[PATCH RESEND] arm64: arch_timer: fix the infinite recursion when enable ftrace and erratum workaround

Discussion:

Ding Tianhong

2017-07-09 08:30:54 UTC

When enable preempt and debug ftrace, and perform the following steps, the
system will hang:
mount -t debugfs nodev /sys/kernel/debug/
cd /sys/kernel/debug/tracing/
echo function_graph > current_tracer

This is because tracing the preempt_disable/enable calls would cause
trace_clock() which would get local timer to go into infinite recursion
when enable the arch timer erratum workaround for some chips, so Prevent
tracing of preempt_disable/enable() in arch_timer_reg_read_stable().

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 <***@huawei.com>
---
arch/arm64/include/asm/arch_timer.h | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/arch/arm64/include/asm/arch_timer.h b/arch/arm64/include/asm/arch_timer.h
index 74d08e4..67bb7a4 100644
--- a/arch/arm64/include/asm/arch_timer.h
+++ b/arch/arm64/include/asm/arch_timer.h
@@ -65,13 +65,13 @@ struct arch_timer_erratum_workaround {
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); \
} \

--
1.9.0

Hanjun Guo

2017-07-10 06:12:29 UTC

Permalink

Post by Ding Tianhong
When enable preempt and debug ftrace, and perform the following steps, the
mount -t debugfs nodev /sys/kernel/debug/
cd /sys/kernel/debug/tracing/
echo function_graph > current_tracer
This is because tracing the preempt_disable/enable calls would cause
trace_clock() which would get local timer to go into infinite recursion
when enable the arch timer erratum workaround for some chips, so Prevent
tracing of preempt_disable/enable() in arch_timer_reg_read_stable().
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")
---
arch/arm64/include/asm/arch_timer.h | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/arch/arm64/include/asm/arch_timer.h b/arch/arm64/include/asm/arch_timer.h
index 74d08e4..67bb7a4 100644
--- a/arch/arm64/include/asm/arch_timer.h
+++ b/arch/arm64/include/asm/arch_timer.h
@@ -65,13 +65,13 @@ struct arch_timer_erratum_workaround {
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); \
} \

This fixes my system hang issue when I using function tracer with
enabled timer errata on D03,

Tested-by: Hanjun Guo <***@linaro.org>

Thanks
Hanjun

Mark Rutland

2017-07-10 11:22:29 UTC

Permalink

As I mentioned before, the patch itself looks fine to me, but the commit
message is somewhat difficult to read.

Can we please change this to:

arm64: arch_timer: avoid infinite recursion when ftrace is enabled

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.

With that commit message:

Acked-by: Mark Rutland <***@arm.com>

Daniel, Thomas, would you be happy to fold that in when picking this? Or
would you prefer that I fix this up and resend?

Thanks,
Mark.

Post by Ding Tianhong
Fixes: 6acc71ccac71 ("arm64: arch_timer: Allows a CPU-specific erratum to only affect a subset of CPUs")
---
arch/arm64/include/asm/arch_timer.h | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/arch/arm64/include/asm/arch_timer.h b/arch/arm64/include/asm/arch_timer.h
index 74d08e4..67bb7a4 100644
--- a/arch/arm64/include/asm/arch_timer.h
+++ b/arch/arm64/include/asm/arch_timer.h
@@ -65,13 +65,13 @@ struct arch_timer_erratum_workaround {
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); \
} \
--
1.9.0

Marc Zyngier

2017-07-10 13:54:27 UTC

Permalink

Post by Mark Rutland

As I mentioned before, the patch itself looks fine to me, but the commit
message is somewhat difficult to read.
arm64: arch_timer: avoid infinite recursion when ftrace is enabled
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
$ 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.
Daniel, Thomas, would you be happy to fold that in when picking this? Or
would you prefer that I fix this up and resend?

With the proposed changes:

Acked-by: Marc Zyngier <***@arm.com>

M.

--
Jazz is not dead. It just smells funny...

Ding Tianhong

2017-07-26 02:42:49 UTC

Permalink

Post by Mark Rutland

Hi Danial, Thomas:

It looks didn't merge to the mainline tree yet, should I update the commit and
resend this patch again?

Thanks
Ding

Post by Mark Rutland
Thanks,
Mark.

_______________________________________________
linux-arm-kernel mailing list
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
.