[v6] ppc: spapr: virtual NVDIMM support

[PATCH v6 0/4] ppc: spapr: virtual NVDIMM support

Posted by Shivaprasad G Bhat 4 years, 2 months ago

The patchset attempts to implement the virtual NVDIMM for pseries.

PAPR semantics is such that each NVDIMM device is comprising of
multiple SCM(Storage Class Memory) blocks. The hypervisor is expected
to prepare the FDT for the NVDIMM device and send guest a hotplug
interrupt with new type RTAS_LOG_V6_HP_TYPE_PMEM currently handled by
the upstream kernel. In response to that interrupt, the guest requests
the hypervisor to bind each of the SCM blocks of the NVDIMM device
using hcalls. There can be SCM block unbind requests in case of driver
errors or unplug(not supported now) use cases. The NVDIMM label
read/writes are done through hcalls.

Since each virtual NVDIMM device is divided into multiple SCM blocks,
the bind, unbind, and queries using hcalls on those blocks can come
independently. This doesnt fit well into the qemu device semantics,
where the map/unmap are done at the (whole)device/object level
granularity. The patchset uses the existing NVDIMM class structures
for the implementation. The bind/unbind is left to happen at the
device_add/del phase itself instead of at hcalls on-demand.

The guest kernel makes bind/unbind requests for the virtual NVDIMM
device at the region level granularity. Without interleaving, each
virtual NVDIMM device is presented as separate region. Hence it is
safe to do bind/unbind everything during the object_add/del.

The free device-memory region which is used for memory hotplug are
done using multiple LMBs of size(256MiB) and are expected to be
aligned to 256 MiB. As the SCM blocks are mapped to the same region,
the SCM blocks also need to be aligned to this size for the subsequent
memory hotplug to work. The minimum SCM block size is set to this size
for that reason and can be made user configurable in future if required.

The first patch moves around the existing static function to common
area for using it in the subsequent patches. Second patch adds new uuid
property to the nvdimm device. Third patch adds FDT entries and basic
device support, the fourth patch adds the hcalls implementation.

The patches are also available at
https://github.com/ShivaprasadGBhat/qemu.git - pseries-nvdimm-v6 branch
and can be used with the upstream kernel. ndctl can be used for
configuring the nvdimms inside the guest.
This is how it can be used ..
Ex :
For coldplug, the device to be added in qemu command line as shown below
-object
memory-backend-file,id=memnvdimm0,prealloc=yes,mem-path=/tmp/nvdimm0,share=yes,size=1073872896
-device
nvdimm,label-size=128k,uuid=75a3cdd7-6a2f-4791-8d15-fe0a920e8e9e,memdev=memnvdimm0,id=nvdimm0,slot=0

For hotplug, the device to be added from monitor as below
object_add
memory-backend-file,id=memnvdimm0,prealloc=yes,mem-path=/tmp/nvdimm0,share=yes,size=1073872896
device_add
nvdimm,label-size=128k,uuid=75a3cdd7-6a2f-4791-8d15-fe0a920e8e9e,memdev=memnvdimm0,id=nvdimm0,slot=0

---
v5: https://lists.nongnu.org/archive/html/qemu-devel/2020-01/msg07472.html
Changes from v5:
- Moved most of the nvdimm code from spapr.c to spapr_nvdimm.c
- Addressed all style/logic comments.
v4: https://lists.gnu.org/archive/html/qemu-devel/2019-12/msg03455.html
Changes from v4:
- The nvdimm occupied GPA area is marked as available for hotplug, the
existing code takes care of if the dimm device is actually present there
or used by nvdimm.
- fixed all comments for hcall implementation code on style/logic issues.
v3: https://lists.gnu.org/archive/html/qemu-devel/2019-10/msg03452.html
Changes from v3:
- Moved NVDIMM uuid property addition to new patch.
- Moved the SCM hcalls to new file
- Changed the metadata read/write hcalls to use st/ldX_be_p macros.
- Fixed all comments on v3
v2: https://lists.gnu.org/archive/html/qemu-devel/2019-05/msg02785.html
Changes from v2:
- Creating the drc indices for the nvdimm devices in advance as
suggested based on the number of user specified max slots property.
- Removed the hard dependency on -machine nvdimm=on, enabled by
default on the current latest pseries machine version.
- Renamed the functions to spapr_dt_X as suggested.
- Metadata is byteswapped before read/write to take care of endianness
semantics during the hcall.
v1 : http://lists.nongnu.org/archive/html/qemu-devel/2019-02/msg01545.html
Changes from v1:
- Rebased to upstream, this required required dt_populate implementation
for nvdimm hotplug support
- Added uuid option to nvdimm device
- Removed the memory region sizing down code as suggested by Igor,
now erroring out if NVDIMM size excluding the label area is not
aligned to 256MB, so patch 2 from previous series no longer needed.
- Removed un-implemented hcalls
- Changed the hcalls to different kinds of checks and return
different values.
- Addressed comments for v1
---

Shivaprasad G Bhat (4):
mem: move nvdimm_device_list to utilities
nvdimm: add uuid property to nvdimm
spapr: Add NVDIMM device support
spapr: Add Hcalls to support PAPR NVDIMM device

--
Signature

Re: [PATCH v6 0/4] ppc: spapr: virtual NVDIMM support

Posted by David Gibson 4 years, 2 months ago

On Sun, Feb 09, 2020 at 10:55:51PM -0600, Shivaprasad G Bhat wrote:
> The patchset attempts to implement the virtual NVDIMM for pseries.
> 
> PAPR semantics is such that each NVDIMM device is comprising of
> multiple SCM(Storage Class Memory) blocks. The hypervisor is expected
> to prepare the FDT for the NVDIMM device and send guest a hotplug
> interrupt with new type RTAS_LOG_V6_HP_TYPE_PMEM currently handled by
> the upstream kernel. In response to that interrupt, the guest requests
> the hypervisor to bind each of the SCM blocks of the NVDIMM device
> using hcalls. There can be SCM block unbind requests in case of driver
> errors or unplug(not supported now) use cases. The NVDIMM label
> read/writes are done through hcalls.
> 
> Since each virtual NVDIMM device is divided into multiple SCM blocks,
> the bind, unbind, and queries using hcalls on those blocks can come
> independently. This doesnt fit well into the qemu device semantics,
> where the map/unmap are done at the (whole)device/object level
> granularity. The patchset uses the existing NVDIMM class structures
> for the implementation. The bind/unbind is left to happen at the
> device_add/del phase itself instead of at hcalls on-demand.
> 
> The guest kernel makes bind/unbind requests for the virtual NVDIMM
> device at the region level granularity. Without interleaving, each
> virtual NVDIMM device is presented as separate region. Hence it is
> safe to do bind/unbind everything during the object_add/del.
> 
> The free device-memory region which is used for memory hotplug are
> done using multiple LMBs of size(256MiB) and are expected to be
> aligned to 256 MiB. As the SCM blocks are mapped to the same region,
> the SCM blocks also need to be aligned to this size for the subsequent
> memory hotplug to work. The minimum SCM block size is set to this size
> for that reason and can be made user configurable in future if required.
> 
> The first patch moves around the existing static function to common
> area for using it in the subsequent patches. Second patch adds new uuid
> property to the nvdimm device. Third patch adds FDT entries and basic
> device support, the fourth patch adds the hcalls implementation.
> 
> The patches are also available at
> https://github.com/ShivaprasadGBhat/qemu.git - pseries-nvdimm-v6 branch
> and can be used with the upstream kernel. ndctl can be used for
> configuring the nvdimms inside the guest.
> This is how it can be used ..
> Ex :
> For coldplug, the device to be added in qemu command line as shown below
> -object 
> memory-backend-file,id=memnvdimm0,prealloc=yes,mem-path=/tmp/nvdimm0,share=yes,size=1073872896
> -device 
> nvdimm,label-size=128k,uuid=75a3cdd7-6a2f-4791-8d15-fe0a920e8e9e,memdev=memnvdimm0,id=nvdimm0,slot=0
> 
> For hotplug, the device to be added from monitor as below
> object_add 
> memory-backend-file,id=memnvdimm0,prealloc=yes,mem-path=/tmp/nvdimm0,share=yes,size=1073872896
> device_add 
> nvdimm,label-size=128k,uuid=75a3cdd7-6a2f-4791-8d15-fe0a920e8e9e,memdev=memnvdimm0,id=nvdimm0,slot=0

Applied to ppc-for-5.0, thanks.

> 
> ---
> v5: https://lists.nongnu.org/archive/html/qemu-devel/2020-01/msg07472.html
> Changes from v5:
>      - Moved most of the nvdimm code from spapr.c to spapr_nvdimm.c
>      - Addressed all style/logic comments.
> v4: https://lists.gnu.org/archive/html/qemu-devel/2019-12/msg03455.html
> Changes from v4:
>      - The nvdimm occupied GPA area is marked as available for hotplug, the
>        existing code takes care of if the dimm device is actually present there
>        or used by nvdimm.
>      - fixed all comments for hcall implementation code on style/logic issues.
> v3: https://lists.gnu.org/archive/html/qemu-devel/2019-10/msg03452.html
> Changes from v3:
>      - Moved NVDIMM uuid property addition to new patch.
>      - Moved the SCM hcalls to new file
>      - Changed the metadata read/write hcalls to use st/ldX_be_p macros.
>      - Fixed all comments on v3
> v2: https://lists.gnu.org/archive/html/qemu-devel/2019-05/msg02785.html
> Changes from v2:
>      - Creating the drc indices for the nvdimm devices in advance as
>        suggested based on the number of user specified max slots property.
>      - Removed the hard dependency on -machine nvdimm=on, enabled by
>        default on the current latest pseries machine version.
>      - Renamed the functions to spapr_dt_X as suggested.
>      - Metadata is byteswapped before read/write to take care of endianness
>        semantics during the hcall.
> v1 : http://lists.nongnu.org/archive/html/qemu-devel/2019-02/msg01545.html
> Changes from v1:
>      - Rebased to upstream, this required required dt_populate implementation
>        for nvdimm hotplug support
>      - Added uuid option to nvdimm device
>      - Removed the memory region sizing down code as suggested by Igor,
>        now erroring out if NVDIMM size excluding the label area is not
>        aligned to 256MB, so patch 2 from previous series no longer needed.
>      - Removed un-implemented hcalls
>      - Changed the hcalls to different kinds of checks and return
>        different values.
>      - Addressed comments for v1
> ---
> 
> Shivaprasad G Bhat (4):
>       mem: move nvdimm_device_list to utilities
>       nvdimm: add uuid property to nvdimm
>       spapr: Add NVDIMM device support
>       spapr: Add Hcalls to support PAPR NVDIMM device
> 
> 
>  default-configs/ppc64-softmmu.mak |    1 
>  hw/acpi/nvdimm.c                  |   28 --
>  hw/mem/Kconfig                    |    2 
>  hw/mem/nvdimm.c                   |   40 +++
>  hw/ppc/Makefile.objs              |    2 
>  hw/ppc/spapr.c                    |   69 +++++
>  hw/ppc/spapr_drc.c                |   19 +
>  hw/ppc/spapr_events.c             |    4 
>  hw/ppc/spapr_nvdimm.c             |  475 +++++++++++++++++++++++++++++++++++++
>  include/hw/mem/nvdimm.h           |    7 +
>  include/hw/ppc/spapr.h            |    8 +
>  include/hw/ppc/spapr_drc.h        |    9 +
>  include/hw/ppc/spapr_nvdimm.h     |   37 +++
>  include/qemu/nvdimm-utils.h       |    7 +
>  util/Makefile.objs                |    1 
>  util/nvdimm-utils.c               |   29 ++
>  16 files changed, 699 insertions(+), 39 deletions(-)
>  create mode 100644 hw/ppc/spapr_nvdimm.c
>  create mode 100644 include/hw/ppc/spapr_nvdimm.h
>  create mode 100644 include/qemu/nvdimm-utils.h
>  create mode 100644 util/nvdimm-utils.c
> 

-- 
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson