Re: [PATCH v2] xen/acpi: upload power and performance related data from a PVH dom0

["Jürgen Groß" <jgross@suse.com>]

[-- Attachment #1.1.1: Type: text/plain, Size: 4110 bytes --]

On 16.09.24 22:50, Jason Andryuk wrote:
> From: Roger Pau Monne <roger.pau@citrix.com>
> 
> When running as a PVH dom0 the ACPI MADT is crafted by Xen in order to
> report the correct numbers of vCPUs that dom0 has, so the host MADT is
> not provided to dom0.  This creates issues when parsing the power and
> performance related data from ACPI dynamic tables, as the ACPI
> Processor UIDs found on the dynamic code are likely to not match the
> ones crafted by Xen in the dom0 MADT.
> 
> Xen would rely on Linux having filled at least the power and
> performance related data of the vCPUs on the system, and would clone
> that information in order to setup the remaining pCPUs on the system
> if dom0 vCPUs < pCPUs.  However when running as PVH dom0 it's likely
> that none of dom0 CPUs will have the power and performance data
> filled, and hence the Xen ACPI Processor driver needs to fetch that
> information by itself.
> 
> In order to do so correctly, introduce a new helper to fetch the _CST
> data without taking into account the system capabilities from the
> CPUID output, as the capabilities reported to dom0 in CPUID might be
> different from the ones on the host.
> 
> Note that the newly introduced code will only fetch the _CST, _PSS,
> _PPC and _PCT from a single CPU, and clone that information for all the
> other Processors.  This won't work on an heterogeneous system with
> Processors having different power and performance related data between
> them.
> 
> Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>
> Signed-off-by: Jason Andryuk <jason.andryuk@amd.com>
> ---
> v2:
> Add second buffer for _CST.  Call was failing with
> AE_BUFFER_OVERFLOW(0x000b)
> 
> Running a PVH Dom0 on AMD, I needed this v2 change to get the C-State
> information uploaded.
> 
> ---
>   drivers/xen/pcpu.c               |   3 +-
>   drivers/xen/xen-acpi-processor.c | 230 ++++++++++++++++++++++++++++---
>   include/xen/xen.h                |   2 +-
>   3 files changed, 216 insertions(+), 19 deletions(-)
> 
> diff --git a/drivers/xen/pcpu.c b/drivers/xen/pcpu.c
> index c63f317e3df3..dc9f2c14bf62 100644
> --- a/drivers/xen/pcpu.c
> +++ b/drivers/xen/pcpu.c

...

> @@ -354,24 +511,44 @@ read_acpi_id(acpi_handle handle, u32 lvl, void *context, void **rv)
>   	default:
>   		return AE_OK;
>   	}
> -	if (invalid_phys_cpuid(acpi_get_phys_id(handle,
> -						acpi_type == ACPI_TYPE_DEVICE,
> -						acpi_id))) {
> +
> +	if (!xen_processor_present(acpi_id)) {
>   		pr_debug("CPU with ACPI ID %u is unavailable\n", acpi_id);
>   		return AE_OK;
>   	}
> -	/* There are more ACPI Processor objects than in x2APIC or MADT.
> -	 * This can happen with incorrect ACPI SSDT declerations. */
> -	if (acpi_id >= nr_acpi_bits) {
> -		pr_debug("max acpi id %u, trying to set %u\n",
> -			 nr_acpi_bits - 1, acpi_id);
> -		return AE_OK;
> -	}
> +
>   	/* OK, There is a ACPI Processor object */
>   	__set_bit(acpi_id, acpi_id_present);
>   
>   	pr_debug("ACPI CPU%u w/ PBLK:0x%lx\n", acpi_id, (unsigned long)pblk);
>   
> +	if (!pr_initialized) {
> +		struct acpi_processor *pr = context;
> +		int rc;
> +
> +		/*
> +		 * There's no CPU on the system that has any performance or
> +		 * power related data, initialize all the required fields by
> +		 * fetching that info here.
> +		 *
> +		 * Note such information is only fetched once, and then reused
> +		 * for all pCPUs.  This won't work on heterogeneous systems
> +		 * with different Cx anb/or Px states between CPUs.
> +		 */
> +
> +		pr->handle = handle;
> +
> +		rc = acpi_processor_get_performance_info(pr);
> +		if (rc)
> +			pr_debug("ACPI CPU%u failed to get performance data\n",
> +				 acpi_id);

Is it really normal to get a failure here? Shouldn't the reaction
be a little bit more visible in this case?

And can you just continue processing?

> +		rc = xen_acpi_processor_evaluate_cst(handle, &pr->power);
> +		if (rc)
> +			pr_debug("ACPI CPU%u failed to get _CST data\n", acpi_id);

Same again. Is pr_debug() enough?


Juergen

[-- Attachment #1.1.2: OpenPGP public key --]
[-- Type: application/pgp-keys, Size: 3743 bytes --]

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 495 bytes --]