[PATCH] drivers: staging: lustre: Use mult if units not specified

[Chris Rorvick]

On Tue, Dec 16, 2014 at 5:35 AM, Dan Carpenter <dan.carpenter at oracle.com> wrote:
> On Tue, Dec 16, 2014 at 11:14:35AM +0000, Dilger, Andreas wrote:
> >
> > Sorry, that isn't right.  Chris' patch is actually doing the right thing
> > to check for units > 1.
>
> It's not right because it discards the negative.

I don't think this patch introduces a bug.  If anything, it was already
there.  It looked to me like the value passed in to `mult' was assumed
to be positive and was simply being used as a flag to indicate whether
`buffer' started with a '-' when units were passed.

For example, say the value passed in is "-2K" and the `mult' is 1.  The
check for '-' will negate `mult' making it -1.  Then the units
conditional will override mult with `-units' (i.e., -1024.)

Now say we pass "-2" with `mult' equal to 1024.  The result is same, but
the path is a bit different.  `mult' will again be negated due to
`buffer' beginning with '-', but then it will be left alone at the units
check.

In both of the above cases the negative sign is properly accounted for.

> >  The proposed change above discards "mult"
> > entirely, which breaks the users of this function that are not in this
> > file (e.g. osc_cached_mb_seq_write() or ll_max_cached_mb_seq_write())
> > that have tunables in units of MB by default, but can also use parameters
> > with units like "4.5G" for convenience.
>
> I think you are confusing lprocfs_write_frac_helper() and
> lprocfs_write_frac_u64_helper().  There is only one caller for this
> function.

By this logic, lprocfs_write_frac_u64_helper() should just be removed
and it's code should be folded into lprocfs_write_u64_helper(), no?

Regards,

Chris