From: Jason Gunthorpe <jgg@ziepe.ca>
To: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Cc: davem@davemloft.net, gregkh@linuxfoundation.org,
Dave Ertman <david.m.ertman@intel.com>,
netdev@vger.kernel.org, linux-rdma@vger.kernel.org,
nhorman@redhat.com, sassmann@redhat.com,
ranjani.sridharan@linux.intel.com,
pierre-louis.bossart@linux.intel.com,
Tony Nguyen <anthony.l.nguyen@intel.com>,
Andrew Bowers <andrewx.bowers@intel.com>
Subject: Re: [net-next 2/9] ice: Create and register virtual bus for RDMA
Date: Fri, 17 Apr 2020 20:49:43 -0300 [thread overview]
Message-ID: <20200417234943.GM26002@ziepe.ca> (raw)
In-Reply-To: <20200417171034.1533253-3-jeffrey.t.kirsher@intel.com>
On Fri, Apr 17, 2020 at 10:10:27AM -0700, Jeff Kirsher wrote:
> +/**
> + * ice_peer_vdev_release - function to map to virtbus_devices release callback
> + * @vdev: pointer to virtbus_device to free
> + */
> +static void ice_peer_vdev_release(struct virtbus_device *vdev)
> +{
> + struct iidc_virtbus_object *vbo;
> +
> + vbo = container_of(vdev, struct iidc_virtbus_object, vdev);
> + kfree(vbo);
> +}
> +
> +/**
> + * ice_init_peer_devices - initializes peer devices
> + * @pf: ptr to ice_pf
> + *
> + * This function initializes peer devices on the virtual bus.
> + */
> +int ice_init_peer_devices(struct ice_pf *pf)
> +{
> + struct ice_vsi *vsi = pf->vsi[0];
> + struct pci_dev *pdev = pf->pdev;
> + struct device *dev = &pdev->dev;
> + int status = 0;
> + unsigned int i;
> +
> + /* Reserve vector resources */
> + status = ice_reserve_peer_qvector(pf);
> + if (status < 0) {
> + dev_err(dev, "failed to reserve vectors for peer drivers\n");
> + return status;
> + }
> + for (i = 0; i < ARRAY_SIZE(ice_peers); i++) {
> + struct ice_peer_dev_int *peer_dev_int;
> + struct ice_peer_drv_int *peer_drv_int;
> + struct iidc_qos_params *qos_info;
> + struct iidc_virtbus_object *vbo;
> + struct msix_entry *entry = NULL;
> + struct iidc_peer_dev *peer_dev;
> + struct virtbus_device *vdev;
> + int j;
> +
> + /* structure layout needed for container_of's looks like:
> + * ice_peer_dev_int (internal only ice peer superstruct)
> + * |--> iidc_peer_dev
> + * |--> *ice_peer_drv_int
> + *
> + * iidc_virtbus_object (container_of parent for vdev)
> + * |--> virtbus_device
> + * |--> *iidc_peer_dev (pointer from internal struct)
> + *
> + * ice_peer_drv_int (internal only peer_drv struct)
> + */
> + peer_dev_int = kzalloc(sizeof(*peer_dev_int), GFP_KERNEL);
> + if (!peer_dev_int)
> + return -ENOMEM;
> +
> + vbo = kzalloc(sizeof(*vbo), GFP_KERNEL);
> + if (!vbo) {
> + kfree(peer_dev_int);
> + return -ENOMEM;
> + }
> +
> + peer_drv_int = kzalloc(sizeof(*peer_drv_int), GFP_KERNEL);
> + if (!peer_drv_int) {
> + kfree(peer_dev_int);
> + kfree(vbo);
> + return -ENOMEM;
> + }
The lifetimes of all this memory look really suspect. The vbo holds a
pointer to the peer_dev but who ensures it it freed after all the vbo
kref's are released so there isn't a dangling pointer in
vbo->peer_dev?
One allocation is much simpler to understand:
struct iidc_virtbus_object {
struct virbus_device vdev;
[public members]
}
struct iidc_virtbus_object_private {
struct iidc_virtbus_object vobj;
[private members]
}
And just kzalloc a single iidc_virtbus_object_private
> + peer_dev->msix_entries = entry;
> + ice_peer_state_change(peer_dev_int, ICE_PEER_DEV_STATE_INIT,
> + false);
> +
> + vdev = &vbo->vdev;
> + vdev->name = ice_peers[i].name;
> + vdev->release = ice_peer_vdev_release;
> + vdev->dev.parent = &pdev->dev;
> +
> + status = virtbus_register_device(vdev);
> + if (status) {
> + kfree(peer_dev_int);
> + kfree(peer_drv_int);
> + vdev = NULL;
To me this feels very unnatural, virtbus_register_device() does the
kfree for the vbo if it fails so this function can't have a the normal
goto error unwind and ends up open coding the error unwinds in each if
above.
> +/* Following APIs are implemented by peer drivers and invoked by device
> + * owner
> + */
> +struct iidc_peer_ops {
> + void (*event_handler)(struct iidc_peer_dev *peer_dev,
> + struct iidc_event *event);
> +
> + /* Why we have 'open' and when it is expected to be called:
> + * 1. symmetric set of API w.r.t close
> + * 2. To be invoked form driver initialization path
> + * - call peer_driver:open once device owner is fully
> + * initialized
> + * 3. To be invoked upon RESET complete
> + */
> + int (*open)(struct iidc_peer_dev *peer_dev);
> +
> + /* Peer's close function is to be called when the peer needs to be
> + * quiesced. This can be for a variety of reasons (enumerated in the
> + * iidc_close_reason enum struct). A call to close will only be
> + * followed by a call to either remove or open. No IDC calls from the
> + * peer should be accepted until it is re-opened.
> + *
> + * The *reason* parameter is the reason for the call to close. This
> + * can be for any reason enumerated in the iidc_close_reason struct.
> + * It's primary reason is for the peer's bookkeeping and in case the
> + * peer want to perform any different tasks dictated by the reason.
> + */
> + void (*close)(struct iidc_peer_dev *peer_dev,
> + enum iidc_close_reason reason);
The open and close op looks really weird
Jason
next prev parent reply other threads:[~2020-04-17 23:49 UTC|newest]
Thread overview: 22+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-04-17 17:10 [net-next 0/9][pull request] 100GbE Intel Wired LAN Driver Updates 2020-04-17 Jeff Kirsher
2020-04-17 17:10 ` [net-next 1/9] Implementation of Virtual Bus Jeff Kirsher
2020-04-17 19:14 ` Pierre-Louis Bossart
2020-04-17 19:51 ` Jason Gunthorpe
2020-04-20 23:16 ` Ertman, David M
2020-04-21 0:44 ` Jason Gunthorpe
2020-04-21 23:27 ` Ertman, David M
2020-04-18 12:50 ` Greg KH
2020-04-20 22:59 ` Ertman, David M
2020-04-21 6:52 ` Greg KH
2020-04-20 23:46 ` Ranjani Sridharan
2020-04-17 17:10 ` [net-next 2/9] ice: Create and register virtual bus for RDMA Jeff Kirsher
2020-04-17 19:17 ` Jason Gunthorpe
2020-04-20 23:47 ` Kirsher, Jeffrey T
2020-04-17 23:49 ` Jason Gunthorpe [this message]
2020-04-17 17:10 ` [net-next 3/9] ice: Complete RDMA peer registration Jeff Kirsher
2020-04-17 17:10 ` [net-next 4/9] ice: Support resource allocation requests Jeff Kirsher
2020-04-17 17:10 ` [net-next 5/9] ice: Enable event notifications Jeff Kirsher
2020-04-17 17:10 ` [net-next 6/9] ice: Allow reset operations Jeff Kirsher
2020-04-17 17:10 ` [net-next 7/9] ice: Pass through communications to VF Jeff Kirsher
2020-04-17 17:10 ` [net-next 8/9] i40e: Move client header location Jeff Kirsher
2020-04-17 17:10 ` [net-next 9/9] i40e: Register a virtbus device to provide RDMA Jeff Kirsher
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=20200417234943.GM26002@ziepe.ca \
--to=jgg@ziepe.ca \
--cc=andrewx.bowers@intel.com \
--cc=anthony.l.nguyen@intel.com \
--cc=davem@davemloft.net \
--cc=david.m.ertman@intel.com \
--cc=gregkh@linuxfoundation.org \
--cc=jeffrey.t.kirsher@intel.com \
--cc=linux-rdma@vger.kernel.org \
--cc=netdev@vger.kernel.org \
--cc=nhorman@redhat.com \
--cc=pierre-louis.bossart@linux.intel.com \
--cc=ranjani.sridharan@linux.intel.com \
--cc=sassmann@redhat.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).