cksystemsteaching
Thread-safe malloc grading task
Quickly added the grading for this task.
This is ready for merge.
Changes
Previously
treiber-stack task:
- check lr/sc RISC-V encoding
- check lr/sc assembly instruction format
- test push() and check if everything is in memory
- test pop() and check if everything pushed is popped correctly
Now
threadsafe-malloc task:
- check lr/sc RISC-V encoding
- check lr/sc assembly instruction format
- test if syscall brk (on malloc) does not force a context switch anymore
- test thread-safe malloc
treiber-stack task (requires passing threadsafe-malloc):
- test push() and check if everything is in memory
- test pop() and check if everything pushed is popped correctly
How it works
Right now malloc() does the following:
- load (get bump pointer)
- ecall (syscall brk: increase program break)
- store (set bump pointer to new program break)
However, syscalls currently force a context switch. This means that if thread 1 does malloc followed by thread 2 doing a malloc, they both get the same address resulting in an overwrite one way or another. This is what is being tested and is also required to be solved (very simple to do).
Notes
I said earlier today that this breaks older assignments. Well, this is only the case if you make syscalls not force a context switch in general, instead of just doing that for brk. By trying a new implementation all older assignments still work without having to change anything that is related to any syscall other than brk (as in: it is not much work). TL;DR: Older assignments do not break.
And on another note: Since the old treiber-stack grading task was "only" testing on Mipster (not Hypster), I decided to keep it that way. Another reason being that the threadsafe test requires a timing in instructions that I so far only achieved in Mipster.
I need to change the threadsafe-malloc.c test to somehow force a context switch between the lr and sc calls (since the brk syscall is not expected to do that anymore, there is a new test to check exactly that). Probably by counting or guessing the amount of instructions left to execute and adding the malloc call to the end of that so lr is included but sc is not (timer related context switch before sc).
Maybe I can do something like compile code and output as binary, then change said binary to add a bunch of nops before the malloc-jal so the above happens and then load this "hacked" binary back in and execute it.
@CAS-ual-TY Looks good! Just keep the file permissions of Makefile as is. Should not be executable...