December 2002 - Darwin-kernel

Re: reading a file in the kernel
by Dan Markarian 16 Dec '02

16 Dec '02

Hi Michael, it seems that there is no way to pass arbitrary buffers You can, certainly, but not with this technique. I believe the IOConnectMethod*() variable-sized structures are limited to a maximum of 4 kB. These structures are copied after all. You would just pass the buffer address and buffer length down to your user client. The user client can then temporarily map the memory to read its contents. { IOMemoryDescriptor * buffer; buffer = IOMemoryDescriptor::withAddress( aUserBufferAddress, aUserBufferLength, kIODirectionOut, fUserTask ); if ( buffer ) { if ( buffer->prepare() == kIOReturnSuccess ) { /* * Use buffer->readBytes() or buffer->map() here. */ buffer->complete(); } buffer->release(); } } Dan On Friday, December 13, 2002, at 03:35 PM, Michael George wrote: On Thursday, December 12, 2002, at 04:57 PM, Amanda Walker wrote: On 12/12/02 4:06 PM, Michael George <george(a)auroravideosys.com> wrote: This is what I was thinking of doing as a fallback. However, using "IOConnectMethod*()" I can pass scalars and structures. However, the data I want to pass in will be of arbitrary length, so I can't define that length in the getTargetAndMethodForIndex() data table. Set the length to 0xffffffff in the table. IOConnectMethod() can handle variable length buffers just fine. I don't think this is the case anymore. I tried it and I got some wild-high IOReturn value. I did some other tests and it seems that: If the size of the struct as defined in getTargetAndMethodForIndex() is too large (somewhere between 1Kb (which succeeded) and 512Kb (which failed)) I get 0xfffffecd as the result. It doesn't matter what size I really pass when I make the IOConnectMethod*() call. Needless to say, 0xffffffff is much bigger than .5Mb... It also seems that if the size for a structure I send in the IOConnectMethod*() calls isn't the same as the size declared in getTargetAndMethodForIndex(), then I get 0xe00002c2 (bad argument) back. Together, it seems that there is no way to pass arbitrary buffers through IOConnectMethod*() calls... Can anyone else confirm or correct this? -Michael _______________________________________________ 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

Strange CPU value in Top
by Jean-Edouard BABIN 14 Dec '02

14 Dec '02

Hi, I have burn a cd with toast, my burner is FireWire, and when toast was testing the CD, my mac comme very very slow, i did a top on the terminal and i have seen this: 0 kernel_tas 202.0% 27:34.32 31 0 - - - 35.2M- 311M- 376 Toast Tita 35.2% 12:47.73 5 144 192 4.60M- 16.0M 18.1M- 94.5M- I don't understand... (i have one cpu..) Is it a 'normal' bug because the mac was too slow ? Or it is a kernel bug ? (i am using os x 10.2.2 on an imac and toast 5.2) -- Jean-Edouard BABIN Mail: Jeb(a)jeb.com.fr Web : www.Jeb.com.fr IRC : Jeb @ Undernet _______________________________________________ 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: howto compile kernel for ktrace
by Pelle Johansson 14 Dec '02

14 Dec '02

torsdagen den 12 december 2002 kl 17.58 skrev Michael Marti: " If you go to the CVS Web interface for xnu, you'll find a popup menu that lists all of the tags. " And this helps a lot. cvs status -v on any file in the xnu module (README is a simple choice) will do the same. Can be good to know. -- Pelle Johansson <morth(a)morth.nu> _______________________________________________ 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: reading a file in the kernel
by Jim Magee 14 Dec '02

14 Dec '02

On Friday, December 13, 2002, at 05:36 PM, Amanda Walker wrote: On 12/13/02 3:35 PM, Michael George <george(a)auroravideosys.com> wrote: too large (somewhere between 1Kb (which succeeded) and 512Kb (which failed)) I get 0xfffffecd as the result. It doesn't matter what size I really pass when I make the IOConnectMethod*() call. Needless to say, 0xffffffff is much bigger than .5Mb... It also seems that if the size for a structure I send in the IOConnectMethod*() calls isn't the same as the size declared in getTargetAndMethodForIndex(), then I get 0xe00002c2 (bad argument) back. Well, I have a kext, all of whose user client methods have 0xffffffff as the length for structure arguments, and all of whose methods work fine with variably-sized buffers under both 10.1.x and 10.2.x. Now, in my case the buffers are small (in the realm of "no more than a few K at a time", so there may well be static limits that I am not running into. The io_struct_inband_t MIG type is declared to be "up to 4K" in the device (IOKit) MIG interface definitions. --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: reading a file in the kernel
by Michael George 13 Dec '02

13 Dec '02

On Thursday, December 12, 2002, at 04:57 PM, Amanda Walker wrote: On 12/12/02 4:06 PM, Michael George <george(a)auroravideosys.com> wrote: This is what I was thinking of doing as a fallback. However, using "IOConnectMethod*()" I can pass scalars and structures. However, the data I want to pass in will be of arbitrary length, so I can't define that length in the getTargetAndMethodForIndex() data table. Set the length to 0xffffffff in the table. IOConnectMethod() can handle variable length buffers just fine. I don't think this is the case anymore. I tried it and I got some wild-high IOReturn value. I did some other tests and it seems that: If the size of the struct as defined in getTargetAndMethodForIndex() is too large (somewhere between 1Kb (which succeeded) and 512Kb (which failed)) I get 0xfffffecd as the result. It doesn't matter what size I really pass when I make the IOConnectMethod*() call. Needless to say, 0xffffffff is much bigger than .5Mb... It also seems that if the size for a structure I send in the IOConnectMethod*() calls isn't the same as the size declared in getTargetAndMethodForIndex(), then I get 0xe00002c2 (bad argument) back. Together, it seems that there is no way to pass arbitrary buffers through IOConnectMethod*() calls... Can anyone else confirm or correct this? -Michael _______________________________________________ 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

Working With SCSI Architecture Model Devices (December 2001)?
by Heck, Steve 13 Dec '02

13 Dec '02

I am having trouble getting the sample src from Working With SCSI Architecture Model Devices (December 2001) to run on 10.2. When I try to run the application I am getting crash at (EXC_BAD_ACCESS) at: task = ( *interface )->CreateSCSITask ( interface ); Also getting one Warning: implicit declaration of function 'mach_port_deallocate' -> what header file is missing? Has anyone gotten this src to work? I am trying to access a Medium Changer and Tape drive. Is there any other sample src available? _______________________________________________ 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: 3DFX Voodoo 4 Driver
by Sean McBride 13 Dec '02

13 Dec '02

Sam Hart (samh(a)granada-learning.com) wrote on Thu, 12 Dec 2002 22:46:12 +0000: >I guess the reason Darwin doesn't support them is because it doesn't >have to be backward compatible with OS 9 ( ?? ) Well, there is all sorts of open source IONDRV stuff in the Darwin source, so I don't think its that. Maybe you found a bug. Wouldn't be the first. :) _______________________________________________ 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: 3DFX Voodoo 4 Driver
by Sam Hart 12 Dec '02

12 Dec '02

On Wednesday, December 11, 2002, at 07:27 PM, Sean McBride wrote: Sam Hart (samh(a)granada-learning.com) wrote on Tue, 10 Dec 2002 09:40:49 +0000: lot of .mac sites ... strange ;) ). What I am wanting to know is, does anyone know anything about writing a video driver in Darwin / OS X or writing a 3DFX driver in general, as I am not above getting my hands dirty and attempting one ( how bad can it be ? ). It is likely that the card has an EEPROM with an OS 9 ndrv driver on it. OS X supports these (and only these) OS 9 drivers, with some caveats (search IOKitFundamentals for 'ndrv'). So its likely that the card works at all because of this ndrv. Unless there really is a OS X native IOFramebuffer for 3dfx cards, but I don't think so. no, I don't either I guess the reason Darwin doesn't support them is because it doesn't have to be backward compatible with OS 9 ( ?? ) Anyway, writing an IOFramebuffer is what you'd need to do. But without docs on the 3dfx hardware, it'd be near impossible I dare say. you're probably right down the Apple store for me then ; p _______________________________________________ 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: reading a file in the kernel
by Michael George 12 Dec '02

12 Dec '02

On Wednesday, December 11, 2002, at 03:48 PM, Jim Magee wrote: If you go to this length, just pass the file contents in, rather than the file descriptor number. Trying to convert from a filedescriptor number to real file data from within a KEXT will require you to access all sorts of kernel-private data structures which are not guaranteed against release-to-release changes. This is what I was thinking of doing as a fallback. However, using "IOConnectMethod*()" I can pass scalars and structures. However, the data I want to pass in will be of arbitrary length, so I can't define that length in the getTargetAndMethodForIndex() data table. I thought there was a way to pass buffers of dynamic length, but I can't find it now... -Michael _______________________________________________ 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

dlil Inject Problem
by Michael Cashwell 12 Dec '02

12 Dec '02

Greetings, I've got an interface filter NKE that's trying to inject an output packet that its filter function either consumed earlier or created. The packet is logically a roughly 1000 byte long ping response. The problem is that dlil_inject_if_output() returns 22 (EINVAL) and refuses to send the packet. I've added code to dump the mbuf chain and it consists of 2 parts (linked via the mbuf's m_next field). The first part is 14 bytes long and looks like a valid etherframe header (2 MAC addresses and a $0800 IP type). The second part (996 bytes long) looks like a valid IP header and then the ping payload. I've looked through the docs and the xnu source for dlil_inject_if_output() and the EINVAL return is either coming from another interface filter or from the driver itself. Either way I don't know what is invalid about the arguments. The last data point is that since the docs for dlil_inject_if_output() say the packet must have its frame header I used M_PREPEND() and a bcopy(... mtod()) to put the header in. (I then dump that to see the two-buffer mbuf chain.) I have to assume that M_PREPEND decided to link in a new buffer at the front of the mbuf. Is having a linked mbuf chain my problem? Is that supposed to only be used for representing multiple packets not different parts of one packet? If so then why did M_PREPEND do that? If not then what else might be wrong? 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