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

Re: panic in thread_invoke
by Brian Bergstrand 21 Jun '03

21 Jun '03

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On Friday, June 20, 2003, at 02:04 AM, Herry Wiputra wrote: > Hi, > > I am having a problem with my kext keep panicing in thread_invoke(). > > The exact error message is > panic: thread_invoke: preemption_level 1 > > In my code, I have a function which already hold a network funnel, > then enter this while loop in which it will sleep until an event > happen > and spawn a new thread to handle some computation. > > The new thread will also try to grab the network_funnel and will > release it > when it returns. > > My kext kept panicing on the second loop, after it spawns a new thread > and sleep again waiting for an event to happen. > > Does anyone ever comes across this issue before?? > Yes. This panic means that you (the thread) is trying to block while preemption is disabled. This usually means that you are trying to grab mutex, a semaphore, ore some other type of blocking lock while holding a spinlock. In your particular case, I would say you are calling an OS function that blocks from your socket callback function. This is a huge no-no. In your socket callback function, you cannot block at all. This means no memory allocation (unless you use kalloc_noblock() and are willing handle a nil return), no mutexes, semaphores, and no calling any BSD function since that will likely cause a funnel grab attempt (which may block). There are probably more of these cases too. The way I get around this, is to spawn a thread at kext load and have it wait on a semaphore. Then my socket callback queues incoming sockets on a list (protected by a spinlock). I then signal the semaphore and the waiting service thread wakes up and starts pulling sockets off of the queue. So basically about the only things that are safe to use in your socket callback are spinlocks and semaphore signals (but not waits!). If you have any more questions, let me know. HTH. > Thanks for your help. > > Regards, > Herry Wiputra > Brian Bergstrand <http://www.classicalguitar.net/brian/> PGP Key: <http://www.classicalguitar.net/brian/misc/public_key.txt> Egotism is the anesthetic that dulls the pain of stupidity. - Frank Leahy -----BEGIN PGP SIGNATURE----- Version: PGP 8.0.2 iQA/AwUBPvPFCnnR2Fu2x7aiEQIDDwCdFSsVn33qH6AI9MYBm6CdVKLpauQAoMBh luEP4OiCs6SOnh5qchxGcct1 =74qv -----END 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: panic in thread_invoke
by Jim Magee 20 Jun '03

20 Jun '03

On Friday, June 20, 2003, at 12:04 AM, Herry Wiputra wrote: I am having a problem with my kext keep panicing in thread_invoke(). The exact error message is panic: thread_invoke: preemption_level 1 You are either sleeping with preemption disabled explicitly, or implicitly (by holding a real simple lock when you sleep). In my code, I have a function which already hold a network funnel, then enter this while loop in which it will sleep until an event happen and spawn a new thread to handle some computation. The new thread will also try to grab the network_funnel and will release it when it returns. My kext kept panicing on the second loop, after it spawns a new thread and sleep again waiting for an event to happen. Does anyone ever comes across this issue before?? I'd have to see the exact code to provide any more feedback than above. But look for a simple lock that you forget to unlock before sleeping. --Jim _______________________________________________ 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

panic in thread_invoke
by Herry Wiputra 20 Jun '03

20 Jun '03

Hi, I am having a problem with my kext keep panicing in thread_invoke(). The exact error message is panic: thread_invoke: preemption_level 1 In my code, I have a function which already hold a network funnel, then enter this while loop in which it will sleep until an event happen and spawn a new thread to handle some computation. The new thread will also try to grab the network_funnel and will release it when it returns. My kext kept panicing on the second loop, after it spawns a new thread and sleep again waiting for an event to happen. Does anyone ever comes across this issue before?? Thanks for your help. Regards, Herry Wiputra _______________________________________________ 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: clock_delay_for_interval
by Eric Long 19 Jun '03

19 Jun '03

> For some reason, the actual implementation of this routine (on PowerPC > - it doesn't exist on Darwin/x86) is called "delay_for_interval()" > (without the "clock" prefix). I'm not sure whether this was > intentional for some reason, or just an error. But I think that > exposing this routine in the header at all was a mistake. There are > other, more supported ways, of doing this that a preferable. Ok. I noticed after I'd written my last post that the clock_delay_for_interval() prototype was within a #ifdef __APPLE_API_PRIVATE section. I was kind of misled into thinking it was open for use by this url: <http://developer.apple.com/techpubs/macosx/Darwin/General/KernelProgramming /services/chapter_16_section_4.html>. > What kind of KEXT is this? If it's an IOKit KEXT, you should be using > IODelay(). If it is a non-IOKit KEXT, you are probably already > knee-deep in "no-compatibility-guarantees" code. But the routine > delay() is much more likely to be supported long term than the > implementation routine you found and tried. This is just an internal non-IOKit tool. I didn't see a prototype for delay() in the Kernel framework. I see #define DELAY(n) delay(n) in <ppc/param.h> but not the actual prototype for delay(). This works the same as the toolbox Delay() function? Where is the prototype declared? Thanks, Eric _______________________________________________ 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: clock_delay_for_interval
by Jim Magee 18 Jun '03

18 Jun '03

On Tuesday, June 17, 2003, at 11:55 AM, Eric Long wrote: The kext links against the kernel framework, yet when I try to load the kext I get this message: kld(): Undefined symbols: _clock_delay_for_interval I'm running on 10.2.6. The kext loaded fine before I added this call. Anyone know why this symbol is missing? For some reason, the actual implementation of this routine (on PowerPC - it doesn't exist on Darwin/x86) is called "delay_for_interval()" (without the "clock" prefix). I'm not sure whether this was intentional for some reason, or just an error. But I think that exposing this routine in the header at all was a mistake. There are other, more supported ways, of doing this that a preferable. What kind of KEXT is this? If it's an IOKit KEXT, you should be using IODelay(). If it is a non-IOKit KEXT, you are probably already knee-deep in "no-compatibility-guarantees" code. But the routine delay() is much more likely to be supported long term than the implementation routine you found and tried. --Jim PS: Kernel documentation is coming. But only for KEXTs that fit into nice, neat, categories (IOKit, Filesystem, Network, etc..). Anything outside of these categories will, to some extent, still be playing with fire. _______________________________________________ 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: Loading KEXT at boot time
by Jonas Maebe 18 Jun '03

18 Jun '03

On woensdag, jun 18, 2003, at 16:00 Europe/Brussels, Joseph Gabriel Echeverria wrote: I'm having trouble getting a kernel extenstion to load at boot time. It is a non-IOKity kernel extionsion, and I want it to load every time the system boots. I set the OSBundleRequired key to Console, but that didn't work. Does any one have experience with this? Use a startupitem. For an example, look at /System/Library/NetworkExtensions/NetworkExtensions 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

Re: Loading KEXT at boot time
by Stéphane Sudre 18 Jun '03

18 Jun '03

On mercredi, juin 18, 2003, at 16:00 Europe/Paris, Joseph Gabriel Echeverria wrote: I'm having trouble getting a kernel extenstion to load at boot time. It is a non-IOKity kernel extionsion, and I want it to load every time the system boots. I set the OSBundleRequired key to Console, but that didn't work. Does any one have experience with this? IMHO, you need to write a Startup Item. Documentation: file:///Developer/Documentation/Darwin/howto/system_starter_howto/ system_starter_howto.html Samples: Look at items in /System/Library/StartupItems _______________________________________________ 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

Loading KEXT at boot time
by Joseph Gabriel Echeverria 18 Jun '03

18 Jun '03

I'm having trouble getting a kernel extenstion to load at boot time. It is a non-IOKity kernel extionsion, and I want it to load every time the system boots. I set the OSBundleRequired key to Console, but that didn't work. Does any one have experience with this? Thanks, Joey _______________________________________________ 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

clock_delay_for_interval
by Eric Long 17 Jun '03

17 Jun '03

In <kern/clock.h> it lists a prototype for this function: extern void clock_delay_for_interval( uint32_t interval, uint32_t scale_factor); I am attempting to use it in a kext like this: clock_delay_for_interval(200, NSEC_PER_USEC); The kext links against the kernel framework, yet when I try to load the kext I get this message: > kld(): Undefined symbols: > _clock_delay_for_interval I'm running on 10.2.6. The kext loaded fine before I added this call. Anyone know why this symbol is missing? I also noticed these functions: > /* > * Configure the clock system. > */ > extern void clock_config(void); > /* > * Initialize the clock system. > */ > extern void clock_init(void); > > /* > * Initialize the clock ipc service facility. > */ > extern void clock_service_create(void); > Do I need to do anything with these before I call clock_delay_for_interval()? Is there documentation around to explain kernel APIs? I looked in the Kernel Programming docs, but they're pretty sparse. Other than poking around in the sources to figure out these things, it seems like there's very little information around. Maybe I just don't know where to look. Thanks for any input, Eric _______________________________________________ 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

Filesystem questions
by Brian Bergstrand 15 Jun '03

15 Jun '03

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 I have two questions (well really one question and one problem) on kernel filesystems. 1. I'm working on a filesystem that prefers 8K blocks (it's fully 64bit and the indoes are double the size of UFS inodes). I noticed the following comment in the UFS code: /* * Buffer cache does not handle multiple pages in a buf when * invalidating incore buffer in pageout. There are no locks * in the pageout path. So there is a danger of loosing data when * block allocation happens at the same time a pageout of buddy * page occurs. incore() returns buf with both * pages, this leads vnode-pageout to incorrectly flush of entire. * buf. Till the low level ffs code is modified to deal with these * do not mount any FS more than 4K size. */ /* * Can't mount filesystems with a fragment size less than DIRBLKSIZ */ /* * Don't mount dirty filesystems, except for the root filesystem */ if ((fs->fs_bsize > PAGE_SIZE) || (fs->fs_fsize < DIRBLKSIZ) || ((!(mp->mnt_flag & MNT_ROOTFS)) && (!fs->fs_clean))) { Is this still true for the Jaguar (and beyond) buffer cache? What can I do to work around the problem? 2. I've ported the Ext2 filesystem to Mac OS X. Current there is a bug open <https://sourceforge.net/tracker/ index.php?func=detail&aid=742939&group_id=64713&atid=508406> that is due to a problem copying RSRC forks. Specifically, the Apple Double header for the Carbon FS layer RSRC file is being written as all 0's. I've tracked this down to something in between my VOP_WRITE and VOP_STRATEGY calls. The rsrc file is written as an 82 byte buffer containing the Apple Double header followed by bigger chunks of the actual data. For some reason, the 82 byte header is being written as 0's (but the data itself is fine). I can't figure out what the problem would be; I can copy files smaller than 82 bytes with no problem. I can also copy an existing RSRC fork (from a UFS FS) using cp with no problem. I'm beginning to think it may be a problem with the Carbon FS layer, but writing to UFS works fine, so I don't know. Does anyone have any clues? The code is available on Sourceforge <http://cvs.sourceforge.net/cgi-bin/viewcvs.cgi/ext2fsx/src/gnu/ext2fs/ >. TIA. Brian Bergstrand <http://www.classicalguitar.net/brian/> PGP Key: <http://www.classicalguitar.net/brian/misc/public_key.txt> Oh, I don't blame Congress. If I had $600 billion at my disposal, I'd be irresponsible, too. - Lichty & Wagner -----BEGIN PGP SIGNATURE----- Version: PGP 8.0.2 iQA/AwUBPuy2C3nR2Fu2x7aiEQLcOACg4riUrolD97eocx5t5Y80ar6iJa8An16S 0CjnBW5E8iH0S9QuxT1o4C91 =ySpQ -----END 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