[PATCH v3 0/2] Improvements to StorageD3Enable

[PATCH v3 0/2] Improvements to StorageD3Enable

[Mario Limonciello]

A number of AMD based OEM systems have problems coming out of s2idle,
which is rooted in that the NVME device power is cut off during s2idle.

That alone is not a bug - the architecture used on Cezanne, Renoir and
Picasso expects this.

Many of these systems do include the StorageD3Enable property, but it is
located in the PCI device itself not in a root port sibling like on Intel.

Intel confirmed that this during pre-production it was placed there, and
actually for production using the PCI device itself is sufficient.

During the course of discussions on the merits of different approaches it
was mentioned that although originally introduced for NVME devices, the
Microsoft specification makes allusions to non-PCI based ACPI storage
devices as well, so a proposal was created to move this into the ACPI
subsystem.

If at a later time different firmware solutions decide to advertise this
functionality, it may make sense to move out of acpi into a more generic
location.  However both AMD's and Intel's solutions for s2idle also rely
upon calling other ACPI drivers and adopting another solution will require
coming up with alternatives for those as well.

Mario Limonciello (2):
  nvme: Look for StorageD3Enable on companion ACPI device instead
  acpi: Move check for _DSD StorageD3Enable property to acpi

 drivers/acpi/device_pm.c | 24 +++++++++++++++++++
 drivers/nvme/host/pci.c  | 50 +---------------------------------------
 include/linux/acpi.h     |  5 ++++
 3 files changed, 30 insertions(+), 49 deletions(-)

-- 
2.25.1

[PATCH v3 0/2] Improvements to StorageD3Enable

[Mario Limonciello]

A number of AMD based OEM systems have problems coming out of s2idle,
which is rooted in that the NVME device power is cut off during s2idle.

That alone is not a bug - the architecture used on Cezanne, Renoir and
Picasso expects this.

Many of these systems do include the StorageD3Enable property, but it is
located in the PCI device itself not in a root port sibling like on Intel.

Intel confirmed that this during pre-production it was placed there, and
actually for production using the PCI device itself is sufficient.

During the course of discussions on the merits of different approaches it
was mentioned that although originally introduced for NVME devices, the
Microsoft specification makes allusions to non-PCI based ACPI storage
devices as well, so a proposal was created to move this into the ACPI
subsystem.

If at a later time different firmware solutions decide to advertise this
functionality, it may make sense to move out of acpi into a more generic
location.  However both AMD's and Intel's solutions for s2idle also rely
upon calling other ACPI drivers and adopting another solution will require
coming up with alternatives for those as well.

Mario Limonciello (2):
  nvme: Look for StorageD3Enable on companion ACPI device instead
  acpi: Move check for _DSD StorageD3Enable property to acpi

 drivers/acpi/device_pm.c | 24 +++++++++++++++++++
 drivers/nvme/host/pci.c  | 50 +---------------------------------------
 include/linux/acpi.h     |  5 ++++
 3 files changed, 30 insertions(+), 49 deletions(-)

-- 
2.25.1


_______________________________________________
Linux-nvme mailing list
Linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme

Re: [PATCH v3 0/2] Improvements to StorageD3Enable

[Julian Sikorski]

Am 28.05.21 um 18:01 schrieb Mario Limonciello:
> A number of AMD based OEM systems have problems coming out of s2idle,
> which is rooted in that the NVME device power is cut off during s2idle.
> 
> That alone is not a bug - the architecture used on Cezanne, Renoir and
> Picasso expects this.
> 
> Many of these systems do include the StorageD3Enable property, but it is
> located in the PCI device itself not in a root port sibling like on Intel.
> 
> Intel confirmed that this during pre-production it was placed there, and
> actually for production using the PCI device itself is sufficient.
> 
> During the course of discussions on the merits of different approaches it
> was mentioned that although originally introduced for NVME devices, the
> Microsoft specification makes allusions to non-PCI based ACPI storage
> devices as well, so a proposal was created to move this into the ACPI
> subsystem.
> 
> If at a later time different firmware solutions decide to advertise this
> functionality, it may make sense to move out of acpi into a more generic
> location.  However both AMD's and Intel's solutions for s2idle also rely
> upon calling other ACPI drivers and adopting another solution will require
> coming up with alternatives for those as well.
> 
> Mario Limonciello (2):
>    nvme: Look for StorageD3Enable on companion ACPI device instead
>    acpi: Move check for _DSD StorageD3Enable property to acpi
> 
>   drivers/acpi/device_pm.c | 24 +++++++++++++++++++
>   drivers/nvme/host/pci.c  | 50 +---------------------------------------
>   include/linux/acpi.h     |  5 ++++
>   3 files changed, 30 insertions(+), 49 deletions(-)
> 
Are these patches meant to replace the blanket Renoir simple suspend patch?
https://gitlab.com/belegdol/kernel-ark/-/commit/a83315923e87a169ceb6c839a84ee0455868218f
I can test it on an Asus Zenbook 14 UM425IA if needed, you should be 
also be able to get a considerable amount of feedback from 
https://gitlab.freedesktop.org/drm/amd/-/issues/1230

Best regards,
Julian

_______________________________________________
Linux-nvme mailing list
Linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme