April 2003 - Darwin-kernel - lists.apple.com

Re: unkillable process ?
by Jonas Maebe 04 Apr '03

04 Apr '03

On vrijdag, apr 4, 2003, at 03:35 Europe/Brussels, jmagee(a)apple.com wrote: That "U" means that the process (or more appropriately, one of the threads in the process) is blocked in the kernel in an "uninterruptible" state. It is unacceptable for the kernel/drivers to so mark a thread and then non make forward progress on it. Do you have any non-standard kernel extensions? What system call was thread making when it got into this state? You can find the latter by running the sample tool on the process (or attaching with gdb). When I have a problem like this, most of the time the process is blocked in statfs. Yesterday, I had this problem with MPlayer when I first added a file to its playlist, quit the application, restarted it and then clicked again on that file in the playlist. kill -9 seems to work only sporadically in such a case (it seems that the kernel blocks a long time, then returns and the program immediately tries again, so the kill only works if you accidentally do it right at a moment the program isn't blocked in the kernel - it seems to signals aren't queued). However, attaching with gdb does seem to work consistently, it seems the ptrace syscall does get queued somehow, so gdb attaches the first time the system call ends. After you attach, you can do a kill. FWIW, here's the sample I did yesterday (on MPlayer, I hope it doesn't wrap too badly): *** 500 Thread_0e03 500 start 500 _start 500 main 500 NSApplicationMain 500 -[NSApplication run] 500 -[NSApplication nextEventMatchingMask:untilDate:inMode:dequeue:] 500 _DPSNextEvent 500 BlockUntilNextEventMatchingListInMode 500 ReceiveNextEventCommon 500 RunCurrentEventLoopInMode 500 CFRunLoopRunSpecific 500 __CFRunLoopRun 500 __CFRunLoopDoSource1 500 __CFMachPortPerform 500 VNDiskArbHandleMessage 500 ServerToClient_server 500 _XDiskArbDiskAppearedWithMountpoint_rpc 500 DiskArbDiskAppearedWithMountpoint_rpc 500 DiskMountCallback(char*, unsigned, char*) 500 VNDiskNotification(int, char*, unsigned, char*, int) 500 MountVolume(char const*, statfs*, unsigned char) 500 statfs 500 statfs [STACK TOP] *** I'm not sure why those mount calls are in there, because I didn't mount a remote volume on this machine. I did mount a volume from this machine on another one - namely the volume on which the file I had deleted resided. I don't know whether that had anything to do with it. Anyway, in this state the program couldn't be killed using kill -9 or with the force quit command (unless you do it all the time and get lucky). Jonas _______________________________________________ darwin-kernel mailing list | darwin-kernel(a)lists.apple.com Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/darwin-kernel Do not post admin requests to the list. They will be ignored.

1 0

Attaching to /dev/kmem with gdb
by Brian Tabone 04 Apr '03

04 Apr '03

List, I have heard of old school UNIX hacks attaching to live kernels with gdb, allowing them to see values of variables and changing those values on the fly. I am curious as to how one could do this on Darwin. I attempted to attach to /dev/kmem via gdb, as root. All I would get back was a complaint that /dev/kmem could not be found (I looked and its there all-right). What I can gather is that /dev/kmem is a character special file, and is not visible to gdb in a direct fashion. I used the following gdb -s /mach.sym /dev/kmem Now I know you can't attach directly to a running kernel as you would a program, since halting the kernel would halt you. But, from what I understand you can attach to the running image to examine the memory, without halting that image. Perhaps I am mistaken? -Sincerely Brian Tabone _______________________________________________ darwin-kernel mailing list | darwin-kernel(a)lists.apple.com Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/darwin-kernel Do not post admin requests to the list. They will be ignored.

1 0

Re: Non-executable stack
by Brian Tabone 04 Apr '03

04 Apr '03

If I remember correctly, part of the reason that the stack grows down from the top is to make it easier for both the stack and heap to grow. Since both stack and heap can grow, it would be difficult to have them be placed back to back in virtual memory. To do that, you would have to make all addresses in either the stack or heap relative to some movable starting point. That would be an address translation headache. Instead, both have their starting points bound by fixed address regions. The heap grows up, the stack grows down and they can grow until they collide, which will occur when you have exhausted your VM space. -Brian Tabone On Thursday, April 3, 2003, at 12:08 AM, Robert Plantz wrote: On Wednesday, April 2, 2003, at 07:10 PM, Justin Walker wrote: tisdagen den 1 april 2003 kl 05.54 skrev Jim Magee: Out of curiosity, I've always wondered why the ABI grows the stack downwards. It seems to be just as easy to grow it upwards, which would make buffer overflows much harder, if not impossible. Habit, mostly, I think. Most of this stuff grew up on PDP11's, and the PDP11 architecture gave downward growing stacks (among other things, interrupts and subroutine calls predecremented stack pointers; and the register specifiers in instructions made it easy to 'comply' with that mode of operation). With modernXXXmore recent architectures like PowerPC where a stack is more ABI than ISA, you can, as it were, go either way (:-}). Perhaps I'm being too simplistic, but if I wanted to make maximum use of memory, I would store my program code at one end and start my stack at the other, then let them grow toward each other. --Bob _______________________________________________ darwin-kernel mailing list | darwin-kernel(a)lists.apple.com Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/darwin-kernel Do not post admin requests to the list. They will be ignored. _______________________________________________ darwin-kernel mailing list | darwin-kernel(a)lists.apple.com Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/darwin-kernel Do not post admin requests to the list. They will be ignored.

1 0

Re: unkillable process ?
by jmagee＠apple.com 04 Apr '03

04 Apr '03

On Thursday, April 3, 2003, at 12:12PM, Pejvan BEIGUI wrote: I've got a real and weird problem, which seems to come from the kernel, and hence I haven't been able to find the answer anywhere else, i'm asking it here: How can anyone explain this: > [2:10:22pm] _pejvan_ ~ >> kill -9 1026 > [2:21:32pm] _pejvan_ ~ >> kill -9 1026 > [2:21:33pm] _pejvan_ ~ >> kill -9 1026 > [2:21:33pm] _pejvan_ ~ >> kill -9 1026 > [2:21:34pm] _pejvan_ ~ >> sudo kill -9 1026 > Password: > [2:21:42pm] _pejvan_ ~ >> sudo kill -9 1026 > [2:21:44pm] _pejvan_ ~ >> sudo kill -9 1026 > [2:21:45pm] _pejvan_ ~ >> ps -U pejvan | grep 1026 > 1026 ?? Us 0:00.02 /Users/pejvan/devel/[...] That "U" means that the process (or more appropriately, one of the threads in the process) is blocked in the kernel in an "uninterruptible" state. It is unacceptable for the kernel/drivers to so mark a thread and then non make forward progress on it. Do you have any non-standard kernel extensions? What system call was thread making when it got into this state? You can find the latter by running the sample tool on the process (or attaching with gdb). I really need to be able to kill this process, since it totally blocks the launching of any graphical app when it crashes, and there's no way for me to restart the computer in clean way: $ sudo shutdown now will fail and after two or three minutes, I get in the text mode output, with a message telling me that there have been some IPC failure, which leads me to the question: does the process crash and blocks some stuff? (which is totally wrong on a preemptive OS) does the IPC mechanism fail somewhere and make this process crash? (which is weirder) What you asked for [with that shutdown command] was not to reboot the machine, but instead to bring it back down to single-user-mode. Two things of note about that: 1. the un-interruptible process will still be blocked, more than likely, unless the shutdown killed off the facility that it was blocked waiting on (not likely). 2. Mac OS X/Darwin does not like being brought down to single-user mode after it was already running in multi-user-mode. Many of the key system services are implemented in user space. Bringing the system down to single-user-mode killed those off. The IPC messages you see are because you killed off the mach_init process that manages the namespace of public mach ports. But you also killed off the dynamic pager, which manages your swap files, and it doesn't like to get restarted either. You probably wanted "shutdown -r now" where the "-r" is to force a reboot. _______________________________________________ darwin-kernel mailing list | darwin-kernel(a)lists.apple.com Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/darwin-kernel Do not post admin requests to the list. They will be ignored.

1 0

unkillable process ?
by Pejvan BEIGUI 03 Apr '03

03 Apr '03

Hi everybody, I've got a real and weird problem, which seems to come from the kernel, and hence I haven't been able to find the answer anywhere else, i'm asking it here: How can anyone explain this: > [2:10:22pm] _pejvan_ ~ >> kill -9 1026 > [2:21:32pm] _pejvan_ ~ >> kill -9 1026 > [2:21:33pm] _pejvan_ ~ >> kill -9 1026 > [2:21:33pm] _pejvan_ ~ >> kill -9 1026 > [2:21:34pm] _pejvan_ ~ >> sudo kill -9 1026 > Password: > [2:21:42pm] _pejvan_ ~ >> sudo kill -9 1026 > [2:21:44pm] _pejvan_ ~ >> sudo kill -9 1026 > [2:21:45pm] _pejvan_ ~ >> ps -U pejvan | grep 1026 > 1026 ?? Us 0:00.02 /Users/pejvan/devel/[...] I really need to be able to kill this process, since it totally blocks the launching of any graphical app when it crashes, and there's no way for me to restart the computer in clean way: $ sudo shutdown now will fail and after two or three minutes, I get in the text mode output, with a message telling me that there have been some IPC failure, which leads me to the question: does the process crash and blocks some stuff? (which is totally wrong on a preemptive OS) does the IPC mechanism fail somewhere and make this process crash? (which is weirder) Anyway, since the kernel isn't able to kill it, it seems that this is some kind of bug in the OS. Am I wrong? any suggestions suggestion about killing this process? is there any related bug reported about this? Thanks for your help, Pejvan PS: this process is launched by a normal user, and hence run in user land. _______________________________________________ darwin-kernel mailing list | darwin-kernel(a)lists.apple.com Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/darwin-kernel Do not post admin requests to the list. They will be ignored.

1 0

Re: Non-executable stack
by Justin Walker 03 Apr '03

03 Apr '03

On Wednesday, Apr 2, 2003, at 22:08 US/Pacific, Robert Plantz wrote: On Wednesday, April 2, 2003, at 07:10 PM, Justin Walker wrote: tisdagen den 1 april 2003 kl 05.54 skrev Jim Magee: Out of curiosity, I've always wondered why the ABI grows the stack downwards. It seems to be just as easy to grow it upwards, which would make buffer overflows much harder, if not impossible. Habit, mostly, I think. Most of this stuff grew up on PDP11's, and the PDP11 architecture gave downward growing stacks (among other things, interrupts and subroutine calls predecremented stack pointers; and the register specifiers in instructions made it easy to 'comply' with that mode of operation). With modernXXXmore recent architectures like PowerPC where a stack is more ABI than ISA, you can, as it were, go either way (:-}). Perhaps I'm being too simplistic, but if I wanted to make maximum use of memory, I would store my program code at one end and start my stack at the other, then let them grow toward each other. That is too simplistic. The main issue is that large chunks of address space are "reserved" for shared libraries and shared data segemnts. In addition, your private data segment has to go somewhere as well. There are a lot of claims on that single, linear, array of memory. Regards, Justin -- Justin C. Walker, Curmudgeon-At-Large * Institute for General Semantics | When LuteFisk is outlawed | Only outlaws will have | LuteFisk *--------------------------------------*-------------------------------* _______________________________________________ darwin-kernel mailing list | darwin-kernel(a)lists.apple.com Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/darwin-kernel Do not post admin requests to the list. They will be ignored.

1 0

Re: Non-executable stack
by David Blanton 03 Apr '03

03 Apr '03

On 4/2/03 11:08 PM, "Robert Plantz" <plantz(a)SONOMA.EDU> wrote: > On Wednesday, April 2, 2003, at 07:10 PM, Justin Walker wrote: > >> >>> tisdagen den 1 april 2003 kl 05.54 skrev Jim Magee: >> >>> Out of curiosity, I've always wondered why the ABI grows the stack >>> downwards. It seems to be just as easy to grow it upwards, which >>> would make buffer overflows much harder, if not impossible. >> >> Habit, mostly, I think. Most of this stuff grew up on PDP11's, and >> the PDP11 architecture gave downward growing stacks (among other >> things, interrupts and subroutine calls predecremented stack pointers; >> and the register specifiers in instructions made it easy to 'comply' >> with that mode of operation). With modernXXXmore recent architectures >> like PowerPC where a stack is more ABI than ISA, you can, as it were, >> go either way (:-}). > > Perhaps I'm being too simplistic, but if I wanted to make maximum > use of memory, I would store my program code at one end and > start my stack at the other, then let them grow toward each > other. > > --Bob Interesting, that is how Mac OS use to do it. Heap at one end, stack at the other and then sniff out for collisions. David Blanton "Cocoa FNG" _______________________________________________ darwin-kernel mailing list | darwin-kernel(a)lists.apple.com Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/darwin-kernel Do not post admin requests to the list. They will be ignored.

1 0

Re: CPU scheduling
by Markus Hitter 03 Apr '03

03 Apr '03

Am Donnerstag, 03.04.03 um 11:39 Uhr schrieb Srinidhi Rao: I am working on a project which needs 100% CPU time when it is running. With some luck, you can get 99.5 % of the CPU. For lengthy computations, there is no need to hog the CPU since you get all available computing time anyway. For realtime needs, search for realtime threads, often used in audio/video applications; e.g. the CoreAudio mailing list archives. Also, what is the timeslice allocated to a process in normal condition? It's one of the first messages you see when booting in verbose mode (hold down Cmd-V at startup). On all versions of OS X I've seen it so far, the quantum is 10 ms. One way I could think is to have a Real time thread in my application. But will this assure me continuous CPU time? No. Or better: just for a few seconds. Even realtime threads are kicked if they hog the CPU for more than some seconds. Note that hogging the CPU renders a computer completely unresponsive exept for a force-reboot. (Source: discussions on CoreAudio) Still can't imagine which type of application whould need _all_ CPU. What are you doing if you run on a slightly slower system? If you want to control a robot, I'd write a kernel extension doing the I/O with some emergency stop functionality. For feeding the kext I'd use a realtime thread inside a normal userland application. (guessing based on your signature). HTH, Markus - - - - - - - - - - - - - - - - - - - Dipl. Ing. Markus Hitter http://www.jump-ing.de/ _______________________________________________ darwin-kernel mailing list | darwin-kernel(a)lists.apple.com Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/darwin-kernel Do not post admin requests to the list. They will be ignored.

1 0

Re: Disk idling and energy saving in darwin
by Felix Kronlage 03 Apr '03

03 Apr '03

On Wed, Apr 02, 2003 at 03:14:28PM -0500, Joey Echeverria wrote: > This may not be quite the right list for this, darwin-userlevel@ would have been the correct list. > but I was wondering if someone could tell me how to set a hard disk to > idle after a timeout in darwin. I can do this in OS X easily using System Preferences, but I > can't figure out how to do the same energy saving things in darwin. you are looking for 'pmset'. The manpage of pmset(1) will help you. -fkr -- gpg-fingerprint: 076E 1E87 3E05 1C7F B1A0 8A48 0D31 9BD3 D9AC 74D0 |Felix Kronlage | whois -h whois.ripe.net FKR-RIPE | |http://www.hazardous.org/fkr/ | fkr(a){grummel.net|opendarwin.org} | [demime 0.98b removed an attachment of type application/pgp-signature] _______________________________________________ darwin-kernel mailing list | darwin-kernel(a)lists.apple.com Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/darwin-kernel Do not post admin requests to the list. They will be ignored.

1 0

Re: Disk idling and energy saving in darwin
by Jim Riggs 03 Apr '03

03 Apr '03

This may not be quite the right list for this, but I was wondering if someone could tell me how to set a hard disk to idle after a timeout in darwin. I can do this in OS X easily using System Preferences, but I can't figure out how to do the same energy saving things in darwin. Ideally I would be able to set a specific timeout after which to turn of the disk, backlight, etc. Any help would be greatly appreciated. These settings are stored in the System Configuration. Take a look at the System Configuration plists in /var/db/SystemConfiguration. Specifically, the file you need is /var/db/SystemConfiguration/com.apple.PowerManagement.xml If it does not exist, you can create it by hand or copy it from an OS X box. Below is a copy of mine that I copied from OS X and set the values as needed (beware of e-mail client line breaks). My disk is set to not sleep (0), but you can change that to whatever is appropriate for your setup. <?xml version="1.0" encoding="UTF-8"?> <!DOCTYPE plist PUBLIC "-//Apple Computer//DTD PLIST 1.0//EN" "http://www.apple. com/DTDs/PropertyList-1.0.dtd"> <plist version="1.0"> <dict> <key>AC Power</key> <dict> <key>Automatic Restart On Power Loss</key> <integer>1</integer> <key>Disk Sleep Timer</key> <integer>0</integer> <key>Display Sleep Timer</key> <integer>15</integer> <key>System Sleep Timer</key> <integer>0</integer> <key>Wake On LAN</key> <integer>1</integer> <key>Wake On Modem Ring</key> <integer>0</integer> </dict> </dict> </plist> _______________________________________________ darwin-kernel mailing list | darwin-kernel(a)lists.apple.com Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/darwin-kernel Do not post admin requests to the list. They will be ignored.

1 0