Re: [PATCH 3/3] pci: gt96100eth use pci probing

[Jeff Garzik <jeff@garzik.org>]

Jiri Slaby wrote:
>>> +unprobe:
>>> +    for (j = i; j > 0; j--) {
>>> +        struct gt96100_if_t *gtif = &gtifs[j - 1];
>>> +        gt96100_remove1(gtif);
>>> +    }
>>> +    kfree(gtifs);
>> upon failure, you fail to set drvdata back to NULL
> What is the purpose of setting this to NULL, other drivers don't do that too?

A simple grep(1) shows well over 300 cases that do this.

And it's just logical:  don't leave a pointer hanging around, after it 
has been kfree'd.


>>> +        struct gt96100_private *gp = netdev_priv(gtif->dev);
>>> +        unregister_netdev(gtif->dev);
>>> +        dmafree(RX_HASH_TABLE_SIZE, gp->hash_table_dma);
>>> +        dmafree(PKT_BUF_SZ*RX_RING_SIZE, gp->rx_buff);
>>> +        dmafree(sizeof(gt96100_rd_t) * RX_RING_SIZE
>>> +            + sizeof(gt96100_td_t) * TX_RING_SIZE,
>>> +            gp->rx_ring);
>>> +        free_netdev(gtif->dev);
>>> +        release_region(gtif->iobase, gp->io_size);
>> shouldn't this be using pci_request_regions() / pci_release_regions() ?
> There are GT96100_ETH{0,1}_BASEs instead of bars, 

Indeed.  I stand corrected.

	Jeff