Re: [PATCH net-next] net: mana: Add new device attributes for mana

[Jason Gunthorpe]

On Wed, Apr 17, 2024 at 11:01:08PM -0700, Shradha Gupta wrote:

> > > > > +static ssize_t mana_attr_show(struct device *dev,
> > > > > +			      struct device_attribute *attr, char *buf)
> > > > > +{
> > > > > +	struct pci_dev *pdev = to_pci_dev(dev);
> > > > > +	struct gdma_context *gc = pci_get_drvdata(pdev);
> > > > > +	struct mana_context *ac = gc->mana.driver_data;
> > > > > +
> > > > > +	if (strcmp(attr->attr.name, "mport") == 0)
> > > > > +		return snprintf(buf, PAGE_SIZE, "%d\n", ac->num_ports);
> > > > > +	else if (strcmp(attr->attr.name, "adapter_mtu") == 0)
> > > > > +		return snprintf(buf, PAGE_SIZE, "%d\n", gc->adapter_mtu);
> > > > > +	else if (strcmp(attr->attr.name, "msix") == 0)
> > > > > +		return snprintf(buf, PAGE_SIZE, "%d\n", gc->max_num_msix);
> > > > > +	else
> > > > > +		return -EINVAL;
> > > > > +
> > > > 
> > > > That is not how sysfs should be implemented at all, please find a
> > > > good example to copy from. Every attribute should use its own function
> > > > with the macros to link it into an attributes group and sysfs_emit
> > > > should be used for printing
> > > 
> > > Not sure if this file drivers/infiniband/hw/usnic/usnic_ib_sysfs.c is a good
> > > example or not.
> > 
> > The first question should be, what are these values used for? You can
> > then decide on debugfs or sysfs. debugfs is easier to use, and you
> > avoid any ABI, which will make long term support easier.
> 
> Hi Andrew,
> We want to eventually use these attributes to make the device settings configurable
> and also improve debuggability for MANA devices. I feel having these attributes 
> in sysfs would make more sense as we plan to extend the attribute list and also make
> them settable.

>From an RDMA perspective this is all available from other APIs already
at least and I wouldn't want to see new sysfs unless there is a netdev
justification.

Jason