The panic is only there for debug purposes at first expiry. Watchdog timeout behavior (second expiry) is not changed. I added the panic as a config option, enabled by default. François-Nicolas
>From 5e1680aa0df9f49459cdc7211ba80d6934f65fbd Mon Sep 17 00:00:00 2001 From: Francois-Nicolas Muller <[email protected]> Date: Tue, 20 Jan 2015 14:55:42 +0100 Subject: [PATCH] Adding TCO watchdog warning interrupt handling. This feature is useful to root cause watchdog expiration. It is activated by boot parameter 'warn_irq' (disabled by default). Upon first expiration of the TCO watchdog, a warning interrupt is fired, then the interrupt handler dumps registers and call stack of all available cpus. TCO watchdog reloads with 2.4 seconds timeout for second expiration. If CONFIG_ITCO_WARN_PANIC is set, the warning interrupt also calls panic() which notifies the panic handlers then reboots the platform, depending on CONFIG_PANIC_TIMEOUT value : - If CONFIG_PANIC_TIMEOUT is zero or greater than 3 seconds, TCO watchdog will reset the platform if second expiration happens before TCO has been kicked again. - If CONFIG_PANIC_TIMEOUT is < 0, platform will reboot immediately (emergency restart procedure). - If CONFIG_PANIC_TIMEOUT is 1 or 2 seconds, platform will reboot after 1 or 2 seconds delay (emergency restart procedure). Change-Id: I48dcb9d38218c8218e35f9969f064b9d5cf316f1 Signed-off-by: Francois-Nicolas Muller <[email protected]> --- drivers/watchdog/Kconfig | 13 +++++++++ drivers/watchdog/iTCO_wdt.c | 66 +++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 79 insertions(+) diff --git a/drivers/watchdog/Kconfig b/drivers/watchdog/Kconfig index 79d2589..15c3807 100644 --- a/drivers/watchdog/Kconfig +++ b/drivers/watchdog/Kconfig @@ -674,6 +674,19 @@ config ITCO_VENDOR_SUPPORT devices. At this moment we only have additional support for some SuperMicro Inc. motherboards. +config ITCO_WARNING_PANIC + bool "Intel TCO Timer/Watchdog panic on warning interrupt" + depends on ITCO_WDT + default y + ---help--- + Force a call to panic() when TCO warning interrupt occurs. + + Warning interrupt happens if warn_irq module parameter is set and + TCO timer first expires. + + If not set, only cpu backtraces are dumped, no call to panic() and + no notification of panic are done. + config IT8712F_WDT tristate "IT8712F (Smart Guardian) Watchdog Timer" depends on X86 diff --git a/drivers/watchdog/iTCO_wdt.c b/drivers/watchdog/iTCO_wdt.c index e802a54..e7c5169 100644 --- a/drivers/watchdog/iTCO_wdt.c +++ b/drivers/watchdog/iTCO_wdt.c @@ -49,6 +49,8 @@ /* Module and version information */ #define DRV_NAME "iTCO_wdt" #define DRV_VERSION "1.11" +#define DRV_NAME_ACPI "iTCO_wdt_wirq" +#define TCO_CLASS DRV_NAME /* Includes */ #include <linux/module.h> /* For module specific items */ @@ -68,6 +70,9 @@ #include <linux/pm.h> /* For suspend/resume */ #include <linux/mfd/core.h> #include <linux/mfd/lpc_ich.h> +#include <linux/nmi.h> +#include <linux/acpi.h> +#include <acpi/actypes.h> #include "iTCO_vendor.h" @@ -107,6 +112,12 @@ static struct { /* this is private data for the iTCO_wdt device */ bool started; } iTCO_wdt_private; +static const struct acpi_device_id iTCO_wdt_ids[] = { + {"8086229C", 0}, + {"", 0}, +}; +MODULE_DEVICE_TABLE(acpi, iTCO_wdt_ids); + /* module parameters */ #define WATCHDOG_TIMEOUT 30 /* 30 sec default heartbeat */ static int heartbeat = WATCHDOG_TIMEOUT; /* in seconds */ @@ -126,6 +137,13 @@ module_param(turn_SMI_watchdog_clear_off, int, 0); MODULE_PARM_DESC(turn_SMI_watchdog_clear_off, "Turn off SMI clearing watchdog (depends on TCO-version)(default=1)"); +static bool warn_irq; +module_param(warn_irq, bool, 0); +MODULE_PARM_DESC(warn_irq, + "Dump all cpus backtraces at first watchdog timer expiration (default=0)"); + +static bool warn_irq_panic = CONFIG_ITCO_WARNING_PANIC; + /* * Some TCO specific functions */ @@ -200,6 +218,35 @@ static int iTCO_wdt_unset_NO_REBOOT_bit(void) return ret; /* returns: 0 = OK, -EIO = Error */ } +static u32 iTCO_wdt_wirq(acpi_handle gpe_device, u32 gpe, void *context) +{ + trigger_all_cpu_backtrace(); + if (warn_irq_panic) + panic("Kernel Watchdog"); + + return IRQ_HANDLED; +} + +static int iTCO_wdt_acpi_add(struct acpi_device *device) +{ + unsigned long long gpe; + acpi_status status; + + status = acpi_evaluate_integer(device->handle, "_GPE", NULL, &gpe); + if (ACPI_FAILURE(status)) + return -EINVAL; + + status = acpi_install_gpe_handler(NULL, gpe, ACPI_GPE_EDGE_TRIGGERED, + iTCO_wdt_wirq, NULL); + if (ACPI_FAILURE(status)) + return -ENODEV; + + acpi_enable_gpe(NULL, gpe); + + pr_debug("interrupt=SCI GPE=0x%02llx", gpe); + return 0; +} + static int iTCO_wdt_start(struct watchdog_device *wd_dev) { unsigned int val; @@ -628,6 +675,15 @@ static struct platform_driver iTCO_wdt_driver = { }, }; +static struct acpi_driver iTCO_wdt_acpi_driver = { + .name = DRV_NAME_ACPI, + .class = TCO_CLASS, + .ids = iTCO_wdt_ids, + .ops = { + .add = iTCO_wdt_acpi_add, + }, +}; + static int __init iTCO_wdt_init_module(void) { int err; @@ -638,12 +694,22 @@ static int __init iTCO_wdt_init_module(void) if (err) return err; + if (warn_irq) { + err = acpi_bus_register_driver(&iTCO_wdt_acpi_driver); + if (err) { + platform_driver_unregister(&iTCO_wdt_driver); + return err; + } + } + return 0; } static void __exit iTCO_wdt_cleanup_module(void) { platform_driver_unregister(&iTCO_wdt_driver); + if (warn_irq) + acpi_bus_unregister_driver(&iTCO_wdt_acpi_driver); pr_info("Watchdog Module Unloaded\n"); } -- 1.7.9.5 -----Original Message----- From: Guenter Roeck [mailto:[email protected]] Sent: Thursday, January 15, 2015 3:49 PM To: Muller, Francois-nicolas Cc: Darren Hart; '[email protected]'; Rafael Wysocki; Linux ACPI Mailing List; [email protected]; Wim Van Sebroeck Subject: Re: [PATCH] TCO Watchdog warning interrupt driver creation On 01/15/2015 05:27 AM, Muller, Francois-nicolas wrote: > TCO driver is anyway auto-loaded by mfd driver lpc_ich as a sub-function of > it. > > The aim of my patch is only to add warning interrupt support in TCO driver. > For this it need the GPE number which is exposed by Bios in acpi tables. > So the patch registers also the TCO driver as an acpi driver to be able to > retrieve this value. > > The acpi code part is only required by the interrupt handling support, not > needed for the loading of the driver itself (already done by lpc_ich). > > In this context I don't see the point of dissociating the driver part from > the loading, unless the loading would need to be reverted later. > Ok, makes sense. Regarding the patch itself, I'll leave it up to Wim to decide what to do. Personally I dislike the notion of panicing as response to a watchdog timeout. Guenter -- To unsubscribe from this list: send the line "unsubscribe linux-watchdog" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html
