July 2003 - Darwin-kernel - lists.apple.com

RE: Kernel Data Alignment
by David Chatterton 10 Jul '03

10 Jul '03

Markus, > > Am Donnerstag, 10.07.03 um 01:43 Uhr schrieb David Chatterton: > > > Most Darwin structure appear not to use 64bit values, > > size_t is 32bits for example. > > How about padding with dummy variables? > Sure, its just a bit frustrating since this has been ported to several other UNIX & Windows platforms. David _______________________________________________ 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: Kernel Data Alignment
by Markus Hitter 10 Jul '03

10 Jul '03

Am Donnerstag, 10.07.03 um 01:43 Uhr schrieb David Chatterton: Most Darwin structure appear not to use 64bit values, size_t is 32bits for example. How about padding with dummy variables? For sizeof(size_t) == 4: typedef struct { size_t mysize; char dummy[4]; } ... For sizeof(size_t) == 8: typedef struct { size_t mysize; } ... Maybe you even can automate this with preprocessor directives. HTH, Markus - - - - - - - - - - - - - - - - - - - Dipl. Ing. Markus Hitter http://www.jump-ing.de/ _______________________________________________ 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

Kernel Data Alignment
by David Chatterton 09 Jul '03

09 Jul '03

We are having problems with data alignment of 64bit types within structures. From the Mach-O docs, it appears that MacOSX code defaults to the Power model which does not align 64bit values on a 64bit boundary, however other platforms we are communicating with do and so our structures do not match. Most Darwin structure appear not to use 64bit values, size_t is 32bits for example. So is there any problem with compiling our kext with align=natural and interacting with kernel structures? A compiler option is obviously easier than #pragma align=natural around our OTW structures, or adding padding where needed. We tried to use __attribute__((aligned(8))) without much success, ie it made no difference. Is this expected to work (it was used on the same code for an intel based Linux with successs)? And what about in 10.3 with more 64bit support? I'm assuming that we will start to see more 64bit types in the kernel, are there any plans regarding alignment? Thanks, David --- David Chatterton Phone: +61 3 8530 7777 Adacel Technologies Fax: +61 3 9596 2960 david.chatterton(a)adacel.com www.adacel.com/global _______________________________________________ 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

Duplicate entries in a file system...
by Marek Kozubal 09 Jul '03

09 Jul '03

Hi! I am writing a file system that is designed to look like HFS+ to Mac OS X. I am having a problem that I just cannot find the cause to. Everything works fine when using the file system from the posix/command line side of things. However, from Carbon (ie the Finder, File Buddy, etc), trying to do various tasks in the Finder causes a second entry for the parent folder that I'm working with to appear. This then causes the Finder to think that they are in fact different folders. An example of this: Open a folder Select a file inside of it Copy the file Paste the file into the same directory... What normally happens is that the Finder adds the word "copy" to the file name and makes a duplicate, however what I am seeing is the following: After the paste command but before the copy "starts" a second duplicate of the parent directory appears, and then I get a dialog saying that the item already exists, and asks me to replace or cancel. This implies that the Finder no longer believes that the parent directories of the file are the same and so doesn't add the word copy to the file name. Also using the "duplicated" entry generally works as expected, w/the "original" one not working and not having any of its contents updated. (Its possible that the duplicate and original are backwards, there's no easy way to tell which is which, but one works and one does not.) Another thing that is weird is that if I made a new folder or copy a file to the same level as these directories suddenly everything works fine for the rest of this mount, but unmounting it and remounting it will usually return the problem. I have looked at vfs_vget(), vop_lookup(), and vop_getattrlist() for any bugs that might return the wrong CNID, or parent CNID, or vnode, but this is all consistant and always returns what is expected. So does anyone know what would cause Carbon/Finder to create this duplicate entry in the Finder window? Like I said, the posix side works fine so I think its something w/the carbon layer. I have not tested a Cocoa application to see if it has a simular problem. Thanks in advance. --- Marek Kozubal marek(a)portents.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: dlil inject output problem
by Joshua Graessley 07 Jul '03

07 Jul '03

There are a few problems here. I believe that your call to output will cause the packet to go through your output filter again. I'm not sure why you're calling m_copym. You could just pass m to dlil_inject_if_output and return EJUSTRETURN. -josh On Monday, July 7, 2003, at 10:13, Hale, Darrian J wrote: Hello, Has anyone ever used dlil_inject_if_output or dlil_inject_pr_output successfully? I cant seem to get either one to work correctly. Whenever I use either of them i seem to get repeat packets. I don't get this problem using dlil_inject_xx_input however. A simple test I tried for the interface filter is as follows: u_long if_hook_id; // Do some startup to get if_hook_id // Function that gets called on outbound packet int if_hook_output(caddr_t cookie, struct ifnet **ifp, struct mbuf **m) { int ret = 0; struct mbuf *mbf = NULL; mbf = m_copym(*m,0,M_COPYALL,M_WAIT); if((ret = dlil_inject_if_output(mbf,if_hook_id)) != 0) printf("dlil_inject_if_output failure, errno %d...\n",ret); return -1; } Thanks for your help, Darrian _______________________________________________ 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

dlil inject output problem
by Hale, Darrian J 07 Jul '03

07 Jul '03

Hello, Has anyone ever used dlil_inject_if_output or dlil_inject_pr_output successfully? I cant seem to get either one to work correctly. Whenever I use either of them i seem to get repeat packets. I don't get this problem using dlil_inject_xx_input however. A simple test I tried for the interface filter is as follows: u_long if_hook_id; // Do some startup to get if_hook_id // Function that gets called on outbound packet int if_hook_output(caddr_t cookie, struct ifnet **ifp, struct mbuf **m) { int ret = 0; struct mbuf *mbf = NULL; mbf = m_copym(*m,0,M_COPYALL,M_WAIT); if((ret = dlil_inject_if_output(mbf,if_hook_id)) != 0) printf("dlil_inject_if_output failure, errno %d...\n",ret); return -1; } Thanks for your help, Darrian _______________________________________________ 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: Asynchronous I/O in Panther
by Jim Magee 03 Jul '03

03 Jul '03

On Monday, June 30, 2003, at 8:42 PM, Sam Vaughan wrote: Asynchronous I/O has been working just fine for a long time now. So just what are they getting at here? POSIX-style async I/O interfaces as the kernel level. --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: Analysis of Darwin kernel design
by Jim Magee 03 Jul '03

03 Jul '03

On Tuesday, July 1, 2003, at 6:59 AM, Kim-Thomas Moeller wrote: Is the source of the Darwin MIG interface generator publicly available? In what extent does it differ from GNU MIG? Yes, MIG is open source. It is part of the bootstrap_cmds project. As for differences - this mostly comes from the two origins. The GNU MIG is mostly from CMU, where our MIG is derived from OSF's version. That means: Our MIG has (currently unused) support for OSF RPC-style kernel interfaces in addition to the traditional mach_msg() style interfaces. We have limited support for DCE-style Network Data Representation (NDR) support - as our MIG is primarily used internal to a single machine (although recent contributions have added back some NDR support for network use) Otherwise, they are pretty similar (last I checked the GNU version). --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

bdevsw_add() is broken
by Sam Vaughan 03 Jul '03

03 Jul '03

According to a comment in bdevsw_add() in xnu/bsd/kern/bsd_stubs.c, a fix was applied to address Radar bug 2842228. I don't know the history behind the fix, but it seems to have completely broken the function. The change appears to be trying to skip the first item in the bdevsw table by using an initial index value of 1. However the items are being examined via the devsw pointer, not by indexing bdevsw. This means that the index doesn't correspond to the item being examined. The result is a bit of a mess. When I call bdevsw_add() with an index of -1, the for() loop returns immediately because the first entry in bdevsw is empty. Since index is 1 though, devsw then gets set to bdevsw[1], which isn't empty. The second memcmp() fails and the function returns -1. If someone could kindly confirm for me that this is wrong, I'll post a bug to Radar. Thanks, Sam _______________________________________________ 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

IORegistryEntrySetCFProperty in AudioDriverPlugInStreamSetProperty
by Alain Crétet 02 Jul '03

02 Jul '03

I have write an audio driver plugin that work correctly with Mac OS X 10.2.3, but with 10.2.6 I have the following problem. I use a specific control in the Audio stream plugin driver to change a value in the driver (See sample code below). In the AudioDriverPlugInStreamSetProperty, I use the IORegistryEntrySetCFProperty function that return 0xE00002BC I think that the 0xE00002BC error code is : #define kIOReturnError iokit_common_err(0x2bc) // general error Could someone explain why this error is returned with Mac OS X 10.2.6 ? I have the same problem with Mac OS X 10.2.3 and G4 DP 1.2 Ghz, not with G4 DP 867 and G3 350. Thanks in advance for your help, Alain This is a sample code of the audio driver plugin : // **************************************************************************** // AudioDriverPlugInStreamSetProperty // **************************************************************************** OSStatus AudioDriverPlugInStreamSetProperty( AudioDeviceID PmDevice, io_object_t PmIOAudioStream, const AudioTimeStamp* PmWhen, UInt32 PmChannel, AudioDevicePropertyID PmPropertyID, UInt32 PmPropertyDataSize, const void* PmPropertyData) { OSStatus LcResult; SInt32 *LcSInt32Ptr; SInt32 LcSInt32; CFNumberRef LcCFNumberRef; LcResult = kAudioHardwareUnknownPropertyError; switch (PmPropertyID) { // ******************************* // Mpeg header // ******************************* case kPropertyMpegHeader: LcResult = noErr; LcSInt32Ptr = (SInt32*) PmPropertyData; LcSInt32 = *LcSInt32Ptr; #ifdef DEBUG_SETSTREAM fprintf( stdout, "%s() : *** --> kPropertyMpegHeader : set PropertyData = 0x%0lX \n", __FUNCTION__, LcSInt32 ); fflush(stdout); #endif LcCFNumberRef = CFNumberCreate( CFAllocatorGetDefault(), kCFNumberSInt32Type, &LcSInt32 ); LcResult = IORegistryEntrySetCFProperty( gConnectionReferenceToControlMpegHeader, CFSTR("IOAudioControlValue"), LcCFNumberRef ); CFRelease(LcCFNumberRef); #ifdef DEBUG_SETSTREAM fprintf( stdout, "%s() : IORegistryEntrySetCFProperty(0x%0X, IOAudioControlValue) LcResult = 0x%0lX %ld \n", __FUNCTION__, gConnectionReferenceToControlMpegHeader, LcResult, LcResult ); fflush(stdout); #endif break; default: break; } return LcResult; } Someone tell me to replace the CFNumberCreate with CFDataCreate. I try it, and I have no error in the IORegistryEntrySetCFProperty function but the value in the IORegistry did not change. This is a sample code of my driver : // **************************************************************************** // **************************************************************************** bool myAudioEngine::initHardware(IOService *provider) { ... // ****************************************** // Create CUSTOM control : Mpeg header // ****************************************** mControlMpegHeader = IOAudioLevelControl::create( CVAL_HEADMPEG_MIN, // Initial value CVAL_HEADMPEG_MIN, // min value : CVAL_HEADMPEG_MAX, // max value : CVAL_HEADMPEG_MIN, // min dB CVAL_HEADMPEG_MAX, // max dB 0, "Mpeg header", kPropertyMpegHeaderCtrlID, // control ID - driver-defined 0, // subType kPropertyMpegHeader ); if (!mControlMpegHeader) { ERRORLOG( " IOAudioLevelControl::mControlMpegHeader() \n" ); return; } // -------------------------------------------------------------------------- mControlMpegHeader->setValueChangeHandler( (IOAudioControl::IntValueChangeHandler)setMpegHeaderChangeHandler, this ); PmAudioStream->addDefaultAudioControl(mControlMpegHeader); mControlMpegHeader->release(); ... } // **************************************************************************** // MPEG HEADER CHANGE HANDLER // **************************************************************************** IOReturn myAudioEngine::setMpegHeaderChangeHandler( IOService *target, IOAudioControl *mpegHeaderControl, SInt32 oldValue, SInt32 newValue) { IOReturn result = kIOReturnBadArgument; myAudioEngine *audioEngine; DOUT( DBG_MESSAGE, (" ** %s() : target = %p mpegHeaderControl = %p \n", __FUNCTION__, target, mpegHeaderControl )); audioEngine = (myAudioEngine *)target; if (audioEngine) { result = audioEngine->setMpegHeaderChanged( mpegHeaderControl, oldValue, newValue); } return result; } // **************************************************************************** // MPEG HEADER CHANGED // **************************************************************************** IOReturn myAudioEngine::setMpegHeaderChanged( IOAudioControl *mpegHeaderControl, SInt32 oldValue, SInt32 newValue) { if (mpegHeaderControl) { DOUT( DBG_CLOCK, (" ** %s() : Channel %ld, MpegHeader = 0x%0lX \n", __FUNCTION__, mpegHeaderControl->getChannelID(), newValue )); // Set Mpeg Header mMpegHeader = newValue; } return kIOReturnSuccess; } _______________________________________________ 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