Re: [net PATCH v2] octeontx2-af: Move validation of ptp pointer before its usage

[Sai Krishna Gajula <saikrishnag@marvell.com>]

> -----Original Message-----
> From: Dan Carpenter <dan.carpenter@linaro.org>
> Sent: Friday, June 9, 2023 6:48 PM
> To: Sai Krishna Gajula <saikrishnag@marvell.com>
> Cc: davem@davemloft.net; edumazet@google.com; kuba@kernel.org;
> pabeni@redhat.com; netdev@vger.kernel.org; linux-kernel@vger.kernel.org;
> Sunil Kovvuri Goutham <sgoutham@marvell.com>;
> maciej.fijalkowski@intel.com; Naveen Mamindlapalli
> <naveenm@marvell.com>
> Subject: Re: [net PATCH v2] octeontx2-af: Move validation of ptp
> pointer before its usage
> 
> On Fri, Jun 09, 2023 at 05:28:06PM +0530, Sai Krishna wrote:
> > @@ -428,7 +427,7 @@ static int ptp_probe(struct pci_dev *pdev,
> >  	return 0;
> >
> >  error_free:
> > -	devm_kfree(dev, ptp);
> > +	kfree(ptp);
> 
> Yeah.  It's strange any time we call devm_kfree()...  So there is something
> here which I have not understood.
We moved from devm_kzalloc/devm_kfree  to kzalloc/kfree as per review comments from Maciej.
> 
> >
> >  error:
> >  	/* For `ptp_get()` we need to differentiate between the case
> 
> This probe function is super weird how it returns success on the failure path.
> One concern, I had initially was that if anything returns -EPROBE_DEFER then
> we cannot recover.  That's not possible in the current code, but it makes me
> itch...  But here is a different crash.
> 

In few circumstances, the PTP device is probed before the AF device in the driver. In such instance, -EPROBE_DEFER is used.
-- EDEFER_PROBE is useful when probe order changes. Ex: AF driver probes before PTP.

> drivers/net/ethernet/marvell/octeontx2/af/ptp.c
>    432  error:
>    433          /* For `ptp_get()` we need to differentiate between the case
>    434           * when the core has not tried to probe this device and the case
> when
>    435           * the probe failed.  In the later case we pretend that the
>    436           * initialization was successful and keep the error in
>    437           * `dev->driver_data`.
>    438           */
>    439          pci_set_drvdata(pdev, ERR_PTR(err));
>    440          if (!first_ptp_block)
>    441                  first_ptp_block = ERR_PTR(err);
> 
> first_ptp_block is NULL for unprobed, an error pointer for probe failure, or
> valid pointer.

This is correct.
> 
>    442
>    443          return 0;
>    444  }
> 
> drivers/net/ethernet/marvell/octeontx2/af/ptp.c
>    201  struct ptp *ptp_get(void)
>    202  {
>    203          struct ptp *ptp = first_ptp_block;
>                             ^^^^^^^^^^^^^^^^^^^^^^
> 
>    204
>    205          /* Check PTP block is present in hardware */
>    206          if (!pci_dev_present(ptp_id_table))
>    207                  return ERR_PTR(-ENODEV);
>    208          /* Check driver is bound to PTP block */
>    209          if (!ptp)
>    210                  ptp = ERR_PTR(-EPROBE_DEFER);
>    211          else
>    212                  pci_dev_get(ptp->pdev);
>                                     ^^^^^^^^^ if first_ptp_block is an error pointer this will
> Oops.

Will fix this in v3 patch.

Thanks,
Sai
> 
>    213
>    214          return ptp;
>    215  }
> 
> regards,
> dan carpenter