Re: [PATCH 4/8] atmel_nand: Clean up and fix probe() error path

[Haavard Skinnemoen <haavard.skinnemoen@atmel.com>]

On Sat, 7 Jun 2008 18:08:07 +0200
"Andrew Victor" <avictor.za@gmail.com> wrote:

> hi Haavard,
> 
> > This fixes several bugs in the atmel_nand_probe() error path, including
> > at least one memory leak.
> 
> There is an issue with the current at91_nand driver here.

Hmm, right.

> The driver currently ioremap()'s the ECC controller's registers, but
> on the AT91 the ECC controller is part of the System Peripherals and
> so it is already mapped at startup.
> I don't think it can/should be remapped twice.

I don't see why that's really a problem -- it may get a different
virtual address and possibly waste a TLB entry, but it should work.

> The AT91 patch (on maxim.org.za) current does:
>     regs = platform_get_resource(pdev, IORESOURCE_MEM, 1);
>          ....
>     host->ecc = (void __force __iomem *) (AT91_VA_BASE_SYS - AT91_BASE_SYS);
>     host->ecc += regs->start;
> instead of the ioremap().  But this is not portable to the AVR32.

How about sticking a "void __iomem *ecc_regs" field in struct
atmel_nand_data and allow platforms to specify a pre-mapped pointer?
Then the driver can do

	if (host->board->ecc_regs)
		host->ecc = host->board->ecc_regs;
	else if (regs)
		host->ecc = ioremap(regs->start, regs->end - regs->start + 1);

	if (hard_ecc && !host->ecc)
		printk("Hardware ECC not available\n");

Right? I think we use a similar trick in the atmel_serial driver too.

Haavard