| Message ID | 1310426484-30083-1-git-send-email-khilman@ti.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Mon, Jul 11, 2011 at 4:21 PM, Kevin Hilman <khilman@ti.com> wrote: > Fix boot crash in watchdog driver when runtime PM is disabled. > > When runtime PM is disabled, devices should be left enabled so that > all device accesses in drivers will succeed even though the runtime PM > get/put calls are noops. > > This is already the case for all devices, but the WDT init has its > own hwmod postsetup which was not taking this into account. > > This fixes a boot crash where the first device access in the WDT probe > function would crash a booting kernel. > > Cc: Paul Walmsley <paul@pwsan.com> > Signed-off-by: Kevin Hilman <khilman@ti.com> > --- > Tony, as this is a boot crash, it should probably be queued for v3.0. > However, since OMAP2PLUS_TYPICAL always enables runtime PM, this is > unlikely to be seen unless specifically testing with runtime PM disabled. I forgot to mention that I added this fix to my for_3.0/pm-fixes-3 branch. Kevin
On 7/11/2011 4:21 PM, Kevin Hilman wrote: > Fix boot crash in watchdog driver when runtime PM is disabled. > > When runtime PM is disabled, devices should be left enabled so that > all device accesses in drivers will succeed even though the runtime PM > get/put calls are noops. > > This is already the case for all devices, but the WDT init has its > own hwmod postsetup which was not taking this into account. > > This fixes a boot crash where the first device access in the WDT probe > function would crash a booting kernel. > > Cc: Paul Walmsley<paul@pwsan.com> > Signed-off-by: Kevin Hilman<khilman@ti.com> > --- > Tony, as this is a boot crash, it should probably be queued for v3.0. > However, since OMAP2PLUS_TYPICAL always enables runtime PM, this is > unlikely to be seen unless specifically testing with runtime PM disabled. > This patch is exactly same as the one I submitted on Jan 5, 2011 for the same issue. Here is the patchwork link for the same. https://patchwork.kernel.org/patch/453481/ Paul had a valid concern about the boot-loader leaving the watchdog enabled. More on this in below thread. http://www.spinics.net/lists/arm-kernel/msg110937.html > arch/arm/mach-omap2/io.c | 4 ++++ > 1 files changed, 4 insertions(+), 0 deletions(-) > > diff --git a/arch/arm/mach-omap2/io.c b/arch/arm/mach-omap2/io.c > index 441e79d..95c08aa 100644 > --- a/arch/arm/mach-omap2/io.c > +++ b/arch/arm/mach-omap2/io.c > @@ -395,7 +395,11 @@ void __init omap2_init_common_infrastructure(void) > * XXX ideally we could detect whether the MPU WDT was currently > * enabled here and make this conditional > */ > +#ifdef CONFIG_PM_RUNTIME > postsetup_state = _HWMOD_STATE_DISABLED; > +#else > + postsetup_state = _HWMOD_STATE_ENABLED; > +#endif > omap_hwmod_for_each_by_class("wd_timer", > _set_hwmod_postsetup_state, > &postsetup_state);
Santosh Shilimkar <santosh.shilimkar@ti.com> writes: > On 7/11/2011 4:21 PM, Kevin Hilman wrote: >> Fix boot crash in watchdog driver when runtime PM is disabled. >> >> When runtime PM is disabled, devices should be left enabled so that >> all device accesses in drivers will succeed even though the runtime PM >> get/put calls are noops. >> >> This is already the case for all devices, but the WDT init has its >> own hwmod postsetup which was not taking this into account. >> >> This fixes a boot crash where the first device access in the WDT probe >> function would crash a booting kernel. >> >> Cc: Paul Walmsley<paul@pwsan.com> >> Signed-off-by: Kevin Hilman<khilman@ti.com> >> --- >> Tony, as this is a boot crash, it should probably be queued for v3.0. >> However, since OMAP2PLUS_TYPICAL always enables runtime PM, this is >> unlikely to be seen unless specifically testing with runtime PM disabled. >> > This patch is exactly same as the one I submitted on Jan 5, 2011 for > the same issue. Here is the patchwork link for the same. > > https://patchwork.kernel.org/patch/453481/ > > Paul had a valid concern about the boot-loader leaving the watchdog > enabled. More on this in below thread. > > http://www.spinics.net/lists/arm-kernel/msg110937.html > Paul's concern is valid, and could be addressed by making that init code actually check if the WDT is currently active (as suggested by the comment there.) In the mean time, this simple fix should be applied to prevent a boot crash, IMO. Kevin
On Tue, 12 Jul 2011, Kevin Hilman wrote: > Paul's concern is valid, and could be addressed by making that init code > actually check if the WDT is currently active (as suggested by the > comment there.) > > In the mean time, this simple fix should be applied to prevent a boot > crash, IMO. Thing is, applying this fix will cause crashes if the bootloader enables the watchdog, but the kernel does not. It just trades one source of crashes for another. The root cause of this problem is that the drivers are incorrectly relying on runtime PM to "enable" devices. My personal opinion is that we should be doing what SSB, PCI, etc. do, and have a omap_device_enable() that is called by the driver. Basically, we need a real omap_bus/omap_device. Meanwhile, switching to lazy device reset as Tony suggested should mask the issue. The patches for this are 95% done; the intention is to post them for early 3.2. Or if people are really gung-ho, 3.1 rc. - Paul
diff --git a/arch/arm/mach-omap2/io.c b/arch/arm/mach-omap2/io.c index 441e79d..95c08aa 100644 --- a/arch/arm/mach-omap2/io.c +++ b/arch/arm/mach-omap2/io.c @@ -395,7 +395,11 @@ void __init omap2_init_common_infrastructure(void) * XXX ideally we could detect whether the MPU WDT was currently * enabled here and make this conditional */ +#ifdef CONFIG_PM_RUNTIME postsetup_state = _HWMOD_STATE_DISABLED; +#else + postsetup_state = _HWMOD_STATE_ENABLED; +#endif omap_hwmod_for_each_by_class("wd_timer", _set_hwmod_postsetup_state, &postsetup_state);
Fix boot crash in watchdog driver when runtime PM is disabled. When runtime PM is disabled, devices should be left enabled so that all device accesses in drivers will succeed even though the runtime PM get/put calls are noops. This is already the case for all devices, but the WDT init has its own hwmod postsetup which was not taking this into account. This fixes a boot crash where the first device access in the WDT probe function would crash a booting kernel. Cc: Paul Walmsley <paul@pwsan.com> Signed-off-by: Kevin Hilman <khilman@ti.com> --- Tony, as this is a boot crash, it should probably be queued for v3.0. However, since OMAP2PLUS_TYPICAL always enables runtime PM, this is unlikely to be seen unless specifically testing with runtime PM disabled. arch/arm/mach-omap2/io.c | 4 ++++ 1 files changed, 4 insertions(+), 0 deletions(-)