Hi, I'm running a multi-threaded application using pthreads + m5threads using SE mode. I see that as the simulation progresses, the memory consumption keeps increasing for the process until it gets killed. the process map shows that the address space is getting filled up with the application's code segment (r---s-), (hello-world-pthreads for example) and with eveery new thread I see a new mapping in the address space, like so.
00007f84a757e000 8268 6220 0 r--s- hello-world-pthread 00007f84a7d91000 8268 6860 0 r--s- hello-world-pthread 00007f84a85a4000 8268 6860 0 r--s- hello-world-pthread 00007f84a8db7000 8268 6920 0 r--s- hello-world-pthread 00007f84a95ca000 8268 6908 0 r--s- hello-world-pthread 00007f84a9ddd000 8268 6860 0 r--s- hello-world-pthread 00007f84aa5f0000 8268 6860 0 r--s- hello-world-pthread 00007f84aae03000 8268 6860 0 r--s- hello-world-pthread ... ... ... I checked the m5threads implementation - the TCB is allocated using an mmap. The mmap is getting trapped correctly in gem5. I've uncommented the munmap in m5threads pthread_join to be safe to ensure this is unmapped and released. I suspect the actual segment mapping is done by the clone syscall (Sorry not an expert in this, I may be wrong) and typically when the thread completes, these mappings should be freed? Any insights on how I can fix this? If anyone has faced this issue before? Would appreciate pointers to this. Thank you, -Andronicus
_______________________________________________ gem5-users mailing list -- gem5-users@gem5.org To unsubscribe send an email to gem5-users-le...@gem5.org %(web_page_url)slistinfo%(cgiext)s/%(_internal_name)s
