September 2003 - Darwin-kernel

SCSI device driver configure() function not called
by Jeff Collins 19 Sep '03

19 Sep '03

After some problems, I finally got a stub SCSI HBA driver to load on OSX using the developers kit from December 2002 by setting com.apple.kernel.mach = 6.1 (Thanks to apple folks for the proper advice on setting the version to 6.1). Please forgive my ignorance, but now I do not know how to get the configure() function called. The stub driver is loaded, and it has successfully brought 68 1 0x10f7b000 0x9000 0x8000 com.apple.iokit.IOSCSIParallelFamily (1.2.4) <35> 69 0 0x10f84000 0x2000 0x1000 ods (1.0d1) <68 11> the IOSCSIParallelFamily driver into memory as well. However, I do not know how to get the SCSI layer to call the configure() function of my driver. Any help would be greatly appreciated. Jeff Collins _______________________________________________ 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: wanted: sosend() howto
by Stefan Werner 19 Sep '03

19 Sep '03

Hi, On Friday, Sep 19, 2003, at 07:00 Europe/Berlin, darwin-kernel-request(a)lists.apple.com wrote: Date: Thu, 18 Sep 2003 23:05:28 +0100 To: darwin-kernel(a)lists.apple.com From: Quinn <eskimo1(a)apple.com> Subject: Re: wanted: sosend() howto In general Apple recommends that you avoid doing networking from kernel code because it requires you to use BSD in-kernel 'APIs', and there's no promise of binary compatibility for those APIs. A better solution is to use a user space daemon to do the networking on behalf of your kernel extension. This is, for example, what our WebDAV VFS plug-in does. If you can explain a little more about the big picture of your kernel extension, I might be able to suggest an alternative design that will make the networking simpler. I am working on an OS X based audio over Ethernet solution. Typically, audio drivers in OS X are kernel extensions (there is a way of doing it in user space like the iSight driver, but after evaluating it I came to the conclusion that the kernel driver API suits my needs better), and I am using the NetworkAudioDevice driver from opendarwin.org as a reference. This driver is using sosend() for transmitting TCP packets, and I now have it working with UDP packets (thanks to help from this list :). A user space daemon would be an option (cause I don't like working in kernel space, debugging is a lot harder than in user space), as long as I get the same precise timing as I do with the kernel driver. I'm confident that the context switches won't be a problem, being in the range of nanoseconds, but I need the guarantee that my network packet is being transmitted as soon as possible and not being delayed because the scheduler put my daemon to sleep. If there is sample code and maybe some documentation on how to establish efficient kernel/user space communication and an data transfer, I'd be grateful for any links. Stefan _______________________________________________ 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: DMA questions (invalidate_dcache and flush_dcache)
by Ryan Rempel 18 Sep '03

18 Sep '03

On Thursday, September 18, 2003, at 05:18 PM, Quinn wrote: At 17:29 -0500 12/9/03, Ryan Rempel wrote: I don't know if I actually need to do this. The drivers are the AppleCurio and AppleMaceEthernet drivers (i.e. the external SCSI and built-in ethernet on the older machines). The calls are in the original drivers from Darwin 0.9 (that is, I didn't put them there), so I'm not sure if they're absolutely required. I haven't tried removing them to see what happens yet. I'd suggest you remove them. As a general rule, if you're using I/O Kit memory descriptors correctly, you shouldn't need to flush the process cache manually. Either the machine is cache coherent (in which case flushing is unnecessary) or I/O Kit takes care of it for you. What sort of driver are you maintaining? They are the external SCSI drivers and built-in ethernet drivers for the older machines (the 7300 - 9600). I'll try taking them out and see what happens :-) _______________________________________________ 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: DMA questions (invalidate_dcache and flush_dcache)
by Quinn 18 Sep '03

18 Sep '03

At 17:29 -0500 12/9/03, Ryan Rempel wrote: I don't know if I actually need to do this. The drivers are the AppleCurio and AppleMaceEthernet drivers (i.e. the external SCSI and built-in ethernet on the older machines). The calls are in the original drivers from Darwin 0.9 (that is, I didn't put them there), so I'm not sure if they're absolutely required. I haven't tried removing them to see what happens yet. I'd suggest you remove them. As a general rule, if you're using I/O Kit memory descriptors correctly, you shouldn't need to flush the process cache manually. Either the machine is cache coherent (in which case flushing is unnecessary) or I/O Kit takes care of it for you. What sort of driver are you maintaining? S+E -- Quinn "The Eskimo!" <http://www.apple.com/developer/> Apple Developer Technical Support * Networking, Communications, Hardware _______________________________________________ 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: wanted: sosend() howto
by Quinn 18 Sep '03

18 Sep '03

At 10:20 +0200 17/9/03, Stefan Werner wrote: Sure, but what else am I supposed to do to send UDP packets? The usual user APIs like bind() or sendto() are not available in the kernel. In general Apple recommends that you avoid doing networking from kernel code because it requires you to use BSD in-kernel 'APIs', and there's no promise of binary compatibility for those APIs. A better solution is to use a user space daemon to do the networking on behalf of your kernel extension. This is, for example, what our WebDAV VFS plug-in does. If you can explain a little more about the big picture of your kernel extension, I might be able to suggest an alternative design that will make the networking simpler. S+E -- Quinn "The Eskimo!" <http://www.apple.com/developer/> Apple Developer Technical Support * Networking, Communications, Hardware _______________________________________________ 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: wanted: sosend() howto
by Stefan Werner 17 Sep '03

17 Sep '03

Hi, On Wednesday, Sep 17, 2003, at 07:00 Europe/Berlin, darwin-kernel-request(a)lists.apple.com wrote: I can understand your struggle. I don't find the 'sosocket()' call anywhere in the xnu source base. Are you sure you have that right? Sorry, my bad. I meant sosend(), as written in the subject. A) Using in-the-kernel calls like this is risky, in that there is no guarantee that the semantics of the call will remain untouched as the kernel evolves. This is just FYI. Sure, but what else am I supposed to do to send UDP packets? The usual user APIs like bind() or sendto() are not available in the kernel. B) Use the source, Luke. The 'xnu' project contains the code that both implements and uses these calls, and you should be able to make use of it to get the information you need by reading code. I tried, but it's going to take a long time to figure out just how I have to write 5 lines of code to send my tiny packet. Source code is not a replacement for a man page, otherwise cars came with a screwdriver instead of a manual. In particular, check the "aurp" code in the "bsd/netat" directory. This code sets up a UDP tunnel between the local system and a remote system to transmit AppleTalk packets; it should have a lot in common with your efforts. I "greped" over the xnu source for sosend() and SOCK_DGRAM, and I think I found the aurp piece. I am slowly figuring out what the arguments of the sosend() might be, but I couldn't figure out how to create and use a struct mbuf. C) As a guide to the code, you should consider looking at Wright/Stevens, "TCP/IP Illustrated, V.2", Addison-Wesley, 1995. I'll try and check a library. From: Wally Crooze <wallycrooze(a)pobox.com> The interface is found in xnu/bsd/sys/socketvar.h. To create a socket you call socreate. sosend is used to send both UDP and TCP packets. This is determined by arguments passed to socreate. I got that far too, but my packets are being rejected. My calls so far are: error = socreate (AF_INET, &sock, SOCK_DGRAM, IPPROTO_UDP); ... error = sosend (sock, (struct sockaddr*) &sockaddr, &uio, NULL, NULL, 0); But it always returns 61 (connection refused?). This is some source code that was originally for TCP packets and it works fine for TCP, but I wasn't able to turn it into working code for UCP. MacOSX uses the FreeBSD implementation of socket calls and is documented in "The Design and Implementation of the 4.4BSD Operating System" published by Addison Wesley. Am I the only one who think it's strange that proper documentation for the innerts of a free operating system that has been distributed for years over the internet is only available in books but not online or in man pages? Stefan _______________________________________________ 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: wanted: sosend() howto
by Wally Crooze 17 Sep '03

17 Sep '03

Stefan Werner wrote: I want to send UDP packets from kernel space and am currently struggling with the sosocket() call. It appears to be close to undocumented and almost an hour of research on google did not turn up much. I'd be very happy if I could get a simple code snippet that will allow me to send a void* buffer (I haven't figured out how to use mbuf yet) to a remote IP/Port. I found a little code in the nfs source code, but I couldn't quite figure out where exactly the payload is and what the other values mean. The interface is found in xnu/bsd/sys/socketvar.h. To create a socket you call socreate. sosend is used to send both UDP and TCP packets. This is determined by arguments passed to socreate. MacOSX uses the FreeBSD implementation of socket calls and is documented in "The Design and Implementation of the 4.4BSD Operating System" published by Addison Wesley. -- Wally Crooze +-------------------------------------------------------+ | ,-_|\ Wally Crooze <wallycrooze(a)pobox.com> | | / \ | | \_,-._/ | | v | +-------------------------------------------------------+ _______________________________________________ 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: wanted: sosend() howto
by Justin C. Walker 16 Sep '03

16 Sep '03

On Tuesday, September 16, 2003, at 08:56 AM, Stefan Werner wrote: Hi, I want to send UDP packets from kernel space and am currently struggling with the sosocket() call. It appears to be close to undocumented and almost an hour of research on google did not turn up much. I'd be very happy if I could get a simple code snippet that will allow me to send a void* buffer (I haven't figured out how to use mbuf yet) to a remote IP/Port. I found a little code in the nfs source code, but I couldn't quite figure out where exactly the payload is and what the other values mean. I can understand your struggle. I don't find the 'sosocket()' call anywhere in the xnu source base. Are you sure you have that right? Some thoughts: A) Using in-the-kernel calls like this is risky, in that there is no guarantee that the semantics of the call will remain untouched as the kernel evolves. This is just FYI. B) Use the source, Luke. The 'xnu' project contains the code that both implements and uses these calls, and you should be able to make use of it to get the information you need by reading code. There is no documentation, per se, for the kernel (cf. (A)). In particular, check the "aurp" code in the "bsd/netat" directory. This code sets up a UDP tunnel between the local system and a remote system to transmit AppleTalk packets; it should have a lot in common with your efforts. Note that the code is a bit encrusted with the dust of the ages; I don't know that it has been run or tested since the Rhapsody days. C) As a guide to the code, you should consider looking at Wright/Stevens, "TCP/IP Illustrated, V.2", Addison-Wesley, 1995. It is a very good "walk-through" of the code implementing the IP support in earier BSD variants. A lot of nits are different, and Darwin has the DLIL layer that is not mentioned in the book, but it will provide you with enough information that the Darwin code should look familiar. The 'mbuf' structure and use is detailed in Ch. 2. Regards, Justin -- /~\ The ASCII Justin C. Walker, Curmudgeon-at-Large \ / Ribbon Campaign X Help cure HTML Email / \ _______________________________________________ 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

wanted: sosend() howto
by Stefan Werner 16 Sep '03

16 Sep '03

Hi, I want to send UDP packets from kernel space and am currently struggling with the sosocket() call. It appears to be close to undocumented and almost an hour of research on google did not turn up much. I'd be very happy if I could get a simple code snippet that will allow me to send a void* buffer (I haven't figured out how to use mbuf yet) to a remote IP/Port. I found a little code in the nfs source code, but I couldn't quite figure out where exactly the payload is and what the other values mean. Thanks a lot, Stefan _______________________________________________ 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: DMA questions (invalidate_dcache and flush_dcache)
by Ryan Rempel 12 Sep '03

12 Sep '03

On Friday, September 12, 2003, at 10:12 AM, Benjamin Herrenschmidt wrote: On Monday, September 8, 2003, at 10:19 PM, Ryan Rempel wrote: I'm updating a driver for the new memory mapping limitations, and I've got a couple of DMA-related questions. The driver uses a couple of functions in <osfmk/ppc/pmap.h>, namely: extern void invalidate_dcache(vm_offset_t va, unsigned length, boolean_t phys); extern void flush_dcache(vm_offset_t va, unsigned length, boolean_t phys); The calls to invalidate_dcache are accompanied by this comment "Purge cache references for the DBDMA entry we are about to look at". The calls to flush_dcache are accompanied by this comnment "Push DMA commands to physical memory". Do you actually need to do that at all ? What driver is this ? Those machines are supposed to be cache coherent... I know there seem to be cache coherency bugs with earlier hardware, though I couldn't get any useful detail from Apple, I had some weird problems when a DBDMA wrote in a middle of a cache line with the MESH driver on a 8500 and possibly others. So if you are trying to deal with such issues, you need to flush/invalidate your data as well and make sure everything is cache line aligned. I don't know if I actually need to do this. The drivers are the AppleCurio and AppleMaceEthernet drivers (i.e. the external SCSI and built-in ethernet on the older machines). The calls are in the original drivers from Darwin 0.9 (that is, I didn't put them there), so I'm not sure if they're absolutely required. I haven't tried removing them to see what happens yet. _______________________________________________ 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