2018-03-02 05:32:18

by David Wang

[permalink] [raw]
Subject: [PATCH] Delete some unusefull operations for centaur CPU/platform

For Centaur CPU, the ucode will make sure that each CPU core can keep cache
coherency with each other when the CPU core entering to any C state. So the
cache flush operations when enter C3 is not necessary and will cause large
C3 enter/exit latency.
And the bus master disable operation when CPU core entering C3 state is not
needed too. Because the chipset will automatically do this operation.

Signed-off-by: David Wang <[email protected]>
---
arch/x86/kernel/acpi/cstate.c | 12 ++++++++++++
1 file changed, 12 insertions(+)

diff --git a/arch/x86/kernel/acpi/cstate.c b/arch/x86/kernel/acpi/cstate.c
index dde437f..1cd357b 100644
--- a/arch/x86/kernel/acpi/cstate.c
+++ b/arch/x86/kernel/acpi/cstate.c
@@ -51,6 +51,18 @@ void acpi_processor_power_init_bm_check(struct acpi_processor_flags *flags,
if (c->x86_vendor == X86_VENDOR_INTEL &&
(c->x86 > 0xf || (c->x86 == 6 && c->x86_model >= 0x0f)))
flags->bm_control = 0;
+
+ if (c->x86_vendor == X86_VENDOR_CENTAUR) {
+ /*
+ * on all centaur CPUs, sw need not execute cache flush operation
+ * when entering C3 type State.
+ *
+ * On all Centaur platforms, sw need not execute ARB_DISABLE while
+ * entering C3 type state.
+ */
+ flags->bm_check = 1;
+ flags->bm_control = 0;
+ }
}
EXPORT_SYMBOL(acpi_processor_power_init_bm_check);

--
1.9.1



2018-03-02 11:31:51

by Rafael J. Wysocki

[permalink] [raw]
Subject: Re: [PATCH] Delete some unusefull operations for centaur CPU/platform

On Friday, March 2, 2018 5:11:48 AM CET David Wang wrote:
> For Centaur CPU, the ucode will make sure that each CPU core can keep cache
> coherency with each other when the CPU core entering to any C state. So the
> cache flush operations when enter C3 is not necessary and will cause large
> C3 enter/exit latency.
> And the bus master disable operation when CPU core entering C3 state is not
> needed too. Because the chipset will automatically do this operation.
>
> Signed-off-by: David Wang <[email protected]>

I've queued up the previous version of your patch with some changes to
comments/subject/changelog made by me for v4.17.

Please see linux-next.

> ---
> arch/x86/kernel/acpi/cstate.c | 12 ++++++++++++
> 1 file changed, 12 insertions(+)
>
> diff --git a/arch/x86/kernel/acpi/cstate.c b/arch/x86/kernel/acpi/cstate.c
> index dde437f..1cd357b 100644
> --- a/arch/x86/kernel/acpi/cstate.c
> +++ b/arch/x86/kernel/acpi/cstate.c
> @@ -51,6 +51,18 @@ void acpi_processor_power_init_bm_check(struct acpi_processor_flags *flags,
> if (c->x86_vendor == X86_VENDOR_INTEL &&
> (c->x86 > 0xf || (c->x86 == 6 && c->x86_model >= 0x0f)))
> flags->bm_control = 0;
> +
> + if (c->x86_vendor == X86_VENDOR_CENTAUR) {
> + /*
> + * on all centaur CPUs, sw need not execute cache flush operation
> + * when entering C3 type State.
> + *
> + * On all Centaur platforms, sw need not execute ARB_DISABLE while
> + * entering C3 type state.
> + */
> + flags->bm_check = 1;
> + flags->bm_control = 0;
> + }
> }
> EXPORT_SYMBOL(acpi_processor_power_init_bm_check);
>
>



2018-03-04 09:43:22

by Ingo Molnar

[permalink] [raw]
Subject: Re: [PATCH] Delete some unusefull operations for centaur CPU/platform

* David Wang <[email protected]> wrote:

> For Centaur CPU, the ucode will make sure that each CPU core can keep cache
> coherency with each other when the CPU core entering to any C state. So the
> cache flush operations when enter C3 is not necessary and will cause large
> C3 enter/exit latency.
> And the bus master disable operation when CPU core entering C3 state is not
> needed too. Because the chipset will automatically do this operation.
>
> Signed-off-by: David Wang <[email protected]>
> ---
> arch/x86/kernel/acpi/cstate.c | 12 ++++++++++++
> 1 file changed, 12 insertions(+)
>
> diff --git a/arch/x86/kernel/acpi/cstate.c b/arch/x86/kernel/acpi/cstate.c
> index dde437f..1cd357b 100644
> --- a/arch/x86/kernel/acpi/cstate.c
> +++ b/arch/x86/kernel/acpi/cstate.c
> @@ -51,6 +51,18 @@ void acpi_processor_power_init_bm_check(struct acpi_processor_flags *flags,
> if (c->x86_vendor == X86_VENDOR_INTEL &&
> (c->x86 > 0xf || (c->x86 == 6 && c->x86_model >= 0x0f)))
> flags->bm_control = 0;
> +
> + if (c->x86_vendor == X86_VENDOR_CENTAUR) {
> + /*
> + * on all centaur CPUs, sw need not execute cache flush operation
> + * when entering C3 type State.
> + *
> + * On all Centaur platforms, sw need not execute ARB_DISABLE while
> + * entering C3 type state.
> + */
> + flags->bm_check = 1;
> + flags->bm_control = 0;
> + }

Please clean up capitalization and spelling of the comment, and please also
explain the ACPI-idle settings a bit better, i.e. something like:

/*
* On all Centaur CPUs the kernel does not have to
* execute a cache flush operation (WBINVD) when
* entering C3 state.
*
* On all Centaur platforms the kernel does not have to
* disable bus-master arbitration (ARB_DISABLE) when
* entering C3 state.
*/

Also, two questions:

1)

I'm wondering about this logic in acpi_processor_power_verify_c3():

if (pr->flags.bm_check) {
if (!pr->flags.bm_control) {
if (pr->flags.has_cst != 1) {
/* bus mastering control is necessary */
ACPI_DEBUG_PRINT((ACPI_DB_INFO,
"C3 support requires BM control\n"));
return;

So ->has_cst is set on Centaur CPUs?

2)

There's this comment in acpi_idle_enter_bm():

/*
* disable bus master
* bm_check implies we need ARB_DIS
* bm_control implies whether we can do ARB_DIS
*
* That leaves a case where bm_check is set and bm_control is
* not set. In that case we cannot do much, we enter C3
* without doing anything.
*/
if (pr->flags.bm_control) {

It says 'bm_check == 1 implies we need ARB_DIS' - but we do the opposite in the
Centaur case? It it me who is confused or the comment?

Thanks,

Ingo