Re: [PATCH v9 4/7] nfc: pn533: Split pn533 init & nfc_register

[Jakub Kicinski <jakub.kicinski@netronome.com>]

On Tue, 15 Oct 2019 11:51:24 +0200, Lars Poeschel wrote:
> > > -	priv->nfc_dev = nfc_allocate_device(&pn533_nfc_ops, protocols,
> > > -					   priv->ops->tx_header_len +
> > > -					   PN533_CMD_DATAEXCH_HEAD_LEN,
> > > -					   priv->ops->tx_tail_len);
> > > -	if (!priv->nfc_dev) {
> > > -		rc = -ENOMEM;
> > > -		goto destroy_wq;
> > > -	}
> > > -
> > > -	nfc_set_parent_dev(priv->nfc_dev, parent);
> > > -	nfc_set_drvdata(priv->nfc_dev, priv);
> > > -
> > > -	rc = nfc_register_device(priv->nfc_dev);
> > > -	if (rc)
> > > -		goto free_nfc_dev;  
> > 
> > Aren't you moving too much out of here? Looking at commit 32ecc75ded72
> > ("NFC: pn533: change order operations in dev registation") it seems like
> > IRQ handler may want to access the data structures, do this change not
> > reintroduce the problem?  
> 
> Yes, you are right, there could be a problem if an irq gets served
> before the driver is registered to the nfc subsystem.
> Well, but the purpose of this patch is exactly that: Prevent use of nfc
> subsystem before the chip is fully initialized.
> To address this, I would not change the part above, but move the
> request_threaded_irq to the very bottom in pn533_i2c_probe, after the
> call to pn53x_register_nfc. So it is not possible to use nfc before the
> chip is initialized and irqs don't get served before the driver is
> registered to nfc subsystem.
> Thank you for this!
> I will include this in v10 of the patchset.

You can run nfc_allocate_device() etc. early, then allocate the IRQ,
and then run nfc_register_device(), would that work? Is that what you
have in mind?