Re: [PATCH rdma-next v1 02/15] RDMA/core: Replace the ib_port_data hw_stats pointers with a ib_port pointer

[Leon Romanovsky]

On Mon, Jun 07, 2021 at 12:23:23PM +0200, Greg KH wrote:
> On Mon, Jun 07, 2021 at 11:17:27AM +0300, Leon Romanovsky wrote:
> > From: Jason Gunthorpe <jgg@xxxxxxxxxx>
> > 
> > It is much saner to store a pointer to the kobject structure that contains
> > the cannonical stats pointer than to copy the stats pointers into a public
> > structure.
> > 
> > Future patches will require the sysfs pointer for other purposes.
> > 
> > Signed-off-by: Jason Gunthorpe <jgg@xxxxxxxxxx>
> > Signed-off-by: Leon Romanovsky <leonro@xxxxxxxxxx>
> > ---
> >  drivers/infiniband/core/core_priv.h |  1 +
> >  drivers/infiniband/core/nldev.c     |  8 ++------
> >  drivers/infiniband/core/sysfs.c     | 14 +++++++++++---
> >  include/rdma/ib_verbs.h             |  3 ++-
> >  4 files changed, 16 insertions(+), 10 deletions(-)
> > 
> > diff --git a/drivers/infiniband/core/core_priv.h b/drivers/infiniband/core/core_priv.h
> > index 29809dd30041..ec5c2c3db423 100644
> > --- a/drivers/infiniband/core/core_priv.h
> > +++ b/drivers/infiniband/core/core_priv.h
> > @@ -378,6 +378,7 @@ struct net_device *rdma_read_gid_attr_ndev_rcu(const struct ib_gid_attr *attr);
> >  
> >  void ib_free_port_attrs(struct ib_core_device *coredev);
> >  int ib_setup_port_attrs(struct ib_core_device *coredev);
> > +struct rdma_hw_stats *ib_get_hw_stats_port(struct ib_device *ibdev, u32 port_num);
> >  
> >  int rdma_compatdev_set(u8 enable);
> >  
> > diff --git a/drivers/infiniband/core/nldev.c b/drivers/infiniband/core/nldev.c
> > index 01316926cef6..e9b4b2cccaa0 100644
> > --- a/drivers/infiniband/core/nldev.c
> > +++ b/drivers/infiniband/core/nldev.c
> > @@ -2066,7 +2066,8 @@ static int stat_get_doit_default_counter(struct sk_buff *skb,
> >  	}
> >  
> >  	port = nla_get_u32(tb[RDMA_NLDEV_ATTR_PORT_INDEX]);
> > -	if (!rdma_is_port_valid(device, port)) {
> > +	stats = ib_get_hw_stats_port(device, port);
> > +	if (!stats) {
> >  		ret = -EINVAL;
> >  		goto err;
> >  	}
> > @@ -2088,11 +2089,6 @@ static int stat_get_doit_default_counter(struct sk_buff *skb,
> >  		goto err_msg;
> >  	}
> >  
> > -	stats = device->port_data ? device->port_data[port].hw_stats : NULL;
> > -	if (stats == NULL) {
> > -		ret = -EINVAL;
> > -		goto err_msg;
> > -	}
> >  	mutex_lock(&stats->lock);
> >  
> >  	num_cnts = device->ops.get_hw_stats(device, stats, port, 0);
> > diff --git a/drivers/infiniband/core/sysfs.c b/drivers/infiniband/core/sysfs.c
> > index d11ceff2b4e4..b153dee1e0fa 100644
> > --- a/drivers/infiniband/core/sysfs.c
> > +++ b/drivers/infiniband/core/sysfs.c
> > @@ -1031,8 +1031,6 @@ static void setup_hw_stats(struct ib_device *device, struct ib_port *port,
> >  			goto err;
> >  		port->hw_stats_ag = hsag;
> >  		port->hw_stats = stats;
> > -		if (device->port_data)
> > -			device->port_data[port_num].hw_stats = stats;
> >  	} else {
> >  		struct kobject *kobj = &device->dev.kobj;
> >  		ret = sysfs_create_group(kobj, hsag);
> > @@ -1053,6 +1051,14 @@ static void setup_hw_stats(struct ib_device *device, struct ib_port *port,
> >  	kfree(stats);
> >  }
> >  
> > +struct rdma_hw_stats *ib_get_hw_stats_port(struct ib_device *ibdev,
> > +					   u32 port_num)
> > +{
> > +	if (!ibdev->port_data || !rdma_is_port_valid(ibdev, port_num))
> > +		return NULL;
> > +	return ibdev->port_data[port_num].sysfs->hw_stats;
> > +}
> > +
> >  static int add_port(struct ib_core_device *coredev, int port_num)
> >  {
> >  	struct ib_device *device = rdma_device_to_ibdev(&coredev->dev);
> > @@ -1171,6 +1177,8 @@ static int add_port(struct ib_core_device *coredev, int port_num)
> >  		setup_hw_stats(device, p, port_num);
> >  
> >  	list_add_tail(&p->kobj.entry, &coredev->port_list);
> > +	if (device->port_data && is_full_dev)
> > +		device->port_data[port_num].sysfs = p;
> 
> You are saving off a pointer to a reference counted structure without
> incrementing the reference count on it?  That's brave, and really wrong.

This is done to ensure that device->port_data[port_num].sysfs is not set
before "p" is properly configured. From reference point of view  "sysfs"
is equal to "p" and the latter already initialized.

Thanks