Re: [RESEND PATCH] memory: Freescale CoreNet Coherency Fabric error reporting driver

On Sun, 2014-06-29 at 23:58 -0500, Bhushan Bharat-R65777 wrote:

quoted

-----Original Message-----
From: Wood Scott-B07421
Sent: Wednesday, June 04, 2014 10:38 PM
To: Bhushan Bharat-R65777
Cc: Greg Kroah-Hartman; linuxppc-dev@lists.ozlabs.org; linux-
kernel@vger.kernel.org
Subject: Re: [RESEND PATCH] memory: Freescale CoreNet Coherency Fabric error
reporting driver

On Wed, 2014-06-04 at 12:04 -0500, Bhushan Bharat-R65777 wrote:

quoted

quoted

-----Original Message-----
From: Wood Scott-B07421
Sent: Wednesday, June 04, 2014 10:12 PM
To: Bhushan Bharat-R65777
Cc: Greg Kroah-Hartman; linuxppc-dev@lists.ozlabs.org; linux-
kernel@vger.kernel.org
Subject: Re: [RESEND PATCH] memory: Freescale CoreNet Coherency
Fabric error reporting driver

On Wed, 2014-06-04 at 03:17 -0500, Bhushan Bharat-R65777 wrote:

quoted

quoted

+static int ccf_remove(struct platform_device *pdev) {
+	struct ccf_private *ccf = dev_get_drvdata(&pdev->dev);
+
+	switch (ccf->info->version) {
+	case CCF1:
+		iowrite32be(0, &ccf->err_regs->errdis);
+		break;
+
+	case CCF2:
+		iowrite32be(0, &ccf->err_regs->errinten);

Do you think it is same to disable detection bits in ccf->err_regs-

errdis?

quoted

Disabling the interrupt is what we're aiming for here, but ccf1
doesn't provide a way to do that separate from disabling detection.

What I wanted to say that do we also need to disable detection (set
ERRDET_LAE | ERRDET_CV bits in errdis) apart from clearing errinten on
ccf2 ?

I don't think we "need" to.  You could argue that we should for consistency,
though I think there's value in errors continuing to be detected even without
the driver (e.g. can dump the registers in a debugger).

Yes this comment was for consistency. Also IIUC, the state which is left when the driver is removed is not default reset behavior.

How many drivers leave the hardware in pristine reset state when
exiting?  And you could argue that having detection off by default is
poor hardware design (enabling interrupts is another matter of course).

If we want errors to be detected then should not we have a sysfs interface?

That may be useful but it's beyond the scope of what I'm doing with this
patch.  We currently don't log machine checks anywhere but via printk
either.

BTW, I thought I had sent v2 of this, but I don't see it anywhere...
I'll respin soon.

-Scott