Unlike the clocks management code for runtime PM, the code used for
system suspend does not check the pm_clock_entry.status field.
If pm_clk_acquire() failed, ce->status will be PCE_STATUS_ERROR, and
ce->clk will be a negative error code (e.g. 0xfffffffe = -2 = -ENOENT).
Depending on the clock implementation, suspend or resume may crash with:
Unable to handle kernel NULL pointer dereference at virtual address 00000026
(CCF clk_disable() has an IS_ERR_OR_NULL() check, while CCF clk_enable()
only has a NULL check; pre-CCF implementations may behave differently)
While just checking for PCE_STATUS_ERROR would be sufficient, it doesn't
hurt to use the same state machine as is done for runtime PM, as this
makes the two versions more similar, and eligible for a future
consolidation.
Signed-off-by: Geert Uytterhoeven <[email protected]>
---
This crash started to happen on armadillo-legacy during s2ram if
CONFIG_PM_RUNTIME is not set after applying "[PATCH v2 07/11] ARM:
shmobile: r8a7740/armadillo legacy: Add A4MP pm domain support"
(http://www.spinics.net/linux/lists/arm-kernel/msg365438.html), as
there's no NULL clock for the HDMI device.
Most existing code calling pm_clk_suspend()/pm_clk_resume() is protected
by a check for CONFIG_PM_RUNTIME (davinci, keystone, omap1,
drivers/sh/pm_runtime.c), so it was not affected by this bug.
Exceptions are:
- arch/arm/mach-shmobile/pm-r8a7779.c (marzen),
- arch/arm/mach-shmobile/pm-rmobile.c (r8a7740/armadillo and
sh7372/mackerel),
but it's difficult to assess from the code whether the bug is really
triggered on these platforms.
Grygorii Strashko's "[PATCH v1 2/4] ARM: keystone: pm: switch to use
generic pm domains" is not affected, as pm_clk_add_clk() is only called
for existing clocks.
If it crashes on marzen or mackerel, I think this fix needs to be
applied to stable, too. I don't have access to marzen or mackerel boards,
though.
How to test:
- Build a kernel with CONFIG_PM_SLEEP/CONFIG_SUSPEND enabled, but
CONFIG_PM_RUNTIME disabled,
- echo 0 > /sys/module/printk/parameters/console_suspend,
- echo mem > /sys/power/state,
- wake up using e.g. gpio-keys or serial console activity.
---
drivers/base/power/clock_ops.c | 19 +++++++++++++++----
1 file changed, 15 insertions(+), 4 deletions(-)
diff --git a/drivers/base/power/clock_ops.c b/drivers/base/power/clock_ops.c
index b99e6c06ee678ecb..78369305e0698109 100644
--- a/drivers/base/power/clock_ops.c
+++ b/drivers/base/power/clock_ops.c
@@ -368,8 +368,13 @@ int pm_clk_suspend(struct device *dev)
spin_lock_irqsave(&psd->lock, flags);
- list_for_each_entry_reverse(ce, &psd->clock_list, node)
- clk_disable(ce->clk);
+ list_for_each_entry_reverse(ce, &psd->clock_list, node) {
+ if (ce->status < PCE_STATUS_ERROR) {
+ if (ce->status == PCE_STATUS_ENABLED)
+ clk_disable(ce->clk);
+ ce->status = PCE_STATUS_ACQUIRED;
+ }
+ }
spin_unlock_irqrestore(&psd->lock, flags);
@@ -385,6 +390,7 @@ int pm_clk_resume(struct device *dev)
struct pm_subsys_data *psd = dev_to_psd(dev);
struct pm_clock_entry *ce;
unsigned long flags;
+ int ret;
dev_dbg(dev, "%s()\n", __func__);
@@ -394,8 +400,13 @@ int pm_clk_resume(struct device *dev)
spin_lock_irqsave(&psd->lock, flags);
- list_for_each_entry(ce, &psd->clock_list, node)
- __pm_clk_enable(dev, ce->clk);
+ list_for_each_entry(ce, &psd->clock_list, node) {
+ if (ce->status < PCE_STATUS_ERROR) {
+ ret = __pm_clk_enable(dev, ce->clk);
+ if (!ret)
+ ce->status = PCE_STATUS_ENABLED;
+ }
+ }
spin_unlock_irqrestore(&psd->lock, flags);
--
1.9.1
On Wednesday, October 01, 2014 08:38:17 PM Geert Uytterhoeven wrote:
> Unlike the clocks management code for runtime PM, the code used for
> system suspend does not check the pm_clock_entry.status field.
> If pm_clk_acquire() failed, ce->status will be PCE_STATUS_ERROR, and
> ce->clk will be a negative error code (e.g. 0xfffffffe = -2 = -ENOENT).
>
> Depending on the clock implementation, suspend or resume may crash with:
>
> Unable to handle kernel NULL pointer dereference at virtual address 00000026
>
> (CCF clk_disable() has an IS_ERR_OR_NULL() check, while CCF clk_enable()
> only has a NULL check; pre-CCF implementations may behave differently)
>
> While just checking for PCE_STATUS_ERROR would be sufficient, it doesn't
> hurt to use the same state machine as is done for runtime PM, as this
> makes the two versions more similar, and eligible for a future
> consolidation.
>
> Signed-off-by: Geert Uytterhoeven <[email protected]>
> ---
> This crash started to happen on armadillo-legacy during s2ram if
> CONFIG_PM_RUNTIME is not set after applying "[PATCH v2 07/11] ARM:
> shmobile: r8a7740/armadillo legacy: Add A4MP pm domain support"
> (http://www.spinics.net/linux/lists/arm-kernel/msg365438.html), as
> there's no NULL clock for the HDMI device.
>
> Most existing code calling pm_clk_suspend()/pm_clk_resume() is protected
> by a check for CONFIG_PM_RUNTIME (davinci, keystone, omap1,
> drivers/sh/pm_runtime.c), so it was not affected by this bug.
>
> Exceptions are:
> - arch/arm/mach-shmobile/pm-r8a7779.c (marzen),
> - arch/arm/mach-shmobile/pm-rmobile.c (r8a7740/armadillo and
> sh7372/mackerel),
> but it's difficult to assess from the code whether the bug is really
> triggered on these platforms.
>
> Grygorii Strashko's "[PATCH v1 2/4] ARM: keystone: pm: switch to use
> generic pm domains" is not affected, as pm_clk_add_clk() is only called
> for existing clocks.
>
> If it crashes on marzen or mackerel, I think this fix needs to be
> applied to stable, too. I don't have access to marzen or mackerel boards,
> though.
>
> How to test:
> - Build a kernel with CONFIG_PM_SLEEP/CONFIG_SUSPEND enabled, but
> CONFIG_PM_RUNTIME disabled,
> - echo 0 > /sys/module/printk/parameters/console_suspend,
> - echo mem > /sys/power/state,
> - wake up using e.g. gpio-keys or serial console activity.
Do I think correctly that this would be 3.18 material?
> ---
> drivers/base/power/clock_ops.c | 19 +++++++++++++++----
> 1 file changed, 15 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/base/power/clock_ops.c b/drivers/base/power/clock_ops.c
> index b99e6c06ee678ecb..78369305e0698109 100644
> --- a/drivers/base/power/clock_ops.c
> +++ b/drivers/base/power/clock_ops.c
> @@ -368,8 +368,13 @@ int pm_clk_suspend(struct device *dev)
>
> spin_lock_irqsave(&psd->lock, flags);
>
> - list_for_each_entry_reverse(ce, &psd->clock_list, node)
> - clk_disable(ce->clk);
> + list_for_each_entry_reverse(ce, &psd->clock_list, node) {
> + if (ce->status < PCE_STATUS_ERROR) {
> + if (ce->status == PCE_STATUS_ENABLED)
> + clk_disable(ce->clk);
> + ce->status = PCE_STATUS_ACQUIRED;
> + }
> + }
>
> spin_unlock_irqrestore(&psd->lock, flags);
>
> @@ -385,6 +390,7 @@ int pm_clk_resume(struct device *dev)
> struct pm_subsys_data *psd = dev_to_psd(dev);
> struct pm_clock_entry *ce;
> unsigned long flags;
> + int ret;
>
> dev_dbg(dev, "%s()\n", __func__);
>
> @@ -394,8 +400,13 @@ int pm_clk_resume(struct device *dev)
>
> spin_lock_irqsave(&psd->lock, flags);
>
> - list_for_each_entry(ce, &psd->clock_list, node)
> - __pm_clk_enable(dev, ce->clk);
> + list_for_each_entry(ce, &psd->clock_list, node) {
> + if (ce->status < PCE_STATUS_ERROR) {
> + ret = __pm_clk_enable(dev, ce->clk);
> + if (!ret)
> + ce->status = PCE_STATUS_ENABLED;
> + }
> + }
>
> spin_unlock_irqrestore(&psd->lock, flags);
>
>
--
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.
Hi Rafael,
On Wed, Oct 1, 2014 at 9:47 PM, Rafael J. Wysocki <[email protected]> wrote:
> On Wednesday, October 01, 2014 08:38:17 PM Geert Uytterhoeven wrote:
>> Unlike the clocks management code for runtime PM, the code used for
>> system suspend does not check the pm_clock_entry.status field.
>> If pm_clk_acquire() failed, ce->status will be PCE_STATUS_ERROR, and
>> ce->clk will be a negative error code (e.g. 0xfffffffe = -2 = -ENOENT).
>>
>> Depending on the clock implementation, suspend or resume may crash with:
>>
>> Unable to handle kernel NULL pointer dereference at virtual address 00000026
>>
>> (CCF clk_disable() has an IS_ERR_OR_NULL() check, while CCF clk_enable()
>> only has a NULL check; pre-CCF implementations may behave differently)
>>
>> While just checking for PCE_STATUS_ERROR would be sufficient, it doesn't
>> hurt to use the same state machine as is done for runtime PM, as this
>> makes the two versions more similar, and eligible for a future
>> consolidation.
>>
>> Signed-off-by: Geert Uytterhoeven <[email protected]>
>> ---
>> This crash started to happen on armadillo-legacy during s2ram if
>> CONFIG_PM_RUNTIME is not set after applying "[PATCH v2 07/11] ARM:
>> shmobile: r8a7740/armadillo legacy: Add A4MP pm domain support"
>> (http://www.spinics.net/linux/lists/arm-kernel/msg365438.html), as
>> there's no NULL clock for the HDMI device.
>>
>> Most existing code calling pm_clk_suspend()/pm_clk_resume() is protected
>> by a check for CONFIG_PM_RUNTIME (davinci, keystone, omap1,
>> drivers/sh/pm_runtime.c), so it was not affected by this bug.
>>
>> Exceptions are:
>> - arch/arm/mach-shmobile/pm-r8a7779.c (marzen),
>> - arch/arm/mach-shmobile/pm-rmobile.c (r8a7740/armadillo and
>> sh7372/mackerel),
>> but it's difficult to assess from the code whether the bug is really
>> triggered on these platforms.
>>
>> Grygorii Strashko's "[PATCH v1 2/4] ARM: keystone: pm: switch to use
>> generic pm domains" is not affected, as pm_clk_add_clk() is only called
>> for existing clocks.
>>
>> If it crashes on marzen or mackerel, I think this fix needs to be
>> applied to stable, too. I don't have access to marzen or mackerel boards,
>> though.
>>
>> How to test:
>> - Build a kernel with CONFIG_PM_SLEEP/CONFIG_SUSPEND enabled, but
>> CONFIG_PM_RUNTIME disabled,
>> - echo 0 > /sys/module/printk/parameters/console_suspend,
>> - echo mem > /sys/power/state,
>> - wake up using e.g. gpio-keys or serial console activity.
>
> Do I think correctly that this would be 3.18 material?
Yes indeed, so Simon can queue up the R-Mobile PM domain bits that will
trigger this on armadillo for 3.19.
If it can be triggered on marzen or mackerel now, I think we need it in
stable, too.
Thanks!
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]
In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
Upstream commit a968bed78b549b4c61d4a46e59161fc1f60f96a6
Author: Geert Uytterhoeven <[email protected]>
Date: Wed Oct 1 20:38:17 2014 +0200
PM / clk: Fix crash in clocks management code if !CONFIG_PM_RUNTIME
Simon tried it on mackerel. It did not crash, but caused a warning and
backtrace:
WARNING: CPU: 0 PID: 1420 at drivers/sh/clk/core.c:240
__clk_disable+0x80/0x90()
Trying to disable clock c04bd4d8 with 0 usecount
which got fixed by the patch. So I think it should be applied to
-stable (v3.14 and up).
Thanks!
On Wed, Oct 1, 2014 at 8:38 PM, Geert Uytterhoeven
<[email protected]> wrote:
> Unlike the clocks management code for runtime PM, the code used for
> system suspend does not check the pm_clock_entry.status field.
> If pm_clk_acquire() failed, ce->status will be PCE_STATUS_ERROR, and
> ce->clk will be a negative error code (e.g. 0xfffffffe = -2 = -ENOENT).
>
> Depending on the clock implementation, suspend or resume may crash with:
>
> Unable to handle kernel NULL pointer dereference at virtual address 00000026
>
> (CCF clk_disable() has an IS_ERR_OR_NULL() check, while CCF clk_enable()
> only has a NULL check; pre-CCF implementations may behave differently)
>
> While just checking for PCE_STATUS_ERROR would be sufficient, it doesn't
> hurt to use the same state machine as is done for runtime PM, as this
> makes the two versions more similar, and eligible for a future
> consolidation.
>
> Signed-off-by: Geert Uytterhoeven <[email protected]>
> ---
> This crash started to happen on armadillo-legacy during s2ram if
> CONFIG_PM_RUNTIME is not set after applying "[PATCH v2 07/11] ARM:
> shmobile: r8a7740/armadillo legacy: Add A4MP pm domain support"
> (http://www.spinics.net/linux/lists/arm-kernel/msg365438.html), as
> there's no NULL clock for the HDMI device.
>
> Most existing code calling pm_clk_suspend()/pm_clk_resume() is protected
> by a check for CONFIG_PM_RUNTIME (davinci, keystone, omap1,
> drivers/sh/pm_runtime.c), so it was not affected by this bug.
>
> Exceptions are:
> - arch/arm/mach-shmobile/pm-r8a7779.c (marzen),
> - arch/arm/mach-shmobile/pm-rmobile.c (r8a7740/armadillo and
> sh7372/mackerel),
> but it's difficult to assess from the code whether the bug is really
> triggered on these platforms.
>
> Grygorii Strashko's "[PATCH v1 2/4] ARM: keystone: pm: switch to use
> generic pm domains" is not affected, as pm_clk_add_clk() is only called
> for existing clocks.
>
> If it crashes on marzen or mackerel, I think this fix needs to be
> applied to stable, too. I don't have access to marzen or mackerel boards,
> though.
>
> How to test:
> - Build a kernel with CONFIG_PM_SLEEP/CONFIG_SUSPEND enabled, but
> CONFIG_PM_RUNTIME disabled,
> - echo 0 > /sys/module/printk/parameters/console_suspend,
> - echo mem > /sys/power/state,
> - wake up using e.g. gpio-keys or serial console activity.
> ---
> drivers/base/power/clock_ops.c | 19 +++++++++++++++----
> 1 file changed, 15 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/base/power/clock_ops.c b/drivers/base/power/clock_ops.c
> index b99e6c06ee678ecb..78369305e0698109 100644
> --- a/drivers/base/power/clock_ops.c
> +++ b/drivers/base/power/clock_ops.c
> @@ -368,8 +368,13 @@ int pm_clk_suspend(struct device *dev)
>
> spin_lock_irqsave(&psd->lock, flags);
>
> - list_for_each_entry_reverse(ce, &psd->clock_list, node)
> - clk_disable(ce->clk);
> + list_for_each_entry_reverse(ce, &psd->clock_list, node) {
> + if (ce->status < PCE_STATUS_ERROR) {
> + if (ce->status == PCE_STATUS_ENABLED)
> + clk_disable(ce->clk);
> + ce->status = PCE_STATUS_ACQUIRED;
> + }
> + }
>
> spin_unlock_irqrestore(&psd->lock, flags);
>
> @@ -385,6 +390,7 @@ int pm_clk_resume(struct device *dev)
> struct pm_subsys_data *psd = dev_to_psd(dev);
> struct pm_clock_entry *ce;
> unsigned long flags;
> + int ret;
>
> dev_dbg(dev, "%s()\n", __func__);
>
> @@ -394,8 +400,13 @@ int pm_clk_resume(struct device *dev)
>
> spin_lock_irqsave(&psd->lock, flags);
>
> - list_for_each_entry(ce, &psd->clock_list, node)
> - __pm_clk_enable(dev, ce->clk);
> + list_for_each_entry(ce, &psd->clock_list, node) {
> + if (ce->status < PCE_STATUS_ERROR) {
> + ret = __pm_clk_enable(dev, ce->clk);
> + if (!ret)
> + ce->status = PCE_STATUS_ENABLED;
> + }
> + }
>
> spin_unlock_irqrestore(&psd->lock, flags);
>
> --
> 1.9.1
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]
In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
On Thu, Oct 30, 2014 at 10:06:25AM +0100, Geert Uytterhoeven wrote:
> Upstream commit a968bed78b549b4c61d4a46e59161fc1f60f96a6
> Author: Geert Uytterhoeven <[email protected]>
> Date: Wed Oct 1 20:38:17 2014 +0200
>
> PM / clk: Fix crash in clocks management code if !CONFIG_PM_RUNTIME
>
>
> Simon tried it on mackerel. It did not crash, but caused a warning and
> backtrace:
>
> WARNING: CPU: 0 PID: 1420 at drivers/sh/clk/core.c:240
> __clk_disable+0x80/0x90()
> Trying to disable clock c04bd4d8 with 0 usecount
>
> which got fixed by the patch. So I think it should be applied to
> -stable (v3.14 and up).
>
Thanks, I'm queuing it for the 3.16.
Cheers,
--
Lu?s
> Thanks!
>
> On Wed, Oct 1, 2014 at 8:38 PM, Geert Uytterhoeven
> <[email protected]> wrote:
> > Unlike the clocks management code for runtime PM, the code used for
> > system suspend does not check the pm_clock_entry.status field.
> > If pm_clk_acquire() failed, ce->status will be PCE_STATUS_ERROR, and
> > ce->clk will be a negative error code (e.g. 0xfffffffe = -2 = -ENOENT).
> >
> > Depending on the clock implementation, suspend or resume may crash with:
> >
> > Unable to handle kernel NULL pointer dereference at virtual address 00000026
> >
> > (CCF clk_disable() has an IS_ERR_OR_NULL() check, while CCF clk_enable()
> > only has a NULL check; pre-CCF implementations may behave differently)
> >
> > While just checking for PCE_STATUS_ERROR would be sufficient, it doesn't
> > hurt to use the same state machine as is done for runtime PM, as this
> > makes the two versions more similar, and eligible for a future
> > consolidation.
> >
> > Signed-off-by: Geert Uytterhoeven <[email protected]>
> > ---
> > This crash started to happen on armadillo-legacy during s2ram if
> > CONFIG_PM_RUNTIME is not set after applying "[PATCH v2 07/11] ARM:
> > shmobile: r8a7740/armadillo legacy: Add A4MP pm domain support"
> > (http://www.spinics.net/linux/lists/arm-kernel/msg365438.html), as
> > there's no NULL clock for the HDMI device.
> >
> > Most existing code calling pm_clk_suspend()/pm_clk_resume() is protected
> > by a check for CONFIG_PM_RUNTIME (davinci, keystone, omap1,
> > drivers/sh/pm_runtime.c), so it was not affected by this bug.
> >
> > Exceptions are:
> > - arch/arm/mach-shmobile/pm-r8a7779.c (marzen),
> > - arch/arm/mach-shmobile/pm-rmobile.c (r8a7740/armadillo and
> > sh7372/mackerel),
> > but it's difficult to assess from the code whether the bug is really
> > triggered on these platforms.
> >
> > Grygorii Strashko's "[PATCH v1 2/4] ARM: keystone: pm: switch to use
> > generic pm domains" is not affected, as pm_clk_add_clk() is only called
> > for existing clocks.
> >
> > If it crashes on marzen or mackerel, I think this fix needs to be
> > applied to stable, too. I don't have access to marzen or mackerel boards,
> > though.
> >
> > How to test:
> > - Build a kernel with CONFIG_PM_SLEEP/CONFIG_SUSPEND enabled, but
> > CONFIG_PM_RUNTIME disabled,
> > - echo 0 > /sys/module/printk/parameters/console_suspend,
> > - echo mem > /sys/power/state,
> > - wake up using e.g. gpio-keys or serial console activity.
> > ---
> > drivers/base/power/clock_ops.c | 19 +++++++++++++++----
> > 1 file changed, 15 insertions(+), 4 deletions(-)
> >
> > diff --git a/drivers/base/power/clock_ops.c b/drivers/base/power/clock_ops.c
> > index b99e6c06ee678ecb..78369305e0698109 100644
> > --- a/drivers/base/power/clock_ops.c
> > +++ b/drivers/base/power/clock_ops.c
> > @@ -368,8 +368,13 @@ int pm_clk_suspend(struct device *dev)
> >
> > spin_lock_irqsave(&psd->lock, flags);
> >
> > - list_for_each_entry_reverse(ce, &psd->clock_list, node)
> > - clk_disable(ce->clk);
> > + list_for_each_entry_reverse(ce, &psd->clock_list, node) {
> > + if (ce->status < PCE_STATUS_ERROR) {
> > + if (ce->status == PCE_STATUS_ENABLED)
> > + clk_disable(ce->clk);
> > + ce->status = PCE_STATUS_ACQUIRED;
> > + }
> > + }
> >
> > spin_unlock_irqrestore(&psd->lock, flags);
> >
> > @@ -385,6 +390,7 @@ int pm_clk_resume(struct device *dev)
> > struct pm_subsys_data *psd = dev_to_psd(dev);
> > struct pm_clock_entry *ce;
> > unsigned long flags;
> > + int ret;
> >
> > dev_dbg(dev, "%s()\n", __func__);
> >
> > @@ -394,8 +400,13 @@ int pm_clk_resume(struct device *dev)
> >
> > spin_lock_irqsave(&psd->lock, flags);
> >
> > - list_for_each_entry(ce, &psd->clock_list, node)
> > - __pm_clk_enable(dev, ce->clk);
> > + list_for_each_entry(ce, &psd->clock_list, node) {
> > + if (ce->status < PCE_STATUS_ERROR) {
> > + ret = __pm_clk_enable(dev, ce->clk);
> > + if (!ret)
> > + ce->status = PCE_STATUS_ENABLED;
> > + }
> > + }
> >
> > spin_unlock_irqrestore(&psd->lock, flags);
> >
> > --
> > 1.9.1
>
> Gr{oetje,eeting}s,
>
> Geert
>
> --
> Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]
>
> In personal conversations with technical people, I call myself a hacker. But
> when I'm talking to journalists I just say "programmer" or something like that.
> -- Linus Torvalds
> --
> To unsubscribe from this list: send the line "unsubscribe stable" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html