Skip to contentSkip to conversation

tracing: silence GCC 9 array bounds warning

PATCH In review Miguel Ojeda 2 participants 4 messages May 22, 2019 Tracing

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 tracing: silence GCC 9 array bounds warning

Starting with GCC 9, -Warray-bounds detects cases when memset is called starting on a member of a struct but the size to be cleared ends up writing over further members.

Such a call happens in the trace code to clear, at once, all members after and including seq on struct trace_iterator:

In function 'memset',
inlined from 'ftrace_dump' at kernel/trace/trace.c:8914:3:
./include/linux/string.h:344:9: warning: '__builtin_memset' offset [8505, 8560] from the object at 'iter' is out of the bounds of referenced subobject 'seq' with type 'struct trace_seq' at offset 4368 [-Warray-bounds]
344 | return __builtin_memset(p, c, size);
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~

In order to avoid GCC complaining about it, we compute the address ourselves by adding the offsetof distance instead of referring directly to the member.

Since there are two places doing this clear (trace.c and trace_kdb.c), take the chance to move the workaround into a single place in the internal header.

---
Steven, let me know if you still prefer (char *) or anything else (or just change it in your side if you want :-)

 kernel/trace/trace.c     |  6 +-----
 kernel/trace/trace.h     | 14 ++++++++++++++
 kernel/trace/trace_kdb.c |  6 +-----
 3 files changed, 16 insertions(+), 10 deletions(-)
kernel/trace/trace.c +1 -5
index 2c92b3d9ea30..1c80521fd436 100644
@@ -8910,12 +8910,8 @@ void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
 
 		cnt++;
 
-		/* reset all but tr, trace, and overruns */
-		memset(&iter.seq, 0,
-		       sizeof(struct trace_iterator) -
-		       offsetof(struct trace_iterator, seq));
+		trace_iterator_reset(&iter);
 		iter.iter_flags |= TRACE_FILE_LAT_FMT;
-		iter.pos = -1;
 
 		if (trace_find_next_entry_inc(&iter) != NULL) {
 			int ret;
kernel/trace/trace.h +14 -0
index 1974ce818ddb..f9a12003f137 100644
@@ -1967,4 +1967,18 @@ static inline void tracer_hardirqs_off(unsigned long a0, unsigned long a1) { }
 
 extern struct trace_iterator *tracepoint_print_iter;
 
+/* reset all but tr, trace, and overruns */
+static __always_inline void trace_iterator_reset(struct trace_iterator *iter)
+{
+	/*
+	 * We do not simplify the start address to &iter->seq in order to let
+	 * GCC 9 know that we really want to overwrite more members than
+	 * just iter->seq (-Warray-bounds).
+	 */
+	const size_t offset = offsetof(struct trace_iterator, seq);
+	memset((char *)(iter) + offset, 0, sizeof(struct trace_iterator) - offset);
+
+	iter->pos = -1;
+}
+
 #endif /* _LINUX_KERNEL_TRACE_H */
kernel/trace/trace_kdb.c +1 -5
index 6c1ae6b752d1..cca65044c14c 100644
@@ -37,12 +37,8 @@ static void ftrace_dump_buf(int skip_entries, long cpu_file)
 	if (skip_entries)
 		kdb_printf("(skipping %d entries)\n", skip_entries);
 
-	/* reset all but tr, trace, and overruns */
-	memset(&iter.seq, 0,
-		   sizeof(struct trace_iterator) -
-		   offsetof(struct trace_iterator, seq));
+	trace_iterator_reset(&iter);
 	iter.iter_flags |= TRACE_FILE_LAT_FMT;
-	iter.pos = -1;
 
 	if (cpu_file == RING_BUFFER_ALL_CPUS) {
 		for_each_tracing_cpu(cpu) {
Signature · 1 line
2.17.1
maintainer
#1 lore raw

kernel/trace/trace.h static inline void tracer_hardirqs_off(unsigned long a0, unsigned long a1) { } · line 1976

Quoted 7 lines · On Wed, 22 May 2019 11:58:10 +0200 Miguel Ojeda…
kernel/trace/trace.h +7 -0
+/* reset all but tr, trace, and overruns */
+static __always_inline void trace_iterator_reset(struct trace_iterator *iter)
+{
+	/*
+	 * We do not simplify the start address to &iter->seq in order to let
+	 * GCC 9 know that we really want to overwrite more members than
+	 * just iter->seq (-Warray-bounds).

This comment is fine for the change log, but here it is too specific. Why does one care about GCC 9 when we are at version GCC 21? I care more about why we are clearing the data and less about the way we are doing it.

A comment like:

/*
* Reset the state of the trace_iterator so that it can read
* consumed data. Normally, the trace_iterator is used for
* reading the data when it is not consumed, and must retain
* state.
*/

That is more useful than why we have the offset hack.

kernel/trace/trace.h static inline void tracer_hardirqs_off(unsigned long a0, unsigned long a1) { } · line 1978

Quoted 2 lines
kernel/trace/trace.h +2 -0
+	 */
+	const size_t offset = offsetof(struct trace_iterator, seq);

Need a empty line between these two.

-- Steve

Quoted 6 lines
kernel/trace/trace.h +5 -0
+	memset((char *)(iter) + offset, 0, sizeof(struct trace_iterator) - offset);
+
+	iter->pos = -1;
+}
+
 #endif /* _LINUX_KERNEL_TRACE_H */
#2 lore raw
Quoted 4 lines · Steven Rostedt

On Wed, 22 May 2019 11:58:10 +0200
Miguel Ojeda <[email protected]> wrote:

Quoted 12 lines

kernel/trace/trace.h static inline void tracer_hardirqs_off(unsigned long a0, unsigned long a1) { } · line 1976

Quoted 7 lines
kernel/trace/trace.h +7 -0
+/* reset all but tr, trace, and overruns */
+static __always_inline void trace_iterator_reset(struct trace_iterator *iter)
+{
+     /*
+      * We do not simplify the start address to &iter->seq in order to let
+      * GCC 9 know that we really want to overwrite more members than
+      * just iter->seq (-Warray-bounds).

This comment is fine for the change log, but here it is too specific. Why does one care about GCC 9 when we are at version GCC 21? I care more about why we are clearing the data and less about the way we are doing it.

Since the code is not written the obvious way on purpose, the idea is to document why that is so -- otherwise the reader may wonder (and possibly re-introduce it back). Specifying when the warning started appearing tends to be clarifying, too.

The commit message explains the change itself, but the comment explains why the current code is written like that.

Quoted 10 lines

A comment like:

/*
* Reset the state of the trace_iterator so that it can read
* consumed data. Normally, the trace_iterator is used for
* reading the data when it is not consumed, and must retain
* state.
*/

That is more useful than why we have the offset hack.

That comment would be great in the function's description, in my opinion, and it is a great addition to have nevertheless. I re-used the existing comment for that to keep the change as minimal as possible (and nevertheless I am not qualified to write it since I have not studied the tracing code). In other words, I'm not saying there are no further improvements :-)

Cheers,
Miguel

maintainer
#3 lore raw
Quoted 18 lines · On Wed, 22 May 2019 15:11:10 +0200 Miguel Ojeda…
Quoted 4 lines · Steven Rostedt

On Wed, 22 May 2019 11:58:10 +0200
Miguel Ojeda <[email protected]> wrote:

Quoted 12 lines

kernel/trace/trace.h static inline void tracer_hardirqs_off(unsigned long a0, unsigned long a1) { } · line 1976

Quoted 7 lines
kernel/trace/trace.h +7 -0
+/* reset all but tr, trace, and overruns */
+static __always_inline void trace_iterator_reset(struct trace_iterator *iter)
+{
+     /*
+      * We do not simplify the start address to &iter->seq in order to let
+      * GCC 9 know that we really want to overwrite more members than
+      * just iter->seq (-Warray-bounds).  

This comment is fine for the change log, but here it is too specific. Why does one care about GCC 9 when we are at version GCC 21? I care more about why we are clearing the data and less about the way we are doing it.

Quoted 7 lines

Since the code is not written the obvious way on purpose, the idea is to document why that is so -- otherwise the reader may wonder (and possibly re-introduce it back). Specifying when the warning started appearing tends to be clarifying, too.

The commit message explains the change itself, but the comment explains why the current code is written like that.

Could also be shorten to: "Keep gcc from complaining about overwriting more than just one member in the structure."

Quoted 10 lines
Quoted 10 lines

A comment like:

/*
* Reset the state of the trace_iterator so that it can read
* consumed data. Normally, the trace_iterator is used for
* reading the data when it is not consumed, and must retain
* state.
*/

Quoted 8 lines
Quoted 1 line

That is more useful than why we have the offset hack.

That comment would be great in the function's description, in my opinion, and it is a great addition to have nevertheless. I re-used the existing comment for that to keep the change as minimal as possible (and nevertheless I am not qualified to write it since I have not studied the tracing code). In other words, I'm not saying there are no further improvements :-)

I put it there so you have an idea what the rational was ;-)

-- Steve