November 2001 - Darwin-kernel

Re: Message Queues
by Umesh Vaishampayan 30 Nov '01

30 Nov '01

On Monday, November 26, 2001, at 02:07 PM, Sean Morrell wrote: Are the BSD (I believe originally system V) message queues implemented in Darwin? Just making sure. I don't see the header "msg.h". Thanks, Sean No. They are not. --Umesh -- Umesh Vaishampayan Apple Computer, Inc. Mac OS X Kernel Ph: (408) 974 0229

1 0

Re: Limitations Within a Driver
by Godfrey van der Linden 27 Nov '01

27 Nov '01

Sean this project of yours is definitely going around in circles. Can you describe in great detail the problem you are trying to solve and lets see if we can find a clean design for you. X has any number of mechanismns that can provide the sort of services you want but there is usually one or two mechanisms that stand out once the problem is clearly defined. So I can assume you wish to write a X Audio driver framework that will be in some manner software configurable. What are you trying to do at configuration time? Godfrey At 11:20 -0800 01-11-26, Sean Morrell wrote: justin(a)mac.com writes: You can probably get this to work, but I think, in general, it's A Bad Idea. What, exactly, are you trying to achieve? Oh, I was just looking for another way of configuring my device driver from user space. The sysctl mechanism was suggested by a gentlemen a few days ago. But I don't believe it will work in my situation. This device is one of a series. Several devices of the same model can be connected on a single system. And there is no way of identifying each device uniquely, other than USB address. So I would need a set of settings for every (model, USB address). I don't see sysctl being applicable here. Tell me if I'm wrong. I just thought it might be easier to have the configuration done via a pipe or message queue. Any suggestions? Regards, >Justin

1 0

Re: Limitations Within a Driver
by justin＠mac.com 26 Nov '01

26 Nov '01

On Monday, November 26, 2001, at 11:20 , Sean Morrell wrote: justin(a)mac.com writes: You can probably get this to work, but I think, in general, it's A Bad Idea. What, exactly, are you trying to achieve? Oh, I was just looking for another way of configuring my device driver from user space. The sysctl mechanism was suggested by a gentlemen a few days ago. Sysctls are probably a bad idea; in particular, they deal primarily with the BSD part of the kernel, and making them visible to the IOKit side is going to be a hassle. In addition, even BSDs don't typically use sysctls for this purpose. But I don't believe it will work in my situation. This device is one of a series. Several devices of the same model can be connected on a single system. And there is no way of identifying each device uniquely, other than USB address. So I would need a set of settings for every (model, USB address). I don't see sysctl being applicable here. Tell me if I'm wrong. I just thought it might be easier to have the configuration done via a pipe or message queue. Any suggestions? Pipes and message queues are, again, BSD-isms, and something you'd want to avoid in an IOKit gizmo. Pipes, FWIW, are typically used between user-mode processes, not in the kernel. For your purposes, there are two possibilities: - if your driver is going to look like an existing BSD-like thing (e.g., there's a device node for it in the file system), you can use ioctls to communicate with it. This requires it be open, which may not suit you. - IOKit provides something called the UserClient interface, which may be what you want. Search the archives for details (I'm no expert here). You'll want to look in the darwin-developers archives as well (e.g., at www.darwinfo.org) Regards, Justin -- /~\ The ASCII Justin C. Walker, Curmudgeon-at-large \ / Ribbon Campaign X Against HTML / \ Email!

1 0

Re: Limitations Within a Driver
by Louis Gerbarg 26 Nov '01

26 Nov '01

I would only recommend using sysctl if your code lives inside the bsd section of the kernel proper. I would tend to suggest you use get and setProperties (and IORegistryEntryCreateCFProperty and friends in userland) if you need a low bandwidth channel, or an IOUserClient if you need a highbandwidth/low latency communication system. Louis On Monday, November 26, 2001, at 01:09 PM, Sean Morrell wrote: louis(a)apple.com writes: File I/O (meaning anything that goes through a file descripter, including sockets and pipes) is just not a good thing from inside a driver. From a strictly technical point you can do it, but it is tricky, expensive, and fragile. In short it is a bad idea. If you are asking about BSD style signals, they are a process based concept, thread delivery is difficult in userland when you are multithreaded. In the kernel they just don't work like that. So what must I use to communicate between my user land configuration app and the driver? I have been told to use sysctl. Is this my only option? Can Mach IPC be used from user space? Is it recommended? Any other options? Louis On Sunday, November 25, 2001, at 10:22 PM, Sean Morrell wrote: Can named pipes, sockets and signals be used from within a driver, specifically an I/O Kit kernel audio driver? I have been told that disk I/O cannot be performed from within a driver? Is this true? If so, does this exclude the use of a disk based IPC such as a named pipe? Thanks, Sean _______________________________________________ darwin-kernel mailing list darwin-kernel(a)lists.apple.com http://www.lists.apple.com/mailman/listinfo/darwin-kernel

1 0

Message Queues
by Sean Morrell 26 Nov '01

26 Nov '01

Are the BSD (I believe originally system V) message queues implemented in Darwin? Just making sure. I don't see the header "msg.h". Thanks, Sean

1 0

Re: Limitations Within a Driver
by Fernando Urbina 26 Nov '01

26 Nov '01

"Sean Morrell" said: > And there is no way of identifying each device uniquely, > other than USB address. You are not guaranteed that the USB address will be the same across reboots. I would use the locationID (GetLocationID) for each device as the way to uniquely identify a USB device. - Fernando -- Fernando Urbina mailto:nano@apple.com USB Technology Team work: 719.534.9387 Apple Computer, Inc. fax: 719.532.0981

1 0

Re: Limitations Within a Driver
by Sean Morrell 26 Nov '01

26 Nov '01

louis(a)apple.com writes: >File I/O (meaning anything that goes through a file descripter, >including sockets and pipes) is just not a good thing from inside a >driver. From a strictly technical point you can do it, but it is tricky, >expensive, and fragile. In short it is a bad idea. If you are asking >about BSD style signals, they are a process based concept, thread >delivery is difficult in userland when you are multithreaded. In the >kernel they just don't work like that. So what must I use to communicate between my user land configuration app and the driver? I have been told to use sysctl. Is this my only option? Can Mach IPC be used from user space? Is it recommended? Any other options? > > >Louis > >On Sunday, November 25, 2001, at 10:22 PM, Sean Morrell wrote: > >> Can named pipes, sockets and signals be used from within a driver, >> specifically an I/O Kit kernel audio driver? I have been told that disk >> I/O cannot be performed from within a driver? Is this true? If so, >> does >> this exclude the use of a disk based IPC such as a named pipe? >> >> Thanks, >> >> Sean >> _______________________________________________ >> darwin-kernel mailing list >> darwin-kernel(a)lists.apple.com >> http://www.lists.apple.com/mailman/listinfo/darwin-kernel

1 0

Re: Limitations Within a Driver
by Louis Gerbarg 26 Nov '01

26 Nov '01

File I/O (meaning anything that goes through a file descripter, including sockets and pipes) is just not a good thing from inside a driver. From a strictly technical point you can do it, but it is tricky, expensive, and fragile. In short it is a bad idea. If you are asking about BSD style signals, they are a process based concept, thread delivery is difficult in userland when you are multithreaded. In the kernel they just don't work like that. Louis On Sunday, November 25, 2001, at 10:22 PM, Sean Morrell wrote: Can named pipes, sockets and signals be used from within a driver, specifically an I/O Kit kernel audio driver? I have been told that disk I/O cannot be performed from within a driver? Is this true? If so, does this exclude the use of a disk based IPC such as a named pipe? Thanks, Sean _______________________________________________ darwin-kernel mailing list darwin-kernel(a)lists.apple.com http://www.lists.apple.com/mailman/listinfo/darwin-kernel

1 0

Do mlock() and madvise() work?
by Ian Ollmann 26 Nov '01

26 Nov '01

Hello all, I am hoping someone familiar with vm on darwin can answer this question. Simply put, if I use mlock or madvise from userland, are they going to do anything, or is this currently a noop? The OS claims it is very liberal about how much space you can lock down, currently LONG_MAX bytes. Normally I would assume it would work, but this seems like just the sort of thing a kernel engineer would decide was a bad idea for religious reasons and procrastinate implementing ad infinitum ... so I thought I would ask. This feature is important for low latency real time audio. If the audio latency is less than the seek time on the hard drive, then vm will fail to deliver the data on time and the sound will break if important data comes to be paged out. Also, on a related note, is there any way to hint to the kernel that a region of memory is likely to be needed soon, and have it do an asynchronous prefetch on it to get it into DRAM before it is needed. Should I use madvise() for this? Should I just set up a thread that blocks on a queue and pass addresses to it to touch? What is the best way to go here? Ian --------------------------------------------------- Ian Ollmann, Ph.D. iano(a)cco.caltech.edu ---------------------------------------------------

1 0

Re: Limitations Within a Driver
by Sean Morrell 26 Nov '01

26 Nov '01

justin(a)mac.com writes: >You can probably get this to work, but I think, in general, it's A Bad >Idea. What, exactly, are you trying to achieve? Oh, I was just looking for another way of configuring my device driver from user space. The sysctl mechanism was suggested by a gentlemen a few days ago. But I don't believe it will work in my situation. This device is one of a series. Several devices of the same model can be connected on a single system. And there is no way of identifying each device uniquely, other than USB address. So I would need a set of settings for every (model, USB address). I don't see sysctl being applicable here. Tell me if I'm wrong. I just thought it might be easier to have the configuration done via a pipe or message queue. Any suggestions? > > >Regards, > >Justin > >On Sunday, November 25, 2001, at 10:22 , Sean Morrell wrote: > >> Can named pipes, sockets and signals be used from within a driver, >> specifically an I/O Kit kernel audio driver? I have been told that disk >> I/O cannot be performed from within a driver? Is this true? If so, >> does >> this exclude the use of a disk based IPC such as a named pipe? >> >> Thanks, >> >> Sean >> _______________________________________________ >> darwin-kernel mailing list >> darwin-kernel(a)lists.apple.com >> http://www.lists.apple.com/mailman/listinfo/darwin-kernel >> >> >--- >Justin C. Walker, Curmudgeon-At-Large * >Institute for General Semantics | Men are from Earth. > | Women are from Earth. > | Deal with it. >*--------------------------------------*-------------------------------* >_______________________________________________ >darwin-kernel mailing list >darwin-kernel(a)lists.apple.com >http://www.lists.apple.com/mailman/listinfo/darwin-kernel

1 0