November 2002 - Darwin-kernel

Performance problems under 10.2
by Tyler Ward 09 Nov '02

09 Nov '02

This is a new thread, but here goes..... Has anyone noticed strange performance problems under 10.2? I have noticed that things which were once fast now use inordinate amounts of system time. For instance, setiathome (the unix command line version) uses almost 25% system time, whereas before it used only a few percent system time. Monitoring it with "top -es 3" shows that it is performing almost 1,000 context switches per second. For instance, here's the setiathome line from my top output.... 3757 setiathome 49.9% 36:17:02 21150597 3129 34 306551 306542 38928334 306576 30614876 Thirty million context switches, and 38 million syscalls. I'm pretty sure this didn't happen before, and it certainly didn't use 20% or more system time. I've also noticed that setiathome noticeably slows my computer while it is running alongside some other relatively intensive programs, whereas it didn't do so before. Furthermore, it seems that the whole system is just less responsive when one program starts misbehaving (this is very subjective, I know) possibly because too much time is being spent in the kernel, thus preventing really fine grained multitasking. Does anyone know what's up with this? I don't think I'm imagining this, what has changed? Tyler Ward tjw19(a)columbia.edu _______________________________________________ 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: Missing Kernel Header Files
by Justin C. Walker 08 Nov '02

08 Nov '02

You can find a discussion of this fascinating subject in the archives for this list or the darwin-development list. Short answer: get the files you want from the CVS repository, making sure you check out with a tag that matches your kernel [uname -a; "xnu" -> "Apple"]. Regards, Justin On Thursday, November 7, 2002, at 02:16 PM, Michael Cashwell wrote: Greetings, I've got an NKE using a PF_SYSTEM / SYSPROTO_EVENT socket for network kernel events. I'm interested in both KEV_DL_SUBCLASS and KEV_INET_SUBCLASS. (My goal is to be notified of network interfaces coming and going and media and IP address changes.) The former is found in net/if.h but the latter is in the xnu sources in netinet/in_var.h which doesn't exist in the version A kernel headers. The same thing seems to be true for KEV_INET6_SUBCLASS in netinet6/in6_var.h and KEV_ATALK_SUBCLASS in netat/at_var.h. These headers also define matching event structs which I need to interpret the event messages, so it's more than just simple defines that are missing. Were these files deliberately left out of the framework headers? Should I not be using them? If not then I'm faced with either copying them from xnu into the framework headers (a pain to have to do on each development system) or copy them in as a source file in my project (a risk that a future update will add them to the framework and I'll then be using old ones). Neither is ideal. What have others done in this situation? Also, if it was an oversight is a bug report the correct way to get them added to the kernel framework in a future release? Thanks, Mike _______________________________________________ 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. -- Justin C. Walker, Curmudgeon-At-Large * Institute for General Semantics | It's not whether you win or lose... | It's whether *I* win or lose. *--------------------------------------*-------------------------------* _______________________________________________ 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: allocate huge amounts of kernel memory
by Heiko Panther 08 Nov '02

08 Nov '02

Hi Umesh, > IOMalloc() is just a wrapper around kalloc(). kalloc() allocates > memory from a sub map which is limited to 16MB. Out of curiosity: Why do you use that sub map, and don't allocate straight from kernel_map? (Which is, by the way, my workaround.) Heiko Panther _______________________________________________ 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

Missing Kernel Header Files
by Michael Cashwell 07 Nov '02

07 Nov '02

Greetings, I've got an NKE using a PF_SYSTEM / SYSPROTO_EVENT socket for network kernel events. I'm interested in both KEV_DL_SUBCLASS and KEV_INET_SUBCLASS. (My goal is to be notified of network interfaces coming and going and media and IP address changes.) The former is found in net/if.h but the latter is in the xnu sources in netinet/in_var.h which doesn't exist in the version A kernel headers. The same thing seems to be true for KEV_INET6_SUBCLASS in netinet6/in6_var.h and KEV_ATALK_SUBCLASS in netat/at_var.h. These headers also define matching event structs which I need to interpret the event messages, so it's more than just simple defines that are missing. Were these files deliberately left out of the framework headers? Should I not be using them? If not then I'm faced with either copying them from xnu into the framework headers (a pain to have to do on each development system) or copy them in as a source file in my project (a risk that a future update will add them to the framework and I'll then be using old ones). Neither is ideal. What have others done in this situation? Also, if it was an oversight is a bug report the correct way to get them added to the kernel framework in a future release? Thanks, Mike _______________________________________________ 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

Atomic functions problem
by Christian Pasin 07 Nov '02

07 Nov '02

I'm trying to use an atomic function in a NDRV driver (for example AddAtomic), but these functions seem not to work correctly . My code is like the following: SInt32 gMyInterruptDepth ; gMyInterruptDepth=0x00000010; DebugValue(AddAtomic(1,&gMyInterruptDepth)); DebugValue(gMyInterruptDepth); (Note DebugValue print a value on system.log) I obtain these results: ....FFFFFFCB (the value returned by AddAtomic) ....10 (the value of my variable is not changed) I used also IncrementAtomic8 but the results were the same. I'm compiling and executing the code on a MacOsX 10.1.5 with a G4 processor, but I tried also on a Jaguar version with dual processor. Why don't atomic functions work correctly? Thanks Christian christian(a)villagetronic.com _______________________________________________ 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: allocate huge amounts of kernel memory
by Umesh Vaishampayan 05 Nov '02

05 Nov '02

On Monday, November 4, 2002, at 07:54 AM, Heiko Panther wrote: > Hi Umesh, > > >> my IOKit driver wants to allocate really a lot of buffers with > >> IOMalloc(). At about 4 MB of alloc'd memory, further calls to > >> IOMalloc() fail. Is there some kind of limit defined? How can I > >> circumvent this problem? > > > Hope you realize that's 4MB of kernel *wired* memory.... > > Of course. > For my testing purposes, I can afford wasting some wired memory. Could > you tell me why it fails, and how I can get more of that precious wired > memory? > > Thank you- > Heiko Panther UTSL. IOMalloc() is just a wrapper around kalloc(). kalloc() allocates memory from a sub map which is limited to 16MB. May be in your case other kalloc() consumers already gobbled up 12MB... Hence you get a failure at around 4MB. --Umesh -- Umesh Vaishampayan Apple Computer, Inc. Mac OS X Kernel Ph: (408) 974 0229 _______________________________________________ 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: Address Space Limit - 2.25GB Per Process?
by Umesh Vaishampayan 05 Nov '02

05 Nov '02

On Saturday, November 2, 2002, at 05:53 PM, Jeremy T. Fox wrote: > > Each user process can have 4GB - 4K of VM address space. However you > do > > not have a larger than 1.5GB "hole" in that address space. If g77 > must > > work in "contiguous VM", I think 1.5GB is the limit you will have to > > work in. > > From the other responses, it looks like the hole is 2.25GB, at least > in Jaguar. So the 4GB of memory address space is most likely only > going to be used by people working on multiple documents in one > program, versus people working on one big document, which presumably > need to be in contiguous memory, I would think. Only if the program does not do "map needed portions of the file" and insisted on mapping the whole file at once.... --Umesh -- Umesh Vaishampayan Apple Computer, Inc. Mac OS X Kernel Ph: (408) 974 0229 _______________________________________________ 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: allocate huge amounts of kernel memory
by Heiko Panther 04 Nov '02

04 Nov '02

Hi Umesh, >> my IOKit driver wants to allocate really a lot of buffers with >> IOMalloc(). At about 4 MB of alloc'd memory, further calls to >> IOMalloc() fail. Is there some kind of limit defined? How can I >> circumvent this problem? > Hope you realize that's 4MB of kernel *wired* memory.... Of course. For my testing purposes, I can afford wasting some wired memory. Could you tell me why it fails, and how I can get more of that precious wired memory? Thank you- Heiko Panther _______________________________________________ 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: Address Space Limit - 2.25GB Per Process?
by Jeremy T. Fox 03 Nov '02

03 Nov '02

> Each user process can have 4GB - 4K of VM address space. However you do > not have a larger than 1.5GB "hole" in that address space. If g77 must > work in "contiguous VM", I think 1.5GB is the limit you will have to > work in. >From the other responses, it looks like the hole is 2.25GB, at least in Jaguar. So the 4GB of memory address space is most likely only going to be used by people working on multiple documents in one program, versus people working on one big document, which presumably need to be in contiguous memory, I would think. Hopefully this contiguous hole limitation will be eliminated if Apple moves to a 64-bit address space. Jeremy _______________________________________________ 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: allocate huge amounts of kernel memory
by Umesh Vaishampayan 01 Nov '02

01 Nov '02

On Friday, November 1, 2002, at 06:40 AM, Heiko Panther wrote: > Hi there, > > my IOKit driver wants to allocate really a lot of buffers with > IOMalloc(). At about 4 MB of alloc'd memory, further calls to > IOMalloc() fail. Is there some kind of limit defined? How can I > circumvent this problem? > > Thanks for pointers, > > Heiko Panther > Hope you realize that's 4MB of kernel *wired* memory.... --Umesh -- Umesh Vaishampayan Apple Computer, Inc. Mac OS X Kernel Ph: (408) 974 0229 _______________________________________________ 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