Question 1

How do you design a real-time embedded system to meet hard deadline requirements?

Accepted Answer

Use a Real-Time OS (RTOS like FreeRTOS), fixed-priority preemptive scheduling, avoid dynamic memory allocation (use static pools), minimize ISR duration, and perform WCET (Worst-Case Execution Time) analysis. RTOS guarantees: deterministic scheduling, priority inversion protection (priority inheritance mutexes). No malloc: heap fragmentation is non-deterministic. ISRs must be short — defer work to tasks. WCET analysis: prove statically that the worst-case path meets the deadline.

Question 2

When would you implement a custom memory allocator instead of using malloc?

Accepted Answer

When you need deterministic allocation time, fragmentation control, pool allocation for fixed-size objects, or when malloc's overhead is too high for real-time or high-frequency allocation patterns. Pool allocator: pre-allocate N objects of size S — O(1) alloc/free, zero fragmentation. Arena/bump allocator: allocate linearly, free everything at once — great for request-scoped allocations. jemalloc and tcmalloc are production alternatives to glibc malloc with better multi-threading performance.

Question 3

What is ABI (Application Binary Interface) compatibility and why does it matter for shared libraries?

Accepted Answer

ABI is the binary contract between compiled modules — struct layout, calling conventions, name mangling; breaking ABI means existing compiled code stops working with the new library version without recompilation. API break: source code must be updated. ABI break: compiled binaries break silently or crash. Safe ABI changes: add functions, add struct fields at the end. Breaking: remove fields, reorder fields, change types, change function signatures. Use symbol versioning and soname bumping for major changes.

Question 4

How would you ensure memory safety in a large C codebase?

Accepted Answer

Enforce strict ownership conventions (each allocation has one owner), use ASan/Valgrind in CI, static analysis (Coverity, clang-tidy, cppcheck), fuzz testing (AFL, libFuzzer), and thorough code review. Defense in depth: conventions (RAII-like patterns) catch most bugs. ASan catches the rest at test time. Static analysis finds common patterns without running code. Fuzzing finds edge cases no human test would cover. For new safety-critical code, consider Rust interop via FFI.

Question 5

How would you implement zero-copy data transfer in a C networking application?

Accepted Answer

Use sendfile() to transfer file data directly from the file descriptor to a socket without copying to user space; use mmap + write for shared memory between processes; or use kernel bypass (DPDK) for ultra-low latency. Normal file serving: disk → kernel buffer → user buffer → socket buffer (2 copies). sendfile(): disk → kernel buffer → socket buffer (0 user-space copies). mmap: file is directly in virtual address space. DPDK: NIC DMA directly to user-space ring buffer, bypassing the kernel entirely for ~1µs latency.

C Programming — Tech Lead Interview Questions

More Tech Lead Topics

C Programming at Other Levels