When booting up preempt rt kernel on stratix10 platform, there is
below calltrace:
BUG: sleeping function called from invalid context at kernel/locking/rtmutex.c:969
in_atomic(): 1, irqs_disabled(): 128, non_block: 0, pid: 1, name: swapper/0
Preemption disabled at:
[<ffff8000100b25b0>] __setup_irq+0xe0/0x730
CPU: 0 PID: 1 Comm: swapper/0 Not tainted 5.10.78-rt54-yocto-preempt-rt #1
Hardware name: SoCFPGA Stratix 10 SoCDK (DT)
Call trace:
......
___might_sleep+0x144/0x180
rt_spin_lock+0x34/0x9c
regmap_lock_spinlock+0x24/0x40
regmap_write+0x48/0x84
a10_eccmgr_irq_unmask+0x34/0x40
......
altr_edac_a10_probe+0x16c/0x2b0
platform_drv_probe+0x60/0xb4
......
ret_from_fork+0x10/0x38
Because regmap_write is invoked under preemption disabled status, and
might trigger sleep.
Recently, the commit 67021f25d952("regmap: teach regmap to use raw
spinlocks if requested in the config ") add an option for regmap to use
raw spinlock. So, enable raw spinlock in preempt-rt kernel to avoid the
might sleep issue.
Note: The commit 67021f25d952 is only in kernel v5.15, not included in
earlier versions. So, if intend to fix this issue in earlier preempt-rt
kernel, it is need to backport the 2 patches together, otherwise there
will be building issue.
Signed-off-by: Meng Li <[email protected]>
---
v2:
- improve the patch title.
- improve the commit log. Clear which commit is depended by this patch.
---
drivers/mfd/altera-sysmgr.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/drivers/mfd/altera-sysmgr.c b/drivers/mfd/altera-sysmgr.c
index 5d3715a28b28..27271cec5d53 100644
--- a/drivers/mfd/altera-sysmgr.c
+++ b/drivers/mfd/altera-sysmgr.c
@@ -83,6 +83,9 @@ static struct regmap_config altr_sysmgr_regmap_cfg = {
.fast_io = true,
.use_single_read = true,
.use_single_write = true,
+#ifdef CONFIG_PREEMPT_RT
+ .use_raw_spinlock = true,
+#endif
};
/**
--
2.17.1
On Tue, Nov 16, 2021 at 11:54 AM Meng Li <[email protected]> wrote:
> diff --git a/drivers/mfd/altera-sysmgr.c b/drivers/mfd/altera-sysmgr.c
> index 5d3715a28b28..27271cec5d53 100644
> --- a/drivers/mfd/altera-sysmgr.c
> +++ b/drivers/mfd/altera-sysmgr.c
> @@ -83,6 +83,9 @@ static struct regmap_config altr_sysmgr_regmap_cfg = {
> .fast_io = true,
> .use_single_read = true,
> .use_single_write = true,
> +#ifdef CONFIG_PREEMPT_RT
> + .use_raw_spinlock = true,
> +#endif
I think you should remove the #ifdef here: if PREEMPT_RT is disabled, the flag
has no effect because spinlock behaves the same way as raw_spinlock. If
anything else starts requiring the use of raw spinlocks, then we probably
want the flag to be set here as well.
Arnd
> -----Original Message-----
> From: Arnd Bergmann <[email protected]>
> Sent: Tuesday, November 16, 2021 8:02 PM
> To: Li, Meng <[email protected]>
> Cc: [email protected]; Lee Jones <[email protected]>; Arnd
> Bergmann <[email protected]>; Linux Kernel Mailing List <linux-
> [email protected]>
> Subject: Re: [PATCH] mfd: altera-sysmgr: enable raw spinlock feature for
> preempt-rt kernel
>
> [Please note: This e-mail is from an EXTERNAL e-mail address]
>
> On Tue, Nov 16, 2021 at 11:54 AM Meng Li <[email protected]> wrote:
> > diff --git a/drivers/mfd/altera-sysmgr.c b/drivers/mfd/altera-sysmgr.c
> > index 5d3715a28b28..27271cec5d53 100644
> > --- a/drivers/mfd/altera-sysmgr.c
> > +++ b/drivers/mfd/altera-sysmgr.c
> > @@ -83,6 +83,9 @@ static struct regmap_config altr_sysmgr_regmap_cfg =
> {
> > .fast_io = true,
> > .use_single_read = true,
> > .use_single_write = true,
> > +#ifdef CONFIG_PREEMPT_RT
> > + .use_raw_spinlock = true,
> > +#endif
>
> I think you should remove the #ifdef here: if PREEMPT_RT is disabled, the
> flag has no effect because spinlock behaves the same way as raw_spinlock. If
> anything else starts requiring the use of raw spinlocks, then we probably
> want the flag to be set here as well.
>
Thanks for your suggestion, and I also agree with the spinlock action when PREEMPT_RT is disabled.
But please allow me to explain why I keep the "ifdef"
1. although I send this patch to mainline upstream, I only want to fix this issue in RT kernel.
Moreover, the commit 67021f25d952("regmap: teach regmap to use raw spinlocks if requested in the config ") is also for RT kernel even if it doesn't use "ifdef CONFIG_PREEMPT_RT"
My ideal is that if this patch is merged into mainline, Linux-rt maintainer will not spend extra effort to focus on this patch. After all, this fixing is more related with driver.
In addition, I found out there are other patches with "ifdef CONFIG_PREEMPT_RT" merged by mainline, so I also send this patch to mainline, not Linux-rt.
2. I check regmap.c code that is related with use_raw_spinlock. If PREEMPT_RT is disabled and use_raw_spinlock is set as true, the else case will not be entered any longer.
In other words, in mainline standard kernel, if use_raw_spinlock is set as true, raw spinlock will be used forever, and the code in else case will become useless.
I feel it is a little unreasonable. So, I keep the "ifdef"
if ((bus && bus->fast_io) ||
config->fast_io) {
if (config->use_raw_spinlock) {
raw_spin_lock_init(&map->raw_spinlock);
map->lock = regmap_lock_raw_spinlock;
map->unlock = regmap_unlock_raw_spinlock;
lockdep_set_class_and_name(&map->raw_spinlock,
lock_key, lock_name);
} else {
spin_lock_init(&map->spinlock);
map->lock = regmap_lock_spinlock;
map->unlock = regmap_unlock_spinlock;
lockdep_set_class_and_name(&map->spinlock,
lock_key, lock_name);
}
} else {
mutex_init(&map->mutex);
map->lock = regmap_lock_mutex;
map->unlock = regmap_unlock_mutex;
map->can_sleep = true;
lockdep_set_class_and_name(&map->mutex,
lock_key, lock_name);
}
Thanks,
Limeng
> Arnd
On Tue, 16 Nov 2021, Li, Meng wrote:
>
>
> > -----Original Message-----
> > From: Arnd Bergmann <[email protected]>
> > Sent: Tuesday, November 16, 2021 8:02 PM
> > To: Li, Meng <[email protected]>
> > Cc: [email protected]; Lee Jones <[email protected]>; Arnd
> > Bergmann <[email protected]>; Linux Kernel Mailing List <linux-
> > [email protected]>
> > Subject: Re: [PATCH] mfd: altera-sysmgr: enable raw spinlock feature for
> > preempt-rt kernel
> >
> > [Please note: This e-mail is from an EXTERNAL e-mail address]
> >
> > On Tue, Nov 16, 2021 at 11:54 AM Meng Li <[email protected]> wrote:
> > > diff --git a/drivers/mfd/altera-sysmgr.c b/drivers/mfd/altera-sysmgr.c
> > > index 5d3715a28b28..27271cec5d53 100644
> > > --- a/drivers/mfd/altera-sysmgr.c
> > > +++ b/drivers/mfd/altera-sysmgr.c
> > > @@ -83,6 +83,9 @@ static struct regmap_config altr_sysmgr_regmap_cfg =
> > {
> > > .fast_io = true,
> > > .use_single_read = true,
> > > .use_single_write = true,
> > > +#ifdef CONFIG_PREEMPT_RT
> > > + .use_raw_spinlock = true,
> > > +#endif
> >
> > I think you should remove the #ifdef here: if PREEMPT_RT is disabled, the
> > flag has no effect because spinlock behaves the same way as raw_spinlock. If
> > anything else starts requiring the use of raw spinlocks, then we probably
> > want the flag to be set here as well.
> >
>
> Thanks for your suggestion, and I also agree with the spinlock action when PREEMPT_RT is disabled.
> But please allow me to explain why I keep the "ifdef"
> 1. although I send this patch to mainline upstream, I only want to fix this issue in RT kernel.
> Moreover, the commit 67021f25d952("regmap: teach regmap to use raw spinlocks if requested in the config ") is also for RT kernel even if it doesn't use "ifdef CONFIG_PREEMPT_RT"
> My ideal is that if this patch is merged into mainline, Linux-rt maintainer will not spend extra effort to focus on this patch. After all, this fixing is more related with driver.
> In addition, I found out there are other patches with "ifdef CONFIG_PREEMPT_RT" merged by mainline, so I also send this patch to mainline, not Linux-rt.
>
> 2. I check regmap.c code that is related with use_raw_spinlock. If PREEMPT_RT is disabled and use_raw_spinlock is set as true, the else case will not be entered any longer.
> In other words, in mainline standard kernel, if use_raw_spinlock is set as true, raw spinlock will be used forever, and the code in else case will become useless.
> I feel it is a little unreasonable. So, I keep the "ifdef"
> if ((bus && bus->fast_io) ||
> config->fast_io) {
> if (config->use_raw_spinlock) {
> raw_spin_lock_init(&map->raw_spinlock);
> map->lock = regmap_lock_raw_spinlock;
> map->unlock = regmap_unlock_raw_spinlock;
> lockdep_set_class_and_name(&map->raw_spinlock,
> lock_key, lock_name);
> } else {
> spin_lock_init(&map->spinlock);
> map->lock = regmap_lock_spinlock;
> map->unlock = regmap_unlock_spinlock;
> lockdep_set_class_and_name(&map->spinlock,
> lock_key, lock_name);
> }
> } else {
> mutex_init(&map->mutex);
> map->lock = regmap_lock_mutex;
> map->unlock = regmap_unlock_mutex;
> map->can_sleep = true;
> lockdep_set_class_and_name(&map->mutex,
> lock_key, lock_name);
> }
>
I dislike #ifery as a general rule. So with that in mind - if it's
not required, I'd prefer that it's removed.
--
Lee Jones [李琼斯]
Senior Technical Lead - Developer Services
Linaro.org │ Open source software for Arm SoCs
Follow Linaro: Facebook | Twitter | Blog
> -----Original Message-----
> From: Lee Jones <[email protected]>
> Sent: Tuesday, November 16, 2021 11:39 PM
> To: Li, Meng <[email protected]>
> Cc: Arnd Bergmann <[email protected]>; [email protected]; Linux
> Kernel Mailing List <[email protected]>
> Subject: Re: [PATCH] mfd: altera-sysmgr: enable raw spinlock feature for
> preempt-rt kernel
>
> [Please note: This e-mail is from an EXTERNAL e-mail address]
>
> On Tue, 16 Nov 2021, Li, Meng wrote:
>
> >
> >
> > > -----Original Message-----
> > > From: Arnd Bergmann <[email protected]>
> > > Sent: Tuesday, November 16, 2021 8:02 PM
> > > To: Li, Meng <[email protected]>
> > > Cc: [email protected]; Lee Jones <[email protected]>;
> > > Arnd Bergmann <[email protected]>; Linux Kernel Mailing List <linux-
> > > [email protected]>
> > > Subject: Re: [PATCH] mfd: altera-sysmgr: enable raw spinlock feature
> > > for preempt-rt kernel
> > >
> > > [Please note: This e-mail is from an EXTERNAL e-mail address]
> > >
> > > On Tue, Nov 16, 2021 at 11:54 AM Meng Li <[email protected]>
> wrote:
> > > > diff --git a/drivers/mfd/altera-sysmgr.c
> > > > b/drivers/mfd/altera-sysmgr.c index 5d3715a28b28..27271cec5d53
> > > > 100644
> > > > --- a/drivers/mfd/altera-sysmgr.c
> > > > +++ b/drivers/mfd/altera-sysmgr.c
> > > > @@ -83,6 +83,9 @@ static struct regmap_config
> > > > altr_sysmgr_regmap_cfg =
> > > {
> > > > .fast_io = true,
> > > > .use_single_read = true,
> > > > .use_single_write = true,
> > > > +#ifdef CONFIG_PREEMPT_RT
> > > > + .use_raw_spinlock = true,
> > > > +#endif
> > >
> > > I think you should remove the #ifdef here: if PREEMPT_RT is
> > > disabled, the flag has no effect because spinlock behaves the same
> > > way as raw_spinlock. If anything else starts requiring the use of
> > > raw spinlocks, then we probably want the flag to be set here as well.
> > >
> >
> > Thanks for your suggestion, and I also agree with the spinlock action when
> PREEMPT_RT is disabled.
> > But please allow me to explain why I keep the "ifdef"
> > 1. although I send this patch to mainline upstream, I only want to fix this
> issue in RT kernel.
> > Moreover, the commit 67021f25d952("regmap: teach regmap to use raw
> spinlocks if requested in the config ") is also for RT kernel even if it doesn't
> use "ifdef CONFIG_PREEMPT_RT"
> > My ideal is that if this patch is merged into mainline, Linux-rt maintainer
> will not spend extra effort to focus on this patch. After all, this fixing is more
> related with driver.
> > In addition, I found out there are other patches with "ifdef
> CONFIG_PREEMPT_RT" merged by mainline, so I also send this patch to
> mainline, not Linux-rt.
> >
> > 2. I check regmap.c code that is related with use_raw_spinlock. If
> PREEMPT_RT is disabled and use_raw_spinlock is set as true, the else case
> will not be entered any longer.
> > In other words, in mainline standard kernel, if use_raw_spinlock is set as
> true, raw spinlock will be used forever, and the code in else case will become
> useless.
> > I feel it is a little unreasonable. So, I keep the "ifdef"
> > if ((bus && bus->fast_io) ||
> > config->fast_io) {
> > if (config->use_raw_spinlock) {
> > raw_spin_lock_init(&map->raw_spinlock);
> > map->lock = regmap_lock_raw_spinlock;
> > map->unlock = regmap_unlock_raw_spinlock;
> > lockdep_set_class_and_name(&map->raw_spinlock,
> > lock_key, lock_name);
> > } else {
> > spin_lock_init(&map->spinlock);
> > map->lock = regmap_lock_spinlock;
> > map->unlock = regmap_unlock_spinlock;
> > lockdep_set_class_and_name(&map->spinlock,
> > lock_key, lock_name);
> > }
> > } else {
> > mutex_init(&map->mutex);
> > map->lock = regmap_lock_mutex;
> > map->unlock = regmap_unlock_mutex;
> > map->can_sleep = true;
> > lockdep_set_class_and_name(&map->mutex,
> > lock_key, lock_name);
> > }
> >
>
> I dislike #ifery as a general rule. So with that in mind - if it's not required, I'd
> prefer that it's removed.
>
Ok！
There is no real difference if remove the #ifery.
I will check the standard kernel and then sent v3 RR.
Thanks,
Limeng
> --
> Lee Jones [李琼斯]
> Senior Technical Lead - Developer Services Linaro.org │ Open source
> software for Arm SoCs Follow Linaro: Facebook | Twitter | Blog