August 2006 - Darwin-dev - lists.apple.com

Re: Get the caller of socket_lock()
by Laurent Dumont 02 Aug '06

02 Aug '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com User-agent: Microsoft-Entourage/11.1.0.040913 Michael, On 10.4.x for i386 there is a slightly different debug mechanism for the socket lock and unlock. Instead of reserved3 and reserverd4, there is circular buffers with the 4 last lock/unlocks for the socket: u_int32_t lock_lr[4]; int next_lock_lr; u_int32_t unlock_lr[4]; int next_unlock_lr; "next_lock_lr" and "next_unlock_lr" provide the number (0 to 3) for the next debug link register slot to be used. You can work your way from n-1 and find the last four lock/unlock recorded for that socket. Laurent on 8/2/06 11:23, Michael Tuexen at Michael.Tuexen(a)lurchi.franken.de wrote: > Dear all, > > we are currently debugging a locking issue on Mac OS X on Intel. It > is related to > socket locks and we are able to break into the debugger to inspect > the socket > structures. On PPC one can find the last caller of the socket_lock() > function, > but is there a similar way in doing it in Intel, possibly by > inspecting the reserved3 > or reserved4 fields in the socket structure? What is stored in these > fields? > > Thank you very much for your help in advance. > > Best regards > Michael > > _______________________________________________ > 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/laurent%40apple.com > > This email sent to laurent(a)apple.com _______________________________________________ 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: Non-priv way to collect job stats under Intel?
by almisr 02 Aug '06

02 Aug '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com an os BUG :) sure ... _______________________________________________ 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… Thanks for the link.. but I encountered the same result, it didn't return the CPU usage (just zeros), so I wonder if it suffers from the same "can't get stats from a running process" limitation as getrusage(). I pulled it into gdb and looked at the extern_proc p_rtime and p_cputicks fields. (Heck, I showed the whole struct and every time-like field was empty except __p_starttime.) I think this is an OS bug. Please file a radar bug about this issue, as an external need carries more weight. -pmb This email sent to site_archiver(a)lists.apple.com

1 0

Re:Non-priv way to collect job stats under Intel?
by Peter Bierman 02 Aug '06

02 Aug '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com sysctl 3 man page.) I grabbed some source described here: http://ezine.daemonnews.org/200205/sysctl-netbsd.html I had to change some "struct proc" declarations to "struct extern_proc", but then it compiles. Running it against my local user didn't correctly sum up my CPU usage, but it did correctly count my processes. -pmb _______________________________________________ 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… >Does sysctl on kern.proc work for this? (See the example in the Thanks for the link.. but I encountered the same result, it didn't return the CPU usage (just zeros), so I wonder if it suffers from the same "can't get stats from a running process" limitation as getrusage(). I pulled it into gdb and looked at the extern_proc p_rtime and p_cputicks fields. (Heck, I showed the whole struct and every time-like field was empty except __p_starttime.) I think this is an OS bug. Please file a radar bug about this issue, as an external need carries more weight. This email sent to site_archiver(a)lists.apple.com

1 0

Re: Non-priv way to collect job stats under Intel?
by almisr 02 Aug '06

02 Aug '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com sorry for your mistery but i've just losed 5 minutes with sysctl and it's work ... Best Regards Ideally, given a pid, I'd like an API returning the following ps values in a struct: %cpu %mem command lim lstart nice ppid pri sl state time uid xstat To solve my immediate problem, I'd settle for lstart state time xstat Does sysctl on kern.proc work for this? (See the example in the sysctl 3 man page.) I grabbed some source described here: http://ezine.daemonnews.org/200205/sysctl-netbsd.html I had to change some "struct proc" declarations to "struct extern_proc", but then it compiles. Running it against my local user didn't correctly sum up my CPU usage, but it did correctly count my processes. Thanks Jeffrey Johnson Macintosh Development Wavefunction, 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… Assuming that you're willing to accept the stipulation that you can only look at your own processes, what sorts of information about said processes would you expect/want such an API to provide? To put it another way, if you could have your idea "process inspection API", what would might it look like? Thanks for the link.. but I encountered the same result, it didn't return the CPU usage (just zeros), so I wonder if it suffers from the same "can't get stats from a running process" limitation as getrusage(). I pulled it into gdb and looked at the extern_proc p_rtime and p_cputicks fields. (Heck, I showed the whole struct and every time-like field was empty except __p_starttime.) This code looks familiar- I think I tried sysctl() for times on PPC long ago and then had to go to task_from_pid() to get real answers. :-). This email sent to site_archiver(a)lists.apple.com

1 0

Get the caller of socket_lock()
by Michael Tuexen 02 Aug '06

02 Aug '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Dear all, Thank you very much for your help in advance. we are currently debugging a locking issue on Mac OS X on Intel. It is related to socket locks and we are able to break into the debugger to inspect the socket structures. On PPC one can find the last caller of the socket_lock() function, but is there a similar way in doing it in Intel, possibly by inspecting the reserved3 or reserved4 fields in the socket structure? What is stored in these fields? Best regards Michael _______________________________________________ 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:Non-priv way to collect job stats under Intel?
by Jeffrey Johnson 02 Aug '06

02 Aug '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On 8/1/06 at 1:24 PM, bierman(a)apple.com (Peter Bierman) wrote: >> >Assuming that you're willing to accept the stipulation that you can >> >only look at your own processes, what sorts of information about said >>>processes would you expect/want such an API to provide? To put it >>>another way, if you could have your idea "process inspection API", >>>what would might it look like? >> >>Ideally, given a pid, I'd like an API returning the following ps >>values in a struct: >> >>%cpu %mem command lim lstart nice ppid pri sl state time uid xstat >> >>To solve my immediate problem, I'd settle for >> >>lstart state time xstat > > >Does sysctl on kern.proc work for this? (See the example in the >sysctl 3 man page.) > >I grabbed some source described here: >http://ezine.daemonnews.org/200205/sysctl-netbsd.html > >I had to change some "struct proc" declarations to "struct >extern_proc", but then it compiles. Running it against my local user >didn't correctly sum up my CPU usage, but it did correctly count my >processes. Thanks for the link.. but I encountered the same result, it didn't return the CPU usage (just zeros), so I wonder if it suffers from the same "can't get stats from a running process" limitation as getrusage(). I pulled it into gdb and looked at the extern_proc p_rtime and p_cputicks fields. (Heck, I showed the whole struct and every time-like field was empty except __p_starttime.) This code looks familiar- I think I tried sysctl() for times on PPC long ago and then had to go to task_from_pid() to get real answers. :-). Thanks Jeffrey Johnson Macintosh Development Wavefunction, 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… This email sent to site_archiver(a)lists.apple.com

1 0

Re: Non-priv way to collect job stats under Intel?
by almisr 02 Aug '06

02 Aug '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com 1. system_cmds src ? 2. sysctl could give what u want man 3 sysctl ------------------------------------- int mib[2], freq; u_int miblen = 2; size_t len; float speed; char *u; mib[0] = CTL_HW ; mib[1] = HW_CPU_FREQ ; len = sizeof(freq); if (sysctl(mib, miblen, &freq, &len, NULL, 0) == -1) { //err } ------------------------------------- int i, mib[4]; size_t len; struct kinfo_proc kp; /* Fill out the first three components of the mib */ len = 4; sysctlnametomib("kern.proc.pid", mib, &len); /* Fetch and print entries for pid's < 100 */ for (i = 0; i < 100; i++) { mib[3] = i; len = sizeof(kp); if (sysctl(mib, 4, &kp, &len, NULL, 0) == -1) perror("sysctl"); else if (len > 0) printkproc(&kp); } 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

How to create pass-through (stub) dylib without using flat namespaces?
by Simon Urbanek 02 Aug '06

02 Aug '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com A.c: void dummy() { foo1(); foo2(); ... } [just to make sure all symbols are there - could be empty, too] gcc -dynamiclib -o libA.dylib A.c -L. -lB prog.c: int main(int argc, char **argv) { foo1(); return 0; } gcc -o prog prog.c -L. -lA Any help is highly appreciated. Thanks, Simon _______________________________________________ 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 need to create a dylib (A) that links to another dylib (B) such that linking to A is sufficient to get the symbols of B (without linking directly to B). i.e. something like this: B.c: void foo1() {... }; void foo2() {... }; [...] gcc -dynamiclib -o libB.dylib B.c As-is it won't work, because the symbols in libB are visible only to libA and not to the prog. So I need some way to make all symbols in B exported through A. I cannot use flat name space - so please discard that right away. In fact I cannot change any flags of prog, so using - undefined dynamic_lookup for prog is not an option, either. I know that a solution exists, because that is what Apple's frameworks do if they have sub-frameworks or libs (e.g. have a look at vecLib - it contains no symbols itself but collects all symbols of the libraries it is linked to) and they use two-level namespaces. This is exactly what I need. I couldn't find anything in ld/gcc/dyld, though. This email sent to site_archiver(a)lists.apple.com

1 0

Re:Non-priv way to collect job stats under Intel?
by Peter Bierman 01 Aug '06

01 Aug '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com %cpu %mem command lim lstart nice ppid pri sl state time uid xstat To solve my immediate problem, I'd settle for I grabbed some source described here: http://ezine.daemonnews.org/200205/sysctl-netbsd.html -pmb _______________________________________________ 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… >Assuming that you're willing to accept the stipulation that you can >only look at your own processes, what sorts of information about said processes would you expect/want such an API to provide? To put it another way, if you could have your idea "process inspection API", what would might it look like? Ideally, given a pid, I'd like an API returning the following ps values in a struct: lstart state time xstat Does sysctl on kern.proc work for this? (See the example in the sysctl 3 man page.) I had to change some "struct proc" declarations to "struct extern_proc", but then it compiles. Running it against my local user didn't correctly sum up my CPU usage, but it did correctly count my processes. This email sent to site_archiver(a)lists.apple.com

1 0

Re:Non-priv way to collect job stats under Intel?
by Jeffrey Johnson 01 Aug '06

01 Aug '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On 8/1/06 at 12:03 PM, darwin-dev-request(a)lists.apple.com wrote: >On Aug 1, 2006, at 10:20 AM, Erik Paulson wrote: > >> It's unfortunate that progrms like Jeffrey's and mine that previously >> did not need any root privileges or setuid helpers now need them, >> especially >> just to do something like find out the status of a process that our >> programs spawned - even though we can still kill them, suspend them, >> and >> find out their exit status. > >Assuming that you're willing to accept the stipulation that you can >only look at your own processes, what sorts of information about said >processes would you expect/want such an API to provide? To put it >another way, if you could have your idea "process inspection API", >what would might it look like? Ideally, given a pid, I'd like an API returning the following ps values in a struct: %cpu %mem command lim lstart nice ppid pri sl state time uid xstat To solve my immediate problem, I'd settle for lstart state time xstat Even removing the long-standing getrusage() "bug"/limitation so I could use it to get ru_utime and ru_stime on running child processes would help :-) Thanks Jeffrey Johnson Macintosh Development Wavefunction, 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… This email sent to site_archiver(a)lists.apple.com

1 0