This patch implements kexec based hibernate/resume. This is based on
the facility provided by kexec_jump. The states save/restore code of
ordinary kexec_jump is overridden by hibernate/resume specific
code. The ACPI methods are called at specified environment to conform
the ACPI specification. A new reboot command is added to go to ACPI S4
state from user space.
Signed-off-by: Huang Ying <[email protected]>
---
include/linux/kexec.h | 4
include/linux/reboot.h | 1
include/linux/suspend.h | 1
kernel/power/disk.c | 244 +++++++++++++++++++++++++++++++++++++++++++++++-
kernel/sys.c | 5
5 files changed, 251 insertions(+), 4 deletions(-)
--- a/kernel/power/disk.c
+++ b/kernel/power/disk.c
@@ -21,6 +21,7 @@
#include <linux/console.h>
#include <linux/cpu.h>
#include <linux/freezer.h>
+#include <linux/kexec.h>
#include "power.h"
@@ -365,13 +366,13 @@ int hibernation_platform_enter(void)
}
/**
- * power_down - Shut the machine down for hibernation.
+ * hibernate_power_down - Shut the machine down for hibernation.
*
* Use the platform driver, if configured so; otherwise try
* to power off or reboot.
*/
-static void power_down(void)
+void hibernate_power_down(void)
{
switch (hibernation_mode) {
case HIBERNATION_TEST:
@@ -461,7 +462,7 @@ int hibernate(void)
error = swsusp_write(flags);
swsusp_free();
if (!error)
- power_down();
+ hibernate_power_down();
} else {
pr_debug("PM: Image restored successfully.\n");
swsusp_free();
@@ -478,6 +479,243 @@ int hibernate(void)
return error;
}
+#ifdef CONFIG_KEXEC
+static int kexec_snapshot(struct notifier_block *nb,
+ unsigned long cmd, void *arg)
+{
+ int error;
+ int platform_mode = (hibernation_mode == HIBERNATION_PLATFORM);
+
+ if (cmd != KJUMP_CMD_HIBERNATE_WRITE_IMAGE)
+ return NOTIFY_DONE;
+
+ pm_prepare_console();
+
+ error = pm_notifier_call_chain(PM_HIBERNATION_PREPARE);
+ if (error)
+ goto Exit;
+
+ error = freeze_processes();
+ if (error) {
+ error = -EBUSY;
+ goto Exit;
+ }
+
+ if (hibernation_test(TEST_FREEZER) ||
+ hibernation_testmode(HIBERNATION_TESTPROC)) {
+ error = -EAGAIN;
+ goto Resume_process;
+ }
+
+ error = platform_start(platform_mode);
+ if (error)
+ goto Resume_process;
+
+ suspend_console();
+ error = device_suspend(PMSG_FREEZE);
+ if (error)
+ goto Resume_console;
+
+ if (hibernation_test(TEST_DEVICES)) {
+ error = -EAGAIN;
+ goto Resume_devices;
+ }
+
+ error = platform_pre_snapshot(platform_mode);
+ if (error)
+ goto Resume_devices;
+
+ if (hibernation_test(TEST_PLATFORM)) {
+ error = -EAGAIN;
+ goto Resume_devices;
+ }
+
+ error = disable_nonboot_cpus();
+ if (error)
+ goto Resume_devices;
+
+ if (hibernation_test(TEST_CPUS) ||
+ hibernation_testmode(HIBERNATION_TEST)) {
+ error = -EAGAIN;
+ goto Enable_cpus;
+ }
+
+ local_irq_disable();
+ /* At this point, device_suspend() has been called, but *not*
+ * device_power_down(). We *must* device_power_down() now.
+ * Otherwise, drivers for some devices (e.g. interrupt
+ * controllers) become desynchronized with the actual state of
+ * the hardware at resume time, and evil weirdness ensues.
+ */
+ error = device_power_down(PMSG_FREEZE);
+ if (error)
+ goto Enable_irqs;
+
+ if (hibernation_test(TEST_CORE)) {
+ error = -EAGAIN;
+ goto Power_up;
+ }
+
+ return NOTIFY_STOP;
+
+ Power_up:
+ device_power_up();
+ Enable_irqs:
+ local_irq_enable();
+ Enable_cpus:
+ enable_nonboot_cpus();
+ Resume_devices:
+ platform_finish(platform_mode);
+ device_resume();
+ Resume_console:
+ resume_console();
+ Resume_process:
+ thaw_processes();
+ Exit:
+ pm_notifier_call_chain(PM_POST_HIBERNATION);
+ pm_restore_console();
+ return notifier_from_errno(error);
+}
+
+static int kexec_prepare_write_image(struct notifier_block *nb,
+ unsigned long cmd, void *arg)
+{
+ int platform_mode = (hibernation_mode == HIBERNATION_PLATFORM);
+
+ if (cmd != KJUMP_CMD_HIBERNATE_WRITE_IMAGE)
+ return NOTIFY_DONE;
+
+ device_power_up();
+ local_irq_enable();
+ enable_nonboot_cpus();
+ platform_finish(platform_mode);
+ device_resume();
+ resume_console();
+ thaw_processes();
+ pm_restore_console();
+ return NOTIFY_STOP;
+}
+
+static int kexec_prepare_resume(struct notifier_block *nb,
+ unsigned long cmd, void *arg)
+{
+ int error;
+ int platform_mode = (hibernation_mode == HIBERNATION_PLATFORM);
+
+ if (cmd != KJUMP_CMD_HIBERNATE_RESUME)
+ return NOTIFY_DONE;
+
+ pm_prepare_console();
+ error = pm_notifier_call_chain(PM_RESTORE_PREPARE);
+ if (error)
+ goto Exit;
+
+ error = freeze_processes();
+ if (error) {
+ error = -EBUSY;
+ goto Exit;
+ }
+
+ suspend_console();
+ error = device_suspend(PMSG_PRETHAW);
+ if (error)
+ goto Resume_console;
+
+ error = platform_pre_restore(platform_mode);
+ if (error)
+ goto Resume_devices;
+
+ error = disable_nonboot_cpus();
+ if (error)
+ goto Resume_devices;
+ local_irq_disable();
+ /* At this point, device_suspend() has been called, but *not*
+ * device_power_down(). We *must* device_power_down() now.
+ * Otherwise, drivers for some devices (e.g. interrupt controllers)
+ * become desynchronized with the actual state of the hardware
+ * at resume time, and evil weirdness ensues.
+ */
+ error = device_power_down(PMSG_PRETHAW);
+ if (error)
+ goto Enable_irqs;
+
+ return NOTIFY_STOP;
+
+ Enable_irqs:
+ local_irq_enable();
+ enable_nonboot_cpus();
+ Resume_devices:
+ platform_restore_cleanup(platform_mode);
+ device_resume();
+ Resume_console:
+ resume_console();
+ thaw_processes();
+ Exit:
+ pm_notifier_call_chain(PM_POST_RESTORE);
+ pm_restore_console();
+ return notifier_from_errno(error);
+}
+
+static int kexec_resume(struct notifier_block *nb,
+ unsigned long cmd, void *arg)
+{
+ int platform_mode = (hibernation_mode == HIBERNATION_PLATFORM);
+
+ if (cmd != KJUMP_CMD_HIBERNATE_RESUME)
+ return NOTIFY_DONE;
+
+ platform_leave(platform_mode);
+ device_power_up();
+ local_irq_enable();
+ enable_nonboot_cpus();
+ platform_finish(platform_mode);
+ device_resume();
+ resume_console();
+ thaw_processes();
+ pm_notifier_call_chain(PM_POST_HIBERNATION);
+ pm_restore_console();
+ return NOTIFY_STOP;
+}
+
+static struct notifier_block nb_kexec_snapshot = {
+ .notifier_call = kexec_snapshot,
+};
+
+static struct notifier_block nb_kexec_prepare_write_image = {
+ .notifier_call = kexec_prepare_write_image,
+};
+
+static struct notifier_block nb_kexec_prepare_resume = {
+ .notifier_call = kexec_prepare_resume,
+};
+
+static struct notifier_block nb_kexec_resume = {
+ .notifier_call = kexec_resume,
+};
+
+static int kexec_hibernation_register(void)
+{
+ int ret;
+
+ ret = blocking_notifier_chain_register(&kjump_chain_pre,
+ &nb_kexec_snapshot);
+ if (ret)
+ return ret;
+ ret = blocking_notifier_chain_register(&kjump_chain_post,
+ &nb_kexec_prepare_write_image);
+ if (ret)
+ return ret;
+ ret = blocking_notifier_chain_register(&kjump_chain_pre,
+ &nb_kexec_prepare_resume);
+ if (ret)
+ return ret;
+ ret = blocking_notifier_chain_register(&kjump_chain_post,
+ &nb_kexec_resume);
+ return ret;
+}
+
+late_initcall(kexec_hibernation_register);
+#endif
/**
* software_resume - Resume from a saved image.
--- a/include/linux/suspend.h
+++ b/include/linux/suspend.h
@@ -198,6 +198,7 @@ extern unsigned long get_safe_page(gfp_t
extern void hibernation_set_ops(struct platform_hibernation_ops *ops);
extern int hibernate(void);
+extern void hibernate_power_down(void);
#else /* CONFIG_HIBERNATION */
static inline int swsusp_page_is_forbidden(struct page *p) { return 0; }
static inline void swsusp_set_page_free(struct page *p) {}
--- a/kernel/sys.c
+++ b/kernel/sys.c
@@ -442,6 +442,11 @@ asmlinkage long sys_reboot(int magic1, i
unlock_kernel();
return ret;
}
+
+ case LINUX_REBOOT_CMD_HIBERNATE_POWER_DOWN:
+ hibernate_power_down();
+ unlock_kernel();
+ return -EINVAL;
#endif
default:
--- a/include/linux/reboot.h
+++ b/include/linux/reboot.h
@@ -33,6 +33,7 @@
#define LINUX_REBOOT_CMD_RESTART2 0xA1B2C3D4
#define LINUX_REBOOT_CMD_SW_SUSPEND 0xD000FCE2
#define LINUX_REBOOT_CMD_KEXEC 0x45584543
+#define LINUX_REBOOT_CMD_HIBERNATE_POWER_DOWN 0xE4390DF3
#ifdef __KERNEL__
--- a/include/linux/kexec.h
+++ b/include/linux/kexec.h
@@ -127,7 +127,9 @@ extern int kexec_vcall(struct kimage *im
unsigned int argc, va_list args);
extern int kexec_jump(struct kimage *image, unsigned long *cmd_ret,
unsigned long cmd);
-#define KJUMP_CMD_NONE 0
+#define KJUMP_CMD_NONE 0x0
+#define KJUMP_CMD_HIBERNATE_WRITE_IMAGE 0x6b630001
+#define KJUMP_CMD_HIBERNATE_RESUME 0x6b630002
#ifdef CONFIG_COMPAT
extern asmlinkage long compat_sys_kexec_load(unsigned long entry,
unsigned long nr_segments,
On Friday, 7 of December 2007, Huang, Ying wrote:
> This patch implements kexec based hibernate/resume. This is based on
> the facility provided by kexec_jump. The states save/restore code of
> ordinary kexec_jump is overridden by hibernate/resume specific
> code.
Can you explain in more details how this works?
> The ACPI methods are called at specified environment to conform
> the ACPI specification. A new reboot command is added to go to ACPI S4
> state from user space.
Well, I still don't like the amount of duplicated code introduced by this patch.
Also, IMO it should be using the mutual exclusion mechanisms used by the
existing hibernation code, ie. pm_mutex and the snapshot_device_available
atomic variable.
> Signed-off-by: Huang Ying <[email protected]>
>
> ---
> include/linux/kexec.h | 4
> include/linux/reboot.h | 1
> include/linux/suspend.h | 1
> kernel/power/disk.c | 244 +++++++++++++++++++++++++++++++++++++++++++++++-
> kernel/sys.c | 5
> 5 files changed, 251 insertions(+), 4 deletions(-)
>
> --- a/kernel/power/disk.c
> +++ b/kernel/power/disk.c
> @@ -21,6 +21,7 @@
> #include <linux/console.h>
> #include <linux/cpu.h>
> #include <linux/freezer.h>
> +#include <linux/kexec.h>
>
> #include "power.h"
>
> @@ -365,13 +366,13 @@ int hibernation_platform_enter(void)
> }
>
> /**
> - * power_down - Shut the machine down for hibernation.
> + * hibernate_power_down - Shut the machine down for hibernation.
> *
> * Use the platform driver, if configured so; otherwise try
> * to power off or reboot.
> */
>
> -static void power_down(void)
> +void hibernate_power_down(void)
> {
> switch (hibernation_mode) {
> case HIBERNATION_TEST:
> @@ -461,7 +462,7 @@ int hibernate(void)
> error = swsusp_write(flags);
> swsusp_free();
> if (!error)
> - power_down();
> + hibernate_power_down();
> } else {
> pr_debug("PM: Image restored successfully.\n");
> swsusp_free();
> @@ -478,6 +479,243 @@ int hibernate(void)
> return error;
> }
>
> +#ifdef CONFIG_KEXEC
> +static int kexec_snapshot(struct notifier_block *nb,
> + unsigned long cmd, void *arg)
> +{
> + int error;
> + int platform_mode = (hibernation_mode == HIBERNATION_PLATFORM);
> +
> + if (cmd != KJUMP_CMD_HIBERNATE_WRITE_IMAGE)
> + return NOTIFY_DONE;
> +
> + pm_prepare_console();
> +
> + error = pm_notifier_call_chain(PM_HIBERNATION_PREPARE);
> + if (error)
> + goto Exit;
> +
> + error = freeze_processes();
> + if (error) {
> + error = -EBUSY;
> + goto Exit;
> + }
> +
> + if (hibernation_test(TEST_FREEZER) ||
> + hibernation_testmode(HIBERNATION_TESTPROC)) {
> + error = -EAGAIN;
> + goto Resume_process;
> + }
> +
> + error = platform_start(platform_mode);
> + if (error)
> + goto Resume_process;
> +
> + suspend_console();
> + error = device_suspend(PMSG_FREEZE);
> + if (error)
> + goto Resume_console;
> +
> + if (hibernation_test(TEST_DEVICES)) {
> + error = -EAGAIN;
> + goto Resume_devices;
> + }
> +
> + error = platform_pre_snapshot(platform_mode);
> + if (error)
> + goto Resume_devices;
> +
> + if (hibernation_test(TEST_PLATFORM)) {
> + error = -EAGAIN;
> + goto Resume_devices;
> + }
> +
> + error = disable_nonboot_cpus();
> + if (error)
> + goto Resume_devices;
> +
> + if (hibernation_test(TEST_CPUS) ||
> + hibernation_testmode(HIBERNATION_TEST)) {
> + error = -EAGAIN;
> + goto Enable_cpus;
> + }
> +
> + local_irq_disable();
> + /* At this point, device_suspend() has been called, but *not*
> + * device_power_down(). We *must* device_power_down() now.
> + * Otherwise, drivers for some devices (e.g. interrupt
> + * controllers) become desynchronized with the actual state of
> + * the hardware at resume time, and evil weirdness ensues.
> + */
> + error = device_power_down(PMSG_FREEZE);
> + if (error)
> + goto Enable_irqs;
> +
> + if (hibernation_test(TEST_CORE)) {
> + error = -EAGAIN;
> + goto Power_up;
> + }
> +
> + return NOTIFY_STOP;
> +
> + Power_up:
> + device_power_up();
> + Enable_irqs:
> + local_irq_enable();
> + Enable_cpus:
> + enable_nonboot_cpus();
> + Resume_devices:
> + platform_finish(platform_mode);
> + device_resume();
> + Resume_console:
> + resume_console();
> + Resume_process:
> + thaw_processes();
> + Exit:
> + pm_notifier_call_chain(PM_POST_HIBERNATION);
> + pm_restore_console();
> + return notifier_from_errno(error);
> +}
> +
> +static int kexec_prepare_write_image(struct notifier_block *nb,
> + unsigned long cmd, void *arg)
> +{
> + int platform_mode = (hibernation_mode == HIBERNATION_PLATFORM);
> +
> + if (cmd != KJUMP_CMD_HIBERNATE_WRITE_IMAGE)
> + return NOTIFY_DONE;
> +
> + device_power_up();
> + local_irq_enable();
> + enable_nonboot_cpus();
> + platform_finish(platform_mode);
> + device_resume();
> + resume_console();
> + thaw_processes();
> + pm_restore_console();
> + return NOTIFY_STOP;
> +}
> +
> +static int kexec_prepare_resume(struct notifier_block *nb,
> + unsigned long cmd, void *arg)
> +{
> + int error;
> + int platform_mode = (hibernation_mode == HIBERNATION_PLATFORM);
> +
> + if (cmd != KJUMP_CMD_HIBERNATE_RESUME)
> + return NOTIFY_DONE;
> +
> + pm_prepare_console();
> + error = pm_notifier_call_chain(PM_RESTORE_PREPARE);
> + if (error)
> + goto Exit;
> +
> + error = freeze_processes();
> + if (error) {
> + error = -EBUSY;
> + goto Exit;
> + }
> +
> + suspend_console();
> + error = device_suspend(PMSG_PRETHAW);
> + if (error)
> + goto Resume_console;
> +
> + error = platform_pre_restore(platform_mode);
> + if (error)
> + goto Resume_devices;
> +
> + error = disable_nonboot_cpus();
> + if (error)
> + goto Resume_devices;
> + local_irq_disable();
> + /* At this point, device_suspend() has been called, but *not*
> + * device_power_down(). We *must* device_power_down() now.
> + * Otherwise, drivers for some devices (e.g. interrupt controllers)
> + * become desynchronized with the actual state of the hardware
> + * at resume time, and evil weirdness ensues.
> + */
> + error = device_power_down(PMSG_PRETHAW);
> + if (error)
> + goto Enable_irqs;
> +
> + return NOTIFY_STOP;
> +
> + Enable_irqs:
> + local_irq_enable();
> + enable_nonboot_cpus();
> + Resume_devices:
> + platform_restore_cleanup(platform_mode);
> + device_resume();
> + Resume_console:
> + resume_console();
> + thaw_processes();
> + Exit:
> + pm_notifier_call_chain(PM_POST_RESTORE);
> + pm_restore_console();
> + return notifier_from_errno(error);
> +}
> +
> +static int kexec_resume(struct notifier_block *nb,
> + unsigned long cmd, void *arg)
> +{
> + int platform_mode = (hibernation_mode == HIBERNATION_PLATFORM);
> +
> + if (cmd != KJUMP_CMD_HIBERNATE_RESUME)
> + return NOTIFY_DONE;
> +
> + platform_leave(platform_mode);
> + device_power_up();
> + local_irq_enable();
> + enable_nonboot_cpus();
> + platform_finish(platform_mode);
> + device_resume();
> + resume_console();
> + thaw_processes();
> + pm_notifier_call_chain(PM_POST_HIBERNATION);
> + pm_restore_console();
> + return NOTIFY_STOP;
> +}
> +
> +static struct notifier_block nb_kexec_snapshot = {
> + .notifier_call = kexec_snapshot,
> +};
> +
> +static struct notifier_block nb_kexec_prepare_write_image = {
> + .notifier_call = kexec_prepare_write_image,
> +};
> +
> +static struct notifier_block nb_kexec_prepare_resume = {
> + .notifier_call = kexec_prepare_resume,
> +};
> +
> +static struct notifier_block nb_kexec_resume = {
> + .notifier_call = kexec_resume,
> +};
> +
> +static int kexec_hibernation_register(void)
> +{
> + int ret;
> +
> + ret = blocking_notifier_chain_register(&kjump_chain_pre,
> + &nb_kexec_snapshot);
> + if (ret)
> + return ret;
> + ret = blocking_notifier_chain_register(&kjump_chain_post,
> + &nb_kexec_prepare_write_image);
> + if (ret)
> + return ret;
> + ret = blocking_notifier_chain_register(&kjump_chain_pre,
> + &nb_kexec_prepare_resume);
> + if (ret)
> + return ret;
> + ret = blocking_notifier_chain_register(&kjump_chain_post,
> + &nb_kexec_resume);
> + return ret;
> +}
> +
> +late_initcall(kexec_hibernation_register);
> +#endif
>
> /**
> * software_resume - Resume from a saved image.
> --- a/include/linux/suspend.h
> +++ b/include/linux/suspend.h
> @@ -198,6 +198,7 @@ extern unsigned long get_safe_page(gfp_t
>
> extern void hibernation_set_ops(struct platform_hibernation_ops *ops);
> extern int hibernate(void);
> +extern void hibernate_power_down(void);
> #else /* CONFIG_HIBERNATION */
> static inline int swsusp_page_is_forbidden(struct page *p) { return 0; }
> static inline void swsusp_set_page_free(struct page *p) {}
> --- a/kernel/sys.c
> +++ b/kernel/sys.c
> @@ -442,6 +442,11 @@ asmlinkage long sys_reboot(int magic1, i
> unlock_kernel();
> return ret;
> }
> +
> + case LINUX_REBOOT_CMD_HIBERNATE_POWER_DOWN:
> + hibernate_power_down();
> + unlock_kernel();
> + return -EINVAL;
> #endif
>
> default:
> --- a/include/linux/reboot.h
> +++ b/include/linux/reboot.h
> @@ -33,6 +33,7 @@
> #define LINUX_REBOOT_CMD_RESTART2 0xA1B2C3D4
> #define LINUX_REBOOT_CMD_SW_SUSPEND 0xD000FCE2
> #define LINUX_REBOOT_CMD_KEXEC 0x45584543
> +#define LINUX_REBOOT_CMD_HIBERNATE_POWER_DOWN 0xE4390DF3
>
>
> #ifdef __KERNEL__
> --- a/include/linux/kexec.h
> +++ b/include/linux/kexec.h
> @@ -127,7 +127,9 @@ extern int kexec_vcall(struct kimage *im
> unsigned int argc, va_list args);
> extern int kexec_jump(struct kimage *image, unsigned long *cmd_ret,
> unsigned long cmd);
> -#define KJUMP_CMD_NONE 0
> +#define KJUMP_CMD_NONE 0x0
> +#define KJUMP_CMD_HIBERNATE_WRITE_IMAGE 0x6b630001
> +#define KJUMP_CMD_HIBERNATE_RESUME 0x6b630002
> #ifdef CONFIG_COMPAT
> extern asmlinkage long compat_sys_kexec_load(unsigned long entry,
> unsigned long nr_segments,
>
>
On Dec 7, 2007 8:52 PM, Rafael J. Wysocki <[email protected]> wrote:
> On Friday, 7 of December 2007, Huang, Ying wrote:
> > This patch implements kexec based hibernate/resume. This is based on
> > the facility provided by kexec_jump. The states save/restore code of
> > ordinary kexec_jump is overridden by hibernate/resume specific
> > code.
>
> Can you explain in more details how this works?
Two blocking notifier chain named kjump_chain_pre and kjump_chain_post
are defined, the basic procedure of kexec jump is as follow:
call functions in kjump_chain_pre
jump to peer kernel
call functions in kjump_chain_post
A command is the first parameter of functions in chain. If A command
is processed in a function, the function will execute and stop the
chain (return NOTIFY_STOP), otherwise it will do nothing (return
NOTIFY_DONE). If no function has interest in the command, the default
behavior will be executed (kexec_vcall_pre, kexec_vcall_post).
So for each command the procedure is as follow:
KJUMP_CMD_HIBERNATE_WRITE_IMAGE:
[chain] kexec_snapshot
jump to kexeced kernel
[chain] kexec_prepare_write_image /* in kexeced kernel */
KJUMP_HIBERNATE_RESUME:
[chain] kexec_prepare_resume /* in kexeced kernel */
jump to kexec kernel
[chain] kexec_resume
> > The ACPI methods are called at specified environment to conform
> > the ACPI specification. A new reboot command is added to go to ACPI S4
> > state from user space.
>
> Well, I still don't like the amount of duplicated code introduced by this patch.
Yes, there are too many duplicated code. They should be merged. But I
want to delay the merging until the kexec based hibernation code goes
more stable.
> Also, IMO it should be using the mutual exclusion mechanisms used by the
> existing hibernation code, ie. pm_mutex and the snapshot_device_available
> atomic variable.
Now the kexec_lock is used as a mutex between kexec related
operations. It seems reasonable to use pm_mutex and maybe
snapshot_device_available to eliminate potential conflict between
kexec based hibernation and u/swsusp.
Best Regards,
Huang Ying