Re: [PATCH v3 6/7] watchdog: dw_wdt: Add pre-timeouts support

From: Serge Semin
Date: Sat May 30 2020 - 02:51:59 EST

Next message: Mauro Carvalho Chehab: "[PATCH v2 02/41] media: atomisp: get rid of the hrt/hive_isp_css_mm_hrt abstraction layer"
Previous message: Wesley Cheng: "Re: [RFC v3 1/3] usb: dwc3: Resize TX FIFOs to meet EP bursting requirements"
In reply to: Guenter Roeck: "Re: [PATCH v3 6/7] watchdog: dw_wdt: Add pre-timeouts support"
Next in thread: Serge Semin: "[PATCH v3 1/7] dt-bindings: watchdog: Convert DW WDT binding to DT schema"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

On Fri, May 29, 2020 at 04:02:19PM -0700, Guenter Roeck wrote:
> On Tue, May 26, 2020 at 06:41:22PM +0300, Serge Semin wrote:
> > DW Watchdog can rise an interrupt in case if IRQ request mode is enabled
> > and timer reaches the zero value. In this case the IRQ lane is left
> > pending until either the next watchdog kick event (watchdog restart) or
> > until the WDT_EOI register is read or the device/system reset. This
> > interface can be used to implement the pre-timeout functionality
> > optionally provided by the Linux kernel watchdog devices.
> >
> > IRQ mode provides a two stages timeout interface. It means the IRQ is
> > raised when the counter reaches zero, while the system reset occurs only
> > after subsequent timeout if the timer restart is not performed. Due to
> > this peculiarity the pre-timeout value is actually set to the achieved
> > hardware timeout, while the real watchdog timeout is considered to be
> > twice as much of it. This applies a significant limitation on the
> > pre-timeout values, so current implementation supports either zero value,
> > which disables the pre-timeout events, or non-zero values, which imply
> > the pre-timeout to be at least half of the current watchdog timeout.
> >
> > Note that we ask the interrupt controller to detect the rising-edge
> > pre-timeout interrupts to prevent the high-level-IRQs flood, since
> > if the pre-timeout happens, the IRQ lane will be left pending until
> > it's cleared by the timer restart.
> >
> > Signed-off-by: Serge Semin <Sergey.Semin@xxxxxxxxxxxxxxxxxxxx>
> > Cc: Alexey Malahov <Alexey.Malahov@xxxxxxxxxxxxxxxxxxxx>
> > Cc: Thomas Bogendoerfer <tsbogend@xxxxxxxxxxxxxxxx>
> > Cc: Arnd Bergmann <arnd@xxxxxxxx>
> > Cc: Rob Herring <robh+dt@xxxxxxxxxx>
> > Cc: linux-mips@xxxxxxxxxxxxxxx
> > Cc: devicetree@xxxxxxxxxxxxxxx
>

> Nitpick below, but I don't really know what to do about it, so
>
> Reviewed-by: Guenter Roeck <linux@xxxxxxxxxxxx>

Right. Semantically platform_get_irq_optional() should never return zero even
though the comment above that function definition states otherwise. I'll fix the
conditional statement to check for > 0 you've commented below and resend with
your tag attached. Thanks.

-Sergey

>
> > ---
> >
> > Changelog v2:
> > - Rearrange SoBs.
> > - Make the Pre-timeout IRQ being optionally supported.
> > ---
> > drivers/watchdog/dw_wdt.c | 138 +++++++++++++++++++++++++++++++++++---
> > 1 file changed, 130 insertions(+), 8 deletions(-)
> >
> > diff --git a/drivers/watchdog/dw_wdt.c b/drivers/watchdog/dw_wdt.c
> > index efbc36872670..3cd7c485cd70 100644
> > --- a/drivers/watchdog/dw_wdt.c
> > +++ b/drivers/watchdog/dw_wdt.c
> > @@ -22,6 +22,7 @@
> > #include <linux/kernel.h>
> > #include <linux/module.h>
> > #include <linux/moduleparam.h>
> > +#include <linux/interrupt.h>
> > #include <linux/of.h>
> > #include <linux/pm.h>
> > #include <linux/platform_device.h>
> > @@ -36,6 +37,8 @@
> > #define WDOG_CURRENT_COUNT_REG_OFFSET 0x08
> > #define WDOG_COUNTER_RESTART_REG_OFFSET 0x0c
> > #define WDOG_COUNTER_RESTART_KICK_VALUE 0x76
> > +#define WDOG_INTERRUPT_STATUS_REG_OFFSET 0x10
> > +#define WDOG_INTERRUPT_CLEAR_REG_OFFSET 0x14
> > #define WDOG_COMP_PARAMS_1_REG_OFFSET 0xf4
> > #define WDOG_COMP_PARAMS_1_USE_FIX_TOP BIT(6)
> >
> > @@ -59,6 +62,11 @@ module_param(nowayout, bool, 0);
> > MODULE_PARM_DESC(nowayout, "Watchdog cannot be stopped once started "
> > "(default=" __MODULE_STRING(WATCHDOG_NOWAYOUT) ")");
> >
> > +enum dw_wdt_rmod {
> > + DW_WDT_RMOD_RESET = 1,
> > + DW_WDT_RMOD_IRQ = 2
> > +};
> > +
> > struct dw_wdt_timeout {
> > u32 top_val;
> > unsigned int sec;
> > @@ -70,6 +78,7 @@ struct dw_wdt {
> > struct clk *clk;
> > struct clk *pclk;
> > unsigned long rate;
> > + enum dw_wdt_rmod rmod;
> > struct dw_wdt_timeout timeouts[DW_WDT_NUM_TOPS];
> > struct watchdog_device wdd;
> > struct reset_control *rst;
> > @@ -86,6 +95,20 @@ static inline int dw_wdt_is_enabled(struct dw_wdt *dw_wdt)
> > WDOG_CONTROL_REG_WDT_EN_MASK;
> > }
> >
> > +static void dw_wdt_update_mode(struct dw_wdt *dw_wdt, enum dw_wdt_rmod rmod)
> > +{
> > + u32 val;
> > +
> > + val = readl(dw_wdt->regs + WDOG_CONTROL_REG_OFFSET);
> > + if (rmod == DW_WDT_RMOD_IRQ)
> > + val |= WDOG_CONTROL_REG_RESP_MODE_MASK;
> > + else
> > + val &= ~WDOG_CONTROL_REG_RESP_MODE_MASK;
> > + writel(val, dw_wdt->regs + WDOG_CONTROL_REG_OFFSET);
> > +
> > + dw_wdt->rmod = rmod;
> > +}
> > +
> > static unsigned int dw_wdt_find_best_top(struct dw_wdt *dw_wdt,
> > unsigned int timeout, u32 *top_val)
> > {
> > @@ -145,7 +168,11 @@ static unsigned int dw_wdt_get_timeout(struct dw_wdt *dw_wdt)
> > break;
> > }
> >
> > - return dw_wdt->timeouts[idx].sec;
> > + /*
> > + * In IRQ mode due to the two stages counter, the actual timeout is
> > + * twice greater than the TOP setting.
> > + */
> > + return dw_wdt->timeouts[idx].sec * dw_wdt->rmod;
> > }
> >
> > static int dw_wdt_ping(struct watchdog_device *wdd)
> > @@ -164,7 +191,20 @@ static int dw_wdt_set_timeout(struct watchdog_device *wdd, unsigned int top_s)
> > unsigned int timeout;
> > u32 top_val;
> >
> > - timeout = dw_wdt_find_best_top(dw_wdt, top_s, &top_val);
> > + /*
> > + * Note IRQ mode being enabled means having a non-zero pre-timeout
> > + * setup. In this case we try to find a TOP as close to the half of the
> > + * requested timeout as possible since DW Watchdog IRQ mode is designed
> > + * in two stages way - first timeout rises the pre-timeout interrupt,
> > + * second timeout performs the system reset. So basically the effective
> > + * watchdog-caused reset happens after two watchdog TOPs elapsed.
> > + */
> > + timeout = dw_wdt_find_best_top(dw_wdt, DIV_ROUND_UP(top_s, dw_wdt->rmod),
> > + &top_val);
> > + if (dw_wdt->rmod == DW_WDT_RMOD_IRQ)
> > + wdd->pretimeout = timeout;
> > + else
> > + wdd->pretimeout = 0;
> >
> > /*
> > * Set the new value in the watchdog. Some versions of dw_wdt
> > @@ -175,25 +215,47 @@ static int dw_wdt_set_timeout(struct watchdog_device *wdd, unsigned int top_s)
> > writel(top_val | top_val << WDOG_TIMEOUT_RANGE_TOPINIT_SHIFT,
> > dw_wdt->regs + WDOG_TIMEOUT_RANGE_REG_OFFSET);
> >
> > + /* Kick new TOP value into the watchdog counter if activated. */
> > + if (watchdog_active(wdd))
> > + dw_wdt_ping(wdd);
> > +
> > /*
> > * In case users set bigger timeout value than HW can support,
> > * kernel(watchdog_dev.c) helps to feed watchdog before
> > * wdd->max_hw_heartbeat_ms
> > */
> > if (top_s * 1000 <= wdd->max_hw_heartbeat_ms)
> > - wdd->timeout = timeout;
> > + wdd->timeout = timeout * dw_wdt->rmod;
> > else
> > wdd->timeout = top_s;
> >
> > return 0;
> > }
> >
> > +static int dw_wdt_set_pretimeout(struct watchdog_device *wdd, unsigned int req)
> > +{
> > + struct dw_wdt *dw_wdt = to_dw_wdt(wdd);
> > +
> > + /*
> > + * We ignore actual value of the timeout passed from user-space
> > + * using it as a flag whether the pretimeout functionality is intended
> > + * to be activated.
> > + */
> > + dw_wdt_update_mode(dw_wdt, req ? DW_WDT_RMOD_IRQ : DW_WDT_RMOD_RESET);
> > + dw_wdt_set_timeout(wdd, wdd->timeout);
> > +
> > + return 0;
> > +}
> > +
> > static void dw_wdt_arm_system_reset(struct dw_wdt *dw_wdt)
> > {
> > u32 val = readl(dw_wdt->regs + WDOG_CONTROL_REG_OFFSET);
> >
> > - /* Disable interrupt mode; always perform system reset. */
> > - val &= ~WDOG_CONTROL_REG_RESP_MODE_MASK;
> > + /* Disable/enable interrupt mode depending on the RMOD flag. */
> > + if (dw_wdt->rmod == DW_WDT_RMOD_IRQ)
> > + val |= WDOG_CONTROL_REG_RESP_MODE_MASK;
> > + else
> > + val &= ~WDOG_CONTROL_REG_RESP_MODE_MASK;
> > /* Enable watchdog. */
> > val |= WDOG_CONTROL_REG_WDT_EN_MASK;
> > writel(val, dw_wdt->regs + WDOG_CONTROL_REG_OFFSET);
> > @@ -231,6 +293,7 @@ static int dw_wdt_restart(struct watchdog_device *wdd,
> > struct dw_wdt *dw_wdt = to_dw_wdt(wdd);
> >
> > writel(0, dw_wdt->regs + WDOG_TIMEOUT_RANGE_REG_OFFSET);
> > + dw_wdt_update_mode(dw_wdt, DW_WDT_RMOD_RESET);
> > if (dw_wdt_is_enabled(dw_wdt))
> > writel(WDOG_COUNTER_RESTART_KICK_VALUE,
> > dw_wdt->regs + WDOG_COUNTER_RESTART_REG_OFFSET);
> > @@ -246,9 +309,19 @@ static int dw_wdt_restart(struct watchdog_device *wdd,
> > static unsigned int dw_wdt_get_timeleft(struct watchdog_device *wdd)
> > {
> > struct dw_wdt *dw_wdt = to_dw_wdt(wdd);
> > + unsigned int sec;
> > + u32 val;
> > +
> > + val = readl(dw_wdt->regs + WDOG_CURRENT_COUNT_REG_OFFSET);
> > + sec = val / dw_wdt->rate;
> >
> > - return readl(dw_wdt->regs + WDOG_CURRENT_COUNT_REG_OFFSET) /
> > - dw_wdt->rate;
> > + if (dw_wdt->rmod == DW_WDT_RMOD_IRQ) {
> > + val = readl(dw_wdt->regs + WDOG_INTERRUPT_STATUS_REG_OFFSET);
> > + if (!val)
> > + sec += wdd->pretimeout;
> > + }
> > +
> > + return sec;
> > }
> >
> > static const struct watchdog_info dw_wdt_ident = {
> > @@ -257,16 +330,41 @@ static const struct watchdog_info dw_wdt_ident = {
> > .identity = "Synopsys DesignWare Watchdog",
> > };
> >
> > +static const struct watchdog_info dw_wdt_pt_ident = {
> > + .options = WDIOF_KEEPALIVEPING | WDIOF_SETTIMEOUT |
> > + WDIOF_PRETIMEOUT | WDIOF_MAGICCLOSE,
> > + .identity = "Synopsys DesignWare Watchdog",
> > +};
> > +
> > static const struct watchdog_ops dw_wdt_ops = {
> > .owner = THIS_MODULE,
> > .start = dw_wdt_start,
> > .stop = dw_wdt_stop,
> > .ping = dw_wdt_ping,
> > .set_timeout = dw_wdt_set_timeout,
> > + .set_pretimeout = dw_wdt_set_pretimeout,
> > .get_timeleft = dw_wdt_get_timeleft,
> > .restart = dw_wdt_restart,
> > };
> >
> > +static irqreturn_t dw_wdt_irq(int irq, void *devid)
> > +{
> > + struct dw_wdt *dw_wdt = devid;
> > + u32 val;
> > +
> > + /*
> > + * We don't clear the IRQ status. It's supposed to be done by the
> > + * following ping operations.
> > + */
> > + val = readl(dw_wdt->regs + WDOG_INTERRUPT_STATUS_REG_OFFSET);
> > + if (!val)
> > + return IRQ_NONE;
> > +
> > + watchdog_notify_pretimeout(&dw_wdt->wdd);
> > +
> > + return IRQ_HANDLED;
> > +}
> > +
> > #ifdef CONFIG_PM_SLEEP
> > static int dw_wdt_suspend(struct device *dev)
> > {
> > @@ -447,6 +545,31 @@ static int dw_wdt_drv_probe(struct platform_device *pdev)
> > goto out_disable_pclk;
> > }
> >
> > + /* Enable normal reset without pre-timeout by default. */
> > + dw_wdt_update_mode(dw_wdt, DW_WDT_RMOD_RESET);
> > +
> > + /*
> > + * Pre-timeout IRQ is optional, since some hardware may lack support
> > + * of it. Note we must request rising-edge IRQ, since the lane is left
> > + * pending either until the next watchdog kick event or up to the
> > + * system reset.
> > + */
> > + ret = platform_get_irq_optional(pdev, 0);
> > + if (ret >= 0) {
>
> I keep seeing notes that an interrupt value of 0 is invalid.
>
> > + ret = devm_request_irq(dev, ret, dw_wdt_irq,
> > + IRQF_SHARED | IRQF_TRIGGER_RISING,
> > + pdev->name, dw_wdt);
> > + if (ret)
> > + goto out_disable_pclk;
> > +
> > + dw_wdt->wdd.info = &dw_wdt_pt_ident;
> > + } else {
> > + if (ret == -EPROBE_DEFER)
> > + goto out_disable_pclk;
> > +
> > + dw_wdt->wdd.info = &dw_wdt_ident;
> > + }
> > +
> > reset_control_deassert(dw_wdt->rst);
> >
> > ret = dw_wdt_init_timeouts(dw_wdt, dev);
> > @@ -454,7 +577,6 @@ static int dw_wdt_drv_probe(struct platform_device *pdev)
> > goto out_disable_clk;
> >
> > wdd = &dw_wdt->wdd;
> > - wdd->info = &dw_wdt_ident;
> > wdd->ops = &dw_wdt_ops;
> > wdd->min_timeout = dw_wdt_get_min_timeout(dw_wdt);
> > wdd->max_hw_heartbeat_ms = dw_wdt_get_max_timeout_ms(dw_wdt);

Next message: Mauro Carvalho Chehab: "[PATCH v2 02/41] media: atomisp: get rid of the hrt/hive_isp_css_mm_hrt abstraction layer"
Previous message: Wesley Cheng: "Re: [RFC v3 1/3] usb: dwc3: Resize TX FIFOs to meet EP bursting requirements"
In reply to: Guenter Roeck: "Re: [PATCH v3 6/7] watchdog: dw_wdt: Add pre-timeouts support"
Next in thread: Serge Semin: "[PATCH v3 1/7] dt-bindings: watchdog: Convert DW WDT binding to DT schema"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]