OS Internals & Processes

Threads inside one process share the same address space; separate processes do not share memory by default. In the Linux kernel both are the same primitive — a Task — differing mainly in what they share. Processes can still exchange data through sockets, pipes, or signals, or share memory explicitly via shm_open or mmap.

Yes, it is real. top reports per-process CPU as a sum across all cores, so 146% means the process is using roughly 1.46 cores at once — entirely normal for a multi-threaded program on a multi-core box. Without more context you cannot say it is a problem; in the general case it is fine. Treating any value over 100% as impossible is the red-flag answer.

A file descriptor is a small integer the OS hands back when you open a resource, used for later reads, writes, and other operations. It identifies not just disk files but also pipes, sockets, devices, and memory-mapped files. By convention 0 is stdin, 1 is stdout, and 2 is stderr — every process starts with those three open.

Find the PID with ps aux | grep or top, then send a signal with kill <pid>. Plain kill sends SIGTERM (15), asking the process to shut down gracefully and release resources. kill -9 sends SIGKILL, which the process cannot catch or handle — it dies instantly with no cleanup, so you use it only when SIGTERM is ignored.

Application code runs in user space — a restricted CPU mode that sees only the process's own virtual memory and cannot run privileged instructions. The kernel runs in kernel space with full access to devices, memory management, and the whole instruction set. The split isolates the OS so a buggy or malicious app cannot corrupt the system; crossing it requires a syscall.

A page is the fixed-size unit (commonly 4 KB) the OS uses to manage virtual memory. The kernel maps each virtual page to a physical frame, to a file, or to swap, and per-page permission bits protect regions from being written or executed. Paging lets processes share physical RAM safely and gives each one its own isolated address space.

It depends on swap. With swap enabled, the system can limp along for a long time, paging memory out to disk. With swap off, the kernel's OOM killer starts terminating processes to reclaim memory — and not necessarily yours. The OOM killer decides based on resident memory (RSS, what is actually mapped to physical RAM), not the larger virtual size.

A syscall is a program's request for an OS service it cannot do itself — reading a file, opening a socket, forking. Making one traps the CPU from user mode into kernel mode: it saves user state, switches to the kernel's privileged context, runs the handler, then returns. For example fork clones a process and epoll lets one goroutine wait on many file descriptors becoming readable.

A mutex grants exclusive access, blocking waiters (the OS deschedules them). A semaphore allows up to N holders, generalizing a mutex. A spinlock busy-waits without sleeping — cheap only for very short critical sections on multiple cores. Advisory file locks (flock) coordinate across processes. On Linux a futex is the fast userspace primitive most of these are built on.