Skip to contentSkip to conversation

selftests/bpf: Fix insn count expectations in chained global subprogs test

PATCH In review Jiayuan Chen 3 participants 3 messages Aug 13

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 selftests/bpf: Fix insn count expectations in chained global subprogs test

The test asserts exact per-subprog insns_self/insns_total values and the aggregate processed count, but these depend on compiler codegen. The expected values match -mcpu=v3 output; under -mcpu=v4 clang emits a sign-extending load in global_good() instead of a load plus two shifts, so the subprog is verified in 3 insns instead of 5 and the test fails on the cpuv4 flavor.

Rewrite the involved subprogs as naked functions with hand-written instruction sequences using only v1 instructions, so the verified insn counts no longer depend on cpu version or compiler codegen.

Fixes c2e6c7de8830 ("bpf: Show more useful info in stack depth stats")
---
 .../bpf/progs/verifier_global_subprogs.c      | 56 +++++++++++++++----
 1 file changed, 45 insertions(+), 11 deletions(-)
tools/testing/selftests/bpf/progs/verifier_global_subprogs.c +45 -11
index 7b65eea97ebc..3352dc177ec1 100644
@@ -25,9 +25,22 @@ __noinline long global_bad(void)
 	return arr[unkn_idx]; /* BOOM */
 }
 
-__noinline long global_good(void)
+/* Insn counts asserted in chained_global_func_calls_success() depend on
+ * the exact instruction sequences of this function and its callers, so
+ * they are written in inline asm to be independent of compiler codegen.
+ */
+__naked __noinline long global_good(void)
 {
-	return arr[0];
+	/* return (long)arr[0]; 5 insns (ld_imm64 counts as one) */
+	asm volatile (
+		"r1 = %[arr] ll;"
+		"r0 = *(u32 *)(r1 + 0);"
+		"r0 <<= 32;"
+		"r0 s>>= 32;"
+		"exit;"
+		:
+		: __imm_addr(arr)
+		: __clobber_all);
 }
 
 __noinline long global_calls_bad(void)
@@ -35,9 +48,13 @@ __noinline long global_calls_bad(void)
 	return global_good() + global_bad() /* does BOOM indirectly */;
 }
 
-__noinline long global_calls_good_only(void)
+__naked __noinline long global_calls_good_only(void)
 {
-	return global_good();
+	/* return global_good(); 2 insns */
+	asm volatile (
+		"call global_good;"
+		"exit;"
+		::: __clobber_all);
 }
 
 __noinline long global_dead(void)
@@ -56,13 +73,30 @@ __msg("subprog 0 (chained_global_func_calls_success) main insns_self 7 insns_tot
 __msg("subprog {{[0-9]+}} (global_calls_good_only) global insns_self 2 insns_total 2 stack")
 __msg("subprog {{[0-9]+}} (global_good) global insns_self 5 insns_total 5 stack")
 __msg("processed 14 insns")
-int chained_global_func_calls_success(void)
-{
-	int sum = 0;
-
-	if (call_dead_subprog)
-		sum += global_dead();
-	return global_calls_good_only() + sum;
+__naked int chained_global_func_calls_success(void)
+{
+	/* int sum = 0;
+	 * if (call_dead_subprog)
+	 *	sum += global_dead();
+	 * return global_calls_good_only() + sum;
+	 *
+	 * call_dead_subprog is known false, so 7 insns are processed:
+	 * the branch to l0 is always taken and global_dead() is dead code.
+	 */
+	asm volatile (
+		"r6 = 0;"
+		"r1 = %[call_dead_subprog] ll;"
+		"r1 = *(u8 *)(r1 + 0);"
+		"if r1 == 0 goto l0_%=;"
+		"call global_dead;"
+		"r6 = r0;"
+	"l0_%=:"
+		"call global_calls_good_only;"
+		"r0 += r6;"
+		"exit;"
+		:
+		: __imm_addr(call_dead_subprog)
+		: __clobber_all);
 }
 
 SEC("?raw_tp")
Signature · 1 line
2.43.0
#1 lore raw
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [Low] The newly introduced multi-line comments violate the mandatory BPF kernel comment style by placing…

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Low] The newly introduced multi-line comments violate the mandatory BPF kernel comment style by placing text on the same line as the opening /*. --

--- Patch [1]: [PATCH net-next v1] selftests/bpf: Fix insn count expectations in chained global subprogs test --- commit 0f391b5a2f4f9b30c1e20dfe376da213cfa9c650
Author: Jiayuan Chen <[email protected]>

selftests/bpf: Fix insn count expectations in chained global subprogs test

This commit rewrites the involved subprograms as naked functions with hand-written instruction sequences using only v1 instructions. This ensures that the verified instruction counts no longer depend on CPU version or compiler codegen, resolving failures on the cpuv4 flavor.

Fixes c2e6c7de8830 ("bpf: Show more useful info in stack depth stats")

tools/testing/selftests/bpf/progs/verifier_global_subprogs.c __noinline long global_bad(void) · line 31

Quoted 13 lines
tools/testing/selftests/bpf/progs/verifier_global_subprogs.c +4 -1
index 7b65eea97ebc2..3352dc177ec15 100644
@@ -25,9 +25,22 @@ __noinline long global_bad(void)
 	return arr[unkn_idx]; /* BOOM */
 }
 
-__noinline long global_good(void)
+/* Insn counts asserted in chained_global_func_calls_success() depend on
+ * the exact instruction sequences of this function and its callers, so
+ * they are written in inline asm to be independent of compiler codegen.
+ */

[Severity: Low]
This isn't a bug, but does this violate the preferred BPF multi-line comment style? The BPF subsystem guidelines specify that multi-line comments should have the opening /* on its own line.

Quoted 1 line
tools/testing/selftests/bpf/progs/verifier_global_subprogs.c +1 -0
+__naked __noinline long global_good(void)

[ ... ]

Quoted 11 lines
tools/testing/selftests/bpf/progs/verifier_global_subprogs.c +0 -7
@@ -56,13 +73,30 @@ __msg("subprog 0 (chained_global_func_calls_success) main insns_self 7 insns_tot
 __msg("subprog {{[0-9]+}} (global_calls_good_only) global insns_self 2 insns_total 2 stack")
 __msg("subprog {{[0-9]+}} (global_good) global insns_self 5 insns_total 5 stack")
 __msg("processed 14 insns")
-int chained_global_func_calls_success(void)
-{
-	int sum = 0;
-
-	if (call_dead_subprog)
-		sum += global_dead();
-	return global_calls_good_only() + sum;

tools/testing/selftests/bpf/progs/verifier_global_subprogs.c __msg("subprog 0 (chained_global_func_calls_success) main insns_self 7 insns_tot · line 85

Quoted 10 lines
tools/testing/selftests/bpf/progs/verifier_global_subprogs.c +10 -0
+__naked int chained_global_func_calls_success(void)
+{
+	/* int sum = 0;
+	 * if (call_dead_subprog)
+	 *	sum += global_dead();
+	 * return global_calls_good_only() + sum;
+	 *
+	 * call_dead_subprog is known false, so 7 insns are processed:
+	 * the branch to l0 is always taken and global_dead() is dead code.
+	 */

[Severity: Low]
This isn't a bug, but it looks like this comment block also places text on the same line as the opening /*. Could this be updated to place the /* on its own line to match the rest of the BPF subsystem?

Quoted 1 line
tools/testing/selftests/bpf/progs/verifier_global_subprogs.c +1 -0
+	asm volatile (

[ ... ]

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
maintainer
#2 lore raw
In reply to Jiayuan Chen
Quoted 7 lines · Jiayuan Chen

The test asserts exact per-subprog insns_self/insns_total values and the aggregate processed count, but these depend on compiler codegen. The expected values match -mcpu=v3 output; under -mcpu=v4 clang emits a sign-extending load in global_good() instead of a load plus two shifts, so the subprog is verified in 3 insns instead of 5 and the test fails on the cpuv4 flavor.

Quoted 6 lines

Rewrite the involved subprogs as naked functions with hand-written instruction sequences using only v1 instructions, so the verified insn counts no longer depend on cpu version or compiler codegen.

Fixes c2e6c7de8830 ("bpf: Show more useful info in stack depth stats")

$subj should have been bpf-next to route to BPF CI. Anyway, it was fixed in the tree via:

9786d424a36b ("selftests/bpf: Fix chained_global_func_calls_success() for cpu v4for-next")

Thanks,
Daniel