Re: [PATCH 6/21] advansys: Move to scsi hotplug initialisation model

[Jeff Garzik <jeff@garzik.org>]

Matthew Wilcox wrote:
> @@ -18511,13 +18493,24 @@ advansys_board_found(int iop, struct device *dev, int bus_type)
>  	}
>  
>  	if (err_code != 0)
> -		goto err_free_irq;
> +		goto err_free_wide_mem;
>  
>  	ASC_DBG_PRT_SCSI_HOST(2, shost);
>  
> +	ret = scsi_add_host(shost, dev);
> +	if (ret)
> +		goto err_free_wide_mem;
> +
> +	scsi_scan_host(shost);
>  	return shost;
>  
> - err_free_irq:
> + err_free_wide_mem:
> +	kfree(boardp->orig_carrp);
> +	kfree(boardp->orig_reqp);
> +	while ((sgp = boardp->adv_sgblkp) != NULL) {
> +		boardp->adv_sgblkp = sgp->next_sgblkp;
> +		kfree(sgp);
> +	}
>  	free_irq(shost->irq, boardp);
>   err_free_dma:
>  	if (shost->dma_channel != NO_ISA_DMA)

When this patch is considered in context of replies to patch #3, it's 
OK...  it un-breaks driver which was broken in patch #3.

But like I noted, this should be done -before- PCI hotplug conversion, 
for the obvious reasons.

Generally the process involves:

* call foo_detect from module init
* for each scsi host, call foo_release from module exit
* do the scsi conversion as you've presented it here

and then in later patches, rip out the PCI code from foo_detect and 
foo_release, etc.

	Jeff