Re: [PATCH 3/9] intel_idle: Relocate definitions of cpuidle callbacks

[Qian Cai <cai@lca.pw>]

On Thu, Feb 13, 2020 at 11:00:26PM +0100, Rafael J. Wysocki wrote:
> From: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com>
> 
> Move the definitions of intel_idle() and intel_idle_s2idle() before
> the definitions of cpuidle_state structures referring to them to
> avoid having to use additional declarations of them (and drop those
> declarations).
> 
> No functional impact.
> 
> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
> ---
>  drivers/idle/intel_idle.c | 154 ++++++++++++++++++++++------------------------
>  1 file changed, 75 insertions(+), 79 deletions(-)
> 
> diff --git a/drivers/idle/intel_idle.c b/drivers/idle/intel_idle.c
> index 5adc058c705d..e0332d567735 100644
> --- a/drivers/idle/intel_idle.c
> +++ b/drivers/idle/intel_idle.c
[]
> +/**
> + * intel_idle - Ask the processor to enter the given idle state.
> + * @dev: cpuidle device of the target CPU.
> + * @drv: cpuidle driver (assumed to point to intel_idle_driver).
> + * @index: Target idle state index.
> + *
> + * Use the MWAIT instruction to notify the processor that the CPU represented by
> + * @dev is idle and it can try to enter the idle state corresponding to @index.
> + *
> + * If the local APIC timer is not known to be reliable in the target idle state,
> + * enable one-shot tick broadcasting for the target CPU before executing MWAIT.
> + *
> + * Optionally call leave_mm() for the target CPU upfront to avoid wakeups due to
> + * flushing user TLBs.
> + *
> + * Must be called under local_irq_disable().
> + */
> +static __cpuidle int intel_idle(struct cpuidle_device *dev,
> +				struct cpuidle_driver *drv, int index)
> +{
> +	struct cpuidle_state *state = &drv->states[index];
> +	unsigned long eax = flg2MWAIT(state->flags);
> +	unsigned long ecx = 1; /* break on interrupt flag */
> +	bool uninitialized_var(tick);

This will generate an UBSAN warning because Clang could poison all
uninitialized stack variables to 0xAA due to CONFIG_INIT_STACK_ALL=y, so one
issue is that,

bool uninitialized_var(x);

would always broken on Clang like this,

[   92.140611] UBSAN: invalid-load in drivers/idle/intel_idle.c:135:7
[   92.143111] load of value 170 is not a valid value for type 'bool' (aka '_Bool')
[   92.145657] CPU: 0 PID: 0 Comm: swapper/0 Not tainted 5.7.0-rc6-next-20200522+ #3
[   92.147424] Hardware name: HP ProLiant BL660c Gen9, BIOS I38 10/17/2018
[   92.149869] Call Trace:
[   92.149869]  dump_stack+0x10b/0x17f
[   92.149869]  __ubsan_handle_load_invalid_value+0xd2/0x110
[   92.149869]  intel_idle+0x54/0xf0
[   92.156202]  cpuidle_enter_state+0x120/0x4f0
[   92.156202]  cpuidle_enter+0x5b/0xa0
[   92.156202]  call_cpuidle+0x25/0x50
[   92.156202]  do_idle+0x1eb/0x2c0
[   92.156202]  cpu_startup_entry+0x25/0x30
[   92.156202]  rest_init+0x26f/0x280
[   92.156202]  arch_call_rest_init+0x17/0x1e
[   92.156202]  start_kernel+0x598/0x633
[   92.156202]  x86_64_start_reservations+0x24/0x26
[   92.156202]  x86_64_start_kernel+0x116/0x1c1
[   92.156202]  secondary_startup_64+0xb6/0xc0

However, I am wondering if it is correct to let "tick" uninitialized to begin
with. If this condition is true,

!static_cpu_has(X86_FEATURE_ARAT) && lapic_timer_always_reliable

Then, we could in the final branch to use the uninitialized value.

if (!static_cpu_has(X86_FEATURE_ARAT) && tick)

Isn't that possible?

> +	int cpu = smp_processor_id();
> +
> +	/*
> +	 * leave_mm() to avoid costly and often unnecessary wakeups
> +	 * for flushing the user TLB's associated with the active mm.
> +	 */
> +	if (state->flags & CPUIDLE_FLAG_TLB_FLUSHED)
> +		leave_mm(cpu);
> +
> +	if (!static_cpu_has(X86_FEATURE_ARAT) && !lapic_timer_always_reliable) {
> +		/*
> +		 * Switch over to one-shot tick broadcast if the target C-state
> +		 * is deeper than C1.
> +		 */
> +		if ((eax >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK) {
> +			tick = true;
> +			tick_broadcast_enter();
> +		} else {
> +			tick = false;
> +		}
> +	}
> +
> +	mwait_idle_with_hints(eax, ecx);
> +
> +	if (!static_cpu_has(X86_FEATURE_ARAT) && tick)
> +		tick_broadcast_exit();
> +
> +	return index;
> +}