| Message ID | 20230307113621.64153-1-gautam.dawar@amd.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | sfc: add vDPA support for EF100 devices | expand |
On Tue, Mar 7, 2023 at 7:36 PM Gautam Dawar <gautam.dawar@amd.com> wrote: > > Hi All, > > This series adds the vdpa support for EF100 devices. Would you mind posting some performance numbers for this device? Thanks > For now, only a network class of vdpa device is supported and > they can be created only on a VF. Each EF100 VF can have one > of the three function personalities (EF100, vDPA & None) at > any time with EF100 being the default. A VF's function personality > is changed to vDPA while creating the vdpa device using vdpa tool. > > A vDPA management device is created per VF to allow selection of > the desired VF for vDPA device creation. The MAC address for the > target net device must be set either by specifying at the vdpa > device creation time via the `mac` parameter of the `vdpa dev add` > command or should be specified as the hardware address of the virtual > function using `devlink port function set hw_addr` command before > creating the vdpa device with the former taking precedence. > > Changes since v1: > > - To ensure isolation between DMA initiated by userspace (guest OS) > and the host MCDI buffer, ummap VF's MCDI DMA buffer and use PF's > IOMMU domain instead for executing vDPA VF's MCDI commands. > - As a result of above change, it is no more necessary to check for > MCDI buffer's IOVA range overlap with the guest buffers. Accordingly, > the DMA config operations and the rbtree/list implementation to store > IOVA mappings have been dropped. > - Support vDPA only if running Firmware supports CLIENT_CMD_VF_PROXY > capability. > - Added .suspend config operation and updated get_vq_state/set_vq_state > to support Live Migration. Also, features VIRTIO_F_ORDER_PLATFORM and > VIRTIO_F_IN_ORDER have been masked off in get_device_features() to > allow Live Migration as QEMU SVQ doesn't support them yet. > - Removed the minimum version (v6.1.0) requirement of QEMU as > VIRTIO_F_IN_ORDER is not exposed > - Fetch the vdpa device MAC address from the underlying VF hw_addr (if > set via `devlink port function set hw_addr` command) > - Removed the mandatory requirement of specifying mac address while > creating vdpa device > - Moved create_vring_ctx() and get_doorbell_offset() in dev_add() > - Moved IRQ allocation at the time of vring creation > - Merged vring_created member of struct ef100_vdpa_vring_info as one > of the flags in vring_state > - Simplified .set_status() implementation > - Removed un-necessary vdpa_state checks against > EF100_VDPA_STATE_INITIALIZED > - Removed userspace triggerable warning in kick_vq() > - Updated year 2023 in copyright banner of new files > > Gautam Dawar (14): > sfc: add function personality support for EF100 devices > sfc: implement MCDI interface for vDPA operations > sfc: update MCDI headers for CLIENT_CMD_VF_PROXY capability bit > sfc: evaluate vdpa support based on FW capability CLIENT_CMD_VF_PROXY > sfc: implement init and fini functions for vDPA personality > sfc: implement vDPA management device operations > sfc: implement vdpa device config operations > sfc: implement vdpa vring config operations > sfc: implement device status related vdpa config operations > sfc: implement filters for receiving traffic > sfc: use PF's IOMMU domain for running VF's MCDI commands > sfc: unmap VF's MCDI buffer when switching to vDPA mode > sfc: update vdpa device MAC address > sfc: register the vDPA device > > drivers/net/ethernet/sfc/Kconfig | 8 + > drivers/net/ethernet/sfc/Makefile | 1 + > drivers/net/ethernet/sfc/ef10.c | 2 +- > drivers/net/ethernet/sfc/ef100.c | 7 +- > drivers/net/ethernet/sfc/ef100_netdev.c | 26 +- > drivers/net/ethernet/sfc/ef100_nic.c | 183 +- > drivers/net/ethernet/sfc/ef100_nic.h | 26 +- > drivers/net/ethernet/sfc/ef100_vdpa.c | 543 +++ > drivers/net/ethernet/sfc/ef100_vdpa.h | 224 ++ > drivers/net/ethernet/sfc/ef100_vdpa_ops.c | 793 ++++ > drivers/net/ethernet/sfc/mcdi.c | 108 +- > drivers/net/ethernet/sfc/mcdi.h | 9 +- > drivers/net/ethernet/sfc/mcdi_filters.c | 51 +- > drivers/net/ethernet/sfc/mcdi_functions.c | 9 +- > drivers/net/ethernet/sfc/mcdi_functions.h | 3 +- > drivers/net/ethernet/sfc/mcdi_pcol.h | 4390 ++++++++++++++++++++- > drivers/net/ethernet/sfc/mcdi_vdpa.c | 259 ++ > drivers/net/ethernet/sfc/mcdi_vdpa.h | 83 + > drivers/net/ethernet/sfc/net_driver.h | 21 + > drivers/net/ethernet/sfc/ptp.c | 4 +- > 20 files changed, 6574 insertions(+), 176 deletions(-) > create mode 100644 drivers/net/ethernet/sfc/ef100_vdpa.c > create mode 100644 drivers/net/ethernet/sfc/ef100_vdpa.h > create mode 100644 drivers/net/ethernet/sfc/ef100_vdpa_ops.c > create mode 100644 drivers/net/ethernet/sfc/mcdi_vdpa.c > create mode 100644 drivers/net/ethernet/sfc/mcdi_vdpa.h > > -- > 2.30.1 >
On 3/10/23 10:39, Jason Wang wrote: > Caution: This message originated from an External Source. Use proper caution when opening attachments, clicking links, or responding. > > > On Tue, Mar 7, 2023 at 7:36 PM Gautam Dawar <gautam.dawar@amd.com> wrote: >> Hi All, >> >> This series adds the vdpa support for EF100 devices. > Would you mind posting some performance numbers for this device? Sure, will do with the next version of this patch series. Thanks > Thanks > >> For now, only a network class of vdpa device is supported and >> they can be created only on a VF. Each EF100 VF can have one >> of the three function personalities (EF100, vDPA & None) at >> any time with EF100 being the default. A VF's function personality >> is changed to vDPA while creating the vdpa device using vdpa tool. >> >> A vDPA management device is created per VF to allow selection of >> the desired VF for vDPA device creation. The MAC address for the >> target net device must be set either by specifying at the vdpa >> device creation time via the `mac` parameter of the `vdpa dev add` >> command or should be specified as the hardware address of the virtual >> function using `devlink port function set hw_addr` command before >> creating the vdpa device with the former taking precedence. >> >> Changes since v1: >> >> - To ensure isolation between DMA initiated by userspace (guest OS) >> and the host MCDI buffer, ummap VF's MCDI DMA buffer and use PF's >> IOMMU domain instead for executing vDPA VF's MCDI commands. >> - As a result of above change, it is no more necessary to check for >> MCDI buffer's IOVA range overlap with the guest buffers. Accordingly, >> the DMA config operations and the rbtree/list implementation to store >> IOVA mappings have been dropped. >> - Support vDPA only if running Firmware supports CLIENT_CMD_VF_PROXY >> capability. >> - Added .suspend config operation and updated get_vq_state/set_vq_state >> to support Live Migration. Also, features VIRTIO_F_ORDER_PLATFORM and >> VIRTIO_F_IN_ORDER have been masked off in get_device_features() to >> allow Live Migration as QEMU SVQ doesn't support them yet. >> - Removed the minimum version (v6.1.0) requirement of QEMU as >> VIRTIO_F_IN_ORDER is not exposed >> - Fetch the vdpa device MAC address from the underlying VF hw_addr (if >> set via `devlink port function set hw_addr` command) >> - Removed the mandatory requirement of specifying mac address while >> creating vdpa device >> - Moved create_vring_ctx() and get_doorbell_offset() in dev_add() >> - Moved IRQ allocation at the time of vring creation >> - Merged vring_created member of struct ef100_vdpa_vring_info as one >> of the flags in vring_state >> - Simplified .set_status() implementation >> - Removed un-necessary vdpa_state checks against >> EF100_VDPA_STATE_INITIALIZED >> - Removed userspace triggerable warning in kick_vq() >> - Updated year 2023 in copyright banner of new files >> >> Gautam Dawar (14): >> sfc: add function personality support for EF100 devices >> sfc: implement MCDI interface for vDPA operations >> sfc: update MCDI headers for CLIENT_CMD_VF_PROXY capability bit >> sfc: evaluate vdpa support based on FW capability CLIENT_CMD_VF_PROXY >> sfc: implement init and fini functions for vDPA personality >> sfc: implement vDPA management device operations >> sfc: implement vdpa device config operations >> sfc: implement vdpa vring config operations >> sfc: implement device status related vdpa config operations >> sfc: implement filters for receiving traffic >> sfc: use PF's IOMMU domain for running VF's MCDI commands >> sfc: unmap VF's MCDI buffer when switching to vDPA mode >> sfc: update vdpa device MAC address >> sfc: register the vDPA device >> >> drivers/net/ethernet/sfc/Kconfig | 8 + >> drivers/net/ethernet/sfc/Makefile | 1 + >> drivers/net/ethernet/sfc/ef10.c | 2 +- >> drivers/net/ethernet/sfc/ef100.c | 7 +- >> drivers/net/ethernet/sfc/ef100_netdev.c | 26 +- >> drivers/net/ethernet/sfc/ef100_nic.c | 183 +- >> drivers/net/ethernet/sfc/ef100_nic.h | 26 +- >> drivers/net/ethernet/sfc/ef100_vdpa.c | 543 +++ >> drivers/net/ethernet/sfc/ef100_vdpa.h | 224 ++ >> drivers/net/ethernet/sfc/ef100_vdpa_ops.c | 793 ++++ >> drivers/net/ethernet/sfc/mcdi.c | 108 +- >> drivers/net/ethernet/sfc/mcdi.h | 9 +- >> drivers/net/ethernet/sfc/mcdi_filters.c | 51 +- >> drivers/net/ethernet/sfc/mcdi_functions.c | 9 +- >> drivers/net/ethernet/sfc/mcdi_functions.h | 3 +- >> drivers/net/ethernet/sfc/mcdi_pcol.h | 4390 ++++++++++++++++++++- >> drivers/net/ethernet/sfc/mcdi_vdpa.c | 259 ++ >> drivers/net/ethernet/sfc/mcdi_vdpa.h | 83 + >> drivers/net/ethernet/sfc/net_driver.h | 21 + >> drivers/net/ethernet/sfc/ptp.c | 4 +- >> 20 files changed, 6574 insertions(+), 176 deletions(-) >> create mode 100644 drivers/net/ethernet/sfc/ef100_vdpa.c >> create mode 100644 drivers/net/ethernet/sfc/ef100_vdpa.h >> create mode 100644 drivers/net/ethernet/sfc/ef100_vdpa_ops.c >> create mode 100644 drivers/net/ethernet/sfc/mcdi_vdpa.c >> create mode 100644 drivers/net/ethernet/sfc/mcdi_vdpa.h >> >> -- >> 2.30.1 >>