On Sun, Aug 08, 2010 at 10:43:41PM +0200, Sascha Hauer wrote:

On Sun, Aug 08, 2010 at 11:32:44AM +0300, Baruch Siach wrote:

quoted

Hi Russell,

On Sun, Aug 08, 2010 at 09:19:56AM +0100, Russell King - ARM Linux wrote:

quoted

On Fri, Aug 06, 2010 at 03:53:09PM +0200, Sascha Hauer wrote:

quoted

+		err = ecc_stat & ecc_bit_mask;
+		if (err > err_limit) {
+			printk(KERN_WARNING "UnCorrectable RS-ECC Error\n");
+			return -1;

Someone's being lazy.

The code at nand_read_subpage() (drivers/mtd/nand/nand_base.c) expects the 
.correct callback to return -1 on an uncorrectable error:

                stat = chip->ecc.correct(mtd, p, &chip->buffers->ecccode[i], &chip->buffers->ecccalc[i]);
                if (stat == -1)
                        mtd->ecc_stats.failed++;
                else
                        mtd->ecc_stats.corrected += stat;

Then this should be changed to check for stat < 0. I found some drivers
in the tree returning an errno value in their .correct function instead
of -1.

That's the whole danger of the whole 'return -1' evilness in the kernel.
The long established convention in the kernel is that negative numbers
returned from functions are negative errno codes.

As soon as you decide that you want a function to return -1 to indicate
an error rather than a real errno code, you lose clarity on which
functions need '-1' and which are proper negative errno codes, and then
you end up with people returning negative errno codes for functions
which should be -1, and people returning -1 for functions which should
be negative errno codes.

If you want a function to return -1 for error, then it probably makes
sense to create a ECC_CORRECT_FAILED definition which happens to be -1.
Or some other number.  But don't use plain '-1' - it looks far too much
like "I was lazy, I couldn't be bothered to look up a proper errno."

We know full well that _lots_ of people submit stuff with 'return -1'
statements where they should be proper negative errno codes, so it's
something people are having a great deal of trouble with already.  Let's
not further confuse everyone by having functions expecting called methods
to do a plain "return -1;" on error.