On platforms with an arch timer erratum workaround, it's possible for
arch_timer_reg_read_stable() to recurse into itself when certain
tracing options are enabled, leading to stack overflows and related
problems.
For example, when PREEMPT_TRACER and FUNCTION_GRAPH_TRACER are
selected, it's possible to trigger this with:
$ mount -t debugfs nodev /sys/kernel/debug/
$ echo function_graph > /sys/kernel/debug/tracing/current_tracer
The problem is that in such cases, preempt_disable() instrumentation
attempts to acquire a timestamp via trace_clock(), resulting in a call
back to arch_timer_reg_read_stable(), and hence recursion.
This patch changes arch_timer_reg_read_stable() to use
preempt_{disable,enable}_notrace(), which avoids this.
This problem is similar to the fixed by upstream commit
96b3d28bf4
("sched/clock: Prevent tracing recursion in sched_clock_cpu()").
Fixes: 6acc71ccac71 ("arm64: arch_timer: Allows a CPU-specific erratum to only affect a subset of CPUs")
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
u64 _val; \
if (needs_unstable_timer_counter_workaround()) { \
const struct arch_timer_erratum_workaround *wa; \
- preempt_disable(); \
+ preempt_disable_notrace(); \
wa = __this_cpu_read(timer_unstable_counter_workaround); \
if (wa && wa->read_##reg) \
_val = wa->read_##reg(); \
else \
_val = read_sysreg(reg); \
- preempt_enable(); \
+ preempt_enable_notrace(); \
} else { \
_val = read_sysreg(reg); \
} \