December 2005 - Darwin-kernel

Re: kernel dump server
by Peter Lovell 27 Dec '05

27 Dec '05

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com On Dec 27, 2005, at 5:25 PM, Andrew Gallatin wrote: I've worked around this in the past by deleting the default route, and then setting the default route to point to the machine running the panic server.. This obviously will not work in all cases, but it may be enough to help you. Hi Drew, that will be OK for some scenarios but not for the general ones we have. Thanks.....Peter _______________________________________________ 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.… Peter Lovell writes: Hi Derek, what I'm seeing is that the client is not sending any arp request at all. It just starts out by sending UDP packets from 1069 to port 1069 on server, using the default router. And looking at kdp_udp.c, I don't see any ARP resolution code. It just takes whatever is the default gateway (set at kdp_set_gateway_mac, called from rt_setgate in route.c) and sends the request there. SInce we're on a managed network, I'm not at all optimistic about having any configuration changes made on the router. Drew The simple approach, which also works for only some of our usage, is to put the clients behind a mini-router. That will then forward the packets properly, which the "big network" router refuses to do. In the slightly longer term we'll just fix the kernel. We're already custom-building so this will not be the massive pain it would be for most people. This email sent to site_archiver(a)lists.apple.com

1 0

Re: kernel dump server
by Andrew Gallatin 27 Dec '05

27 Dec '05

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Peter Lovell writes: > Hi Derek, > what I'm seeing is that the client is not sending any arp request at all. It just starts out by sending UDP packets from 1069 to port 1069 on server, using the default router. > > And looking at kdp_udp.c, I don't see any ARP resolution code. It just takes whatever is the default gateway (set at kdp_set_gateway_mac, called from rt_setgate in route.c) and sends the request there. SInce we're on a managed network, I'm not at all optimistic about having any configuration changes made on the router. > I've worked around this in the past by deleting the default route, and then setting the default route to point to the machine running the panic server.. This obviously will not work in all cases, but it may be enough to help you. Drew _______________________________________________ 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: kernel dump server
by Peter Lovell 27 Dec '05

27 Dec '05

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Hi Derek, what I'm seeing is that the client is not sending any arp request at all. It just starts out by sending UDP packets from 1069 to port 1069 on server, using the default router. And looking at kdp_udp.c, I don't see any ARP resolution code. It just takes whatever is the default gateway (set at kdp_set_gateway_mac, called from rt_setgate in route.c) and sends the request there. SInce we're on a managed network, I'm not at all optimistic about having any configuration changes made on the router. Thanks for this. I'm off to try talking to the network folks. Regards.....Peter On Tuesday, December 27, 2005, at 10:21AM, Derek Kumar <drk(a)apple.com> wrote: >On Dec 26, 2005, at 7:16 PM, Peter Lovell wrote: > >> Hi all, >> I'm setting up a kernel dump server, as described in technote 2118 >> (thanks Quinn). >> >> However, I'm having a problem in that the client seems to be unable >> to find the server. Before banging on this some more in the >> morning, I thought I'd just check to see if there is any change >> needed operating in Tiger rather than Panther, with launchd >> starting to replace the tasks previously handled by xinetd. I don't >> think there should be, and the server does indeed seem to have a >> UDP listening port 1069 as I'd expect. The client tries fifteen >> times (or so) and then just times out. >> >> I suspect that there may be local switch/router issues related to >> UDP (although both are on the same local network) but thought I'd >> just check the possibility of changes for Tiger. >I'll second Justin Walker's recommendations, and note a couple of >known issues, in the event that they apply to your configuration: > >A late Panther update added the ability to bind the kernel debugger >to a non-en0 interface via a kernel boot-arg; however, that also >appears to have introduced a race where the debugger can occasionally >match to the wrong interface; you should be able to work around this >by adding the "kdp_match_name=en0" boot-arg (if you wish to debug >over en0, which is the default). (Radar 3844652). > >Are you on a NAT-ted network? If, for some reason, the ARP resolution >broadcast for the server fails, the debugger will attempt to route >the crashdump packets via the cached default route (the debugger's >"network stack" is a fairly minimalist, polled implementation, and >doesn't query routing tables and so on, which may or may not be valid >post-panic), and if the router doesn't forward packets arriving at >the internal interface addressed to internal nodes, you may see a >timeout. You can either try and determine why the resolution fails, >or reconfigure your router as appropriate. >A future update should address/ameliorate these issues. Feel free to >contact me with a packet trace if you can't determine the source of >the problem. > >Derek > > > _______________________________________________ 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: kernel dump server
by Derek Kumar 27 Dec '05

27 Dec '05

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com On Dec 26, 2005, at 7:16 PM, Peter Lovell wrote: Derek _______________________________________________ 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.… Hi all, I'm setting up a kernel dump server, as described in technote 2118 (thanks Quinn). However, I'm having a problem in that the client seems to be unable to find the server. Before banging on this some more in the morning, I thought I'd just check to see if there is any change needed operating in Tiger rather than Panther, with launchd starting to replace the tasks previously handled by xinetd. I don't think there should be, and the server does indeed seem to have a UDP listening port 1069 as I'd expect. The client tries fifteen times (or so) and then just times out. I suspect that there may be local switch/router issues related to UDP (although both are on the same local network) but thought I'd just check the possibility of changes for Tiger. I'll second Justin Walker's recommendations, and note a couple of known issues, in the event that they apply to your configuration: A late Panther update added the ability to bind the kernel debugger to a non-en0 interface via a kernel boot-arg; however, that also appears to have introduced a race where the debugger can occasionally match to the wrong interface; you should be able to work around this by adding the "kdp_match_name=en0" boot-arg (if you wish to debug over en0, which is the default). (Radar 3844652). Are you on a NAT-ted network? If, for some reason, the ARP resolution broadcast for the server fails, the debugger will attempt to route the crashdump packets via the cached default route (the debugger's "network stack" is a fairly minimalist, polled implementation, and doesn't query routing tables and so on, which may or may not be valid post-panic), and if the router doesn't forward packets arriving at the internal interface addressed to internal nodes, you may see a timeout. You can either try and determine why the resolution fails, or reconfigure your router as appropriate. A future update should address/ameliorate these issues. Feel free to contact me with a packet trace if you can't determine the source of the problem. This email sent to site_archiver(a)lists.apple.com

1 0

Re: kernel dump server
by Justin C. Walker 27 Dec '05

27 Dec '05

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com On Dec 26, 2005, at 16:16 , Peter Lovell wrote: I've not tried this, so I'm just asking obvious questions :-} Cheers, Justin -- Justin C. Walker, Curmudgeon at Large Institute for General Semantics ----------- My wife 'n kids 'n dogs are gone, I can't get Jesus on the phone, But Ol' Milwaukee's Best is my best friend. ----------- _______________________________________________ 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.… Hi all, I'm setting up a kernel dump server, as described in technote 2118 (thanks Quinn). However, I'm having a problem in that the client seems to be unable to find the server. Before banging on this some more in the morning, I thought I'd just check to see if there is any change needed operating in Tiger rather than Panther, with launchd starting to replace the tasks previously handled by xinetd. I don't think there should be, and the server does indeed seem to have a UDP listening port 1069 as I'd expect. The client tries fifteen times (or so) and then just times out. I suspect that there may be local switch/router issues related to UDP (although both are on the same local network) but thought I'd just check the possibility of changes for Tiger. What does tcpdump show? Are the attempts showing up as arps or as udp packets that are being ignored? Does tcpdump on the server show the packets being received? Don't forget to disable promiscuous mode. This email sent to site_archiver(a)lists.apple.com

1 0

kernel dump server
by Peter Lovell 27 Dec '05

27 Dec '05

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Thanks.....Peter _______________________________________________ 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.… Hi all, I'm setting up a kernel dump server, as described in technote 2118 (thanks Quinn). However, I'm having a problem in that the client seems to be unable to find the server. Before banging on this some more in the morning, I thought I'd just check to see if there is any change needed operating in Tiger rather than Panther, with launchd starting to replace the tasks previously handled by xinetd. I don't think there should be, and the server does indeed seem to have a UDP listening port 1069 as I'd expect. The client tries fifteen times (or so) and then just times out. I suspect that there may be local switch/router issues related to UDP (although both are on the same local network) but thought I'd just check the possibility of changes for Tiger. This email sent to site_archiver(a)lists.apple.com

1 0

OSMalloc Fails (returns 0)
by Russ Seehafer 25 Dec '05

25 Dec '05

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Domainkey-signature: a=rsa-sha1; q=dns; c=nofws; s=beta; d=gmail.com; h=received:mime-version:to:message-id:content-type:from:subject:date:x-mailer:sender; b=YC8zIF+HsrJpNG3gJ6rKM3Kl0WcSVVQHpUvjcBIp3OEnbU425tcVSpE/4QcHm5O8z+sIXTbWOPaJ8pR90wNpVTlD0oSEjipqOwt1KQc01YOPBpvqPWWuRwhqWuAikMg8GdU6PtdHNRmA7wggtYYSGIE75GIbRLtTsu3F1D3jagg= _______________________________________________ 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: Threading and Synchronization in Mass storage driver
by Herb Petschauer 24 Dec '05

24 Dec '05

site_archiver(a)lists.apple.com Delivered-To: Darwin-kernel(a)lists.apple.com Domainkey-signature: a=rsa-sha1; q=dns; c=nofws; s=beta; d=gmail.com; h=received:message-id:date:from:to:subject:in-reply-to:mime-version:content-type:content-transfer-encoding:content-disposition:references; b=uIDVgATCTK+30Cb3nU0RHq1Hw7QKAteuJTFiJu2nGDJF+j/B01p74WNrQNwjmbj8Y5WxCdNZsKG/6i0s9MrwUkrP0eU9quMhiYNpdET6k9hWfX3t9W6CbGFJ7L/t86Q/LYomtOUBmIDaOvUpRdbjazO51q56r2qdB1ABAHNy1SQ= On 12/16/05, Herb Petschauer <bogvardi(a)gmail.com> wrote: > Your panic begins at: > > /Users/yogeshp/PROJECTS/SnapShot/DarwinCode/xnu-517.7.21/iokit/Kernel/IOLib.c:140 > > -H. > > > On 16 Dec 2005 19:19:35 +0530, Yogesh P <pahilwan.yogesh(a)spsoftindia.com> wrote: > > Hi Herb, > > > > I have setup a 2 machine debugging to trace out the exact cause of the > > crash while using the IOLockSleep() synchronization primitive. > > > > During 2 machine debugging, I have found that the threads get created > > successfully but when the call goes to IOLockSleep(request_lock, (void*) > > thread_id, false)the driver gets crashed. > > > > I put the stack backtrace for the same crash as follows: > > > > gdb() > > Breakpoint 1, com_CoreSnap_driver::init(OSDictionary*) (this=0x1c7be20, > > properties=0x1d74a60) at Spsoft_CoreSnapDriver.cpp:64 > > 64 IOLog("\n Inside Init function...\n"); > > (gdb) n > > Current language: auto; currently c++ > > Line number 65438 out of range; Spsoft_CoreSnapDriver.cpp has 1568 > > lines. > > (gdb) n > > Breakpoint 1, com_CoreSnap_driver::init(OSDictionary*) (this=0x1b110a0, > > properties=0x1d74880) at Spsoft_CoreSnapDriver.cpp:64 > > 64 IOLog("\n Inside Init function...\n"); > > (gdb) n > > Line number 65438 out of range; Spsoft_CoreSnapDriver.cpp has 1568 > > lines. > > gdb) n > > 78 for(i = 0; i < NUM_HANDLER_THREADS; i++) > > (gdb) n > > 80 thr_id[i] = i; > > (gdb) n > > 82 IOLog("\nInit():: thr_id = %d\n", thr_id[i]); > > (gdb) n > > Line number 65438 out of range; Spsoft_CoreSnapDriver.cpp has 1568 > > lines. > > (gdb) n > > 84 IOThread thread_t = > > IOCreateThread(com_CoreSnap_driver::handle_request_loop, (void*) > > &thr_id[i]); > > (gdb) n > > 86 if(thread_t) > > (gdb) n > > 88 IOLog("\nInit():: Thread created successfully\n"); > > (gdb) n > > Line number 65438 out of range; Spsoft_CoreSnapDriver.cpp has 1568 > > lines. > > (gdb) n > > Program received signal SIGTRAP, Trace/breakpoint trap. > > Debugger (message=0x2d4440 "panic") at > > /Users/yogeshp/PROJECTS/SnapShot/DarwinCode/xnu-517.7.21/osfmk/ppc/model_dep.c:627 > > 627 TRAP_DEBUGGER; > > (gdb) bt > > #0 Debugger (message=0x2d4440 "panic") at > > /Users/yogeshp/PROJECTS/SnapShot/DarwinCode/xnu-517.7.21/osfmk/ppc/model_dep.c:627 > > #1 0x00034d84 in panic (str=0x2d43c8 "{%d} Assertion failed: file > > \"%s\", line %d: %s\n") at > > /Users/yogeshp/PROJECTS/SnapShot/DarwinCode/xnu-517.7.21/osfmk/kern/debug.c:198 > > #2 0x0004201c in assert_wait (event=0x0, interruptible=0) at > > /Users/yogeshp/PROJECTS/SnapShot/DarwinCode/xnu-517.7.21/osfmk/kern/sched_prim.c:619 > > #3 0x00042424 in thread_sleep_mutex (event=0x1927d98, mutex=0x0, > > interruptible=1) at > > /Users/yogeshp/PROJECTS/SnapShot/DarwinCode/xnu-517.7.21/osfmk/kern/sched_prim.c:749 > > #4 0x20e65530 in com_CoreSnap_driver::handle_request_loop(void*) > > (data=0xcb63cb4) at > > /System/Library/Frameworks/Kernel.framework/Headers/IOKit/IOLocks.h:112 > > #5 0x002777d4 in ioThreadStart () at > > /Users/yogeshp/PROJECTS/SnapShot/DarwinCode/xnu-517.7.21/iokit/Kernel/IOLib.c:140 > > Current language: auto; currently c > > (gdb) n > > 635 splx(spl); > > (gdb) n > > 637 } > > (gdb) n > > panic (str=0x2d43c8 "{%d} Assertion failed: file \"%s\", line %d: %s\n") > > at > > /Users/yogeshp/PROJECTS/SnapShot/DarwinCode/xnu-517.7.21/osfmk/kern/debug.c:202 > > 202 PANIC_LOCK(); > > (gdb) n > > 204 PANIC_UNLOCK(); > > (gdb) n > > 203 panicstr = (char *)0; > > (gdb) n > > 204 PANIC_UNLOCK(); > > (gdb) n > > 205 thread->wait_queue = save_waits[cpu_number()]; /* > > Restore the wait queue */ > > (gdb) n > > 206 if (return_on_panic) { > > (gdb) n > > 205 thread->wait_queue = save_waits[cpu_number()]; /* > > Restore the wait queue */ > > (gdb) n > > 206 if (return_on_panic) { > > (gdb) n > > 211 kdb_printf("panic: We are hanging here...\n"); > > (gdb) n > > 212 while(1); > > (gdb) n > > > > Please suggest so that I can pinpoint the issue. > > > > Thanks in advance, > > Yogesh Pahilwan And back to bottom posting... Ok, I downloaded the xnu-517.7.21 source (aka OS X, 10.3.5), and I opened IOLib.c and guess which line is line 140? static void ioThreadStart( void ) { IOThreadFunc fcn; void * arg; fcn = threadArgFcn; arg = threadArgArg; lock_done( threadArgLock); (*fcn)(arg); IOExitThread(); } Hint: (*fcn)(arg); So, if you haven't modified the kernel source and built your own custom kernel I'm wondering if your kernel threads are actually working. Try scaling back to your threading code unit test and see if your threads are indeed being created correctly. But I would really recommend you read Cliff and Mike's comments in this thread as they are more valid (i.e. it looks a lot like you are trying to re-invent the wheel, poorly). Read up on the IOCommandPool, IOWorkLoop at the very minimum. Please! Cheers, H. _______________________________________________ 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: Notify Application when all system items have been loaded
by Peter Lovell 23 Dec '05

23 Dec '05

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com On Dec 23, 2005, at 3:29 PM, Carl Smith wrote: Thanks Carl Regards.....Peter _______________________________________________ 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 was wondering if anyone knows of some way that my app can tell when all the OS items have been loaded and started at startup. My problem is that I have a NKE and an application. My NKE attaches itself to all the interfaces when it is loaded, my application loads my NKE, and my application is started automatically at startup. What I have been seeing is that my application loads my NKE before all the interfaces are up. In other words my NKE will attach itself to the built in Ethernet but not the Airport card. Now if I shut down my app, unload my NKE and start everything again, manually without a system restart then my NKE will see all the interfaces, including the Airport. So what I would like is some way that my app knows not to load the NKE until the OS is finished with all it's overhead. Is there some API call or notification I can use in my application to check the status of the OS and it's start up routine? Hi Carl, I think there are a couple of issues you have to work with before going too far down this path. There are solutions, but I suspect that your current path has some bumps in it. The main thing to consider is that the list of interfaces is dynamic. So, although you may wait until the "startup" stuff is complete, that might not be satisfactory for what you're trying to do. For example, the system might start without Airport because it's turned off by the user. Does it matter to you if it's enabled after an hour or so? What happens if Airport is there and then it's turned off? And even Ethernet "en0" might be configured off. I suspect that you really don't want to wait for "startup complete" but instead need to respond to interface transitions as and when they occur. The system configuration framework will be your friend in that endeavour. This email sent to site_archiver(a)lists.apple.com

1 0

Notify Application when all system items have been loaded
by Carl Smith 23 Dec '05

23 Dec '05

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Thread-index: AcYH/4+LT/brIQ8ySduYYmOVJQR9UA== Thread-topic: Notify Application when all system items have been loaded _______________________________________________ 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