Menu
▾
▴
dynamicpower-devel — Dynamic Power Management development
You can subscribe to this list here.
| 2003 |
Jan
|
Feb
|
Mar
|
Apr
|
May
|
Jun
|
Jul
|
Aug
|
Sep
|
Oct
|
Nov
(1) |
Dec
(1) |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2004 |
Jan
|
Feb
|
Mar
(2) |
Apr
(1) |
May
(1) |
Jun
(1) |
Jul
(1) |
Aug
(1) |
Sep
|
Oct
(1) |
Nov
|
Dec
(4) |
| 2005 |
Jan
(10) |
Feb
|
Mar
(14) |
Apr
(8) |
May
(2) |
Jun
|
Jul
(1) |
Aug
(7) |
Sep
|
Oct
(1) |
Nov
(1) |
Dec
|
| 2006 |
Jan
|
Feb
|
Mar
|
Apr
(1) |
May
|
Jun
|
Jul
|
Aug
|
Sep
|
Oct
(1) |
Nov
(3) |
Dec
|
| 2007 |
Jan
(2) |
Feb
(2) |
Mar
(5) |
Apr
(12) |
May
(2) |
Jun
|
Jul
(10) |
Aug
(2) |
Sep
(2) |
Oct
|
Nov
(2) |
Dec
(4) |
| 2008 |
Jan
(2) |
Feb
(1) |
Mar
(1) |
Apr
|
May
|
Jun
(1) |
Jul
(3) |
Aug
|
Sep
(13) |
Oct
(6) |
Nov
(27) |
Dec
(37) |
| 2009 |
Jan
(13) |
Feb
(14) |
Mar
(11) |
Apr
(19) |
May
(41) |
Jun
(21) |
Jul
(29) |
Aug
(8) |
Sep
(2) |
Oct
|
Nov
|
Dec
(4) |
| 2010 |
Jan
(6) |
Feb
(5) |
Mar
(5) |
Apr
|
May
(2) |
Jun
(1) |
Jul
|
Aug
(2) |
Sep
(1) |
Oct
(2) |
Nov
|
Dec
|
|
From: Manish R. <man...@st...> - 2007-04-03 11:57:22
|
Hi, I am implementing power management on arm926ejs based soc. When I take system to deep sleep (mem standby ) and wakeup, Then in some cases ( approx 1 out of 10), I get dump written at end of mail. In rest of cases it run successfully. Pls note than I wakeup succesfully and I get some prints from end of kernel/power/main.c. In mem standby entry and exit, save & restore mmu registers, peripheral registers etc. Can anybody give me any clue? I am not able to put any breakpoint on data abort handler after deep sleep. Thanks Regards Manish Dump is as follows: ------------------- Bad mode in data abort handler detected: mode UND_32 Internal error: Oops - bad mode: 0 [#3] Modules linked in: CPU: 0 PC is at 0xffff0380 LR is at update_mmu_cache+0x1b0/0x1c4 pc : [<ffff0380>] lr : [<c002a030>] Not tainted sp : c1b49e30 ip : 00000000 fp : c1b49ebc r10: c001e4e0 r9 : c937f4f8 r8 : 0000d000 r7 : 00000000 r6 : 00000005 r5 : c9822548 r4 : 00000000 r3 : 00001000 r2 : 00001875 r1 : 0000e000 r0 : 0000d6c0 Flags: nzcv IRQs off FIQs on Mode UND_32 Segment user Control: 5317F Table: 00530000 DAC: 00000015 Process busybox (pid: 227, stack limit = 0xc1b481a0) Stack: (0xc1b49e30 to 0xc1b4a000) 9e20: 0000d6c0 0000e000 00001875 00001000 9e40: 00000000 c9822548 00000005 00000000 0000d000 c937f4f8 c001e4e0 c1b49ebc 9e60: 00000000 c1b49e30 c002a030 ffff0380 0000009b ffffffff 0000008c c001b570 9e80: 00000080 00000000 00000000 c937f50c 00000005 00000005 09b8f05f c03e11e0 9ea0: c1b48000 c9822548 0000d000 c001e4e0 c1b49f2c c1b49ec0 c00759fc c0029e90 9ec0: c1b49ef4 c98222a0 c98222ac 0000081f c1b49efc 00000000 c02300a8 c0530000 9ee0: c0530000 c9c9a834 00000034 00000000 00000000 c1b49f00 c002a5dc 00000002 9f00: c1b49f2c ffffffff c001e514 c9822548 c93ac820 0000d334 c1b49fb0 c001e4e0 9f20: c1b49f6c c1b49f30 c002a2dc c0075580 c1b855a0 0000009e 00000000 00000000 9f40: c02977b8 ffffffff 0000d334 0000d334 00000000 c1b49fb0 00000000 40023000 9f60: c1b49f9c c1b49f70 c002a504 c002a208 c004d49c c98222e0 c1b49fac ffffffff 9f80: 00000000 0000d334 00000000 00000000 c1b49fac c1b49fa0 c002a59c c002a4e4 9fa0: 00000000 c1b49fb0 c0021d68 c002a594 4000d810 40017628 bed9de94 0000001c 9fc0: 00000000 00000000 0000d334 00000000 00000000 00000000 40023000 00000000 9fe0: 00000000 bed9de90 4000d4c8 0000d334 60000010 ffffffff 00000000 00000000 Backtrace: [<c0029e80>] (update_mmu_cache+0x0/0x1c4) from [<c00759fc>] (__handle_mm_fault+0x48c/0x92c) [<c0075570>] (__handle_mm_fault+0x0/0x92c) from [<c002a2dc>] (do_page_fault+0xe4/0x230) [<c002a1f8>] (do_page_fault+0x0/0x230) from [<c002a504>] (do_translation_fault+0x30/0xb0) [<c002a4d4>] (do_translation_fault+0x0/0xb0) from [<c002a59c>] (do_PrefetchAbort+0x18/0x1c) r8 = 00000000 r7 = 00000000 r6 = 0000D334 r5 = 00000000 r4 = FFFFFFFF [<c002a584>] (do_PrefetchAbort+0x0/0x1c) from [<c0021d68>] (ret_from_exception+0x0/0x10) Code: 00000000 00000000 00000000 00000000 (e88d4001) <6>note: busybox[227] exited with preempt_count 1 |
|
From: Manish R. <man...@st...> - 2007-03-30 13:45:29
|
Hi, While coming out of mem (writing mem into /sys/power/state) mode, code is not coming out of thaw_processes (kernel/power/process.c) All prints before and after appears. Can somebody help me in this regard? Thanks Regards Manish -----Original Message----- From: dyn...@li... [mailto:dyn...@li...] On Behalf Of dyn...@li... Sent: Thursday, March 22, 2007 12:41 AM To: dyn...@li... Subject: Dynamicpower-devel Digest, Vol 5, Issue 3 Send Dynamicpower-devel mailing list submissions to dyn...@li... To subscribe or unsubscribe via the World Wide Web, visit https://lists.sourceforge.net/lists/listinfo/dynamicpower-devel or, via email, send a message with subject or body 'help' to dyn...@li... You can reach the person managing the list at dyn...@li... When replying, please edit your Subject line so it is more specific than "Re: Contents of Dynamicpower-devel digest..." Today's Topics: 1. Re: Sleep and deep sleep state (Todd Poynor) ---------------------------------------------------------------------- Message: 1 Date: Tue, 20 Mar 2007 22:49:37 -0700 From: "Todd Poynor" <tod...@gm...> Subject: Re: [Dynamicpower-devel] Sleep and deep sleep state To: "Manish RATHI" <man...@st...> Cc: dyn...@li... Message-ID: <aeb...@ma...> Content-Type: text/plain; charset=ISO-8859-1; format=flowed On 3/19/07, Manish RATHI <man...@st...> wrote: > If I update only xtime not jiffies then any maturing timer started > before going sleep mode behave wrongly. I thought I had previously seen Linux kernel developers come down against jiffies updates for sleep intervals, but it seems there's examples of both in the tree for different architectures. If there's actions on the timer list that need to occur at some precise relative time after the system just woke up from an arbitrarily long period of complete inactivity then I'd be interested to hear about these; I'm guessing it doesn't much matter either way. If there's an action that needs to occur at some absolute time then it would normally be triggered by an RTC wakeup that pulls the system out of sleep at the appropriate time (if still asleep). > In OMAP/PXA is it possible to take system in sleep mode thru sysfs if > yes then ehere this sysfs file Reside? See http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=blob;f=Do cumentation/ABI/testing/sysfs-power;h=dcff4d0623add0e7708c642d0cfe210b0c1f48 ab;hb=HEAD for the generic description. Look for arch-specific code that registers pm_ops, for example, http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=blob;f=ar ch/arm/mach-pxa/pm.c;h=b4d8276d6050b1e34d16c06153228f03c0b113a3;hb=HEAD for the PXA implementation. -- Todd ------------------------------ ------------------------------------------------------------------------- Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT & business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV ------------------------------ _______________________________________________ Dynamicpower-devel mailing list Dyn...@li... https://lists.sourceforge.net/lists/listinfo/dynamicpower-devel End of Dynamicpower-devel Digest, Vol 5, Issue 3 ************************************************ |
|
From: Todd P. <tod...@gm...> - 2007-03-21 05:49:38
|
On 3/19/07, Manish RATHI <man...@st...> wrote: > If I update only xtime not jiffies then any maturing timer started before > going sleep mode behave wrongly. I thought I had previously seen Linux kernel developers come down against jiffies updates for sleep intervals, but it seems there's examples of both in the tree for different architectures. If there's actions on the timer list that need to occur at some precise relative time after the system just woke up from an arbitrarily long period of complete inactivity then I'd be interested to hear about these; I'm guessing it doesn't much matter either way. If there's an action that needs to occur at some absolute time then it would normally be triggered by an RTC wakeup that pulls the system out of sleep at the appropriate time (if still asleep). > In OMAP/PXA is it possible to take system in sleep mode thru sysfs if yes > then ehere this sysfs file > Reside? See http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=blob;f=Documentation/ABI/testing/sysfs-power;h=dcff4d0623add0e7708c642d0cfe210b0c1f48ab;hb=HEAD for the generic description. Look for arch-specific code that registers pm_ops, for example, http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=blob;f=arch/arm/mach-pxa/pm.c;h=b4d8276d6050b1e34d16c06153228f03c0b113a3;hb=HEAD for the PXA implementation. -- Todd |
|
From: Manish R. <man...@st...> - 2007-03-20 06:08:36
|
Hi, If I update only xtime not jiffies then any maturing timer started before going sleep mode behave wrongly. What's ur opinion? In OMAP/PXA is it possible to take system in sleep mode thru sysfs if yes then ehere this sysfs file Reside? Regards Manish -----Original Message----- From: Todd Poynor [mailto:tod...@qa...] Sent: Friday, March 16, 2007 12:17 AM To: Manish RATHI; dyn...@li... Subject: Re: [Dynamicpower-devel] Sleep and deep sleep state On Thu, 15 Mar 2007 1:11 am, Manish RATHI wrote: > Hi, > Is there any interface (sysfs or any other) in linux by which I can > take system into sleep mode? Standard feature of linux, several embedded board families such as TI OMAPs and Intel/Marvell PXA2xx implement the pm_ops that hook up to the generic interface. > Is there any reference code in which timer is updated by sleep > duration? Is it OK if I directly update Jiffies in my code. Adjusting jiffies for sleep intervals is not standard practice in linux (jiffies are relative OS time that can be considered to be frozen during OS inactivity). Adjust wall time (xtime) according to RTC or other timesource that keeps time during the sleep. See existing rtc drivers suspend/resume callbacks. --todd |
|
From: Todd P. <tod...@qa...> - 2007-03-15 18:56:41
|
On Thu, 15 Mar 2007 1:11 am, Manish RATHI wrote: > Hi, > Is there any interface (sysfs or any other) in linux by which I can > take system into sleep mode? Standard feature of linux, several embedded board families such as TI OMAPs and Intel/Marvell PXA2xx implement the pm_ops that hook up to the generic interface. > Is there any reference code in which timer is updated by sleep > duration? Is it OK if I directly update > Jiffies in my code. Adjusting jiffies for sleep intervals is not standard practice in linux (jiffies are relative OS time that can be considered to be frozen during OS inactivity). Adjust wall time (xtime) according to RTC or other timesource that keeps time during the sleep. See existing rtc drivers suspend/resume callbacks. --todd |
|
From: Manish R. <man...@st...> - 2007-03-15 08:06:19
|
Hi, Is there any interface (sysfs or any other) in linux by which I can take = system into sleep mode? Is there any reference code in which timer is updated by sleep = duration? Is it OK if I directly update Jiffies in my code. Thanks Regards Manish |
|
From: Todd P. <tod...@gm...> - 2007-02-07 19:25:25
|
On 2/6/07, Manish RATHI <man...@st...> wrote: > Hi, > Is there any plan to merge dpm work into main kernel tree? I've been away from this for a while so I might not be aware of the current status. Although portions of DPM have been discussed (namely the infamous Operating Points brouhaha), it is unlikely that other portions of DPM would be accepted in their current form. > I want to implement DPM over my arm based platform. > I've read white paper of Montavista and IBM of DPM. > >From which place I shd start? Grab the patches for OMAP etc. and see what is needed. So far as I know there's no porting documentation. > Is there any code available in arm linux for sleep and deep sleep mode entry and exit? Some arch-specific low power mode code is implemented in the main kernel for platforms like OMAP, XScale PXA, etc. Check the code for the pm ops for those platforms. -- Todd |
|
From: Manish R. <man...@st...> - 2007-02-06 08:32:35
|
Hi, Is there any plan to merge dpm work into main kernel tree? I want to implement DPM over my arm based platform.=20 I've read white paper of Montavista and IBM of DPM. >From which place I shd start? Is there any code available in arm linux for sleep and deep sleep mode = entry and exit? Regards Manish |
|
From: Todd P. <tod...@gm...> - 2007-01-17 06:38:15
|
On 1/16/07, Henry Prao <ra...@gm...> wrote: > I am researching the policy optimization for hard disk. I get patch > from sourceforge.net, but the patch is just fro CPU. I want to know if > there is any patch for hard disk. The Active Block I/O Scheduling System (ABISS) http://abiss.sourceforge.net/ is one solution for disk I/O policy that could be managed with power consumption in mind, if that's the sort of thing you're after. -- Todd |
|
From: Henry P. <ra...@gm...> - 2007-01-17 04:37:58
|
I am researching the policy optimization for hard disk. I get patch from sourceforge.net, but the patch is just fro CPU. I want to know if there is any patch for hard disk. |
|
From: Scott E. P. <pr...@mo...> - 2006-11-14 17:16:53
|
I think most systems using DPM have some kind of load-based policy manager, which adjusts the system OP based on the total system load. On an embedded system, with dedicated applications, it would be possible to have a finer-grained policy manager that tied the system OP to the specific use case or workload being run. However, on a general-purpose system, running an unconstrained workload, you're probably limited to doing reactive power control based on watching the system load. scott | From: <zha...@bj...> | | Thanks,Dave | | I'm also in trouble with questions as follows: | The new proc-fs interface can associate "op" with applications,but,if an | application have many states,like a JAVA game, | energy will be supplied when playing while going to sleep when not playing | for a long time.Is there a way to deal with this? | Another question,also take the JAVA game for example,users can get their | preferred game from Internet or other sources.How can | DPM manages these "Extended" applications which manufacturer has not | specific their "op" at the factory? | I think,if the core's load is heavy,it should switch to a more high speed | "op",and to low speed "op" when is light automately and | intelligently.Then,scaling the CPU's load is a problem.Any means can be | used to achieve this? | | Thanks and best wishes! | | Barry | Beijing,China | | | Actually, the new version of DPM I'm working on now just does that. You | can | set the operating point you want on a per process basis and any time that | process is exec'd or context switched the associated operating point | will be set by the kernel. | | There is a new procfs interface to simply set the operating point for a | process. | It makes it very simple to associate operating points with specific | applications on the system. | | On embedded system where typically all the applications are installed | at the factory the manufacturer knows which operating point they'd like | for each process. | | With the new interface they can set them at boot time and the system will | handle all the operating state changes invisibly to the applications. | | Dave | | On 11/8/06, zha...@bj... <zha...@bj...> | wrote: | > I'm very interested in DPM for linux.After read the document-"Dynamic | Power | > Management for Embedded Systems V1.1" and the source file of DPM patch | for | > 2.6.16 kernel and pxa27x,a question comes to me:how can DPM manage the | > applications up layer? Maybe the "sysfs" or "proc" interface can be | > utilized. But,a optimal idea is that - DPM is transparent to all | > applications.That is to say, we should realize the "Policy Manager" | refered | > by this document,but how? Leave the linux core monitoring the tasks | > schedule and then trigger "operation point" switch? | > | > Thanks and best wishes! | > | > | > | > ------------------------------------------------------------------------- | > Using Tomcat but need to do more? Need to support web services, security? | > Get stuff done quickly with pre-integrated technology to make your job | easier | > Download IBM WebSphere Application Server v.1.0.1 based on Apache | Geronimo | > http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642 | > _______________________________________________ | > Dynamicpower-devel mailing list | > Dyn...@li... | > https://lists.sourceforge.net/lists/listinfo/dynamicpower-devel | > | | | | | | ------------------------------------------------------------------------- | Using Tomcat but need to do more? Need to support web services, security? | Get stuff done quickly with pre-integrated technology to make your job easier | Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo | http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642 | _______________________________________________ | Dynamicpower-devel mailing list | Dyn...@li... | https://lists.sourceforge.net/lists/listinfo/dynamicpower-devel | -- scott preece motorola mobile devices, il67, 1800 s. oak st., champaign, il 61820 e-mail: pr...@mo... fax: +1-217-384-8550 phone: +1-217-384-8589 cell: +1-217-433-6114 pager: 217...@vt... |
|
From: <zha...@bj...> - 2006-11-13 06:10:30
|
Thanks,Dave I'm also in trouble with questions as follows: The new proc-fs interface can associate "op" with applications,but,if an application have many states,like a JAVA game, energy will be supplied when playing while going to sleep when not playing for a long time.Is there a way to deal with this? Another question,also take the JAVA game for example,users can get their preferred game from Internet or other sources.How can DPM manages these "Extended" applications which manufacturer has not specific their "op" at the factory? I think,if the core's load is heavy,it should switch to a more high speed "op",and to low speed "op" when is light automately and intelligently.Then,scaling the CPU's load is a problem.Any means can be used to achieve this? Thanks and best wishes! Barry Beijing,China Actually, the new version of DPM I'm working on now just does that. You can set the operating point you want on a per process basis and any time that process is exec'd or context switched the associated operating point will be set by the kernel. There is a new procfs interface to simply set the operating point for a process. It makes it very simple to associate operating points with specific applications on the system. On embedded system where typically all the applications are installed at the factory the manufacturer knows which operating point they'd like for each process. With the new interface they can set them at boot time and the system will handle all the operating state changes invisibly to the applications. Dave On 11/8/06, zha...@bj... <zha...@bj...> wrote: > I'm very interested in DPM for linux.After read the document-"Dynamic Power > Management for Embedded Systems V1.1" and the source file of DPM patch for > 2.6.16 kernel and pxa27x,a question comes to me:how can DPM manage the > applications up layer? Maybe the "sysfs" or "proc" interface can be > utilized. But,a optimal idea is that - DPM is transparent to all > applications.That is to say, we should realize the "Policy Manager" refered > by this document,but how? Leave the linux core monitoring the tasks > schedule and then trigger "operation point" switch? > > Thanks and best wishes! > > > > ------------------------------------------------------------------------- > Using Tomcat but need to do more? Need to support web services, security? > Get stuff done quickly with pre-integrated technology to make your job easier > Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo > http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642 > _______________________________________________ > Dynamicpower-devel mailing list > Dyn...@li... > https://lists.sourceforge.net/lists/listinfo/dynamicpower-devel > |
|
From: <zha...@bj...> - 2006-11-09 05:03:57
|
I'm very interested in DPM for linux.After read the document-"Dynamic Power Management for Embedded Systems V1.1" and the source file of DPM patch for 2.6.16 kernel and pxa27x,a question comes to me:how can DPM manage the applications up layer? Maybe the "sysfs" or "proc" interface can be utilized. But,a optimal idea is that - DPM is transparent to all applications.That is to say, we should realize the "Policy Manager" refered by this document,but how? Leave the linux core monitoring the tasks schedule and then trigger "operation point" switch? Thanks and best wishes! |
|
From: rajesh k. <rku...@gm...> - 2006-10-12 15:36:00
|
Hi amit,
I have some problem understanding the explaination about dpm from ur
mail
http://sourceforge.net/mailarchive/forum.php?thread_id=6330950&forum_id=36261
for the question
"What is beginning of the idle task being run by DPM?"
u have said that we don't want calls to quick_idle and full_idle
before entering md_idle
i didn't understand why this shouldn't happen , i also didn't get the
part of setting dpm_startup to NULL
a few lines below this ,( point no 12) , u have said that changing the
operating point frequently causes the system to hang or go into a
infinite while loop.
i want to know why this happens.
Please help me.
|
|
From: Todd P. <tp...@mv...> - 2005-11-30 03:32:06
|
By popular request, a userspace DPM helper library named dpmlib is released, providing functions that access the DPM sysfs interface to create operating points, etc. See http://sourceforge.net/projects/dynamicpower/ file releases if you are interested. -- Todd |
|
From: Todd P. <tp...@mv...> - 2005-10-02 20:44:37
|
The SourceForge Dynamic Power Management project is pleased to add support for the Intel XScale PXA27x "Bulverde" platform on which a number of popular cell phones and PDAs are based. See http://sourceforge.net/projects/dynamicpower Core * Resynced with 2.6.13 kernel.org. * Various cleanups, including: Remove extra dpm_unlock in dpm_quick_idle(); dpm_set_opt return error codes from arch set_opt function. Thanks to Armin Kuster for many of these. * Allow dpm_driver_scale() to be called in atomic contexts. * Update constraints sysfs attributes for changed function prototypes. * Kill dpm_md.init function pointer, not used. OMAP * Resynced with patch-2.6.13-omap1 * Cleanups, remove ARM speed validation code that needed dups of private arch stuff. PXA27x * First release. * Has issues with frequent FCS + DVS (that is, combo CPU frequency and voltage scale), as between idle-task and task states. v=1500 for all operating points, or policies that keep the same operating point at those frequent transitions, can workaround. * Partial pxafb scaling support, needs other patches not upstream yet to work well (such as overlays). * No pcmcia scaling yet. * No camera scaling, but it's pretty much useless on this h/w anyway. -- Todd |
|
From: Todd P. <tp...@mv...> - 2005-08-23 20:17:08
|
Livio Tenze wrote: > You sed me that " the dpm_set_os function which is placed in the idle > loop and scheduler by default": but does the standard implementation > (the standard code) exploit operating states different from "idle" and > "task"? In other words: in my montavista kernel I have 13 states: relock > idle-task idle sleep task-4 task-3 task-2 task-1 task task+1 task+2 > task+3 task+4. Are these states exploited by the linux standard kernel > or new dpm_set_os calls are needed to use all these states? If new calls > are needed: which states are used by default? > When you say that it is necessary to "and add dpm_set_os in the right > place to trigger the transition": do you mean in an external program > (e.g. in a dpm manager), into the applications code (if I am not wrong > these applications are labeled ad power-aware into the documentation I > read) or directly in the kernel code? > Do you know whether a power manager implementation (to manage the > policies) does exist? DPM provides the needed dpm_set_os() calls in the kernel. You can modify your apps to set their own task states as needed, or use the default, or manage task states from an external program. The power policy manager would typically set policies and other chores such as trigger system suspends. I'm not aware of GPL'ed power managers hooked up to DPM. -- Todd |
|
From: Livio T. <liv...@en...> - 2005-08-19 06:22:37
|
Hello, Matthew! Thanks a lot for your help! I would have some questions too. You sed me that " the dpm_set_os function which is placed in the idle loop and scheduler by default": but does the standard implementation (the standard code) exploit operating states different from "idle" and "task"? In other words: in my montavista kernel I have 13 states: relock idle-task idle sleep task-4 task-3 task-2 task-1 task task+1 task+2 task+3 task+4. Are these states exploited by the linux standard kernel or new dpm_set_os calls are needed to use all these states? If new calls are needed: which states are used by default? When you say that it is necessary to "and add dpm_set_os in the right place to trigger the transition": do you mean in an external program (e.g. in a dpm manager), into the applications code (if I am not wrong these applications are labeled ad power-aware into the documentation I read) or directly in the kernel code? Do you know whether a power manager implementation (to manage the policies) does exist? Thanks a lot for your help and for your useful hints!!!! Livius Matthew Locke wrote: > As Todd mentioned, there are a lot of details and the only > documentation is the stuff you've seen. Here is a quick summary > starting from the top: > > Policies are a mapping of operating points to operating states. > Linux changes the operating state using the dpm_set_os function which > is placed in the idle loop and scheduler by default. > When the operating state is changed, dpm will switch to the operating > point that is mapped to that operating state. > It is expected that people will define additional operating states and > add dpm_set_os in the right place to trigger the transition. > Device constraints are not mandatory. > > Note; I left out classes for simplicity. They may or may not be > necessary depending your device. > > On Aug 17, 2005, at 2:39 AM, Livio Tenze wrote: > >> Hi all! >> >> I am new in the DPM field. I would like to find a good document to >> learn the DPM functionalities? May you suggest me? >> I read some articles and part of montavista documentation, but some >> subjects are not clear: in particular I do not understand how the DPM >> changes the operationg point. I am working on a freescale MX21 >> processor: I set the operating point, the operating classes and the >> policies. When I enable a policy, how does the system change the >> operating state? According to which rule? >> Is it mandatory to set the the device constraints? >> >> Thanks to all! >> Livius >> >> >> ------------------------------------------------------- >> SF.Net email is Sponsored by the Better Software Conference & EXPO >> September 19-22, 2005 * San Francisco, CA * Development Lifecycle >> Practices >> Agile & Plan-Driven Development * Managing Projects & Teams * Testing >> & QA >> Security * Process Improvement & Measurement * >> http://www.sqe.com/bsce5sf >> _______________________________________________ >> Dynamicpower-devel mailing list >> Dyn...@li... >> https://lists.sourceforge.net/lists/listinfo/dynamicpower-devel >> > |
|
From: Todd P. <tp...@mv...> - 2005-08-18 19:41:46
|
Livio Tenze wrote: ... > I read some articles and part of montavista documentation, but some > subjects are not clear: in particular I do not understand how the DPM > changes the operationg point. I am working on a freescale MX21 In addition to Matthew's overall explanation (and he's one of the original DPM designers, I'm just a johnny-come-lately), some specifics about your situation. MX21 is not currently a part of the 2.6 patches maintained here at sourceforge. You may have MontaVista's 2.4-based product and may therefore have access to support channels that can help with board specifics. > Is it mandatory to set the the device constraints? Are you asking whether you can get away with ignoring the device constraints and set whatever operating points you want? If so, it depends on the board-specific behavior of the affected devices, and I don't know MX21 well enough to say, but I do know various MX21 drivers place lower bounds on HCLK. Assuming the device constraints are properly set up, the effect of setting clocks sourced by those devices outside the constrained range might be to wedge the device until the next power cycle, or may cause the device to fail to function properly until the clocks are put back into the range needed, or might be undefined. -- Todd |
|
From: Matthew L. <ml...@em...> - 2005-08-18 18:56:02
|
As Todd mentioned, there are a lot of details and the only documentation is the stuff you've seen. Here is a quick summary starting from the top: Policies are a mapping of operating points to operating states. Linux changes the operating state using the dpm_set_os function which is placed in the idle loop and scheduler by default. When the operating state is changed, dpm will switch to the operating point that is mapped to that operating state. It is expected that people will define additional operating states and add dpm_set_os in the right place to trigger the transition. Device constraints are not mandatory. Note; I left out classes for simplicity. They may or may not be necessary depending your device. On Aug 17, 2005, at 2:39 AM, Livio Tenze wrote: > Hi all! > > I am new in the DPM field. I would like to find a good document to > learn the DPM functionalities? May you suggest me? > I read some articles and part of montavista documentation, but some > subjects are not clear: in particular I do not understand how the DPM > changes the operationg point. I am working on a freescale MX21 > processor: I set the operating point, the operating classes and the > policies. When I enable a policy, how does the system change the > operating state? According to which rule? > Is it mandatory to set the the device constraints? > > Thanks to all! > Livius > > > ------------------------------------------------------- > SF.Net email is Sponsored by the Better Software Conference & EXPO > September 19-22, 2005 * San Francisco, CA * Development Lifecycle > Practices > Agile & Plan-Driven Development * Managing Projects & Teams * Testing > & QA > Security * Process Improvement & Measurement * > http://www.sqe.com/bsce5sf > _______________________________________________ > Dynamicpower-devel mailing list > Dyn...@li... > https://lists.sourceforge.net/lists/listinfo/dynamicpower-devel > |
|
From: Todd P. <tp...@mv...> - 2005-08-18 17:47:19
|
Livio Tenze wrote: > Hi all! > > I am new in the DPM field. I would like to find a good document to learn > the DPM functionalities? May you suggest me? > I read some articles and part of montavista documentation, but some > subjects are not clear: in particular I do not understand how the DPM > changes the operationg point. I am working on a freescale MX21 > processor: I set the operating point, the operating classes and the > policies. When I enable a policy, how does the system change the > operating state? According to which rule? > Is it mandatory to set the the device constraints? Hi, it's a complicated subject that I probably do need to explain better, esp. since system PM policy frameworks and interfaces are an active subject of discussion in the community these days. I'll try to get to this soon. -- Todd |
|
From: Livio T. <liv...@en...> - 2005-08-17 09:39:15
|
Hi all! I am new in the DPM field. I would like to find a good document to learn the DPM functionalities? May you suggest me? I read some articles and part of montavista documentation, but some subjects are not clear: in particular I do not understand how the DPM changes the operationg point. I am working on a freescale MX21 processor: I set the operating point, the operating classes and the policies. When I enable a policy, how does the system change the operating state? According to which rule? Is it mandatory to set the the device constraints? Thanks to all! Livius |
|
From: Todd P. <tp...@mv...> - 2005-08-09 18:56:56
|
A new effort to interest the Linux power management communities in a portion of the capabilities offered by Dynamic Power Management (DPM) has begun. Patches to add the concept of "power parameters" (integer valued things that the machine-dependent backend knows how to interpret, clock values, dividers, voltages, etc.) and "operating points" (collections of power parameters with mutually agreeable values) are proposed. The working title of the patches is the "PowerOP" subsystem. PowerOP could be a new layer beneath cpufreq (which manages a "cpu frequency" abstraction that maps onto a subset of the power parameters), but the topic of integrating directly into cpufreq will also be discussed. If nothing else, this would allow for the lowest-level DPM interfaces, and more importantly the machine-dependent code for setting the various power parameters, to gain acceptance in the community and be shared among DPM and cpufreq. If you have any interest in how this shapes up, either to better support hardware that you produce or to better manage the power/performance levels of platforms from your software, then please do join in the discussion on LKML, cpufreq, and linux-pm lists. The threads have "PowerOP" in the title. Thanks, -- Todd |
|
From: Todd P. <tp...@mv...> - 2005-07-02 02:29:22
|
Dynamic Power Management has been released for Linux 2.6.12 with these additional changes: Core: * Speed up processing device constraints and driver scale callbacks by putting these in their own lists (and using notifier interface for callbacks). * Use sched_clock() for stats timings, remove dpm_md_time and friends and standardize on microseconds for timing units. (Thanks to Amit Choudhary for reminder) * Compile fixes for CONFIG_PM=n. OMAP fixes from Vladimir Barinov: * Fix loops_per_jiffie: scale all clock rates to MHz, convert to new clock scheme. * System often hangs if the DPLL freq changed from low to high with large step. But system never hangs if the step is minimal (12MHz) and never hangs if we changing from high to low. Workaround. * Move omap_cfg_reg() to init section. * Add clocking support for 216MHz. The TI OMAP community is adding cpufreq support for OMAP and fixing up various clocking API changes needed for DPM as well, so OMAP DPM can clean up some temporary hacks soon. Intel XScale PXA27x support is coming soon. http://sourceforge.net/projects/dynamicpower -- Todd |
|
From: Todd P. <tp...@mv...> - 2005-05-06 02:43:53
|
Vladimir A. Barinov wrote: > Hi Todd, > > This is a patch for hang after freq scale issue. > System often hangs if the DPLL freq changed from low > to high with large step. But system never hangs if the step > is minimal (12MHz) and never hangs if we changing from high to low. > This patch is a workaround but there is no other way to solve > this issue on linux-omap at this moment. > The delay 0x1FF does care about jump from 12MHz to 194MHz. > > Also: > 1. Removed previous code for omap1710 because > the code described above does care about hangs/garbage produce. > > 2. Proposed to move omap_cfg_reg() to init section to avoid > unnecessary callbacks of this function. > > 3. Added DPM capability for 216MHz to be close with resent clock.c. Thanks Vladimir, merging these to the next version. The magic delay constant 0x1FF is a little unnerving, but oh well. -- Todd |
236 messages has been excluded from this view by a project administrator.
