Re: [PATCH -mm] x86 port lockless MCE quirky bank0

[Hugh Dickins <hugh@veritas.com>]

On Tue, 17 May 2005, Andrew Morton wrote:
> 
> Yes.  I'll drop the following from -mm:
> 
> x86-port-lockless-mce-preparation.patch
> x86-port-lockless-mce-implementation.patch
> x86-port-lockless-mce-implementation-fix.patch
> x86-port-lockless-mce-implementation-fix-2.patch

The right decision for now, I think - thanks.

I presume they may return later on, so I'd better confess:
I lied when I said my patch fixed the P6 bank0 issue, I was confused
between rebuildings and rebootings.  It should have fixed the issue,
but the patch which actually fixed it was one earlier, which had an
off-by-one ("> quirky_bank0") which I'd corrected before posting.

The ">= quirky_bank0" patch was not enough to fix the issue because...
mce_cpu_quirks (and mce_init and mce_cpu_features) are never called at
startup (on Intel or Centaur), because the logic in machine_check_init
(see below) to call mcheck_init is broken.

Which explains why I got the freeze at resume not at startup,
and casts doubt on how much any of it has got tested so far.

Hugh

void __devinit machine_check_init(struct cpuinfo_x86 *c)
{
	if (mce_dont_init)
		return;

	switch (c->x86_vendor) {

		case X86_VENDOR_INTEL:
			if (c->x86==5)
				intel_p5_mcheck_init(c);
			break;

		case X86_VENDOR_CENTAUR:
			if (c->x86==5)
				winchip_mcheck_init(c);
			break;

		default:
			machine_check_vector = do_machine_check;
			wmb();
			mcheck_init(c);
			break;
	}
}