From mboxrd@z Thu Jan 1 00:00:00 1970 From: tony@atomide.com (Tony Lindgren) Date: Mon, 19 Dec 2011 15:20:49 -0800 Subject: buggy usage of pm_idle by omap1 code In-Reply-To: <20111219213445.GL14542@n2100.arm.linux.org.uk> References: <20111219173518.GH6464@atomide.com> <20111219213445.GL14542@n2100.arm.linux.org.uk> Message-ID: <20111219232049.GU6464@atomide.com> To: linux-arm-kernel@lists.infradead.org List-Id: linux-arm-kernel.lists.infradead.org * Russell King - ARM Linux [111219 13:02]: > On Mon, Dec 19, 2011 at 09:35:19AM -0800, Tony Lindgren wrote: > > Good catch. Here's a patch to deal with that issue, want to take this > > into your idle series? I'll try out your idle series today at some > > point. > > Two points: > > 1. It's good practice to call cpu_idle_wait() after modifying pm_idle, > even on UP - it allows static checking of this. Hmm, looks like we're not using that anywhere right now. Should we have a function or macro set_arm_idle() that does that automatically? > > From: Tony Lindgren > > Date: Mon, 19 Dec 2011 09:11:11 -0800 > > Subject: [PATCH] ARM: OMAP1: Fix pm_idle during suspend > > > > Commit 9ccdac3662dbf3c75e8f8851a214bdf7d365a4bd ([ARM] idle: > > clean up pm_idle calling, obey hlt_counter) removed a check > > for NULL pm_idle. > > > > On omap1 the system hits an equivalent of suspend during idle. > > If we are suspending, we need to make sure so we don't call > > omap1_pm_idle during the suspend process. > > > > Fix this by setting the pm_idle function to a an empty > > function while suspending. > > 2. what's wrong with calling disable_hlt() to prevent the pm_idle function > being called? Why solve it this complicated way of changing the pm_idle > pointer when there's some infrastructure which can fix it for you already? Thanks that allows removal of few lines of code. Updated patch below. Regards, Tony From: Tony Lindgren Date: Mon, 19 Dec 2011 09:11:11 -0800 Subject: [PATCH] Subject: [PATCH] ARM: OMAP1: Fix pm_idle during suspend Commit 9ccdac3662dbf3c75e8f8851a214bdf7d365a4bd ([ARM] idle: clean up pm_idle calling, obey hlt_counter) removed a check for NULL pm_idle. On omap1 the system hits an equivalent of suspend during idle. If we are suspending, we need to make sure so we don't call omap1_pm_idle during the suspend process. Fix this by using disable_hlt and enable_hlt for the suspend as pointed out by Russell King . This allows us to remove the saved_idle. Reported-by: Nicolas Pitre Signed-off-by: Tony Lindgren --- a/arch/arm/mach-omap1/pm.c +++ b/arch/arm/mach-omap1/pm.c @@ -583,8 +583,6 @@ static void omap_pm_init_proc(void) #endif /* DEBUG && CONFIG_PROC_FS */ -static void (*saved_idle)(void) = NULL; - /* * omap_pm_prepare - Do preliminary suspend work. * @@ -592,8 +590,7 @@ static void (*saved_idle)(void) = NULL; static int omap_pm_prepare(void) { /* We cannot sleep in idle until we have resumed */ - saved_idle = pm_idle; - pm_idle = NULL; + disable_hlt(); return 0; } @@ -630,7 +627,7 @@ static int omap_pm_enter(suspend_state_t state) static void omap_pm_finish(void) { - pm_idle = saved_idle; + enable_hlt(); }