Re: [Regression, post-3.5] System suspend broken on the Mackerel board

[Paul Mundt <lethal@linux-sh.org>]

On Sat, Jul 28, 2012 at 12:53:11AM +0200, Rafael J. Wysocki wrote:
> Unfortunately, your commit
> 
> commit ca5481c68e9fbcea62bb3c78ae6cccf99ca8fb73
> Author: Paul Mundt <lethal@linux-sh.org>
> Date:   Tue Jul 10 12:08:14 2012 +0900
> 
>     sh: pfc: Rudimentary pinctrl-backed GPIO support.
> 
> breaks system suspend on the Mackerel board (.config attached).  The system
> simply doesn't suspend and instead it hangs somewhere while suspending
> devices (apparently before running the "late" callbacks).
> 
> If the above commit is reverted, system suspend works normally.

On Tue, Jul 31, 2012 at 08:57:02PM -0700, kuninori.morimoto.gx@renesas.com wrote:
> gpio: sh7724_pfc handling gpio 0 -> 486
> core: sh7724_pfc support registered
> HW Breakpoints: SH-4A UBC support registered
> autorequest GPIO-53
> ------------[ cut here ]------------
> WARNING: at /opt/usr/src/WORK/morimoto/gitlinux/linux-2.6/drivers/gpio/gpiolib.3
> Modules linked in:
> 
> Pid : 1, Comm:          swapper
> CPU : 0                 Not tainted  (3.5.0-rc6+ #1407)
> 
> PC is at gpio_ensure_requested+0x30/0x78
> PR is at gpio_ensure_requested+0x30/0x78

Morimoto-san's logs off-list made it clear what happened. Both of these
platforms are going gpio_request() calls at arch_initcall() time which
completely screwed up the ordering of the pfc core. We seem to -ENODEV
out in one place due to missing a pfc pointer initialization elsewhere
resulting in -EPROBE_DEFER from gpiolib.

Turns out we can just collapse the probe/init stuff anyways, so this
ought to fix it. I've verified that it fixes Morimoto-san's issue, my
expectation is that the mackerel case is likewise getting tripped up but
no one bothered implementing any error detecting logic for gpio_request()
failing, so it doesn't fail gracefully.

I'll be pushing this out to Linus shortly:

---

commit 1e32dfe323d156d5d7b25b9feffe015d19713db2
Author: Paul Mundt <lethal@linux-sh.org>
Date:   Wed Aug 1 16:27:38 2012 +0900

    sh: pfc: Fix up init ordering mess.
    
    Commit ca5481c68e9fbcea62bb3c78ae6cccf99ca8fb73 ("sh: pfc: Rudimentary
    pinctrl-backed GPIO support.") introduced a regression for platforms that
    were doing early GPIO API calls (from arch_initcall() or earlier),
    leading to a situation where our two-stage registration logic would trip
    itself up and we'd -ENODEV out of the pinctrl registration path,
    resulting in endless -EPROBE_DEFER errors. Further lack of checking any
    sort of errors from gpio_request() resulted in boot time warnings,
    tripping on the FLAG_REQUESTED test-and-set in gpio_ensure_requested().
    
    As it turns out there's no particular need to bother with the two-stage
    registration, as the platform bus is already available at the point that
    we have to start caring. As such, it's easiest to simply fold these
    together in to a single init path, the ordering of which is ensured
    through the platform's mux registration, as usual.
    
    Reported-by: Rafael J. Wysocki <rjw@sisk.pl>
    Reported-by: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com>
    Signed-off-by: Paul Mundt <lethal@linux-sh.org>

diff --git a/drivers/sh/pfc/pinctrl.c b/drivers/sh/pfc/pinctrl.c
index 814b292..2804eaa 100644
--- a/drivers/sh/pfc/pinctrl.c
+++ b/drivers/sh/pfc/pinctrl.c
@@ -325,20 +325,6 @@ static struct pinctrl_desc sh_pfc_pinctrl_desc = {
 	.confops	= &sh_pfc_pinconf_ops,
 };
 
-int sh_pfc_register_pinctrl(struct sh_pfc *pfc)
-{
-	sh_pfc_pmx = kzalloc(sizeof(struct sh_pfc_pinctrl), GFP_KERNEL);
-	if (unlikely(!sh_pfc_pmx))
-		return -ENOMEM;
-
-	spin_lock_init(&sh_pfc_pmx->lock);
-
-	sh_pfc_pmx->pfc = pfc;
-
-	return 0;
-}
-EXPORT_SYMBOL_GPL(sh_pfc_register_pinctrl);
-
 static inline void __devinit sh_pfc_map_one_gpio(struct sh_pfc *pfc,
 						 struct sh_pfc_pinctrl *pmx,
 						 struct pinmux_gpio *gpio,
@@ -505,7 +491,7 @@ static struct platform_device sh_pfc_pinctrl_device = {
 	.id		= -1,
 };
 
-static int __init sh_pfc_pinctrl_init(void)
+static int sh_pfc_pinctrl_init(void)
 {
 	int rc;
 
@@ -519,10 +505,22 @@ static int __init sh_pfc_pinctrl_init(void)
 	return rc;
 }
 
+int sh_pfc_register_pinctrl(struct sh_pfc *pfc)
+{
+	sh_pfc_pmx = kzalloc(sizeof(struct sh_pfc_pinctrl), GFP_KERNEL);
+	if (unlikely(!sh_pfc_pmx))
+		return -ENOMEM;
+
+	spin_lock_init(&sh_pfc_pmx->lock);
+
+	sh_pfc_pmx->pfc = pfc;
+
+	return sh_pfc_pinctrl_init();
+}
+EXPORT_SYMBOL_GPL(sh_pfc_register_pinctrl);
+
 static void __exit sh_pfc_pinctrl_exit(void)
 {
 	platform_driver_unregister(&sh_pfc_pinctrl_driver);
 }
-
-subsys_initcall(sh_pfc_pinctrl_init);
 module_exit(sh_pfc_pinctrl_exit);