mbox series

[0/5] cxl/pmem: Add core infrastructure for PMEM support

Message ID 162336395765.2462439.11368504490069925374.stgit@dwillia2-desk3.amr.corp.intel.com (mailing list archive)
Headers show
Series cxl/pmem: Add core infrastructure for PMEM support | expand

Message

Dan Williams June 10, 2021, 10:25 p.m. UTC
CXL Memory Expander devices (CXL 2.0 Type-3) support persistent memory
in addition to volatile memory expansion. The most significant changes
this requires of the existing LIBNVDIMM infrastructure, compared to what
was needed to support ACPI NFIT defined PMEM, is the ability to
dynamically provision regions in addition to namespaces, and a formal
model for hotplug.

Before region provisioning can be added the CXL enabling needs to
enumerate "nvdimm" devices on a CXL nvdimm-bus. This is modeled as a
CXL-nvdimm-bridge device (bridging CXL to nvdimm) and an associated
driver to activate and deactivate that bus-bridge. Once the bridge is
registered it scans for CXL nvdimm devices registered by endpoints.  The
CXL core bus is used as a rendezvous for nvdimm bridges and endpoints
allowing them to be registered and enabled in any order.

At the end of this series the ndctl utility can see CXL nvdimm resources
just like any other nvdimm bus.

    # ndctl list -BDiu -b CXL
    {
      "provider":"CXL",
      "dev":"ndbus1",
      "dimms":[
        { 
          "dev":"nmem1",
          "state":"disabled"
        },
        { 
          "dev":"nmem0",
          "state":"disabled"
        }
      ]
    }

Follow-on patches extend the nvdimm core label support for CXL region
and namespace labels. For now just add the machinery to register the
bus and nvdimm base objects.

---

Dan Williams (5):
      cxl/core: Add cxl-bus driver infrastructure
      cxl/pmem: Add initial infrastructure for pmem support
      libnvdimm: Export nvdimm shutdown helper, nvdimm_delete()
      libnvdimm: Drop unused device power management support
      cxl/pmem: Register 'pmem' / cxl_nvdimm  devices


 drivers/cxl/Kconfig        |   13 ++
 drivers/cxl/Makefile       |    2 
 drivers/cxl/acpi.c         |   37 +++++-
 drivers/cxl/core.c         |  281 ++++++++++++++++++++++++++++++++++++++++++++
 drivers/cxl/cxl.h          |   57 +++++++++
 drivers/cxl/mem.h          |    2 
 drivers/cxl/pci.c          |   23 +++-
 drivers/cxl/pmem.c         |  230 ++++++++++++++++++++++++++++++++++++
 drivers/nvdimm/bus.c       |   64 ++++++----
 drivers/nvdimm/dimm_devs.c |   18 +++
 include/linux/libnvdimm.h  |    1 
 11 files changed, 697 insertions(+), 31 deletions(-)
 create mode 100644 drivers/cxl/pmem.c

base-commit: 40ba17afdfabb01688c61565dbe02a916241bc05

Comments

Jonathan Cameron June 11, 2021, 12:59 p.m. UTC | #1
On Thu, 10 Jun 2021 15:25:57 -0700
Dan Williams <dan.j.williams@intel.com> wrote:

> CXL Memory Expander devices (CXL 2.0 Type-3) support persistent memory
> in addition to volatile memory expansion. The most significant changes
> this requires of the existing LIBNVDIMM infrastructure, compared to what
> was needed to support ACPI NFIT defined PMEM, is the ability to
> dynamically provision regions in addition to namespaces, and a formal
> model for hotplug.
> 
> Before region provisioning can be added the CXL enabling needs to
> enumerate "nvdimm" devices on a CXL nvdimm-bus. This is modeled as a
> CXL-nvdimm-bridge device (bridging CXL to nvdimm) and an associated
> driver to activate and deactivate that bus-bridge. Once the bridge is
> registered it scans for CXL nvdimm devices registered by endpoints.  The
> CXL core bus is used as a rendezvous for nvdimm bridges and endpoints
> allowing them to be registered and enabled in any order.
> 
> At the end of this series the ndctl utility can see CXL nvdimm resources
> just like any other nvdimm bus.
> 
>     # ndctl list -BDiu -b CXL
>     {
>       "provider":"CXL",
>       "dev":"ndbus1",
>       "dimms":[
>         { 
>           "dev":"nmem1",
>           "state":"disabled"
>         },
>         { 
>           "dev":"nmem0",
>           "state":"disabled"
>         }
>       ]
>     }
> 
> Follow-on patches extend the nvdimm core label support for CXL region
> and namespace labels. For now just add the machinery to register the
> bus and nvdimm base objects.

Hi Dan,

Nice set. I won't claim to know all that much about the nvdimm
side of things. With that in mind and the fact all my comments
were trivial stuff you can clean up without me reading it again.

Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>

for whole set.

Thanks,

Jonathan

> 
> ---
> 
> Dan Williams (5):
>       cxl/core: Add cxl-bus driver infrastructure
>       cxl/pmem: Add initial infrastructure for pmem support
>       libnvdimm: Export nvdimm shutdown helper, nvdimm_delete()
>       libnvdimm: Drop unused device power management support
>       cxl/pmem: Register 'pmem' / cxl_nvdimm  devices
> 
> 
>  drivers/cxl/Kconfig        |   13 ++
>  drivers/cxl/Makefile       |    2 
>  drivers/cxl/acpi.c         |   37 +++++-
>  drivers/cxl/core.c         |  281 ++++++++++++++++++++++++++++++++++++++++++++
>  drivers/cxl/cxl.h          |   57 +++++++++
>  drivers/cxl/mem.h          |    2 
>  drivers/cxl/pci.c          |   23 +++-
>  drivers/cxl/pmem.c         |  230 ++++++++++++++++++++++++++++++++++++
>  drivers/nvdimm/bus.c       |   64 ++++++----
>  drivers/nvdimm/dimm_devs.c |   18 +++
>  include/linux/libnvdimm.h  |    1 
>  11 files changed, 697 insertions(+), 31 deletions(-)
>  create mode 100644 drivers/cxl/pmem.c
> 
> base-commit: 40ba17afdfabb01688c61565dbe02a916241bc05