Hi,
Here is the 2nd version of the patches to reject rethook if RCU is
not watching. The 1st version is here;
https://lore.kernel.org/all/165189881197.175864.14757002789194211860.stgit@devnote2/
This is actually related to the idle function tracing issue
reported by Jiri on LKML (*)
(*) https://lore.kernel.org/bpf/[email protected]/
Jiri reported that fprobe (and rethook) based kprobe-multi bpf
trace kicks "suspicious RCU usage" warning. This is because the
RCU operation is used in the kprobe-multi handler. However, I
also found that the similar issue exists in the rethook because
the rethook uses RCU operation.
I added a new patch [1/2] to test this issue by fprobe_example.ko.
(with this patch, it can avoid using printk() which also involves
the RCU operation.)
------
# insmod fprobe_example.ko symbol=arch_cpu_idle use_trace=1 stackdump=0
fprobe_init: Planted fprobe at arch_cpu_idle
# rmmod fprobe_example.ko
=============================
WARNING: suspicious RCU usage
5.18.0-rc5-00019-gcae4ec21e87a-dirty #30 Not tainted
-----------------------------
include/trace/events/lock.h:37 suspicious rcu_dereference_check() usage!
other info that might help us debug this:
rcu_scheduler_active = 2, debug_locks = 1
RCU used illegally from extended quiescent state!
no locks held by swapper/0/0.
stack backtrace:
CPU: 0 PID: 0 Comm: swapper/0 Not tainted 5.18.0-rc5-00019-gcae4ec21e87a-dirty #30
------
After applying [2/2] fix (which avoid initializing rethook on
function entry if !rcu_watching()), this warning was gone.
------
# insmod fprobe_example.ko symbol=arch_cpu_idle use_trace=1 stackdump=0
fprobe_init: Planted fprobe at arch_cpu_idle
# rmmod fprobe_example.ko
fprobe_exit: fprobe at arch_cpu_idle unregistered. 225 times hit, 230 times missed
------
Note that you can test this program until the arch_cpu_idle()
is marked as noinstr. After that, the function can not be
traced.
Thank you,
---
Masami Hiramatsu (Google) (2):
fprobe: samples: Add use_trace option and show hit/missed counter
rethook: Reject getting a rethook if RCU is not watching
kernel/trace/rethook.c | 9 +++++++++
samples/fprobe/fprobe_example.c | 21 +++++++++++++++++----
2 files changed, 26 insertions(+), 4 deletions(-)
--
Signature
On Wed, Jun 08, 2022 at 01:10:52AM +0900, Masami Hiramatsu (Google) wrote:
> Hi,
>
> Here is the 2nd version of the patches to reject rethook if RCU is
> not watching. The 1st version is here;
>
> https://lore.kernel.org/all/165189881197.175864.14757002789194211860.stgit@devnote2/
>
> This is actually related to the idle function tracing issue
> reported by Jiri on LKML (*)
>
> (*) https://lore.kernel.org/bpf/[email protected]/
>
> Jiri reported that fprobe (and rethook) based kprobe-multi bpf
> trace kicks "suspicious RCU usage" warning. This is because the
> RCU operation is used in the kprobe-multi handler. However, I
> also found that the similar issue exists in the rethook because
> the rethook uses RCU operation.
>
> I added a new patch [1/2] to test this issue by fprobe_example.ko.
> (with this patch, it can avoid using printk() which also involves
> the RCU operation.)
>
> ------
> # insmod fprobe_example.ko symbol=arch_cpu_idle use_trace=1 stackdump=0
> fprobe_init: Planted fprobe at arch_cpu_idle
> # rmmod fprobe_example.ko
>
> =============================
> WARNING: suspicious RCU usage
> 5.18.0-rc5-00019-gcae4ec21e87a-dirty #30 Not tainted
> -----------------------------
> include/trace/events/lock.h:37 suspicious rcu_dereference_check() usage!
>
> other info that might help us debug this:
>
> rcu_scheduler_active = 2, debug_locks = 1
>
>
> RCU used illegally from extended quiescent state!
> no locks held by swapper/0/0.
>
> stack backtrace:
> CPU: 0 PID: 0 Comm: swapper/0 Not tainted 5.18.0-rc5-00019-gcae4ec21e87a-dirty #30
> ------
>
> After applying [2/2] fix (which avoid initializing rethook on
> function entry if !rcu_watching()), this warning was gone.
>
> ------
> # insmod fprobe_example.ko symbol=arch_cpu_idle use_trace=1 stackdump=0
> fprobe_init: Planted fprobe at arch_cpu_idle
> # rmmod fprobe_example.ko
> fprobe_exit: fprobe at arch_cpu_idle unregistered. 225 times hit, 230 times missed
> ------
>
> Note that you can test this program until the arch_cpu_idle()
> is marked as noinstr. After that, the function can not be
> traced.
>
> Thank you,
>
> ---
>
> Masami Hiramatsu (Google) (2):
> fprobe: samples: Add use_trace option and show hit/missed counter
> rethook: Reject getting a rethook if RCU is not watching
LGTM
Acked-by: Jiri Olsa <[email protected]>
jirka
>
>
> kernel/trace/rethook.c | 9 +++++++++
> samples/fprobe/fprobe_example.c | 21 +++++++++++++++++----
> 2 files changed, 26 insertions(+), 4 deletions(-)
>
> --
> Signature