RE: drivers/vdpa/vdpa.c:296 vdpa_mgmtdev_fill() warn: should '(((1))) << (mdev->id_table[i]->device)' be a 64 bit type?

[Parav Pandit <parav@nvidia.com>]

Hi Joe,

> From: Joe Perches <joe@perches.com>
> Sent: Thursday, November 25, 2021 10:23 PM
> 
> On Thu, 2021-11-25 at 11:15 -0500, Michael S. Tsirkin wrote:
> > On Thu, Nov 25, 2021 at 10:03:28AM +0300, Dan Carpenter wrote:
> > > tree:   https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
> master
> > > head:   5d9f4cf36721aba199975a9be7863a3ff5cd4b59
> > > commit: 33b347503f014ebf76257327cbc7001c6b721956 vdpa: Define vdpa
> > > mgmt device, ops and a netlink interface
> > > config: i386-randconfig-m021-20211118
> > > (https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fd
> > > ownload.01.org%2F0day-
> ci%2Farchive%2F20211125%2F202111250317.VUsGtdo
> > > x-
> lkp%40intel.com%2Fconfig&amp;data=04%7C01%7Cparav%40nvidia.com%7Cf
> > >
> 4d5aa1e5dfe431788f908d9b03408a1%7C43083d15727340c1b7db39efd9ccc17
> a%7
> > >
> C0%7C0%7C637734559803525095%7CUnknown%7CTWFpbGZsb3d8eyJWIjoi
> MC4wLjAw
> > >
> MDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&amp;sda
> ta=
> > >
> 0XNeoKTYFnhqSH7Zc3MEVYoz00Ox2XAwrRSShA9XpdM%3D&amp;reserved=0
> )
> > > compiler: gcc-9 (Debian 9.3.0-22) 9.3.0
> > >
> > > If you fix the issue, kindly add following tag as appropriate
> > > Reported-by: kernel test robot <lkp@intel.com>
> > > Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
> > >
> > > smatch warnings:
> > > drivers/vdpa/vdpa.c:296 vdpa_mgmtdev_fill() warn: should '(((1))) <<
> (mdev->id_table[i]->device)' be a 64 bit type?
> > >
> > > vim +296 drivers/vdpa/vdpa.c
> > >
> > > 33b347503f014eb Parav Pandit 2021-01-05  280  static int
> vdpa_mgmtdev_fill(const struct vdpa_mgmt_dev *mdev, struct sk_buff *msg,
> > > 33b347503f014eb Parav Pandit 2021-01-05  281  			     u32
> portid, u32 seq, int flags)
> > > 33b347503f014eb Parav Pandit 2021-01-05  282  {
> > > 33b347503f014eb Parav Pandit 2021-01-05  283  	u64
> supported_classes = 0;
> > >
> > > ^^^^^^^^^^^^^^^^^^^^^ This is a u64
> > >
> > > 33b347503f014eb Parav Pandit 2021-01-05  284  	void *hdr;
> > > 33b347503f014eb Parav Pandit 2021-01-05  285  	int i = 0;
> > > 33b347503f014eb Parav Pandit 2021-01-05  286  	int err;
> > > 33b347503f014eb Parav Pandit 2021-01-05  287
> > > 33b347503f014eb Parav Pandit 2021-01-05  288  	hdr =
> genlmsg_put(msg, portid, seq, &vdpa_nl_family, flags,
> VDPA_CMD_MGMTDEV_NEW);
> > > 33b347503f014eb Parav Pandit 2021-01-05  289  	if (!hdr)
> > > 33b347503f014eb Parav Pandit 2021-01-05  290  		return -
> EMSGSIZE;
> > > 33b347503f014eb Parav Pandit 2021-01-05  291  	err =
> vdpa_nl_mgmtdev_handle_fill(msg, mdev);
> > > 33b347503f014eb Parav Pandit 2021-01-05  292  	if (err)
> > > 33b347503f014eb Parav Pandit 2021-01-05  293  		goto msg_err;
> > > 33b347503f014eb Parav Pandit 2021-01-05  294
> > > 33b347503f014eb Parav Pandit 2021-01-05  295  	while (mdev-
> >id_table[i].device) {
> > > 33b347503f014eb Parav Pandit 2021-01-05 @296
> 	supported_classes |= BIT(mdev->id_table[i].device);
> > >
> > > Which suggests that this should be BIT_ULL().
>
I will respond to this part in reply to Michael.
 
> True but it seems all of these tables have only 1 or at most 3 non-null
> terminated entries.
> 
> It looks as if mdev->id_table itself could be NULL though:
> 
> drivers/vdpa/virtio_pci/vp_vdpa.c:      .id_table       = NULL, /* only dynamic ids
> */
> 
Below is not possible, because above id_table is for the PCI driver table and above id_table is per mgmt device for id table type struct virtio_device_id.

> which _might_ cause a null dereference.  Dunno if that's possible.
>