Re: [PATCH v13 15/18] EDAC: Add memory repair control feature

[Jonathan Cameron <Jonathan.Cameron@Huawei.com>]

On Wed, 9 Oct 2024 13:41:16 +0100
<shiju.jose@huawei.com> wrote:

> From: Shiju Jose <shiju.jose@huawei.com>
> 
> Add generic EDAC memory repair control, eg. PPR(Post Package Repair),
> memory sparing etc, control driver in order to control memory repairs
> in the system. Supports sPPR(soft PPR), hPPR(hard PPR), soft/hard memory
> sparing, memory sparing at cacheline/row/bank/rank granularity etc.
> Device with memory repair features registers with EDAC device driver,
> which retrieves memory repair descriptor from EDAC memory repair driver and
> exposes the sysfs repair control attributes to userspace in
> /sys/bus/edac/devices/<dev-name>/mem_repairX/.
> 
> The common memory repair control interface abstracts the control of an
> arbitrary memory repair functionality to a common set of functions.
> The sysfs memory repair attribute nodes would be present only if the client
> driver has implemented the corresponding attribute callback function and
> passed in ops to the EDAC device driver during registration.
> 
> Signed-off-by: Shiju Jose <shiju.jose@huawei.com>
The question inline that we discussed offlist.

Whether it makes sense to potentially have one device provide
several mem_repairX differing in granularity (and may type) of
repair, or one mem_repairX that has a control over granularity?

The CXL spec has it designed as separate control interfaces but
I'm not sure if we should follow that precedence or not.

> ---
>  .../ABI/testing/sysfs-edac-mem-repair         | 152 +++++++++
>  drivers/edac/Makefile                         |   2 +-
>  drivers/edac/edac_device.c                    |  31 ++
>  drivers/edac/mem_repair.c                     | 317 ++++++++++++++++++
>  include/linux/edac.h                          |  67 ++++
>  5 files changed, 568 insertions(+), 1 deletion(-)
>  create mode 100644 Documentation/ABI/testing/sysfs-edac-mem-repair
>  create mode 100755 drivers/edac/mem_repair.c
> 
> diff --git a/Documentation/ABI/testing/sysfs-edac-mem-repair b/Documentation/ABI/testing/sysfs-edac-mem-repair
> new file mode 100644
> index 000000000000..9a8712ed9d47
> --- /dev/null
> +++ b/Documentation/ABI/testing/sysfs-edac-mem-repair
> @@ -0,0 +1,152 @@
> +What:		/sys/bus/edac/devices/<dev-name>/mem_repairX
> +Date:		Oct 2024
> +KernelVersion:	6.12
> +Contact:	linux-edac@vger.kernel.org
> +Description:
> +		The sysfs EDAC bus devices /<dev-name>/mem_repairX subdirectory
> +		belongs to the memory media repair features control, such as
> +		PPR (Post Package Repair), memory sparing etc, where<dev-name>
> +		directory corresponds to a device registered with the EDAC
> +		device driver for the memory repair features.
> +		/mem_repairX belongs to either sPPR (Soft PPR) or hPPR (Hard PPR)
> +		feature of PPR feature, hard or soft memory sparing etc. The memory
> +		sparing is a repair function that replaces a portion of memory
> +		(spared memory) with a portion of functional memory. The memory
> +		sparing has cacheline/row/bank/rank sparing granularities.
> +		The sysfs memory repair attr nodes would be only present if a
> +		memory repair feature is supported.
> +
> +What:		/sys/bus/edac/devices/<dev-name>/mem_repairX/repair_type
> +Date:		Oct 2024
> +KernelVersion:	6.12
> +Contact:	linux-edac@vger.kernel.org
> +Description:
> +		(RO) Type of the repair instance. For eg. sPPR, hPPR, cacheline/
> +		row/bank/rank memory sparing etc.
So this is the open question for me with this feature.
Do we do a monolithic 'device' that does all repair types for which we pick a mode
or do we (as here) allow for one mem_repairX for each supported type?

I don't particularly mind but it is a design question I'd like input on
from a wider audience.

> +
> +What:		/sys/bus/edac/devices/<dev-name>/mem_repairX/hpa
> +Date:		Oct 2024
> +KernelVersion:	6.12
> +Contact:	linux-edac@vger.kernel.org
> +Description:
> +		(WO) Set HPA (Host Physical Address) for memory repair.

Can we not just read back what was written?  Seems like userspace
might expect that?