From: "Saleem, Shiraz" <shiraz.saleem@intel.com>
To: Jason Gunthorpe <jgg@nvidia.com>
Cc: "dledford@redhat.com" <dledford@redhat.com>,
"kuba@kernel.org" <kuba@kernel.org>,
"davem@davemloft.net" <davem@davemloft.net>,
"linux-rdma@vger.kernel.org" <linux-rdma@vger.kernel.org>,
"netdev@vger.kernel.org" <netdev@vger.kernel.org>,
"Ertman, David M" <david.m.ertman@intel.com>,
"Nguyen, Anthony L" <anthony.l.nguyen@intel.com>,
"Williams, Dan J" <dan.j.williams@intel.com>,
"Hefty, Sean" <sean.hefty@intel.com>
Subject: RE: [PATCH v4 resend 01/23] iidc: Introduce iidc.h
Date: Mon, 12 Apr 2021 14:51:18 +0000 [thread overview]
Message-ID: <2ee289f620154810921df2bc2c903192@intel.com> (raw)
In-Reply-To: <20210407180529.GA547588@nvidia.com>
> Subject: Re: [PATCH v4 resend 01/23] iidc: Introduce iidc.h
>
> On Tue, Apr 06, 2021 at 07:14:40PM -0500, Shiraz Saleem wrote:
> > +/* Structure representing auxiliary driver tailored information about
> > +the core
> > + * PCI dev, each auxiliary driver using the IIDC interface will have
> > +an
> > + * instance of this struct dedicated to it.
> > + */
> > +struct iidc_core_dev_info {
> > + struct pci_dev *pdev; /* PCI device of corresponding to main function */
> > + struct auxiliary_device *adev;
> > + /* KVA / Linear address corresponding to BAR0 of underlying
> > + * pci_device.
> > + */
> > + u8 __iomem *hw_addr;
> > + int cdev_info_id;
> > +
> > + u8 ftype; /* PF(false) or VF (true) */
>
> Where is ftype initialized?
Today it is just pf. But the upcoming Intel ethernet VF driver will set it to true.
>
> > + u16 vport_id;
> > + enum iidc_rdma_protocol rdma_protocol;
>
> This duplicates the aux device name, not really sure why it is needed. One user just
> uses it to make the string, the rest is entangled with the devlink and doesn't need
> to be stored here.
It is used to pass the type of protocol at drv.probe() in aux RDMA driver.
>
> > + struct iidc_qos_params qos_info;
> > + struct net_device *netdev;
> > +
> > + struct msix_entry *msix_entries;
> > + u16 msix_count; /* How many vectors are reserved for this device */
> > +
> > + /* Following struct contains function pointers to be initialized
> > + * by core PCI driver and called by auxiliary driver
> > + */
> > + const struct iidc_core_ops *ops;
> > +};
>
> I spent a while trying to understand this struct and why it exists and..
>
> > +
> > +struct iidc_auxiliary_dev {
> > + struct auxiliary_device adev;
> > + struct iidc_core_dev_info *cdev_info;
>
> This cdev_info should just be a 'struct ice_pf *' and the "struct iidc_core_dev_info"
> should be deleted entirely. You'll notice this ends up looking nearly exactly like
> mlx5 does after this.
It was intentionally designed to be core device object carving out only pieces of the PF
information required by the rdma driver. The next near-term PCI driver using IIDC can also
this. Why expose the whole PF? This is a design choice no? Why do we need to follow mlx5?
>
> You can see it clearly based on how this gets initialized:
>
> cdev_info->pdev = pf->pdev;
> cdev_info->hw_addr = (u8 __iomem *)pf->hw.hw_addr;
>
> struct ice_vsi *vsi = ice_get_main_vsi(pf);
> cdev_info->vport_id = vsi->vsi_num;
> cdev_info->netdev = vsi->netdev;
> cdev_info->msix_count = pf->num_rdma_msix;
> cdev_info->msix_entries = &pf->msix_entries[pf-
> >rdma_base_vector];
>
> ice_setup_dcb_qos_info(pf, cdev_info->qos_info);
>
> Since the main place this cdev_info appears is in the ops API between the two
> modules, it looks to me like this is being designed in this obfuscated way to try
> and create a stable ABI between two modules.
>
> Remove all the stable module ABI hackery before you resend it.
>
I don't follow what the hackery is. Just because we use cdev_info in the .ops callbacks as opposed to ice_pf?
This is a private interface for Intel RDMA/PCI drivers and yes it is designed to be forward
looking especially since when we have near-term plans to use it.
Can you explain what you mean by stable module ABI hackery?
Shiraz
next prev parent reply other threads:[~2021-04-12 14:51 UTC|newest]
Thread overview: 31+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-04-07 0:14 [PATCH v4 resend 00/23] Add Intel Ethernet Protocol Driver for RDMA (irdma) Shiraz Saleem
2021-04-07 0:14 ` [PATCH v4 resend 01/23] iidc: Introduce iidc.h Shiraz Saleem
2021-04-07 18:05 ` Jason Gunthorpe
2021-04-12 14:51 ` Saleem, Shiraz [this message]
2021-04-12 16:12 ` Jason Gunthorpe
2021-04-07 0:14 ` [PATCH v4 resend 02/23] ice: Initialize RDMA support Shiraz Saleem
2021-04-07 0:14 ` [PATCH v4 resend 03/23] ice: Implement iidc operations Shiraz Saleem
2021-04-07 0:14 ` [PATCH v4 resend 04/23] ice: Register auxiliary device to provide RDMA Shiraz Saleem
2021-04-07 17:45 ` Jason Gunthorpe
2021-04-12 14:51 ` Saleem, Shiraz
2021-04-07 0:14 ` [PATCH v4 resend 05/23] ice: Add devlink params support Shiraz Saleem
2021-04-07 0:14 ` [PATCH v4 resend 06/23] i40e: Prep i40e header for aux bus conversion Shiraz Saleem
2021-04-07 0:14 ` [PATCH v4 resend 07/23] i40e: Register auxiliary devices to provide RDMA Shiraz Saleem
2021-04-07 0:14 ` [PATCH v4 resend 08/23] RDMA/irdma: Register auxiliary driver and implement private channel OPs Shiraz Saleem
2021-04-07 0:14 ` [PATCH v4 resend 09/23] RDMA/irdma: Implement device initialization definitions Shiraz Saleem
2021-04-07 0:14 ` [PATCH v4 resend 10/23] RDMA/irdma: Implement HW Admin Queue OPs Shiraz Saleem
2021-04-07 0:14 ` [PATCH v4 resend 11/23] RDMA/irdma: Add HMC backing store setup functions Shiraz Saleem
2021-04-07 0:14 ` [PATCH v4 resend 12/23] RDMA/irdma: Add privileged UDA queue implementation Shiraz Saleem
2021-04-07 0:14 ` [PATCH v4 resend 13/23] RDMA/irdma: Add QoS definitions Shiraz Saleem
2021-04-07 0:14 ` [PATCH v4 resend 14/23] RDMA/irdma: Add connection manager Shiraz Saleem
2021-04-07 0:14 ` [PATCH v4 resend 15/23] RDMA/irdma: Add PBLE resource manager Shiraz Saleem
2021-04-07 0:14 ` [PATCH v4 resend 16/23] RDMA/irdma: Implement device supported verb APIs Shiraz Saleem
2021-04-07 0:14 ` [PATCH v4 resend 17/23] RDMA/irdma: Add RoCEv2 UD OP support Shiraz Saleem
2021-04-07 0:14 ` [PATCH v4 resend 18/23] RDMA/irdma: Add user/kernel shared libraries Shiraz Saleem
2021-04-07 0:14 ` [PATCH v4 resend 19/23] RDMA/irdma: Add miscellaneous utility definitions Shiraz Saleem
2021-04-07 0:14 ` [PATCH v4 resend 20/23] RDMA/irdma: Add dynamic tracing for CM Shiraz Saleem
2021-04-07 0:15 ` [PATCH v4 resend 21/23] RDMA/irdma: Add ABI definitions Shiraz Saleem
2021-04-07 15:23 ` Jason Gunthorpe
2021-04-07 20:57 ` Saleem, Shiraz
2021-04-07 0:15 ` [PATCH v4 resend 22/23] RDMA/irdma: Add irdma Kconfig/Makefile and remove i40iw Shiraz Saleem
2021-04-07 0:15 ` [PATCH v4 resend 23/23] RDMA/irdma: Update MAINTAINERS file Shiraz Saleem
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=2ee289f620154810921df2bc2c903192@intel.com \
--to=shiraz.saleem@intel.com \
--cc=anthony.l.nguyen@intel.com \
--cc=dan.j.williams@intel.com \
--cc=davem@davemloft.net \
--cc=david.m.ertman@intel.com \
--cc=dledford@redhat.com \
--cc=jgg@nvidia.com \
--cc=kuba@kernel.org \
--cc=linux-rdma@vger.kernel.org \
--cc=netdev@vger.kernel.org \
--cc=sean.hefty@intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).