March 2010 - Darwin-kernel - lists.apple.com

Re: EVFILT_TIMER just broken
by A.M. 12 Mar '10

12 Mar '10

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com On Mar 11, 2010, at 6:15 AM, Godfrey van der Linden wrote: > I've now explored every variant I can think of for using kevent with EVFILT_TIMER. If I read kevent(2) properly the default implementation for the TIMER is to set a repeating timeout that can fire every ident timeperiods and delivers a EVFILT_TIMER event. > > Yet only ever get one hit, immediately and the event never fires again. Is this a documentation problem or am I missing something? > > Here is the code that demonstrates the problem, not that it never returns after the initial immediate and incorrect kevents. > > I'm running a vanilla x86_64 macbook pro with 10.6.2. > > Please help, this problem has been screwing me over for the last week. > > Godfrey > > /* > gcc -std=c99 -Wall -Wextra -Werror -g -o kevent_timer kevent_timer.c > */ > > #include <sys/cdefs.h> > > #include <assert.h> > #include <errno.h> > #include <stdio.h> > #include <stdbool.h> > #include <strings.h> > #include <sys/time.h> > > #include <sys/event.h> > > #define kLongTimer_ms (1000) > #define kShortTimer_ms ( 10) > > static struct timeval sStart; > static int sKQ; > > static void > logKEvent(const struct kevent64_s *kep, int ind, const char * const pfx); > > int main(int argc __unused, char *argv[] __unused) > { > sKQ = kqueue(); > > gettimeofday(&sStart, NULL); > > uintptr_t nch = 0; > struct kevent64_s timers[2]; > > EV_SET64(&timers[nch], kLongTimer_ms, > EVFILT_TIMER, EV_ADD, 0, 0, nch, 0, 0); nch++; > EV_SET64(&timers[nch], kShortTimer_ms, > EVFILT_TIMER, EV_ADD, 0, 0, nch, 0, 0); nch++; > Your code is broken here. The "ident" field of the struct is supposed to contain the timer number, not the time interval- that's put in the "data" field with fflags holding the time units. This is all in the man page. Here is some sample code that works: #include <sys/cdefs.h> #include <assert.h> #include <errno.h> #include <stdio.h> #include <stdbool.h> #include <strings.h> #include <sys/time.h> #include <sys/event.h> #include <stdlib.h> int main() { int akq=kqueue(); struct kevent64_s aTimerEvent; EV_SET64(&aTimerEvent,1,EVFILT_TIMER,EV_ADD|EV_ENABLE,NOTE_SECONDS,1,0,0,0); //add the event int aSuccess=kevent64(akq,&aTimerEvent,1,NULL,0,0,NULL); if(aSuccess<0) abort(); while(1) { struct kevent64_s anEvent; int anEventCount=kevent64(akq,NULL,0,&anEvent,1,0,NULL); if(anEventCount<0) abort(); if(anEventCount==1) { printf("timer!\n"); } } } Cheers, M _______________________________________________ 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 Debug Kit 10.6.2
by Stéphane Sudre 11 Mar '10

11 Mar '10

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com On 10-03-11, at 12:51, Terry Lambert wrote: OK, good to know. _______________________________________________ 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 Mar 11, 2010, at 2:55 AM, Stéphane Sudre wrote: Is there any reason why the Kernel Debug Kit 10.6.2 is not available at http://developer.apple.com/sdk/? You're looking on the sdk page, where there is (usually, eventually) a link added for convenience, instead of the kernel debug kits download page: <http://developer.apple.com/hardwaredrivers/download/kerneldebugkits.html > (it's the first link in the download list). I already had access to it via the ADC Member site but I was surprised it was not publicly visible as I was not aware of the above URL. http://developer.apple.com/sdk/ is the first result you get when you type "Kernel Debug Kit" in the search field at developer.apple.com. This email sent to site_archiver(a)lists.apple.com

1 0

Re: Kernel Debug Kit 10.6.2
by Terry Lambert 11 Mar '10

11 Mar '10

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com (it's the first link in the download list). Hope that helps, -- Terry _______________________________________________ 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 Mar 11, 2010, at 2:55 AM, Stéphane Sudre wrote: Is there any reason why the Kernel Debug Kit 10.6.2 is not available at http://developer.apple.com/sdk/? You're looking on the sdk page, where there is (usually, eventually) a link added for convenience, instead of the kernel debug kits download page: <http://developer.apple.com/hardwaredrivers/download/kerneldebugkits.html > This email sent to site_archiver(a)lists.apple.com

1 0

EVFILT_TIMER just broken
by Godfrey van der Linden 11 Mar '10

11 Mar '10

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com I've now explored every variant I can think of for using kevent with EVFILT_TIMER. If I read kevent(2) properly the default implementation for the TIMER is to set a repeating timeout that can fire every ident timeperiods and delivers a EVFILT_TIMER event. Yet only ever get one hit, immediately and the event never fires again. Is this a documentation problem or am I missing something? Here is the code that demonstrates the problem, not that it never returns after the initial immediate and incorrect kevents. I'm running a vanilla x86_64 macbook pro with 10.6.2. Please help, this problem has been screwing me over for the last week. Godfrey /* gcc -std=c99 -Wall -Wextra -Werror -g -o kevent_timer kevent_timer.c */ #include <sys/cdefs.h> #include <assert.h> #include <errno.h> #include <stdio.h> #include <stdbool.h> #include <strings.h> #include <sys/time.h> #include <sys/event.h> #define kLongTimer_ms (1000) #define kShortTimer_ms ( 10) static struct timeval sStart; static int sKQ; static void logKEvent(const struct kevent64_s *kep, int ind, const char * const pfx); int main(int argc __unused, char *argv[] __unused) { sKQ = kqueue(); gettimeofday(&sStart, NULL); uintptr_t nch = 0; struct kevent64_s timers[2]; EV_SET64(&timers[nch], kLongTimer_ms, EVFILT_TIMER, EV_ADD, 0, 0, nch, 0, 0); nch++; EV_SET64(&timers[nch], kShortTimer_ms, EVFILT_TIMER, EV_ADD, 0, 0, nch, 0, 0); nch++; struct kevent64_s *chList = timers; bool seenLong = false, seenShort = false; do { struct kevent64_s evList[2]; bzero(evList, sizeof(evList)); for (int i = 0; i < (int) nch; i++) logKEvent(&chList[i], i, "change"); int nev = kevent64(sKQ, chList, nch, evList, 2, 0, NULL); chList = NULL; nch = 0; for (int i = 0; i < nev; i++) { struct kevent64_s *kevp = &evList[i]; logKEvent(kevp, i, "kev"); assert(kevp->filter == EVFILT_TIMER); if (kevp->udata == 0) seenLong = true; else if (kevp->udata == 1) seenShort = true; else assert(false); }; if (seenLong && seenShort && !(timers[0].flags & EV_CLEAR)) { seenLong = seenShort = false; chList = timers; nch = sizeof(timers)/sizeof(timers[0]); for (int i = 0; i < (int) nch; i++) chList[i].flags = EV_ADD|EV_CLEAR; }; } while ( true ); // } while ( !seenLong || !seenShort ); } static void logKEvent(const struct kevent64_s *kep, int ind, const char * const pfx) { static const char *sFilterNames[] = { "EVNULL", "EVFILT_READ", "EVFILT_WRITE", "EVFILT_AIO", "EVFILT_VNODE", "EVFILT_PROC", "EVFILT_SIGNAL", "EVFILT_TIMER", "EVFILT_MACHPORT", "EVFILT_FS", "EVFILT_USER", "EVFILT_SESSION" }; // 1 2 3 4 5 6 7 // 12345678901234567890123456789012345678901234567890123456789012345678901234567 // ADD|DELETE|ENABLE|DISABLE|ONESHOT|CLEAR|RECEIPT|DISPATCH|BAD0x0100|BAD0x0200| // 8 9 10 11 // 890123456789012345678901234567890123456789 // BAD0x0400|BAD0x0800|FLAG1|FLAG2|ERROR|EOF0 #define kMaxFlagCharacters 119 static const char *sFlagNames[] = { "ADD", "DELETE", "ENABLE", "DISABLE", "ONESHOT", "CLEAR", "RECEIPT", "DISPATCH", "BAD0x0100", "BAD0x0200", "BAD0x0400", "BAD0x0800", "FLAG1", "FLAG2", "ERROR", "EOF" }; char flagBuffer[kMaxFlagCharacters], *cp = flagBuffer; int cc; #define kNumFilterNames (sizeof(sFilterNames)/sizeof(sFilterNames[0])) #define kNumFlagNames (sizeof(sFlagNames) /sizeof(sFlagNames[0])) #define kNumFlagChars (sizeof(flagBuffer) /sizeof(flagBuffer[0])) cp[0] = '\0'; for (size_t i = 0; i < kNumFlagNames; i++) { cc = 0; if ( kep->flags & (1 << i) ) cc = sprintf(cp, "%s|", sFlagNames[i]); cp += cc; } cp[-1] = '\0'; // Overwrite trailing | assert(cp < &flagBuffer[kNumFlagChars]); unsigned filter = - kep->filter; assert(filter < kNumFilterNames); struct timeval delta; gettimeofday(&delta, NULL); timersub(&delta, &sStart, &delta); fprintf(stderr, "%7ldus: %s[%d] = { %lld, %s, %s, %x, %lld }\n", delta.tv_sec * 1000 * 1000 + delta.tv_usec, pfx, ind, kep->ident, sFilterNames[filter], flagBuffer, kep->fflags, kep->data); } _______________________________________________ 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

Kernel Debug Kit 10.6.2
by Stéphane Sudre 11 Mar '10

11 Mar '10

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.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.… Is there any reason why the Kernel Debug Kit 10.6.2 is not available at http://developer.apple.com/sdk/? This email sent to site_archiver(a)lists.apple.com

1 0

EVFILT_TIMER does not behave as expected
by Godfrey van der Linden 11 Mar '10

11 Mar '10

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com G'day, All. I've just spent a few days tracking down a bug in my kevent based event handling system and the timers just don't work as expected, that is ONESHOT USECOND timers always return immediately! I set up my initial change list as struct kevent timers[2]; int nch = 0; EV_SET(&timers[nch], kLongTimerUS, EVFILT_TIMER, EV_ADD|EV_ONESHOT, NOTE_USECONDS, 0, (void *) nch); nch++; EV_SET(&timers[nch], kShortTimerUS, EVFILT_TIMER, EV_ADD|EV_ONESHOT, NOTE_USECONDS, 0, (void *) nch); nch++; Then I call chList = timers; for (;;) { int nev = kevent(kq, chList, (int) nch, &evList, 1, NULL); chList = NULL; nch = 0; logKEvent(&evList); } Finally the results I get are 6us: timers[0] = { 1000000, EVFILT_TIMER, ADD|ONESHOT, 2, 0 } 32us: timers[1] = { 10000, EVFILT_TIMER, ADD|ONESHOT, 2, 0 } 46us: kev[0] = { 1000000, EVFILT_TIMER, ADD|ONESHOT, 0, 1 } 53us: kev[0] = { 10000, EVFILT_TIMER, ADD|ONESHOT, 0, 1 } Where the first two lines log the creation of the change list, where the second two lines are kevents returned from my kevent call. The beginning of the line is the number of micro seconds from the start of the program, Notice I asked for two timeouts one of 1s another of 10ms. They both fire within 53us of the test running! That I didn't expect! Below I have included the a minimum proof of the problem. /* gcc -std=c99 -Wall -Wextra -Werror -g -o kevent_bug kevent_bug.c */ #include <sys/cdefs.h> #include <assert.h> #include <errno.h> #include <stdio.h> #include <stdbool.h> #include <strings.h> #include <sys/time.h> #include <sys/event.h> #define kLongTimerUS (1000 * 1000) #define kShortTimerUS ( 10 * 1000) static struct timeval sStart; static void logKEvent(const struct kevent *kep, int ind, const char * const pfx); int main(int argc __unused, char *argv[] __unused) { int kq = kqueue(); gettimeofday(&sStart, NULL); struct kevent timers[2]; uintptr_t nch = 0; EV_SET(&timers[nch], kLongTimerUS, EVFILT_TIMER, EV_ADD|EV_ONESHOT, NOTE_USECONDS, 0, (void *) nch); nch++; EV_SET(&timers[nch], kShortTimerUS, EVFILT_TIMER, EV_ADD|EV_ONESHOT, NOTE_USECONDS, 0, (void *) nch); nch++; for (unsigned i = 0; i < nch; i++) logKEvent(&timers[i], i, "timers"); bool seenLong = false, seenShort = false; const struct kevent *chList = timers; do { struct kevent evList; bzero(&evList, sizeof(evList)); int nev = kevent(kq, chList, (int) nch, &evList, 1, NULL); nch = 0; chList = NULL; // Disable the change list logKEvent(&evList, 0, "kev"); // Log the event assert(nev == 1); assert(evList.filter == EVFILT_TIMER); uintptr_t udata = (uintptr_t) evList.udata; if (udata == 0) seenLong = true; else if (udata == 1) seenShort = true; else assert(false); } while ( !seenLong || !seenShort ); } static void logKEvent(const struct kevent *kep, int ind, const char * const pfx) { static const char *sFilterNames[] = { "EVNULL", "EVFILT_READ", "EVFILT_WRITE", "EVFILT_AIO", "EVFILT_VNODE", "EVFILT_PROC", "EVFILT_SIGNAL", "EVFILT_TIMER", "EVFILT_MACHPORT", "EVFILT_FS", "EVFILT_USER", "EVFILT_SESSION" }; // 1 2 3 4 5 6 7 // 12345678901234567890123456789012345678901234567890123456789012345678901234567 // ADD|DELETE|ENABLE|DISABLE|ONESHOT|CLEAR|RECEIPT|DISPATCH|BAD0x0100|BAD0x0200| // 8 9 10 11 // 890123456789012345678901234567890123456789 // BAD0x0400|BAD0x0800|FLAG1|FLAG2|ERROR|EOF0 #define kMaxFlagCharacters 119 static const char *sFlagNames[] = { "ADD", "DELETE", "ENABLE", "DISABLE", "ONESHOT", "CLEAR", "RECEIPT", "DISPATCH", "BAD0x0100", "BAD0x0200", "BAD0x0400", "BAD0x0800", "FLAG1", "FLAG2", "ERROR", "EOF" }; char flagBuffer[kMaxFlagCharacters], *cp = flagBuffer; int cc; #define kNumFilterNames (sizeof(sFilterNames)/sizeof(sFilterNames[0])) #define kNumFlagNames (sizeof(sFlagNames) /sizeof(sFlagNames[0])) #define kNumFlagChars (sizeof(flagBuffer) /sizeof(flagBuffer[0])) cp[0] = '\0'; for (size_t i = 0; i < kNumFlagNames; i++) { cc = 0; if ( kep->flags & (1 << i) ) cc = sprintf(cp, "%s|", sFlagNames[i]); cp += cc; } cp[-1] = '\0'; // Overwrite trailing | assert(cp < &flagBuffer[kNumFlagChars]); unsigned filter = - kep->filter; assert(filter < kNumFilterNames); struct timeval delta; gettimeofday(&delta, NULL); timersub(&delta, &sStart, &delta); fprintf(stderr, "%7ldus: %s[%d] = { %ld, %s, %s, %x, %ld }\n", delta.tv_sec * 1000 * 1000 + delta.tv_usec, pfx, ind, kep->ident, sFilterNames[filter], flagBuffer, kep->fflags, kep->data); } _______________________________________________ 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

dtrace error on 10.6: no struct aiocb definition is available
by mogambo 09 Mar '10

09 Mar '10

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=domainkey-signature:mime-version:received:date:message-id:subject :from:to:content-type; bh=GHED0sbv1yPlYkU1Rf1DKNCLX+c4+gZIPL86pBAz+Z0=; b=Cl7IZMxVZDyU0tQMFu3aoU1JWfOsdJvsyUN7QICGfIiLGI1zBueEMI7XT/dq7MWCtW a/OLVUEiQ7qjOY00amHVmNHg2jpcB6/FSc67wDVwRE53ykfC6Za2jXLfdK1jZ9eGjpGA f9X2YvUqv+JMYh4LTtgzK/eIscoEwiJ2hFTmc= Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma; h=mime-version:date:message-id:subject:from:to:content-type; b=JsYb/HK8Fsaj/Voo0wixX/B0y34GO86vVveYjaY+d3SlBo4FbMWLMvgttReah8llBU 3RU7JiveY+Tnzhu90z8Cq4ogLHKdUkJ/Iv4+1yLPERn0kz1XCAbZARTvLPKSL8lAmzj4 VdFi+LI1NKiiZN+Bni4jCAZez89EO3fKEjYpc= _______________________________________________ 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: Thread Performance Tuning
by Wade Tregaskis 08 Mar '10

08 Mar '10

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.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: KAUTH_PROCESS_CANSIGNAL unimplemented
by Terry Lambert 08 Mar '10

08 Mar '10

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com I am not terriffically sure I want to implement this as specified. FWIW. What are you trying to use it for? -- Terry On Mar 6, 2010, at 1:11 AM, Dave Keck <davekeck(a)gmail.com> wrote: To save others the time I spent learning the hard way: The KAUTH_PROCESS_CANSIGNAL action of the Kauth subsystem is unimplemented in all public versions of OS X. (At time of writing, that's through 10.6.) The only mention of this is within the kernel source. Bug #7724502 _______________________________________________ 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/tlambert%40apple.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 tlambert(a)apple.com This email sent to site_archiver(a)lists.apple.com

1 0

Re: Thread Performance Tuning
by babazaroni＠gmail.com 07 Mar '10

07 Mar '10

site_archiver(a)lists.apple.com Delivered-To: darwin-kernel(a)lists.apple.com Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=domainkey-signature:mime-version:received:in-reply-to:references :date:message-id:subject:from:to:content-type :content-transfer-encoding; bh=Cu7aknwtFm2CshoZqfhCw+Wy91DZjtGb+ZQ8Ls36NP4=; b=iuzhDY1DUHoqLg+d8AVpli/bcjXZl/AXauRvGMEeTgQmTNgqDaHbyoOyC3ZkTUlV6+ toqOWWIjT4He90vSq1CRTRfifDdftRVwF+5POfvNDSSTBEbMTUA76sUyEtU4NDfcNs7s X+7ljqD9JokmTHnuSguCBovB/oAEKFpMh4DY4= Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma; h=mime-version:in-reply-to:references:date:message-id:subject:from:to :content-type:content-transfer-encoding; b=QBgE785AqhRZIZDXxhb0WykzkyCpQIrriWiGo6JEKSjpasV25wJLVN/VT99hQF8UZf a5n7fqbn73FnQnf9nNFHcymC1b1W7LLuVjLzWqf1kKHsv/muDKw9FSQmsZ94zVreySyB /tTCdGx7xEh8UCmD+96A/2fukUFk+dtxhiIB8= Looking closer, it seems that this is not a cpu swapping problem, but possibly a swapping out, cache miss type problem. Looking at the over all User and Sytem load shows: When audio distorting: The user cpu is 26, while the system is 53. When audio clean: The user cpu at 50 and the system is 32 and audio is clean. Something has forced the system to pay much more attention to my process. On Sun, Mar 7, 2010 at 12:27 PM, <babazaroni(a)gmail.com> wrote: > Hello, > > I'm starting a user-level thread with posix_threads at the highest > SCHED_FIFO priority, on a Core 2 Duo 2ghz Mackbook running 10.5. > > This thread receives audio over udp from a core audio server in > another process, does some processing, then sends it back to the core > audio server to be played. > > I can increase the workload of this thread to find out where > distortion of the audio begins. This happens when the work load is > set to 1200 and Activity Monitor shows 55% cpu usage. (A single work > load is crudely written deinterleaving of a 64 sample 8 channel audio > frame). > > Now the interesting part is if I start FireFox, play a youtube video, > then exit, I can increase the work load to 2500 and 95% cpu usage > before distortion happens. > > My uneducated guess was the thread is being forced to another core, > but I want to hear why you guys think, and how can I get this > performance all the time. > > I've looked into the Thread Affinity API, but the doc seems to say > this works only on systems with multiple-L2 caches, one for each cpu > and I don't think a Core 2 duo qualifies. > > I could start a true kernel thread if that will give me more control > of this, but would rather leave in user land for now. > > Also interested in any tools that will shed light on what's going on. > > 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.… This email sent to site_archiver(a)lists.apple.com

1 0