2022-06-23 01:29:54

by Tony W Wang-oc

Subject: [PATCH V1] x86/cstate: Add Zhaoxin/Centaur ACPI Cx FFH MWAIT support

Recent Zhaoxin/Centaur CPUs support X86_FEATURE_MWAIT that implies
the MONITOR/MWAIT instructions can be used for ACPI Cx state.
The BIOS declares Cx state in _CST object to use FFH on Zhaoxin/Centaur
systems. So let function ffh_cstate_init() support These CPUs too.

Signed-off-by: Tony W Wang-oc <[email protected]>
---
arch/x86/kernel/acpi/cstate.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/arch/x86/kernel/acpi/cstate.c b/arch/x86/kernel/acpi/cstate.c
index 7945eae..d4185e1 100644
--- a/arch/x86/kernel/acpi/cstate.c
+++ b/arch/x86/kernel/acpi/cstate.c
@@ -213,7 +213,9 @@ static int __init ffh_cstate_init(void)

if (c->x86_vendor != X86_VENDOR_INTEL &&
c->x86_vendor != X86_VENDOR_AMD &&
- c->x86_vendor != X86_VENDOR_HYGON)
+ c->x86_vendor != X86_VENDOR_HYGON &&
+ c->x86_vendor != X86_VENDOR_CENTAUR &&
+ c->x86_vendor != X86_VENDOR_ZHAOXIN)
return -1;

cpu_cstate_entry = alloc_percpu(struct cstate_entry);
--
2.7.4

2022-06-23 16:25:37

by Dave Hansen

Subject: Re: [PATCH V1] x86/cstate: Add Zhaoxin/Centaur ACPI Cx FFH MWAIT support

On 6/22/22 18:26, Tony W Wang-oc wrote:
> Recent Zhaoxin/Centaur CPUs support X86_FEATURE_MWAIT that implies
> the MONITOR/MWAIT instructions can be used for ACPI Cx state.
> The BIOS declares Cx state in _CST object to use FFH on Zhaoxin/Centaur
> systems. So let function ffh_cstate_init() support These CPUs too.
>
> Signed-off-by: Tony W Wang-oc <[email protected]>
> ---
>  arch/x86/kernel/acpi/cstate.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/arch/x86/kernel/acpi/cstate.c b/arch/x86/kernel/acpi/cstate.c
> index 7945eae..d4185e1 100644
> --- a/arch/x86/kernel/acpi/cstate.c
> +++ b/arch/x86/kernel/acpi/cstate.c
> @@ -213,7 +213,9 @@ static int __init ffh_cstate_init(void)
>
>      if (c->x86_vendor != X86_VENDOR_INTEL &&
>          c->x86_vendor != X86_VENDOR_AMD &&
> -        c->x86_vendor != X86_VENDOR_HYGON)
> +        c->x86_vendor != X86_VENDOR_HYGON &&
> +        c->x86_vendor != X86_VENDOR_CENTAUR &&
> +        c->x86_vendor != X86_VENDOR_ZHAOXIN)
>          return -1;

Many of the changelogs that add new vendors here go on about particular
C states declared in the _CST object and contents of CPUID leaf 5.

Why do we even _have_ a vendor check here? Shouldn't the code just be
going and doing the validation of the _CST object and CPUID that the
changelogs blather on about?

Intel certainly made the original sin on this one (see 991528d7348), but
I hope _something_ changed in the 16 years since that patch went in.

2022-06-24 03:38:03

by Tony W Wang-oc

Subject: Re: [PATCH V1] x86/cstate: Add Zhaoxin/Centaur ACPI Cx FFH MWAIT support

On 23/6/2022 23:55, Dave Hansen wrote:
> On 6/22/22 18:26, Tony W Wang-oc wrote:
>> Recent Zhaoxin/Centaur CPUs support X86_FEATURE_MWAIT that implies
>> the MONITOR/MWAIT instructions can be used for ACPI Cx state.
>> The BIOS declares Cx state in _CST object to use FFH on Zhaoxin/Centaur
>> systems. So let function ffh_cstate_init() support These CPUs too.
>>
>> Signed-off-by: Tony W Wang-oc <[email protected]>
>> ---
>>  arch/x86/kernel/acpi/cstate.c | 4 +++-
>>  1 file changed, 3 insertions(+), 1 deletion(-)
>>
>> diff --git a/arch/x86/kernel/acpi/cstate.c b/arch/x86/kernel/acpi/cstate.c
>> index 7945eae..d4185e1 100644
>> --- a/arch/x86/kernel/acpi/cstate.c
>> +++ b/arch/x86/kernel/acpi/cstate.c
>> @@ -213,7 +213,9 @@ static int __init ffh_cstate_init(void)
>>
>>      if (c->x86_vendor != X86_VENDOR_INTEL &&
>>          c->x86_vendor != X86_VENDOR_AMD &&
>> -        c->x86_vendor != X86_VENDOR_HYGON)
>> +        c->x86_vendor != X86_VENDOR_HYGON &&
>> +        c->x86_vendor != X86_VENDOR_CENTAUR &&
>> +        c->x86_vendor != X86_VENDOR_ZHAOXIN)
>>          return -1;
>
> Many of the changelogs that add new vendors here go on about particular
> C states declared in the _CST object and contents of CPUID leaf 5.
>
> Why do we even _have_ a vendor check here? Shouldn't the code just be
> going and doing the validation of the _CST object and CPUID that the
> changelogs blather on about?
>

Yes, agree!

Will change as below. Please help to check if it is OK. Thanks a lot.
static int __init ffh_cstate_init(void)
{
- struct cpuinfo_x86 *c = &boot_cpu_data;
-
- if (c->x86_vendor != X86_VENDOR_INTEL &&
- c->x86_vendor != X86_VENDOR_AMD &&
- c->x86_vendor != X86_VENDOR_HYGON)
+ if (!boot_cpu_has(X86_FEATURE_MWAIT))
return -1;

> Intel certainly made the original sin on this one (see 991528d7348), but
> I hope _something_ changed in the 16 years since that patch went in.
> .
>

--
Sincerely
TonyWWang-oc