Message ID | 1464183151-4912-2-git-send-email-wsa@the-dreams.de (mailing list archive) |
---|---|
State | Accepted |
Delegated to: | Geert Uytterhoeven |
Headers | show |
Hi Wolfram, On 25.05.2016 16:32, Wolfram Sang wrote: > From: Wolfram Sang <wsa+renesas@sang-engineering.com> > > The change adds a simple watchdog pretimeout framework infrastructure, > its purpose is to allow users to select a desired handling of watchdog > pretimeout events, which may be generated by a watchdog driver. > > By design every watchdog pretimeout governor may be compiled as a > kernel module, a user selects a default watchdog pretimeout > governor during compilation stage and can select another governor in > runtime. > > Watchdogs with WDIOF_PRETIMEOUT capability now have two device > attributes in sysfs: read/write pretimeout_governor attribute and read > only pretimeout_available_governors attribute. > > Watchdogs with no WDIOF_PRETIMEOUT capability has no changes in > sysfs. > > Signed-off-by: Vladimir Zapolskiy <vladimir_zapolskiy@mentor.com> > Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com> > --- > > Changes since Vladimir's last version: > > * rebased > adapt to the internal data reorganization, especially the now private > struct device *dev > * dropped can_sleep support! > The additional lock, list, and workqueue made the code quite complex. The > only user was the userspace governor which can be reworked to let the > watchdog device code do the bottom half. In addition, I am not fully > convinced sending a uevent is the proper thing to do, but this needs to > be discussed in another thread. Removing this support makes the code much > easier to follow (locking!), saves 30% of LoC + a list + a workqueue. The thing is that I'm particularly interested in 1) sleeping governors, 2) userspace notification of any appropriate kind, but preferably not by adding a clumsy .poll callback, uevent is the best IMHO. The userspace sleeping governor is the only one proposed for a mainline, however the whole idea of having a framework is to allow users to write their own private governors, and that's exactly what we need and use. So the original complexity has its state-of-the-art grounds, and for sake of getting a solid picture for reviewers and users it is better to introduce sleeping functionality right from the beginning. I know it is quite complex, probably it might be better to add it to the series as a separate patch? > * moved pretimeout registration from watchdog_core to watchdog_dev > Let's handle it exactly where the device is created, so we have access to > the now private device pointer for adding the sysfs files. > * don't export watchdog_(un)register_pretimeout since they are linked to the > core anyhow > * whitespace cleanups > Thanks for pushing it, but do you think that the authorship of the code can be preserved? Feel free to ask me to rebase the change and so on, patch review procedure is well established and I'm pretty sure I can cope with it. I believe the main problem with the original code since the time when rebase was not required is that it didn't receive any formal technical review from Guenter or Wim, but I'm glad to know that someone else is interested in it. Merge window is closing, so it's good time for me to rebase the change and resend it, do you have any objections? > drivers/watchdog/Kconfig | 8 + > drivers/watchdog/Makefile | 6 +- > drivers/watchdog/watchdog_dev.c | 8 + > drivers/watchdog/watchdog_pretimeout.c | 269 +++++++++++++++++++++++++++++++++ > drivers/watchdog/watchdog_pretimeout.h | 35 +++++ > include/linux/watchdog.h | 10 ++ > 6 files changed, 334 insertions(+), 2 deletions(-) > create mode 100644 drivers/watchdog/watchdog_pretimeout.c > create mode 100644 drivers/watchdog/watchdog_pretimeout.h > -- With best wishes, Vladimir
Hi Vladimir, great to see you still have capacity for this series :) > The thing is that I'm particularly interested in > > 1) sleeping governors, > 2) userspace notification of any appropriate kind, but preferably not by > adding a clumsy .poll callback, uevent is the best IMHO. I am totally open that poll might not be a good idea, but why do you think uevent is best? (Disclaimer: I don't do much userspace code) > The userspace sleeping governor is the only one proposed for a mainline, > however the whole idea of having a framework is to allow users to write > their own private governors, and that's exactly what we need and use. One reason I decided to drop 'can_sleep' is that I guessed 98% of users will be happy with the panic, noop, and userspace governers. 2% might need custom governors from which maybe not even all need to sleep. Chances are high IMO that these govenors will be out-of-tree code, so having all this additional complexity for some out-of-tree govenors was questionable to me. I wondered if it would make sense to let those govenors do the bottom half handling themselves. There was also a technical reason: The dev pointer was first moved to watchdog_device private data before it was ultimately removed. So, while trying to fix this, the code got more and more complicated which led me to the decision to go the other way around: make the code simpler so it will be easier maintainable in the future. > So the original complexity has its state-of-the-art grounds, and for > sake of getting a solid picture for reviewers and users it is better to > introduce sleeping functionality right from the beginning. I still wonder if bottom half handling shouldn't be put to the governors which need that. > I know it is quite complex, probably it might be better to add it to > the series as a separate patch? That might help the initial review. > Thanks for pushing it, but do you think that the authorship of the > code can be preserved? I changed the authorship because I did one fundamental change to your original design. Not knowing if you'd approve of that, I didn't want to put your sticker on something you might not even like. > Feel free to ask me to rebase the change and so on, patch review procedure > is well established and I'm pretty sure I can cope with it. No doubt about that. I had some ideas and thought it is easier to talk over code. If you want to rebase it, too, I'd be happy to check what you came up with to solve the problems. I might still argue that I prefer the less-code approach, but it will be Guenter's / Wim's decision, of course. And I apologize for not contacting you beforehand which would have been friendly. I got a rush on hacking it and wanted to show what I came up with. No offence, sorry! Thanks, Wolfram
On Thu, May 26, 2016 at 06:41:36PM +0200, Wolfram Sang wrote: [ ... ] > > No doubt about that. I had some ideas and thought it is easier to talk > over code. If you want to rebase it, too, I'd be happy to check what you > came up with to solve the problems. I might still argue that I prefer > the less-code approach, but it will be Guenter's / Wim's decision, of > course. > I have a large back-log of patches to review. Simple patches with less code will get preferential treating. The more complex, the higher the likelyhood that the patches get pushed to the end of the queue. Giving a quick glance, I liked Wolfram's patches because they seemed to be simple and straightforward. I hope the next version won't add too much additional complexity. Guenter
On Thu, May 26, 2016 at 03:37:17PM -0700, Guenter Roeck wrote: > On Thu, May 26, 2016 at 06:41:36PM +0200, Wolfram Sang wrote: > [ ... ] > > > > No doubt about that. I had some ideas and thought it is easier to talk > > over code. If you want to rebase it, too, I'd be happy to check what you > > came up with to solve the problems. I might still argue that I prefer > > the less-code approach, but it will be Guenter's / Wim's decision, of > > course. > > > I have a large back-log of patches to review. Simple patches with less code > will get preferential treating. The more complex, the higher the likelyhood > that the patches get pushed to the end of the queue. > > Giving a quick glance, I liked Wolfram's patches because they seemed to > be simple and straightforward. I hope the next version won't add too much > additional complexity. Vladimir, did you have time to look into this? I can offer to resend this series on-top of v4.7-rc1 with the core patch re-attributed to Vladimir. Then we can review the basic stuff now, and can discuss/deal with the bottom half handling incrementally. Sounds good?
Hi Wolfram, On 05.06.2016 12:48, Wolfram Sang wrote: > On Thu, May 26, 2016 at 03:37:17PM -0700, Guenter Roeck wrote: >> On Thu, May 26, 2016 at 06:41:36PM +0200, Wolfram Sang wrote: >> [ ... ] >>> >>> No doubt about that. I had some ideas and thought it is easier to talk >>> over code. If you want to rebase it, too, I'd be happy to check what you >>> came up with to solve the problems. I might still argue that I prefer >>> the less-code approach, but it will be Guenter's / Wim's decision, of >>> course. >>> >> I have a large back-log of patches to review. Simple patches with less code >> will get preferential treating. The more complex, the higher the likelyhood >> that the patches get pushed to the end of the queue. >> >> Giving a quick glance, I liked Wolfram's patches because they seemed to >> be simple and straightforward. I hope the next version won't add too much >> additional complexity. > > Vladimir, did you have time to look into this? yes, I managed to rebase and split my change to smaller pieces to encourage reviewers. My plan is to do thorough testing of the change and send it for review tomorrow or on Tuesday, please participate in review. > I can offer to resend this series on-top of v4.7-rc1 with the core patch > re-attributed to Vladimir. Then we can review the basic stuff now, and > can discuss/deal with the bottom half handling incrementally. > > Sounds good? > -- Best wishes, Vladimir
diff --git a/drivers/watchdog/Kconfig b/drivers/watchdog/Kconfig index 3902c9ca7f099d..909d1021de5cbc 100644 --- a/drivers/watchdog/Kconfig +++ b/drivers/watchdog/Kconfig @@ -1774,4 +1774,12 @@ config USBPCWATCHDOG Most people will say N. +comment "Watchdog Pretimeout Governors" + +config WATCHDOG_PRETIMEOUT_GOV + bool "Enable watchdog pretimeout governors" + default n + help + The option allows to select watchdog pretimeout governors. + endif # WATCHDOG diff --git a/drivers/watchdog/Makefile b/drivers/watchdog/Makefile index 2cbc9709852d0e..820860cf3e8d62 100644 --- a/drivers/watchdog/Makefile +++ b/drivers/watchdog/Makefile @@ -3,8 +3,10 @@ # # The WatchDog Timer Driver Core. -watchdog-objs += watchdog_core.o watchdog_dev.o -obj-$(CONFIG_WATCHDOG_CORE) += watchdog.o +obj-$(CONFIG_WATCHDOG_CORE) += watchdog.o + +watchdog-y += watchdog_core.o watchdog_dev.o +watchdog-$(CONFIG_WATCHDOG_PRETIMEOUT_GOV) += watchdog_pretimeout.o # Only one watchdog can succeed. We probe the ISA/PCI/USB based # watchdog-cards first, then the architecture specific watchdog diff --git a/drivers/watchdog/watchdog_dev.c b/drivers/watchdog/watchdog_dev.c index 3595cffa24ea49..5d028f94a90743 100644 --- a/drivers/watchdog/watchdog_dev.c +++ b/drivers/watchdog/watchdog_dev.c @@ -49,6 +49,7 @@ #include <linux/uaccess.h> /* For copy_to_user/put_user/... */ #include "watchdog_core.h" +#include "watchdog_pretimeout.h" /* * struct watchdog_core_data - watchdog core internal data @@ -911,6 +912,12 @@ int watchdog_dev_register(struct watchdog_device *wdd) return PTR_ERR(dev); } + ret = watchdog_register_pretimeout(wdd, dev); + if (ret) { + device_destroy(&watchdog_class, devno); + watchdog_cdev_unregister(wdd); + } + return ret; } @@ -924,6 +931,7 @@ int watchdog_dev_register(struct watchdog_device *wdd) void watchdog_dev_unregister(struct watchdog_device *wdd) { + watchdog_unregister_pretimeout(wdd); device_destroy(&watchdog_class, wdd->wd_data->cdev.dev); watchdog_cdev_unregister(wdd); } diff --git a/drivers/watchdog/watchdog_pretimeout.c b/drivers/watchdog/watchdog_pretimeout.c new file mode 100644 index 00000000000000..87a10ebeaacc7e --- /dev/null +++ b/drivers/watchdog/watchdog_pretimeout.c @@ -0,0 +1,269 @@ +/* + * Watchdog pretimout governor framework + * + * Copyright (C) 2015 Mentor Graphics + * Copyright (C) 2016 Renesas Electronics Corporation + * Copyright (C) 2016 Sang Engineering, Wolfram Sang + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + */ + +#include <linux/module.h> +#include <linux/slab.h> +#include <linux/watchdog.h> + +#include "watchdog_pretimeout.h" + +/* The mutex protects governor list and serializes external interfaces */ +static DEFINE_MUTEX(governor_lock); + +/* List of registered watchdog pretimeout governors */ +static LIST_HEAD(governor_list); + +/* The spinlock protects wdd->gov */ +static DEFINE_SPINLOCK(pretimeout_lock); + +struct governor_priv { + struct watchdog_governor *gov; + bool is_default; + struct list_head entry; +}; + +static struct governor_priv *find_governor_by_name(const char *gov_name) +{ + struct governor_priv *priv; + + list_for_each_entry(priv, &governor_list, entry) + if (!strncmp(gov_name, priv->gov->name, WATCHDOG_GOV_NAME_LEN)) + return priv; + + return NULL; +} + +static struct watchdog_governor *find_default_governor(void) +{ + struct governor_priv *priv; + + list_for_each_entry(priv, &governor_list, entry) + if (priv->is_default) + return priv->gov; + + return NULL; +} + +void watchdog_notify_pretimeout(struct watchdog_device *wdd) +{ + unsigned long flags; + + if (!wdd) + return; + + spin_lock_irqsave(&pretimeout_lock, flags); + + if (wdd->gov) + wdd->gov->pretimeout(wdd); + + spin_unlock_irqrestore(&pretimeout_lock, flags); +} +EXPORT_SYMBOL_GPL(watchdog_notify_pretimeout); + +int watchdog_register_governor(struct watchdog_governor *gov) +{ + struct governor_priv *priv; + + if (!gov || !gov->pretimeout || strlen(gov->name) >= WATCHDOG_GOV_NAME_LEN) + return -EINVAL; + + priv = kzalloc(sizeof(*priv), GFP_KERNEL); + if (!priv) + return -ENOMEM; + + mutex_lock(&governor_lock); + + if (find_governor_by_name(gov->name)) { + kfree(priv); + mutex_unlock(&governor_lock); + return -EBUSY; + } + + priv->gov = gov; + + if (!strncmp(WATCHDOG_PRETIMEOUT_DEFAULT_GOV, + gov->name, WATCHDOG_GOV_NAME_LEN)) { + priv->is_default = true; + } + + list_add(&priv->entry, &governor_list); + + mutex_unlock(&governor_lock); + + return 0; +} +EXPORT_SYMBOL_GPL(watchdog_register_governor); + +void watchdog_unregister_governor(struct watchdog_governor *gov) +{ + struct governor_priv *priv; + + if (!gov) + return; + + mutex_lock(&governor_lock); + + list_for_each_entry(priv, &governor_list, entry) { + if (priv->gov == gov) { + list_del(&priv->entry); + kfree(priv); + break; + } + } + + mutex_unlock(&governor_lock); +} +EXPORT_SYMBOL_GPL(watchdog_unregister_governor); + +static ssize_t pretimeout_governor_store(struct device *dev, + struct device_attribute *attr, + const char *buf, size_t count) +{ + struct watchdog_device *wdd = dev_get_drvdata(dev); + struct governor_priv *priv; + + mutex_lock(&governor_lock); + + priv = find_governor_by_name(buf); + if (!priv) { + mutex_unlock(&governor_lock); + return -EINVAL; + } + + if (!try_module_get(priv->gov->owner)) { + mutex_unlock(&governor_lock); + return -ENODEV; + } + + spin_lock_irq(&pretimeout_lock); + if (wdd->gov) + module_put(wdd->gov->owner); + wdd->gov = priv->gov; + spin_unlock_irq(&pretimeout_lock); + + mutex_unlock(&governor_lock); + + return count; +} + +static ssize_t pretimeout_governor_show(struct device *dev, + struct device_attribute *devattr, char *buf) +{ + struct watchdog_device *wdd = dev_get_drvdata(dev); + char gov_name[WATCHDOG_GOV_NAME_LEN] = { 0 }; + int count; + + mutex_lock(&governor_lock); + + spin_lock_irq(&pretimeout_lock); + if (wdd->gov) + strncpy(gov_name, wdd->gov->name, WATCHDOG_GOV_NAME_LEN); + spin_unlock_irq(&pretimeout_lock); + + count = sprintf(buf, "%s\n", gov_name); + + mutex_unlock(&governor_lock); + + return count; +} +static DEVICE_ATTR_RW(pretimeout_governor); + +static ssize_t pretimeout_available_governors_show(struct device *dev, + struct device_attribute *devattr, char *buf) +{ + struct governor_priv *priv; + int count = 0; + + mutex_lock(&governor_lock); + + list_for_each_entry(priv, &governor_list, entry) + count += sprintf(buf + count, "%s\n", priv->gov->name); + + mutex_unlock(&governor_lock); + + return count; +} +static DEVICE_ATTR_RO(pretimeout_available_governors); + +static struct attribute *wdd_pretimeout_attrs[] = { + &dev_attr_pretimeout_governor.attr, + &dev_attr_pretimeout_available_governors.attr, + NULL, +}; + +static umode_t wdd_pretimeout_attr_is_visible(struct kobject *kobj, + struct attribute *attr, int n) +{ + struct device *dev = container_of(kobj, struct device, kobj); + struct watchdog_device *wdd = dev_get_drvdata(dev); + + if (wdd->info->options & WDIOF_PRETIMEOUT) + return attr->mode; + + return 0; +} + +static const struct attribute_group wdd_pretimeout_group = { + .is_visible = wdd_pretimeout_attr_is_visible, + .attrs = wdd_pretimeout_attrs, +}; + +static const struct attribute_group *wdd_pretimeout_groups[] = { + &wdd_pretimeout_group, + NULL, +}; + +int watchdog_register_pretimeout(struct watchdog_device *wdd, struct device *dev) +{ + struct watchdog_governor *gov; + int ret; + + if (!(wdd->info->options & WDIOF_PRETIMEOUT)) + return 0; + + ret = sysfs_create_groups(&dev->kobj, wdd_pretimeout_groups); + if (ret) + return ret; + + mutex_lock(&governor_lock); + gov = find_default_governor(); + if (gov && !try_module_get(gov->owner)) { + mutex_unlock(&governor_lock); + return -ENODEV; + } + + spin_lock_irq(&pretimeout_lock); + wdd->gov = gov; + spin_unlock_irq(&pretimeout_lock); + + mutex_unlock(&governor_lock); + + return 0; +} + +void watchdog_unregister_pretimeout(struct watchdog_device *wdd) +{ + if (!(wdd->info->options & WDIOF_PRETIMEOUT)) + return; + + spin_lock_irq(&pretimeout_lock); + if (wdd->gov) + module_put(wdd->gov->owner); + wdd->gov = NULL; + spin_unlock_irq(&pretimeout_lock); +} + +MODULE_AUTHOR("Vladimir Zapolskiy <vladimir_zapolskiy@mentor.com>"); +MODULE_DESCRIPTION("Watchdog pretimeout governor framework"); +MODULE_LICENSE("GPL"); diff --git a/drivers/watchdog/watchdog_pretimeout.h b/drivers/watchdog/watchdog_pretimeout.h new file mode 100644 index 00000000000000..38965cdd23bebe --- /dev/null +++ b/drivers/watchdog/watchdog_pretimeout.h @@ -0,0 +1,35 @@ +#ifndef __WATCHDOG_PRETIMEOUT_H +#define __WATCHDOG_PRETIMEOUT_H + +#define WATCHDOG_GOV_NAME_LEN 20 + +struct module; +struct watchdog_device; + +struct watchdog_governor { + const char name[WATCHDOG_GOV_NAME_LEN]; + void (*pretimeout)(struct watchdog_device *wdd); + struct module *owner; +}; + +/* Interfaces to watchdog pretimeout governors */ +int watchdog_register_governor(struct watchdog_governor *gov); +void watchdog_unregister_governor(struct watchdog_governor *gov); + +/* Interfaces to watchdog_core.c */ +#ifdef CONFIG_WATCHDOG_PRETIMEOUT_GOV + +#define WATCHDOG_PRETIMEOUT_DEFAULT_GOV "none" + +int watchdog_register_pretimeout(struct watchdog_device *wdd, struct device *dev); +void watchdog_unregister_pretimeout(struct watchdog_device *wdd); +#else +static inline int watchdog_register_pretimeout(struct watchdog_device *wdd, + struct device *dev) +{ + return 0; +} +static inline void watchdog_unregister_pretimeout(struct watchdog_device *wdd) {} +#endif + +#endif diff --git a/include/linux/watchdog.h b/include/linux/watchdog.h index 51732d6c9555de..4b2fa45370fceb 100644 --- a/include/linux/watchdog.h +++ b/include/linux/watchdog.h @@ -19,6 +19,7 @@ struct watchdog_ops; struct watchdog_device; struct watchdog_core_data; +struct watchdog_governor; /** struct watchdog_ops - The watchdog-devices operations * @@ -59,6 +60,7 @@ struct watchdog_ops { * watchdog device. * @info: Pointer to a watchdog_info structure. * @ops: Pointer to the list of watchdog operations. + * @gov: Pointer to watchdog pretimeout governor. * @bootstatus: Status of the watchdog device at boot. * @timeout: The watchdog devices timeout value (in seconds). * @min_timeout:The watchdog devices minimum timeout value (in seconds). @@ -93,6 +95,7 @@ struct watchdog_device { const struct attribute_group **groups; const struct watchdog_info *info; const struct watchdog_ops *ops; + struct watchdog_governor *gov; unsigned int bootstatus; unsigned int timeout; unsigned int min_timeout; @@ -180,4 +183,11 @@ extern int watchdog_init_timeout(struct watchdog_device *wdd, extern int watchdog_register_device(struct watchdog_device *); extern void watchdog_unregister_device(struct watchdog_device *); +/* drivers/watchdog/watchdog_pretimeout.c */ +#ifdef CONFIG_WATCHDOG_PRETIMEOUT_GOV +void watchdog_notify_pretimeout(struct watchdog_device *wdd); +#else +static inline void watchdog_notify_pretimeout(struct watchdog_device *wdd) {} +#endif + #endif /* ifndef _LINUX_WATCHDOG_H */