Re: [PATCH] crypto: caam: fix error reporting

[Kim Phillips <kim.phillips@freescale.com>]

On Mon, 3 Nov 2014 11:18:36 +0200
Cristian Stoica <cristian.stoica@freescale.com> wrote:

> On 10/31/2014 08:22 PM, Kim Phillips wrote:
> > On Fri, 31 Oct 2014 18:57:33 +0200
> > Cristian Stoica <cristian.stoica@freescale.com> wrote:
> > 
> > If this issue was brought up by h/w, the appropriate new error codes
> > should be being introduced.
> 
> If you have the new error codes please send them to me and I'll make an
> update.

I was mainly inquiring as to the motive of the patch.  fwiw, I don't
see error codes with the most significant bit set in the latest
documentation (which is available from STC).

> > Otherwise, I'm assuming it was brought up by a static code analyser,
> > which technically could be ignored, but...
> 
> Actually, our static code analyzer did not see this one.

ok, so the patch technically isn't fixing anything broken, then.

> >> -	/*
> >> -	 * If there is no further error handling function, just
> >> -	 * print the error code, error string and exit. Otherwise
> >> -	 * call the handler function.
> >> -	 */
> > 
> > why remove the comment?  It's still valid.
> 
> The comment was disagreeing with the new code, so I just removed it.

the new code just added a new condition, which doesn't invalidate
the comment.  And simply removing the comment as opposed to amending
it is a bit overkill.

> >> -	if (!status_src[ssrc].report_ssed)
> >> -		dev_err(jrdev, "%08x: %s: \n", status, status_src[ssrc].error);
> >> -	else
> >> +	if (status_src[ssrc].report_ssed)
> >>  		status_src[ssrc].report_ssed(jrdev, status, error);
> >> +	else if (error)
> >> +		dev_err(jrdev, "%d: %s\n", ssrc, error);
> >> +	else
> >> +		dev_err(jrdev, "%d: unknown error code\n", ssrc);
> > 
> > This is simpler:
> > 
> > diff --git a/drivers/crypto/caam/error.c b/drivers/crypto/caam/error.c
> > index 6531054..6f4a148 100644
> > --- a/drivers/crypto/caam/error.c
> > +++ b/drivers/crypto/caam/error.c
> > @@ -224,7 +224,12 @@ void caam_jr_strstatus(struct device *jrdev, u32 status)
> >                 { report_cond_code_status, "Condition Code" },
> >         };
> >         u32 ssrc = status >> JRSTA_SSRC_SHIFT;
> > -       const char *error = status_src[ssrc].error;
> > +       const char *error;
> > +
> > +       if (ssrc >= ARRAY_SIZE(status_src)) {
> > +               dev_err(jrdev, "unknown error status source %d\n", ssrc);
> > +               return;
> > +       }
> 
> It is indeed simpler but does it consider also the missing error codes
> at index 1 and 5? Just checking for an upper bound is not enough.

no, the existing code already handles that.  Note that newer
documentation fills the 1 and 5 slots, too.

> On the other hand, if the error field is only three bits wide instead of
> four as stated by the documentation, a better fix means using a three
> bit mask instead of reporting an invalid error code.

true, but then we'd introduce a direct discrepancy with the
documentation, and thus h/w.

Kim