Why an access needs intervention
Normal paged access depends on suitable mappings and permissions. Nonresident pages, mappings requiring setup, or protection violations can need system intervention, with exact exception categories defined by the architecture.
A fault is therefore not automatically a programming error. Conversely, an invalid address does not become legitimate merely by loading a page. The operating system must interpret both hardware information and its own address-space records.
A valid request can resume after handling
For a valid page backed by external storage, handling obtains a frame, retrieves contents, updates the mapping, and resumes the access under the system’s rules. Other ready work can run while I/O is pending.
Backing contents may come from swap or a program file. Frame shortages can additionally require replacement and reclamation, explaining why one apparent memory access can occasionally incur much greater delay than an ordinary RAM access.
References: [1]
Separating faults, I/O, and TLB misses
A missing TLB entry can refer to a page already resident in memory. Looking up the page table and filling the translation is distinct from retrieving page contents, so these events should not be conflated.
Linux getrusage distinguishes faults handled without I/O from those requiring it. Fault counts alone therefore do not reveal disk traffic or total delay; diagnosis needs the actual categories and observed costs.