Dear Jaeyong,
While I was browsing source code, I found two confusing variables: At genode/base/include/base/thread.h Thread_base::Context {
long stack[]; addr_t stack_base; }Although, they are commented a bit, it is hard to understand what are they doing and what's the difference. Could you tell me about them more?
please let me first refer you to the brief description of the concept behind the code:
http://genode.org/documentation/release-notes/10.02#New_thread-context_manag...
Each 'Thread_base' object is accompanied by a 'Thread_base::Context' object, which holds the thread's stack and information about thread-local storage (i.e., the UTCB) and similar meta data. In contrast to 'Thread_base' objects, which are plain C++ objects that can reside anywhere (on BSS, heap, or even on a stack), thread context objects are placed within the so-called thread-context area, which is a region of the process' virtual address space managed by the process itself. Each thread owns a 1MB portion of the context area. Only a tiny bit of this portion is actually used for the thread context. The gaps between the thread contexts are not populated with memory. This way, stack overruns will result in an unresolvable page fault (similar to a segmentation fault) rather than corrupted memory.
Each thread context consists of two parts, meta data expressed by the 'Thread_base::Context' type and the stack. Whereas the meta data has the same size and layout for each thread (well, the main thread is an exception here), the stack size differs from thread to thread. The metadata is always positioned such that it starts after the stack. Because the stack always grows from higher to lower addresses, an overrunning stack won't overwrite its meta data. The 'stack_base' variable points to the "other end" of the stack - the lower one which would get crossed if the stack overruns.
I hope this description could be of help. Do you have a suggestion about a comment that would have helped you to get a better understanding in the first place? I'd like to make the situation easier for future readers of the code.
Best regards Norman