site_archiver@lists.apple.com Delivered-To: darwin-dev@lists.apple.com Am 08.11.2005 um 21:03 schrieb Sam Hart <samh@granada-learning.com>: I think it is a bug in the 32bit handling too, albeit less dramatic. The process b0rked with a bus error rather than an "Out Of Memory". Here is a more correct version of said code: #include <stdio.h> #include <string.h> #include <stdlib.h> #define BUFFER_SIZE (1024*1024) int main(int argc, const char * argv[]) { long long int i = 0, total = 0; while (++i) { char *bp = malloc(BUFFER_SIZE * sizeof(char)); return 0; } Does this version crash with a bus error as well? Mike -- Mike Fischer Softwareentwicklung, EDV-Beratung Schulung, Vertrieb Address: Bundesstrasse 9, D-20146 Hamburg, Germany Phone: +49 (0)40/45038886, Fax: +49 (0)40/45038887 _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev@lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.appl... Given the source snippet Bob originally posted on this list I think it might simply be one of the following bugs in his code: - The stack gets overwritten when i > ARRAY_SIZE. - It tries to copy from NULL after malloc failed. - The strcpy function will copy an unspecified amount of data because buf is not a \0 terminated C string. Depending what comes after buf on the stack this will overwrite more or less dynamic memory which might also cause a crash. strncpy or memset would have been better. Oh and some other non crashing bugs or oddities in that code are: - The display of the allocated memory dosn't include the initial allocation of bigbuf so it is off by about 8MB (for 64-bit, or 4MB for 32-bit). - The choice of 1000000 is somewhat arbitrary, given that the page size of the VM system is 4096 bytes. I'd use 1MB or 1024*1024. if (bp == NULL) { printf("Internal malloc failed!\n"); break; // When this happens then the situation will not get better. Give up right away. } else { memset(bp, 'A', BUFFER_SIZE * sizeof(char)); // Access the memory so the VM system will actually allocate // physical memory, not just logical address space. // Note: it would be enough to touch one byte of every memory // page but the speed difference will not justify the more // complicated code as the VM thrashing will hide any // small perfomance gains made here. total = i; // Keep track of the total. if (i % 1000 == 0) // Don't print too much. { printf("MB allocated: %lld\n",i); } } } printf("Done! Total MB allocated: %lld\n", total); while (1) { } // So it doesn't exit! Really bad style to use an infinite loop but ok for this test! This email sent to site_archiver@lists.apple.com