackmap.c
> > > +++ b/kernel/bpf/stackmap.c
> > > @@ -31,6 +31,11 @@ struct bpf_stack_map {
> > >       struct stack_map_bucket *buckets[] __counted_by(n_buckets);
> > >  };
> > >
> > > +struct bpf_perf_callchain_entry {
> > > +     u64 nr;
> > > +     u64 ip[PERF_MAX_STACK_DEPTH];
> > > +};
> > > +

we shouldn't introduce another type, there is perf_callchain_entry in
linux/perf_event.h, what's the problem with using that?

> > >  static inline bool stack_map_use_build_id(struct bpf_map *map)
> > >  {
> > >       return (map->map_flags & BPF_F_STACK_BUILD_ID);
> > > @@ -305,6 +310,7 @@ BPF_CALL_3(bpf_get_stackid, struct pt_regs *, regs, struct bpf_map *, map,
> > >       bool user = flags & BPF_F_USER_STACK;
> > >       struct perf_callchain_entry *trace;
> > >       bool kernel = !user;
> > > +     struct bpf_perf_callchain_entry entry = { 0 };
> >
> > so IIUC having entries on stack we do not need to do preempt_disable
> > you had in the previous version, right?
> >
> > I saw Andrii's justification to have this on the stack, I think it's
> > fine, but does it have to be initialized? it seems that only used
> > entries are copied to map
>
> No. We're not adding 1k stack consumption.

Right, and I thought we concluded as much last time, so it's a bit
surprising to see this in this patch.

Tao, you should go with 3 entries per CPU used in a stack-like
fashion. And then passing that entry into get_perf_callchain() (to
avoid one extra copy).

>
> pw-bot: cr

Re: [RFC PATCH bpf-next v2 2/2] bpf: Pass external callchain entry to get_perf_callchain

[Tao Chen]

在 2025/10/17 04:39, Andrii Nakryiko 写道:
> On Tue, Oct 14, 2025 at 8:02 AM Alexei Starovoitov
> <alexei.starovoitov@gmail.com> wrote:
>>
>> On Tue, Oct 14, 2025 at 5:14 AM Jiri Olsa <olsajiri@gmail.com> wrote:
>>>
>>> On Tue, Oct 14, 2025 at 06:01:28PM +0800, Tao Chen wrote:
>>>> As Alexei noted, get_perf_callchain() return values may be reused
>>>> if a task is preempted after the BPF program enters migrate disable
>>>> mode. Drawing on the per-cpu design of bpf_perf_callchain_entries,
>>>> stack-allocated memory of bpf_perf_callchain_entry is used here.
>>>>
>>>> Signed-off-by: Tao Chen <chen.dylane@linux.dev>
>>>> ---
>>>>   kernel/bpf/stackmap.c | 19 +++++++++++--------
>>>>   1 file changed, 11 insertions(+), 8 deletions(-)
>>>>
>>>> diff --git a/kernel/bpf/stackmap.c b/kernel/bpf/stackmap.c
>>>> index 94e46b7f340..acd72c021c0 100644
>>>> --- a/kernel/bpf/stackmap.c
>>>> +++ b/kernel/bpf/stackmap.c
>>>> @@ -31,6 +31,11 @@ struct bpf_stack_map {
>>>>        struct stack_map_bucket *buckets[] __counted_by(n_buckets);
>>>>   };
>>>>
>>>> +struct bpf_perf_callchain_entry {
>>>> +     u64 nr;
>>>> +     u64 ip[PERF_MAX_STACK_DEPTH];
>>>> +};
>>>> +
> 
> we shouldn't introduce another type, there is perf_callchain_entry in
> linux/perf_event.h, what's the problem with using that?

perf_callchain_entry uses flexible array, DEFINE_PER_CPU seems do not
create buffer for this, for ease of use, the size of the ip array has 
been explicitly defined.

struct perf_callchain_entry {
         u64                             nr;
         u64                             ip[]; /* 
/proc/sys/kernel/perf_event_max_stack */
};

> 
>>>>   static inline bool stack_map_use_build_id(struct bpf_map *map)
>>>>   {
>>>>        return (map->map_flags & BPF_F_STACK_BUILD_ID);
>>>> @@ -305,6 +310,7 @@ BPF_CALL_3(bpf_get_stackid, struct pt_regs *, regs, struct bpf_map *, map,
>>>>        bool user = flags & BPF_F_USER_STACK;
>>>>        struct perf_callchain_entry *trace;
>>>>        bool kernel = !user;
>>>> +     struct bpf_perf_callchain_entry entry = { 0 };
>>>
>>> so IIUC having entries on stack we do not need to do preempt_disable
>>> you had in the previous version, right?
>>>
>>> I saw Andrii's justification to have this on the stack, I think it's
>>> fine, but does it have to be initialized? it seems that only used
>>> entries are copied to map
>>
>> No. We're not adding 1k stack consumption.
> 
> Right, and I thought we concluded as much last time, so it's a bit
> surprising to see this in this patch.
> 

Ok, I feel like I'm missing some context from our previous exchange.

> Tao, you should go with 3 entries per CPU used in a stack-like
> fashion. And then passing that entry into get_perf_callchain() (to
> avoid one extra copy).
>

Got it. It is more clearer, will change it in v3.

>>
>> pw-bot: cr


-- 
Best Regards
Tao Chen

Re: [RFC PATCH bpf-next v2 2/2] bpf: Pass external callchain entry to get_perf_callchain

[Andrii Nakryiko]

On Sat, Oct 18, 2025 at 12:51 AM Tao Chen <chen.dylane@linux.dev> wrote:
>
> 在 2025/10/17 04:39, Andrii Nakryiko 写道:
> > On Tue, Oct 14, 2025 at 8:02 AM Alexei Starovoitov
> > <alexei.starovoitov@gmail.com> wrote:
> >>
> >> On Tue, Oct 14, 2025 at 5:14 AM Jiri Olsa <olsajiri@gmail.com> wrote:
> >>>
> >>> On Tue, Oct 14, 2025 at 06:01:28PM +0800, Tao Chen wrote:
> >>>> As Alexei noted, get_perf_callchain() return values may be reused
> >>>> if a task is preempted after the BPF program enters migrate disable
> >>>> mode. Drawing on the per-cpu design of bpf_perf_callchain_entries,
> >>>> stack-allocated memory of bpf_perf_callchain_entry is used here.
> >>>>
> >>>> Signed-off-by: Tao Chen <chen.dylane@linux.dev>
> >>>> ---
> >>>>   kernel/bpf/stackmap.c | 19 +++++++++++--------
> >>>>   1 file changed, 11 insertions(+), 8 deletions(-)
> >>>>
> >>>> diff --git a/kernel/bpf/stackmap.c b/kernel/bpf/stackmap.c
> >>>> index 94e46b7f340..acd72c021c0 100644
> >>>> --- a/kernel/bpf/stackmap.c
> >>>> +++ b/kernel/bpf/stackmap.c
> >>>> @@ -31,6 +31,11 @@ struct bpf_stack_map {
> >>>>        struct stack_map_bucket *buckets[] __counted_by(n_buckets);
> >>>>   };
> >>>>
> >>>> +struct bpf_perf_callchain_entry {
> >>>> +     u64 nr;
> >>>> +     u64 ip[PERF_MAX_STACK_DEPTH];
> >>>> +};
> >>>> +
> >
> > we shouldn't introduce another type, there is perf_callchain_entry in
> > linux/perf_event.h, what's the problem with using that?
>
> perf_callchain_entry uses flexible array, DEFINE_PER_CPU seems do not
> create buffer for this, for ease of use, the size of the ip array has
> been explicitly defined.
>
> struct perf_callchain_entry {
>          u64                             nr;
>          u64                             ip[]; /*
> /proc/sys/kernel/perf_event_max_stack */
> };
>

Ok, fair enough, but instead of casting between perf_callchain_entry
and bpf_perf_callchain_entry, why not put perf_callchain_entry inside
bpf_perf_callchain_entry as a first struct and pass a pointer to it.
That looks a bit more appropriate? Though I'm not sure if compiler
will complain about that flex array...

But on related note, I looked briefly at how perf gets those
perf_callchain_entries, and it does seem like it also has a small
stack of entries, so maybe we don't really need to invent anything
here. See PERF_NR_CONTEXTS and how that's used.

If instead of disabling preemption we disable migration, then I think
we should be good with relying on perf's callchain management, or am I
missing something?

> >
> >>>>   static inline bool stack_map_use_build_id(struct bpf_map *map)
> >>>>   {
> >>>>        return (map->map_flags & BPF_F_STACK_BUILD_ID);
> >>>> @@ -305,6 +310,7 @@ BPF_CALL_3(bpf_get_stackid, struct pt_regs *, regs, struct bpf_map *, map,
> >>>>        bool user = flags & BPF_F_USER_STACK;
> >>>>        struct perf_callchain_entry *trace;
> >>>>        bool kernel = !user;
> >>>> +     struct bpf_perf_callchain_entry entry = { 0 };
> >>>
> >>> so IIUC having entries on stack we do not need to do preempt_disable
> >>> you had in the previous version, right?
> >>>
> >>> I saw Andrii's justification to have this on the stack, I think it's
> >>> fine, but does it have to be initialized? it seems that only used
> >>> entries are copied to map
> >>
> >> No. We're not adding 1k stack consumption.
> >
> > Right, and I thought we concluded as much last time, so it's a bit
> > surprising to see this in this patch.
> >
>
> Ok, I feel like I'm missing some context from our previous exchange.
>
> > Tao, you should go with 3 entries per CPU used in a stack-like
> > fashion. And then passing that entry into get_perf_callchain() (to
> > avoid one extra copy).
> >
>
> Got it. It is more clearer, will change it in v3.
>
> >>
> >> pw-bot: cr
>
>
> --
> Best Regards
> Tao Chen

Re: [RFC PATCH bpf-next v2 2/2] bpf: Pass external callchain entry to get_perf_callchain

[Tao Chen]

在 2025/10/22 00:37, Andrii Nakryiko 写道:
> On Sat, Oct 18, 2025 at 12:51 AM Tao Chen <chen.dylane@linux.dev> wrote:
>>
>> 在 2025/10/17 04:39, Andrii Nakryiko 写道:
>>> On Tue, Oct 14, 2025 at 8:02 AM Alexei Starovoitov
>>> <alexei.starovoitov@gmail.com> wrote:
>>>>
>>>> On Tue, Oct 14, 2025 at 5:14 AM Jiri Olsa <olsajiri@gmail.com> wrote:
>>>>>
>>>>> On Tue, Oct 14, 2025 at 06:01:28PM +0800, Tao Chen wrote:
>>>>>> As Alexei noted, get_perf_callchain() return values may be reused
>>>>>> if a task is preempted after the BPF program enters migrate disable
>>>>>> mode. Drawing on the per-cpu design of bpf_perf_callchain_entries,
>>>>>> stack-allocated memory of bpf_perf_callchain_entry is used here.
>>>>>>
>>>>>> Signed-off-by: Tao Chen <chen.dylane@linux.dev>
>>>>>> ---
>>>>>>    kernel/bpf/stackmap.c | 19 +++++++++++--------
>>>>>>    1 file changed, 11 insertions(+), 8 deletions(-)
>>>>>>
>>>>>> diff --git a/kernel/bpf/stackmap.c b/kernel/bpf/stackmap.c
>>>>>> index 94e46b7f340..acd72c021c0 100644
>>>>>> --- a/kernel/bpf/stackmap.c
>>>>>> +++ b/kernel/bpf/stackmap.c
>>>>>> @@ -31,6 +31,11 @@ struct bpf_stack_map {
>>>>>>         struct stack_map_bucket *buckets[] __counted_by(n_buckets);
>>>>>>    };
>>>>>>
>>>>>> +struct bpf_perf_callchain_entry {
>>>>>> +     u64 nr;
>>>>>> +     u64 ip[PERF_MAX_STACK_DEPTH];
>>>>>> +};
>>>>>> +
>>>
>>> we shouldn't introduce another type, there is perf_callchain_entry in
>>> linux/perf_event.h, what's the problem with using that?
>>
>> perf_callchain_entry uses flexible array, DEFINE_PER_CPU seems do not
>> create buffer for this, for ease of use, the size of the ip array has
>> been explicitly defined.
>>
>> struct perf_callchain_entry {
>>           u64                             nr;
>>           u64                             ip[]; /*
>> /proc/sys/kernel/perf_event_max_stack */
>> };
>>
> 
> Ok, fair enough, but instead of casting between perf_callchain_entry
> and bpf_perf_callchain_entry, why not put perf_callchain_entry inside
> bpf_perf_callchain_entry as a first struct and pass a pointer to it.
> That looks a bit more appropriate? Though I'm not sure if compiler
> will complain about that flex array...
> 
> But on related note, I looked briefly at how perf gets those
> perf_callchain_entries, and it does seem like it also has a small
> stack of entries, so maybe we don't really need to invent anything
> here. See PERF_NR_CONTEXTS and how that's used.
> 

It seems so. The implementation of get_callchain_entry and 
put_callchain_entry is similar to a simple stack management.

struct perf_callchain_entry *get_callchain_entry(int *rctx)
{
         int cpu;
         struct callchain_cpus_entries *entries;

         *rctx = get_recursion_context(this_cpu_ptr(callchain_recursion));
         if (*rctx == -1)
                 return NULL;

         entries = rcu_dereference(callchain_cpus_entries);
         if (!entries) {
  
put_recursion_context(this_cpu_ptr(callchain_recursion), *rctx);
                 return NULL;
         }

         cpu = smp_processor_id();

         return (((void *)entries->cpu_entries[cpu]) +
                 (*rctx * perf_callchain_entry__sizeof()));
}

void
put_callchain_entry(int rctx)
{
         put_recursion_context(this_cpu_ptr(callchain_recursion), rctx);
}

> If instead of disabling preemption we disable migration, then I think
> we should be good with relying on perf's callchain management, or am I
> missing something?
> 

Peter proposed refactoring the get_perf_callchain interface in v3, i 
think after that we can still use perf callchain, but with the 
refactored API, it can prevent being overwritten by preemption.

BPF_CALL_3(bpf_get_stackid, struct pt_regs *, regs, struct bpf_map *, map
{
	...
	entry = bpf_get_perf_callchain;
	__bpf_get_stackid(map, entry, flags);
	bpf_put_callchain_entry(entry);
	...
}

A simple specific implementation is as follows:

diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index fd1d91017b9..1c7573f085f 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -67,6 +67,7 @@ struct perf_callchain_entry_ctx {
  	u32				nr;
  	short				contexts;
  	bool				contexts_maxed;
+	bool				add_mark;
  };

  typedef unsigned long (*perf_copy_f)(void *dst, const void *src,
@@ -1718,9 +1719,18 @@ DECLARE_PER_CPU(struct perf_callchain_entry, 
perf_callchain_entry);

  extern void perf_callchain_user(struct perf_callchain_entry_ctx 
*entry, struct pt_regs *regs);
  extern void perf_callchain_kernel(struct perf_callchain_entry_ctx 
*entry, struct pt_regs *regs);
+
+extern void __init_perf_callchain_ctx(struct perf_callchain_entry_ctx *ctx,
+				      struct perf_callchain_entry *entry,
+				      u32 max_stack, bool add_mark);
+
+extern void __get_perf_callchain_kernel(struct perf_callchain_entry_ctx 
*ctx, struct pt_regs *regs);
+extern void __get_perf_callchain_user(struct perf_callchain_entry_ctx 
*ctx, struct pt_regs *regs);
+
+
  extern struct perf_callchain_entry *
  get_perf_callchain(struct pt_regs *regs, bool kernel, bool user,
-		   u32 max_stack, bool crosstask, bool add_mark);
+		   u32 max_stack, bool crosstask);
  extern int get_callchain_buffers(int max_stack);
  extern void put_callchain_buffers(void);
  extern struct perf_callchain_entry *get_callchain_entry(int *rctx);
diff --git a/kernel/bpf/stackmap.c b/kernel/bpf/stackmap.c
index 4d53cdd1374..3f2543e5244 100644
--- a/kernel/bpf/stackmap.c
+++ b/kernel/bpf/stackmap.c
@@ -297,14 +297,38 @@ static long __bpf_get_stackid(struct bpf_map *map,
  	return id;
  }

+static struct perf_callchain_entry *entry
+bpf_get_perf_callchain(int *rctx, struct pt_regs *regs, bool kernel, 
bool user, int max_stack)
+{
+	struct perf_callchain_entry_ctx ctx;
+	struct perf_callchain_entry *entry;
+
+	entry = get_callchain_entry(&rctx);
+	if (unlikely(!entry))
+		return NULL;
+
+	__init_perf_callchain_ctx(&ctx, entry)
+	if (kernel)
+		__get_perf_callchain_kernel(&ctx, regs);
+	if (user)
+		__get_perf_callchain_user(&ctx, regs);
+	return entry;
+}
+
+static bpf_put_perf_callchain(int rctx)
+{
+	put_callchain_entry(rctx);
+}
+
  BPF_CALL_3(bpf_get_stackid, struct pt_regs *, regs, struct bpf_map *, map,
  	   u64, flags)
  {
  	u32 max_depth = map->value_size / stack_map_data_size(map);
  	u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
  	bool user = flags & BPF_F_USER_STACK;
-	struct perf_callchain_entry *trace;
+	struct perf_callchain_entry *entry;
  	bool kernel = !user;
+	int rctx, ret;

  	if (unlikely(flags & ~(BPF_F_SKIP_FIELD_MASK | BPF_F_USER_STACK |
  			       BPF_F_FAST_STACK_CMP | BPF_F_REUSE_STACKID)))
@@ -314,14 +338,14 @@ BPF_CALL_3(bpf_get_stackid, struct pt_regs *, 
regs, struct bpf_map *, map,
  	if (max_depth > sysctl_perf_event_max_stack)
  		max_depth = sysctl_perf_event_max_stack;

-	trace = get_perf_callchain(regs, kernel, user, max_depth,
-				   false, false);
-
-	if (unlikely(!trace))
-		/* couldn't fetch the stack trace */
+	entry = bpf_get_perf_callchain(&rctx, regs, kernel, user, max_depth);
+	if (unlikely(!entry))
  		return -EFAULT;

-	return __bpf_get_stackid(map, trace, flags);
+	ret = __bpf_get_stackid(map, entry, flags);
+	bpf_put_callchain_entry(rctx);
+
+	return ret;
  }

  const struct bpf_func_proto bpf_get_stackid_proto = {
@@ -452,7 +476,7 @@ static long __bpf_get_stack(struct pt_regs *regs, 
struct task_struct *task,
  		trace = get_callchain_entry_for_task(task, max_depth);
  	else
  		trace = get_perf_callchain(regs, kernel, user, max_depth,
-					   crosstask, false);
+					   crosstask);

  	if (unlikely(!trace) || trace->nr < skip) {
  		if (may_fault)
diff --git a/kernel/events/callchain.c b/kernel/events/callchain.c
index 808c0d7a31f..2c36e490625 100644
--- a/kernel/events/callchain.c
+++ b/kernel/events/callchain.c
@@ -216,13 +216,54 @@ static void 
fixup_uretprobe_trampoline_entries(struct perf_callchain_entry *entr
  #endif
  }

+void __init_perf_callchain_ctx(struct perf_callchain_entry_ctx *ctx,
+			       struct perf_callchain_entry *entry,
+			       u32 max_stack, bool add_mark)
+
+{
+	ctx->entry		= entry;
+	ctx->max_stack		= max_stack;
+	ctx->nr			= entry->nr = 0;
+	ctx->contexts		= 0;
+	ctx->contexts_maxed	= false;
+	ctx->add_mark		= add_mark;
+}
+
+void __get_perf_callchain_kernel(struct perf_callchain_entry_ctx *ctx, 
struct pt_regs *regs)
+{
+	if (user_mode(regs))
+		return;
+
+	if (ctx->add_mark)
+		perf_callchain_store_context(ctx, PERF_CONTEXT_KERNEL);
+	perf_callchain_kernel(ctx, regs);
+}
+
+void __get_perf_callchain_user(struct perf_callchain_entry_ctx *ctx, 
struct pt_regs *regs)
+{
+	int start_entry_idx;
+
+	if (!user_mode(regs)) {
+		if (current->flags & (PF_KTHREAD | PF_USER_WORKER))
+			return;
+		regs = task_pt_regs(current);
+	}
+
+	if (ctx->add_mark)
+		perf_callchain_store_context(ctx, PERF_CONTEXT_USER);
+
+	start_entry_idx = ctx->nr;
+	perf_callchain_user(ctx, regs);
+	fixup_uretprobe_trampoline_entries(ctx->entry, start_entry_idx);
+}
+
  struct perf_callchain_entry *
  get_perf_callchain(struct pt_regs *regs, bool kernel, bool user,
-		   u32 max_stack, bool crosstask, bool add_mark)
+		   u32 max_stack, bool crosstask)
  {
  	struct perf_callchain_entry *entry;
  	struct perf_callchain_entry_ctx ctx;
-	int rctx, start_entry_idx;
+	int rctx;

  	/* crosstask is not supported for user stacks */
  	if (crosstask && user && !kernel)
@@ -232,34 +273,14 @@ get_perf_callchain(struct pt_regs *regs, bool 
kernel, bool user,
  	if (!entry)
  		return NULL;

-	ctx.entry		= entry;
-	ctx.max_stack		= max_stack;
-	ctx.nr			= entry->nr = 0;
-	ctx.contexts		= 0;
-	ctx.contexts_maxed	= false;
+	__init_perf_callchain_ctx(&ctx, entry, max_stack, true);

-	if (kernel && !user_mode(regs)) {
-		if (add_mark)
-			perf_callchain_store_context(&ctx, PERF_CONTEXT_KERNEL);
-		perf_callchain_kernel(&ctx, regs);
-	}
-
-	if (user && !crosstask) {
-		if (!user_mode(regs)) {
-			if (current->flags & (PF_KTHREAD | PF_USER_WORKER))
-				goto exit_put;
-			regs = task_pt_regs(current);
-		}
+	if (kernel)
+		__get_perf_callchain_kernel(&ctx, regs);

-		if (add_mark)
-			perf_callchain_store_context(&ctx, PERF_CONTEXT_USER);
-
-		start_entry_idx = entry->nr;
-		perf_callchain_user(&ctx, regs);
-		fixup_uretprobe_trampoline_entries(entry, start_entry_idx);
-	}
+	if (user && !crosstask)
+		__get_perf_callchain_user(&ctx, regs);

-exit_put:
  	put_callchain_entry(rctx);

  	return entry;
diff --git a/kernel/events/core.c b/kernel/events/core.c
index 7541f6f85fc..eb0f110593d 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -8218,7 +8218,7 @@ perf_callchain(struct perf_event *event, struct 
pt_regs *regs)
  		return &__empty_callchain;

  	callchain = get_perf_callchain(regs, kernel, user,
-				       max_stack, crosstask, true);
+				       max_stack, crosstask);
  	return callchain ?: &__empty_callchain;
  }

>>>
>>>>>>    static inline bool stack_map_use_build_id(struct bpf_map *map)
>>>>>>    {
>>>>>>         return (map->map_flags & BPF_F_STACK_BUILD_ID);
>>>>>> @@ -305,6 +310,7 @@ BPF_CALL_3(bpf_get_stackid, struct pt_regs *, regs, struct bpf_map *, map,
>>>>>>         bool user = flags & BPF_F_USER_STACK;
>>>>>>         struct perf_callchain_entry *trace;
>>>>>>         bool kernel = !user;
>>>>>> +     struct bpf_perf_callchain_entry entry = { 0 };
>>>>>
>>>>> so IIUC having entries on stack we do not need to do preempt_disable
>>>>> you had in the previous version, right?
>>>>>
>>>>> I saw Andrii's justification to have this on the stack, I think it's
>>>>> fine, but does it have to be initialized? it seems that only used
>>>>> entries are copied to map
>>>>
>>>> No. We're not adding 1k stack consumption.
>>>
>>> Right, and I thought we concluded as much last time, so it's a bit
>>> surprising to see this in this patch.
>>>
>>
>> Ok, I feel like I'm missing some context from our previous exchange.
>>
>>> Tao, you should go with 3 entries per CPU used in a stack-like
>>> fashion. And then passing that entry into get_perf_callchain() (to
>>> avoid one extra copy).
>>>
>>
>> Got it. It is more clearer, will change it in v3.
>>
>>>>
>>>> pw-bot: cr
>>
>>
>> --
>> Best Regards
>> Tao Chen


-- 
Best Regards
Tao Chen