Milestone 5 - Mutlicore

[Liblor]

Tasks:

• [x] Bring up second core #110
• [x] Basic Inter-Core Communication #113
• [x] Multicore Memory Management #113
• [ ] Refactor code
  • [ ] properly implement bootinfo memory assignment (we currently can only boot core 0 and 1)
  • [x] refactor urpc_receive_bootinfo to return bootinfo
  • [x] remove code duplication in urpc_slave_serve_req (bootinfo)
  • [x] make urpc_slave_serve_req serve one single request, caller loops instead
  • [x] thread yield instead of busy loop
  • [x] write down protocol used for urpc
  • [x] rename comandline_len to comandline_size
  • [x] error handling (malloc)
• [ ] write tests
• [x] merge this PR with thread safety PR
• [x] enable grading functions on 2nd core

Discussion topics:

• aos_rpc_init where to call?
• barriers
• thread safety core1: spawn functions

Milestone summary (demo):

• Booting 2nd core successfully
• init0 and init1 can communicate
• Applications on each core handle page faults
• init0 is able to start programs on core 1
• Your user lvl messaging protocol used for inter-core communication
• How the second core knows which memory belongs to it, and how you manage memory allocations between cores
• What you needed to do to make sure interrupts are delivered to both cores
• How barriers work in your implementation and why

[abertschi]

urpc protocol for this milestone as of writing of this note:

• core0 is master, core1 is slave
• core0 can send core1 one message at a time
• next message can be sent once response received
• shared frame is allocated and mapped into vmemory of init dispatcher on both cores

• shared frame contains the following data structure:

  struct urpc_shared_mem {
  enum urpc_msg_type type;
  enum urpc_status status;
  
  union {
      struct bootinfo_msg bootinfo_msg;
      struct urpc_spawn_request spawn_req;
      struct urpc_spawn_response spawn_resp;
  };
  };

• slave loops in a thread on urpc_shared_mem->status until status indicates that datastructure contains data for slave.
  • urpc_shared_mem->type encodes struct available in union.

how to spawn a process on core1: usecase: binaryA on core0 spawns hello on core1: core0 to core1:

• binaryA -> lmp -> init process server -> urpc -> core1 spawn hello
• core1 response -> urpc -> init process server -> lmp -> binaryA

[abertschi]

• new changes implement work of init on core 1 in one single thread (instead of 2). This avoids potential issue in paging but does not fix it. https://github.com/Liblor/advanced_operating_systems_2020/pull/120

[Liblor]

When I compile the code I get a binary that produces a binary with the following error:

hello.1.0: libc exit NYI!
multicore_test.0.0: spawned hello: pid 99 on core 0
armv8/sbin/init.1.0: got UrpcMasterData
armv8/sbin/init.1.0: name is: /armv8/sbin/hello
kernel 1 PANIC! oops: frame offset invalid

However, if @abertschi builds it, the issue doesn't occur (I tried his binary it worked without a problem). This is pretty weird as we both build using podman

[eikendev]

Have you both removed the build directory and called Hake from scratch?

[eikendev]

When I compile the code I get a binary that produces a binary with the following error:

hello.1.0: libc exit NYI!
multicore_test.0.0: spawned hello: pid 99 on core 0
armv8/sbin/init.1.0: got UrpcMasterData
armv8/sbin/init.1.0: name is: /armv8/sbin/hello
kernel 1 PANIC! oops: frame offset invalid

However, if @abertschi builds it, the issue doesn't occur (I tried his binary it worked without a problem). This is pretty weird as we both build using podman

I rebuilt on ec484ae4aab4ef63f80dc3f718c6715afa139aa3 and get the same error message. @abertschi Can you remove build/ and run make hake && make install to confirm?