> 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
