Re: [PATCH] x86: Don't use MWAIT if explicitly requested

[Zhenzhong Duan]

On 2019/10/18 15:12, Peter Zijlstra wrote:
> On Fri, Oct 18, 2019 at 11:45:54AM +0800, Zhenzhong Duan wrote:
> > If 'idle=nomwait' is specified or process matching what's in
> > processor_idle_dmi_table, we should't use MWAIT at bootup stage before
> > cpuidle driver loaded, even if it's preferred by default on Intel.
> >
> > Add a check so that HALT instruction is used in those cases.
> The comment in idle_setup():
>
> 	/*
> 	 * If the boot option of "idle=nomwait" is added,
> 	 * it means that mwait will be disabled for CPU C2/C3
> 	 * states. In such case it won't touch the variable
> 	 * of boot_option_idle_override.
> 	 */
> 	boot_option_idle_override = IDLE_NOMWAIT;
>
> explicitly states this option is for C2+

Yea, this is confusing. Other place referencing 
boot_option_idle_override tell me

"idle=nomwait" means not using mwait for all cstates. Maybe 'C2/C3' could be

removed from above comment?

See drivers/acpi/processor_idle.c:

ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ if (cx.type == ACPI_STATE_C1 &&
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ (boot_option_idle_override == IDLE_NOMWAIT)) {
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ /*
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ * In most cases the C1 space_id 
obtained from
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ * _CST object is FIXED_HARDWARE access 
mode.
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ * But when the option of idle=halt is 
added,
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ * the entry_method type should be 
changed from
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ * CSTATE_FFH to CSTATE_HALT.
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ * When the option of idle=nomwait is 
added,
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ * the C1 entry_method type should be
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ * CSTATE_HALT.
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ */
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ cx.entry_method = ACPI_CSTATE_HALT;
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ snprintf(cx.desc, ACPI_CX_DESC_LEN, 
"ACPI HLT");
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ }

and drivers/acpi/processor_pdc.c:

ÂÂÂÂÂÂÂ if (boot_option_idle_override == IDLE_NOMWAIT) {
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ /*
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ * If mwait is disabled for CPU C-states, the C2C3_FFH 
access
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ * mode will be disabled in the parameter of _PDC object.
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ * Of course C1_FFH access mode will also be disabled.
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ */
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ union acpi_object *obj;
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ u32 *buffer = NULL;

ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ obj = pdc_in->pointer;
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ buffer = (u32 *)(obj->buffer.pointer);
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ buffer[2] &= ~(ACPI_PDC_C_C2C3_FFH | ACPI_PDC_C_C1_FFH);

ÂÂÂÂÂÂÂ }

> > Signed-off-by: Zhenzhong Duan <zhenzhong.duan@xxxxxxxxxx>
> > Cc: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
> > Cc: Borislav Petkov <bp@xxxxxxxxx>
> > Cc: Ingo Molnar <mingo@xxxxxxxxxx>
> > Cc: "H. Peter Anvin" <hpa@xxxxxxxxx>
> > Cc: Boris Ostrovsky <boris.ostrovsky@xxxxxxxxxx>
> > ---
> >   arch/x86/kernel/process.c | 4 ++++
> >   1 file changed, 4 insertions(+)
> >
> > diff --git a/arch/x86/kernel/process.c b/arch/x86/kernel/process.c
> > index 5e94c43..37fc577 100644
> > --- a/arch/x86/kernel/process.c
> > +++ b/arch/x86/kernel/process.c
> > @@ -667,6 +667,10 @@ static void amd_e400_idle(void)
> >    */
> >   static int prefer_mwait_c1_over_halt(const struct cpuinfo_x86 *c)
> >   {
> > +	/* Don't use MWAIT-C1 if explicitly requested */
> > +	if (boot_option_idle_override == IDLE_NOMWAIT)
> > +		return 0;
> And this is very much about C1...
>
> OTOH, "idle=halt" should be forcing HLT over MWAIT, so did you want to
> write:
>
> 	if (boot_option_idle_override == IDLE_HALT)
> 		return 0;
>
> instead?

I think it's not necessory, if 'idle=halt' specified, 
select_idle_routine() returns early,

prefer_mwait_c1_over_halt() is never called.

Zhenzhong