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

Re: Re: Re: darwin equivalent to linux readahead(2) function ?
by Shawn Erickson 18 Dec '06

18 Dec '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Domainkey-signature: a=rsa-sha1; q=dns; c=nofws; s=beta; d=gmail.com; h=received:message-id:date:from:to:subject:cc:in-reply-to:mime-version:content-type:content-transfer-encoding:content-disposition:references; b=rxe6B24uWxgoF3iHSak2ggxD8Yo+hx/FjuMKMKqCm28v8jrwhMEcbi6a6qjHeKDj5MPofDBb1GuIVQ3K/tPvvhEWrX2q0tpmQubYBk7JxSw9pRcXA/HlBwekwZe0ROBDtftNizX2U0gDvKfAaUfV3Ucm4yzta/VKgBlHtKvnigA= On 12/18/06, Philippe Devallois <phdevallois(a)intego.com> wrote: > fcntl(inFD,F_RDAHEAD,1); > > man fcntl > > Not so complicated ;) > > Hope this helps, Actually, I'm not sure how to use it :-( I tried like that: #include <stdio.h> #include <unistd.h> #include <sys/stat.h> #include <fcntl.h> int main(int argc, const char* argv[]) { FILE* bigfile; int fd; struct stat sb; char buf[4]; bigfile = fopen("bigfile", "r"); if (!bigfile) return 1; fd = fileno(bigfile); if (fstat(fd, &sb)) { fclose(bigfile); return 1; } fcntl(fd, F_RDAHEAD, 1); read(fd, buf, 4); printf("bigfile size is %lld bytes\n", sb.st_size); fclose(bigfile); return 0; } But it seems it's not the way I want it to behave. Here is the (timed) result of the execution: $ time ./rdaheadtest bigfile size is 118053508 bytes real 0m0.055s user 0m0.002s sys 0m0.015s It obviously has NOT loaded the whole 113 MB file into memory because (1) I haven't heard my hard disk at all and (2) my PowerBook G4 is probably not able to load 113 MB in less than 0.055 s. I must be doing something wrong but the fcntl man is not really verbose about F_RDAHEAD and it's usage. Note that Mac OS X will use any available physical RAM pages to cache file data (unified buffer cache). It is possible that prior tests you ran resulted in the test file getting fully loading into pages sitting in the file cache. Any access to those pages would happen at RAM access speeds and wouldn't involve disk access. For example picking a file I know I haven't touched since last reboot... [serickson@serickson-pmg5:~/Documents/Downloads] [0:506] > dd if=xnu-792.13.8.tar of=/dev/null bs=1m 31+1 records in 31+1 records out 33351680 bytes transferred in 0.651121 secs (51221928 bytes/sec) [serickson@serickson-pmg5:~/Documents/Downloads] [0:507] > dd if=xnu-792.13.8.tar of=/dev/null bs=1m 31+1 records in 31+1 records out 33351680 bytes transferred in 0.056653 secs (588700803 bytes/sec) -Shawn _______________________________________________ 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 12/18/06, Cédric Luthi <cedric.luthi(a)gmail.com> wrote: That would be nice if you could tell me a bit more on how to actually use it. This email sent to site_archiver(a)lists.apple.com

1 0

Using C++ locales on Mac OS X
by Stefan van den Oord 18 Dec '06

18 Dec '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Domainkey-signature: a=rsa-sha1; q=dns; c=nofws; s=beta; d=gmail.com; h=received:message-id:date:from:to:subject:mime-version:content-type; b=cSupmN+5UagWdENXAbnH/SmCAhm/s63bRZWrVcG2Wj1f/r5PHGOzpAbOnN5WoUVaaewT7uHVEzzZsss578iVfUTtHGGDroXNmUH1AR03fPbrxoKkt0ZAaRUbaCET+f3X99s4ZJ2Q4lWv4vSiYyAl+boIvxMUw2PzDx9d+6bwE+c= _______________________________________________ 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: Re: darwin equivalent to linux readahead(2) function ?
by Cédric Luthi 18 Dec '06

18 Dec '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Domainkey-signature: a=rsa-sha1; q=dns; c=nofws; s=beta; d=gmail.com; h=received:message-id:date:from:to:subject:cc:in-reply-to:mime-version:content-type:content-transfer-encoding:content-disposition:references; b=pJEWd+dnPCbIZnCpmpPobwudydRChtGX9huk8U5RN1bQ+x8G7jNggvAAxmWHMAps/CVdZyVUhariYpBL6FqnFsUqFF0i8viNDRJPTFt3E2ec1wSOAPpkRhegoLl96+0HwvG0xF6pmybJhrrBpsp05tRNPWqKcYng4MuRtlzgiNk= On 12/18/06, Philippe Devallois <phdevallois(a)intego.com> wrote: fcntl(inFD,F_RDAHEAD,1); man fcntl Not so complicated ;) Actually, I'm not sure how to use it :-( I tried like that: #include <stdio.h> #include <unistd.h> #include <sys/stat.h> #include <fcntl.h> int main(int argc, const char* argv[]) { FILE* bigfile; int fd; struct stat sb; char buf[4]; bigfile = fopen("bigfile", "r"); if (!bigfile) return 1; fd = fileno(bigfile); if (fstat(fd, &sb)) { fclose(bigfile); return 1; } fcntl(fd, F_RDAHEAD, 1); read(fd, buf, 4); printf("bigfile size is %lld bytes\n", sb.st_size); fclose(bigfile); return 0; } But it seems it's not the way I want it to behave. Here is the (timed) result of the execution: $ time ./rdaheadtest bigfile size is 118053508 bytes real 0m0.055s user 0m0.002s sys 0m0.015s It obviously has NOT loaded the whole 113 MB file into memory because (1) I haven't heard my hard disk at all and (2) my PowerBook G4 is probably not able to load 113 MB in less than 0.055 s. I must be doing something wrong but the fcntl man is not really verbose about F_RDAHEAD and it's usage. That would be nice if you could tell me a bit more on how to actually use it. Regards. Cédric Luthi _______________________________________________ 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… Hope this helps, This email sent to site_archiver(a)lists.apple.com

1 0

Re: darwin equivalent to linux readahead(2) function ?
by Philippe Devallois 18 Dec '06

18 Dec '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Any pointer to a function or even more complicated method to achieve this goal is welcome. fcntl(inFD,F_RDAHEAD,1); man fcntl Not so complicated ;) Hope this helps, -- Philippe _______________________________________________ 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

darwin equivalent to linux readahead(2) function ?
by Cédric Luthi 18 Dec '06

18 Dec '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Domainkey-signature: a=rsa-sha1; q=dns; c=nofws; s=beta; d=gmail.com; h=received:message-id:date:from:to:subject:mime-version:content-type:content-transfer-encoding:content-disposition; b=symRyWWT2t9F9pU68gJGH7Mx0OkEAifEnqMx+xbFPVCgn9pHGkycJ+VNcoQo9DAAlUAZtaP6AVBwDlLmPFrOCqK12QxlJ3tJJIQntgnbdBaJakmrAoM4oMzuPDkde5YNp+Co/lziTJLboTrq1RZNNY3S9P6R3r+BbSvhG6XKESA= Hello list, I was wondering if there exists a darwin equivalent to Linux readahead(2) function. <http://www.die.net/doc/linux/man/man2/readahead.2.html> This is used to "populates the page cache with data from a file so that subsequent reads from that file will not block on disk I/O". I've been googling for readahead darwin, readahead bsd etc. but found nothing relevant :-( Any pointer to a function or even more complicated method to achieve this goal is welcome. Regards. Cédric Luthi _______________________________________________ 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: emulating wdticklerd
by Giuliano Gavazzi 18 Dec '06

18 Dec '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com the watchdog kext is now running on the production server. I found that I had to call PEHaltRestart(kPERestartCPU) to restart from the KEXT. Will let you know how it goes, that is... if the server hangs again! Since I detached from the RAID the disk that I suspect as the cause of the hangs [see my prev. thread], probably nothing will happen... In the meantime if you have any comments you are still welcome to browse the code on my public repository. g _______________________________________________ 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: emulating wdticklerd
by Giuliano Gavazzi 17 Dec '06

17 Dec '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com g _______________________________________________ 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 forgot to say: in the original version there was no userland process. The watchdog task, i.e. the one that resets the reboot timer, was in the KEXT too. The watchdog task would simply reset the reboot task timer and then reset its own timer. The only problem is that before resetting the reboot timer it had to call something that might block, and I do not think this is acceptable in a KEXT (would that not bring the whole system to a halt?). This is unfortunate for two reasons: it is more complex coding, and, above all, it is less secure. Anyone can call and set the reboot timer from userland... I will have to add some shared secret. This email sent to site_archiver(a)lists.apple.com

1 0

emulating wdticklerd
by Giuliano Gavazzi 17 Dec '06

17 Dec '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Giuliano _______________________________________________ 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… Hello again, I am aware that it has already been discussed in the past the possibility to have the MacOSX Server watchdog running on MacOSX to recover from hangs, but the discussion did not really get very far. In 10.4 this task is performed by wdticklerd. It does not appear that wdticklerd is in the open source tree. I have written in the past week about a disk problem that is causing hangs on a server of mine, but one thing I noticed that gave me the idea to write something that could restart the machine when all processes are waiting for the disk subsystem. I have written a small IOService KEXT (com_humph_kext_watchdog) with a IOUserClient to interface it to a userland tool. The KEXT can start and set a timer that when fired will reboot the machine (todo) and the userland tool, watchdog_ctl, has a loop that will resets the timer to a value a minute longer than the loop itself. Before resetting the timer though, it writes out a log line, so if stops there waiting, the timer in the kernel will fire. It remains to be seen what to call that will restart the machine without attempting to write to disk at all. Will reboot(RB_AUTOBOOT| RB_NOSYNC) be enough? If you would like to look at the code it is on a repository. The url to the subversion repository is it.humph.com/versions/public/ (I omitted the http to avoid bots..) user and pass are sguest and sub (just read privs, if I did not do anything wrong...) This email sent to site_archiver(a)lists.apple.com

1 0

Login and Logout items?
by Stephane 15 Dec '06

15 Dec '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com >From what I've read: According to the document available at http://developer.apple.com/ documentation/MacOSX/Conceptual/BPSystemStartup/Articles/ CustomLogin.html, you can define Login _and_ logout items. I know you can define Login Items but Logout Items is new to me. Is this really possible using the loginwindow.plist and how is someone supposed to define a logout item (since this is not explained in the documentation)? If this is just a typo in the documentation, what is the recommended way to start a task at logout? - loginhook are not recommended - launchd user agent are not 10.4 compatible - if logout is delayed by a tool (by canceling the quit AEvent and sending a logout (or shutdown in some cases) event later), an alert is displayed to the user. _______________________________________________ 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: further help on debugging a hang
by Paul Ripke 15 Dec '06

15 Dec '06

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Mail-followup-to: Giuliano Gavazzi <dev+lists(a)humph.com>, darwin-dev(a)lists.apple.com User-agent: Mutt/1.4.2.1i On Wed, Dec 13, 2006 at 09:17:40PM +0100, Giuliano Gavazzi wrote: > Thanks to all for the advice. I knew of the Kernel Debug Kit, my main > problem was to be sure that the correct key to press for an NMI was > the recessed and not the prominent one (the first hit I found was > indeed from someone who was thinking from inside the computer, and > was, incorrectly for us outside, defining the NMI as the convex one...). > > Unfortunately a hang means that I have too much information and not a > nice kernel panic that usually happens near the source of trouble. > > If anyone accustomed to kernel dump's showallstacks would like to > have a look at it: > > http://www.humph.com/tmp/dump.txt > > it's about 350KB 6536 lines. > > I have sent this same file to ACARD for them to examine. I'm no Darwin guru, but given the number of threads sleeping in various hfs_vnop_* routines, it's almost definitely disk related. I'd hope ACARD can give you further instructions to debug the hung system state. At least it doesn't look like a vnode deadlock, IMHO. Cheers, -- stix _______________________________________________ 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