February 2005 - Darwin-kernel

Re: xnu-517.11.1 won't compile.
by Kevin Van Vechten 14 Feb '05

14 Feb '05

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com - Kevin On Feb 10, 2005, at 3:26 PM, Josh Ellithorpe wrote: Anyone gotten this to work... or have a hint for me =) Josh Ellithorpe quest(a)mac.com _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… This is indeed a problem with the source tarball. The osfmk/ppc/machine_routines_asm.s source file is incomplete. I'll try to post a fix with the complete version of this file tomorrow. I have compiled every version of xnu since 10.3 came out however xnu-517.11.1 just won't build for me. All my dependencies are up to date however I get the following error. If someone can help me with this I would really apperciate it. /usr/bin/strip -S -x /usr/local/src/xnu-517.11.1/BUILD/obj/RELEASE_PPC/./mach_kernel.sys -o /usr/local/src/xnu-517.11.1/BUILD/obj/RELEASE_PPC/./mach_kernel; /usr/bin/ld: Undefined symbols: _cacheDisable _cacheInit _clock_get_uptime _cpu_number _mach_absolute_time _machine_idle_ret _ml_get_timebase _ml_read_temp _ml_scom_read _ml_scom_write _ml_set_processor_speed_dfs _ml_set_processor_speed_dpll _ml_set_processor_speed_powertune _ml_thrm_init _set_machine_current_act _current_act _current_thread make[2]: *** [do_build_mach_kernel] Error 1 make[1]: *** [build_all] Error 2 make: *** [all] Error 2 I even grabbed xnu-517.9.5 from the 10.3.7 repository and that built fine... the only dependency that I saw changed between 10.3.7 and 10.3.8 was IOKitUser-184 which I already installed. This email sent to site_archiver(a)lists.apple.com

1 0

Cannot mprotect() dylib code?
by Lassi A.Tuura 11 Feb '05

11 Feb '05

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Hi, The code is being ported from Linux platforms where this works. Lassi -- Children begin by loving their parents. After a time they judge them. Rarely, if ever, do they forgive them. --Oscar Wilde _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… We have a debugging type library that parses and modifies the machine code on the fly for certain functions, mainly to insert a jump to "capture" functions which do some special stuff then invoke the original code. The run-time patching appears to work on functions from the main executable, but not on functions from dylibs -- mprotect() on a page of code from the latter always fails, errno is set to 13 (EACCESS, Permission denied). I also tried vm_protect(mach_task_self(), ...), it also fails. How do I modify code from dylibs on OS X at run-time? Presumably there's a way to do this since fix and continue has been implemented :-) This email sent to site_archiver(a)lists.apple.com

1 0

Re: Darwin-kernel Digest, Vol 2, Issue 34
by Jason Seng 11 Feb '05

11 Feb '05

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Hi Sam, Message: 1 Date: Thu, 10 Feb 2005 15:33:56 -0500 From: Sam Hopkins <sah(a)coraid.com> Subject: AoE driver To: darwin-kernel(a)lists.apple.com, darwin-drivers(a)lists.apple.com Message-ID: <532e610ca32addbf5dca1ffc99229550(a)coraid.com> Content-Type: text/plain; charset="US-ASCII" As for the rexmit timer, I need a way for a function to be called every X ms. So far all I've found towards this end is a way to use IOLock instead of timeout. I don't think this helps me, but perhaps I'm missing something. http://developer.apple.com/samplecode/TimerEventSample/TimerEventSample.sit Cheers, -- J-Sun Seng Networking & Communications / Hardware Developer Technical Support Apple Computer, Inc. #1 Infinite Loop MS: 303-2T Cupertino, CA 95014 _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… I think IOTimerEventSource might fit your need. Have a look at the TimerEventSample samplecode at: This email sent to site_archiver(a)lists.apple.com

1 0

xnu kernel 517.11.1 compile errors
by Maarten Engelen 11 Feb '05

11 Feb '05

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Today has not been a good day. :( I'm trying to compile the xnu 517.11.1 kernel, but I keep getting compile errors. First I got the following 'undefined symbols' error: /usr/bin/ld: Undefined symbols: _cacheDisable _cacheInit _clock_get_uptime _cpu_number _mach_absolute_time _machine_idle_ret _ml_get_timebase _ml_read_temp _ml_scom_read _ml_scom_write _ml_set_processor_speed_dfs _ml_set_processor_speed_dpll _ml_set_processor_speed_powertune _ml_thrm_init _set_machine_current_act _current_act _current_thread make[2]: *** [do_build_mach_kernel] Error 1 make[1]: *** [build_all] Error 2 make: *** [all] Error 2 That was no success. I looked up the symbols ld mentions and I found that the file osfmk/ppc/machine_routines_asm.s is three times as small in xnu 517.11.1 compared to 517.9.5. A lot of code is missing in this file. After merging the two files and adding the missing code to 517.11.1 I tried to compile. It still gives me an undefined symbols error, but it is not as long as before: /usr/bin/ld: Undefined symbols: _ml_set_processor_speed_dfs _ml_set_processor_speed_dpll _ml_set_processor_speed_powertune make[2]: *** [do_build_mach_kernel] Error 1 make[1]: *** [build_all] Error 2 make: *** [all] Error 2 These functions are nowhere to be found in the 517.11.1 and the 517.9.5 code. I presume that the machine_routines_asm.s file is messed up pretty good in the source tarball. I tried the tarball from apple.com, opendarwin.org and the cvs version from opendarwin.org. Nothing works. :( Am I doing something wrong? In case the source tree is broken, is Apple aware of this? Any help is greatly appreciated. Maarten _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… This email sent to site_archiver(a)lists.apple.com

1 0

xnu-517.11.1 won't compile.
by Josh Ellithorpe 10 Feb '05

10 Feb '05

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com I have compiled every version of xnu since 10.3 came out however xnu-517.11.1 just won't build for me. All my dependencies are up to date however I get the following error. If someone can help me with this I would really apperciate it. /usr/bin/strip -S -x /usr/local/src/xnu-517.11.1/BUILD/obj/RELEASE_PPC/./mach_kernel.sys -o /usr/local/src/xnu-517.11.1/BUILD/obj/RELEASE_PPC/./mach_kernel; /usr/bin/ld: Undefined symbols: _cacheDisable _cacheInit _clock_get_uptime _cpu_number _mach_absolute_time _machine_idle_ret _ml_get_timebase _ml_read_temp _ml_scom_read _ml_scom_write _ml_set_processor_speed_dfs _ml_set_processor_speed_dpll _ml_set_processor_speed_powertune _ml_thrm_init _set_machine_current_act _current_act _current_thread make[2]: *** [do_build_mach_kernel] Error 1 make[1]: *** [build_all] Error 2 make: *** [all] Error 2 I even grabbed xnu-517.9.5 from the 10.3.7 repository and that built fine... the only dependency that I saw changed between 10.3.7 and 10.3.8 was IOKitUser-184 which I already installed. Anyone gotten this to work... or have a hint for me =) Josh Ellithorpe quest(a)mac.com _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… This email sent to site_archiver(a)lists.apple.com

1 0

Re: workloop thread priority = 81, lowest in kernel threads.
by Quinn 10 Feb '05

10 Feb '05

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com I read the IOKit funda.pdf , one thing is not clear to me is: TimerInterruptSource are handled using thread_call_enter. so if my timeout handler function is executing, is it possible that secondary interrupt handler can be served on other processor? S+E -- Quinn "The Eskimo!" <http://www.apple.com/developer/> Apple Developer Technical Support * Networking, Communications, Hardware _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… At 21:58 +0000 10/2/05, Parav Pandit wrote: All workloop event sources (timers, commands, interrupts) acquire the mutex associated with the workloop before executing the callback (in workloop parlance they "close the gate"). Thus, regardless of thread priorities, workloop callbacks are always serialised. To take your example, if CPU 0 is executing a timer event source callback, it will have closed the gate. If the system receives an interrupt at that time, it will start up the workloop's thread to execute the interrupt callback. Assume this thread runs on CPU 1. That thread will try to acquire the workloop's mutex before calling the interrupt callback. That mutex is currently held by the thread on CPU 0, so the workloop thread will block. Ergo serialisation. This email sent to site_archiver(a)lists.apple.com

1 0

Re: AoE driver
by Quinn 10 Feb '05

10 Feb '05

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com The I/O Kit integration would go as follows. o Subclass IOATAController and include all of your core logic in that subclass. $ ioreg [...] | +-o pci@f4000000 <class IOPlatformDevice, registered, matc[...] | | +-o AppleMacRiscPCI <class AppleMacRiscPCI, !registered,[...] | | +-o ata-6@D <class IOPCIDevice, registered, matched, a[...] | | | +-o AppleKauaiATA <class AppleKauaiATA, !registered,[...] | | | +-o ATADeviceNub@0 <class ATADeviceNub, registered[...] | | | +-o IOATABlockStorageDriver <class IOATABlockSto[...] | | | +-o IOATABlockStorageDevice <class IOATABlockS[...] | | | +-o IOBlockStorageDriver <class IOBlockStora[...] | | | +-o IBM-IC35L120AVVA07-0 Media <class IOMe[...] | | | +-o IOMediaBSDClient <class IOMediaBSDCl[...] | | | +-o IOApplePartitionScheme <class IOAppl[...] | | | +-o Apple@1 <class IOMedia, registered[...] | | | | +-o IOMediaBSDClient <class IOMediaB[...] | | | +-o Macintosh@2 <class IOMedia, regist[...] | | | | +-o IOMediaBSDClient <class IOMediaB[...] | | | +-o Macintosh@3 <class IOMedia, regist[...] | | | | +-o IOMediaBSDClient <class IOMediaB[...] | | | +-o Macintosh@4 <class IOMedia, regist[...] | | | | +-o IOMediaBSDClient <class IOMediaB[...] | | | +-o Macintosh@5 <class IOMedia, regist[...] | | | | +-o IOMediaBSDClient <class IOMediaB[...] | | | +-o Macintosh@6 <class IOMedia, regist[...] | | | | +-o IOMediaBSDClient <class IOMediaB[...] | | | +-o Macintosh@7 <class IOMedia, regist[...] | | | | +-o IOMediaBSDClient <class IOMediaB[...] | | | +-o Patch Partition@8 <class IOMedia, [...] | | | | +-o IOMediaBSDClient <class IOMediaB[...] | | | +-o X1@9 <class IOMedia, registered, m[...] | | | | +-o IOMediaBSDClient <class IOMediaB[...] | | | +-o X2@10 <class IOMedia, registered, [...] | | | | +-o IOMediaBSDClient <class IOMediaB[...] | | | +-o Guy Smiley@11 <class IOMedia, regi[...] | | | | +-o IOMediaBSDClient <class IOMediaB[...] | | | +-o Snuffy@12 <class IOMedia, register[...] | | | | +-o IOMediaBSDClient <class IOMediaB[...] | | | +-o BigBird@13 <class IOMedia, registe[...] | | | +-o IOMediaBSDClient <class IOMediaB[...] [...] The "AppleKauaiATA" node is the ATA controller, a subclass of IOATAController. $ IOPrintSuperClasses AppleKauaiATA Looking for superclasses of AppleKauaiATA OSObject MacIOATA IOATAController IOService IORegistryEntry The "ATADeviceNub" is the ATA nub, a subclass of IOATANub. $ IOPrintSuperClasses ATADeviceNub Looking for superclasses of ATADeviceNub OSObject IOATADevice IOService IORegistryEntry Everything else is provided via the magic of I/O Kit. <http://developer.apple.com/samplecode/IOPrintSuperClasses/IOPrintSuperClass…> S+E -- Quinn "The Eskimo!" <http://www.apple.com/developer/> Apple Developer Technical Support * Networking, Communications, Hardware _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… At 16:20 -0500 10/2/05, Dan Markarian wrote: We do have an ATA block storage interface at lower level in case that is more of a fit. I don't have a lot of experience with ATAoE, but if it represents an ATA interface to the system, you probably want to implement an ATA controller, rather than a generic block storage device. This is what we recommend for similar situations, such as iSCSI. o Have your ATA controller match on IOResources. This is an abstract node in the I/O Registry that we provide so that software-based devices can load. o Have your ATA controller publish nubs into the registry, one for each ATA device. These nubs would typically be a subclass of IOATADevice. These are the points where the higher levels of the block storage system connect to your component. If you look at the I/O Registry for a typical Mac, you'll see this in action. Here's a dump from my machine. The "ata-6" node represents the PCI device that is the ATA hardware. This is the provider for the ATA controller. In your case you don't have physical hardware, so this would be IOResources. btw The "ioreg" tool is built in to the system. You can get the IOPrintSuperClasses tool from the developer web site. This email sent to site_archiver(a)lists.apple.com

1 0

workloop thread priority = 81, lowest in kernel threads.
by Parav Pandit 10 Feb '05

10 Feb '05

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Hi, Let me ask u a simple questions. 1. workloop is to handle interrupts, timer interrupts and the commands. some important work , asked by the primary interupts and other asked by commands. Right? Assuming this is right. So when work is important workloop thread should have higher priority than any other kernel thead. Workloop thread is created with IOCreateThread API. IOCreateThread creates the kernel thread with priority of 81. thread_call_enter() has high priority as 94. so if I have a thread created with thread_call_enter() than it will get high priority than my secondary interrupt handler. is it true? We can see this from remote-gdb. I read the IOKit funda.pdf , one thing is not clear to me is: TimerInterruptSource are handled using thread_call_enter. so if my timeout handler function is executing, is it possible that secondary interrupt handler can be served on other processor? or CloseGate() is executed after timeout hanlder is finished? and seconadary interrupt will be executed. Regards, Parav Pandit ________________________________________________________________________ Yahoo! India Matrimony: Find your life partner online Go to: http://yahoo.shaadi.com/india-matrimony _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… This email sent to site_archiver(a)lists.apple.com

1 0

Re: AoE driver
by Dan Markarian 10 Feb '05

10 Feb '05

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Hi Sam, Dan On 10 Feb 2005, at 3:33 PM, Sam Hopkins wrote: Hello all, I'm writing an AoE (ATA over Ethernet) driver for Darwin that will enable Ethernet attached ATA devices like the Coraid EtherDrive blade (coraid.com) to be used. Our current linux and freebsd drivers have three main components: o block device interface (per AoE device) o network retransmit timer (per AoE device) o network packet handler for ethernet type 0x88a2 The AoE driver is primarily responsible for translating block read/write requests into AoE frames, retransmitting them when responses don't come back in a certain time, and acknowledging the requests as complete to the system when they do. I've read a lot of documentation on the IO Kit system, but its internals are sufficiently different from what we've previously done that I can't see a clear way to proceed. I'm confused by the driver/nub relationship. I think I need to write a provider for something in the IOStorageFamily, but I can't figure out what, nor am I clear on what my interfaces are for block reads/writes. Perhaps someone can point to some clarifying documentation here (not "IO Kit Fundamentals" -- I've read that). As for the rexmit timer, I need a way for a function to be called every X ms. So far all I've found towards this end is a way to use IOLock instead of timeout. I don't think this helps me, but perhaps I'm missing something. Lastly, I need to register an Ethernet protocol handler. I read the paper on NKEs, but am not sure whether I need a protocol handler NKE or a data link NKE. If the former, is there an ethernet domain I use in net_add_proto? If the latter, it seems I need to register a separate filter per interface? Any help would be appreciated. Cheers, Sam _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… IOBlockStorageDevice in IOStorageFamily is the high level interface for a block storage device. It is independent of transport. It accepts block reads and writes across doAsyncReadWrite(). These are abstract methods left up to the developer to implement as appropriate to their transport. We do have an ATA block storage interface at lower level in case that is more of a fit. IOBlockStorageDevice is the device object, that is, the nub. The driver object is what publishes one device object for each hardware device. It may be a subclass of IOService in the case of an agnostic transport family. It contains the guts of your driver implementation. It is the device object that represents the hardware device and what we communicate across when we see someone publish one. _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/ markarian%40apple.com This email sent to markarian(a)apple.com This email sent to site_archiver(a)lists.apple.com

1 0

Re: AoE driver
by Josh Graessley 10 Feb '05

10 Feb '05

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Please don't cross post. -josh On Feb 10, 2005, at 12:33 PM, Sam Hopkins wrote: Hello all, I'm writing an AoE (ATA over Ethernet) driver for Darwin that will enable Ethernet attached ATA devices like the Coraid EtherDrive blade (coraid.com) to be used. Our current linux and freebsd drivers have three main components: o block device interface (per AoE device) o network retransmit timer (per AoE device) o network packet handler for ethernet type 0x88a2 The AoE driver is primarily responsible for translating block read/write requests into AoE frames, retransmitting them when responses don't come back in a certain time, and acknowledging the requests as complete to the system when they do. I've read a lot of documentation on the IO Kit system, but its internals are sufficiently different from what we've previously done that I can't see a clear way to proceed. I'm confused by the driver/nub relationship. I think I need to write a provider for something in the IOStorageFamily, but I can't figure out what, nor am I clear on what my interfaces are for block reads/writes. Perhaps someone can point to some clarifying documentation here (not "IO Kit Fundamentals" -- I've read that). As for the rexmit timer, I need a way for a function to be called every X ms. So far all I've found towards this end is a way to use IOLock instead of timeout. I don't think this helps me, but perhaps I'm missing something. Lastly, I need to register an Ethernet protocol handler. I read the paper on NKEs, but am not sure whether I need a protocol handler NKE or a data link NKE. If the former, is there an ethernet domain I use in net_add_proto? If the latter, it seems I need to register a separate filter per interface? Any help would be appreciated. Cheers, Sam _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.… Not sure about the IOKit side of things, but I might be able to help on the third part, getting packets from an interface. The easiest way to get packets is to register a protocol with the interface. When you do this, you can specify a demux descriptor that describes the types of frames you're interested in receiving. Grab the darwin xnu sources and look at xnu/bsd/net/ether_inet_pr_module.c. In addition to the demux descriptor, you also specify a function that will be called when a packet matching that type is received on the interface. The functions you will need are defined in dlil.h. Look for something like dlil_proto_attach or dlil_attach_protocol. The kernel is changing drastically between Panther and Tiger. The changes to implement improved SMP scalability necessitates a number of changes to kernel extensions. Any work on the networking side that you do for Panther will need to be modified significantly to work on Tiger. IOKit already has a much better abstraction model that insulates most IOKit drivers from these changes. _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/jgraessley% 40apple.com This email sent to jgraessley(a)apple.com This email sent to site_archiver(a)lists.apple.com

1 0