Re: [PATCH v2] ata: Compact Flash driver for Netlogic XLR/XLS

["Jayachandran C." <jayachandranc@netlogicmicro.com>]

On Mon, Jun 11, 2012 at 09:02:15PM +0400, Sergei Shtylyov wrote:
> Hello.
> 
> On 11-06-2012 12:28, Jayachandran C wrote:
> 
> >From: Kamlakant Patel<kamlakant.patel@broadcom.com>
> 
> >Add a platform ATA driver for the CF interface on Netlogic XLR/XLS
> >MIPS SoCs. Chipselect 6 on the peripheral IO bus on these SoCs can
> >be configured to act as a Compact Flash interface.
> 
> >The driver expects three resources to be passed to it:
> >  0 - memory region corresponding to the CF chipselect
> >  1 - memory region of the flash interrupt ack register in the
> >      flash configuration region
> >  2 - and the IRQ resource for the Compact Flash.
> 
> >Signed-off-by: Kamlakant Patel<kamlakant.patel@broadcom.com>
> >Signed-off-by: Jayachandran C<jayachandranc@netlogicmicro.com>
> 
>    Jayachandran, after our chat on #mipslinux I thought your next
> driver will be for drivers/pcmcia/ since you're not going to use the
> True-IDE mode of the PCMCIA interface.

Since the CF is expected to be present at boot-time (and initialized by
firmware), we can just use a simple PATA driver instead of using the full
PCMCIA infrastructure.  But we are looking at that option as well.

> 
> >diff --git a/drivers/ata/pata_xlr_cf.c b/drivers/ata/pata_xlr_cf.c
> >new file mode 100644
> >index 0000000..7d699c4
> >--- /dev/null
> >+++ b/drivers/ata/pata_xlr_cf.c
> >@@ -0,0 +1,170 @@
> [...]
> >+#define XLR_CF_REG_BASE	0x1f0
> >+#define XLR_CF_REG_CTRL	0x3f6
> 
>    These port addresses are the offsets in the PCMCIA I/O space of
> the I/O card you insert. In True IDE mode they are replaced by -CEx
> signals IIRC.
> 
> >+
> >+struct pata_xlr_priv {
> >+	void __iomem	*xlr_cf_ackreg;
> 
>    Why not just make that register pointer the private data directly?

That would lose the __iomem attribute.

> >+static bool pata_xlr_irq_check(struct ata_port *port)
> >+{
> >+	struct pata_xlr_priv *priv = port->private_data;
> >+	unsigned int reg;
> >+
> >+	reg = readl(priv->xlr_cf_ackreg);
> >+	return (reg != 0);
> 
>    Parens not needed.
> 
> >+static int __devinit pata_xlr_cf_probe(struct platform_device *pdev)
> >+{
> >+	struct ata_host *host;
> >+	struct ata_port *ap;
> >+	struct ata_ioports *ioaddr;
> >+	struct resource *io_res, *irq_res, *ack_res;
> >+	struct pata_xlr_priv *priv;
> >+	struct device *dev =&pdev->dev;
> >+	void __iomem *iodata_addr;
> >+	int irq = 0;
> >+
> >+	/* Simple resource validation */
> >+	if (pdev->num_resources != 3) {
> >+		dev_err(&pdev->dev, "invalid number of resources\n");
> >+		return -EINVAL;
> >+	}
> >+
> >+	/* Get the I/O base */
> >+	io_res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
> >+	if (io_res == NULL)
> >+		return -EINVAL;
> >+
> >+	ack_res = platform_get_resource(pdev, IORESOURCE_MEM, 1);
> >+	if (ack_res == NULL)
> >+		return -EINVAL;
> >+
> >+	/* And the IRQ */
> >+	irq_res = platform_get_resource(pdev, IORESOURCE_IRQ, 0);
> 
>    You can use paltform_get_irq() and save on 'irq_res'.
> 
> >+	if (irq_res && irq_res->start>  0)
> >+		irq = irq_res->start;
> >+
> >+	ioaddr = &ap->ioaddr;
> >+	ioaddr->data_addr = iodata_addr;
> 
>    This is incorrect and overriden by ata_sff_std_ports() later.
> 
> >+	ioaddr->cmd_addr = ioaddr->data_addr + XLR_CF_REG_BASE;
> >+	ioaddr->altstatus_addr = ioaddr->data_addr + XLR_CF_REG_CTRL;
> >+	ioaddr->ctl_addr = ioaddr->data_addr + XLR_CF_REG_CTRL;
> >+
> >+	ata_sff_std_ports(ioaddr);
> >+
> >+	ata_port_desc(ap, "mmio cmd 0x%x ctl 0x%x",
> 
>    Why not use "0x%p" without the casts?
> 
> >+			(unsigned int)ap->ioaddr.cmd_addr,
> >+			(unsigned int)ap->ioaddr.ctl_addr);
> >+
> >+	/* activate */
> >+	return ata_host_activate(host, irq, irq ? ata_sff_interrupt : NULL,
> >+				 IRQF_DISABLED, &pata_xlr_sht);
> 
>    IRQF_DISABLED is a nop now. Don't add another use of this deprecated flag.

Thanks for the comments, will post an updated patch (after we try the 
PCMCIA option as well).

Regards,
JC.