September 2004 - Darwin-dev

Re: how to test atomicity?
by Matt Watson 19 Sep '04

19 Sep '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com pthread_t t1, t2; int iX = 0; assert(!pthread_create(&t1, NULL, test_atomic_inc, &iX)); assert(!pthread_create(&t2, NULL, test_atomic_dec, &iX)); assert(!pthread_join(t1, NULL)); assert(!pthread_join(t2, NULL)); assert(!iX); matt. On Sep 19, 2004, at 8:22 AM, jmzorko(a)mac.com wrote: #include <pthread.h> #include <iostream> using namespace std; typedef int AtomicCountType; // code lifted from linux asm volatile( "1: lwarx %0,0,%1\n" " addic %0,%0,1\n" // addic allows r0, addi doesn't " stwcx. %0,0,%1\n" " bne- 1b" : "=&r" (tmp) : "r" ( p ) : "cc", "memory"); return( tmp ); } // code lifted from linux asm volatile( "2: lwarx %0,0,%1\n" " addic %0,%0,-1\n" // addic allows r0, addi doesn't " stwcx. %0,0,%1\n" " bne- 2b" : "=&r" (tmp) : "r" ( p ) : "cc", "memory"); return( tmp ); } inline static AtomicCountType AtomicBump( volatile AtomicCountType *p, AtomicCountType AddMe ) { int tmp; // code lifted from linux asm volatile( "3: lwarx %0,0,%1\n" " addic %0,%0,( AddMe )\n" // addic allows r0, addi doesn't " stwcx. %0,0,%1\n" " bne- 3b" : "=&r" (tmp) : "r" ( p ) : "cc", "memory"); return( tmp ); } void test_atomic_inc( void *pX ) { int *piX = static_cast< int * >( pX ); for ( int iIndex = 0; iIndex < 1000000000; iIndex ++ ) { //AtomicIncrement( piX ); ( *piX ) ++; } pthread_exit( 0 ); } void test_atomic_dec( void *pX ) { int *piX = static_cast< int * >( pX ); for ( int iIndex = 0; iIndex < 1000000000; iIndex ++ ) { //AtomicDecrement( piX ); ( *piX ) --; } pthread_exit( 0 ); } int main( int iArgc, char *pcArgv[] ) { // our one true int ... int iX = 0; // ... and some other stuff int iStatus = 0; // create 500 threads, half doing incs, half doing decs pthread_t threads[ 500 ] = { 0 }; for ( int iIndex = 0; iIndex < 500; iIndex ++ ) { int iCreatedOK = pthread_create( &threads[ iIndex ], NULL, iIndex % 2 == 0 ? test_atomic_inc : test_atomic_dec, &iX ); if ( iCreatedOK != 0 ) { cout << "can't make thread " << iIndex << "\n"; break; } else { pthread_join( threads[ iIndex ], ( void ** )&iStatus ); } } // test to make sure iX is still 0 i.e. operations were really atomic cout << "test " << ( iX == 0 ? "passed: " : "failed: " ) << iX << "\n"; pthread_exit( 0 ); } _______________________________________________ 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… You're not creating 500 simultaneous threads, since pthread_join waits for the thread to complete and the loop does create, join, create, join... Also, you don't need 500 threads to test atomicity. Try just creating 2 threads in main(), like: inline static AtomicCountType AtomicIncrement( volatile AtomicCountType *p ) { int tmp; inline static AtomicCountType AtomicDecrement( volatile AtomicCountType *p ) { int tmp; This email sent to site_archiver(a)lists.apple.com

1 0

how to test atomicity?
by jmzorko＠mac.com 19 Sep '04

19 Sep '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Hello, all ... #include <pthread.h> #include <iostream> using namespace std; typedef int AtomicCountType; return( tmp ); } return( tmp ); } inline static AtomicCountType AtomicBump( volatile AtomicCountType *p, AtomicCountType AddMe ) { int tmp; return( tmp ); } void test_atomic_inc( void *pX ) { int *piX = static_cast< int * >( pX ); for ( int iIndex = 0; iIndex < 1000000000; iIndex ++ ) { //AtomicIncrement( piX ); ( *piX ) ++; } pthread_exit( 0 ); } void test_atomic_dec( void *pX ) { int *piX = static_cast< int * >( pX ); for ( int iIndex = 0; iIndex < 1000000000; iIndex ++ ) { //AtomicDecrement( piX ); ( *piX ) --; } pthread_exit( 0 ); } int main( int iArgc, char *pcArgv[] ) { // our one true int ... int iX = 0; // ... and some other stuff int iStatus = 0; // create 500 threads, half doing incs, half doing decs pthread_t threads[ 500 ] = { 0 }; if ( iCreatedOK != 0 ) { cout << "can't make thread " << iIndex << "\n"; break; } else { pthread_join( threads[ iIndex ], ( void ** )&iStatus ); } } pthread_exit( 0 ); } Regards, John Falling You - exploring the beauty of voice and sound New album, "Touch", available now http://www.magnatune.com/artists/falling_you _______________________________________________ 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'm porting a class library / framework to OSX 10.3 / Darwin PPC. One of the things needed is atomic increments, decrements and swaps. With the help of others here, and various web sites, I found the lwarx / stwcx pair and some assembly doing exactly what I want. Cool! Now, the hard part -- I want to test these atomic incs / decs against their non-atomic counterparts. I have a 12" PowerBook G4 and a dual-G4 desktop, so I can test atomic ops on both (esp. the dual G4). I wrote some C++ code to launch 500 pthreads, half doing increments, the other half doing decrements, on the same variable. Each thread does it's thing a certain # of times, and if the incs and decs are atomic, the variable should be the same when the test ends as when it started. ... at least, this is what i'm thinking. The problem is that my test passes even w/o using the atomic inc / decs. I was hoping to see the atomic version pass and the non-atomic version fail, but noooo. The code (not much at all, actually) follows. Either I don't understand the problem (I think I grok it pretty well), my code is not really testing it (maybe), or the compiler (Apple's gcc3.3) is too smart for me and laughs heartily at my pathetic and weak attempts to subvert it. Can anyone tell me which of the above is true? inline static AtomicCountType AtomicIncrement( volatile AtomicCountType *p ) { int tmp; // code lifted from linux asm volatile( "1: lwarx %0,0,%1\n" " addic %0,%0,1\n" // addic allows r0, addi doesn't " stwcx. %0,0,%1\n" " bne- 1b" : "=&r" (tmp) : "r" ( p ) : "cc", "memory"); inline static AtomicCountType AtomicDecrement( volatile AtomicCountType *p ) { int tmp; // code lifted from linux asm volatile( "2: lwarx %0,0,%1\n" " addic %0,%0,-1\n" // addic allows r0, addi doesn't " stwcx. %0,0,%1\n" " bne- 2b" : "=&r" (tmp) : "r" ( p ) : "cc", "memory"); // code lifted from linux asm volatile( "3: lwarx %0,0,%1\n" " addic %0,%0,( AddMe )\n" // addic allows r0, addi doesn't " stwcx. %0,0,%1\n" " bne- 3b" : "=&r" (tmp) : "r" ( p ) : "cc", "memory"); for ( int iIndex = 0; iIndex < 500; iIndex ++ ) { int iCreatedOK = pthread_create( &threads[ iIndex ], NULL, iIndex % 2 == 0 ? test_atomic_inc : test_atomic_dec, &iX ); // test to make sure iX is still 0 i.e. operations were really atomic cout << "test " << ( iX == 0 ? "passed: " : "failed: " ) << iX << "\n"; This email sent to site_archiver(a)lists.apple.com

1 0

Re: Using MIG RPC between 2 threads inside a process
by Steve Checkoway 19 Sep '04

19 Sep '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com That's a very good point, I hadn't thought of that. - Steve _______________________________________________ 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… On Sep 18, 2004, at 4:50 PM, Paul Forgey wrote: If you happen to signal the conditional between the CheckMyCondition and pthread_cond_wait, the consumer will go into the wait with requests in the queue. Signal while holding the mutex and two things will happen: the consumer will notice work in the queue before waiting, or it will be waiting. This email sent to site_archiver(a)lists.apple.com

1 0

Re: bind
by Felix Kronlage 19 Sep '04

19 Sep '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Sep 17, 2004, at 20:49, Damien Bobillot wrote: _______________________________________________ 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… Originally, we had intended to leave the BIND-8 library in libSystem and remove the BIND-8 headers (resolv.h and nameser.h) but we ended up keeping the old headers and adding the BIND_8_COMPAT switch so that developers that really, really, had no other choice except to keep building with the BIND-8 version could do so. If at all possible, please avoid using BIND-8. It will eventually go away. It's important to keep the BIND 8 version, not for developers (they can rewritte their code) but for administrators who want to compile and use applications not updated to BIND 9 (administrators are often not able to rewritte code). but is it really the OS-vendor's responsibility to drag along old stuff, because certain applications are not maintained and updated to work in recent enviroments? (not that I'm calling here for removing bind 8 compat, this is more of a general thing) felix -- gpg-fingerprint: 076E 1E87 3E05 1C7F B1A0 8A48 0D31 9BD3 D9AC 74D0 http://fkr.hazardous.org | http://opendarwin.org/~fkr/ | FKR-RIPE This email sent to site_archiver(a)lists.apple.com

1 0

Re: Trick to compiling Apple3Com905x?
by Felix Kronlage 19 Sep '04

19 Sep '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Sep 18, 2004, at 23:44, Terry Simons wrote: _______________________________________________ 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… The PearPC project has a working pre-compiled PPC KEXT that I have used a bit - Possibly the same one you found Walt? They also have sources. I don't know if they've modified them or not, but the binary works for me. http://pearpc.sourceforge.net/apsl.html I started looking at that a while ago, but running the kext on my quicksilver with 10.3.4 caused it to crash :/ I did not swap cards back then, I will do this and report back. -fkr -- gpg-fingerprint: 076E 1E87 3E05 1C7F B1A0 8A48 0D31 9BD3 D9AC 74D0 http://fkr.hazardous.org | http://opendarwin.org/~fkr/ | FKR-RIPE This email sent to site_archiver(a)lists.apple.com

1 0

Re: Using MIG RPC between 2 threads inside a process
by Paul Forgey 18 Sep '04

18 Sep '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Sep 18, 2004, at 1:01 PM, Paul Forgey wrote: On Sep 18, 2004, at 2:01 AM, Steve Checkoway wrote: _______________________________________________ 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/paulf%40aphrodite.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… I know this is starting to stray off the original question, but I just realized you will indeed want to signal while the mutex is being held, and it's for the same reasons I gave that you want to make it a habit to do so. If you happen to signal the conditional between the CheckMyCondition and pthread_cond_wait, the consumer will go into the wait with requests in the queue. Signal while holding the mutex and two things will happen: the consumer will notice work in the queue before waiting, or it will be waiting. In this case, I don't think it makes too much difference either. In general, I'd say make a habit of signaling the conditional while holding the mutex. Since conditionals have no state, the intention for pairing them with mutexes is to insure somebody is around to block on the conditional when it's signaled if you acquire the mutex before doing so. Here the mutex is released in the consumer thread while it isn't dealing with the queue or waiting for the conditional. This is good because other producer threads could otherwise needlessly block. Since it is properly looping on the predicate (CheckMyCondition()), which does have state, it can get away with doing that. A client might or might not loop over and over but one thing you forgot was to unlock after you signal. pthread_mutex_lock(&lock); DoSomeWork(); pthread_cond_signal(&cond); pthread_mutex_unlock(&lock); It might be better to signal after unlocking but since pthread_cond_signal doesn't wake up the other threads right away, I'm not sure it makes a difference. I'll bet someone here could say which way is best. This email sent to paulf(a)aphrodite.com This email sent to site_archiver(a)lists.apple.com

1 0

Re: Trick to compiling Apple3Com905x?
by Terry Simons 18 Sep '04

18 Sep '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com http://pearpc.sourceforge.net/apsl.html - Terry On Sep 18, 2004, at 1:11 PM, Andreas Monitzer wrote: _______________________________________________ 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… The PearPC project has a working pre-compiled PPC KEXT that I have used a bit - Possibly the same one you found Walt? They also have sources. I don't know if they've modified them or not, but the binary works for me. From: Walt Pawley <walt(a)wump.org> I happened to have a 3Com 3C905B-TX ethernet card and wanted to plug it into a Beige G3 using Mac OS 10.2.6. I tried compiling the Darwin distributions of both the -3 and -7 drivers but both failed in the same manner on several different systems. I know it has been successfully compiled because I found a working version of a kext which I'm using now. This email sent to site_archiver(a)lists.apple.com

1 0

unlink: invalid argument
by Jonas Maebe 18 Sep '04

18 Sep '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Hello, <<<<<<<<<<<<<<<<<< bigmac:~ jonas$ cat sd90mac #!/bin/bash export DISPLAY=:0 ssh -CX jmaebe@sd90mac <<<<<<<<<<<<<<<<<<< 15104 rm CALL getuid 15104 rm RET getuid 213/0xd5 15104 rm CALL ioctl(0,FIODTYPE,0xbffffbe0) 15104 rm RET ioctl 0 15104 rm CALL lstat(0xbffffd8f,0xbffffb80) 15104 rm NAMI ".Xauthority" 15104 rm RET lstat 0 15104 rm CALL access(0xbffffd8f,0x2) 15104 rm NAMI ".Xauthority" 15104 rm RET access 0 15104 rm CALL unlink(0xbffffd8f) 15104 rm NAMI ".Xauthority" 15104 rm RET unlink -1 errno 22 Invalid argument 15104 rm CALL write(0x2,0xbffff2a0,0x4) 15104 rm GIO fd 2 wrote 4 bytes "rm: " 15104 rm RET write 4 15104 rm CALL write(0x2,0xbffff2f0,0xb) 15104 rm GIO fd 2 wrote 11 bytes ".Xauthority" 15104 rm RET write 11/0xb 15104 rm CALL write(0x2,0x9010791c,0x2) 15104 rm GIO fd 2 wrote 2 bytes ": " 15104 rm RET write 2 15104 rm CALL write(0x2,0xbffff300,0x11) 15104 rm GIO fd 2 wrote 17 bytes "Invalid argument " <<<<<<<<<<<<<<<<<<< Jonas _______________________________________________ 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 have a directory in my home directory which is shared via NFS to several Linux clients. Every now and then, the following happens (bigmac = my machine, sd90mac = a Linux Fedora Core 2 machine): bigmac:~ jonas$ ./sd90mac Warning: No xauth data; using fake authentication data for X11 forwarding. /usr/bin/X11/xauth: unable to link authority file /user/jmaebe/.Xauthority, use /user/jmaebe/.Xauthority-n jmaebe@SD90MAC:~$ <<<<<<<<<<<<<<<<<<< (the warning about the fake authentication data is because X11's not running locally currently). I can also not delete the .Xauthority file on my own machine when this happens. The relevant part of the ktrace of "rm .Xauthority": What I can do to fix this, is to move the file to another directory (either locally or on the linux box, using the "mv" command) and then delete the file without any problems. At the next login, this file is recreated without a problem and it will keep working fine for a while. I only have that problem with that particular file and if I don't fix it, also with the replacement ".Xauthority-n" file after a while. Now, if this is caused by file system corruption, it happens "consistently", in the sense that I had this problem on my old G4 and also on my new G5 (which have different hard disks). I have no idea what exactly triggers it however (it happens seemingly randomly). I'm running 10.3.5, but it has happened during all incarnations of 10.3. I don't think it happened under 10.2.x, but I'm not sure anymore. Is there any way to get more details as to what exactly causes unlink to return "invalid argument" for this file? This email sent to site_archiver(a)lists.apple.com

1 0

Re: Using MIG RPC between 2 threads inside a process
by Paul Forgey 18 Sep '04

18 Sep '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Sep 18, 2004, at 2:01 AM, Steve Checkoway wrote: _______________________________________________ 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… In this case, I don't think it makes too much difference either. In general, I'd say make a habit of signaling the conditional while holding the mutex. Since conditionals have no state, the intention for pairing them with mutexes is to insure somebody is around to block on the conditional when it's signaled if you acquire the mutex before doing so. Here the mutex is released in the consumer thread while it isn't dealing with the queue or waiting for the conditional. This is good because other producer threads could otherwise needlessly block. Since it is properly looping on the predicate (CheckMyCondition()), which does have state, it can get away with doing that. A client might or might not loop over and over but one thing you forgot was to unlock after you signal. pthread_mutex_lock(&lock); DoSomeWork(); pthread_cond_signal(&cond); pthread_mutex_unlock(&lock); It might be better to signal after unlocking but since pthread_cond_signal doesn't wake up the other threads right away, I'm not sure it makes a difference. I'll bet someone here could say which way is best. This email sent to site_archiver(a)lists.apple.com

1 0

Re: Trick to compiling Apple3Com905x?
by Andreas Monitzer 18 Sep '04

18 Sep '04

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com From: Walt Pawley <walt(a)wump.org> Hi, andy _______________________________________________ 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 happened to have a 3Com 3C905B-TX ethernet card and wanted to plug it into a Beige G3 using Mac OS 10.2.6. I tried compiling the Darwin distributions of both the -3 and -7 drivers but both failed in the same manner on several different systems. I know it has been successfully compiled because I found a working version of a kext which I'm using now. I stumbled upon the same problem and found some x86 assembler in there, it seems that this driver was only written for that platform. The card works fine using Debian Linux on my Mac, though. This email sent to site_archiver(a)lists.apple.com

1 0