January 2010 - Darwin-dev - lists.apple.com

Re: Run loops: Mach messages vs. kqueues
by Mo McRoberts 20 Jan '10

20 Jan '10

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Wed, Jan 20, 2010 at 16:35, A.M. <agentm(a)themactionfaction.com> wrote: > You are absolutely correct. I, too, was disappointed that CFRunLoops poll mach ports only- especially since gnustep polls file descriptors, which is much more useful. With a kqueue-backed runloop, one can poll file descriptors, signals, mach ports, timers, vnodes, and audit sessions. The runloop becomes a central dispatch loop, just as it should be. I'd suggest that GNUstep polls FDs because it has to (because there are no Mach ports on most platforms GNUstep runs on), rather than as a feature :) M. _______________________________________________ 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: Run loops: Mach messages vs. kqueues
by David Leimbach 20 Jan '10

20 Jan '10

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(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:cc:content-type; bh=73Y0/wRQcW50cLp95znljItd4diS7rvv1UgSx/Lj8hI=; b=kxT+lM7EjsOupUzLFUf3/ymDUuwv8LmQEbnlTyZAgGwkShKSjmpVhv7jDZDAOY5KLl ZNHESY5wDzvfHhO/m2uuej7ALHdnbNuKj1ykEE4cw/R0qt8sGZ8EWrHjenWvKbp7bQVc sFXKohnyu2s/o0AHk95piNjbCy2WBeoufpopI= 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 :cc:content-type; b=LDQFKtij+48x1QyrG/T3eMzckTdjBiaJH//DRoUl+32D51/QJXOuaj9wZtMzm71CU3 d9WDuD2ZLIWaRjl8uDphhHyd4l6AOyuz3Fh2x3JDG3875fmqJ4Zy/CK+bfgNw03/mrK9 GnagG3Pr/eVfDdCEHupvZgstc97LUdYwIBrrI= _______________________________________________ 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: Run loops: Mach messages vs. kqueues
by Damien Sorresso 20 Jan '10

20 Jan '10

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Jan 20, 2010, at 1:38 AM, Dave Keck wrote: > Hello, > > Is there a reason that CFRunLoop is implemented atop Mach messages > rather than a kqueue? If it were implemented atop a kqueue, not only > would CFRunLoop maintain support for waking when a Mach message > arrives, but it could also be awoken when data is available on an > arbitrary file descriptor. (As far as I'm aware, the only way to do > this currently is to spawn a secondary thread to monitor the > descriptor and message the original thread to wake it.) > > It seems a kqueue-CFRunLoop would improve efficiency quite a bit at no > cost; I've written a lot of code that uses the secondary thread > technique, and I know there's a lot of system APIs that do the same > (CFFileDescriptor and CFSocket to name a few.) > > I did notice that kqueues just gained support for monitoring Mach > messages in 10.6 though, so perhaps CFRunLoop has just been waiting > for support in the kernel? David, CFRunLoop has been around for a very long time, since the 10.0 days, well before kqueue was implemented on Mac OS X. I would not be surprised if moving it to a kqueue-based firing mechanism would introduce binary incompatibilities. (Also, I believe that CFMachPort has been around just as long as CFRunLoop; it wasn't just added in Snow Leopard.) In any case, you really want a GCD run loop. -- Damien Sorresso BSD Engineering Apple 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: Run loops: Mach messages vs. kqueues
by A.M. 20 Jan '10

20 Jan '10

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Jan 20, 2010, at 4:38 AM, Dave Keck wrote: > Hello, > > Is there a reason that CFRunLoop is implemented atop Mach messages > rather than a kqueue? If it were implemented atop a kqueue, not only > would CFRunLoop maintain support for waking when a Mach message > arrives, but it could also be awoken when data is available on an > arbitrary file descriptor. (As far as I'm aware, the only way to do > this currently is to spawn a secondary thread to monitor the > descriptor and message the original thread to wake it.) > > It seems a kqueue-CFRunLoop would improve efficiency quite a bit at no > cost; I've written a lot of code that uses the secondary thread > technique, and I know there's a lot of system APIs that do the same > (CFFileDescriptor and CFSocket to name a few.) > > I did notice that kqueues just gained support for monitoring Mach > messages in 10.6 though, so perhaps CFRunLoop has just been waiting > for support in the kernel? Hi Dave, You are absolutely correct. I, too, was disappointed that CFRunLoops poll mach ports only- especially since gnustep polls file descriptors, which is much more useful. With a kqueue-backed runloop, one can poll file descriptors, signals, mach ports, timers, vnodes, and audit sessions. The runloop becomes a central dispatch loop, just as it should be. By the way, CFRunLoop's sources are available here: http://www.opensource.apple.com/source/CF/CF-550.13/CFRunLoop.c In the end, I simply wrote by own runloop which runs on all threads except the primary thread. Cheers, M _______________________________________________ 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

Run loops: Mach messages vs. kqueues
by Dave Keck 20 Jan '10

20 Jan '10

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=domainkey-signature:mime-version:sender:received:from:date :x-google-sender-auth:message-id:subject:to:content-type; bh=hZIPdeKDTUTHx7KMvgP0soLfS6P+xPOWND/jVxv75yU=; b=sn5JRd7xQ3gLQLSulov5M+RZEo3qA15QoPeyes6y8dsY4TVba0vkMOEOiIpvZ5twvi kJS0lkKHKyepw4FXG8N60IU8ihgDUPVqLCzJgIz4hlKexMYK4wjw9fcE3STQFy3x+8Ko suool2PA80NQKKBUiBX0rYbiaxT1EZ6JykJAw= Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma; h=mime-version:sender:from:date:x-google-sender-auth:message-id :subject:to:content-type; b=dg5KdS34v39xBmr8nBsA3w9SUqF0Oq2XKhpm7qWm9fOUpenraZa0GLW0VOp5DYNi5Z pfgCRG88akU40nIfPDDPOUMS6xrijlnv5hSyd0RoMngpByj0Eds9QnffzKPFumY9YxNU dIZHDpiATGo+YMe1/nQIOaezdC/okpJcI1TEQ= Hello, Is there a reason that CFRunLoop is implemented atop Mach messages rather than a kqueue? If it were implemented atop a kqueue, not only would CFRunLoop maintain support for waking when a Mach message arrives, but it could also be awoken when data is available on an arbitrary file descriptor. (As far as I'm aware, the only way to do this currently is to spawn a secondary thread to monitor the descriptor and message the original thread to wake it.) It seems a kqueue-CFRunLoop would improve efficiency quite a bit at no cost; I've written a lot of code that uses the secondary thread technique, and I know there's a lot of system APIs that do the same (CFFileDescriptor and CFSocket to name a few.) I did notice that kqueues just gained support for monitoring Mach messages in 10.6 though, so perhaps CFRunLoop has just been waiting for support in the kernel? Thanks for any insight! David _______________________________________________ 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

Setting processor affinity on MacOS X?
by Alison Cassidy 20 Jan '10

20 Jan '10

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Hi all, I have a really quick tho' obscure question. I have a need to run a thread on a particular processor core. Under normal conditions, that's not something people do, but as a diagnostic developer, I have a specific need to target stress-testing of specific CPU cores. Long story .... :) I've been using CHUD's utilBindThreadToCPU() with a lot of success to-date - everything works as it should. However, as CHUD is not suitable for a shipping software product, is there a public - or indeed private - API available in MacOS X.6+ which would effectively do the same thing? Best regards, -- Allie Alison Cassidy - Diagnostics Architecture, Apple 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: OSAtomic
by Ariel Burton 18 Jan '10

18 Jan '10

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com User-agent: Mutt/1.4.1i Cant't you do it in C like this: int32_t /* return result of operation */ OSAtomicAnd32 ( int32_t theAmount, volatile int32_t *p ) { int32_t mem_val; int32_t new_val; do { mem_val = *p; new_val = mem_val & theAmount; } while ( ! OSAtomicCompareAndSwap32 ( mem_val, new_val, p ) ); return new_val; } If p is volatile, then won't the assignment to mem_val be repeated on each iteration? If the value of *p changes in the meantime, that is, between the time it is read here and the time it is read again in OSAtomicCompareAndSwap32, the operation will fail and the attempt repeated. If it succeeds, the required effect has been achieved. This should have the same ABA properties as the assemebler. Ariel == On Mon 18 Jan 11:30, Greg Parker wrote: > On Jan 18, 2010, at 11:04 AM, Derek Darwin Lists wrote: > > I believe you are misreading/mistranslating the x86 assembly implementation of OSAtomicAnd32. The compare and swap operation (cmpxchg) in the "commpage" will populate GPR EAX with the value in memory on a mismatch, so the loop on failure will restart with the "latest" value of the memory operand after the failed comparison. > > That's right. You can't do that with C code and OSAtomicCompareAndSwap(), but x86 OSAtomicAnd32() is built atop other assembly code that is not identical to OSAtomicCompareAndSwap(). > > > -- > Greg Parker gparker(a)apple.com Runtime Wrangler > > > _______________________________________________ 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: OSAtomic
by Greg Parker 18 Jan '10

18 Jan '10

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Jan 18, 2010, at 11:04 AM, Derek Darwin Lists wrote: > I believe you are misreading/mistranslating the x86 assembly implementation of OSAtomicAnd32. The compare and swap operation (cmpxchg) in the "commpage" will populate GPR EAX with the value in memory on a mismatch, so the loop on failure will restart with the "latest" value of the memory operand after the failed comparison. That's right. You can't do that with C code and OSAtomicCompareAndSwap(), but x86 OSAtomicAnd32() is built atop other assembly code that is not identical to OSAtomicCompareAndSwap(). -- Greg Parker gparker(a)apple.com Runtime Wrangler _______________________________________________ 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: OSAtomic
by Derek Darwin Lists 18 Jan '10

18 Jan '10

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(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:cc:content-type; bh=pqnAX4rngrjSPZOcLGW2K1RGGOxRWupqwY0DvDOavmU=; b=Zsd6Q/zH2EhZz/GBcfasGCyJmz2QJRgFJrQiIEck/yHziOumkLmYitvVPzd9E2wquV q6n/EY76I3nHO7nPfxLemg4qJTOBjMuYxdqeUpHNvklT3+iQGRGoU3dBVZ9uHohitvGL XIJfO3MDvyG2bVyjEgTdysHSE7/Fz/xwTTtp8= 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 :cc:content-type; b=nCxp1kzAAmNLzm79RFr7DqROcC7KT5NGEaK93QI82+JDHJVrRneExla4VAtBHS3wAN HgCDajFRccPFwNMJNgIhS4JNr4vxC2GpZUKG57Y3Bgekao+ITVWCJEKYeLBKUzklJfa4 ir2V7GxIpfM/XhhCAMMS4q0Lezj5bk2gou6KM= _______________________________________________ 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

OSAtomic
by ariel.burton＠totalviewtech.com 18 Jan '10

18 Jan '10

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com User-agent: nail 11.25 7/29/05 Hello, I've been looking at the set of operations defined in OSAtomic.h. The manual page describes the logical operations (and, or, and xor) as being "layered on top of the OSAtomicCompareAndSwap() primitives". From what I can tell looking at the assmember for OSAtomicAnd32, it is implemented (at least on x86) something like this: int32_t /* return result of operation */ OSAtomicAnd32 ( int32_t theAmount, volatile int32_t *p ) { int32_t mem_val = *p; /* POINT 1 */ int32_t new_val = mem_val & theAmount; /* PONT 2 */ while ( ! OSAtomicCompareAndSwap ( mem_val, new_val, p ) ) { } return new_val; } What this appears to be doing is reading and recording the contents of the memory location (mem_val). It then computes what the new value should be (new_val). Now to ensure atomicity, it then calls the atomic compare and swap operation with mem_val and new_val. The atomic swap should be performed if and only if the contents of the memory location are mem_val. What I don't understand is this. The contents of memory could change through some other agency after the first read, but before the call to OSAtomicCompareAndSwap, that between points 1 and 2 above. That's ok if the value of the memory location returns to mem_val at some point. But what if the value doesn't revert back to mem_val? In that case, the operation will loop indefinitely. In view of all this, I feel there must be some assumuptions about how these routines are to be used. Are they supposed to be general-purpose, or restricted to particular patterns of usage? Many thanks, Ariel Burton _______________________________________________ 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