Messages in this thread | | | Subject | Re: [PATCH 1/2] MCE, AMD: Make MC2 decoding part of amd_decoder_ops as well | From | Joe Perches <> | Date | Mon, 17 Dec 2012 12:16:16 -0800 |
| |
On Mon, 2012-12-17 at 21:05 +0100, Borislav Petkov wrote: > On Mon, Dec 17, 2012 at 11:57:01AM -0800, Joe Perches wrote: > > It'd be better to change the pr_cont uses to pr_emerg > > > > [] > > > > > +static void decode_mc2_mce(struct mce *m) > > > +{ > > > + u16 ec = EC(m->status); > > > + u8 xec = XEC(m->status, xec_mask); > > > + > > > + pr_emerg(HW_ERR "MC2 Error: ");
> > > > Remove this and > > > > > + if (fam_ops->mc2_mce(ec, xec)) > > > + ; > > > + else > > > + pr_emerg(HW_ERR "Corrupted MC2 MCE info?\n"); > > > } > > > > And make this > > > > if (!fam_ops->mc2_mce(ec, xec)) > > pr_emerg(etc...); > > No, this is not how we do this here. We do pr_emerg in the main per-bank > function, i.e. mc0, mc1, mc2... and we finish the line in the respective > function with pr_cont. > > If your fear is line interleaving, then this shouldn't happen in most > cases because we're in atomic #MC context and nothing else is executing > in that case. (And I haven't seen it interleave in all my testing so > far). > > If it does interleave when we have a non-critical error detected and > reported in process context, then this whole decoding code needs a lot > more work than this. >
You also have paths where you start a pr_emerg and do not terminate it when there is no apparent error at all.
| |