[Xenomai-help] kernel warnings and oopses calling gpiolib functions from xenomai kernel space

[Xenomai-help] kernel warnings and oopses calling gpiolib functions from xenomai kernel space

[Jeremy Brown]

We're having what we suspect are related problems with two simple
xenomai experiments.  Has anyone here seen these, and/or have a
workaround?

---------------
Problem 1:

Trying to set the value of a GPIO pin from a periodic RT task in a
kernel module, we're getting a kernel warning.  (This is on an
ARM-based BeagleBoard.)   The task function is simple:
void timer_func(void* arg) {
  rt_task_set_periodic(NULL, TM_NOW, (period_usecs / 2) * 1000LL);
  while (1) {
    rt_task_wait_period(NULL);
    output_value = !output_value;
    gpio_set_value(gpio_out, output_value);
  }
}

The warning is:
[ 1411.807952] ------------[ cut here ]------------
[ 1411.812652] WARNING: at kernel/sched.c:5377 gpio_set+0x38/0xac()
[ 1411.818695] Modules linked in: xk_gpio_writer mailbox_mach rtc_twl
pegasus mailbox rtc_core sg [last unloaded: xuk_gpio_driver]
[ 1411.830383] [<c00448d0>] (unwind_backtrace+0x0/0xe8) from
[<c006bce0>] (warn_slowpath_common+0x48/0x60)
[ 1411.839874] [<c006bce0>] (warn_slowpath_common+0x48/0x60) from
[<c0056b18>] (gpio_set+0x38/0xac)
[ 1411.848754] [<c0056b18>] (gpio_set+0x38/0xac) from [<c0321084>]
(__gpio_set_value+0x30/0x34)
[ 1411.857269] [<c0321084>] (__gpio_set_value+0x30/0x34) from
[<bf0411b4>] (timer_func+0x58/0x64 [xk_gpio_writer])
[ 1411.867462] [<bf0411b4>] (timer_func+0x58/0x64 [xk_gpio_writer])
from [<c00c29d8>] (xnarch_thread_trampoline+0x20/0x2c)
[ 1411.878356] [<c00c29d8>] (xnarch_thread_trampoline+0x20/0x2c) from
[<c00c155c>] (__xnpod_schedule+0x290/0x818)
[ 1411.888427] [<c00c155c>] (__xnpod_schedule+0x290/0x818) from
[<ffffffff>] (0xffffffff)
[ 1411.896423] ---[ end trace 3a72a39f63a09558 ]---

We have verified that gpio_cansleep is 0 for the selected GPIO, so it
should be safe to use it from an interrupt context.   We also have a
similar non-xenomai kernel module that uses an hrtimer, and it causes
no warnings.  The period_usecs we use is 7071us for both experiments.

--------------
Problem 2:

Trying to read or set the value of GPIO pins from an RT kernel module
interrupt handler, we get occasional kernel oopses which get
attributed to randomly-selected processes.  We believe this is
happening under heavy IRQ pressure.

The code is:
int irq_func(rtdm_irq_t *irq_handle) {
  int output_value = gpio_get_value(gpio_in);
  gpio_set_value(gpio_out, output_value);
  return RTDM_IRQ_HANDLED;
}

An example oops, which happened to a /bin/sleep process, is:
[16163.048004] Unable to handle kernel paging request at virtual
address 40076f4c
[16163.055328] pgd = c1300000
[16163.058074] [40076f4c] *pgd=86bd6031, *pte=00000000, *ppte=00000000
[16163.064422] Internal error: Oops: 80000007 [#1] PREEMPT
[16163.069671] last sysfs file: /sys/class/gpio/unexport
[16163.074768] Modules linked in: xk_gpio_responder rtc_twl
mailbox_mach pegasus mailbox rtc_core sg [last unloaded:
xuk_gpio_driver]
[16163.086669] CPU: 0    Tainted: G        W   (2.6.33.5-l3rix20100618 #3)
[16163.093353] PC is at 0x40076f4c
[16163.096527] LR is at 0xa5b9
[16163.099334] pc : [<40076f4c>]    lr : [<0000a5b9>]    psr: 20000030
[16163.099334] sp : beaa06e8  ip : 40076f4d  fp : 00000000
[16163.110900] r10: 00000001  r9 : 00000000  r8 : beaa0730
[16163.116180] r7 : 40113c10  r6 : 40113000  r5 : 40113454  r4 : 401135d0
[16163.122741] r3 : 000150f4  r2 : 00000004  r1 : 00000000  r0 : 401135d0
[16163.129333] Flags: nzCv  IRQs on  FIQs on  Mode USER_32  ISA Thumb
Segment user
[16163.136779] Control: 10c5387d  Table: 81300019  DAC: 00000015
[16163.142578] Process sleep (pid: 20005, stack limit = 0xc136e2f0)
[16163.148620] ---[ end trace fb7968b6c29437fc ]---
[16163.153289] note: sleep[20005] exited with preempt_count 1

Again, a similar non-xenomai linux interrupt handler does not seem to
cause this problem under the same IRQ load.

--------------
System info:
uname -a:
"Linux beagleboard 2.6.33-5-l3rix20100618 #3 PREEMPT Fri Jun 18
17:00:34 EDT 2010 armv7l GNU/Linux"
Xenomai version is a git checkout, master branch of
git://xenomai.org/xenomai-2.5.git
commit a85eb493aad1dd9712be657cd397833f2ae8e9e4
committed on 2010-05-26

--------------

Thanks in advance for any and all help!

Jeremy

Re: [Xenomai-help] kernel warnings and oopses calling gpiolib functions from xenomai kernel space

[Gilles Chanteperdrix]

Jeremy Brown wrote:
> Thanks in advance for any and all help!

Hi,

from your post, we can conclude that gpio_set_value is a Linux kernel
function which can yet not be used in Xenomai context (you could
probably even get more explicit messages by enabling I-pipe debugging).
I am going to try and propose a fix for this issue, will send you a
patch ASAP.

Note that the issue is more an Adeos I-pipe issue than a Xenomai issue,
so I am more interested in knowing what version of adeos than what
version of Xenomai you are using. But chances that the problem exists
with all versions anyway.

Regards.

-- 
					    Gilles.

Re: [Xenomai-help] kernel warnings and oopses calling gpiolib functions from xenomai kernel space

[Gilles Chanteperdrix]

Gilles Chanteperdrix wrote:
> Jeremy Brown wrote:
>> Thanks in advance for any and all help!
> 
> Hi,
> 
> from your post, we can conclude that gpio_set_value is a Linux kernel
> function which can yet not be used in Xenomai context (you could
> probably even get more explicit messages by enabling I-pipe debugging).
> I am going to try and propose a fix for this issue, will send you a
> patch ASAP.
> 
> Note that the issue is more an Adeos I-pipe issue than a Xenomai issue,
> so I am more interested in knowing what version of adeos than what
> version of Xenomai you are using. But chances that the problem exists
> with all versions anyway.

Could you try the following patch?

diff --git a/arch/arm/plat-omap/gpio.c b/arch/arm/plat-omap/gpio.c
index 589a4bb..f1d85ff 100644
--- a/arch/arm/plat-omap/gpio.c
+++ b/arch/arm/plat-omap/gpio.c
@@ -195,7 +195,7 @@ struct gpio_bank {
 #endif
 	u32 level_mask;
 	u32 toggle_mask;
-	spinlock_t lock;
+	ipipe_spinlock_t lock;
 	struct gpio_chip chip;
 	struct clk *dbck;
 	u32 mod_usage;

-- 
					    Gilles.

Re: [Xenomai-help] kernel warnings and oopses calling gpiolib functions from xenomai kernel space

[Jeremy Brown]

Will do, but with the holiday here, probably no earlier than Tuesday.

Thanks for the quick response!

Jeremy


On Sat, Jul 3, 2010 at 10:20 AM, Gilles Chanteperdrix
<gilles.chanteperdrix@xenomai.org> wrote:
> Gilles Chanteperdrix wrote:
>> Jeremy Brown wrote:
>>> Thanks in advance for any and all help!
>>
>> Hi,
>>
>> from your post, we can conclude that gpio_set_value is a Linux kernel
>> function which can yet not be used in Xenomai context (you could
>> probably even get more explicit messages by enabling I-pipe debugging).
>> I am going to try and propose a fix for this issue, will send you a
>> patch ASAP.
>>
>> Note that the issue is more an Adeos I-pipe issue than a Xenomai issue,
>> so I am more interested in knowing what version of adeos than what
>> version of Xenomai you are using. But chances that the problem exists
>> with all versions anyway.
>
> Could you try the following patch?
>
> diff --git a/arch/arm/plat-omap/gpio.c b/arch/arm/plat-omap/gpio.c
> index 589a4bb..f1d85ff 100644
> --- a/arch/arm/plat-omap/gpio.c
> +++ b/arch/arm/plat-omap/gpio.c
> @@ -195,7 +195,7 @@ struct gpio_bank {
>  #endif
>        u32 level_mask;
>        u32 toggle_mask;
> -       spinlock_t lock;
> +       ipipe_spinlock_t lock;
>        struct gpio_chip chip;
>        struct clk *dbck;
>        u32 mod_usage;
>
> --
>                                            Gilles.
>

Re: [Xenomai-help] kernel warnings and oopses calling gpiolib functions from xenomai kernel space

[Jeremy Brown]

To the best of our ability to tell, this patch resolved our problem
--- thank you!

Jeremy

PS The pipe patch version is whatever prepare-kernel applied by
default, which I think for our xenomai checkout and kernel version is
adeos-ipipe-2.6.33-arm-1.16-01.patch

On Sat, Jul 3, 2010 at 10:20 AM, Gilles Chanteperdrix
<gilles.chanteperdrix@xenomai.org> wrote:
> Gilles Chanteperdrix wrote:
>> Jeremy Brown wrote:
>>> Thanks in advance for any and all help!
>>
>> Hi,
>>
>> from your post, we can conclude that gpio_set_value is a Linux kernel
>> function which can yet not be used in Xenomai context (you could
>> probably even get more explicit messages by enabling I-pipe debugging).
>> I am going to try and propose a fix for this issue, will send you a
>> patch ASAP.
>>
>> Note that the issue is more an Adeos I-pipe issue than a Xenomai issue,
>> so I am more interested in knowing what version of adeos than what
>> version of Xenomai you are using. But chances that the problem exists
>> with all versions anyway.
>
> Could you try the following patch?
>
> diff --git a/arch/arm/plat-omap/gpio.c b/arch/arm/plat-omap/gpio.c
> index 589a4bb..f1d85ff 100644
> --- a/arch/arm/plat-omap/gpio.c
> +++ b/arch/arm/plat-omap/gpio.c
> @@ -195,7 +195,7 @@ struct gpio_bank {
>  #endif
>        u32 level_mask;
>        u32 toggle_mask;
> -       spinlock_t lock;
> +       ipipe_spinlock_t lock;
>        struct gpio_chip chip;
>        struct clk *dbck;
>        u32 mod_usage;
>
> --
>                                            Gilles.
>