| Message ID | 20220824234211.1625026-1-pshier@google.com (mailing list archive) |
|---|---|
| State | Handled Elsewhere, archived |
| Headers | show |
| Series | [v2] intel_idle: avoid tracing MSR writes while idling | expand |
On Wed, Aug 24, 2022 at 04:42:11PM -0700, Peter Shier wrote: > With commit bf5835bcdb963 ("intel_idle: Disable IBRS during long idle"), > enabling wrmsr trace with CONFIG_LOCKDEP causes "suspicious > rcu_dereference_check() usage" warning because do_trace_write_msr does not > use trace_write_msr_rcuidle. > > Change intel_idle_ibrs to use native_wrmsr to avoid tracing > > Signed-off-by: Peter Shier <pshier@google.com> > --- Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> > drivers/idle/intel_idle.c | 4 ++-- > 1 file changed, 2 insertions(+), 2 deletions(-) > > diff --git a/drivers/idle/intel_idle.c b/drivers/idle/intel_idle.c > index 3e101719689a..df129c73786f 100644 > --- a/drivers/idle/intel_idle.c > +++ b/drivers/idle/intel_idle.c > @@ -187,12 +187,12 @@ static __cpuidle int intel_idle_ibrs(struct cpuidle_device *dev, > int ret; > > if (smt_active) > - wrmsrl(MSR_IA32_SPEC_CTRL, 0); > + native_wrmsr(MSR_IA32_SPEC_CTRL, 0); > > ret = __intel_idle(dev, drv, index); > > if (smt_active) > - wrmsrl(MSR_IA32_SPEC_CTRL, spec_ctrl); > + native_wrmsr(MSR_IA32_SPEC_CTRL, spec_ctrl); > > return ret; > } > --
On Thu, Aug 25, 2022 at 1:42 AM Peter Shier <pshier@google.com> wrote: > > With commit bf5835bcdb963 ("intel_idle: Disable IBRS during long idle"), > enabling wrmsr trace with CONFIG_LOCKDEP causes "suspicious > rcu_dereference_check() usage" warning because do_trace_write_msr does not > use trace_write_msr_rcuidle. > > Change intel_idle_ibrs to use native_wrmsr to avoid tracing > > Sample warning: > ============================ > WARNING: suspicious RCU usage > 6.0.0-dbg-DEV #7 Tainted: G S O > ----------------------------- > arch/x86/include/asm/msr-trace.h:48 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/59/0. > > stack backtrace: > CPU: 59 PID: 0 Comm: swapper/59 Tainted: G S O 6.0.0-dbg-DEV #7 > Call Trace: > dump_stack_lvl > dump_stack > lockdep_rcu_suspicious > trace_write_msr > do_trace_write_msr > intel_idle_ibrs > cpuidle_enter_state > cpuidle_enter > do_idle > cpu_startup_entry > start_secondary > secondary_startup_64_no_verify > > Tested on skylake using: > echo "msr:write_msr" >/sys/kernel/debug/tracing/set_event > with and without patch. > > Signed-off-by: Peter Shier <pshier@google.com> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> > --- > > v2: > - changed to avoid tracing > > v1: https://lore.kernel.org/all/20220823234353.937002-1-pshier@google.com/ > > drivers/idle/intel_idle.c | 4 ++-- > 1 file changed, 2 insertions(+), 2 deletions(-) > > diff --git a/drivers/idle/intel_idle.c b/drivers/idle/intel_idle.c > index 3e101719689a..df129c73786f 100644 > --- a/drivers/idle/intel_idle.c > +++ b/drivers/idle/intel_idle.c > @@ -187,12 +187,12 @@ static __cpuidle int intel_idle_ibrs(struct cpuidle_device *dev, > int ret; > > if (smt_active) > - wrmsrl(MSR_IA32_SPEC_CTRL, 0); > + native_wrmsr(MSR_IA32_SPEC_CTRL, 0); > > ret = __intel_idle(dev, drv, index); > > if (smt_active) > - wrmsrl(MSR_IA32_SPEC_CTRL, spec_ctrl); > + native_wrmsr(MSR_IA32_SPEC_CTRL, spec_ctrl); > > return ret; > } > --
diff --git a/drivers/idle/intel_idle.c b/drivers/idle/intel_idle.c index 3e101719689a..df129c73786f 100644 --- a/drivers/idle/intel_idle.c +++ b/drivers/idle/intel_idle.c @@ -187,12 +187,12 @@ static __cpuidle int intel_idle_ibrs(struct cpuidle_device *dev, int ret; if (smt_active) - wrmsrl(MSR_IA32_SPEC_CTRL, 0); + native_wrmsr(MSR_IA32_SPEC_CTRL, 0); ret = __intel_idle(dev, drv, index); if (smt_active) - wrmsrl(MSR_IA32_SPEC_CTRL, spec_ctrl); + native_wrmsr(MSR_IA32_SPEC_CTRL, spec_ctrl); return ret; }
With commit bf5835bcdb963 ("intel_idle: Disable IBRS during long idle"), enabling wrmsr trace with CONFIG_LOCKDEP causes "suspicious rcu_dereference_check() usage" warning because do_trace_write_msr does not use trace_write_msr_rcuidle. Change intel_idle_ibrs to use native_wrmsr to avoid tracing Sample warning: ============================ WARNING: suspicious RCU usage 6.0.0-dbg-DEV #7 Tainted: G S O ----------------------------- arch/x86/include/asm/msr-trace.h:48 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/59/0. stack backtrace: CPU: 59 PID: 0 Comm: swapper/59 Tainted: G S O 6.0.0-dbg-DEV #7 Call Trace: dump_stack_lvl dump_stack lockdep_rcu_suspicious trace_write_msr do_trace_write_msr intel_idle_ibrs cpuidle_enter_state cpuidle_enter do_idle cpu_startup_entry start_secondary secondary_startup_64_no_verify Tested on skylake using: echo "msr:write_msr" >/sys/kernel/debug/tracing/set_event with and without patch. Signed-off-by: Peter Shier <pshier@google.com> --- v2: - changed to avoid tracing v1: https://lore.kernel.org/all/20220823234353.937002-1-pshier@google.com/ drivers/idle/intel_idle.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) --