September 2005 - Darwin-dev

Corefoundation && Linux
by Jeffrey Drake 20 Sep '05

20 Sep '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Domainkey-signature: a=rsa-sha1; q=dns; c=nofws; s=beta; d=gmail.com; h=received:message-id:date:from:reply-to:to:subject:mime-version:content-type; b=rDc2LqzeF805bdgZZlEWN3LmPQVEIaAx+PmFWE+7eorhlw9XXBkvA7oJwmfLqpUbz6aoTZDQf9S403yZptosXoKSMA6lhGNs3rZDvXGVFwtAHExkmCIyBXFQlWnDZAEW3tUBIk/a24JU1fe3Y7+wKQ3SdZe61PvfeDFSrLcEpdA= _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0

Re: Is my program leaking?
by Marc Van Olmen 19 Sep '05

19 Sep '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com I'm wondering..... because I'm callling mach_make_memory_entry should I call mach_port_deallocate on the memory_entry that is returned? regards, mvo On Sep 19, 2005, at 9:13 AM, Marc Van Olmen wrote: hi, Here the code: static void * AllocateVirtualBuffer(OS_MemorySize pBufferLength) { kern_return_t error; vm_address_t originalAddress = NULL; vm_address_t realAddress = NULL; mach_port_t memoryEntry; vm_size_t memoryEntryLength; vm_address_t virtualAddress = NULL; OS_Int32 errorCount; // Success! return (void *)realAddress; #ifdef mkDebug // Here's a little test... *(char *)realAddress = '?'; if (*(char *)virtualAddress != '?') OS_Debug_Log("VirtualRingBuffer: Test 1: vm magic failed\n"); *(char *)(virtualAddress + 1) = '!'; if (*(char *)(realAddress + 1) != '!') OS_Debug_Log("VirtualRingBuffer: Test 2: vm magic failed\n"); #endif errorReturn: if (realAddress) vm_deallocate(mach_task_self(), realAddress, pBufferLength); if (virtualAddress) vm_deallocate(mach_task_self(), virtualAddress, pBufferLength); return NULL; } void FreeVirtualBuffer(void *fBuffer, UInt32 pBufferLength) { kern_return_t error; On Sep 19, 2005, at 2:12 AM, plumber wrote: Should I be worried? Is my application leaking memory? How can I trace where the leak is happening? take a look here /Developer/Applications/Performance Tools/ Best Regards _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… I figured out that it is "leaking" in some code I used for a long time, it is a code that allocates a buffer that is circular by using a sort of Virtual memory technique... I picked this code up from the internet and it seems that causes the leak... I'm guessing that the vm_deallocate is not releasing all the memory. The current leak doesn't seem to be traceable by mallocdebug because it isn't using any of those calls... I was using that tool before... / /---------------------------------------------------------------------- ------------------ // AllocateVirtualBuffer / /---------------------------------------------------------------------- ------------------ // We want to find where we can get 2 * fBufferLength bytes of contiguous address space. // So let's just allocate that space, remember its address, and deallocate it. // (This doesn't actually have to touch all of that memory so it's not terribly expensive.) error = vm_allocate(mach_task_self(), &originalAddress, 2 * pBufferLength, TRUE); if (error) { #ifdef mkDebug OS_Debug_Break("vm_allocate initial chunk", error); #endif return NULL; } error = vm_deallocate(mach_task_self(), originalAddress, 2 * pBufferLength); if (error) { #ifdef mkDebug OS_Debug_Break("vm_deallocate initial chunk", error); #endif return NULL; } // Then allocate a "real" block of memory at the same address, but with the normal pBufferLength. realAddress = originalAddress; error = vm_allocate(mach_task_self(), &realAddress, pBufferLength, FALSE); if (error) { #ifdef mkDebug OS_Debug_Break("vm_allocate real chunk", error); #endif return NULL; } if (realAddress != originalAddress) { #ifdef mkDebug OS_Debug_Log("allocateVirtualBuffer: vm_allocate 2nd time didn't return same address (%p vs %p)\n", originalAddress, realAddress); #endif goto errorReturn; } // Then make a memory entry for the area we just allocated. memoryEntryLength = pBufferLength; error = mach_make_memory_entry(mach_task_self(), &memoryEntryLength, realAddress, VM_PROT_READ | VM_PROT_WRITE, &memoryEntry, NULL); if (error) { #ifdef mkDebug OS_Debug_Break("mach_make_memory_entry", error); #endif goto errorReturn; } if (!memoryEntry) { #ifdef mkDebug OS_Debug_Log("mach_make_memory_entry: returned memoryEntry of NULL\n"); #endif goto errorReturn; } if (memoryEntryLength != pBufferLength) { #ifdef mkDebug OS_Debug_Log("mach_make_memory_entry: size changed (from %0x to %0x)\n", pBufferLength, memoryEntryLength); #endif goto errorReturn; } // And map the area immediately after the first block, with length pBufferLength, to that memory entry. virtualAddress = realAddress + pBufferLength; errorCount = 0; while (errorCount <10) { error = vm_map(mach_task_self(), &virtualAddress, pBufferLength, 0, FALSE, memoryEntry, 0, FALSE, VM_PROT_READ | VM_PROT_WRITE, VM_PROT_READ | VM_PROT_WRITE, VM_INHERIT_DEFAULT); if (error) { #ifdef mkDebug OS_Debug_Break("vm_map", error); #endif // TODO Retry from the beginning, instead of failing completely. There is a tiny (but > 0) probability that someone // will allocate this space out from under us. virtualAddress = NULL; goto errorReturn; } else { break; } errorCount++; } if (virtualAddress != realAddress + pBufferLength) { #ifdef mkDebug OS_Debug_Log("vm_map: didn't return correct address (%p vs %p)\n", realAddress + pBufferLength, virtualAddress); #endif goto errorReturn; } / /---------------------------------------------------------------------- ------------------ // FreeVirtualBuffer / /---------------------------------------------------------------------- ------------------ // We can conveniently deallocate both the vm_allocated memory and // the vm_mapped region at the same time. error = vm_deallocate(mach_task_self(), (vm_address_t)fBuffer, pBufferLength * 2); if (error) { #ifdef mkDebug OS_Debug_Break("vm_deallocate in dealloc", error); #endif } } This email sent to site_archiver(a)lists.apple.com

1 0

Re: IODefaultMatchCategory exists?
by Fernando Urbina 19 Sep '05

19 Sep '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Sep 19, 2005, at 7:37 AM, Dan Joumaa wrote: bInterfaceClass: 255 bInterfaceProtocol: 255 bInterfaceSubClass: 255 CFBundleIdentifier: com.ProjectNitro.PrismWireless idProduct: 8723 idVendor: 1643 IOClass: PrismWireless IOKitDebug: 65535 IOProviderClass: IOUSBDevice -- Fernando Urbina USB Technology Team Apple Computer, Inc. _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… And nothing else happens. My driver's init command isn't called, nothing. It just loads the kext and it's silent. For those who want to know what the used personality is: How did you come up with this personality? This is not a valid set of matching criteria for an IOUSBDevice. I suggest you look at Q&A 1076: http://developer.apple.com/qa/qa2001/qa1076.html. In the future, for USB questions, it would be worth your while to ask in the USB list @ http://lists.apple.com. This email sent to site_archiver(a)lists.apple.com

1 0

IODefaultMatchCategory exists?
by Dan Joumaa 19 Sep '05

19 Sep '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Hello, bInterfaceClass: 255 bInterfaceProtocol: 255 bInterfaceSubClass: 255 CFBundleIdentifier: com.ProjectNitro.PrismWireless idProduct: 8723 idVendor: 1643 IOClass: PrismWireless IOKitDebug: 65535 IOProviderClass: IOUSBDevice --ness _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… I am writing a wireless driver for Prism2 chipsets, though I'm having a small problem. I picked up some info from a slightly modified Prism2 configuration off USB Prober (such as configuration descriptors, etc.) Now that I've modified it to work with my plist, I kextload'd the driver. tail comes up with this: Sep 18 17:00:47 localhost kernel: OHCI Root Hub Simulation: family specific matching fails Sep 18 17:00:47 localhost kernel: IOUSBDevice: family specific matching fails Sep 18 17:00:47 localhost kernel: OHCI Root Hub Simulation: family specific matching fails Sep 18 17:00:47 localhost kernel: OHCI Root Hub Simulation: family specific matching fails Sep 18 17:00:47 localhost kernel: EHCI Root Hub Simulation: family specific matching fails Sep 18 17:00:47 localhost kernel: iPod : family specific matching fails Sep 18 17:00:47 localhost kernel: Matching service count = 1 Sep 18 17:00:47 localhost kernel: IOUSBCompositeDevice: match category IODefaultMatchCategory exists And nothing else happens. My driver's init command isn't called, nothing. It just loads the kext and it's silent. For those who want to know what the used personality is: This email sent to site_archiver(a)lists.apple.com

1 0

Re: Is my program leaking?
by Marc Van Olmen 19 Sep '05

19 Sep '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com hi, Here the code: static void * AllocateVirtualBuffer(OS_MemorySize pBufferLength) { kern_return_t error; vm_address_t originalAddress = NULL; vm_address_t realAddress = NULL; mach_port_t memoryEntry; vm_size_t memoryEntryLength; vm_address_t virtualAddress = NULL; OS_Int32 errorCount; // Success! return (void *)realAddress; #ifdef mkDebug // Here's a little test... *(char *)realAddress = '?'; if (*(char *)virtualAddress != '?') OS_Debug_Log("VirtualRingBuffer: Test 1: vm magic failed\n"); *(char *)(virtualAddress + 1) = '!'; if (*(char *)(realAddress + 1) != '!') OS_Debug_Log("VirtualRingBuffer: Test 2: vm magic failed\n"); #endif errorReturn: if (realAddress) vm_deallocate(mach_task_self(), realAddress, pBufferLength); if (virtualAddress) vm_deallocate(mach_task_self(), virtualAddress, pBufferLength); return NULL; } void FreeVirtualBuffer(void *fBuffer, UInt32 pBufferLength) { kern_return_t error; On Sep 19, 2005, at 2:12 AM, plumber wrote: Should I be worried? Is my application leaking memory? How can I trace where the leak is happening? take a look here /Developer/Applications/Performance Tools/ Best Regards _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… I figured out that it is "leaking" in some code I used for a long time, it is a code that allocates a buffer that is circular by using a sort of Virtual memory technique... I picked this code up from the internet and it seems that causes the leak... I'm guessing that the vm_deallocate is not releasing all the memory. The current leak doesn't seem to be traceable by mallocdebug because it isn't using any of those calls... I was using that tool before... / /----------------------------------------------------------------------- ----------------- // AllocateVirtualBuffer / /----------------------------------------------------------------------- ----------------- // We want to find where we can get 2 * fBufferLength bytes of contiguous address space. // So let's just allocate that space, remember its address, and deallocate it. // (This doesn't actually have to touch all of that memory so it's not terribly expensive.) error = vm_allocate(mach_task_self(), &originalAddress, 2 * pBufferLength, TRUE); if (error) { #ifdef mkDebug OS_Debug_Break("vm_allocate initial chunk", error); #endif return NULL; } error = vm_deallocate(mach_task_self(), originalAddress, 2 * pBufferLength); if (error) { #ifdef mkDebug OS_Debug_Break("vm_deallocate initial chunk", error); #endif return NULL; } // Then allocate a "real" block of memory at the same address, but with the normal pBufferLength. realAddress = originalAddress; error = vm_allocate(mach_task_self(), &realAddress, pBufferLength, FALSE); if (error) { #ifdef mkDebug OS_Debug_Break("vm_allocate real chunk", error); #endif return NULL; } if (realAddress != originalAddress) { #ifdef mkDebug OS_Debug_Log("allocateVirtualBuffer: vm_allocate 2nd time didn't return same address (%p vs %p)\n", originalAddress, realAddress); #endif goto errorReturn; } // Then make a memory entry for the area we just allocated. memoryEntryLength = pBufferLength; error = mach_make_memory_entry(mach_task_self(), &memoryEntryLength, realAddress, VM_PROT_READ | VM_PROT_WRITE, &memoryEntry, NULL); if (error) { #ifdef mkDebug OS_Debug_Break("mach_make_memory_entry", error); #endif goto errorReturn; } if (!memoryEntry) { #ifdef mkDebug OS_Debug_Log("mach_make_memory_entry: returned memoryEntry of NULL\n"); #endif goto errorReturn; } if (memoryEntryLength != pBufferLength) { #ifdef mkDebug OS_Debug_Log("mach_make_memory_entry: size changed (from %0x to %0x)\n", pBufferLength, memoryEntryLength); #endif goto errorReturn; } // And map the area immediately after the first block, with length pBufferLength, to that memory entry. virtualAddress = realAddress + pBufferLength; errorCount = 0; while (errorCount <10) { error = vm_map(mach_task_self(), &virtualAddress, pBufferLength, 0, FALSE, memoryEntry, 0, FALSE, VM_PROT_READ | VM_PROT_WRITE, VM_PROT_READ | VM_PROT_WRITE, VM_INHERIT_DEFAULT); if (error) { #ifdef mkDebug OS_Debug_Break("vm_map", error); #endif // TODO Retry from the beginning, instead of failing completely. There is a tiny (but > 0) probability that someone // will allocate this space out from under us. virtualAddress = NULL; goto errorReturn; } else { break; } errorCount++; } if (virtualAddress != realAddress + pBufferLength) { #ifdef mkDebug OS_Debug_Log("vm_map: didn't return correct address (%p vs %p)\n", realAddress + pBufferLength, virtualAddress); #endif goto errorReturn; } / /----------------------------------------------------------------------- ----------------- // FreeVirtualBuffer / /----------------------------------------------------------------------- ----------------- // We can conveniently deallocate both the vm_allocated memory and // the vm_mapped region at the same time. error = vm_deallocate(mach_task_self(), (vm_address_t)fBuffer, pBufferLength * 2); if (error) { #ifdef mkDebug OS_Debug_Break("vm_deallocate in dealloc", error); #endif } } This email sent to site_archiver(a)lists.apple.com

1 0

Re: Is my program leaking?
by plumber 19 Sep '05

19 Sep '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Should I be worried? Is my application leaking memory? How can I trace where the leak is happening? take a look here /Developer/Applications/Performance Tools/ Best Regards _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0

Is my program leaking?
by Marc Van Olmen 18 Sep '05

18 Sep '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com hi, I have a program that every minute Does something the following: Open QuickTime vOut playsmovie. Closes QuickTime vOut. The system performs a little bit slower in that case..... Should I be worried? Is my application leaking memory? How can I trace where the leak is happening? regards marc _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… Now the memory (real and VM) of my Application stays stable also the Kernel memory real and VM stays stable. But I noticed that Activity Memory is telling me that global Free memory is going down.... Decrease happens until 30MB and then pageouts starts to happen.... When I quit the application the Global Free memory is back to it's level from before my app was running. This email sent to site_archiver(a)lists.apple.com

1 0

/dev/random in Darwin 6 vs fcntl
by Walt Pawley 13 Sep '05

13 Sep '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com I've been messing with Darwinports to provide more modern versions of some functions under Mac OS X.2.x, ie. Jaguar. I managed to get BIND9 running but not without a few "loose ends". In particular, it seems that in named, version 9.3.1, an fcntl call to make /dev/random access for named non-blocking, fails. As near as I can tell, this is because Darwin<7 does not have any support for fcntl in the /dev/random code. Perhaps someone who knows what what they're doing with all this stuff could verify this, or not? Perhaps the same or similar someone(s) could advise whether this is a real problem or a basically ignorable information message (it doesn't keep named from running or doing it's basic work as near as I can tell from some small amount of testing)? if it is a problem, would it be a major problem to fit a more modern /dev/random onto an existing Mac OS X.2.x? Perhaps using mknod? -- Walter M. Pawley <walt(a)wump.org> Wump Research & Company 676 River Bend Road, Roseburg, OR 97470 541-672-8975 _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0

net/if_tun.h on Darwin?
by André-John Mas 12 Sep '05

12 Sep '05

site_archiver(a)lists.apple.com Delivered-To: Darwin-dev(a)lists.apple.com Hi, Andre _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… I wanted to use the TSP Client (http://www.hexago.com/index.php? pgID=27) to tunnel IPv6 through my router, but it seems that in the Darwin port this feature is not supported. Trying to build with FreeBSD as the target it would seem that net/if_tun.h is not available on Darwin. Is there anything that I can install to provide this dependency, or any equivalent that can be used instead? This email sent to site_archiver(a)lists.apple.com

1 0

Re: thread-local storage, especially on x86
by Alexei Kosut 12 Sep '05

12 Sep '05

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com The Darwin kernel offers a function - thread_set_user_ldt() - which allows user-level code to set up a segment selector which can be loaded into %fs and used to address a specified address range. (The name of this function may be a little misleading, since it seems that that segment descriptor is shared by all threads in the calling process.) Comments prefacing that function suggest the possibility (if I understand correctly) of allocating and maintaing this selector on a per-thread basis; it's much easier for me to imagine that behavior being useful than it is for me to imagine a use for the current behavior. Alexei _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… On Sep 11, 2005, at 10:35 AM, Gary Byers wrote: thread_set_user_ldt is per-thread, not per-process. The kernel re- uses the same selector for multiple threads, but changes the segment descriptors in the LDT during a context switch. Each thread gets its own area. Note that the thread_set_user_ldt kernel function and it's Mach trap userspace equivalent are not part of the Mac OS X or Darwin supported API, and may change from release to release. This email sent to site_archiver(a)lists.apple.com

1 0