tracing: Read user data from futex system call trace event

PATCH Superseded Steven Rostedt 3 participants 6 messages Mar 3 – Mar 5 Tracing

lore

Share & tools

lore

Steven Rostedt

author

2026-03-03 21:49 UTC #0 lore raw

cover tracing: Read user data from futex system call trace event

We are looking at the performance of futexes and require a bit more information when tracing them.

The two patches here extend the system call reading of user space to create specific handling of the futex system call. It now reads the user space relevant data (the addr, utime and addr2), as well as parses the flags. This adds a little smarts to the trace event as it only shows the parameters that are relevant, as well as parses utime as either a timespec or as val2 depending on the futex_op.

Here's an example of the new output:

   futex_requeue-1431    [004] .....   154.082233: sys_futex(uaddr: 0x7ffcadad2ff8 (0), FUTEX_WAIT, val: 0, timespec: 0x7f0785f83e90 (0.030000000))
   futex_requeue-1421    [002] .....   154.092286: sys_futex(uaddr: 0x7ffcadad2ff8 (0), FUTEX_CMP_REQUEUE, val: 3, val2: 0x7, uaddr2: 0x7ffcadad2ffc (0), val3: 0)
   futex_requeue-1421    [002] .....   154.092321: sys_futex(uaddr: 0x7ffcadad2ffc (0), FUTEX_WAKE, val: 2147483647)
     futex_waitv-1433    [006] .....   154.111394: sys_futex(uaddr: 0x55ed4a768554 (0), FUTEX_WAKE|FUTEX_PRIVATE_FLAG, val: 1)
     futex_waitv-1435    [002] .....   154.128186: sys_futex(uaddr: 0x7f035fa40000 (0), FUTEX_WAKE, val: 1)

Steven Rostedt (2):
tracing: Have futex syscall trace event show specific user data
tracing: Update futex syscall trace event to show more commands

----

 kernel/trace/trace_syscalls.c | 370 +++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 367 insertions(+), 3 deletions(-)

Steven Rostedt

author

2026-03-03 21:49 UTC #1 lore raw

PATCH 1/2 tracing: Have futex syscall trace event show specific user data

From: Steven Rostedt <[email protected]>

Add specific reporting of the futex system call. This allows for debugging the futex code a bit easier. Instead of just showing the values passed into the futex system call, read the value of the user space memory pointed to by the addr parameter.

Also make the op parameter more readable by parsing the values to show what the command is:

 futex_requeue_p-3251    [002] .....  2101.068479: sys_futex(uaddr: 0x55e79a4da834 (0x80000cb1), FUTEX_LOCK_PI|FUTEX_PRIVATE_FLAG, val: 0)
 futex_requeue_p-3248    [001] .....  2101.068970: sys_futex(uaddr: 0x7f859072f990 (0xcb2), FUTEX_WAIT_BITSET|FUTEX_CLOCK_REALTIME, val: 3250)
 futex_requeue_p-3252    [005] .....  2101.069108: sys_futex(uaddr: 0x55e79a4da838 (0), FUTEX_WAIT_REQUEUE_PI|FUTEX_PRIVATE_FLAG, val: 0, timespec: 0x7ffe61076aa0, uaddr2: 0x55e79a4da834, uaddr2: 94453214586932, val3: 0)
 futex_requeue_p-3252    [005] .....  2101.069410: sys_futex(uaddr: 0x55e79a4da834 (0x80000cb1), FUTEX_LOCK_PI|FUTEX_PRIVATE_FLAG, val: 0)

Signed-off-by Steven Rostedt (Google) <[email protected]>

---
 kernel/trace/trace_syscalls.c | 266 +++++++++++++++++++++++++++++++++-
 1 file changed, 263 insertions(+), 3 deletions(-)

kernel/trace/trace_syscalls.c +263 -3

index e96d0063cbcf..bc60a0497bcc 100644
@@ -11,6 +11,8 @@
 #include <linux/xarray.h>
 #include <asm/syscall.h>
 
+#include <uapi/linux/futex.h>
+
 #include "trace_output.h"
 #include "trace.h"

@@ -237,6 +239,125 @@ sys_enter_openat_print(struct syscall_trace_enter *trace, struct syscall_metadat
 	return trace_handle_return(s);
 }
 
+static const char * __futex_cmds[] =
+{
+	"FUTEX_WAIT", "FUTEX_WAKE", "FUTEX_FD", "FUTEX_REQUEUE",
+	"FUTEX_CMP_REQUEUE", "FUTEX_WAKE_OP", "FUTEX_LOCK_PI",
+	"FUTEX_UNLOCK_PI", "FUTEX_TRYLOCK_PI", "FUTEX_WAIT_BITSET",
+	"FUTEX_WAKE_BITSET", "FUTEX_WAIT_REQUEUE_PI", "FUTEX_CMP_REQUEUE_PI",
+	"FUTEX_LOCK_PI2", };
+
+/* From futex/syscalls.c */
+static __always_inline bool futex_cmd_has_timeout(u32 cmd)
+{
+	switch (cmd) {
+	case FUTEX_WAIT:
+	case FUTEX_LOCK_PI:
+	case FUTEX_LOCK_PI2:
+	case FUTEX_WAIT_BITSET:
+	case FUTEX_WAIT_REQUEUE_PI:
+		return true;
+	}
+	return false;
+}
+
+static __always_inline bool futex_cmd_has_addr2(u32 cmd)
+{
+	switch (cmd) {
+	case FUTEX_REQUEUE:
+	case FUTEX_CMP_REQUEUE:
+	case FUTEX_WAKE_OP:
+	case FUTEX_WAIT_REQUEUE_PI:
+		return true;
+	}
+	return false;
+}
+
+static enum print_line_t
+sys_enter_futex_print(struct syscall_trace_enter *trace, struct syscall_metadata *entry,
+		      struct trace_seq *s, struct trace_event *event, int ent_size)
+{
+	bool done = false;
+	unsigned int op, cmd;
+	void *end = (void *)trace + ent_size;
+	void *ptr;
+	int i;
+
+	/* Set ptr to the user space copied area */
+	ptr = (void *)trace->args + sizeof(unsigned long) * entry->nb_args;
+	if (ptr + 4 > end)
+		ptr = NULL;
+
+	trace_seq_printf(s, "%s(", entry->name);
+
+	for (i = 0; !done && i < entry->nb_args; i++) {
+
+		if (trace_seq_has_overflowed(s))
+			goto end;
+
+		switch (i) {
+		case 0:
+			trace_seq_printf(s, "uaddr: 0x%lx", trace->args[i]);
+			if (ptr) {
+				u32 val = *(u32 *)ptr;
+				if (val < 10)
+					trace_seq_printf(s, " (%u)", val);
+				else
+					trace_seq_printf(s, " (0x%x)", val);
+			}
+			continue;
+		case 1:
+			op = trace->args[i];
+			cmd = op & FUTEX_CMD_MASK;
+			if (cmd <= FUTEX_LOCK_PI2)
+				trace_seq_printf(s, ", %s", __futex_cmds[cmd]);
+			else
+				trace_seq_puts(s, ", UNKNOWN");
+
+			if (op & FUTEX_PRIVATE_FLAG)
+				trace_seq_puts(s, "|FUTEX_PRIVATE_FLAG");
+			if (op & FUTEX_CLOCK_REALTIME)
+				trace_seq_puts(s, "|FUTEX_CLOCK_REALTIME");
+			continue;
+		case 3:
+			if (!futex_cmd_has_timeout(cmd)) {
+
+				if (!futex_cmd_has_addr2(cmd)) {
+					done = true;
+					continue;
+				}
+
+				trace_seq_printf(s, ", val2: 0x%x",
+						 (u32)(long)trace->args[i]);
+				continue;
+			}
+
+			if (!trace->args[i])
+				continue;
+
+			trace_seq_printf(s, ", timespec: 0x%lx",
+					 trace->args[i]);
+
+			continue;
+
+		case 4:
+			if (!futex_cmd_has_addr2(cmd)) {
+				done = true;
+				continue;
+			}
+			trace_seq_printf(s, ", uaddr2: 0x%lx", trace->args[i]);
+		}
+
+		trace_seq_printf(s, ", %s: %lu", entry->args[i],
+				 trace->args[i]);
+	}
+	trace_seq_putc(s, ')');
+end:
+	trace_seq_putc(s, '\n');
+
+	return trace_handle_return(s);
+}
+
 static enum print_line_t
 print_syscall_enter(struct trace_iterator *iter, int flags,
 		    struct trace_event *event)

@@ -267,6 +388,10 @@ print_syscall_enter(struct trace_iterator *iter, int flags,
 		if (!tr || !(tr->trace_flags & TRACE_ITER(VERBOSE)))
 			return sys_enter_openat_print(trace, entry, s, event);
 		break;
+	case __NR_futex:
+		if (!tr || !(tr->trace_flags & TRACE_ITER(VERBOSE)))
+			return sys_enter_futex_print(trace, entry, s, event, iter->ent_size);
+		break;
 	default:
 		break;
 	}

@@ -437,6 +562,69 @@ sys_enter_openat_print_fmt(struct syscall_metadata *entry, char *buf, int len)
 	return pos;
 }
 
+static int __init
+sys_enter_futex_print_fmt(struct syscall_metadata *entry, char *buf, int len)
+{
+	int pos = 0;
+
+	pos += snprintf(buf + pos, LEN_OR_ZERO,
+			"\"uaddr: 0x%%lx (0x%%lx) cmd=%%s%%s%%s");
+	pos += snprintf(buf + pos, LEN_OR_ZERO,
+			"  val: 0x%%x timeout/val2: 0x%%llx");
+	pos += snprintf(buf + pos, LEN_OR_ZERO,
+			" uaddr2: 0x%%lx val3: 0x%%x\", ");
+
+	pos += snprintf(buf + pos, LEN_OR_ZERO,
+			" REC->uaddr,");
+	pos += snprintf(buf + pos, LEN_OR_ZERO,
+			" REC->__value,");
+	pos += snprintf(buf + pos, LEN_OR_ZERO,
+			"  __print_symbolic(REC->op & 0x%x, ", FUTEX_CMD_MASK);
+
+	pos += snprintf(buf + pos, LEN_OR_ZERO,
+			"{%d, \"FUTEX_WAIT\"}, ", FUTEX_WAIT);
+	pos += snprintf(buf + pos, LEN_OR_ZERO,
+			"{%d, \"FUTEX_WAKE\"}, ", FUTEX_WAKE);
+	pos += snprintf(buf + pos, LEN_OR_ZERO,
+			"{%d, \"FUTEX_FD\"}, ", FUTEX_FD);
+	pos += snprintf(buf + pos, LEN_OR_ZERO,
+			"{%d, \"FUTEX_REQUEUE\"}, ", FUTEX_REQUEUE);
+	pos += snprintf(buf + pos, LEN_OR_ZERO,
+			"{%d, \"FUTEX_CMP_REQUEUE\"}, ", FUTEX_CMP_REQUEUE);
+	pos += snprintf(buf + pos, LEN_OR_ZERO,
+			"{%d, \"FUTEX_WAKE_OP\"}, ", FUTEX_WAKE_OP);
+	pos += snprintf(buf + pos, LEN_OR_ZERO,
+			"{%d, \"FUTEX_LOCK_PI\"}, ", FUTEX_LOCK_PI);
+	pos += snprintf(buf + pos, LEN_OR_ZERO,
+			"{%d, \"FUTEX_UNLOCK_PI\"}, ", FUTEX_UNLOCK_PI);
+	pos += snprintf(buf + pos, LEN_OR_ZERO,
+			"{%d, \"FUTEX_TRYLOCK_PI\"}, ", FUTEX_TRYLOCK_PI);
+	pos += snprintf(buf + pos, LEN_OR_ZERO,
+			"{%d, \"FUTEX_WAIT_BITSET\"}, ", FUTEX_WAIT_BITSET);
+	pos += snprintf(buf + pos, LEN_OR_ZERO,
+			"{%d, \"FUTEX_WAKE_BITSET\"}, ", FUTEX_WAKE_BITSET);
+	pos += snprintf(buf + pos, LEN_OR_ZERO,
+			"{%d, \"FUTEX_WAIT_REQUEUE_PI\"}, ", FUTEX_WAIT_REQUEUE_PI);
+	pos += snprintf(buf + pos, LEN_OR_ZERO,
+			"{%d, \"FUTEX_CMP_REQUEUE_PI\"}, ", FUTEX_CMP_REQUEUE_PI);
+	pos += snprintf(buf + pos, LEN_OR_ZERO,
+			"{%d, \"FUTEX_LOCK_PI2\"}),", FUTEX_LOCK_PI2);
+
+	pos += snprintf(buf + pos, LEN_OR_ZERO,
+			" (REC->op & %d) ? \"|FUTEX_PRIVATE_FLAG\" : \"\",",
+			FUTEX_PRIVATE_FLAG);
+	pos += snprintf(buf + pos, LEN_OR_ZERO,
+			" (REC->op & %d) ? \"|FUTEX_CLOCK_REALTIME\" : \"\",",
+			FUTEX_CLOCK_REALTIME);
+
+	pos += snprintf(buf + pos, LEN_OR_ZERO,
+			" REC->val, REC->utime,");
+
+	pos += snprintf(buf + pos, LEN_OR_ZERO,
+			" REC->uaddr, REC->val3");
+	return pos;
+}
+
 static int __init
 __set_enter_print_fmt(struct syscall_metadata *entry, char *buf, int len)
 {

@@ -447,6 +635,8 @@ __set_enter_print_fmt(struct syscall_metadata *entry, char *buf, int len)
 	switch (entry->syscall_nr) {
 	case __NR_openat:
 		return sys_enter_openat_print_fmt(entry, buf, len);
+	case __NR_futex:
+		return sys_enter_futex_print_fmt(entry, buf, len);
 	default:
 		break;
 	}

@@ -523,6 +713,21 @@ static void __init free_syscall_print_fmt(struct trace_event_call *call)
 		kfree(call->print_fmt);
 }
 
+static int __init futex_fields(struct trace_event_call *call, int offset)
+{
+	char *arg;
+	int ret;
+
+	arg = kstrdup("__value", GFP_KERNEL);
+	if (WARN_ON_ONCE(!arg))
+		return -ENOMEM;
+	ret = trace_define_field(call, "u32", arg, offset, sizeof(int), 0,
+				 FILTER_OTHER);
+	if (ret)
+		kfree(arg);
+	return ret;
+}
+
 static int __init syscall_enter_define_fields(struct trace_event_call *call)
 {
 	struct syscall_trace_enter trace;

@@ -544,6 +749,9 @@ static int __init syscall_enter_define_fields(struct trace_event_call *call)
 		offset += sizeof(unsigned long);
 	}
 
+	if (!ret && meta->syscall_nr == __NR_futex)
+		return futex_fields(call, offset);
+
 	if (ret || !meta->user_mask)
 		return ret;

@@ -689,6 +897,48 @@ static int syscall_copy_user_array(char *buf, const char __user *ptr,
 	return 0;
 }
 
+static int
+syscall_get_futex(unsigned long *args, char **buffer, int *size, int buf_size)
+{
+	struct syscall_user_buffer *sbuf;
+	const char __user *ptr;
+	char *buf;
+
+	/* buf_size of zero means user doesn't want user space read */
+	if (!buf_size)
+		return -1;
+
+	/* If the syscall_buffer is NULL, tracing is being shutdown */
+	sbuf = READ_ONCE(syscall_buffer);
+	if (!sbuf)
+		return -1;
+
+	ptr = (char __user *)args[0];
+
+	*buffer = trace_user_fault_read(&sbuf->buf, ptr, 4, NULL, NULL);
+	if (!*buffer)
+		return -1;
+
+	/* Add room for the value */
+	*size += 4;
+
+	buf = *buffer;
+
+	return 0;
+}
+
+static void syscall_put_futex(struct syscall_metadata *sys_data,
+			      struct syscall_trace_enter *entry,
+			      char *buffer)
+{
+	u32 *ptr;
+
+	/* Place the futex key into the storage */
+	ptr = (void *)entry->args + sizeof(unsigned long) * sys_data->nb_args;
+
+	*ptr = *(u32 *)buffer;
+}
+
 static char *sys_fault_user(unsigned int buf_size,
 			    struct syscall_metadata *sys_data,
 			    struct syscall_user_buffer *sbuf,

@@ -905,6 +1155,9 @@ static void ftrace_syscall_enter(void *data, struct pt_regs *regs, long id)
 		if (syscall_get_data(sys_data, args, &user_ptr,
 				     &size, user_sizes, &uargs, tr->syscall_buf_sz) < 0)
 			return;
+	} else if (syscall_nr == __NR_futex) {
+		if (syscall_get_futex(args, &user_ptr, &size, tr->syscall_buf_sz) < 0)
+			return;
 	}
 
 	size += sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;

@@ -921,6 +1174,9 @@ static void ftrace_syscall_enter(void *data, struct pt_regs *regs, long id)
 	if (mayfault)
 		syscall_put_data(sys_data, entry, user_ptr, size, user_sizes, uargs);
 
+	else if (syscall_nr == __NR_futex)
+		syscall_put_futex(sys_data, entry, user_ptr);
+
 	trace_event_buffer_commit(&fbuffer);
 }

@@ -971,14 +1227,18 @@ static int reg_event_syscall_enter(struct trace_event_file *file,
 {
 	struct syscall_metadata *sys_data = call->data;
 	struct trace_array *tr = file->tr;
+	bool enable_faults;
 	int ret = 0;
 	int num;
 
 	num = sys_data->syscall_nr;
 	if (WARN_ON_ONCE(num < 0 || num >= NR_syscalls))
 		return -ENOSYS;
+
+	enable_faults = sys_data->user_mask || num == __NR_futex;
+
 	guard(mutex)(&syscall_trace_lock);
-	if (sys_data->user_mask) {
+	if (enable_faults) {
 		ret = syscall_fault_buffer_enable();
 		if (ret < 0)
 			return ret;

@@ -986,7 +1246,7 @@ static int reg_event_syscall_enter(struct trace_event_file *file,
 	if (!tr->sys_refcount_enter) {
 		ret = register_trace_sys_enter(ftrace_syscall_enter, tr);
 		if (ret < 0) {
-			if (sys_data->user_mask)
+			if (enable_faults)
 				syscall_fault_buffer_disable();
 			return ret;
 		}

@@ -1011,7 +1271,7 @@ static void unreg_event_syscall_enter(struct trace_event_file *file,
 	WRITE_ONCE(tr->enter_syscall_files[num], NULL);
 	if (!tr->sys_refcount_enter)
 		unregister_trace_sys_enter(ftrace_syscall_enter, tr);
-	if (sys_data->user_mask)
+	if (sys_data->user_mask || num == __NR_futex)
 		syscall_fault_buffer_disable();
 }

Signature · 1 line

2.51.0

Steven Rostedt

author

2026-03-03 21:49 UTC #4 lore raw

PATCH 2/2 tracing: Update futex syscall trace event to show more commands

In reply to Steven Rostedt

From: Steven Rostedt <[email protected]>

Make the futex syscall trace event a little more smart. Have it read the futex_op instruction to determine what else it can save and print. For the appropriate options, it will read the utime (timespec) parameter and show its output as well as the uaddr2.

 futex_requeue_p-1154    [004] .....   144.568339: sys_futex(uaddr: 0x5652b178d834 (0x482), FUTEX_UNLOCK_PI|FUTEX_PRIVATE_FLAG, val: 0)
 futex_requeue_p-1162    [002] .....   144.568696: sys_futex(uaddr: 0x7f763b7fece0 (2), FUTEX_WAIT|FUTEX_PRIVATE_FLAG, val: 2)
 futex_requeue_p-1151    [000] .....   144.568700: sys_futex(uaddr: 0x7f763b7fece0 (0), FUTEX_WAKE|FUTEX_PRIVATE_FLAG, val: 1)
 futex_requeue_p-1162    [002] .....   144.568705: sys_futex(uaddr: 0x7f763b7fece0 (0), FUTEX_WAKE|FUTEX_PRIVATE_FLAG, val: 1)
 futex_requeue_p-1151    [000] .....   144.568715: sys_futex(uaddr: 0x7f764369e990 (0x483), FUTEX_WAIT_BITSET|FUTEX_CLOCK_REALTIME, val: 1155)
 futex_requeue_p-1155    [005] .....   144.569420: sys_futex(uaddr: 0x5652b178d838 (0), FUTEX_WAIT_REQUEUE_PI|FUTEX_PRIVATE_FLAG, val: 0, timespec: 0x7ffdacfba500 (143.890024054), uaddr2: 0x5652b178d834 (0), val3: 0)
 futex_requeue_p-1155    [005] .....   144.569454: sys_futex(uaddr: 0x5652b178d834 (0), FUTEX_LOCK_PI|FUTEX_PRIVATE_FLAG, val: 0)

Signed-off-by Steven Rostedt (Google) <[email protected]>

---
 kernel/trace/trace_syscalls.c | 138 +++++++++++++++++++++++++++++-----
 1 file changed, 121 insertions(+), 17 deletions(-)

kernel/trace/trace_syscalls.c +121 ~~-17~~

index bc60a0497bcc..d128dcb218a7 100644
@@ -273,10 +273,18 @@ static __always_inline bool futex_cmd_has_addr2(u32 cmd)
 	return false;
 }
 
+struct futex_data {
+	u32		val1;
+	u32		val2;
+	unsigned long	ts1;
+	unsigned long	ts2;
+};
+
 static enum print_line_t
 sys_enter_futex_print(struct syscall_trace_enter *trace, struct syscall_metadata *entry,
 		      struct trace_seq *s, struct trace_event *event, int ent_size)
 {
+	struct futex_data *data;
 	bool done = false;
 	unsigned int op, cmd;
 	void *end = (void *)trace + ent_size;

@@ -285,8 +293,10 @@ sys_enter_futex_print(struct syscall_trace_enter *trace, struct syscall_metadata
 
 	/* Set ptr to the user space copied area */
 	ptr = (void *)trace->args + sizeof(unsigned long) * entry->nb_args;
-	if (ptr + 4 > end)
-		ptr = NULL;
+	if (ptr + sizeof(*data) > end)
+		data = NULL;
+	else
+		data = ptr;
 
 	trace_seq_printf(s, "%s(", entry->name);

@@ -298,8 +308,8 @@ sys_enter_futex_print(struct syscall_trace_enter *trace, struct syscall_metadata
 		switch (i) {
 		case 0:
 			trace_seq_printf(s, "uaddr: 0x%lx", trace->args[i]);
-			if (ptr) {
-				u32 val = *(u32 *)ptr;
+			if (data) {
+				u32 val = data->val1;
 				if (val < 10)
 					trace_seq_printf(s, " (%u)", val);
 				else

@@ -338,6 +348,15 @@ sys_enter_futex_print(struct syscall_trace_enter *trace, struct syscall_metadata
 			trace_seq_printf(s, ", timespec: 0x%lx",
 					 trace->args[i]);
 
+			if (!data)
+				continue;
+
+			if (!data->ts1 && !data->ts2) {
+				trace_seq_puts(s, " (0)");
+				continue;
+			}
+			trace_seq_printf(s, " (%lu.%09lu)",
+					 data->ts1, data->ts2);
 			continue;
 
 		case 4:

@@ -346,6 +365,12 @@ sys_enter_futex_print(struct syscall_trace_enter *trace, struct syscall_metadata
 				continue;
 			}
 			trace_seq_printf(s, ", uaddr2: 0x%lx", trace->args[i]);
+
+			if (!data)
+				continue;
+
+			trace_seq_printf(s, " (%x)", data->val2);
+			continue;
 		}
 
 		trace_seq_printf(s, ", %s: %lu", entry->args[i],

@@ -570,9 +595,9 @@ sys_enter_futex_print_fmt(struct syscall_metadata *entry, char *buf, int len)
 	pos += snprintf(buf + pos, LEN_OR_ZERO,
 			"\"uaddr: 0x%%lx (0x%%lx) cmd=%%s%%s%%s");
 	pos += snprintf(buf + pos, LEN_OR_ZERO,
-			"  val: 0x%%x timeout/val2: 0x%%llx");
+			"  val: 0x%%x timeout/val2: 0x%%llx (%%lu.%%lu)");
 	pos += snprintf(buf + pos, LEN_OR_ZERO,
-			" uaddr2: 0x%%lx val3: 0x%%x\", ");
+			" uaddr2: 0x%%lx (0x%%lx) val3: 0x%%x\", ");
 
 	pos += snprintf(buf + pos, LEN_OR_ZERO,
 			" REC->uaddr,");

@@ -618,10 +643,12 @@ sys_enter_futex_print_fmt(struct syscall_metadata *entry, char *buf, int len)
 			FUTEX_CLOCK_REALTIME);
 
 	pos += snprintf(buf + pos, LEN_OR_ZERO,
-			" REC->val, REC->utime,");
+			" REC->val, REC->utime, REC->__ts1, REC->__ts2,");
 
 	pos += snprintf(buf + pos, LEN_OR_ZERO,
-			" REC->uaddr, REC->val3");
+			" REC->uaddr,");
+	pos += snprintf(buf + pos, LEN_OR_ZERO,
+			" REC->__value2, REC->val3");
 	return pos;
 }

@@ -724,7 +751,39 @@ static int __init futex_fields(struct trace_event_call *call, int offset)
 	ret = trace_define_field(call, "u32", arg, offset, sizeof(int), 0,
 				 FILTER_OTHER);
 	if (ret)
-		kfree(arg);
+		goto free;
+	offset += sizeof(int);
+
+	arg = kstrdup("__value2", GFP_KERNEL);
+	if (WARN_ON_ONCE(!arg))
+		return -ENOMEM;
+	ret = trace_define_field(call, "u32", arg, offset, sizeof(int), 0,
+				 FILTER_OTHER);
+	if (ret)
+		goto free;
+	offset += sizeof(int);
+
+	arg = kstrdup("__ts1", GFP_KERNEL);
+	if (WARN_ON_ONCE(!arg))
+		return -ENOMEM;
+	ret = trace_define_field(call, "unsigned long", arg, offset,
+				 sizeof(unsigned long), 0, FILTER_OTHER);
+	if (ret)
+		goto free;
+	offset += sizeof(long);
+
+	arg = kstrdup("__ts2", GFP_KERNEL);
+	if (WARN_ON_ONCE(!arg))
+		return -ENOMEM;
+	ret = trace_define_field(call, "unsigned long", arg, offset,
+				 sizeof(unsigned long), 0, FILTER_OTHER);
+	if (ret)
+		goto free;
+
+	return 0;
+
+free:
+	kfree(arg);
 	return ret;
 }

@@ -897,11 +956,51 @@ static int syscall_copy_user_array(char *buf, const char __user *ptr,
 	return 0;
 }
 
+struct tp_futex_data {
+	u32			cmd;
+	const u32		__user *val1;
+	const u32 		__user *val2;
+	void			__user *timeout;
+};
+
+static int syscall_copy_futex(char *buf, const char __user *ptr,
+			      size_t size, void *data)
+{
+	struct tp_futex_data *tp_data = data;
+	struct futex_data *fdata = (void *)buf;
+	int cmd = tp_data->cmd & FUTEX_CMD_MASK;
+	int ret;
+
+	memset(fdata, 0, sizeof(*fdata));
+
+	if (tp_data->val1) {
+		ret = __copy_from_user(&fdata->val1, tp_data->val1, 4);
+		if (ret)
+			return -1;
+	}
+
+	if (tp_data->val2 && futex_cmd_has_addr2(cmd)) {
+		ret = __copy_from_user(&fdata->val2, tp_data->val2, 4);
+		if (ret)
+			return -1;
+	}
+
+	if (tp_data->timeout && futex_cmd_has_timeout(cmd)) {
+		/* Copies both ts1 and ts2 */
+		ret = __copy_from_user(&fdata->ts1, tp_data->timeout,
+				       sizeof(long) * 2);
+		if (ret)
+			return -1;
+	}
+
+	return 0;
+}
+
 static int
 syscall_get_futex(unsigned long *args, char **buffer, int *size, int buf_size)
 {
 	struct syscall_user_buffer *sbuf;
-	const char __user *ptr;
+	struct tp_futex_data tp_data;
 	char *buf;
 
 	/* buf_size of zero means user doesn't want user space read */

@@ -913,14 +1012,18 @@ syscall_get_futex(unsigned long *args, char **buffer, int *size, int buf_size)
 	if (!sbuf)
 		return -1;
 
-	ptr = (char __user *)args[0];
+	tp_data.cmd = args[1];
+	tp_data.val1 = (u32 __user *)args[0];
+	tp_data.val2 = (u32 __user *)args[4];
+	tp_data.timeout = (u64 __user *)args[3];
 
-	*buffer = trace_user_fault_read(&sbuf->buf, ptr, 4, NULL, NULL);
+	*buffer = trace_user_fault_read(&sbuf->buf, NULL, 0,
+					syscall_copy_futex, &tp_data);
 	if (!*buffer)
 		return -1;
 
-	/* Add room for the value */
-	*size += 4;
+	/* Add room for values */
+	*size += sizeof(struct futex_data);
 
 	buf = *buffer;

@@ -931,12 +1034,13 @@ static void syscall_put_futex(struct syscall_metadata *sys_data,
 			      struct syscall_trace_enter *entry,
 			      char *buffer)
 {
-	u32 *ptr;
+	struct futex_data *fdata = (void *)buffer;
+	struct futex_data *data;
 
 	/* Place the futex key into the storage */
-	ptr = (void *)entry->args + sizeof(unsigned long) * sys_data->nb_args;
+	data = (void *)entry->args + sizeof(unsigned long) * sys_data->nb_args;
 
-	*ptr = *(u32 *)buffer;
+	*data = *fdata;
 }
 
 static char *sys_fault_user(unsigned int buf_size,

Signature · 1 line

2.51.0

Steven Rostedt

maintainer

2026-03-03 22:44 UTC #5 lore raw

[PATCH 3/2] tracing: Show TID and flags for PI futex system call trace event

In reply to Steven Rostedt

From: Steven Rostedt <[email protected]>

For the futex system call trace event for FUTEX_LOCK_PI and FUTEX_UNLOCK_PI commands, show the TID from the value (which is usually in hex) as well as translate the flags (DIED and WAITERS).

 pi_mutex_hammer-1098    [000] .....   121.876928: sys_futex(uaddr: 0x560f40cc8180 (0x450) tid: 1104, FUTEX_LOCK_PI|FUTEX_PRIVATE_FLAG, val: 0, timespec: 0x7f2f9d4b1e50 (0.000100000))
 pi_mutex_hammer-1128    [000] .....   121.877120: sys_futex(uaddr: 0x560f40cc8180 (0x8000042a) tid: 1066 (WAITERS), FUTEX_LOCK_PI|FUTEX_PRIVATE_FLAG, val: 0, timespec: 0x7f2f8e493e50 (0.000100000))
 pi_mutex_hammer-1106    [000] .....   121.877242: sys_futex(uaddr: 0x560f40cc8180 (0x80000452) tid: 1106 (WAITERS), FUTEX_UNLOCK_PI|FUTEX_PRIVATE_FLAG, val: 0)

This makes it easier to see the hand off of a mutex and who the owner was.

Signed-off-by Steven Rostedt (Google) <[email protected]>

---
 kernel/trace/trace_syscalls.c | 25 +++++++++++++++++++++++--
 1 file changed, 23 insertions(+), 2 deletions(-)

kernel/trace/trace_syscalls.c +23 -2

index d128dcb218a7..b3eddfee1a6b 100644
@@ -300,6 +300,9 @@ sys_enter_futex_print(struct syscall_trace_enter *trace, struct syscall_metadata
 
 	trace_seq_printf(s, "%s(", entry->name);
 
+	op = trace->args[1];
+	cmd = op & FUTEX_CMD_MASK;
+
 	for (i = 0; !done && i < entry->nb_args; i++) {
 
 		if (trace_seq_has_overflowed(s))

@@ -314,11 +317,29 @@ sys_enter_futex_print(struct syscall_trace_enter *trace, struct syscall_metadata
 					trace_seq_printf(s, " (%u)", val);
 				else
 					trace_seq_printf(s, " (0x%x)", val);
+
+				switch(cmd) {
+				case FUTEX_LOCK_PI:
+				case FUTEX_UNLOCK_PI:
+					trace_seq_printf(s, " tid: %d", val & FUTEX_TID_MASK);
+
+					if (!(val & (FUTEX_OWNER_DIED|FUTEX_WAITERS)))
+						break;
+
+					trace_seq_puts(s, " (");
+					if (val & FUTEX_WAITERS)
+						trace_seq_puts(s, "WAITERS");
+					if (val & FUTEX_OWNER_DIED) {
+						if (op & FUTEX_WAITERS)
+							trace_seq_putc(s, '|');
+						trace_seq_puts(s, "DIED");
+					}
+					trace_seq_putc(s, ')');
+					break;
+				}
 			}
 			continue;
 		case 1:
-			op = trace->args[i];
-			cmd = op & FUTEX_CMD_MASK;
 			if (cmd <= FUTEX_LOCK_PI2)
 				trace_seq_printf(s, ", %s", __futex_cmds[cmd]);
 			else

Signature · 1 line

2.51.0

Peter Zijlstra

2026-03-04 09:07 UTC #2 lore raw

Re: [PATCH 1/2] tracing: Have futex syscall trace event show specific user data

In reply to Steven Rostedt

Quoted 15 lines · Steven Rostedt

From: Steven Rostedt <[email protected]>
Add specific reporting of the futex system call. This allows for debugging the futex code a bit easier. Instead of just showing the values passed into the futex system call, read the value of the user space memory pointed to by the addr parameter.
Also make the op parameter more readable by parsing the values to show what the command is:
 futex_requeue_p-3251    [002] .....  2101.068479: sys_futex(uaddr: 0x55e79a4da834 (0x80000cb1), FUTEX_LOCK_PI|FUTEX_PRIVATE_FLAG, val: 0)
 futex_requeue_p-3248    [001] .....  2101.068970: sys_futex(uaddr: 0x7f859072f990 (0xcb2), FUTEX_WAIT_BITSET|FUTEX_CLOCK_REALTIME, val: 3250)
 futex_requeue_p-3252    [005] .....  2101.069108: sys_futex(uaddr: 0x55e79a4da838 (0), FUTEX_WAIT_REQUEUE_PI|FUTEX_PRIVATE_FLAG, val: 0, timespec: 0x7ffe61076aa0, uaddr2: 0x55e79a4da834, uaddr2: 94453214586932, val3: 0)
 futex_requeue_p-3252    [005] .....  2101.069410: sys_futex(uaddr: 0x55e79a4da834 (0x80000cb1), FUTEX_LOCK_PI|FUTEX_PRIVATE_FLAG, val: 0)

Quoted 4 lines

Signed-off-by Steven Rostedt (Google) <[email protected]>
---
 kernel/trace/trace_syscalls.c | 266 +++++++++++++++++++++++++++++++++-
 1 file changed, 263 insertions(+), 3 deletions(-)

Egads, I really dislike how all sorts of syscall crud is 'duplicated' in this file, rather than near or in the actual syscall definition.

Steven Rostedt

author

2026-03-05 01:37 UTC #3 lore raw

Re: [PATCH 1/2] tracing: Have futex syscall trace event show specific user data

Quoted 16 lines · On Wed, 4 Mar 2026 10:07:48 +0100 Peter Zijlstra <[email protected]> wrote:

Quoted 15 lines · Steven Rostedt
From: Steven Rostedt <[email protected]>
Add specific reporting of the futex system call. This allows for debugging the futex code a bit easier. Instead of just showing the values passed into the futex system call, read the value of the user space memory pointed to by the addr parameter.
Also make the op parameter more readable by parsing the values to show what the command is:
 futex_requeue_p-3251    [002] .....  2101.068479: sys_futex(uaddr: 0x55e79a4da834 (0x80000cb1), FUTEX_LOCK_PI|FUTEX_PRIVATE_FLAG, val: 0)
 futex_requeue_p-3248    [001] .....  2101.068970: sys_futex(uaddr: 0x7f859072f990 (0xcb2), FUTEX_WAIT_BITSET|FUTEX_CLOCK_REALTIME, val: 3250)
 futex_requeue_p-3252    [005] .....  2101.069108: sys_futex(uaddr: 0x55e79a4da838 (0), FUTEX_WAIT_REQUEUE_PI|FUTEX_PRIVATE_FLAG, val: 0, timespec: 0x7ffe61076aa0, uaddr2: 0x55e79a4da834, uaddr2: 94453214586932, val3: 0)
 futex_requeue_p-3252    [005] .....  2101.069410: sys_futex(uaddr: 0x55e79a4da834 (0x80000cb1), FUTEX_LOCK_PI|FUTEX_PRIVATE_FLAG, val: 0)

Quoted 7 lines

Quoted 4 lines
Signed-off-by Steven Rostedt (Google) <[email protected]>
---
 kernel/trace/trace_syscalls.c | 266 +++++++++++++++++++++++++++++++++-
 1 file changed, 263 insertions(+), 3 deletions(-)  
Egads, I really dislike how all sorts of syscall crud is 'duplicated' in this file, rather than near or in the actual syscall definition.

Now I did copy the futex_cmd_has_timeout() which I could just use the futex one by exporting it. But the rest is due to printing the content.

What exactly is "duplicated"? The code is mostly done to display the data, where as the futex syscall code is about implementing it.

What do you suggest in reuse? Or just move some of theses functions into kernel/futex/syscall.c?

-- Steve