January 2004 - Darwin-kernel

Long latencies in realtime threads when doing BSD syscalls -- Bug or expected behaviour?
by Mario Kleiner 30 Jan '04

30 Jan '04

Hello, i'm new to this list, so please apologize if i post to the wrong list, but i couldn't find a clear answer in the list archives. We have written a simple C-timing loop, which is basically: while(true) { some BSD-Syscall like setjmp, getpid, getXXX, fopen, fprintf, fwrite, ... ; usleep(500 milliseconds) } to test realtime behaviour/latencies of OS-X. The loop runs under THREAD TIME CONSTRAINT priority. When we run it for e.g. 120 seconds, most loop passes have a duration of less than 1 ms (usually around 0.6ms), even under a loaded system -- which is good. But every thirty seconds, the sync() syscall of the update daemon triggers spikes in loop duration of up to 14ms in our realtime thread, although it is supposed to preempt other processes when it becomes runnable. To me, it seems that any BSD syscall (especially ones that do I/O) in any (timesharing) thread can produce high latencies in parallel running realtime threads, by delaying their BSD syscalls. If we remove the syscalls from the timing loop, it stays unaffected, so preemption works. >From everything i read on the web and this lists, my assumption is that the cause of this is the kernel_funnel, which seems to prevent concurrent execution of BSD calls in the kernel, but i would like to have some confirmation for this from somebody knowledgeable: So, am i right that the kernel_funnel has this effect on all BSD syscalls, even on uniprocessor machines? Does this mean, that the kernel funnel is automatically released if a thread sleeps in the kernel, but that it is not released, if a thread gets preempted by a realtime thread -- which would cause the realtime thread to get delayed/blocked on the lock, as soon as it does a BSD syscall -- basically some kind of priority inversion? That would make perfect sense to me, but why doesn't the same thing happen to the gettimeofday() syscall that i use for timing and the usleep() when it gets back from sleep? Is there a way to avoid this, e.g. by only using Apples Frameworks or Mach API's instead of Posix stuff? Our current -- ugly -- solution is to kill the update() process during realtime parts of the application. Is there something better, e.g. the equivalent of SCHED_FIFO scheduling on other Unices? And a last question: Is there an equivalent to mlockall() to prevent paging for a realtime app? Sorry for the long post. Any comments would be appreciated. ciao, -mario _______________________________________________ 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

USB DeviceRequest
by Joe Maloof 27 Jan '04

27 Jan '04

Is it safe to call USB device manager synchronously? Can I use IOCommandGate to issue IOUSBDevice_>DeviceRequest? When I tried to send a device request in my IOCommandGate Action, the request will be implemented but I don't get an IOReturn and the driver will hang for ever...Any suggestions will be appreciated!!!! _______________________________________________ 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: System V message queues
by Alex Scotti 24 Jan '04

24 Jan '04

Thanks for your help. The code is at http://www.cameronpond.org/~alex/sysvmsg.tar.gz with the working sysctl stuff now. -Alex On Jan 22, 2004, at 2:49 PM, Justin Walker wrote: On Thursday, January 22, 2004, at 10:55 AM, Justin Walker wrote: On Thursday, January 22, 2004, at 10:12 AM, Alex Scotti wrote: I was trying to port some code over to darwin which needed sysv msgqs. I "ported" over the code in bsd/kern/sysv_msg.c to be a kernel extension. Is anyone interested in this? I don't have issues like adding the necessary sysctl variables worked out yet (tunables are hardcoded for now) - can this even be done in a kext? Responding only to the last question: check IPFirewalls for an indication of how to add and remove sysctl's dynamically. To be just a tad more clear: It's 'IPFirewall', and it's a project available on the darwin (http://developer.apple.com/darwin) website. Follow the links to the source for Darwin. You can also find this on the OpenDarwin site (http://www.opendarwin.org) 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. _______________________________________________ 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: System V message queues
by Justin Walker 22 Jan '04

22 Jan '04

On Thursday, January 22, 2004, at 10:55 AM, Justin Walker wrote: On Thursday, January 22, 2004, at 10:12 AM, Alex Scotti wrote: I was trying to port some code over to darwin which needed sysv msgqs. I "ported" over the code in bsd/kern/sysv_msg.c to be a kernel extension. Is anyone interested in this? I don't have issues like adding the necessary sysctl variables worked out yet (tunables are hardcoded for now) - can this even be done in a kext? Responding only to the last question: check IPFirewalls for an indication of how to add and remove sysctl's dynamically. To be just a tad more clear: It's 'IPFirewall', and it's a project available on the darwin (http://developer.apple.com/darwin) website. Follow the links to the source for Darwin. You can also find this on the OpenDarwin site (http://www.opendarwin.org) 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: System V message queues
by Justin Walker 22 Jan '04

22 Jan '04

On Thursday, January 22, 2004, at 10:12 AM, Alex Scotti wrote: I was trying to port some code over to darwin which needed sysv msgqs. I "ported" over the code in bsd/kern/sysv_msg.c to be a kernel extension. Is anyone interested in this? I don't have issues like adding the necessary sysctl variables worked out yet (tunables are hardcoded for now) - can this even be done in a kext? Responding only to the last question: check IPFirewalls for an indication of how to add and remove sysctl's dynamically. Regards, Justin -- 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

System V message queues
by Alex Scotti 22 Jan '04

22 Jan '04

I was trying to port some code over to darwin which needed sysv msgqs. I "ported" over the code in bsd/kern/sysv_msg.c to be a kernel extension. Is anyone interested in this? I don't have issues like adding the necessary sysctl variables worked out yet (tunables are hardcoded for now) - can this even be done in a kext? -Alex Scotti _______________________________________________ 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: sosend, soreceive, soconnect
by Timothy Standing 22 Jan '04

22 Jan '04

Jim - I spent several weeks last spring tripping over bugs in the KUNC code so I am very glad to hear that there is an alternate mechanism recommended by Apple. So given the fact that I need to launch an executable every time my driver loads, how do you recommend I do this? Can you give me a url to documentation or the name of a project in Darwin which does it correctly? (The KUNC mechanism is very well documented which is probably why a lot of people are using it.) The way I ended up dealing with the "wing and a prayer" aspect of KUNCExecute was to keep calling it on my daemon until it sent my driver a tickle. This isn't very elegant but it works. Thanks in advance for your response. Tim Standing SoftRAID, LLC On Jan 21, 2004, at 10:00 PM, darwin-kernel-request(a)lists.apple.com wrote: On Jan 20, 2004, at 5:27 AM, ollie wrote: You could have your kext either launch the daemon using KUNCExecute or else have it send a kernel event when ever it requires action from your daemon. This will remove all polling from your current design. Please, do NOT use KUNCExecute(). It's a "wing and a prayer" mechanism (there are many situations where it will fail silently) only added for people who were too stubborn to switch to the "driver down from the daemon, not the out from the kernel" model of execution under Mac OS X/Darwin. Each release, the timing of system boot changes, and therefore the timing of when KUNCExecute() will succeed also changes. There are several mechanism available to register the daemon from user-space for launching only when needed. These are much more reliable than the KUNCExecute() mechanism, as the user-level dependencies are always satisfied before the daemon can be registered in the first place. --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

Re: question about mach ports and fork
by Mike Smith 22 Jan '04

22 Jan '04

On Jan 21, 2004, at 3:37 PM, Nathan T. Hjelm wrote: about IOService, however. You can't just open a new IOKit service connection in the child? This is sometimes impossible because the parent was privileged and the child not (so I feel your pain). I have tried to open a new connection to iokit with the child process by calling IOCreateMasterPort again but I get the same result as when I didn't. What do you mean by this? In your forked child, you have to obtain the new master port, then re-lookup the device(s) you're interested in. You can't just get "a new master port". As a general rule, you shouldn't expect to be able to do anything much useful after forking unlesss you're sticking to the basic BSD/POSIX APIs; as Jim's pointed out, most of the other frameworks depend on at-exec initialisation that can't be successfully reproduced later. If you know you haven't changed working directory, re-execing argv[0] with a 'magic' argument is simple enough. = 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

Re: question about mach ports and fork
by Nathan T. Hjelm 21 Jan '04

21 Jan '04

> Actually, only fork() and exec() invalidate the Mach ports. The current > vfork() implementation actually arranges for the Mach task associated > with the parent to be shared with the child (resulting in two procs > running in one Mach task and having equal access to Mach resources like > ports and VM). But it's only an implementation artifact and you > shouldn't depend upon it. You must stick to the list of vfork()-safe > APIs and behaviors (exec(), _exit(), not popping the stack, etc...) to > assure continued operations regardless of the implementation. I just tried that with my library and all of my IOUSBDeviceInterfaces work with both the child and parent. I will have to keep that in mind. Its too bad I can't rely on that because it is exactly what i need (and expected from fork). > There is no [guaranteed safe] way to re-validate them under the same > port name with the same number of references (thus making the port > names sprinkled throughout the process memory valid again). But there > are some common techniques used to pass send rights to these ports from > the parent to the child under a [randomly chosen] new name. And many > of the frameworks allow for instantiation of objects passing an > explicit port to use - rather than looking them anew. I'm not sure > about IOService, however. You can't just open a new IOKit service > connection in the child? This is sometimes impossible because the > parent was privileged and the child not (so I feel your pain). I have tried to open a new connection to iokit with the child process by calling IOCreateMasterPort again but I get the same result as when I didn't. The problem must lie in the IOService which looks like it keeps trak of the master port used to open it. It would be a really simple fix if i could rely on the programmer to call usb_open on a device after the fork call (i think. i know there are a number of ports invalidated that I don't really know about). Thanks for your reply. I am just starting to get a handle on the mach/bsd kernel used in darwin and how it differs from a traditional bsd kernel. -Nathan _______________________________________________ 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: question about mach ports and fork
by Jim Magee 21 Jan '04

21 Jan '04

On Jan 21, 2004, at 3:37 PM, Nathan T. Hjelm wrote: I have tried to open a new connection to iokit with the child process by calling IOCreateMasterPort again but I get the same result as when I didn't. The problem must lie in the IOService which looks like it keeps trak of the master port used to open it. It would be a really simple fix if i could rely on the programmer to call usb_open on a device after the fork call (i think. i know there are a number of ports invalidated that I don't really know about). After fork()ing the memory image is stale (i.e. contain names for Mach ports that the child doesn't actually have). One could argue that all these frameworks should install atfork() handlers and clean up this state. The problem is that it is not really possible to determine the order those handlers have to be called in in order to assure all dependencies are met (the spec says they are called in the same order that they are registered, but the order required to "re-initialize" the frameworks may be quite different than the order to initialize them the first time). So, we haven't gone that route (except for the basic Un*x libraries). Instead, you MUST exec() after fork() in any process that is using any Mac OS X-specific frameworks. You can re-exec the same executable - and pass a parameter to indicate that it has already fork()'ed and doesn't need to again. If you do this, you will re-initialize correctly. --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