Skip to main content
Link
Menu
Expand
(external link)
Document
Search
Copy
Copied
Search cs492-stevens
Lectures
Lecture 11 Annotated
11 - Linux Processes
Outline
Process Address Space
Task Struct
More Reading
Linux Clone Syscall
/proc
Announcements
Reading: MOS 10.3.1, 10.3.2, 10.3.3; LDD 2
Process Address Space
Kernel address space in Linux is NOT separate.
(Apologies for stating otherwise in previous lecture)
The process address space is split into user- and
kernel-space.
Kernel is mapped into same virtual memory range
for every process.
Kernel execution contexts must be maintained
separately for each process
While in user-mode, access to kernel-mode address
range is denied by MMU.
It is possible for kernel to live in different
address space, but split space makes some things
easier (like copying user data)
Linux represents a process as a task.
Equivalent of PCB:
Field categories:
Notable fields:
Kernel can access the calling process via:
User-mode processes can get information about
other processes through the special /proc file
system in Linux.
Recap: fork & exec
Fork is part of the POSIX standard for creating
processes.
Clone creates tasks, which can represent either
processes or threads depending on the amount of
sharing.
Linux will assign a PID based on parameters.
share nothing = fork
share everything = create thread
Linux creates new tasks with the clone syscall.
Some information from task_struct available via:
See #proc in Linux Filesystem Hierarchy
A special macro to get the pointer to the
current user-space process.
generalization of fork
not part of POSIX, not portable!
Starts executing function instead of full copy of
process.
Execution can optionally use new stack.
sharing_flags describes amount of sharing
between caller and callee (parent and child)
Source:
Task is a generalization of an execution context.
Linux uses tasks to represent threads as well.
Scheduling params (priority, CPU time, etc.)
struct thread_info thread_info
Memory image (text, stack, heap, etc.)
unsigned int __state
Signals
void *stack
CPU registers
unsigned int flags
System call state (parameters, result, etc.)
int prio, static_prio, normal_prio
struct mm_struct *mm
pid_t pid
struct task_struct *parent
u64 utime, stime, start_time
char comm[TASK_COMM_LEN]
File descriptor table
Accounting (user time, system time, limits, etc.)
Kernel stack (per-process)
Miscellaneous (PID, process state, etc.)
pointer to memory management struct
name of loaded program in process
During the context switch, access permissions
change with kernel-mode entry, and kernel-mode
address range is accessible.
Task Struct
Linux Clone Syscall
PA 2 autograder almost done.
No in-person lecture on Tuesday.
I willl post a recording on Monday.
1)
2)
3)
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
00000000_00000000
ffff8000_00000000
00007fff_ffffefff
ffffffff_ffffffff
user-space
virtual memory
(switched)
kernel-space
virtual memory
(shared)
...
see Linux mm documentation
see Lorenzo Stoakes's notes
task_struct
current
include/linux/sched.h
arch/x86/include/asm/current.h
fork
clone
kernel_clone
copy_process
wake_up_new_task
source code:
/proc
`-- <PID>
|-- status # process status info
|-- comm # program name
|-- exe # program symlink
|-- maps # mapped memory
...
child_id = clone(function, stack_ptr, sharing_flags, arg);
MOS Figure 10-8
MOS Figure 10-9
Both
and
call
calls
does memory copies
puts process in ready queue
linux-kernel-labs - Processes
linux-kernel-labs - Address Space
Lorenzo Stoakes - VM Notes: Process Address Space
Lorenzo Stoakes - MM Notes: Virtual memory layout
Linux Kernel Docs - Memory Management Documentation
Linux Kernel Docs - Process Addresses
and
kernel_clone
copy_process
wake_up_new_task