Re: [PATCH 1/2] [PCI] Check that MWI bit really did get set

[Andrew Morton <akpm@osdl.org>]

On Sun, 15 Oct 2006 17:16:35 -0700
David Brownell <david-b@pacbell.net> wrote:

> 
> > You, the driver author _do not know_ what pci_set_mwi() does at present, on
> > all platforms, nor do you know what it does in the future. 
> 
> I know that it enables MWI accesses ... or fails.  Beyond that, there
> should be no reason to care.  If the hardware can use a lower-overhead
> type of PCI bus cycle, I want it to do so.  If not, no sweat.
> 

There are two reasons why it can fail:

1: The bus doesn't support MWI.  Here, the caller doesn't care.

2: The bus _does_ support MWI, but the attempt to enable it failed. 
   Here we very much do care, because we're losing performance.

> 
> > This is not a terribly important issue, and it is far from the worst case
> > of missed error-checking which we have in there. 
> 
> The reason I think it's important enough to continue this discussion is
> that as it currently stands, it's a good example of a **BAD** interface
> design ... since it's pointlessly marked as must_check.  (See appended
> patch to fix that issue.)

It's important to continue this discussion so that certain principles can
be set and agreed to.  Because we have a *lot* of unchecked errors in
there.  We would benefit from setting guidelines establishing

- Which sorts of errors should be handled in callers

- Which sorts of errors should be handled (ie: just reported) in callees

- Which sorts of errors should be handled in neither callers nor callees
  (are there any of these?)

- Whether is it ever legitimate for a caller to not check the return code
  from a callee which can return -EFOO.  (I suspect not - it probably
  indicates a misdesign in the callee, as in this case).