Re: [PATCH 04/15] ftrace: pack event structures.

From: Steven Rostedt
Date: Tue Mar 08 2011 - 21:01:39 EST


On Tue, 2011-03-08 at 18:30 -0500, Steven Rostedt wrote:
> David (Miller),
>
> I'm going back through this patch set and looking to incorporate it into
> 2.6.39. But I know how touchy sparc is with "packed" data structures. Do
> you see this patch as hurting sparc?
>
> I wonder if we should change this to something like:
>
> #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
> # define do_event_packed __attirbute__((packed))
> #else
> # define do_event_packed
> #endif
>
> and use "do_event_packed" instead?
>

Here's the modified patch, if you think the current one is not OK:

-- Steve

commit c49fd97767b112a9a7a50c184caf5059f798d5a0
Author: David Sharp <dhsharp@xxxxxxxxxx>
Date: Fri Dec 3 16:13:18 2010 -0800

tracing: Pack event structures.

Ftrace event structures have a 12-byte struct trace_entry at the beginning.
If the structure is aligned, this means that if the first field is 64-bits,
there will be 4 bytes of padding. Ironically, due to the 4-byte ringbuffer
header, this will make 64-bit writes unaligned, if the ring buffer position
is currently 64-bit aligned:
4(rb)+12(ftrace)+4(pad) = 20; 20%8 = 4

Adding __attribute__((packed)) to the event structures removes the extra
space from the trace events, and actually improves alignment of trace
events with a first field that is 64-bits.

About 65 tracepoints have a 4-byte pad at offset 12:
# find events -name format | xargs -n1 awk '
$1=="name:" {name=$2}
$1=="format:"{FS="\t"}
$3=="offset:12;" && $4=="size:4;"{okay=1}
$3=="offset:16;" && !okay {print name}' | wc -l
65

With all 'syscalls' and 'timer' events enabled, this results in a 5%
improvement in a simple 512MB read benchmark with warm caches.

Tested:

setup:
# echo 1 >events/syscalls/enable
# echo 1 >events/timer/enable
# echo 0 > tracing_enabled
off:
# for n in $(seq 10) ; do \
time dd if=/dev/hda3 of=/dev/null bs=1K count=512K ; \
done
on:
# for n in $(seq 10) ; do \
echo 1 >tracing_enabled; \
time dd if=/dev/hda3 of=/dev/null bs=1K count=512K ; \
echo 0 >tracing_enabled; \
echo > trace; \
done

real time mean/median/stdev
w/o patch off: 1.1679/1.164/0.0169
w/o patch on : 1.9432/1.936/0.0274
w/ patch off: 1.1715/1.159/0.0431
w/ patch on : 1.8425/1.836/0.0138
"on" delta: -0.1007 --> -5.2%

Google-Bug-Id: 2895627

Cc: David S. Miller <davem@xxxxxxxxxxxxx>
Signed-off-by: David Sharp <dhsharp@xxxxxxxxxx>
LKML-Reference: <1291421609-14665-5-git-send-email-dhsharp@xxxxxxxxxx>

[
Replaced "__attibute__((packed))" with "ftrace_do_packed" which
is a macro that is defined as the packed attribute only if the
arch has efficient unaligned access.
]

Signed-off-by: Steven Rostedt <rostedt@xxxxxxxxxxx>

diff --git a/include/linux/ftrace_event.h b/include/linux/ftrace_event.h
index 47e3997..c920ab5 100644
--- a/include/linux/ftrace_event.h
+++ b/include/linux/ftrace_event.h
@@ -282,4 +282,16 @@ perf_trace_buf_submit(void *raw_data, int size, int rctx, u64 addr,
}
#endif

+/*
+ * On systems that can access unaligned data efficiently we
+ * pack the data structures. This actually gives a boost
+ * in tracing speed. But leave the structs alone for archs
+ * that do not do as well with packed data structures.
+ */
+#ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
+# define ftrace_do_packed __attribute__((packed))
+#else
+# define ftrace_do_packed
+#endif
+
#endif /* _LINUX_FTRACE_EVENT_H */
diff --git a/include/trace/ftrace.h b/include/trace/ftrace.h
index 3e68366..6b8f366 100644
--- a/include/trace/ftrace.h
+++ b/include/trace/ftrace.h
@@ -48,7 +48,8 @@
#define __array(type, item, len) type item[len];

#undef __dynamic_array
-#define __dynamic_array(type, item, len) u32 __data_loc_##item;
+#define __dynamic_array(type, item, len) \
+ u32 __data_loc_##item __attribute__((aligned(4)));

#undef __string
#define __string(item, src) __dynamic_array(char, item, -1)
@@ -62,7 +63,7 @@
struct trace_entry ent; \
tstruct \
char __data[0]; \
- }; \
+ } ftrace_do_packed; \
\
static struct ftrace_event_class event_class_##name;

diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h
index 9cd025a..ce185fe 100644
--- a/kernel/trace/trace.h
+++ b/kernel/trace/trace.h
@@ -60,7 +60,7 @@ enum trace_type {
struct struct_name { \
struct trace_entry ent; \
tstruct \
- }
+ } ftrace_do_packed

#undef TP_ARGS
#define TP_ARGS(args...) args


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/