Re: [PATCH] rtc: armada38x: add __ro_after_init to armada38x_rtc_ops

[Russell King - ARM Linux <linux@armlinux.org.uk>]

On Tue, Jan 03, 2017 at 09:31:18PM +0000, Russell King - ARM Linux wrote:
> On Tue, Jan 03, 2017 at 01:18:29PM -0800, Kees Cook wrote:
> > On Mon, Jan 2, 2017 at 6:06 AM, Russell King - ARM Linux
> > <linux@armlinux.org.uk> wrote:
> > > On Mon, Dec 26, 2016 at 05:01:02PM +0530, Bhumika Goyal wrote:
> > >> The object armada38x_rtc_ops of type rtc_class_ops structure is not
> > >> modified after getting initialized by armada38x_rtc_probe. Apart from
> > >> getting referenced in init it is also passed as an argument to the function
> > >> devm_rtc_device_register but this argument is of type const struct
> > >> rtc_class_ops *. Therefore add __ro_after_init to its declaration.
> > >
> > > What I'd prefer here is for the structure to be duplicated, with one
> > > copy having the alarm methods and one which does not.  Both can then
> > > be made "const" (so placed into the read-only section at link time)
> > > and the probe function select between the two.
> > >
> > > I think that's a cleaner and better solution, even though it's
> > > slightly larger.
> > >
> > > I'm not a fan of __ro_after_init being used where other solutions are
> > > possible.
> > 
> > Can the pointer that points to the struct rtc_class_ops be made ro_after_init?
> 
> It's passed into the RTC core code, and probably stored in some dynamically
> allocated object, so probably no.  It's the same class of problem as every
> file_operations pointer in the kernel, or the thousand other operations
> structure pointers that a running kernel has.

For the elimination of doubt, this is what I meant in my original email.
As you can see, there's nothing to be marked as __ro_after_init anymore.

 drivers/rtc/rtc-armada38x.c | 24 +++++++++++++++++-------
 1 file changed, 17 insertions(+), 7 deletions(-)

diff --git a/drivers/rtc/rtc-armada38x.c b/drivers/rtc/rtc-armada38x.c
index 9a3f2a6f512e..a4166ccfce36 100644
--- a/drivers/rtc/rtc-armada38x.c
+++ b/drivers/rtc/rtc-armada38x.c
@@ -202,7 +202,7 @@ static irqreturn_t armada38x_rtc_alarm_irq(int irq, void *data)
 	return IRQ_HANDLED;
 }
 
-static struct rtc_class_ops armada38x_rtc_ops = {
+static const struct rtc_class_ops armada38x_rtc_ops = {
 	.read_time = armada38x_rtc_read_time,
 	.set_time = armada38x_rtc_set_time,
 	.read_alarm = armada38x_rtc_read_alarm,
@@ -210,8 +210,15 @@ static struct rtc_class_ops armada38x_rtc_ops = {
 	.alarm_irq_enable = armada38x_rtc_alarm_irq_enable,
 };
 
+static const struct rtc_class_ops armada38x_rtc_ops_noirq = {
+	.read_time = armada38x_rtc_read_time,
+	.set_time = armada38x_rtc_set_time,
+	.read_alarm = armada38x_rtc_read_alarm,
+};
+
 static __init int armada38x_rtc_probe(struct platform_device *pdev)
 {
+	const struct rtc_class_ops *ops;
 	struct resource *res;
 	struct armada38x_rtc *rtc;
 	int ret;
@@ -242,19 +249,22 @@ static __init int armada38x_rtc_probe(struct platform_device *pdev)
 				0, pdev->name, rtc) < 0) {
 		dev_warn(&pdev->dev, "Interrupt not available.\n");
 		rtc->irq = -1;
+	}
+	platform_set_drvdata(pdev, rtc);
+
+	if (rtc->irq != -1) {
+		device_init_wakeup(&pdev->dev, 1);
+		ops = &armada38x_rtc_ops;
+	} else {
 		/*
 		 * If there is no interrupt available then we can't
 		 * use the alarm
 		 */
-		armada38x_rtc_ops.set_alarm = NULL;
-		armada38x_rtc_ops.alarm_irq_enable = NULL;
+		ops = &armada38x_rtc_ops_noirq;
 	}
-	platform_set_drvdata(pdev, rtc);
-	if (rtc->irq != -1)
-		device_init_wakeup(&pdev->dev, 1);
 
 	rtc->rtc_dev = devm_rtc_device_register(&pdev->dev, pdev->name,
-					&armada38x_rtc_ops, THIS_MODULE);
+						ops, THIS_MODULE);
 	if (IS_ERR(rtc->rtc_dev)) {
 		ret = PTR_ERR(rtc->rtc_dev);
 		dev_err(&pdev->dev, "Failed to register RTC device: %d\n", ret);

-- 
RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up
according to speedtest.net.