Message-ID: <a76853a97dda4ccb96c35d4095e4866a@intel.com>
Date:   Tue, 6 Apr 2021 21:05:10 +0000
From:   "Saleem, Shiraz" <shiraz.saleem@...el.com>
To:     "dledford@...hat.com" <dledford@...hat.com>,
        "jgg@...dia.com" <jgg@...dia.com>,
        "kuba@...nel.org" <kuba@...nel.org>,
        "davem@...emloft.net" <davem@...emloft.net>
CC:     "linux-rdma@...r.kernel.org" <linux-rdma@...r.kernel.org>,
        "netdev@...r.kernel.org" <netdev@...r.kernel.org>,
        "Ertman, David M" <david.m.ertman@...el.com>,
        "Nguyen, Anthony L" <anthony.l.nguyen@...el.com>
Subject: RE: [PATCH v4 00/23] Add Intel Ethernet Protocol Driver for RDMA
 (irdma)

> Subject: [PATCH v4 00/23] Add Intel Ethernet Protocol Driver for RDMA (irdma)
> 
> The following patch series introduces a unified Intel Ethernet Protocol Driver for
> RDMA (irdma) for the X722 iWARP device and a new E810 device which supports
> iWARP and RoCEv2. The irdma module replaces the legacy i40iw module for X722
> and extends the ABI already defined for i40iw. It is backward compatible with
> legacy X722 rdma-core provider (libi40iw).
> 
> X722 and E810 are PCI network devices that are RDMA capable. The RDMA block
> of this parent device is represented via an auxiliary device exported to 'irdma'
> using the core auxiliary bus infrastructure recently added for 5.11 kernel.
> The parent PCI netdev drivers 'i40e' and 'ice' register auxiliary RDMA devices with
> private data/ops encapsulated that bind to auxiliary drivers registered in irdma
> module.
> 
> This patchset was initially submitted as an RFC where in we got feedback to come
> up with a generic scheme for RDMA drivers to attach to a PCI device owned by
> netdev PCI driver [1]. Solutions using platform bus and MFD were explored but
> rejected by the community and the consensus was to add a new bus infrastructure
> to support this usage model.
> 
> Further revisions of this series along with the auxiliary bus were submitted [2]. At
> this point, Greg KH requested that we take the auxiliary bus review and revision
> process to an internal mailing list and garner the buy-in of a respected kernel
> contributor, along with consensus of all major stakeholders including Nvidia (for
> mlx5 sub-function use-case) and Intel sound driver. This process took a while and
> stalled further development/review of this netdev/irdma series.
> The auxiliary bus was eventually merged in 5.11.
> 
> Between v1 to v2 of this submission, the IIDC went through a major re-write based
> on the feedback and we hope it is now more in alignment with what the community
> wants.
> 
> This series is built against rdma for-next and currently includes the netdev patches
> for ease of review. This includes updates to 'ice' driver to provide RDMA support
> and converts 'i40e' driver to use the auxiliary bus infrastructure.
> Once the patches are closer to merging, a shared pull request will be submitted.
> 
> v3-->v4:
> * Fixup W=1 warnings in ice patches
> * Fix issues uncovered by pyverbs for create user AH and multicast
> * Fix descriptor set issue for fast register introduced during port to FIELD_PREP
> in v2 submission
> 
> v2-->v3:
> * rebase rdma for-next. Adapt to core change '1fb7f8973f51 ("RDMA: Support
> more than 255 rdma ports")'
> * irdma Kconfig updates to conform with linux coding style.
> * Fix a 0-day build issue
> * Remove rdma_resource_limits selector devlink param. Follow on patch to be
> submitted for it with suggestion from Parav to use devlink resource.
> * Capitalize abbreviations in ice idc. e.g. 'aux' to 'AUX'
> 
> v1-->v2:
> * Remove IIDC channel OPs - open, close, peer_register and peer_unregister.
>   And all its associated FSM in ice PCI core driver.
> * Use device_lock in ice PCI core driver while issuing IIDC ops callbacks.
> * Remove peer_* verbiage from shared IIDC header and rename the structs and
> channel ops
>   with iidc_core*/iidc_auxiliary*.
> * Allocate ib_device at start and register it at the end of drv.probe() in irdma gen2
> auxiliary driver.
> * Use ibdev_* printing extensively throughout most of the driver
>   Remove idev_to_dev, ihw_to_dev macros as no longer required in new print
> scheme.
> * Do not bump ABI ver. to 6 in irdma. Maintain irdma ABI ver. at 5 for legacy i40iw
> user-provider compatibility.
> * Add a boundary check in irdma_alloc_ucontext to fail binding with < 4 user-space
> provider version.
> * Remove devlink from irdma. Add 2 new rdma-related devlink parameters added
> to ice PCI core driver.
> * Use FIELD_PREP/FIELD_GET/GENMASK on get/set of descriptor fields versus
> home grown ones LS_*/RS_*.
> * Bind 2 separate auxiliary drivers in irdma - one for gen1 and one for gen2 and
> future devices.
> * Misc. driver fixes in irdma
> 
> [1] https://patchwork.kernel.org/project/linux-rdma/patch/20190215171107.6464-2-
> shiraz.saleem@...el.com/
> [2] https://lore.kernel.org/linux-rdma/20200520070415.3392210-1-
> jeffrey.t.kirsher@...el.com/
> 

The irdma rdma-core provider will be send out shortly.

Shiraz

Powered by blists - more mailing lists