mei: create proper iAMT watchdog driver
Subsystems (4)
v2 cover mei: create proper iAMT watchdog driver
Instead of integrating the iAMT watchdog within the mei core driver we will create a watchdog device on the mei client bus and create a proper watchdog driver for it.
V2:
1. The watchdog device is no longer dynamically allocated in separate structure
2. Export device internal status via debugfs
3. Address few comments from Guenter
4. Reworked de/registration
The patches should apply and compile on both watchdog-next and char-misc-next trees/branches.
I would prefer this will go via char-misc-next as this is the tree we work against and conflicts should be trivial if at any.
Alexander Usyskin (3):
mei: wd: drop the watchdog code from the core mei driver
watchdog: mei_wdt: register wd device only if required
watchdog: mei_wdt: re-register device on event
Tomas Winkler (4):
mei: drop nfc leftovers from the mei driver
watchdog: mei_wdt: implement MEI iAMT watchdog driver
watchdog: mei_wdt: add status debugfs entry
mei: bus: whitelist the watchdog client
Documentation/misc-devices/mei/mei.txt | 12 +- MAINTAINERS | 1 + drivers/misc/mei/Kconfig | 6 +- drivers/misc/mei/Makefile | 1 - drivers/misc/mei/bus-fixup.c | 29 ++ drivers/misc/mei/client.c | 12 +- drivers/misc/mei/client.h | 4 - drivers/misc/mei/init.c | 10 +- drivers/misc/mei/interrupt.c | 15 - drivers/misc/mei/mei_dev.h | 72 +--- drivers/misc/mei/wd.c | 391 ------------------- drivers/watchdog/Kconfig | 15 + drivers/watchdog/Makefile | 1 + drivers/watchdog/mei_wdt.c | 692 +++++++++++++++++++++++++++++++++ 14 files changed, 753 insertions(+), 508 deletions(-) delete mode 100644 drivers/misc/mei/wd.c create mode 100644 drivers/watchdog/mei_wdt.c
Signature · 1 line
2.4.3
PATCH v2 1/7 mei: drop nfc leftovers from the mei driver
We left few function prototypes in the header file after moving nfc logic to bus.
---
V2: resend
drivers/misc/mei/mei_dev.h | 11 ----------- 1 file changed, 11 deletions(-)
drivers/misc/mei/mei_dev.h +0 -11
@@ -649,17 +649,6 @@ int mei_amthif_irq_read_msg(struct mei_cl *cl, struct mei_cl_cb *complete_list);int mei_amthif_irq_read (struct mei_device *dev,s32 *slots); -/* -* NFC functions -*/ -int mei_nfc_host_init (struct mei_device *dev,struct mei_me_client *me_cl); -void mei_nfc_host_exit (struct mei_device *dev); - -/* -* NFC Client UUID -*/ -extern const uuid_le mei_nfc_guid; -int mei_wd_send (struct mei_device *dev);int mei_wd_stop (struct mei_device *dev);int mei_wd_host_init (struct mei_device *dev,struct mei_me_client *me_cl);
Signature · 1 line
2.4.3
PATCH v2 2/7 mei: wd: drop the watchdog code from the core mei driver
In reply to Tomas WinklerFrom: Alexander Usyskin <[email protected]>
Instead of integrating the iAMT watchdog in the mei core driver we will create a watchdog device on the mei client bus and create a driver for it.
This patch removes the watchdog code from the mei core driver.
---
V2: resend
drivers/misc/mei/Kconfig | 6 +- drivers/misc/mei/Makefile | 1 - drivers/misc/mei/client.c | 12 +- drivers/misc/mei/client.h | 4 - drivers/misc/mei/init.c | 10 +- drivers/misc/mei/interrupt.c | 15 -- drivers/misc/mei/mei_dev.h | 61 +------ drivers/misc/mei/wd.c | 391 ------------------------------------------- 8 files changed, 9 insertions(+), 491 deletions(-) delete mode 100644 drivers/misc/mei/wd.c
drivers/misc/mei/Kconfig +3 -3
@@ -1,6 +1,6 @@ config INTEL_MEI tristate "Intel Management Engine Interface" - depends on X86 && PCI && WATCHDOG_CORE + depends on X86 && PCI help The Intel Management Engine (Intel ME) provides Manageability, Security and Media services for system containing Intel chipsets.
@@ -12,7 +12,7 @@ config INTEL_MEI config INTEL_MEI_ME tristate "ME Enabled Intel Chipsets" select INTEL_MEI - depends on X86 && PCI && WATCHDOG_CORE + depends on X86 && PCI help MEI support for ME Enabled Intel chipsets.
@@ -37,7 +37,7 @@ config INTEL_MEI_ME config INTEL_MEI_TXE tristate "Intel Trusted Execution Environment with ME Interface" select INTEL_MEI - depends on X86 && PCI && WATCHDOG_CORE + depends on X86 && PCI help MEI Support for Trusted Execution Environment device on Intel SoCs
drivers/misc/mei/Makefile +0 -1
@@ -9,7 +9,6 @@ mei-objs += interrupt.o mei-objs += client.o mei-objs += main.o mei-objs += amthif.o -mei-objs += wd.o mei-objs += bus.o mei-objs += bus-fixup.o mei-$(CONFIG_DEBUG_FS) += debugfs.o
drivers/misc/mei/client.c +3 -9
@@ -648,7 +648,7 @@ int mei_cl_unlink(struct mei_cl *cl)if (!cl)return 0 ; -/* wd and amthif might not be initialized */ +/* amthif might not be initialized */ if (!cl->dev)return 0 ;
@@ -679,17 +679,11 @@ void mei_host_client_init(struct work_struct *work)mutex_lock (&dev->device_lock); - me_cl =mei_me_cl_by_uuid (dev, &mei_amthif_guid);if (me_cl)mei_amthif_host_init (dev, me_cl);mei_me_cl_put (me_cl); - me_cl =mei_me_cl_by_uuid (dev, &mei_wd_guid); -if (me_cl) -mei_wd_host_init (dev, me_cl); -mei_me_cl_put (me_cl); - dev->dev_state =MEI_DEV_ENABLED ; dev->reset_count =0 ;mutex_unlock (&dev->device_lock);
@@ -1153,7 +1147,7 @@ err:* * Return: 1 if mei_flow_ctrl_creds >0, 0 - otherwise. */ -int mei_cl_flow_ctrl_creds (struct mei_cl *cl) +static int mei_cl_flow_ctrl_creds (struct mei_cl *cl) {int rets;
@@ -1186,7 +1180,7 @@ int mei_cl_flow_ctrl_creds(struct mei_cl *cl)* 0 on success * -EINVAL when ctrl credits are <= 0 */ -int mei_cl_flow_ctrl_reduce (struct mei_cl *cl) +static int mei_cl_flow_ctrl_reduce (struct mei_cl *cl) {if (WARN_ON (!cl || !cl->me_cl))return -EINVAL ;
drivers/misc/mei/client.h +0 -4
@@ -18,7 +18,6 @@#define _MEI_CLIENT_H_#include <linux/types.h> -#include <linux/watchdog.h> #include <linux/poll.h> #include <linux/mei.h>
@@ -120,9 +119,6 @@ struct mei_cl_cb *mei_cl_alloc_cb(struct mei_cl *cl, size_t length, enum mei_cb_file_ops type, struct file *fp);int mei_cl_flush_queues (struct mei_cl *cl,const struct file *fp); -int mei_cl_flow_ctrl_creds (struct mei_cl *cl); - -int mei_cl_flow_ctrl_reduce (struct mei_cl *cl);/* * MEI input output function prototype */
drivers/misc/mei/init.c +1 -9
@@ -156,8 +156,7 @@ int mei_reset(struct mei_device *dev)mei_cl_all_wakeup (dev);/* remove entry if already in list */ -dev_dbg (dev->dev,"remove iamthif and wd from the file list.\n" ); -mei_cl_unlink (&dev->wd_cl); +dev_dbg (dev->dev,"remove iamthif from the file list.\n" );mei_cl_unlink (&dev->iamthif_cl);mei_amthif_reset_params (dev); }
@@ -165,7 +164,6 @@ int mei_reset(struct mei_device *dev)mei_hbm_reset (dev); dev->rd_msg_hdr =0 ; - dev->wd_pending = false;if (ret) {dev_err (dev->dev,"hw_reset failed ret = %d\n" , ret);
@@ -335,16 +333,12 @@ void mei_stop(struct mei_device *dev)mutex_lock (&dev->device_lock); -mei_wd_stop (dev); - dev->dev_state =MEI_DEV_POWER_DOWN ;mei_reset (dev);/* move device to disabled state unconditionally */ dev->dev_state =MEI_DEV_DISABLED ;mutex_unlock (&dev->device_lock); - -mei_watchdog_unregister (dev); }EXPORT_SYMBOL_GPL (mei_stop);
@@ -394,7 +388,6 @@ void mei_device_init(struct mei_device *dev,init_waitqueue_head (&dev->wait_hw_ready);init_waitqueue_head (&dev->wait_pg);init_waitqueue_head (&dev->wait_hbm_start); -init_waitqueue_head (&dev->wait_stop_wd); dev->dev_state =MEI_DEV_INITIALIZING ; dev->reset_count =0 ;
@@ -407,7 +400,6 @@ void mei_device_init(struct mei_device *dev,INIT_WORK (&dev->init_work, mei_host_client_init);INIT_WORK (&dev->reset_work, mei_reset_work); -INIT_LIST_HEAD (&dev->wd_cl.link);INIT_LIST_HEAD (&dev->iamthif_cl.link);mei_io_list_init (&dev->amthif_cmd_list);mei_io_list_init (&dev->amthif_rd_complete_list);
drivers/misc/mei/interrupt.c +0 -15
@@ -360,21 +360,6 @@ int mei_irq_write_handler(struct mei_device *dev, struct mei_cl_cb *cmpl_list)list_move_tail (&cb->list, &cmpl_list->list); } -if (dev->wd_state ==MEI_WD_STOPPING ) { - dev->wd_state =MEI_WD_IDLE ; -wake_up (&dev->wait_stop_wd); - } - -if (mei_cl_is_connected (&dev->wd_cl)) { -if (dev->wd_pending && -mei_cl_flow_ctrl_creds (&dev->wd_cl) >0 ) { - ret =mei_wd_send (dev); -if (ret) -return ret; - dev->wd_pending = false; - } - } -/* complete control write list CB */ dev_dbg (dev->dev,"complete control write list cb.\n" );list_for_each_entry_safe (cb, next, &dev->ctrl_wr_list.list, list) {
drivers/misc/mei/mei_dev.h +2 -59
121 lines not on this page · raw message
drivers/misc/mei/wd.c deleted +0 -391
391 lines not on this page · raw message
Signature · 1 line
2.4.3
PATCH v2 3/7 watchdog: mei_wdt: implement MEI iAMT watchdog driver
In reply to Tomas WinklerCreate a driver with the generic watchdog interface for the MEI iAMT watchdog device.
---
V2: The watchdog device is no longer dynamically allocated in separate structure
Documentation/misc-devices/mei/mei.txt | 12 +- MAINTAINERS | 1 + drivers/watchdog/Kconfig | 15 ++ drivers/watchdog/Makefile | 1 + drivers/watchdog/mei_wdt.c | 415 +++++++++++++++++++++++++++++++++ 5 files changed, 438 insertions(+), 6 deletions(-) create mode 100644 drivers/watchdog/mei_wdt.c
Documentation/misc-devices/mei/mei.txt +6 -6
@@ -231,15 +231,15 @@ IT knows when a platform crashes even when there is a hard failure on the host. The Intel AMT Watchdog is composed of two parts: 1) Firmware feature - receives the heartbeats and sends an event when the heartbeats stop. - 2) Intel MEI driver - connects to the watchdog feature, configures the - watchdog and sends the heartbeats. + 2) Intel MEI iAMT watchdog driver - connects to the watchdog feature, + configures the watchdog and sends the heartbeats. -The Intel MEI driver uses the kernel watchdog API to configure the Intel AMT -Watchdog and to send heartbeats to it. The default timeout of the +The Intel iAMT watchdog MEI driver uses the kernel watchdog API to configure +the Intel AMT Watchdog and to send heartbeats to it. The default timeout of the watchdog is 120 seconds. -If the Intel AMT Watchdog feature does not exist (i.e. the connection failed), -the Intel MEI driver will disable the sending of heartbeats. +If the Intel AMT is not enabled in the firmware then the watchdog client won't enumerate +on the me client bus and watchdog devices won't be exposed. Supported Chipsets
MAINTAINERS +1 -0
@@ -5666,6 +5666,7 @@ S: Supported F: include/uapi/linux/mei.h F: include/linux/mei_cl_bus.h F: drivers/misc/mei/* +F: drivers/watchdog/mei_wdt.c F: Documentation/misc-devices/mei/* INTEL MIC DRIVERS (mic)
drivers/watchdog/Kconfig +15 -0
@@ -1154,6 +1154,21 @@ config SBC_EPX_C3_WATCHDOG To compile this driver as a module, choose M here: the module will be called sbc_epx_c3. +config INTEL_MEI_WDT + tristate "Intel MEI iAMT Watchdog" + depends on INTEL_MEI && X86 + select WATCHDOG_CORE + ---help--- + A device driver for the Intel MEI iAMT watchdog. + + The Intel AMT Watchdog is an OS Health (Hang/Crash) watchdog. + Whenever the OS hangs or crashes, iAMT will send an event + to any subscriber to this event. The watchdog doesn't reset the + the platform. + + To compile this driver as a module, choose M here: + the module will be called mei_wdt. + # M32R Architecture # M68K Architecture
drivers/watchdog/Makefile +1 -0
@@ -123,6 +123,7 @@ obj-$(CONFIG_MACHZ_WDT) += machzwd.o obj-$(CONFIG_SBC_EPX_C3_WATCHDOG) += sbc_epx_c3.o obj-$(CONFIG_INTEL_SCU_WATCHDOG) += intel_scu_watchdog.o obj-$(CONFIG_INTEL_MID_WATCHDOG) += intel-mid_wdt.o +obj-$(CONFIG_INTEL_MEI_WDT) += mei_wdt.o # M32R Architecture
drivers/watchdog/mei_wdt.c new file +415 -0
415 lines not on this page · raw message
Signature · 1 line
2.4.3
Re: [char-misc-next v2 3/7] watchdog: mei_wdt: implement MEI iAMT watchdog driver
Quoted 5 lines · Tomas Winkler
Create a driver with the generic watchdog interface for the MEI iAMT watchdog device.
Signed-off-by Alexander Usyskin <[email protected]>Signed-off-by Tomas Winkler <[email protected]>
Quoted 509 lines
---
V2: The watchdog device is no longer dynamically allocated in separate structureDocumentation/misc-devices/mei/mei.txt | 12 +- MAINTAINERS | 1 + drivers/watchdog/Kconfig | 15 ++ drivers/watchdog/Makefile | 1 + drivers/watchdog/mei_wdt.c | 415 +++++++++++++++++++++++++++++++++ 5 files changed, 438 insertions(+), 6 deletions(-)create mode 100644
drivers/watchdog/mei_wdt.cDocumentation/misc-devices/mei/mei.txt +6
-6@@ -231,15 +231,15 @@ IT knows when a platform crashes even when there is a hard failure on the host. The Intel AMT Watchdog is composed of two parts: 1) Firmware feature - receives the heartbeats and sends an event when the heartbeats stop. - 2) Intel MEI driver - connects to the watchdog feature, configures the - watchdog and sends the heartbeats. + 2) Intel MEI iAMT watchdog driver - connects to the watchdog feature, + configures the watchdog and sends the heartbeats. -The Intel MEI driver uses the kernel watchdog API to configure the Intel AMT -Watchdog and to send heartbeats to it. The default timeout of the +The Intel iAMT watchdog MEI driver uses the kernel watchdog API to configure +the Intel AMT Watchdog and to send heartbeats to it. The default timeout of the watchdog is 120 seconds. -If the Intel AMT Watchdog feature does not exist (i.e. the connection failed), -the Intel MEI driver will disable the sending of heartbeats. +If the Intel AMT is not enabled in the firmware then the watchdog client won't enumerate +on the me client bus and watchdog devices won't be exposed. Supported ChipsetsMAINTAINERS +1
-0@@ -5666,6 +5666,7 @@ S: Supported F: include/uapi/linux/mei.h F: include/linux/mei_cl_bus.h F: drivers/misc/mei/* +F: drivers/watchdog/mei_wdt.c F: Documentation/misc-devices/mei/* INTEL MIC DRIVERS (mic)drivers/watchdog/Kconfig +15
-0@@ -1154,6 +1154,21 @@ config SBC_EPX_C3_WATCHDOG To compile this driver as a module, choose M here: the module will be called sbc_epx_c3. +config INTEL_MEI_WDT + tristate "Intel MEI iAMT Watchdog" + depends on INTEL_MEI && X86 + select WATCHDOG_CORE + ---help--- + A device driver for the Intel MEI iAMT watchdog. + + The Intel AMT Watchdog is an OS Health (Hang/Crash) watchdog. + Whenever the OS hangs or crashes, iAMT will send an event + to any subscriber to this event. The watchdog doesn't reset the + the platform. + + To compile this driver as a module, choose M here: + the module will be called mei_wdt. + # M32R Architecture # M68K Architecturedrivers/watchdog/Makefile +1
-0@@ -123,6 +123,7 @@ obj-$(CONFIG_MACHZ_WDT) += machzwd.o obj-$(CONFIG_SBC_EPX_C3_WATCHDOG) += sbc_epx_c3.o obj-$(CONFIG_INTEL_SCU_WATCHDOG) += intel_scu_watchdog.o obj-$(CONFIG_INTEL_MID_WATCHDOG) += intel-mid_wdt.o +obj-$(CONFIG_INTEL_MEI_WDT) += mei_wdt.o # M32R Architecturedrivers/watchdog/mei_wdt.c new file +415
-0@@ -0,0 +1,415 @@ +/* +* Intel Management Engine Interface (Intel MEI) Linux driver +* Copyright (c) 2015, Intel Corporation. +* +* This program is free software; you can redistribute it and/or modify it +* under the terms and conditions of the GNU General Public License, +* version 2, as published by the Free Software Foundation. +* +* This program is distributed in the hope it will be useful, but WITHOUT +* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or +* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for +* more details. +*/ + +#include <linux/module.h> +#include <linux/slab.h> +#include <linux/interrupt.h> +#include <linux/watchdog.h> + +#include <linux/uuid.h> +#include <linux/mei_cl_bus.h> + +/* +* iAMT Watchdog Device +*/ +#define INTEL_AMT_WATCHDOG_ID "iamt_wdt" + +#define MEI_WDT_DEFAULT_TIMEOUT 120/* seconds */ +#define MEI_WDT_MIN_TIMEOUT 120/* seconds */ +#define MEI_WDT_MAX_TIMEOUT 65535/* seconds */ + +/* Commands */ +#define MEI_MANAGEMENT_CONTROL 0x02 + +/* MEI Management Control version number */ +#define MEI_MC_VERSION_NUMBER 0x10 + +/* Sub Commands */ +#define MEI_MC_START_WD_TIMER_REQ 0x13 +#define MEI_MC_STOP_WD_TIMER_REQ 0x14 + +/** +* enum mei_wdt_state - internal watchdog state +* +* @MEI_WDT_IDLE: wd is idle and not opened +* @MEI_WDT_START: wd was opened, start was called +* @MEI_WDT_RUNNING: wd is expecting keep alive pings +* @MEI_WDT_STOPPING: wd is stopping and will move to IDLE +*/ +enum mei_wdt_state { +MEI_WDT_IDLE , +MEI_WDT_START , +MEI_WDT_RUNNING , +MEI_WDT_STOPPING , +}; + +/** +* struct mei_wdt - mei watchdog driver +* @wdd: watchdog device +* @refcnt: watchdog device reference counter +* +* @state: watchdog internal state +* @cldev: mei watchdog client device +* @timeout: watchdog current timeout +*/ +struct mei_wdt { +struct watchdog_device wdd; +struct kref refcnt; + +struct mei_cl_device *cldev; +enum mei_wdt_state state; +u16 timeout; +}; + +/* +* struct mei_mc_hdr - Management Control Command Header +* +* @command: Management Control (0x2) +* @bytecount: Number of bytes in the message beyond this byte +* @subcommand: Management Control Subcommand +* @versionnumber: Management Control Version (0x10) +*/ +struct mei_mc_hdr { +u8 command; +u8 bytecount; +u8 subcommand; +u8 versionnumber; +}; + +/** +* struct mei_wdt_start_request watchdog start/ping +* +* @hdr: Management Control Command Header +* @timeout: timeout value +* @reserved: reserved (legacy) +*/ +struct mei_wdt_start_request { +struct mei_mc_hdr hdr; +u16 timeout; +u8 reserved[17 ]; +} __packed; + +/** +* struct mei_wdt_stop_request - watchdog stop +* +* @hdr: Management Control Command Header +*/ +struct mei_wdt_stop_request { +struct mei_mc_hdr hdr; +} __packed; + +/** +* mei_wdt_ping - send wd start/ping command +* +* @wdt: mei watchdog device +* +* Return: 0 on success, +* negative errno code on failure +*/ +static int mei_wdt_ping (struct mei_wdt *wdt) +{ +struct mei_wdt_start_request req; +const size_t req_len =sizeof (req); +int ret; + +memset (&req,0 , req_len); + req.hdr.command =MEI_MANAGEMENT_CONTROL ; + req.hdr.bytecount = req_len - offsetof(struct mei_mc_hdr , subcommand); + req.hdr.subcommand =MEI_MC_START_WD_TIMER_REQ ; + req.hdr.versionnumber =MEI_MC_VERSION_NUMBER ; + req.timeout = wdt->timeout; + + ret =mei_cldev_send (wdt->cldev, (u8 *)&req, req_len); +if (ret <0 ) +return ret; + +return 0 ; +} + +/** +* mei_wdt_stop - send wd stop command +* +* @wdt: mei watchdog device +* +* Return: 0 on success, +* negative errno code on failure +*/ +static int mei_wdt_stop (struct mei_wdt *wdt) +{ +struct mei_wdt_stop_request req; +const size_t req_len =sizeof (req); +int ret; + +memset (&req,0 , req_len); + req.hdr.command =MEI_MANAGEMENT_CONTROL ; + req.hdr.bytecount = req_len - offsetof(struct mei_mc_hdr , subcommand); + req.hdr.subcommand =MEI_MC_STOP_WD_TIMER_REQ ; + req.hdr.versionnumber =MEI_MC_VERSION_NUMBER ; + + ret =mei_cldev_send (wdt->cldev, (u8 *)&req, req_len); +if (ret <0 ) +return ret; + +return 0 ; +} + +static void mei_wdt_release (struct kref *ref) +{ + struct mei_wdt *wdt = container_of(ref, struct mei_wdt, refcnt); + +kfree (wdt); +} + +/** +* mei_wdt_register - register with the watchdog subsystem +* +* @wdt: mei watchdog device +* +* Return: 0 if success, negative errno code for failure +*/ +static int mei_wdt_register (struct mei_wdt *wdt) +{ +struct device *dev; +int ret; + + dev = &wdt->cldev->dev; + +watchdog_set_drvdata (&wdt->wdd, wdt); + + ret =watchdog_register_device (&wdt->wdd); +if (ret) { +dev_err (dev,"unable to register watchdog device = %d.\n" , ret); +return ret; + } + +return 0 ; +} + +/** +* mei_wdt_ops_start - wd start command from the watchdog core. +* +* @wdd: watchdog device +* +* Return: 0 +*/ +static int mei_wdt_ops_start (struct watchdog_device *wdd) +{ +struct mei_wdt *wdt =watchdog_get_drvdata (wdd); + + wdt->state =MEI_WDT_START ; + wdd->timeout = wdt->timeout; +return 0 ; +} + +/** +* mei_wdt_ops_stop - wd stop command from the watchdog core. +* +* @wdd: watchdog device +* +* Return: 0 if success, negative errno code for failure +*/ +static int mei_wdt_ops_stop (struct watchdog_device *wdd) +{ +struct mei_wdt *wdt =watchdog_get_drvdata (wdd); +int ret; + +if (wdt->state !=MEI_WDT_RUNNING ) +return 0 ; + + wdt->state =MEI_WDT_STOPPING ; + + ret =mei_wdt_stop (wdt); +if (ret) +return ret; + + wdt->state =MEI_WDT_IDLE ; + +return 0 ; +} + +/** +* mei_wdt_ops_ping - wd ping command from the watchdog core. +* +* @wdd: watchdog device +* +* Return: 0 if success, negative errno code on failure +*/ +static int mei_wdt_ops_ping (struct watchdog_device *wdd) +{ +struct mei_wdt *wdt =watchdog_get_drvdata (wdd); +int ret; + +if (wdt->state !=MEI_WDT_START && wdt->state !=MEI_WDT_RUNNING ) +return 0 ; + + ret =mei_wdt_ping (wdt); +if (ret) +return ret; + + wdt->state =MEI_WDT_RUNNING ; + +return 0 ; +} + +/** +* mei_wdt_ops_set_timeout - wd set timeout command from the watchdog core. +* +* @wdd: watchdog device +* @timeout: timeout value to set +* +* Return: 0 if success, negative errno code for failure +*/ +static int mei_wdt_ops_set_timeout (struct watchdog_device *wdd, +unsigned int timeout) +{ +struct mei_wdt *wdt =watchdog_get_drvdata (wdd); + +/* valid value is already checked by the caller */ + wdt->timeout = timeout; + wdd->timeout = timeout; + +return 0 ; +} + +static void mei_wdt_ops_ref (struct watchdog_device *wdd) +{ +struct mei_wdt *wdt =watchdog_get_drvdata (wdd); + +kref_get (&wdt->refcnt); +} + +static void mei_wdt_ops_unref (struct watchdog_device *wdd) +{ +struct mei_wdt *wdt =watchdog_get_drvdata (wdd); + +kref_put (&wdt->refcnt, mei_wdt_release); +} + +static const struct watchdog_ops wd_ops = { + .owner =THIS_MODULE , + .start = mei_wdt_ops_start, + .stop = mei_wdt_ops_stop, + .ping = mei_wdt_ops_ping, + .set_timeout = mei_wdt_ops_set_timeout, + .ref = mei_wdt_ops_ref, + .unref = mei_wdt_ops_unref, +}; + +static struct watchdog_info wd_info = { + .identity =INTEL_AMT_WATCHDOG_ID , + .options =WDIOF_KEEPALIVEPING | +WDIOF_SETTIMEOUT | +WDIOF_ALARMONLY , +}; + +static int mei_wdt_probe (struct mei_cl_device *cldev, +const struct mei_cl_device_id *id) +{ +struct mei_wdt *wdt; +int ret; + + wdt =kzalloc (sizeof (struct mei_wdt ),GFP_KERNEL ); +if (!wdt) +return -ENOMEM ; + +kref_init (&wdt->refcnt); + wdt->timeout =MEI_WDT_DEFAULT_TIMEOUT ; + wdt->state =MEI_WDT_IDLE ; + wdt->cldev = cldev; +mei_cldev_set_drvdata (cldev, wdt); + + ret =mei_cldev_enable (cldev); +if (ret <0 ) { +dev_err (&cldev->dev,"Could not enable cl device\n" ); + goto err_out; + } + + wd_info.firmware_version =mei_cldev_ver (cldev); + + wdt->wdd.info = &wd_info; + wdt->wdd.ops = &wd_ops; + wdt->wdd.parent = &cldev->dev; + wdt->wdd.timeout = wdt->timeout; + wdt->wdd.min_timeout =MEI_WDT_MIN_TIMEOUT ; + wdt->wdd.max_timeout =MEI_WDT_MAX_TIMEOUT ; + + ret =mei_wdt_register (wdt); +if (ret) + goto err_disable; + +return 0 ; + +err_disable: +mei_cldev_disable (cldev); + +err_out: +kref_put (&wdt->refcnt, mei_wdt_release); + +return ret; +} + +static int mei_wdt_remove (struct mei_cl_device *cldev) +{ +struct mei_wdt *wdt =mei_cldev_get_drvdata (cldev); + +watchdog_unregister_device (&wdt->wdd); + +kref_put (&wdt->refcnt, mei_wdt_release); + +mei_cldev_disable (cldev); + +return 0 ; +} + +#define MEI_UUID_WD UUID_LE(0x05B79A6F, 0x4628, 0x4D7F, \ + 0x89, 0x9D, 0xA9, 0x15, 0x14, 0xCB, 0x32, 0xAB) + +static struct mei_cl_device_id mei_wdt_tbl[] = { + { .uuid =MEI_UUID_WD , .version =0x1 }, +/* required last entry */ + { } +}; +MODULE_DEVICE_TABLE (mei, mei_wdt_tbl); + +static struct mei_cl_driver mei_wdt_driver = { + .id_table = mei_wdt_tbl, + .name =KBUILD_MODNAME , + + .probe = mei_wdt_probe, + .remove = mei_wdt_remove, +}; + +static int __init mei_wdt_init (void ) +{ +int ret; + + ret =mei_cldev_driver_register (&mei_wdt_driver); +if (ret) { +pr_err (KBUILD_MODNAME ": module registration failed\n" ); +return ret; + } +return 0 ; +} + +static void __exit mei_wdt_exit(void) +{ +mei_cldev_driver_unregister (&mei_wdt_driver); +} + +module_init (mei_wdt_init); +module_exit (mei_wdt_exit); + +MODULE_AUTHOR ("Intel Corporation" ); +MODULE_LICENSE ("GPL" ); +MODULE_DESCRIPTION ("Device driver for Intel MEI iAMT watchdog" );
PATCH v2 4/7 watchdog: mei_wdt: add status debugfs entry
In reply to Tomas WinklerAdd entry for dumping current watchdog internal state
--- drivers/watchdog/mei_wdt.c | 84 ++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 84 insertions(+)
drivers/watchdog/mei_wdt.c +84 -0
@@ -16,6 +16,7 @@#include <linux/slab.h> #include <linux/interrupt.h> #include <linux/watchdog.h> +#include <linux/debugfs.h> #include <linux/uuid.h> #include <linux/mei_cl_bus.h>
@@ -54,6 +55,19 @@ enum mei_wdt_state { MEI_WDT_STOPPING, }; +#if IS_ENABLED (CONFIG_DEBUG_FS ) +static const char *mei_wdt_state_str (enum mei_wdt_state state) +{ +switch (state) { +case MEI_WDT_IDLE :return "IDLE" ; +case MEI_WDT_START :return "START" ; +case MEI_WDT_RUNNING :return "RUNNING" ; +case MEI_WDT_STOPPING :return "STOPPING" ; +default :return "unknown" ; + } +} +#endif /* CONFIG_DEBUG_FS */ +/** * struct mei_wdt - mei watchdog driver * @wdd: watchdog device
@@ -62,6 +76,8 @@ enum mei_wdt_state {* @state: watchdog internal state * @cldev: mei watchdog client device * @timeout: watchdog current timeout +* +* @dbgfs_dir: debugfs dir entry */ struct mei_wdt {struct watchdog_device wdd;
@@ -70,6 +86,10 @@ struct mei_wdt {struct mei_cl_device *cldev;enum mei_wdt_state state;u16 timeout; + +#if IS_ENABLED (CONFIG_DEBUG_FS ) +struct dentry *dbgfs_dir; +#endif /* CONFIG_DEBUG_FS */ };/*
@@ -296,6 +316,66 @@ static void mei_wdt_ops_unref(struct watchdog_device *wdd)kref_put (&wdt->refcnt, mei_wdt_release); } +#if IS_ENABLED (CONFIG_DEBUG_FS ) + +static ssize_t mei_dbgfs_read_state (struct file *file,char __user *ubuf, +size_t cnt,loff_t *ppos) +{ +struct mei_wdt *wdt = file->private_data; +const size_t bufsz =32 ; +char buf[32 ]; +ssize_t pos =0 ; + + pos +=scnprintf (buf + pos, bufsz - pos,"state: %s\n" , +mei_wdt_state_str (wdt->state)); + +return simple_read_from_buffer (ubuf, cnt, ppos, buf, pos); +} + +static const struct file_operations dbgfs_fops_state = { + .open = simple_open, + .read = mei_dbgfs_read_state, + .llseek = generic_file_llseek, +}; + +static void dbgfs_deregister (struct mei_wdt *wdt) +{ +if (!wdt->dbgfs_dir) +return ; +debugfs_remove_recursive (wdt->dbgfs_dir); + wdt->dbgfs_dir =NULL ; +} + +static int dbgfs_register (struct mei_wdt *wdt) +{ +struct dentry *dir, *f; + + dir =debugfs_create_dir (KBUILD_MODNAME ,NULL ); +if (!dir) +return -ENOMEM ; + + wdt->dbgfs_dir = dir; + f =debugfs_create_file ("state" ,S_IRUSR , dir, wdt, &dbgfs_fops_state); +if (!f) + goto err; + +return 0 ; +err: +dbgfs_deregister (wdt); +return -ENODEV ; +} + +#else + +static inline void dbgfs_deregister (struct mei_wdt *wdt) {} + +static inline int dbgfs_register (struct mei_wdt *wdt) +{ +return 0 ; +} +#endif /* CONFIG_DEBUG_FS */ + +static const struct watchdog_ops wd_ops = { .owner =THIS_MODULE , .start = mei_wdt_ops_start,
@@ -348,6 +428,8 @@ static int mei_wdt_probe(struct mei_cl_device *cldev,if (ret) goto err_disable; +dbgfs_register (wdt); +return 0 ; err_disable:
@@ -367,6 +449,8 @@ static int mei_wdt_remove(struct mei_cl_device *cldev)kref_put (&wdt->refcnt, mei_wdt_release); +dbgfs_deregister (wdt); +mei_cldev_disable (cldev);return 0 ;
Signature · 1 line
2.4.3
Guenter Roeck
maintainerRe: [char-misc-next 4/7] watchdog: mei_wdt: add status debugfs entry
Quoted 22 lines · Tomas Winkler
Add entry for dumping current watchdog internal state
Signed-off-by Tomas Winkler <[email protected]>--- drivers/watchdog/mei_wdt.c | 84 ++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 84 insertions(+)drivers/watchdog/mei_wdt.c +1
-0@@ -16,6 +16,7 @@#include <linux/slab.h> #include <linux/interrupt.h> #include <linux/watchdog.h> +#include <linux/debugfs.h> #include <linux/uuid.h> #include <linux/mei_cl_bus.h> @@ -54,6 +55,19 @@ enum mei_wdt_state { MEI_WDT_STOPPING, };
drivers/watchdog/mei_wdt.c enum mei_wdt_state { · line 66
Quoted 10 lines
drivers/watchdog/mei_wdt.c +9
-0+#if IS_ENABLED (CONFIG_DEBUG_FS ) +static const char *mei_wdt_state_str (enum mei_wdt_state state) +{ +switch (state) { +case MEI_WDT_IDLE :return "IDLE" ; +case MEI_WDT_START :return "START" ; +case MEI_WDT_RUNNING :return "RUNNING" ; +case MEI_WDT_STOPPING :return "STOPPING" ; +default :return "unknown" ;
Doesn't this cause checkpatch warnings ?
drivers/watchdog/mei_wdt.c enum mei_wdt_state { · line 70
Quoted 4 lines
drivers/watchdog/mei_wdt.c +4
-0+ } +} +#endif /* CONFIG_DEBUG_FS */ +
Can you move this conditional code into the other #ifdef block below ?
Thanks,
Guenter
Quoted 100 lines
drivers/watchdog/mei_wdt.c +68
-0/** * struct mei_wdt - mei watchdog driver * @wdd: watchdog device @@ -62,6 +76,8 @@ enum mei_wdt_state {* @state: watchdog internal state * @cldev: mei watchdog client device * @timeout: watchdog current timeout +* +* @dbgfs_dir: debugfs dir entry */ struct mei_wdt {struct watchdog_device wdd;@@ -70,6 +86,10 @@ struct mei_wdt {struct mei_cl_device *cldev;enum mei_wdt_state state;u16 timeout; + +#if IS_ENABLED (CONFIG_DEBUG_FS ) +struct dentry *dbgfs_dir; +#endif /* CONFIG_DEBUG_FS */ };/* @@ -296,6 +316,66 @@ static void mei_wdt_ops_unref(struct watchdog_device *wdd)kref_put (&wdt->refcnt, mei_wdt_release); } +#if IS_ENABLED (CONFIG_DEBUG_FS ) + +static ssize_t mei_dbgfs_read_state (struct file *file,char __user *ubuf, +size_t cnt,loff_t *ppos) +{ +struct mei_wdt *wdt = file->private_data; +const size_t bufsz =32 ; +char buf[32 ]; +ssize_t pos =0 ; + + pos +=scnprintf (buf + pos, bufsz - pos,"state: %s\n" , +mei_wdt_state_str (wdt->state)); + +return simple_read_from_buffer (ubuf, cnt, ppos, buf, pos); +} + +static const struct file_operations dbgfs_fops_state = { + .open = simple_open, + .read = mei_dbgfs_read_state, + .llseek = generic_file_llseek, +}; + +static void dbgfs_deregister (struct mei_wdt *wdt) +{ +if (!wdt->dbgfs_dir) +return ; +debugfs_remove_recursive (wdt->dbgfs_dir); + wdt->dbgfs_dir =NULL ; +} + +static int dbgfs_register (struct mei_wdt *wdt) +{ +struct dentry *dir, *f; + + dir =debugfs_create_dir (KBUILD_MODNAME ,NULL ); +if (!dir) +return -ENOMEM ; + + wdt->dbgfs_dir = dir; + f =debugfs_create_file ("state" ,S_IRUSR , dir, wdt, &dbgfs_fops_state); +if (!f) + goto err; + +return 0 ; +err: +dbgfs_deregister (wdt); +return -ENODEV ; +} + +#else + +static inline void dbgfs_deregister (struct mei_wdt *wdt) {} + +static inline int dbgfs_register (struct mei_wdt *wdt) +{ +return 0 ; +} +#endif /* CONFIG_DEBUG_FS */ + +static const struct watchdog_ops wd_ops = { .owner =THIS_MODULE , .start = mei_wdt_ops_start,@@ -348,6 +428,8 @@ static int mei_wdt_probe(struct mei_cl_device *cldev,if (ret) goto err_disable; +dbgfs_register (wdt); +return 0 ; err_disable:
drivers/watchdog/mei_wdt.c static int mei_wdt_remove(struct mei_cl_device *cldev) · line 453
Quoted 5 lines
drivers/watchdog/mei_wdt.c +2
-0@@ -367,6 +449,8 @@ static int mei_wdt_remove(struct mei_cl_device *cldev)kref_put (&wdt->refcnt, mei_wdt_release); +dbgfs_deregister (wdt); +
This should probably come before the call to kref_put() since it uses wdt.
Quoted 4 lines
drivers/watchdog/mei_wdt.c +0
-0mei_cldev_disable (cldev);return 0 ;
RE: [char-misc-next 4/7] watchdog: mei_wdt: add status debugfs entry
Quoted 26 lines · Tomas Winkler
Quoted 5 lines
Add entry for dumping current watchdog internal state
Signed-off-by Tomas Winkler <[email protected]>--- drivers/watchdog/mei_wdt.c | 84++++++++++++++++++++++++++++++++++++++++++++++
Quoted 18 lines
1 file changed, 84 insertions(+)drivers/watchdog/mei_wdt.c +1
-0@@ -16,6 +16,7 @@#include <linux/slab.h> #include <linux/interrupt.h> #include <linux/watchdog.h> +#include <linux/debugfs.h> #include <linux/uuid.h> #include <linux/mei_cl_bus.h> @@ -54,6 +55,19 @@ enum mei_wdt_state { MEI_WDT_STOPPING, };
Quoted 11 lines
drivers/watchdog/mei_wdt.c enum mei_wdt_state { · line 66
Quoted 9 lines
drivers/watchdog/mei_wdt.c +9
-0+#if IS_ENABLED (CONFIG_DEBUG_FS ) +static const char *mei_wdt_state_str (enum mei_wdt_state state) +{ +switch (state) { +case MEI_WDT_IDLE :return "IDLE" ; +case MEI_WDT_START :return "START" ; +case MEI_WDT_RUNNING :return "RUNNING" ; +case MEI_WDT_STOPPING :return "STOPPING" ; +default :return "unknown" ;Doesn't this cause checkpatch warnings ?
It doesn't and also checkpatch.pl is just a help tool.
Quoted 7 lines
drivers/watchdog/mei_wdt.c enum mei_wdt_state { · line 70
Quoted 5 lines
drivers/watchdog/mei_wdt.c +4
-0+ } +} +#endif /* CONFIG_DEBUG_FS */ +Can you move this conditional code into the other #ifdef block below ?
I'm removing the ifdef in the following patch so this is just for this path to not to spill warning the compilation w/o CONFIG_DEBUG_FS set.
.start = mei_wdt_ops_start,
Quoted 11 lines
Quoted 11 lines
drivers/watchdog/mei_wdt.c +2
-0@@ -348,6 +428,8 @@ static int mei_wdt_probe(struct mei_cl_device *cldev,if (ret) goto err_disable; +dbgfs_register (wdt); +return 0 ; err_disable:@@ -367,6 +449,8 @@ static int mei_wdt_remove(struct mei_cl_device *cldev)
Quoted 6 lines
drivers/watchdog/mei_wdt.c static void mei_wdt_ops_unref(struct watchdog_device *wdd) · line 367
Quoted 4 lines
drivers/watchdog/mei_wdt.c +2
-0kref_put (&wdt->refcnt, mei_wdt_release); +dbgfs_deregister (wdt); +This should probably come before the call to
kref_put()since it uses wdt.
Good catch, will resend.
Tomas
Guenter Roeck
maintainerRe: [char-misc-next 4/7] watchdog: mei_wdt: add status debugfs entry
Quoted 37 lines · Tomas
Quoted 26 lines · Tomas Winkler
Quoted 5 lines
Add entry for dumping current watchdog internal state
Signed-off-by Tomas Winkler <[email protected]>--- drivers/watchdog/mei_wdt.c | 84++++++++++++++++++++++++++++++++++++++++++++++
Quoted 18 lines
1 file changed, 84 insertions(+)drivers/watchdog/mei_wdt.c +1
-0@@ -16,6 +16,7 @@#include <linux/slab.h> #include <linux/interrupt.h> #include <linux/watchdog.h> +#include <linux/debugfs.h> #include <linux/uuid.h> #include <linux/mei_cl_bus.h> @@ -54,6 +55,19 @@ enum mei_wdt_state { MEI_WDT_STOPPING, };Quoted 11 lines
drivers/watchdog/mei_wdt.c enum mei_wdt_state { · line 66
Quoted 9 lines
drivers/watchdog/mei_wdt.c +9
-0+#if IS_ENABLED (CONFIG_DEBUG_FS ) +static const char *mei_wdt_state_str (enum mei_wdt_state state) +{ +switch (state) { +case MEI_WDT_IDLE :return "IDLE" ; +case MEI_WDT_START :return "START" ; +case MEI_WDT_RUNNING :return "RUNNING" ; +case MEI_WDT_STOPPING :return "STOPPING" ; +default :return "unknown" ;Doesn't this cause checkpatch warnings ?
Quoted 1 line
It doesn't and also checkpatch.pl is just a help tool.
I should have said "Doesn't this violate the single-statement-per-line" coding style rule, but I guess then you'd argue that "case xxx:" is not a statement.
Guenter
RE: [char-misc-next 4/7] watchdog: mei_wdt: add status debugfs entry
Quoted 13 lines
Quoted 13 lines
Quoted 13 lines
drivers/watchdog/mei_wdt.c enum mei_wdt_state { · line 66
Quoted 11 lines
drivers/watchdog/mei_wdt.c +9
-0+#if IS_ENABLED (CONFIG_DEBUG_FS ) +static const char *mei_wdt_state_str (enum mei_wdt_state state) +{ +switch (state) { +case MEI_WDT_IDLE :return "IDLE" ; +case MEI_WDT_START :return "START" ; +case MEI_WDT_RUNNING :return "RUNNING" ; +case MEI_WDT_STOPPING :return "STOPPING" ; +default :return "unknown" ;Doesn't this cause checkpatch warnings ?
Quoted 5 lines
Quoted 1 line
It doesn't and also checkpatch.pl is just a help tool.
I should have said "Doesn't this violate the single-statement-per-line" coding style rule, but I guess then you'd argue that "case xxx:" is not a statement.
I believe it is reasonably readable , but I don't have a strong feeling about it, if you do I can change it.
Tomas
PATCH v2 5/7 mei: bus: whitelist the watchdog client
In reply to Tomas WinklerThe iAMT WD client has to be whitelisted sice it has two connections and is filtered out by number_of_connections fixup.
Also the API has changed for BDW and SKL but firmware haven't updated the protocol version.
---
V2: resend
drivers/misc/mei/bus-fixup.c | 29 +++++++++++++++++++++++++++++ 1 file changed, 29 insertions(+)
drivers/misc/mei/bus-fixup.c +29 -0
@@ -35,6 +35,9 @@ static const uuid_le mei_nfc_info_guid = MEI_UUID_NFC_INFO;#define MEI_UUID_NFC_HCI UUID_LE(0x0bb17a78, 0x2a8e, 0x4c50, \ 0x94, 0xd4, 0x50, 0x26, 0x67, 0x23, 0x77, 0x5c) +#define MEI_UUID_WD UUID_LE(0x05B79A6F, 0x4628, 0x4D7F, \ + 0x89, 0x9D, 0xA9, 0x15, 0x14, 0xCB, 0x32, 0xAB) +#define MEI_UUID_ANY NULL_UUID_LE/**
@@ -66,6 +69,31 @@ static void blacklist(struct mei_cl_device *cldev) cldev->do_match =0 ; } +/** +* mei_wd - wd client on the bus, change protocol version +* as the API has changed. +* +* @cldev: me clients device +*/ +#if IS_ENABLED (CONFIG_INTEL_MEI_ME ) +#include <linux/pci.h> +#include "hw-me-regs.h" +static void mei_wd (struct mei_cl_device *cldev) +{ +struct pci_dev *pdev =to_pci_dev (cldev->dev.parent); + +dev_dbg (&cldev->dev,"running hook %s\n" , __func__); +if (pdev->device ==MEI_DEV_ID_WPT_LP || + pdev->device ==MEI_DEV_ID_SPT || + pdev->device ==MEI_DEV_ID_SPT_H ) + cldev->me_cl->props.protocol_version =0x2 ; + + cldev->do_match =1 ; +} +#else +static inline void mei_wd (struct mei_cl_device *cldev) {} +#endif /* CONFIG_INTEL_MEI_ME */ +struct mei_nfc_cmd {u8 command;u8 status;
@@ -280,6 +308,7 @@ static struct mei_fixup {MEI_FIXUP (MEI_UUID_ANY , number_of_connections),MEI_FIXUP (MEI_UUID_NFC_INFO , blacklist),MEI_FIXUP (MEI_UUID_NFC_HCI , mei_nfc), + MEI_FIXUP(MEI_UUID_WD, mei_wd), };/**
Signature · 1 line
2.4.3
PATCH v2 6/7 watchdog: mei_wdt: register wd device only if required
In reply to Tomas WinklerFrom: Alexander Usyskin <[email protected]>
For Intel Broadwell and newer platforms, the ME device can inform the host whether the watchdog functionality is activated or not. If the watchdog functionality is not activated then the watchdog interface can be not registered and eliminate unnecessary pings and hence lower the power consumption by avoiding waking up the device.
---
V2: rework unregistration
drivers/watchdog/mei_wdt.c | 180 +++++++++++++++++++++++++++++++++++++++++---- 1 file changed, 167 insertions(+), 13 deletions(-)
drivers/watchdog/mei_wdt.c +167 -13
@@ -16,6 +16,7 @@#include <linux/slab.h> #include <linux/interrupt.h> #include <linux/watchdog.h> +#include <linux/completion.h> #include <linux/debugfs.h> #include <linux/uuid.h>
@@ -38,43 +39,54 @@/* Sub Commands */ #define MEI_MC_START_WD_TIMER_REQ 0x13 +#define MEI_MC_START_WD_TIMER_RES 0x83 +#define MEI_WDT_STATUS_SUCCESS 0 +#define MEI_WDT_WDSTATE_NOT_REQUIRED 0x1#define MEI_MC_STOP_WD_TIMER_REQ 0x14/** * enum mei_wdt_state - internal watchdog state * +* @MEI_WDT_PROBE: wd in probing stage * @MEI_WDT_IDLE: wd is idle and not opened * @MEI_WDT_START: wd was opened, start was called * @MEI_WDT_RUNNING: wd is expecting keep alive pings * @MEI_WDT_STOPPING: wd is stopping and will move to IDLE +* @MEI_WDT_NOT_REQUIRED: wd device is not required */ enum mei_wdt_state { +MEI_WDT_PROBE ,MEI_WDT_IDLE ,MEI_WDT_START ,MEI_WDT_RUNNING ,MEI_WDT_STOPPING , +MEI_WDT_NOT_REQUIRED , }; -#if IS_ENABLED (CONFIG_DEBUG_FS )static const char *mei_wdt_state_str (enum mei_wdt_state state) {switch (state) { +case MEI_WDT_PROBE :return "PROBE" ;case MEI_WDT_IDLE :return "IDLE" ;case MEI_WDT_START :return "START" ;case MEI_WDT_RUNNING :return "RUNNING" ;case MEI_WDT_STOPPING :return "STOPPING" ; +case MEI_WDT_NOT_REQUIRED :return "NOT_REQUIRED" ;default :return "unknown" ; } } -#endif /* CONFIG_DEBUG_FS */ /** * struct mei_wdt - mei watchdog driver * @wdd: watchdog device * @refcnt: watchdog device reference counter +* @reg_lock: watchdog device registration lock +* @is_registered: is watchdog device registered * -* @state: watchdog internal state * @cldev: mei watchdog client device +* @state: watchdog internal state +* @resp_required: ping required response +* @response: ping response * @timeout: watchdog current timeout * * @dbgfs_dir: debugfs dir entry
@@ -82,9 +94,13 @@ static const char *mei_wdt_state_str(enum mei_wdt_state state)struct mei_wdt {struct watchdog_device wdd;struct kref refcnt; +struct mutex reg_lock; +bool is_registered;struct mei_cl_device *cldev;enum mei_wdt_state state; +bool resp_required; +struct completion response;u16 timeout;#if IS_ENABLED (CONFIG_DEBUG_FS )
@@ -121,6 +137,19 @@ struct mei_wdt_start_request { } __packed;/** +* struct mei_wdt_start_response watchdog start/ping response +* +* @hdr: Management Control Command Header +* @status: operation status +* @wdstate: watchdog status bit mask +*/ +struct mei_wdt_start_response { +struct mei_mc_hdr hdr; +u8 status; +u8 wdstate; +} __packed; + +/** * struct mei_wdt_stop_request - watchdog stop * * @hdr: Management Control Command Header
@@ -192,6 +221,23 @@ static void mei_wdt_release(struct kref *ref) }/** +* mei_wdt_unregister - unregister from the watchdog subsystem +* +* @wdt: mei watchdog device +*/ +static void mei_wdt_unregister (struct mei_wdt *wdt) +{ + +mutex_lock (&wdt->reg_lock); +if (wdt->is_registered) { +clear_bit (WDOG_ACTIVE , &wdt->wdd.status); +watchdog_unregister_device (&wdt->wdd); + wdt->is_registered = false; + } +mutex_unlock (&wdt->reg_lock); +} + +/** * mei_wdt_register - register with the watchdog subsystem * * @wdt: mei watchdog device
@@ -205,15 +251,20 @@ static int mei_wdt_register(struct mei_wdt *wdt) dev = &wdt->cldev->dev; -watchdog_set_drvdata (&wdt->wdd, wdt); + wdt->state =MEI_WDT_IDLE ; +mutex_lock (&wdt->reg_lock); +clear_bit (WDOG_UNREGISTERED , &wdt->wdd.status); +watchdog_set_drvdata (&wdt->wdd, wdt); ret =watchdog_register_device (&wdt->wdd);if (ret) {dev_err (dev,"unable to register watchdog device = %d.\n" , ret); -return ret; +watchdog_set_drvdata (&wdt->wdd,NULL ); } + wdt->is_registered = true; +mutex_unlock (&wdt->reg_lock); -return 0 ; +return ret; }/**
@@ -259,6 +310,83 @@ static int mei_wdt_ops_stop(struct watchdog_device *wdd) }/** +* mei_wdt_event_rx - callback for data receive +* +* @cldev: bus device +*/ +static void mei_wdt_event_rx (struct mei_cl_device *cldev) +{ +struct mei_wdt *wdt =mei_cldev_get_drvdata (cldev); +struct mei_wdt_start_response res; +const size_t res_len =sizeof (res); +int ret; + + ret =mei_cldev_recv (wdt->cldev, (u8 *)&res, res_len); +if (ret <0 ) { +dev_err (&cldev->dev,"failure in recv %d\n" , ret); +return ; + } + +/* Empty response can be sent on stop */ +if (ret ==0 ) +return ; + +if (ret <sizeof (struct mei_mc_hdr )) { +dev_err (&cldev->dev,"recv small data %d\n" , ret); +return ; + } + +if (res.hdr.command !=MEI_MANAGEMENT_CONTROL || + res.hdr.versionnumber !=MEI_MC_VERSION_NUMBER ) { +dev_err (&cldev->dev,"wrong command received\n" ); +return ; + } + +if (res.hdr.subcommand !=MEI_MC_START_WD_TIMER_RES ) { +dev_warn (&cldev->dev,"unsupported command %d :%s[%d]\n" , + res.hdr.subcommand, +mei_wdt_state_str (wdt->state), + wdt->state); +return ; + } + +if (wdt->state ==MEI_WDT_RUNNING ) + goto out; + +if (wdt->state ==MEI_WDT_PROBE ) { +if (res.wdstate &MEI_WDT_WDSTATE_NOT_REQUIRED ) { + wdt->state =MEI_WDT_NOT_REQUIRED ; + }else { +/* stop the ping register watchdog device */ +mei_wdt_stop (wdt); +mei_wdt_register (wdt); + } +return ; + } + +dev_warn (&cldev->dev,"not in correct state %s[%d]\n" , +mei_wdt_state_str (wdt->state), wdt->state); + +out: +if (!completion_done (&wdt->response)) +complete (&wdt->response); +} + +/** +* mei_wdt_event - callback for event receive +* +* @cldev: bus device +* @events: event mask +* @context: callback context +*/ +static void mei_wdt_event (struct mei_cl_device *cldev, +u32 events,void *context) +{ +if (events &BIT (MEI_CL_EVENT_RX )) +mei_wdt_event_rx (cldev); +} + +/** * mei_wdt_ops_ping - wd ping command from the watchdog core. * * @wdd: watchdog device
@@ -273,13 +401,18 @@ static int mei_wdt_ops_ping(struct watchdog_device *wdd)if (wdt->state !=MEI_WDT_START && wdt->state !=MEI_WDT_RUNNING )return 0 ; +if (wdt->resp_required) +init_completion (&wdt->response); + + wdt->state =MEI_WDT_RUNNING ; ret =mei_wdt_ping (wdt);if (ret)return ret; - wdt->state =MEI_WDT_RUNNING ; +if (wdt->resp_required) + ret =wait_for_completion_killable (&wdt->response); -return 0 ; +return ret; }/**
@@ -405,8 +538,13 @@ static int mei_wdt_probe(struct mei_cl_device *cldev,kref_init (&wdt->refcnt); wdt->timeout =MEI_WDT_DEFAULT_TIMEOUT ; - wdt->state =MEI_WDT_IDLE ; + wdt->state =MEI_WDT_PROBE ; wdt->cldev = cldev; + wdt->resp_required =mei_cldev_ver (cldev) >0x1 ; +mutex_init (&wdt->reg_lock); + wdt->is_registered = false; +init_completion (&wdt->response); +mei_cldev_set_drvdata (cldev, wdt); ret =mei_cldev_enable (cldev);
@@ -415,6 +553,13 @@ static int mei_wdt_probe(struct mei_cl_device *cldev, goto err_out; } + ret =mei_cldev_register_event_cb (wdt->cldev,BIT (MEI_CL_EVENT_RX ), + mei_wdt_event,NULL ); +if (ret) { +dev_err (&cldev->dev,"Could not register event ret=%d\n" , ret); + goto err_disable; + } + wd_info.firmware_version =mei_cldev_ver (cldev); wdt->wdd.info = &wd_info;
@@ -424,8 +569,13 @@ static int mei_wdt_probe(struct mei_cl_device *cldev, wdt->wdd.min_timeout =MEI_WDT_MIN_TIMEOUT ; wdt->wdd.max_timeout =MEI_WDT_MAX_TIMEOUT ; - ret =mei_wdt_register (wdt); -if (ret) +/* register after ping response */ +if (wdt->resp_required) + ret =mei_wdt_ping (wdt); +else + ret =mei_wdt_register (wdt); + +if (ret <0 ) goto err_disable;dbgfs_register (wdt);
@@ -445,7 +595,11 @@ static int mei_wdt_remove(struct mei_cl_device *cldev) {struct mei_wdt *wdt =mei_cldev_get_drvdata (cldev); -watchdog_unregister_device (&wdt->wdd); +/* Free the caller in case of fw initiated or unexpected reset */ +if (!completion_done (&wdt->response)) +complete (&wdt->response); + +mei_wdt_unregister (wdt);kref_put (&wdt->refcnt, mei_wdt_release);
@@ -460,7 +614,7 @@ static int mei_wdt_remove(struct mei_cl_device *cldev) 0x89, 0x9D, 0xA9, 0x15, 0x14, 0xCB, 0x32, 0xAB)static struct mei_cl_device_id mei_wdt_tbl[] = { - { .uuid =MEI_UUID_WD , .version =0x1 }, + { .uuid =MEI_UUID_WD , .version =MEI_CL_VERSION_ANY },/* required last entry */ { } };
Signature · 1 line
2.4.3
Guenter Roeck
maintainerRe: [char-misc-next v2 6/7] watchdog: mei_wdt: register wd device only if required
Quoted 124 lines · Tomas Winkler
From: Alexander Usyskin <[email protected]>
For Intel Broadwell and newer platforms, the ME device can inform the host whether the watchdog functionality is activated or not. If the watchdog functionality is not activated then the watchdog interface can be not registered and eliminate unnecessary pings and hence lower the power consumption by avoiding waking up the device.
Signed-off-by Alexander Usyskin <[email protected]>Signed-off-by Tomas Winkler <[email protected]>---
V2: rework unregistrationdrivers/watchdog/mei_wdt.c | 180 +++++++++++++++++++++++++++++++++++++++++---- 1 file changed, 167 insertions(+), 13 deletions(-)drivers/watchdog/mei_wdt.c +32
-3@@ -16,6 +16,7 @@#include <linux/slab.h> #include <linux/interrupt.h> #include <linux/watchdog.h> +#include <linux/completion.h> #include <linux/debugfs.h> #include <linux/uuid.h> @@ -38,43 +39,54 @@/* Sub Commands */ #define MEI_MC_START_WD_TIMER_REQ 0x13 +#define MEI_MC_START_WD_TIMER_RES 0x83 +#define MEI_WDT_STATUS_SUCCESS 0 +#define MEI_WDT_WDSTATE_NOT_REQUIRED 0x1#define MEI_MC_STOP_WD_TIMER_REQ 0x14/** * enum mei_wdt_state - internal watchdog state * +* @MEI_WDT_PROBE: wd in probing stage * @MEI_WDT_IDLE: wd is idle and not opened * @MEI_WDT_START: wd was opened, start was called * @MEI_WDT_RUNNING: wd is expecting keep alive pings * @MEI_WDT_STOPPING: wd is stopping and will move to IDLE +* @MEI_WDT_NOT_REQUIRED: wd device is not required */ enum mei_wdt_state { +MEI_WDT_PROBE ,MEI_WDT_IDLE ,MEI_WDT_START ,MEI_WDT_RUNNING ,MEI_WDT_STOPPING , +MEI_WDT_NOT_REQUIRED , }; -#if IS_ENABLED (CONFIG_DEBUG_FS )static const char *mei_wdt_state_str (enum mei_wdt_state state) {switch (state) { +case MEI_WDT_PROBE :return "PROBE" ;case MEI_WDT_IDLE :return "IDLE" ;case MEI_WDT_START :return "START" ;case MEI_WDT_RUNNING :return "RUNNING" ;case MEI_WDT_STOPPING :return "STOPPING" ; +case MEI_WDT_NOT_REQUIRED :return "NOT_REQUIRED" ;default :return "unknown" ; } } -#endif /* CONFIG_DEBUG_FS */ /** * struct mei_wdt - mei watchdog driver * @wdd: watchdog device * @refcnt: watchdog device reference counter +* @reg_lock: watchdog device registration lock +* @is_registered: is watchdog device registered * -* @state: watchdog internal state * @cldev: mei watchdog client device +* @state: watchdog internal state +* @resp_required: ping required response +* @response: ping response * @timeout: watchdog current timeout * * @dbgfs_dir: debugfs dir entry @@ -82,9 +94,13 @@ static const char *mei_wdt_state_str(enum mei_wdt_state state)struct mei_wdt {struct watchdog_device wdd;struct kref refcnt; +struct mutex reg_lock; +bool is_registered;struct mei_cl_device *cldev;enum mei_wdt_state state; +bool resp_required; +struct completion response;u16 timeout;#if IS_ENABLED (CONFIG_DEBUG_FS )@@ -121,6 +137,19 @@ struct mei_wdt_start_request { } __packed;/** +* struct mei_wdt_start_response watchdog start/ping response +* +* @hdr: Management Control Command Header +* @status: operation status +* @wdstate: watchdog status bit mask +*/ +struct mei_wdt_start_response { +struct mei_mc_hdr hdr; +u8 status; +u8 wdstate; +} __packed; + +/** * struct mei_wdt_stop_request - watchdog stop * * @hdr: Management Control Command Header @@ -192,6 +221,23 @@ static void mei_wdt_release(struct kref *ref) }/**
drivers/watchdog/mei_wdt.c static void mei_wdt_release(struct kref *ref) · line 233
Quoted 10 lines
drivers/watchdog/mei_wdt.c +10
-0+* mei_wdt_unregister - unregister from the watchdog subsystem +* +* @wdt: mei watchdog device +*/ +static void mei_wdt_unregister (struct mei_wdt *wdt) +{ + +mutex_lock (&wdt->reg_lock); +if (wdt->is_registered) { +clear_bit (WDOG_ACTIVE , &wdt->wdd.status);
WDOG_ACTIVE is commonly handled by the core. Why does it have to be touched here ?
I am a bit concerned about the entire sequence.
If the following happens
- watchdog character device is opened
- watchdog goes away
- watchdog reappears
- watchdog character device is accessed
the code may access the driver through an old instance of the watchdog character device. The behavior in this situation is pretty much undefined. It may work, but only by luck, not by design.
Quoted 11 lines
drivers/watchdog/mei_wdt.c +7
-0+watchdog_unregister_device (&wdt->wdd); + wdt->is_registered = false; + } +mutex_unlock (&wdt->reg_lock); +} + +/** * mei_wdt_register - register with the watchdog subsystem * * @wdt: mei watchdog device
drivers/watchdog/mei_wdt.c static int mei_wdt_register(struct mei_wdt *wdt) · line 257
Quoted 8 lines
drivers/watchdog/mei_wdt.c +3
-1@@ -205,15 +251,20 @@ static int mei_wdt_register(struct mei_wdt *wdt) dev = &wdt->cldev->dev; -watchdog_set_drvdata (&wdt->wdd, wdt); + wdt->state =MEI_WDT_IDLE ; +mutex_lock (&wdt->reg_lock); +clear_bit (WDOG_UNREGISTERED , &wdt->wdd.status);
Please don't touch watchdog core internal states - this flag will likely go away soon.
Quoted 188 lines
drivers/watchdog/mei_wdt.c +115
-9+watchdog_set_drvdata (&wdt->wdd, wdt); ret =watchdog_register_device (&wdt->wdd);if (ret) {dev_err (dev,"unable to register watchdog device = %d.\n" , ret); -return ret; +watchdog_set_drvdata (&wdt->wdd,NULL ); } + wdt->is_registered = true; +mutex_unlock (&wdt->reg_lock); -return 0 ; +return ret; }/** @@ -259,6 +310,83 @@ static int mei_wdt_ops_stop(struct watchdog_device *wdd) }/** +* mei_wdt_event_rx - callback for data receive +* +* @cldev: bus device +*/ +static void mei_wdt_event_rx (struct mei_cl_device *cldev) +{ +struct mei_wdt *wdt =mei_cldev_get_drvdata (cldev); +struct mei_wdt_start_response res; +const size_t res_len =sizeof (res); +int ret; + + ret =mei_cldev_recv (wdt->cldev, (u8 *)&res, res_len); +if (ret <0 ) { +dev_err (&cldev->dev,"failure in recv %d\n" , ret); +return ; + } + +/* Empty response can be sent on stop */ +if (ret ==0 ) +return ; + +if (ret <sizeof (struct mei_mc_hdr )) { +dev_err (&cldev->dev,"recv small data %d\n" , ret); +return ; + } + +if (res.hdr.command !=MEI_MANAGEMENT_CONTROL || + res.hdr.versionnumber !=MEI_MC_VERSION_NUMBER ) { +dev_err (&cldev->dev,"wrong command received\n" ); +return ; + } + +if (res.hdr.subcommand !=MEI_MC_START_WD_TIMER_RES ) { +dev_warn (&cldev->dev,"unsupported command %d :%s[%d]\n" , + res.hdr.subcommand, +mei_wdt_state_str (wdt->state), + wdt->state); +return ; + } + +if (wdt->state ==MEI_WDT_RUNNING ) + goto out; + +if (wdt->state ==MEI_WDT_PROBE ) { +if (res.wdstate &MEI_WDT_WDSTATE_NOT_REQUIRED ) { + wdt->state =MEI_WDT_NOT_REQUIRED ; + }else { +/* stop the ping register watchdog device */ +mei_wdt_stop (wdt); +mei_wdt_register (wdt); + } +return ; + } + +dev_warn (&cldev->dev,"not in correct state %s[%d]\n" , +mei_wdt_state_str (wdt->state), wdt->state); + +out: +if (!completion_done (&wdt->response)) +complete (&wdt->response); +} + +/** +* mei_wdt_event - callback for event receive +* +* @cldev: bus device +* @events: event mask +* @context: callback context +*/ +static void mei_wdt_event (struct mei_cl_device *cldev, +u32 events,void *context) +{ +if (events &BIT (MEI_CL_EVENT_RX )) +mei_wdt_event_rx (cldev); +} + +/** * mei_wdt_ops_ping - wd ping command from the watchdog core. * * @wdd: watchdog device @@ -273,13 +401,18 @@ static int mei_wdt_ops_ping(struct watchdog_device *wdd)if (wdt->state !=MEI_WDT_START && wdt->state !=MEI_WDT_RUNNING )return 0 ; +if (wdt->resp_required) +init_completion (&wdt->response); + + wdt->state =MEI_WDT_RUNNING ; ret =mei_wdt_ping (wdt);if (ret)return ret; - wdt->state =MEI_WDT_RUNNING ; +if (wdt->resp_required) + ret =wait_for_completion_killable (&wdt->response); -return 0 ; +return ret; }/** @@ -405,8 +538,13 @@ static int mei_wdt_probe(struct mei_cl_device *cldev,kref_init (&wdt->refcnt); wdt->timeout =MEI_WDT_DEFAULT_TIMEOUT ; - wdt->state =MEI_WDT_IDLE ; + wdt->state =MEI_WDT_PROBE ; wdt->cldev = cldev; + wdt->resp_required =mei_cldev_ver (cldev) >0x1 ; +mutex_init (&wdt->reg_lock); + wdt->is_registered = false; +init_completion (&wdt->response); +mei_cldev_set_drvdata (cldev, wdt); ret =mei_cldev_enable (cldev);@@ -415,6 +553,13 @@ static int mei_wdt_probe(struct mei_cl_device *cldev, goto err_out; } + ret =mei_cldev_register_event_cb (wdt->cldev,BIT (MEI_CL_EVENT_RX ), + mei_wdt_event,NULL ); +if (ret) { +dev_err (&cldev->dev,"Could not register event ret=%d\n" , ret); + goto err_disable; + } + wd_info.firmware_version =mei_cldev_ver (cldev); wdt->wdd.info = &wd_info;@@ -424,8 +569,13 @@ static int mei_wdt_probe(struct mei_cl_device *cldev, wdt->wdd.min_timeout =MEI_WDT_MIN_TIMEOUT ; wdt->wdd.max_timeout =MEI_WDT_MAX_TIMEOUT ; - ret =mei_wdt_register (wdt); -if (ret) +/* register after ping response */ +if (wdt->resp_required) + ret =mei_wdt_ping (wdt); +else + ret =mei_wdt_register (wdt); + +if (ret <0 ) goto err_disable;dbgfs_register (wdt);@@ -445,7 +595,11 @@ static int mei_wdt_remove(struct mei_cl_device *cldev) {struct mei_wdt *wdt =mei_cldev_get_drvdata (cldev); -watchdog_unregister_device (&wdt->wdd); +/* Free the caller in case of fw initiated or unexpected reset */ +if (!completion_done (&wdt->response)) +complete (&wdt->response); + +mei_wdt_unregister (wdt);kref_put (&wdt->refcnt, mei_wdt_release);@@ -460,7 +614,7 @@ static int mei_wdt_remove(struct mei_cl_device *cldev) 0x89, 0x9D, 0xA9, 0x15, 0x14, 0xCB, 0x32, 0xAB)static struct mei_cl_device_id mei_wdt_tbl[] = { - { .uuid =MEI_UUID_WD , .version =0x1 }, + { .uuid =MEI_UUID_WD , .version =MEI_CL_VERSION_ANY },/* required last entry */ { } };
RE: [char-misc-next v2 6/7] watchdog: mei_wdt: register wd device only if required
Quoted 127 lines · Tomas Winkler
Quoted 8 lines
From: Alexander Usyskin <[email protected]>
For Intel Broadwell and newer platforms, the ME device can inform the host whether the watchdog functionality is activated or not. If the watchdog functionality is not activated then the watchdog interface can be not registered and eliminate unnecessary pings and hence lower the power consumption by avoiding waking up the device.
Quoted 6 lines
Signed-off-by Alexander Usyskin <[email protected]>Signed-off-by Tomas Winkler <[email protected]>---
V2: rework unregistrationdrivers/watchdog/mei_wdt.c | 180+++++++++++++++++++++++++++++++++++++++++----
drivers/watchdog/mei_wdt.c static const char *mei_wdt_state_str(enum
Quoted 73 lines
1 file changed, 167 insertions(+), 13 deletions(-)drivers/watchdog/mei_wdt.c +15
-3@@ -16,6 +16,7 @@#include <linux/slab.h> #include <linux/interrupt.h> #include <linux/watchdog.h> +#include <linux/completion.h> #include <linux/debugfs.h> #include <linux/uuid.h> @@ -38,43 +39,54 @@/* Sub Commands */ #define MEI_MC_START_WD_TIMER_REQ 0x13 +#define MEI_MC_START_WD_TIMER_RES 0x83 +#define MEI_WDT_STATUS_SUCCESS 0 +#define MEI_WDT_WDSTATE_NOT_REQUIRED 0x1#define MEI_MC_STOP_WD_TIMER_REQ 0x14/** * enum mei_wdt_state - internal watchdog state * +* @MEI_WDT_PROBE: wd in probing stage * @MEI_WDT_IDLE: wd is idle and not opened * @MEI_WDT_START: wd was opened, start was called * @MEI_WDT_RUNNING: wd is expecting keep alive pings * @MEI_WDT_STOPPING: wd is stopping and will move to IDLE +* @MEI_WDT_NOT_REQUIRED: wd device is not required */ enum mei_wdt_state { +MEI_WDT_PROBE ,MEI_WDT_IDLE ,MEI_WDT_START ,MEI_WDT_RUNNING ,MEI_WDT_STOPPING , +MEI_WDT_NOT_REQUIRED , }; -#if IS_ENABLED (CONFIG_DEBUG_FS )static const char *mei_wdt_state_str (enum mei_wdt_state state) {switch (state) { +case MEI_WDT_PROBE :return "PROBE" ;case MEI_WDT_IDLE :return "IDLE" ;case MEI_WDT_START :return "START" ;case MEI_WDT_RUNNING :return "RUNNING" ;case MEI_WDT_STOPPING :return "STOPPING" ; +case MEI_WDT_NOT_REQUIRED :return "NOT_REQUIRED" ;default :return "unknown" ; } } -#endif /* CONFIG_DEBUG_FS */ /** * struct mei_wdt - mei watchdog driver * @wdd: watchdog device * @refcnt: watchdog device reference counter +* @reg_lock: watchdog device registration lock +* @is_registered: is watchdog device registered * -* @state: watchdog internal state * @cldev: mei watchdog client device +* @state: watchdog internal state +* @resp_required: ping required response +* @response: ping response * @timeout: watchdog current timeout * * @dbgfs_dir: debugfs dir entry @@ -82,9 +94,13 @@ static const char *mei_wdt_state_str(enum
mei_wdt_statestate)Quoted 36 lines
drivers/watchdog/mei_wdt.c +17
-0struct mei_wdt {struct watchdog_device wdd;struct kref refcnt; +struct mutex reg_lock; +bool is_registered;struct mei_cl_device *cldev;enum mei_wdt_state state; +bool resp_required; +struct completion response;u16 timeout;#if IS_ENABLED (CONFIG_DEBUG_FS )@@ -121,6 +137,19 @@ struct mei_wdt_start_request { } __packed;/** +* struct mei_wdt_start_response watchdog start/ping response +* +* @hdr: Management Control Command Header +* @status: operation status +* @wdstate: watchdog status bit mask +*/ +struct mei_wdt_start_response { +struct mei_mc_hdr hdr; +u8 status; +u8 wdstate; +} __packed; + +/** * struct mei_wdt_stop_request - watchdog stop * * @hdr: Management Control Command Header @@ -192,6 +221,23 @@ static void mei_wdt_release(struct kref *ref) }
Quoted 14 lines
drivers/watchdog/mei_wdt.c static void mei_wdt_release(struct kref *ref) · line 233
Quoted 11 lines
drivers/watchdog/mei_wdt.c +10
-0/** +* mei_wdt_unregister - unregister from the watchdog subsystem +* +* @wdt: mei watchdog device +*/ +static void mei_wdt_unregister (struct mei_wdt *wdt) +{ + +mutex_lock (&wdt->reg_lock); +if (wdt->is_registered) { +clear_bit (WDOG_ACTIVE , &wdt->wdd.status);
WDOG_ACTIVEis commonly handled by the core. Why does it have to be touched here ?
This have to be cleaned otherwise we ops->start is won't be called on new open()
Quoted 10 lines
I am a bit concerned about the entire sequence.
If the following happens
- watchdog character device is opened
- watchdog goes away
- watchdog reappears
- watchdog character device is accessed
Quoted 3 lines
the code may access the driver through an old instance of the watchdog character device. The behavior in this situation is pretty much undefined. It may work, but only by luck, not by design.
You won't get through as the file descriptor is stale, you have to open the device again. This should work by design, you have pluggable devices.
Quoted 15 lines
Quoted 15 lines
drivers/watchdog/mei_wdt.c +7
-0+watchdog_unregister_device (&wdt->wdd); + wdt->is_registered = false; + } +mutex_unlock (&wdt->reg_lock); +} + +/** * mei_wdt_register - register with the watchdog subsystem * * @wdt: mei watchdog device @@ -205,15 +251,20 @@ static int mei_wdt_register(struct mei_wdt *wdt) dev = &wdt->cldev->dev;
Quoted 7 lines
drivers/watchdog/mei_wdt.c static int mei_wdt_register(struct mei_wdt *wdt) · line 257
Quoted 5 lines
drivers/watchdog/mei_wdt.c +3
-1-watchdog_set_drvdata (&wdt->wdd, wdt); + wdt->state =MEI_WDT_IDLE ; +mutex_lock (&wdt->reg_lock); +clear_bit (WDOG_UNREGISTERED , &wdt->wdd.status);Please don't touch watchdog core internal states
No choice here. This is a bit tricky bit. This is not cleared, by the core as fresh watchdog device upon registering I think in the overall design it would be cleaner if this flag would have positive logic. i.e. WDOG_REGISTERED. The code would be simpler if I would be able just toggle this bit, but there is no event propagated to the user space on this change, so I prefer to remove the whole device as this is something already supported..
- this flag will likely
Quoted 1 line
go away soon
We will have to deal with that when this happens.
Tomas
Guenter Roeck
maintainerRe: [char-misc-next v2 6/7] watchdog: mei_wdt: register wd device only if required
Quoted 154 lines · Tomas
Quoted 127 lines · Tomas Winkler
Quoted 8 lines
From: Alexander Usyskin <[email protected]>
For Intel Broadwell and newer platforms, the ME device can inform the host whether the watchdog functionality is activated or not. If the watchdog functionality is not activated then the watchdog interface can be not registered and eliminate unnecessary pings and hence lower the power consumption by avoiding waking up the device.
Quoted 6 lines
Signed-off-by Alexander Usyskin <[email protected]>Signed-off-by Tomas Winkler <[email protected]>---
V2: rework unregistrationdrivers/watchdog/mei_wdt.c | 180+++++++++++++++++++++++++++++++++++++++++----
drivers/watchdog/mei_wdt.c static const char *mei_wdt_state_str(enum
Quoted 73 lines
1 file changed, 167 insertions(+), 13 deletions(-)drivers/watchdog/mei_wdt.c +15
-3@@ -16,6 +16,7 @@#include <linux/slab.h> #include <linux/interrupt.h> #include <linux/watchdog.h> +#include <linux/completion.h> #include <linux/debugfs.h> #include <linux/uuid.h> @@ -38,43 +39,54 @@/* Sub Commands */ #define MEI_MC_START_WD_TIMER_REQ 0x13 +#define MEI_MC_START_WD_TIMER_RES 0x83 +#define MEI_WDT_STATUS_SUCCESS 0 +#define MEI_WDT_WDSTATE_NOT_REQUIRED 0x1#define MEI_MC_STOP_WD_TIMER_REQ 0x14/** * enum mei_wdt_state - internal watchdog state * +* @MEI_WDT_PROBE: wd in probing stage * @MEI_WDT_IDLE: wd is idle and not opened * @MEI_WDT_START: wd was opened, start was called * @MEI_WDT_RUNNING: wd is expecting keep alive pings * @MEI_WDT_STOPPING: wd is stopping and will move to IDLE +* @MEI_WDT_NOT_REQUIRED: wd device is not required */ enum mei_wdt_state { +MEI_WDT_PROBE ,MEI_WDT_IDLE ,MEI_WDT_START ,MEI_WDT_RUNNING ,MEI_WDT_STOPPING , +MEI_WDT_NOT_REQUIRED , }; -#if IS_ENABLED (CONFIG_DEBUG_FS )static const char *mei_wdt_state_str (enum mei_wdt_state state) {switch (state) { +case MEI_WDT_PROBE :return "PROBE" ;case MEI_WDT_IDLE :return "IDLE" ;case MEI_WDT_START :return "START" ;case MEI_WDT_RUNNING :return "RUNNING" ;case MEI_WDT_STOPPING :return "STOPPING" ; +case MEI_WDT_NOT_REQUIRED :return "NOT_REQUIRED" ;default :return "unknown" ; } } -#endif /* CONFIG_DEBUG_FS */ /** * struct mei_wdt - mei watchdog driver * @wdd: watchdog device * @refcnt: watchdog device reference counter +* @reg_lock: watchdog device registration lock +* @is_registered: is watchdog device registered * -* @state: watchdog internal state * @cldev: mei watchdog client device +* @state: watchdog internal state +* @resp_required: ping required response +* @response: ping response * @timeout: watchdog current timeout * * @dbgfs_dir: debugfs dir entry @@ -82,9 +94,13 @@ static const char *mei_wdt_state_str(enum
mei_wdt_statestate)Quoted 36 lines
drivers/watchdog/mei_wdt.c +17
-0struct mei_wdt {struct watchdog_device wdd;struct kref refcnt; +struct mutex reg_lock; +bool is_registered;struct mei_cl_device *cldev;enum mei_wdt_state state; +bool resp_required; +struct completion response;u16 timeout;#if IS_ENABLED (CONFIG_DEBUG_FS )@@ -121,6 +137,19 @@ struct mei_wdt_start_request { } __packed;/** +* struct mei_wdt_start_response watchdog start/ping response +* +* @hdr: Management Control Command Header +* @status: operation status +* @wdstate: watchdog status bit mask +*/ +struct mei_wdt_start_response { +struct mei_mc_hdr hdr; +u8 status; +u8 wdstate; +} __packed; + +/** * struct mei_wdt_stop_request - watchdog stop * * @hdr: Management Control Command Header @@ -192,6 +221,23 @@ static void mei_wdt_release(struct kref *ref) }Quoted 14 lines
drivers/watchdog/mei_wdt.c static void mei_wdt_release(struct kref *ref) · line 233
Quoted 11 lines
drivers/watchdog/mei_wdt.c +10
-0/** +* mei_wdt_unregister - unregister from the watchdog subsystem +* +* @wdt: mei watchdog device +*/ +static void mei_wdt_unregister (struct mei_wdt *wdt) +{ + +mutex_lock (&wdt->reg_lock); +if (wdt->is_registered) { +clear_bit (WDOG_ACTIVE , &wdt->wdd.status);
WDOG_ACTIVEis commonly handled by the core. Why does it have to be touched here ?This have to be cleaned otherwise we
ops->startis won't be called on newopen()Quoted 8 lines
I am a bit concerned about the entire sequence.
If the following happens
- watchdog character device is opened
- watchdog goes away
- watchdog reappears
- watchdog character device is accessed
Quoted 6 lines
Quoted 3 lines
the code may access the driver through an old instance of the watchdog character device. The behavior in this situation is pretty much undefined. It may work, but only by luck, not by design.
You won't get through as the file descriptor is stale, you have to open the device again. This should work by design, you have pluggable devices.
It can definitely happen today without the third step, ie
- watchdog is unregistered
- watchdog character device is accessed
is possible, making the ref/unref operations necessary, and the reason for introducing WDOG_UNREGISTERED in the first place. See commit e907df327252.
I actually tested that condition. It _can_ happen.
Quoted 29 lines
Quoted 16 lines
Quoted 16 lines
drivers/watchdog/mei_wdt.c +7
-0+watchdog_unregister_device (&wdt->wdd); + wdt->is_registered = false; + } +mutex_unlock (&wdt->reg_lock); +} + +/** * mei_wdt_register - register with the watchdog subsystem * * @wdt: mei watchdog device @@ -205,15 +251,20 @@ static int mei_wdt_register(struct mei_wdt *wdt) dev = &wdt->cldev->dev;Quoted 7 lines
drivers/watchdog/mei_wdt.c static int mei_wdt_register(struct mei_wdt *wdt) · line 257
Quoted 5 lines
drivers/watchdog/mei_wdt.c +3
-1-watchdog_set_drvdata (&wdt->wdd, wdt); + wdt->state =MEI_WDT_IDLE ; +mutex_lock (&wdt->reg_lock); +clear_bit (WDOG_UNREGISTERED , &wdt->wdd.status);Please don't touch watchdog core internal states
No choice here. This is a bit tricky bit. This is not cleared, by the core as fresh watchdog device upon registering I think in the overall design it would be cleaner if this flag would have positive logic. i.e.
WDOG_REGISTERED. The code would be simpler if I would be able just toggle this bit, but there is no event propagated to the user space on this change, so I prefer to remove the whole device as this is something already supported..
Quoted 4 lines
- this flag will likely
Quoted 1 line
go away soon
We will have to deal with that when this happens.
Consider the case where the watchdog was removed, but the character device (or misc device) file descriptor is still open. Now you clear this flag. User space accesses the "stale" file descriptor right after your "clear" operation. Since the flag is no longer set, the watchdog core doesn't know anymore that the watchdog device is unregistered, and calls the driver functions, even though the watchdog is still unregistered.
I don't know what happens if the watchdog is re-registered using the same character device data structure, ie after cdev_init() is called on it again. Maybe a file access using the "old" file descriptor is then rejected by the kernel, and it doesn't find its way into the driver. Maybe I am missing it, but I don't immediately see where the kernel would detect that situation. Instead, it seems to me that the kernel holds a reference to the file operations until the device is closed/released. And how would that close operation happen unless the kernel actually calls the fops release function ? After all, the close operation _does_ have to call the watchdog core, since the watchdog core calls module_put() as well as kref_put(), and clears WDOG_DEV_OPEN.
Did you actually test that case ? That would actually be interesting to know for sure.
Another interesting case occurs after you unregistered the watchdog device, but the character device is still open. Your code clears WATCHDOG_ACTIVE, meaning the watchdog core will believe that the watchdog isn't running and won't even check for other flags such as WDOG_ALLOW_RELEASE, or call the stop function, but will just close the watchdog device unconditionally. That may well work with your driver, but again relies on internal functionality and may have unexpected side effects (such as that the WDOG_ALLOW_RELEASE flag won't be cleared).
Yet another interesting case occurs if the character device is still open after the watchdog was unregistererd and re-registered. In that case, WDOG_DEV_OPEN will still be set, so it will be impossible to open the watchdog device until the old instance is closed. How would that close actually happen ? Do you rely on the kernel detecting that the fd is stale when user space tries to use it, and on calling the release function at that time ?
Either case, with the current watchdog core, there is no guarantee that 'struct watchdog_device' is no longer in use by the watchdog core until the unref function has been called. Until then, the watchdog core still "owns" the structure, and you can not re-use it.
I am working on a set of patches to remedy that situation. With those patches, the watchdog core will no longer use struct watchdog_device after the call to watchdog_unregister_device, and you will be able to re-use it immediately. But we are not there yet.
Overall, it may be better to allocate a new instance of struct watchdog_device whenever you re-register the watchdog, to avoid all those problems. Sorry that I didn't realize that earlier.
Guenter
PATCH v2 7/7 watchdog: mei_wdt: re-register device on event
In reply to Tomas WinklerFrom: Alexander Usyskin <[email protected]>
For Intel SKL platform the ME device can inform the host via asynchronous notification that the watchdog feature was activated on the device. The activation doesn't require reboot.
In that case the driver registers the watchdog device with the kernel. In the same manner the feature can be deactivated without reboot in that case the watchdog device should be unregistered.
---
V2: rework un/registration in runtime
drivers/watchdog/mei_wdt.c | 45 ++++++++++++++++++++++++++++++++++++++++++--- 1 file changed, 42 insertions(+), 3 deletions(-)
drivers/watchdog/mei_wdt.c +42 -3
@@ -87,6 +87,7 @@ static const char *mei_wdt_state_str(enum mei_wdt_state state)* @state: watchdog internal state * @resp_required: ping required response * @response: ping response +* @unregister: unregister worker * @timeout: watchdog current timeout * * @dbgfs_dir: debugfs dir entry
@@ -101,6 +102,7 @@ struct mei_wdt {enum mei_wdt_state state;bool resp_required;struct completion response; +struct work_struct unregister;u16 timeout;#if IS_ENABLED (CONFIG_DEBUG_FS )
@@ -237,6 +239,13 @@ static void mei_wdt_unregister(struct mei_wdt *wdt)mutex_unlock (&wdt->reg_lock); } +static void mei_wdt_unregister_work (struct work_struct *work) +{ + struct mei_wdt *wdt = container_of(work, struct mei_wdt, unregister); + +mei_wdt_unregister (wdt); +} +/** * mei_wdt_register - register with the watchdog subsystem *
@@ -350,8 +359,13 @@ static void mei_wdt_event_rx(struct mei_cl_device *cldev)return ; } -if (wdt->state ==MEI_WDT_RUNNING ) +if (wdt->state ==MEI_WDT_RUNNING ) { +if (res.wdstate &MEI_WDT_WDSTATE_NOT_REQUIRED ) { + wdt->state =MEI_WDT_NOT_REQUIRED ; +schedule_work (&wdt->unregister); + } goto out; + }if (wdt->state ==MEI_WDT_PROBE ) {if (res.wdstate &MEI_WDT_WDSTATE_NOT_REQUIRED ) {
@@ -372,6 +386,21 @@ out:complete (&wdt->response); } +/* +* mei_wdt_notify_event - callback for event notification +* +* @cldev: bus device +*/ +static void mei_wdt_notify_event (struct mei_cl_device *cldev) +{ +struct mei_wdt *wdt =mei_cldev_get_drvdata (cldev); + +if (wdt->state !=MEI_WDT_NOT_REQUIRED ) +return ; + wdt->state =MEI_WDT_IDLE ; +mei_wdt_register (wdt); +} +/** * mei_wdt_event - callback for event receive *
@@ -384,6 +413,9 @@ static void mei_wdt_event(struct mei_cl_device *cldev, {if (events &BIT (MEI_CL_EVENT_RX ))mei_wdt_event_rx (cldev); + +if (events &BIT (MEI_CL_EVENT_NOTIF )) +mei_wdt_notify_event (cldev); }/**
@@ -543,6 +575,7 @@ static int mei_wdt_probe(struct mei_cl_device *cldev, wdt->resp_required =mei_cldev_ver (cldev) >0x1 ;mutex_init (&wdt->reg_lock); wdt->is_registered = false; +INIT_WORK (&wdt->unregister, mei_wdt_unregister_work);init_completion (&wdt->response);mei_cldev_set_drvdata (cldev, wdt);
@@ -553,9 +586,13 @@ static int mei_wdt_probe(struct mei_cl_device *cldev, goto err_out; } - ret =mei_cldev_register_event_cb (wdt->cldev,BIT (MEI_CL_EVENT_RX ), + ret =mei_cldev_register_event_cb (wdt->cldev, +BIT (MEI_CL_EVENT_RX ) | +BIT (MEI_CL_EVENT_NOTIF ), mei_wdt_event,NULL ); -if (ret) { + +/* on legacy devices notification is not supported */ +if (ret && ret != -EOPNOTSUPP ) {dev_err (&cldev->dev,"Could not register event ret=%d\n" , ret); goto err_disable; }
@@ -599,6 +636,8 @@ static int mei_wdt_remove(struct mei_cl_device *cldev)if (!completion_done (&wdt->response))complete (&wdt->response); +cancel_work_sync (&wdt->unregister); +mei_wdt_unregister (wdt);kref_put (&wdt->refcnt, mei_wdt_release);
Signature · 1 line
2.4.3
Guenter Roeck
maintainerRe: [char-misc-next v2 7/7] watchdog: mei_wdt: re-register device on event
Quoted 9 lines · Tomas Winkler
From: Alexander Usyskin <[email protected]>
For Intel SKL platform the ME device can inform the host via asynchronous notification that the watchdog feature was activated on the device. The activation doesn't require reboot.
In that case the driver registers the watchdog device with the kernel. In the same manner the feature can be deactivated without reboot in that case the watchdog device should be unregistered.
Quoted 3 lines
Signed-off-by Alexander Usyskin <[email protected]>Signed-off-by Tomas Winkler <[email protected]>---
I am not really happy about the watchdog device appearing and disappearing dynamically. This wreaks havoc with any standard watchdog application.
Isn't there a better way to handle this ? How about just registering the watchdog device and return an error in the access functions if it is disabled ?
Thanks,
Guenter
Quoted 120 lines
V2: rework un/registration in runtime
drivers/watchdog/mei_wdt.c | 45 ++++++++++++++++++++++++++++++++++++++++++--- 1 file changed, 42 insertions(+), 3 deletions(-)drivers/watchdog/mei_wdt.c +42
-3@@ -87,6 +87,7 @@ static const char *mei_wdt_state_str(enum mei_wdt_state state)* @state: watchdog internal state * @resp_required: ping required response * @response: ping response +* @unregister: unregister worker * @timeout: watchdog current timeout * * @dbgfs_dir: debugfs dir entry @@ -101,6 +102,7 @@ struct mei_wdt {enum mei_wdt_state state;bool resp_required;struct completion response; +struct work_struct unregister;u16 timeout;#if IS_ENABLED (CONFIG_DEBUG_FS )@@ -237,6 +239,13 @@ static void mei_wdt_unregister(struct mei_wdt *wdt)mutex_unlock (&wdt->reg_lock); } +static void mei_wdt_unregister_work (struct work_struct *work) +{ + struct mei_wdt *wdt = container_of(work, struct mei_wdt, unregister); + +mei_wdt_unregister (wdt); +} +/** * mei_wdt_register - register with the watchdog subsystem * @@ -350,8 +359,13 @@ static void mei_wdt_event_rx(struct mei_cl_device *cldev)return ; } -if (wdt->state ==MEI_WDT_RUNNING ) +if (wdt->state ==MEI_WDT_RUNNING ) { +if (res.wdstate &MEI_WDT_WDSTATE_NOT_REQUIRED ) { + wdt->state =MEI_WDT_NOT_REQUIRED ; +schedule_work (&wdt->unregister); + } goto out; + }if (wdt->state ==MEI_WDT_PROBE ) {if (res.wdstate &MEI_WDT_WDSTATE_NOT_REQUIRED ) {@@ -372,6 +386,21 @@ out:complete (&wdt->response); } +/* +* mei_wdt_notify_event - callback for event notification +* +* @cldev: bus device +*/ +static void mei_wdt_notify_event (struct mei_cl_device *cldev) +{ +struct mei_wdt *wdt =mei_cldev_get_drvdata (cldev); + +if (wdt->state !=MEI_WDT_NOT_REQUIRED ) +return ; + wdt->state =MEI_WDT_IDLE ; +mei_wdt_register (wdt); +} +/** * mei_wdt_event - callback for event receive * @@ -384,6 +413,9 @@ static void mei_wdt_event(struct mei_cl_device *cldev, {if (events &BIT (MEI_CL_EVENT_RX ))mei_wdt_event_rx (cldev); + +if (events &BIT (MEI_CL_EVENT_NOTIF )) +mei_wdt_notify_event (cldev); }/** @@ -543,6 +575,7 @@ static int mei_wdt_probe(struct mei_cl_device *cldev, wdt->resp_required =mei_cldev_ver (cldev) >0x1 ;mutex_init (&wdt->reg_lock); wdt->is_registered = false; +INIT_WORK (&wdt->unregister, mei_wdt_unregister_work);init_completion (&wdt->response);mei_cldev_set_drvdata (cldev, wdt);@@ -553,9 +586,13 @@ static int mei_wdt_probe(struct mei_cl_device *cldev, goto err_out; } - ret =mei_cldev_register_event_cb (wdt->cldev,BIT (MEI_CL_EVENT_RX ), + ret =mei_cldev_register_event_cb (wdt->cldev, +BIT (MEI_CL_EVENT_RX ) | +BIT (MEI_CL_EVENT_NOTIF ), mei_wdt_event,NULL ); -if (ret) { + +/* on legacy devices notification is not supported */ +if (ret && ret != -EOPNOTSUPP ) {dev_err (&cldev->dev,"Could not register event ret=%d\n" , ret); goto err_disable; }@@ -599,6 +636,8 @@ static int mei_wdt_remove(struct mei_cl_device *cldev)if (!completion_done (&wdt->response))complete (&wdt->response); +cancel_work_sync (&wdt->unregister); +mei_wdt_unregister (wdt);kref_put (&wdt->refcnt, mei_wdt_release);
Re: [char-misc-next v2 7/7] watchdog: mei_wdt: re-register device on event
Quoted 2 lines
I am not really happy about the watchdog device appearing and disappearing dynamically. This wreaks havoc with any standard watchdog application.
Any software that doesn't handle this has been broken for over fifteen years. We have hotplug PCI and we have PCI watchdog card support. This isn't a new behaviour to anyone outside the embedded single board space.
Quoted 2 lines
Isn't there a better way to handle this ? How about just registering the watchdog device and return an error in the access functions if it is disabled ?
That breaks the existing behaviour of hot pluggable watchdog interfaces and is different to just about any other device in the kernel. Today with any desktop or server distribution you can already trivially arrange for watchdog daemons to start at the point a watchdog is detected dynamically.
Alan
Guenter Roeck
maintainerRe: [char-misc-next v2 7/7] watchdog: mei_wdt: re-register device on event
Quoted 7 lines · One Thousand Gnomes
Quoted 2 lines
I am not really happy about the watchdog device appearing and disappearing dynamically. This wreaks havoc with any standard watchdog application.
Any software that doesn't handle this has been broken for over fifteen years. We have hotplug PCI and we have PCI watchdog card support. This isn't a new behaviour to anyone outside the embedded single board space.
Quoted 8 lines
Quoted 2 lines
Isn't there a better way to handle this ? How about just registering the watchdog device and return an error in the access functions if it is disabled ?
That breaks the existing behaviour of hot pluggable watchdog interfaces and is different to just about any other device in the kernel. Today with any desktop or server distribution you can already trivially arrange for watchdog daemons to start at the point a watchdog is detected dynamically.
Ok, you have a point. Wonder if any distributions are doing that, though. Any idea ?
Guenter
Re: [char-misc-next v2 7/7] watchdog: mei_wdt: re-register device on event
Quoted 8 lines
Quoted 5 lines
That breaks the existing behaviour of hot pluggable watchdog interfaces and is different to just about any other device in the kernel. Today with any desktop or server distribution you can already trivially arrange for watchdog daemons to start at the point a watchdog is detected dynamically.
Ok, you have a point. Wonder if any distributions are doing that, though. Any idea ?
I don't know for any of the out of the box mainstream ones.
There is a second problem if you register the watchdog but don't actually have it enabled. Consider the case where a user has the mei watchdog disabled and a more advanced watchdog card plugged in - in that case the naïvely implemented existing watchdog app may bind to the non-functional watchdog, not the correct one.
Alan
RE: [char-misc-next v2 7/7] watchdog: mei_wdt: re-register device on event
Quoted 10 lines
Quoted 9 lines
Quoted 6 lines
That breaks the existing behaviour of hot pluggable watchdog interfaces and is different to just about any other device in the kernel. Today with any desktop or server distribution you can already trivially arrange for watchdog daemons to start at the point a watchdog is detected dynamically.
Ok, you have a point. Wonder if any distributions are doing that, though. Any idea ?
Quoted 7 lines
I don't know for any of the out of the box mainstream ones.
There is a second problem if you register the watchdog but don't actually have it enabled. Consider the case where a user has the mei watchdog disabled and a more advanced watchdog card plugged in - in that case the naïvely implemented existing watchdog app may bind to the non-functional watchdog, not the correct one.
That's correct, though WDIOF_ALARMONLY or device identity has to be checked.
And if mei is disabled one would probably hit ENODEV, still there will be no uevent to the user space when the device is provisioned again, so still I prefer add/removing the watchdog device.
Thanks
Tomas