xv6-riscv/notes/chapter2-os-organization

110 lines
4.8 KiB
Text

XV6 as most OSs is monolithic, preemptive and time-sharing multiplexing.
Isolation comes from the CPU which provides machine, supervisor and user modes.
- machine: just for the boot time
- supervisor: each time the kernel is active
including when calling a syscall (with `ecall` in RISC-V)
- user: for all user code
Minix, L4 (seL4) and QNX operating systems have been mentioned as micro-kernel OSs.
Memory layout of a process:
[text; data; user stack; heap (large) ; trapframe; trampoline]
0 ↑------------------------------------------------------------↑ 2^38-1
text: instructions
data: global variables
heap: explicitly allocated memory
trapframe: saved process registers when switching in and out of the kernel
trampoline (4 KiB): code to transition in and out of the kernel
trapframe & trampoline: explained in chapter 4
kernel `proc` structure contains:
- (kstack) kernel stack used when the process calls for syscalls
- (pagetable) pointers to physical memory pages actually used by the process,
provided to the hardware for translation (virtual @ ←→ physical @)
- (state) UNUSED, USED, SLEEPING, RUNNABLE, RUNNING or ZOMBIE
- (trapframe) saved process registers when switching in and out of the kernel
- (pid) process identification number
- (ofile) list of opened files
- (name) process name
- (cwd) current directory
- (context) kernel registers used to enter the process
- (sz) size of process's memory
- (parent) pointer to the process's parent proc structure
- (xstate) exit status (given to its parent when it "waits" for it)
- (killed) non-zero when the process has been killed
- (chan) TODO: not currently explained
- (lock) TODO: not currently explained
RISC-V instructions
- ecall: raise hardware privilege level
program counter change to a kernel-defined entry point which then switches to the
process's kernel stack and executes kernel instructions for this syscall
once done, the kernel calls sret
- sret: lower hardware privilege level
A process is the abstraction of memory and CPU for a running program,
giving it the illusion of being alone on the hardware.
A process is:
- an address space to give a running program the illusion of owning the entire memory
- a thread to give a running program the illusion of having a CPU for himself
starting xv6, the different phases
context: booting up the machine
=> paging hardware disabled (virtual memory == physical memory)
=> booting at phy@ 0x80000000 because 0-0x80000000 contains IO devices
=> FYI: stack grows DOWN
boot loader loads xv6 into memory then jumps to _entry (kernel/entry.S:7)
phases for _entry:
1. sets up a 4096-byte stack for each hardware thread (HART)
(hart = "hardware thread" as opposed to software-managed thread context)
these stacks start at the address "start0" defined in C code (kernel/start.c:11)
2. loads stack0+4096 in sp (stack pointer)
(which is the top of the first stack because stacks grow DOWN)
3. jumps to C function "start" (kernel/start.c:14)
phases for "start":
=> main idea: "start" performs machine-mode configuration then jumps to "main"
ex: interruptions, exceptions and Physical Memory Protection configuration
1. configures supervisor mode (related to the `mret` RISC-V instruction)
mret enables to "return" from a mode to previous one
mret in this case is first _configured_ to jump to supervisor mode
a. mstatus (previous mode) is set to "supervisor"
b. mepc (return address) is set to the address of "main" (kernel/main.c:10)
c. satp (page-table register) is set to 0
=> disables virtual address translation in supervisor mode
d. delegates all interruptions and exceptions to supervisor mode
2. sets a timer interrupts on the clock chip
3. changes to supervisor mode with `mret` while jumping to "main" (kernel/main.c:10)
phases for "main":
1. initializes devices, subsystems and a lot of stuff in general
2. calls "userinit" (kernel/proc.c:233) to set up the first "user process"
=> it is just the creation of the process from a kernel point of view ≠ execution
=> the process is then in "RUNNABLE" state
=> the program for this process is in initcode.S (kernel/initcode.S:3)
(but compiled for some reason into the kernel/proc:221 char array)
3. calls the scheduler
executes the only "RUNNABLE" process in the list, made by "userinit"
=> this "initcode" executes the /init application
here is the code:
1. a0 = address to the "/init" string
2. a1 = argv for the future process
3. EXEC syscall to run the init program with provided parameters (a0 & a1)
phases for /init:
1. creates a console device (if needed)
2. opens file descriptors
3. starts the shell on the newly created console
system is up and running, yay!