perf trace: Refactor augmented_raw_syscalls using bpf_for

PATCH Applied Viktor Malik 2 participants 5 messages Jul 7 – Jul 17

Subsystems (3)

Stable branch BPF perf events

lore

Share & tools

lore

Viktor Malik

author

2026-07-07 06:53 UTC #0 lore raw

v4 cover perf trace: Refactor augmented_raw_syscalls using bpf_for

The loop for processing syscall args in augment_raw_syscalls has a history of breaking with Clang updates. In the past, we've seen it break (i.e. stop passing the BPF verifier) between Clang 15 and 16 and now a similar thing happened between Clang 21 and 22. While the issue is mitigated on the main line by a recent verifier update, it remains broken on the 6.12 and 6.18 stable branches, effectively breaking perf trace.

The reason is that the loop is quite complex and the BPF verifier often struggles to prove that it terminates.

This series fixes the issue by replacing the standard for loop by the bpf_for macro, which uses numeric BPF iterator. This should prevent future breakages of this kind since the verifier has much easier job proving that the loop terminates. Small adjustments were necessary for the loop to make it work, see the second commit message for details.

To keep perf compatible with older kernels, the first commit factors out the loop body into a function, which is then called either from bpf_for or from a standard for loop, depending on whether BPF numeric iterators are available.

Changes from v3:
- v3: https://lore.kernel.org/bpf/[email protected]/
- Do not remove any hacks from the BPF program to keep it backwards compatible with older kernels and Clang versions (Sashiko).

Changes from v2:
- v2: https://lore.kernel.org/bpf/[email protected]/T/
- Small refactoring suggested by Namhhyung
- Make size in augment_arg() an int (found by Sashiko)
- Keep the original way of clamping aug_size to TRACE_AUG_MAX_BUF (found by Sashiko)

Changes from v1:
- v1: https://lore.kernel.org/bpf/[email protected]/T/
- Use bpf_for instead of bpf_loop (suggested by Alexei and Andrii)
- Keep the change backwards compatible with older kernels (required by Namhyung)

Viktor Malik (2):
perf trace: Factor out BPF loop body
perf trace: Refactor augmented_raw_syscalls using bpf_for

 .../bpf_skel/augmented_raw_syscalls.bpf.c     | 155 +++++++++++-------
 1 file changed, 98 insertions(+), 57 deletions(-)

Signature · 1 line

2.54.0

Viktor Malik

author

2026-07-07 06:53 UTC #1 lore raw

PATCH v4 1/2 perf trace: Factor out BPF loop body

The BPF program in augmented_raw_syscalls uses a for loop to iterate all syscall arguments. The loop body is quite complex and often poses problems for the BPF verifier. As a preparation step for addressing this issue, factor out the loop body into a separate function.

Signed-off-by Viktor Malik <[email protected]>

Cc [email protected]

---
 .../bpf_skel/augmented_raw_syscalls.bpf.c     | 128 ++++++++++--------
 1 file changed, 73 insertions(+), 55 deletions(-)

tools/perf/util/bpf_skel/augmented_raw_syscalls.bpf.c +73 ~~-55~~

index 2a6e61864ee0..bc036a348079 100644
@@ -429,15 +429,80 @@ static bool pid_filter__has(struct pids_filtered *pids, pid_t pid)
 	return bpf_map_lookup_elem(pids, &pid) != NULL;
 }
 
+/*
+ * Determine what type of argument and how many bytes to read from user space, using the
+ * value in the beauty_map. This is the relation of parameter type and its corresponding
+ * value in the beauty map, and how many bytes we read eventually:
+ *
+ * string: 1			      -> size of string
+ * struct: size of struct	      -> size of struct
+ * buffer: -1 * (index of paired len) -> value of paired len (maximum: TRACE_AUG_MAX_BUF)
+ */
+static inline int augment_arg(struct syscall_enter_args *args, int i,
+			      unsigned int *beauty_map,
+			      struct augmented_arg *payload_offset)
+{
+	int index, value_size = sizeof(struct augmented_arg) - offsetof(struct augmented_arg, value);
+	s64 aug_size, size;
+	bool augmented;
+	void *arg;
+
+	arg = (void *)args->args[i];
+	augmented = false;
+	size = beauty_map[i];
+	aug_size = size; /* size of the augmented data read from user space */
+
+	if (size == 0 || arg == NULL)
+		return 0;
+
+	if (size == 1) { /* string */
+		aug_size = bpf_probe_read_user_str(payload_offset->value, value_size, arg);
+		/* minimum of 0 to pass the verifier */
+		if (aug_size < 0)
+			aug_size = 0;
+
+		augmented = true;
+	} else if (size > 0 && size <= value_size) { /* struct */
+		if (!bpf_probe_read_user(payload_offset->value, size, arg))
+			augmented = true;
+	} else if ((int)size < 0 && size >= -6) { /* buffer */
+		index = -(size + 1);
+		barrier_var(index); // Prevent clang (noticed with v18) from removing the &= 7 trick.
+		index &= 7;	    // Satisfy the bounds checking with the verifier in some kernels.
+		aug_size = args->args[index] > TRACE_AUG_MAX_BUF ? TRACE_AUG_MAX_BUF : args->args[index];
+
+		if (aug_size > 0) {
+			if (!bpf_probe_read_user(payload_offset->value, aug_size, arg))
+				augmented = true;
+		}
+	}
+
+	/* Augmented data size is limited to sizeof(augmented_arg->unnamed union with value field) */
+	if (aug_size > value_size)
+		aug_size = value_size;
+
+	/* write data to payload */
+	if (augmented) {
+		int written = offsetof(struct augmented_arg, value) + aug_size;
+
+		if (written < 0 || written > sizeof(struct augmented_arg))
+			return -1;
+
+		payload_offset->size = aug_size;
+		return written;
+	}
+
+	return 0;
+}
+
 static int augment_sys_enter(void *ctx, struct syscall_enter_args *args)
 {
-	bool augmented, do_output = false;
-	int zero = 0, index, value_size = sizeof(struct augmented_arg) - offsetof(struct augmented_arg, value);
+	bool do_output = false;
+	int zero = 0, written;
 	u64 output = 0; /* has to be u64, otherwise it won't pass the verifier */
-	s64 aug_size, size;
 	unsigned int nr, *beauty_map;
 	struct beauty_payload_enter *payload;
-	void *arg, *payload_offset;
+	void *payload_offset;
 
 	/* fall back to do predefined tail call */
 	if (args == NULL)

@@ -457,58 +522,11 @@ static int augment_sys_enter(void *ctx, struct syscall_enter_args *args)
 	/* copy the sys_enter header, which has the syscall_nr */
 	__builtin_memcpy(&payload->args, args, sizeof(struct syscall_enter_args));
 
-	/*
-	 * Determine what type of argument and how many bytes to read from user space, using the
-	 * value in the beauty_map. This is the relation of parameter type and its corresponding
-	 * value in the beauty map, and how many bytes we read eventually:
-	 *
-	 * string: 1			      -> size of string
-	 * struct: size of struct	      -> size of struct
-	 * buffer: -1 * (index of paired len) -> value of paired len (maximum: TRACE_AUG_MAX_BUF)
-	 */
 	for (int i = 0; i < 6; i++) {
-		arg = (void *)args->args[i];
-		augmented = false;
-		size = beauty_map[i];
-		aug_size = size; /* size of the augmented data read from user space */
-
-		if (size == 0 || arg == NULL)
-			continue;
-
-		if (size == 1) { /* string */
-			aug_size = bpf_probe_read_user_str(((struct augmented_arg *)payload_offset)->value, value_size, arg);
-			/* minimum of 0 to pass the verifier */
-			if (aug_size < 0)
-				aug_size = 0;
-
-			augmented = true;
-		} else if (size > 0 && size <= value_size) { /* struct */
-			if (!bpf_probe_read_user(((struct augmented_arg *)payload_offset)->value, size, arg))
-				augmented = true;
-		} else if ((int)size < 0 && size >= -6) { /* buffer */
-			index = -(size + 1);
-			barrier_var(index); // Prevent clang (noticed with v18) from removing the &= 7 trick.
-			index &= 7;	    // Satisfy the bounds checking with the verifier in some kernels.
-			aug_size = args->args[index] > TRACE_AUG_MAX_BUF ? TRACE_AUG_MAX_BUF : args->args[index];
-
-			if (aug_size > 0) {
-				if (!bpf_probe_read_user(((struct augmented_arg *)payload_offset)->value, aug_size, arg))
-					augmented = true;
-			}
-		}
-
-		/* Augmented data size is limited to sizeof(augmented_arg->unnamed union with value field) */
-		if (aug_size > value_size)
-			aug_size = value_size;
-
-		/* write data to payload */
-		if (augmented) {
-			int written = offsetof(struct augmented_arg, value) + aug_size;
-
-			if (written < 0 || written > sizeof(struct augmented_arg))
-				return 1;
-
-			((struct augmented_arg *)payload_offset)->size = aug_size;
+		written = augment_arg(args, i, beauty_map, (struct augmented_arg *)payload_offset);
+		if (written < 0)
+			return 1;
+		if (written > 0) {
 			output += written;
 			payload_offset += written;
 			do_output = true;

Signature · 1 line

2.54.0

Viktor Malik

author

2026-07-07 06:53 UTC #2 lore raw

PATCH v4 2/2 perf trace: Refactor augmented_raw_syscalls using bpf_for

In reply to Viktor Malik

The loop for processing syscall args in augment_raw_syscalls has a history of breaking with Clang updates, see e.g. commit 013eb043f37b ("perf trace: Fix BPF loading failure (-E2BIG)") from Clang 15 to 16.

Now, a similar thing happened between Clang 21 and 22. While the issue is mitigated on the main line by a recent verifier update, it remains broken on the 6.12 and 6.18 stable branches:

[linux-6.18.y]# sudo perf trace true
libbpf: prog 'sys_enter': BPF program load failed: -E2BIG
libbpf: prog 'sys_enter': -- BEGIN PROG LOAD LOG --
[...]
BPF program is too large. Processed 1000001 insn
processed 1000001 insns (limit 1000000) max_states_per_insn 40 total_states 37941 peak_states 232 mark_read 0 -- END PROG LOAD LOG --
libbpf: prog 'sys_enter': failed to load: -E2BIG
libbpf: failed to load object 'augmented_raw_syscalls_bpf'
libbpf: failed to load BPF skeleton 'augmented_raw_syscalls_bpf': -E2BIG
Error: failed to get syscall or beauty map fd
[...]

The reason is that the loop is quite complex and the BPF verifier often struggles to prove that it terminates.

Fix the issue by replacing the standard for loop with the bpf_for macro, which uses a numeric BPF iterator. This should prevent future breakages of this kind since the verifier has a much easier job proving that the loop terminates.

Small adjustments were necessary for the loop to make it work. The main problem is that the verifier sometimes has problems with bpf_for loops that use a carry-over state, such as the payload_offset and output vars here, since the verifier tries to track their values too precisely and cannot prove loop convergence. To resolve the issue, we (1) explicitly recompute payload_offset in every iteration and (2) use a trick with adding a global zero to output to help the verifier forget its precise state and use a range instead.

Finally, to keep backwards compatibility with older kernel versions that don't have bpf_for (i.e. numeric iterators), fall back to standard loop.

Signed-off-by Viktor Malik <[email protected]>

Fixes a68fd6a6cdd3 ("perf trace: Collect augmented data using BPF")

Cc [email protected]

---
 .../bpf_skel/augmented_raw_syscalls.bpf.c     | 47 ++++++++++++++-----
 1 file changed, 35 insertions(+), 12 deletions(-)

tools/perf/util/bpf_skel/augmented_raw_syscalls.bpf.c +35 ~~-12~~

index bc036a348079..3bc9e28a9b8a 100644
@@ -429,6 +429,8 @@ static bool pid_filter__has(struct pids_filtered *pids, pid_t pid)
 	return bpf_map_lookup_elem(pids, &pid) != NULL;
 }
 
+u64 ZERO = 0;
+
 /*
  * Determine what type of argument and how many bytes to read from user space, using the
  * value in the beauty_map. This is the relation of parameter type and its corresponding

@@ -440,9 +442,10 @@ static bool pid_filter__has(struct pids_filtered *pids, pid_t pid)
  */
 static inline int augment_arg(struct syscall_enter_args *args, int i,
 			      unsigned int *beauty_map,
-			      struct augmented_arg *payload_offset)
+			      struct beauty_payload_enter *payload, u64 offset)
 {
 	int index, value_size = sizeof(struct augmented_arg) - offsetof(struct augmented_arg, value);
+	struct augmented_arg *payload_offset;
 	s64 aug_size, size;
 	bool augmented;
 	void *arg;

@@ -455,6 +458,12 @@ static inline int augment_arg(struct syscall_enter_args *args, int i,
 	if (size == 0 || arg == NULL)
 		return 0;
 
+	/* bounds check for the verifier */
+	if (offset > sizeof(payload->aug_args) - sizeof(payload->aug_args[0]))
+		return -1;
+	barrier_var(offset);
+	payload_offset = (struct augmented_arg *)((void *)&payload->aug_args + offset);
+
 	if (size == 1) { /* string */
 		aug_size = bpf_probe_read_user_str(payload_offset->value, value_size, arg);
 		/* minimum of 0 to pass the verifier */

@@ -498,11 +507,10 @@ static inline int augment_arg(struct syscall_enter_args *args, int i,
 static int augment_sys_enter(void *ctx, struct syscall_enter_args *args)
 {
 	bool do_output = false;
-	int zero = 0, written;
+	int i, zero = 0, written;
 	u64 output = 0; /* has to be u64, otherwise it won't pass the verifier */
 	unsigned int nr, *beauty_map;
 	struct beauty_payload_enter *payload;
-	void *payload_offset;
 
 	/* fall back to do predefined tail call */
 	if (args == NULL)

@@ -514,7 +522,6 @@ static int augment_sys_enter(void *ctx, struct syscall_enter_args *args)
 
 	/* set up payload for output */
 	payload        = bpf_map_lookup_elem(&beauty_payload_enter_map, &zero);
-	payload_offset = (void *)&payload->aug_args;
 
 	if (beauty_map == NULL || payload == NULL)
 		return 1;

@@ -522,14 +529,30 @@ static int augment_sys_enter(void *ctx, struct syscall_enter_args *args)
 	/* copy the sys_enter header, which has the syscall_nr */
 	__builtin_memcpy(&payload->args, args, sizeof(struct syscall_enter_args));
 
-	for (int i = 0; i < 6; i++) {
-		written = augment_arg(args, i, beauty_map, (struct augmented_arg *)payload_offset);
-		if (written < 0)
-			return 1;
-		if (written > 0) {
-			output += written;
-			payload_offset += written;
-			do_output = true;
+	if (bpf_ksym_exists(bpf_iter_num_new)) {
+		bpf_for(i, 0, 6) {
+			written = augment_arg(args, i, beauty_map, payload, output);
+			if (written < 0)
+				return 1;
+			if (written > 0) {
+				output += written;
+				/*
+				 * guide the verifier to forget range of `output`, which
+				 * helps to prove convergence of the loop
+				 */
+				output += ZERO;
+				do_output = true;
+			}
+		}
+	} else {
+		for (i = 0; i < 6; i++) {
+			written = augment_arg(args, i, beauty_map, payload, output);
+			if (written < 0)
+				return 1;
+			if (written > 0) {
+				output += written;
+				do_output = true;
+			}
 		}
 	}

Signature · 1 line

2.54.0

Namhyung Kim

maintainer

2026-07-13 18:00 UTC #3 lore raw

In reply to Viktor Malik

Hello,

Quoted 34 lines · Viktor Malik

The loop for processing syscall args in augment_raw_syscalls has a history of breaking with Clang updates. In the past, we've seen it break (i.e. stop passing the BPF verifier) between Clang 15 and 16 and now a similar thing happened between Clang 21 and 22. While the issue is mitigated on the main line by a recent verifier update, it remains broken on the 6.12 and 6.18 stable branches, effectively breaking perf trace.
The reason is that the loop is quite complex and the BPF verifier often struggles to prove that it terminates.
This series fixes the issue by replacing the standard for loop by the bpf_for macro, which uses numeric BPF iterator. This should prevent future breakages of this kind since the verifier has much easier job proving that the loop terminates. Small adjustments were necessary for the loop to make it work, see the second commit message for details.
To keep perf compatible with older kernels, the first commit factors out the loop body into a function, which is then called either from bpf_for or from a standard for loop, depending on whether BPF numeric iterators are available.
Changes from v3:
- v3: https://lore.kernel.org/bpf/[email protected]/
- Do not remove any hacks from the BPF program to keep it backwards compatible with older kernels and Clang versions (Sashiko).
Changes from v2:
- v2: https://lore.kernel.org/bpf/[email protected]/T/
- Small refactoring suggested by Namhhyung
- Make size in augment_arg() an int (found by Sashiko)
- Keep the original way of clamping aug_size to TRACE_AUG_MAX_BUF (found by Sashiko)

Quoted 5 lines

Changes from v1:
- v1: https://lore.kernel.org/bpf/[email protected]/T/
- Use bpf_for instead of bpf_loop (suggested by Alexei and Andrii)
- Keep the change backwards compatible with older kernels (required by Namhyung)

Andrii, are you ok with this?

Thanks,
Namhyung

Quoted 11 lines

Viktor Malik (2):
perf trace: Factor out BPF loop body
perf trace: Refactor augmented_raw_syscalls using bpf_for
 .../bpf_skel/augmented_raw_syscalls.bpf.c     | 155 +++++++++++-------
 1 file changed, 98 insertions(+), 57 deletions(-)
Signature · 1 line
2.54.0

Namhyung Kim

maintainer Applied

2026-07-17 17:46 UTC #4 lore raw

In reply to Viktor Malik

Quoted 9 lines · Viktor Malik

The loop for processing syscall args in augment_raw_syscalls has a history of breaking with Clang updates. In the past, we've seen it break (i.e. stop passing the BPF verifier) between Clang 15 and 16 and now a similar thing happened between Clang 21 and 22. While the issue is mitigated on the main line by a recent verifier update, it remains broken on the 6.12 and 6.18 stable branches, effectively breaking perf trace.
[...]

Applied to perf-tools-next, thanks!

Best regards,
Namhyung