On some boards is WDI logic of max6370 chip connected via GPIO.
So extend max63xx_wdt driver to allow specifying WDI logic via GPIO.
Signed-off-by: Pali Rohár <[email protected]>
---
Changes in v2:
* Usage of dev_err_probe()
* Fixing assignment of wdt->ping
* Remove clearing of wdt->gpio_wdi
* Move YAML change to separate patch
---
drivers/watchdog/max63xx_wdt.c | 24 ++++++++++++++++++++++++
1 file changed, 24 insertions(+)
diff --git a/drivers/watchdog/max63xx_wdt.c b/drivers/watchdog/max63xx_wdt.c
index 9e1541cfae0d..6e43f9e6d7eb 100644
--- a/drivers/watchdog/max63xx_wdt.c
+++ b/drivers/watchdog/max63xx_wdt.c
@@ -27,6 +27,7 @@
#include <linux/io.h>
#include <linux/slab.h>
#include <linux/property.h>
+#include <linux/gpio/consumer.h>
#define DEFAULT_HEARTBEAT 60
#define MAX_HEARTBEAT 60
@@ -53,6 +54,9 @@ struct max63xx_wdt {
void __iomem *base;
spinlock_t lock;
+ /* GPIOs */
+ struct gpio_desc *gpio_wdi;
+
/* WDI and WSET bits write access routines */
void (*ping)(struct max63xx_wdt *wdt);
void (*set)(struct max63xx_wdt *wdt, u8 set);
@@ -158,6 +162,17 @@ static const struct watchdog_info max63xx_wdt_info = {
.identity = "max63xx Watchdog",
};
+static void max63xx_gpio_ping(struct max63xx_wdt *wdt)
+{
+ spin_lock(&wdt->lock);
+
+ gpiod_set_value_cansleep(wdt->gpio_wdi, 1);
+ udelay(1);
+ gpiod_set_value_cansleep(wdt->gpio_wdi, 0);
+
+ spin_unlock(&wdt->lock);
+}
+
static void max63xx_mmap_ping(struct max63xx_wdt *wdt)
{
u8 val;
@@ -225,10 +240,19 @@ static int max63xx_wdt_probe(struct platform_device *pdev)
return -EINVAL;
}
+ wdt->gpio_wdi = devm_gpiod_get(dev, NULL, GPIOD_FLAGS_BIT_DIR_OUT);
+ if (IS_ERR(wdt->gpio_wdi) && PTR_ERR(wdt->gpio_wdi) != -ENOENT)
+ return dev_err_probe(dev, PTR_ERR(wdt->gpio_wdi),
+ "unable to request gpio: %ld\n",
+ PTR_ERR(wdt->gpio_wdi));
+
err = max63xx_mmap_init(pdev, wdt);
if (err)
return err;
+ if (!IS_ERR(wdt->gpio_wdi))
+ wdt->ping = max63xx_gpio_ping;
+
platform_set_drvdata(pdev, &wdt->wdd);
watchdog_set_drvdata(&wdt->wdd, wdt);
--
2.20.1
On Fri, Apr 29, 2022 at 03:13:49PM +0200, Pali Roh?r wrote:
> @@ -27,6 +27,7 @@
> #include <linux/io.h>
> #include <linux/slab.h>
> #include <linux/property.h>
> +#include <linux/gpio/consumer.h>
It would be better to keep them alphabetically. Anyway, they aren't sorted
originally...
> +static void max63xx_gpio_ping(struct max63xx_wdt *wdt)
> +{
> + spin_lock(&wdt->lock);
Does it really need to acquire the lock? It looks like the lock is to prevent
concurrent accesses to the mmap in max63xx_mmap_ping() and max63xx_mmap_set().
> + gpiod_set_value_cansleep(wdt->gpio_wdi, 1);
> + udelay(1);
Doesn't it need to include <linux/delay.h> for udelay()?
> @@ -225,10 +240,19 @@ static int max63xx_wdt_probe(struct platform_device *pdev)
> return -EINVAL;
> }
>
> + wdt->gpio_wdi = devm_gpiod_get(dev, NULL, GPIOD_FLAGS_BIT_DIR_OUT);
> + if (IS_ERR(wdt->gpio_wdi) && PTR_ERR(wdt->gpio_wdi) != -ENOENT)
Use devm_gpiod_get_optional() to make the intent clear. Also, it gets rid of
the check for -ENOENT.
> + return dev_err_probe(dev, PTR_ERR(wdt->gpio_wdi),
> + "unable to request gpio: %ld\n",
> + PTR_ERR(wdt->gpio_wdi));
It doesn't need to again print for PTR_ERR(wdt->gpio_wdi). dev_err_probe()
prints the error.
> err = max63xx_mmap_init(pdev, wdt);
> if (err)
> return err;
>
> + if (!IS_ERR(wdt->gpio_wdi))
> + wdt->ping = max63xx_gpio_ping;
Thus, the max63xx_gpio_ping() overrides max63xx_mmap_ping() if the GPIO was
provided? It would be better to mention the behavior in the commit message.
Also, could both the assignments of `wdt->gpio_wdi` and `wdt->ping` happen
after max63xx_mmap_init()?
On 5/2/22 20:57, Tzung-Bi Shih wrote:
> On Fri, Apr 29, 2022 at 03:13:49PM +0200, Pali Rohár wrote:
>> @@ -27,6 +27,7 @@
>> #include <linux/io.h>
>> #include <linux/slab.h>
>> #include <linux/property.h>
>> +#include <linux/gpio/consumer.h>
>
> It would be better to keep them alphabetically. Anyway, they aren't sorted
> originally...
>
>> +static void max63xx_gpio_ping(struct max63xx_wdt *wdt)
>> +{
>> + spin_lock(&wdt->lock);
>
> Does it really need to acquire the lock? It looks like the lock is to prevent
> concurrent accesses to the mmap in max63xx_mmap_ping() and max63xx_mmap_set().
>
Actually, that doesn't work at all. spin_lock() directly contradicts
with gpiod_set_value_cansleep().
>> + gpiod_set_value_cansleep(wdt->gpio_wdi, 1);
>> + udelay(1);
>
> Doesn't it need to include <linux/delay.h> for udelay()?
>
>> @@ -225,10 +240,19 @@ static int max63xx_wdt_probe(struct platform_device *pdev)
>> return -EINVAL;
>> }
>>
>> + wdt->gpio_wdi = devm_gpiod_get(dev, NULL, GPIOD_FLAGS_BIT_DIR_OUT);
>> + if (IS_ERR(wdt->gpio_wdi) && PTR_ERR(wdt->gpio_wdi) != -ENOENT)
>
> Use devm_gpiod_get_optional() to make the intent clear. Also, it gets rid of
> the check for -ENOENT.
>
>> + return dev_err_probe(dev, PTR_ERR(wdt->gpio_wdi),
>> + "unable to request gpio: %ld\n",
>> + PTR_ERR(wdt->gpio_wdi));
>
> It doesn't need to again print for PTR_ERR(wdt->gpio_wdi). dev_err_probe()
> prints the error.
>
>> err = max63xx_mmap_init(pdev, wdt);
>> if (err)
>> return err;
>>
>> + if (!IS_ERR(wdt->gpio_wdi))
>> + wdt->ping = max63xx_gpio_ping;
>
> Thus, the max63xx_gpio_ping() overrides max63xx_mmap_ping() if the GPIO was
> provided? It would be better to mention the behavior in the commit message.
>
> Also, could both the assignments of `wdt->gpio_wdi` and `wdt->ping` happen
> after max63xx_mmap_init()?
On Monday 02 May 2022 21:37:16 Guenter Roeck wrote:
> On 5/2/22 20:57, Tzung-Bi Shih wrote:
> > On Fri, Apr 29, 2022 at 03:13:49PM +0200, Pali Rohár wrote:
> > > @@ -27,6 +27,7 @@
> > > #include <linux/io.h>
> > > #include <linux/slab.h>
> > > #include <linux/property.h>
> > > +#include <linux/gpio/consumer.h>
> >
> > It would be better to keep them alphabetically. Anyway, they aren't sorted
> > originally...
> >
> > > +static void max63xx_gpio_ping(struct max63xx_wdt *wdt)
> > > +{
> > > + spin_lock(&wdt->lock);
> >
> > Does it really need to acquire the lock? It looks like the lock is to prevent
> > concurrent accesses to the mmap in max63xx_mmap_ping() and max63xx_mmap_set().
> >
>
> Actually, that doesn't work at all. spin_lock() directly contradicts
> with gpiod_set_value_cansleep().
>
> > > + gpiod_set_value_cansleep(wdt->gpio_wdi, 1);
> > > + udelay(1);
> >
> > Doesn't it need to include <linux/delay.h> for udelay()?
> >
> > > @@ -225,10 +240,19 @@ static int max63xx_wdt_probe(struct platform_device *pdev)
> > > return -EINVAL;
> > > }
> > > + wdt->gpio_wdi = devm_gpiod_get(dev, NULL, GPIOD_FLAGS_BIT_DIR_OUT);
> > > + if (IS_ERR(wdt->gpio_wdi) && PTR_ERR(wdt->gpio_wdi) != -ENOENT)
> >
> > Use devm_gpiod_get_optional() to make the intent clear. Also, it gets rid of
> > the check for -ENOENT.
> >
> > > + return dev_err_probe(dev, PTR_ERR(wdt->gpio_wdi),
> > > + "unable to request gpio: %ld\n",
> > > + PTR_ERR(wdt->gpio_wdi));
> >
> > It doesn't need to again print for PTR_ERR(wdt->gpio_wdi). dev_err_probe()
> > prints the error.
> >
> > > err = max63xx_mmap_init(pdev, wdt);
> > > if (err)
> > > return err;
> > > + if (!IS_ERR(wdt->gpio_wdi))
> > > + wdt->ping = max63xx_gpio_ping;
> >
> > Thus, the max63xx_gpio_ping() overrides max63xx_mmap_ping() if the GPIO was
> > provided? It would be better to mention the behavior in the commit message.
> >
> > Also, could both the assignments of `wdt->gpio_wdi` and `wdt->ping` happen
> > after max63xx_mmap_init()?
>
Hello! I'm going to look at all these issues. Recently I sent max63
watchdog driver also into U-Boot and seems that I mixed DTS and driver
code between U-Boot and Kernel... and tested something mixed.
I will do new testing again, and will check that I'm testing correct
code.
On Wednesday 04 May 2022 00:05:50 Pali Rohár wrote:
> On Monday 02 May 2022 21:37:16 Guenter Roeck wrote:
> > On 5/2/22 20:57, Tzung-Bi Shih wrote:
> > > On Fri, Apr 29, 2022 at 03:13:49PM +0200, Pali Rohár wrote:
> > > > @@ -27,6 +27,7 @@
> > > > #include <linux/io.h>
> > > > #include <linux/slab.h>
> > > > #include <linux/property.h>
> > > > +#include <linux/gpio/consumer.h>
> > >
> > > It would be better to keep them alphabetically. Anyway, they aren't sorted
> > > originally...
> > >
> > > > +static void max63xx_gpio_ping(struct max63xx_wdt *wdt)
> > > > +{
> > > > + spin_lock(&wdt->lock);
> > >
> > > Does it really need to acquire the lock? It looks like the lock is to prevent
> > > concurrent accesses to the mmap in max63xx_mmap_ping() and max63xx_mmap_set().
> > >
> >
> > Actually, that doesn't work at all. spin_lock() directly contradicts
> > with gpiod_set_value_cansleep().
> >
> > > > + gpiod_set_value_cansleep(wdt->gpio_wdi, 1);
> > > > + udelay(1);
> > >
> > > Doesn't it need to include <linux/delay.h> for udelay()?
> > >
> > > > @@ -225,10 +240,19 @@ static int max63xx_wdt_probe(struct platform_device *pdev)
> > > > return -EINVAL;
> > > > }
> > > > + wdt->gpio_wdi = devm_gpiod_get(dev, NULL, GPIOD_FLAGS_BIT_DIR_OUT);
> > > > + if (IS_ERR(wdt->gpio_wdi) && PTR_ERR(wdt->gpio_wdi) != -ENOENT)
> > >
> > > Use devm_gpiod_get_optional() to make the intent clear. Also, it gets rid of
> > > the check for -ENOENT.
> > >
> > > > + return dev_err_probe(dev, PTR_ERR(wdt->gpio_wdi),
> > > > + "unable to request gpio: %ld\n",
> > > > + PTR_ERR(wdt->gpio_wdi));
> > >
> > > It doesn't need to again print for PTR_ERR(wdt->gpio_wdi). dev_err_probe()
> > > prints the error.
> > >
> > > > err = max63xx_mmap_init(pdev, wdt);
> > > > if (err)
> > > > return err;
> > > > + if (!IS_ERR(wdt->gpio_wdi))
> > > > + wdt->ping = max63xx_gpio_ping;
> > >
> > > Thus, the max63xx_gpio_ping() overrides max63xx_mmap_ping() if the GPIO was
> > > provided? It would be better to mention the behavior in the commit message.
> > >
> > > Also, could both the assignments of `wdt->gpio_wdi` and `wdt->ping` happen
> > > after max63xx_mmap_init()?
> >
>
> Hello! I'm going to look at all these issues. Recently I sent max63
> watchdog driver also into U-Boot and seems that I mixed DTS and driver
> code between U-Boot and Kernel... and tested something mixed.
>
> I will do new testing again, and will check that I'm testing correct
> code.
Hello! Now I sent a new version V3. I have tested it on PowerPC P2020
based board where watchdog registers are exported via CPLD and new V3
version is working fine.