Skip to contentSkip to conversation

perf test: Make leafloop workload immune to compiler options

PATCH Superseded James Clark 2 participants 2 messages May 8 perf events

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

James Clark

author reviewer
#0 lore raw

PATCH 1/1 perf test: Make leafloop workload immune to compiler options

Since the leafloop test program was moved into the main Perf binary as a workload, it inherited the same compiler options as Perf. In this case the -fstack-protector option broke the assumption that simple leaf frames don't have a stack frame on Arm. This causes test_arm_callgraph_fp.sh to pass even if the stack isn't augmented with the link register, making the test useless.

Fix it by rewriting the leaf function in assembly seeing as it's so simple. Adding -fno-stack-protector would also work, but wouldn't be robust against other future compiler option additions.

The local variables and 'a' variable were never needed so remove them to simplify.

Assisted-by GitHub-Copilot:GPT-5.5
---
 tools/perf/tests/workloads/leafloop.c | 36 +++++++++++++++++++++++++++--------
 1 file changed, 28 insertions(+), 8 deletions(-)
tools/perf/tests/workloads/leafloop.c +28 -8
index f7561767e32c..58367fd1daec 100644
@@ -6,10 +6,9 @@
 #include "../tests.h"
 
 /* We want to check these symbols in perf script */
-noinline void leaf(volatile int b);
-noinline void parent(volatile int b);
+noinline void leaf(void);
+noinline void parent(void);
 
-static volatile int a;
 static volatile sig_atomic_t done;
 
 static void sighandler(int sig __maybe_unused)
@@ -17,15 +16,36 @@ static void sighandler(int sig __maybe_unused)
 	done = 1;
 }
 
-noinline void leaf(volatile int b)
+#if defined(__aarch64__)
+/*
+ * Write leaf() in assembly so it stays as a minimal leaf function with no
+ * stack frame and won't get silently broken in the future by any Perf wide
+ * compilation options like -fstack-protector-all.
+ */
+asm(
+	".text\n"
+	".global leaf\n"
+	".type leaf, %function\n"
+	"leaf:\n"
+	"	adrp	x1, done\n"
+	"	ldr	w2, [x1, #:lo12:done]\n"
+	"	cbz	w2, leaf\n"
+	"	ret\n"
+);
+
+#else
+
+noinline void leaf(void)
 {
 	while (!done)
-		a += b;
+		;
 }
 
-noinline void parent(volatile int b)
+#endif
+
+noinline void parent(void)
 {
-	leaf(b);
+	leaf();
 }
 
 static int leafloop(int argc, const char **argv)
@@ -39,7 +59,7 @@ static int leafloop(int argc, const char **argv)
 	signal(SIGALRM, sighandler);
 	alarm(sec);
 
-	parent(sec);
+	parent();
 	return 0;
 }
 

---
base-commit: 8c8f2093614373ea8179b562320212a25cf937c0
change-id: 20260508-james-perf-leafloop-stack-c221600eddf2

Best regards,

Signature · 1 line
James Clark <[email protected]>
#1 lore raw
Thank you for your contribution! Sashiko AI review found 3 potential issue(s) to consider: - [Low] LTO build failure due to hardcoded static variable name in inline assembly. - [Low] Missing ELF…

Thank you for your contribution! Sashiko AI review found 3 potential issue(s) to consider:
- [Low] LTO build failure due to hardcoded static variable name in inline assembly.
- [Low] Missing ELF .size directive for inline assembly function.
- [Low] Fragile section management in top-level assembly.
--

commit 57b4e20838dd06862703b6049b087e60606a2dd4
Author: James Clark <[email protected]>

perf test: Make leafloop workload immune to compiler options

This commit replaces the C implementation of the leaf function with an inline assembly version for AArch64. This ensures compiler options like -fstack-protector don't inadvertently add a stack frame, which would make the test_arm_callgraph_fp test pass incorrectly.

Quoted 15 lines
tools/perf/tests/workloads/leafloop.c +2 -3
index f7561767e32cd..58367fd1daece 100644
@@ -6,10 +6,9 @@
 #include "../tests.h"
 
 /* We want to check these symbols in perf script */
-noinline void leaf(volatile int b);
-noinline void parent(volatile int b);
+noinline void leaf(void);
+noinline void parent(void);
 
-static volatile int a;
 static volatile sig_atomic_t done;

[ ... ]

tools/perf/tests/workloads/leafloop.c static void sighandler(int sig __maybe_unused) · line 26

Quoted 9 lines
tools/perf/tests/workloads/leafloop.c +8 -1
-noinline void leaf(volatile int b)
+#if defined(__aarch64__)
+/*
+ * Write leaf() in assembly so it stays as a minimal leaf function with no
+ * stack frame and won't get silently broken in the future by any Perf wide
+ * compilation options like -fstack-protector-all.
+ */
+asm(
+	".text\n"

Could using .text here instead of .pushsection .text and .popsection interfere with the compiler's section tracking?

If the compiler assumes the previous section remains active after the asm block, it might place subsequent data or code in the wrong section under certain compilation flags.

tools/perf/tests/workloads/leafloop.c static void sighandler(int sig __maybe_unused) · line 30

Quoted 4 lines
tools/perf/tests/workloads/leafloop.c +4 -0
+	".global leaf\n"
+	".type leaf, %function\n"
+	"leaf:\n"
+	"	adrp	x1, done\n"

Since done is declared as a static variable at the top of the file, if perf is compiled with link-time optimization (LTO=1), the compiler might partition and rename the C variable to avoid collisions across translation units (e.g., done.lto_priv.0).

Will the hardcoded string "done" in this inline assembly cause an undefined reference linker error under LTO builds?

tools/perf/tests/workloads/leafloop.c static void sighandler(int sig __maybe_unused) · line 33

Quoted 3 lines
tools/perf/tests/workloads/leafloop.c +3 -0
+	"	ldr	w2, [x1, #:lo12:done]\n"
+	"	cbz	w2, leaf\n"
+	"	ret\n"

This isn't a bug, but there doesn't appear to be a size directive (.size leaf, . - leaf) at the end of the function.

While perf might handle 0-sized symbols internally, could omitting the size cause issues with external ELF tools, debuggers, or profilers trying to resolve the symbol bounds?

Quoted 1 line

+);

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1