Hello, Samuel Thibault, le mer. 01 nov. 2023 16:06:57 +0100, a ecrit: > Samuel Thibault, le mer. 01 nov. 2023 15:35:00 +0100, a ecrit: > > Samuel Thibault, le mer. 01 nov. 2023 13:14:17 +0100, a ecrit: > > > Samuel Thibault, le mer. 01 nov. 2023 01:50:40 +0100, a ecrit: > > > > Samuel Thibault, le mar. 31 oct. 2023 04:40:43 +0100, a ecrit: > > > > > (it looks like there are memory leaks in proc, its vminfo keeps > > > > > increasing). > > > > > > > > It seems 64bit-specific: the program below makes proc leak memory, 100 > > > > vminfo lines at a time. Possibly __mach_msg_destroy doesn't actually > > > > properly parse messages to be destroyed, so that in the error case the > > > > server leaks non-inline data? Flavio, perhaps you have an idea? > > > > > > I don't think we have the kernel-to-user equivalent for > > > adjust_msg_type_size? So that we end up pushing twice too much data to > > > userland for port arrays? > > > > I found and fixed the allocation issue in the kernel. > > It seems proc is still leaking, but on the heap this time. This is not > 64bit-specific, the same simple reproducer triggers it: > > while [ "$(echo -n `echo a` )" = a ] ; do : ; done > > or more simply: > > while true ; do echo $(echo -n $(echo a)) > /dev/null ; done
I found the issue, it's because of the quiescence support in libports, which assumes that all threads will sooner or later go through a quiescent state (because it finished processing a message). But that's not true, proc doesn't set a thread timeout, and thus some threads can stay indefinitely stuck in receiving messages. And thus the deferred dereferencing used by ports_destroy_right never gets achieved. I'll push a fix. Samuel
