syzbot
KCSAN: data-race in __migrate_disable / tracing_gen_ctx_irq_test
Status:
moderation: reported on 2026/11/19 00:01
Subsystems:
trace
Labels:
race:benign
prio:low
[Documentation on labels]
Reported-by: syzbot+a67ef08256a54e74811a@syzkaller.appspotmail.com
First crash: 2d13h, last: 2d02h
✨ AI-generated root cause summary:
A data race occurs on the task_struct::migration_disabled counter between task context in __migrate_disable()/__migrate_enable() and interrupt context in tracing_gen_ctx_irq_test(). When task context increments or decrements migration_disabled, preemption and interrupts are not disabled on the fast paths. An asynchronous hardware or timer interrupt firing on the same CPU can trigger a tracepoint or BPF trace print event, invoking migration_disable_value() to read current->migration_disabled using plain unmarked memory accesses rather than READ_ONCE()/data_race().
▶
▼
✨ Full AI root cause analysis
Root Cause Analysis
1. Involved Entities
2. Technical Mechanism and Triggering Flow
-
Task Context (__migrate_disable / __migrate_enable):
-
Interrupt Context (tracing_gen_ctx_irq_test):
-
Data Race Detected by KCSAN:
- The memory access in
migration_disable_value() is a plain, unmarked C read (current->migration_disabled) rather than READ_ONCE() or data_race().
- The writes in
__migrate_disable() and __migrate_enable() are plain, unmarked C modifications.
- Because hardware interrupts can interleave asynchronously with task execution on the same CPU, KCSAN detects the concurrent unmarked read and write to
&task_struct::migration_disabled as a data race.
Sample crash report:
* Struck through repros no longer work on HEAD.