Skip to contentSkip to conversation

Need to remove char pointers from trace events

PATCH Steven Rostedt 3 participants 4 messages Jul 10, 2019

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

I was doing a bit of an audit on trace events and found this:

# cat /debug/tracing/events/i915/intel_disable_plane/format
name: intel_disable_plane
ID: 1358
format:
	field:unsigned short common_type;	offset:0;	size:2;	signed:0;
	field:unsigned char common_flags;	offset:2;	size:1;	signed:0;
	field:unsigned char common_preempt_count;	offset:3;	size:1;	signed:0;
	field:int common_pid;	offset:4;	size:4;	signed:1;
	field:enum pipe pipe;	offset:8;	size:4;	signed:1;
	field:const char * name;	offset:16;	size:8;	signed:0;
	field:u32 frame;	offset:24;	size:4;	signed:0;
	field:u32 scanline;	offset:28;	size:4;	signed:0;

print fmt: "pipe %c, plane %s, frame=%u, scanline=%u", ((REC->pipe) + 'A'), REC->name, REC->frame, REC->scanline

Same goes for intel_update_plane.

The problem here is:

	field:const char * name;	offset:16;	size:8;	signed:0;

print fmt: "pipe %c, plane %s, frame=%u, scanline=%u", ((REC->pipe) + 'A'), REC->name, REC->frame, REC->scanline

Where the TRACE_EVENT() macro has:

	    TP_fast_assign(
			   __entry->pipe = crtc->pipe;
			   __entry->name = plane->name;
			   __entry->frame = crtc->base.dev->driver->get_vblank_counter(crtc->base.dev,
										       crtc->pipe);
			   __entry->scanline = intel_get_crtc_scanline(crtc);
			   ),

	    TP_printk("pipe %c, plane %s, frame=%u, scanline=%u",
		      pipe_name(__entry->pipe), __entry->name,
		      __entry->frame, __entry->scanline)

The issue here is that you record a pointer address to "plane->name" and then sometime in the distant future access that same address. There's usually no guarantee that the contents at that address will exist when the buffer is read.

The proper way to record strings, is to record the string into the ring buffer itself, and not rely on it existing hours or days later.

I recommend the following patch:

-- Steve

drivers/gpu/drm/i915/i915_trace.h +6 -6
index 12893304c8f8..d41d914a16ca 100644
@@ -298,16 +298,16 @@ TRACE_EVENT(intel_update_plane,
 
 	    TP_STRUCT__entry(
 			     __field(enum pipe, pipe)
-			     __field(const char *, name)
 			     __field(u32, frame)
 			     __field(u32, scanline)
 			     __array(int, src, 4)
 			     __array(int, dst, 4)
+			     __string(name, plane->name)
 			     ),
 
 	    TP_fast_assign(
+			   __assign_str(name, plane->name);
 			   __entry->pipe = crtc->pipe;
-			   __entry->name = plane->name;
 			   __entry->frame = crtc->base.dev->driver->get_vblank_counter(crtc->base.dev,
 										       crtc->pipe);
 			   __entry->scanline = intel_get_crtc_scanline(crtc);
@@ -316,7 +316,7 @@ TRACE_EVENT(intel_update_plane,
 			   ),
 
 	    TP_printk("pipe %c, plane %s, frame=%u, scanline=%u, " DRM_RECT_FP_FMT " -> " DRM_RECT_FMT,
-		      pipe_name(__entry->pipe), __entry->name,
+		      pipe_name(__entry->pipe), __get_str(name),
 		      __entry->frame, __entry->scanline,
 		      DRM_RECT_FP_ARG((const struct drm_rect *)__entry->src),
 		      DRM_RECT_ARG((const struct drm_rect *)__entry->dst))
@@ -328,21 +328,21 @@ TRACE_EVENT(intel_disable_plane,
 
 	    TP_STRUCT__entry(
 			     __field(enum pipe, pipe)
-			     __field(const char *, name)
 			     __field(u32, frame)
 			     __field(u32, scanline)
+			     __string(name, plane->name)
 			     ),
 
 	    TP_fast_assign(
+			   __assign_str(name, plane->name);
 			   __entry->pipe = crtc->pipe;
-			   __entry->name = plane->name;
 			   __entry->frame = crtc->base.dev->driver->get_vblank_counter(crtc->base.dev,
 										       crtc->pipe);
 			   __entry->scanline = intel_get_crtc_scanline(crtc);
 			   ),
 
 	    TP_printk("pipe %c, plane %s, frame=%u, scanline=%u",
-		      pipe_name(__entry->pipe), __entry->name,
+		      pipe_name(__entry->pipe), __get_str(name),
 		      __entry->frame, __entry->scanline)
 );
 
#1 lore raw
Quoted 44 lines · Steven Rostedt

I was doing a bit of an audit on trace events and found this:

# cat /debug/tracing/events/i915/intel_disable_plane/format
name: intel_disable_plane
ID: 1358
format:
	field:unsigned short common_type;	offset:0;	size:2;	signed:0;
	field:unsigned char common_flags;	offset:2;	size:1;	signed:0;
	field:unsigned char common_preempt_count;	offset:3;	size:1;	signed:0;
	field:int common_pid;	offset:4;	size:4;	signed:1;
	field:enum pipe pipe;	offset:8;	size:4;	signed:1;
	field:const char * name;	offset:16;	size:8;	signed:0;
	field:u32 frame;	offset:24;	size:4;	signed:0;
	field:u32 scanline;	offset:28;	size:4;	signed:0;

print fmt: "pipe %c, plane %s, frame=%u, scanline=%u", ((REC->pipe) + 'A'), REC->name, REC->frame, REC->scanline

Same goes for intel_update_plane.

The problem here is:

	field:const char * name;	offset:16;	size:8;	signed:0;

print fmt: "pipe %c, plane %s, frame=%u, scanline=%u", ((REC->pipe) + 'A'), REC->name, REC->frame, REC->scanline

Where the TRACE_EVENT() macro has:

	    TP_fast_assign(
			   __entry->pipe = crtc->pipe;
			   __entry->name = plane->name;
			   __entry->frame = crtc->base.dev->driver->get_vblank_counter(crtc->base.dev,
										       crtc->pipe);
			   __entry->scanline = intel_get_crtc_scanline(crtc);
			   ),

	    TP_printk("pipe %c, plane %s, frame=%u, scanline=%u",
		      pipe_name(__entry->pipe), __entry->name,
		      __entry->frame, __entry->scanline)
Quoted 4 lines

The issue here is that you record a pointer address to "plane->name" and then sometime in the distant future access that same address. There's usually no guarantee that the contents at that address will exist when the buffer is read.

The only way those can disappear is if the device goes away. But I have no problem going with your patch. Want to provide a proper commit message for it?

Quoted 65 lines

The proper way to record strings, is to record the string into the ring buffer itself, and not rely on it existing hours or days later.

I recommend the following patch:

-- Steve

drivers/gpu/drm/i915/i915_trace.h +6 -6
index 12893304c8f8..d41d914a16ca 100644
@@ -298,16 +298,16 @@ TRACE_EVENT(intel_update_plane,
 
 	    TP_STRUCT__entry(
 			     __field(enum pipe, pipe)
-			     __field(const char *, name)
 			     __field(u32, frame)
 			     __field(u32, scanline)
 			     __array(int, src, 4)
 			     __array(int, dst, 4)
+			     __string(name, plane->name)
 			     ),
 
 	    TP_fast_assign(
+			   __assign_str(name, plane->name);
 			   __entry->pipe = crtc->pipe;
-			   __entry->name = plane->name;
 			   __entry->frame = crtc->base.dev->driver->get_vblank_counter(crtc->base.dev,
 										       crtc->pipe);
 			   __entry->scanline = intel_get_crtc_scanline(crtc);
@@ -316,7 +316,7 @@ TRACE_EVENT(intel_update_plane,
 			   ),
 
 	    TP_printk("pipe %c, plane %s, frame=%u, scanline=%u, " DRM_RECT_FP_FMT " -> " DRM_RECT_FMT,
-		      pipe_name(__entry->pipe), __entry->name,
+		      pipe_name(__entry->pipe), __get_str(name),
 		      __entry->frame, __entry->scanline,
 		      DRM_RECT_FP_ARG((const struct drm_rect *)__entry->src),
 		      DRM_RECT_ARG((const struct drm_rect *)__entry->dst))
@@ -328,21 +328,21 @@ TRACE_EVENT(intel_disable_plane,
 
 	    TP_STRUCT__entry(
 			     __field(enum pipe, pipe)
-			     __field(const char *, name)
 			     __field(u32, frame)
 			     __field(u32, scanline)
+			     __string(name, plane->name)
 			     ),
 
 	    TP_fast_assign(
+			   __assign_str(name, plane->name);
 			   __entry->pipe = crtc->pipe;
-			   __entry->name = plane->name;
 			   __entry->frame = crtc->base.dev->driver->get_vblank_counter(crtc->base.dev,
 										       crtc->pipe);
 			   __entry->scanline = intel_get_crtc_scanline(crtc);
 			   ),
 
 	    TP_printk("pipe %c, plane %s, frame=%u, scanline=%u",
-		      pipe_name(__entry->pipe), __entry->name,
+		      pipe_name(__entry->pipe), __get_str(name),
 		      __entry->frame, __entry->scanline)
 );
 
Signature · 2 lines
Ville Syrjälä
Intel
#2 lore raw
Quoted 5 lines · On Wed, 10 Jul 2019 18:45:24 +0300 Ville Syrjälä…
Quoted 5 lines
	    TP_printk("pipe %c, plane %s, frame=%u, scanline=%u",
		      pipe_name(__entry->pipe), __entry->name,
		      __entry->frame, __entry->scanline)
Quoted 8 lines
Quoted 4 lines

The issue here is that you record a pointer address to "plane->name" and then sometime in the distant future access that same address. There's usually no guarantee that the contents at that address will exist when the buffer is read.

The only way those can disappear is if the device goes away. But I have no problem going with your patch. Want to provide a proper commit message for it?

Sure, but does that mean the trace data will go away with the device? If not, then you still have the issue.

Also note that perf and trace-cmd will not know how to read that data either, so adding it to the ring buffer gives them access.

I'll send a patch next, thanks!

-- Steve

Quoted 5 lines
Quoted 5 lines

The proper way to record strings, is to record the string into the ring buffer itself, and not rely on it existing hours or days later.

#3 lore raw

✗ Fi.CI.BAT: failure for Need to remove char pointers from trace events

In reply to Steven Rostedt
== Series Details ==

== Series Details ==

Series: Need to remove char pointers from trace events
URL : https://patchwork.freedesktop.org/series/63513/
State : failure

== Summary ==

Applying: Need to remove char pointers from trace events Using index info to reconstruct a base tree...
M drivers/gpu/drm/i915/i915_trace.h
Falling back to patching base and 3-way merge... Auto-merging drivers/gpu/drm/i915/i915_trace.h
CONFLICT (content): Merge conflict in drivers/gpu/drm/i915/i915_trace.h
error: Failed to merge in the changes.
hint: Use 'git am --show-current-patch' to see the failed patch Patch failed at 0001 Need to remove char pointers from trace events When you have resolved this problem, run "git am --continue". If you prefer to skip this patch, run "git am --skip" instead. To restore the original branch and stop patching, run "git am --abort".

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx