New feature proposal "quickwakeup"
Subsystems (3)
Hi All,
The purpose of this feature is to drastically reduce the suspend/resume time for device driver which needs to do periodic job.
In our use case (android smartphone), the system is most of the time in suspend to RAM, and needs to send a low level command every 30s. With current framework it takes about 500ms on omap3430 to resume the full system, and then suspend again. With quickwakup feature, in the resume process after resuming sysdev and re-enabling irq, the driver handler is executed, and then it suspends again.
This new path takes 20ms for us, which leads to good power-saving.
This can be used for battery-level check, or for SIM card polling on GSM network for example.
I need your feedback to know if we should keep it internally or if it can make its way to the mainline kernel.
here is a patch with my current work, it doesn't compile, it's a kind of early prototype.
there are still a lot of open issues, like race condition with other wakeup events ( I use wakelocks on android, but don't know how to have a similar thing on linux ) ...
basically, a driver who wants to use this feature will register two functions :
a check function, called with interrupt disable. the driver should read the irqstatus of his hw and return if the wakeup is for him or not. a callback function, called with interrupt enable, and sysdev resumed. it should block until the end of the task. if it returns an error, the other drivers and userspace are resumed. otherwise it goes to suspend to RAM again.
I rebased my patches on the branch linux-next, I'm not sure if it is the right branch.
Best regards,
Jocelyn
--- include/linux/quickwakeup.h | 33 +++++++++++++++++++++++ kernel/power/Kconfig | 8 +++++ kernel/power/Makefile | 2 +- kernel/power/quickwakeup.c | 60 +++++++++++++++++++++++++++++++++++++++++++ kernel/power/suspend.c | 42 +++++++++++++++++++++-------- 5 files changed, 132 insertions(+), 13 deletions(-) create mode 100755 include/linux/quickwakeup.h create mode 100644 kernel/power/quickwakeup.c
include/linux/quickwakeup.h new file +33 -0
@@ -0,0 +1,33 @@ +/* include/linux/quickwakeup.h +* +* Copyright (C) 2009 Motorola. +* +* This software is licensed under the terms of the GNU General Public +* License version 2, as published by the Free Software Foundation, and +* may be copied, distributed, and modified under those terms. +* +* This program is distributed in the hope that 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. +* +*/ + +struct quickwakeup_ops { +struct list_head list; +int (*qw_callback) (void ); +int (*qw_check)(void ); +int checked; +}; + +#ifdef CONFIG_QUICK_WAKEUP + +int quickwakeup_register (struct quickwakeup_ops *ops); +int quickwakeup_check (void ); +int quickwakeup_execute (void ); +void quickwakeup_unregister (struct quickwakeup_ops *ops); + +#else +static int quickwakeup_register (struct quickwakeup_ops *ops) {return 0 ; }; +void quickwakeup_unregister (struct quickwakeup_ops *ops) {}; +#endif
kernel/power/Kconfig +8 -0
@@ -236,3 +236,11 @@ config PM_RUNTIME and the bus type drivers of the buses the devices are on are responsible for the actual handling of the autosuspend requests and wake-up events. + +config QUICK_WAKEUP + bool "Quick wakeup" + depends on SUSPEND + default n + ---help--- + Allow kernel driver to do periodic jobs without resuming the full system + This option can increase battery life on android powered smartphone.
kernel/power/Makefile +1 -1
@@ -10,5 +10,5 @@ obj-$(CONFIG_SUSPEND) += suspend.o obj-$(CONFIG_PM_TEST_SUSPEND) += suspend_test.o obj-$(CONFIG_HIBERNATION) += hibernate.o snapshot.o swap.o user.o obj-$(CONFIG_HIBERNATION_NVS) += hibernate_nvs.o - +obj-$(CONFIG_QUICK_WAKEUP) += quickwakeup.o obj-$(CONFIG_MAGIC_SYSRQ) += poweroff.o
kernel/power/quickwakeup.c new file +60 -0
60 lines not on this page · raw message
kernel/power/suspend.c +30 -12
54 lines not on this page · raw message
Signature · 1 line
1.6.3.3
Hmm. This is an interesting approach. Just to understand the usecase better, though - my understanding is that Nokia's approach on Omap-based devices has been to treat "suspend" as a runtime state, without actually performing a systemwide transition. What are the constraints that make shifting into a different hardware state more efficient for you?
Signature · 1 line
Matthew Garrett | [email protected]
Quoted 6 lines · Matthew Garrett
Hmm. This is an interesting approach. Just to understand the usecase better, though - my understanding is that Nokia's approach on Omap-based devices has been to treat "suspend" as a runtime state, without actually performing a systemwide transition. What are the constraints that make shifting into a different hardware state more efficient for you?
In fact we are doing "retention" in idle and with the suspend path. It's specific to Android, with the global suspend and wakelock framework. When all wakelocks are released (ie screen is black, no running app), the phone goes to suspend, which freeze userspace and also suspend the device driver. This way you can choose the wakeup sources, and stay in suspend during a long time (a few hours). Without this, you have to track down all user space app that wakeup the phone, and it is hard to sleep more than a few second.
The drawback is that when you want to wakeup, it takes a lot of time to unfreeze the userspace, and resume all the device driver. This is what this feature address (for low-level wakeup only).
But as this patch is not Android specific, I think other platforms may use it too.
Best regards,
Jocelyn
Quoted 8 lines · Falempe Jocelyn
In fact we are doing "retention" in idle and with the suspend path. It's specific to Android, with the global suspend and wakelock framework. When all wakelocks are released (ie screen is black, no running app), the phone goes to suspend, which freeze userspace and also suspend the device driver. This way you can choose the wakeup sources, and stay in suspend during a long time (a few hours). Without this, you have to track down all user space app that wakeup the phone, and it is hard to sleep more than a few second.
What's triggering the application wakeups? If it's just timeouts, can't you simply adjust the range timer slop so that the timeouts don't fire?
Quoted 5 lines
The drawback is that when you want to wakeup, it takes a lot of time to unfreeze the userspace, and resume all the device driver. This is what this feature address (for low-level wakeup only).
But as this patch is not Android specific, I think other platforms may use it too.
Yeah, I can certainly see the appeal if you have that constraint.
Signature · 1 line
Matthew Garrett | [email protected]
Quoted 11 lines · Matthew Garrett
Quoted 8 lines · Falempe Jocelyn
In fact we are doing "retention" in idle and with the suspend path. It's specific to Android, with the global suspend and wakelock framework. When all wakelocks are released (ie screen is black, no running app), the phone goes to suspend, which freeze userspace and also suspend the device driver. This way you can choose the wakeup sources, and stay in suspend during a long time (a few hours). Without this, you have to track down all user space app that wakeup the phone, and it is hard to sleep more than a few second.
Quoted 2 lines
What's triggering the application wakeups? If it's just timeouts, can't you simply adjust the range timer slop so that the timeouts don't fire?
yes that's a good point. I think google chose the suspend way also because it synchronize filesystem, and some driver can enter deeper sleep, knowing that the user won't notice the increased latency. it protects also the phone against a crashed/bad app that would eat 100% cpu always.
Quoted 9 lines
Quoted 6 lines
The drawback is that when you want to wakeup, it takes a lot of time to unfreeze the userspace, and resume all the device driver. This is what this feature address (for low-level wakeup only).
But as this patch is not Android specific, I think other platforms may use it too.Yeah, I can certainly see the appeal if you have that constraint.