Message-Id: <1310588766-16638-2-git-send-email-w.sang@pengutronix.de>
Date:	Wed, 13 Jul 2011 22:26:01 +0200
From:	Wolfram Sang <w.sang@...gutronix.de>
To:	linux-watchdog@...r.kernel.org
Cc:	linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org,
	wim@...ana.be, tim.bird@...sony.com,
	Wolfram Sang <w.sang@...gutronix.de>
Subject: [RFC 1/6] Documentation: watchdog: add guide how to convert drivers to new framework

Signed-off-by: Wolfram Sang <w.sang@...gutronix.de>
---
 .../watchdog/convert_drivers_to_kernel_api.txt     |  195 ++++++++++++++++++++
 1 files changed, 195 insertions(+), 0 deletions(-)
 create mode 100644 Documentation/watchdog/convert_drivers_to_kernel_api.txt

diff --git a/Documentation/watchdog/convert_drivers_to_kernel_api.txt b/Documentation/watchdog/convert_drivers_to_kernel_api.txt
new file mode 100644
index 0000000..9c0cff9
--- /dev/null
+++ b/Documentation/watchdog/convert_drivers_to_kernel_api.txt
@@ -0,0 +1,195 @@
+Converting old watchdog drivers to the watchdog framework
+by Wolfram Sang <w.sang@...gutronix.de>
+=========================================================
+
+Before the watchdog framework came into the kernel, every driver had to
+implement the API on its own. Now, as the framework factored out the common
+components, those drivers can be lightened making it a user of the framework.
+This document shall guide you for this task. The necessary steps are described
+as well as things to look out for.
+
+
+Remove the file_operations struct
+---------------------------------
+
+Old drivers define their own file_operations for actions like open(), write(),
+etc... These are now handled by the framework and just call the driver when
+needed. So, in general, the file operations-struct and assorted functions can
+go. Only very few driver-specific details have to be moved to other functions.
+Here is a overview of the functions and probably needed actions:
+
+- open: Everything dealing with resource management (file-open checks, magic
+  close preparations) can simply go. Device specific stuff needs to go to the
+  driver specific start-function. Note that for some drivers, the start-function
+  also serves as the ping-function. If that is the case and you need start/stop
+  to be balanced (clocks!), you are better off refactoring a separate start-function.
+
+- close: Same hints as for open apply.
+
+- write: Can simply go, all defined behaviour is taken care of by the framework,
+  i.e. ping on write and magic char ('V') handling.
+
+- ioctl: While the driver is allowed to have extensions to the IOCTL interface,
+  the most common ones are handled by the framework, supported by some assistance
+  from the driver:
+
+	WDIOC_GETSUPPORT:
+		Returns the mandatory watchdog_info struct from the driver
+
+	WDIOC_GETSTATUS:
+		Needs the status-callback defined, otherwise returns 0
+
+	WDIOC_GETBOOTSTATUS:
+		Needs the bootstatus member properly set. Make sure it is 0 if you
+		don't have further support!
+
+	WDIOC_SETOPTIONS:
+		No preparations needed
+
+	WDIOC_KEEPALIVE:
+		If wanted, options in watchdog_info need to have WDIOF_KEEPALIVEPING
+		set
+
+	WDIOC_SETTIMEOUT:
+		Options in watchdog_info need to have WDIOF_SETTIMEOUT set
+		and a set_timeout-callback has to be defined. The core will also
+		do limit-checking, if min_timeout and max_timeout in the watchdog
+		device are set. All is optional.
+
+	WDIOC_GETTIMEOUT:
+		No preparations needed
+
+  Other IOCTLs can be served using the ioctl-callback. Note that this is mainly
+  intended for porting old drivers, new drivers should not invent private IOCTLs.
+  Private IOCTLs are processed first. When the callback returns with
+  -ENOIOCTLCMD, the IOCTLs of the framework will be tried, too. Any other error
+  is directly given to the user.
+
+Example conversion:
+
+-static const struct file_operations s3c2410wdt_fops = {
+-       .owner          = THIS_MODULE,
+-       .llseek         = no_llseek,
+-       .write          = s3c2410wdt_write,
+-       .unlocked_ioctl = s3c2410wdt_ioctl,
+-       .open           = s3c2410wdt_open,
+-       .release        = s3c2410wdt_release,
+-};
+
+Check the functions for device-specific stuff and keep it for later
+refactoring. The rest can go.
+
+
+Remove the miscdevice
+---------------------
+
+Since the file_operations are gone now, you can also remove the 'struct
+miscdevice'. The framework will create it on watchdog_dev_register() called by
+watchdog_register_device().
+
+-static struct miscdevice s3c2410wdt_miscdev = {
+-       .minor          = WATCHDOG_MINOR,
+-       .name           = "watchdog",
+-       .fops           = &s3c2410wdt_fops,
+-};
+
+
+Remove obsolete includes and defines
+------------------------------------
+
+Because of the simplifications, a few defines are probably unused now. Remove
+them. Includes can be removed, too. For example:
+
+- #include <linux/fs.h>
+- #include <linux/miscdevice.h> (if MODULE_ALIAS_MISCDEV is not used)
+- #include <linux/uaccess.h> (if no custom IOCTLs are used)
+
+
+Add the watchdog operations
+---------------------------
+
+All possible callbacks are defined in 'struct watchdog_ops'. You can find it
+explained in 'watchdog-kernel-api.txt' in this directory. start(), stop() and
+owner must be set, the rest is optional. You will easily find corresponding
+functions in the old driver. Note that you will now get a pointer to the
+watchdog_device as a parameter to these functions, so you probably have to
+change the function header. Other changes are most likely not needed, because
+here simply happens the direct hardware access. If you have device-specific
+code left from the above steps, it should be refactored into these callbacks.
+
+Here is a simple example:
+
++static struct watchdog_ops s3c2410wdt_ops = {
++       .owner = THIS_MODULE,
++       .start = s3c2410wdt_start,
++       .stop = s3c2410wdt_stop,
++       .ping = s3c2410wdt_keepalive,
++       .set_timeout = s3c2410wdt_set_heartbeat,
++};
+
+A typical function-header change looks like:
+
+-static void s3c2410wdt_keepalive(void)
++static int s3c2410wdt_keepalive(struct watchdog_device *wdd)
+ {
+...
++
++       return 0;
+ }
+
+...
+
+-       s3c2410wdt_keepalive();
++       s3c2410wdt_keepalive(&s3c2410_wdd);
+
+
+Add the watchdog device
+-----------------------
+
+Now we need to create a 'struct watchdog_device' and populate it with the
+necessary information for the framework. The struct is also explained in detail
+in 'watchdog-kernel-api.txt' in this directory. We pass it the mandatory
+watchdog_info struct and the newly created watchdog_ops. Often, old drivers
+have their own record-keeping for things like bootstatus and timeout using
+static variables. Those have to be converted to use the members in
+watchdog_device. Note that the timeout values are unsigned int. Some drivers
+use signed int, so this has to be converted, too.
+
+Here is a simple example for a watchdog device:
+
++static struct watchdog_device s3c2410_wdd = {
++       .info = &s3c2410_wdt_ident,
++       .ops = &s3c2410wdt_ops,
++};
+
+
+Register the watchdog device
+----------------------------
+
+Replace misc_register(&miscdev) with watchdog_register_device(&watchdog_dev).
+Make sure the return value gets checked and the error message, if present,
+still fits. Also convert the unregister case.
+
+-       ret = misc_register(&s3c2410wdt_miscdev);
++       ret = watchdog_register_device(&s3c2410_wdd);
+
+...
+
+-       misc_deregister(&s3c2410wdt_miscdev);
++       watchdog_unregister_device(&s3c2410_wdd);
+
+
+Update the Kconfig-entry
+------------------------
+
+The entry for the driver now needs to select WATCHDOG_CORE:
+
++       select WATCHDOG_CORE
+
+
+Create a patch and send it to upstream
+--------------------------------------
+
+Make sure you understood Documentation/SubmittingPatches and send your patch to
+linux-watchdog@...r.kernel.org. We are looking forward to it :)
+
-- 
1.7.2.5

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists