December 2006 - Darwin-kernel

Re: Socket filter's sf_unregistered func not getting called -- occasionally
by Josh Graessley 03 Dec '06

03 Dec '06

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com -josh On Dec 2, 2006, at 7:14 PM, Greg Miller wrote: Hi folks, I'm playing around with socket filters (sys/kpi_socketfilter.h) on 10.4.8 on Intel, mostly just to experiment and learn. I followed the tcplognke sample code pretty closely (the code from 11/27, http://developer.apple.com/samplecode/tcplognke/) and wrote a small KEXT that basically just logs messages when the sf_notify function is called. However, I noticed that occasionally the KEXT would fail to unload. The call to sflt_unregister() succeeded, but my sf_unregistered function was never getting called back. To eliminate my code as a suspect, I tried this with the tcplognke code itself, and it seems to exhibit the same behavior. I have read the docs and code comments that explain that "[sflt_unregister()] will not detach the socket filter from all sockets it may be attached to at the time, it will just prevent the socket filter from being attached to any new sockets", but the issue seems to be that sometimes the sf_unregistered function is never called, and so the socket filter can never safely be unloaded. Again, this only seems to happen on occasion, and it doesn't seem to be related to Intel (I've seen it on my G5 as well). When it happens, it's typically after maybe 5 - 10 minutes of usage with network traffic. Am I missing something obvious? Is it a known issue that sometimes socket filters won't be able to unload? Thanks in advance. Greg _______________________________________________ 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 Attaching using gdb and running the script showallstacks that comes from the kgmacros will sometimes be enlightening. If I remember correctly, the socket filter will only fail to unregister if one of your socket filter functions is currently running. It has been a year or so since I looked at that stuff though, so I may be off a little, and there's always the possibility of a bug or two. -- Greg Miller jgm(a)google.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/jgraessley% 40apple.com This email sent to jgraessley(a)apple.com smime.p7s

1 0

Socket filter's sf_unregistered func not getting called -- occasionally
by Greg Miller 03 Dec '06

03 Dec '06

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Domainkey-signature: a=rsa-sha1; s=beta; d=google.com; c=nofws; q=dns; h=received:message-id:date:from:to:subject:mime-version: content-type:content-transfer-encoding:content-disposition; b=XbUQkdn49HOxLk5jF4471QigdKbCPlr6D0iDtQ7Ju5PdogXWrN+bgZlX0L4eHnrPy EWgqErq9b5IGTxCmEBvHQ== Hi folks, I'm playing around with socket filters (sys/kpi_socketfilter.h) on 10.4.8 on Intel, mostly just to experiment and learn. I followed the tcplognke sample code pretty closely (the code from 11/27, http://developer.apple.com/samplecode/tcplognke/) and wrote a small KEXT that basically just logs messages when the sf_notify function is called. However, I noticed that occasionally the KEXT would fail to unload. The call to sflt_unregister() succeeded, but my sf_unregistered function was never getting called back. To eliminate my code as a suspect, I tried this with the tcplognke code itself, and it seems to exhibit the same behavior. I have read the docs and code comments that explain that "[sflt_unregister()] will not detach the socket filter from all sockets it may be attached to at the time, it will just prevent the socket filter from being attached to any new sockets", but the issue seems to be that sometimes the sf_unregistered function is never called, and so the socket filter can never safely be unloaded. Again, this only seems to happen on occasion, and it doesn't seem to be related to Intel (I've seen it on my G5 as well). When it happens, it's typically after maybe 5 - 10 minutes of usage with network traffic. Am I missing something obvious? Is it a known issue that sometimes socket filters won't be able to unload? Thanks in advance. Greg -- Greg Miller jgm(a)google.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: Random Kernel Panic during restart after installing kernel extensions.
by Michael Smith 02 Dec '06

02 Dec '06

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com User-agent: Thunderbird 1.5.0.8 (Macintosh/20061025) And restart the machine, NOW sometimes I get a random "kernel panic". Is there something else also mandatory? Your kernel extensions need to have at least one less bug. = Mike _______________________________________________ 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.… Ravi wrote: Hi, I am trying to install three kernel extension to /System/Library/Extensions, they have some dependency in between them. ie A.kext, B.kext and C.kext. B depends on A, C depends on both A and B. After successful installation in my postflight script I update the system cache as: > touch /System/Library/Extensions in the script. This email sent to site_archiver(a)lists.apple.com

1 0

Random Kernel Panic during restart after installing kernel extensions.
by Ravi 02 Dec '06

02 Dec '06

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com And restart the machine, NOW sometimes I get a random "kernel panic". Is there something else also mandatory? Thanks _______________________________________________ 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, I am trying to install three kernel extension to /System/Library/ Extensions, they have some dependency in between them. ie A.kext, B.kext and C.kext. B depends on A, C depends on both A and B. After successful installation in my postflight script I update the system cache as: > touch /System/Library/Extensions in the script. This email sent to site_archiver(a)lists.apple.com

1 0

Re: address space layout of x86_64?
by Derek Kumar 01 Dec '06

01 Dec '06

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com On Dec 1, 2006, at 2:10 PM, Andrew Gallatin wrote: What is the address space layout of 64-bit applications on x86_64? Is the address space below 4GB reserved for kernel mappings? 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.… In the event that this helps with bringup and such for your project...yes, by default 64-bit applications are built with a 4GB PAGEZERO segment--this serves to catch pointer truncation bugs etc., and is also a performance enhancement in that those applications can have the kernel mapped in to that region (the level 4 page map for that task is created with the appropriate kernel page directory pointer table entries, thus avoiding the cost of a CR3 switch on kernel transitions). Currently, you can however choose to build with a smaller page zero segment (which will negate the shared CR3 mechanism)--see the -pagezero_size argument to the linker. This email sent to site_archiver(a)lists.apple.com

1 0

Re: Call function on all processors
by Derek Kumar 01 Dec '06

01 Dec '06

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com On Nov 21, 2006, at 2:26 PM, Paul Litvak wrote: Paul Litvak wrote: > How would you call a function on all cpus from a kernel extension? I'm curious; why would you want to do this? The only reasons I can think of have to do with changing per-CPU (vs. per context) state, and the OS owns all of that state. I am working with the Intel virtualization technology and want to invoke the VMXON instruction on all CPUs. 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.… On 11/17/06, Michael Smith <drivers(a)mu.org> wrote: For the moment, rather than invoking mp_rendezvous or variants (the first is an unsupported export, and mp_rendezvous_no_intrs is a legacy export; additionally, there is currently an issue with the pmap (low level MMU management) layer wherein arbitrarily IPI-ing all processors via those APIs can lead to deadlocks), you can spawn N threads during initialization (where N>= the number of active CPUs, for instance) and on each thread, execute a CPU number check (it looks like cpu_number() is a legacy export, so you may need to obtain the APIC ID instead; disable interrupts to avoid being preempted during the interval of the check) and set the appropriate MSRs/ execute VMXON if you haven't already done so on that processor. Rinse/ repeat (and block) as necessary. This email sent to site_archiver(a)lists.apple.com

1 0

Re: address space layout of x86_64?
by Andrew Gallatin 01 Dec '06

01 Dec '06

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Michael Smith writes: > Andrew Gallatin wrote: > > What is the address space layout of 64-bit applications > > on x86_64? Is the address space below 4GB reserved > > for kernel mappings? > > > > It's my recollection that, on EM64T systems the 64-bit user process > memory map leaves the virtual range 0-0xffffffff unmapped as a mechanism > for catching application (and originally kernel) bugs. Thanks.. for verifying this. Sick hacks, here I come. Sigh.. 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: address space layout of x86_64?
by Michael Smith 01 Dec '06

01 Dec '06

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com User-agent: Thunderbird 1.5.0.8 (Macintosh/20061025) = Mike _______________________________________________ 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.… Andrew Gallatin wrote: What is the address space layout of 64-bit applications on x86_64? Is the address space below 4GB reserved for kernel mappings? It's my recollection that, on EM64T systems the 64-bit user process memory map leaves the virtual range 0-0xffffffff unmapped as a mechanism for catching application (and originally kernel) bugs. This email sent to site_archiver(a)lists.apple.com

1 0

address space layout of x86_64?
by Andrew Gallatin 01 Dec '06

01 Dec '06

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com What is the address space layout of 64-bit applications on x86_64? Is the address space below 4GB reserved for kernel mappings? Thanks, 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: Deleting BSD Flags files
by Jerry Krinock 01 Dec '06

01 Dec '06

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Thread-index: AccVUqaa5SgTF4FFEdunRQAWy6NyrA== Thread-topic: Deleting BSD Flags files User-agent: Microsoft-Entourage/11.2.5.060620 on 06/11/25 22:20, William Kucharski at kucharsk(a)mac.com wrote: > If you don't feel like finding out which flags are set ("ls -lo"), then this > should work from single user-mode: > > chflags noschg nosappnd nosunlnk <file>; rm -f <file> The flags available on my system (10.4.8) seemed to be a little different. The following worked for me: chflags noschg,nouchg,nosappnd,nouappnd,noarch,noopaque,nodump <file> Also, those flag names must be separated as shown, with commas and no spaces. _______________________________________________ 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