Hey Goetz, 2008/11/24 Goetz Babin-Ebell <[EMAIL PROTECTED]>: >> That is okay. But, those are mapped memory, not actually used. Mapped >> memory does not consume memory until it is actually used (i.e. >> swapped in). > > And that is the problem: > It is an embedded device, which means no swap. > And unfortunately with all the running applications and modules > we reached the limit at which every additionally used K can mean > an oops...
The way I understand it, linux overcommits memory: malloc() will succeed, memory will be mapped for the process, but pages will only get allocated when the memory is effectively used (by reading or writing). This means that the totally mapped memory used by all processes together in your system may easily exceed the total amount of physical memory. I would expect this behaviour to be regardless of underlying architecture (although it can be switched off apparently). So, in your case, if big contiguous chunks of memory are not used (such as thread stacks), you should not count them as memory consumption, even if you have no swap space. The result of segfaulting on a page that was mapped but is not currently active in memory, has as usual consequence the "swapping in" of the page, since this is the exactly same mechanism used when having swap space. But in your case, for a new page it does not involve actual swap space, since no content was provided yet for the page. Ofcourse you should be sparing with memory, but therefore it is even more important to understand the actually used memory resources. If you would have the ability to build your application also on i386 I would recommend using valgrind to do some profiling. Regards, koen ------------------------------------------------------------------------- This SF.Net email is sponsored by the Moblin Your Move Developer's challenge Build the coolest Linux based applications with Moblin SDK & win great prizes Grand prize is a trip for two to an Open Source event anywhere in the world http://moblin-contest.org/redirect.php?banner_id=100&url=/ _______________________________________________ witty-interest mailing list witty-interest@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/witty-interest
