Re: [PATCH 1/1] - sgiioc4: fixup use of mmio ops

From: John Keller
Date: Thu Jul 13 2006 - 09:52:57 EST


>
> Jeremy Higdon wrote:
> >> @@ -381,7 +381,7 @@ ide_dma_sgiioc4(ide_hwif_t * hwif, unsig
> >> goto dma_alloc_failure;
> >> }
> >>
> >> - hwif->dma_base = dma_base;
> >> + hwif->dma_base = (unsigned long) ioremap(dma_base, num_ports);
> >> hwif->dmatable_cpu = pci_alloc_consistent(hwif->pci_dev,
> >> IOC4_PRD_ENTRIES * IOC4_PRD_BYTES,
> >> &hwif->dmatable_dma);
>
> adding a bug: must check for NULL ioremap return value

Shall do.


>
>
> >> @@ -607,18 +607,14 @@ ide_init_sgiioc4(ide_hwif_t * hwif)
> >> hwif->ide_dma_lostirq = &sgiioc4_ide_dma_lostirq;
> >> hwif->ide_dma_timeout = &__ide_dma_timeout;
> >>
> >> - /*
> >> - * The IOC4 uses MMIO rather than Port IO.
> >> - * It also needs special workarounds for INB.
> >> - */
> >> - default_hwif_mmiops(hwif);
> >> hwif->INB = &sgiioc4_INB;
> >> }
> >>
> >> static int __devinit
> >> sgiioc4_ide_setup_pci_device(struct pci_dev *dev, ide_pci_device_t * d)
> >> {
> >> - unsigned long base, ctl, dma_base, irqport;
> >> + unsigned long ctl, dma_base, irqport;
> >> + unsigned long bar0, cmd_base, cmd_phys_base, virt_base;
>
> New code should use 'void __iomem *' for ioremap return values, wherever
> possible.

OK.


>
>
> >> ide_hwif_t *hwif;
> >> int h;
> >>
> >> @@ -636,23 +632,27 @@ sgiioc4_ide_setup_pci_device(struct pci_
> >> }
> >>
> >> /* Get the CmdBlk and CtrlBlk Base Registers */
> >> - base = pci_resource_start(dev, 0) + IOC4_CMD_OFFSET;
> >> - ctl = pci_resource_start(dev, 0) + IOC4_CTRL_OFFSET;
> >> - irqport = pci_resource_start(dev, 0) + IOC4_INTR_OFFSET;
> >> + bar0 = pci_resource_start(dev, 0);
> >> + virt_base = (unsigned long) ioremap(bar0, pci_resource_len(dev, 0));
> >> + cmd_base = virt_base + IOC4_CMD_OFFSET;
> >> + ctl = virt_base + IOC4_CTRL_OFFSET;
> >> + irqport = virt_base + IOC4_INTR_OFFSET;
> >> dma_base = pci_resource_start(dev, 0) + IOC4_DMA_OFFSET;
>
> ditto: new bug here too

I'll add the check.


>
>
> >> - if (!request_region(base, IOC4_CMD_CTL_BLK_SIZE, hwif->name)) {
> >> + cmd_phys_base = bar0 + IOC4_CMD_OFFSET;
> >> + if (!request_mem_region(cmd_phys_base, IOC4_CMD_CTL_BLK_SIZE,
> >> + hwif->name)) {
> >> printk(KERN_ERR
> >> - "%s : %s -- ERROR, Port Addresses "
> >> + "%s : %s -- ERROR, Addresses "
> >> "0x%p to 0x%p ALREADY in use\n",
> >> - __FUNCTION__, hwif->name, (void *) base,
> >> - (void *) base + IOC4_CMD_CTL_BLK_SIZE);
> >> + __FUNCTION__, hwif->name, (void *) cmd_phys_base,
> >> + (void *) cmd_phys_base + IOC4_CMD_CTL_BLK_SIZE);
>
> If 'void __iomem *' were used, no casts would be needed here

So, 'void __iomem *' should also be used for physical (non-mapped)
addresses, as in this case?

John
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/