Skip to contentSkip to conversation

selftests/bpf: Return -1 from get_preempt_count() for unsupported architecture

PATCH In review Saket Kumar Bhaskar 3 participants 4 messages Jan 29

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

PATCH 1/1 selftests/bpf: Return -1 from get_preempt_count() for unsupported architecture

Make get_preempt_count() to return -1 for unsupported architectures instead of 0, so callers can distinguish not supported from a valid preempt_count of 0.

---
 tools/testing/selftests/bpf/bpf_experimental.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/testing/selftests/bpf/bpf_experimental.h +1 -1
index 2cd9165c7348..6759d2ae6151 100644
@@ -631,7 +631,7 @@ static inline int get_preempt_count(void)
 #elif defined(bpf_target_arm64)
 	return bpf_get_current_task_btf()->thread_info.preempt.count;
 #endif
-	return 0;
+	return -1;
 }
 
 /* Description
Signature · 1 line
2.52.0
#1 lore raw
Returning -1 here causes incorrect behavior in all the bpf_in_* helper functions that call get_preempt_count(). Since -1 is 0xFFFFFFFF in twos complement, the bitmask operations in the callers…

tools/testing/selftests/bpf/bpf_experimental.h static inline int get_preempt_count(void) · line 632

Quoted 11 lines
tools/testing/selftests/bpf/bpf_experimental.h +1 -1
index a39576c8ba..87706a213a 100644
@@ -628,7 +628,7 @@ static inline int get_preempt_count(void)
 #elif defined(bpf_target_arm64)
 	return bpf_get_current_task_btf()->thread_info.preempt.count;
 #endif
-	return 0;
+	return -1;
 }

Returning -1 here causes incorrect behavior in all the bpf_in_* helper functions that call get_preempt_count(). Since -1 is 0xFFFFFFFF in twos complement, the bitmask operations in the callers produce non-zero results:

bpf_in_nmi(): -1 & NMI_MASK = 0x00F00000 (non-zero)
bpf_in_hardirq(): -1 & HARDIRQ_MASK = 0x000F0000 (non-zero) bpf_in_interrupt(): returns non-zero (falsely indicates interrupt ctx) bpf_in_serving_softirq(): returns non-zero
bpf_in_task(): returns 0 (falsely indicates NOT in task context)

Previously with return 0, these functions would return 0 on unsupported architectures. Now they return false positives indicating the system is in NMI/hardirq/softirq context when it may not be.

Should the callers be updated to check for -1 before performing the bitmask operations? For example:

    static inline int bpf_in_nmi(void)
    {
        int pcnt = get_preempt_count();
        if (pcnt < 0)
            return -1;  /* or 0? */
        return pcnt & NMI_MASK;
    }

---
AI reviewed your patch. Please fix the bug or email reply why it's not a bug.
See: https://github.com/kernel-patches/vmtest/blob/master/ci/claude/README.md

CI run summary: https://github.com/kernel-patches/bpf/actions/runs/21466700415

AI-authorship-score: low
AI-authorship-explanation: Simple one-line change with clear human-authored commit message following standard kernel conventions.
issues-found: 1
issue-severity-score: low
issue-severity-explanation: Incomplete API change causes bpf_in_* functions to return false positives on unsupported architectures, but this only affects selftest code on non-x86/arm64 systems.

reviewer
#3 lore raw
Quoted 7 lines · Yonghong Song
Quoted 4 lines · Saket Kumar Bhaskar

Make get_preempt_count() to return -1 for unsupported architectures instead of 0, so callers can distinguish not supported from a valid preempt_count of 0.

Quoted 5 lines
Quoted 1 line

LGTM.

Dig through further. My above Ack is invalid due the following usage e.g.
static inline int bpf_in_nmi(void)

{
        return get_preempt_count() & NMI_MASK;
}

in bpf_experimental.h. In such cases, returning -1 for get_preempt_count() may cause bpf_in_nmi() to be true, but this is incorrect.