mmio-and-bit-manipulation

MMIO and Bit Manipulation

Purpose

Guide agents through safe memory-mapped I/O: volatile semantics, read-modify-write patterns, bitfield pitfalls, alignment and endianness, and portable register access macros for bare-metal drivers.

When to Use

• Writing peripheral register drivers without HAL
• Fixing intermittent register corruption or stale reads
• Replacing C bitfields with explicit masks
• Porting drivers between little-endian MCUs
• Auditing ISR vs main-line register access

Workflow

1. MMIO fundamentals

Peripheral registers live at fixed addresses in the CPU memory map. The compiler must not cache reads/writes.

#include <stdint.h>

#define PERIPH_BASE   0x40000000U
#define GPIOA_MODER   (*(volatile uint32_t *)(PERIPH_BASE + 0x20000U))

Qualifier	Effect
volatile	Forces load/store each access — required for hardware
const volatile	Read-only hardware (rare)
Plain uint32_t *	Wrong — compiler may optimize away

2. Read-modify-write macros

#define REG32(addr)        (*(volatile uint32_t *)(addr))
#define REG_SET(addr, mask)   (REG32(addr) |= (mask))
#define REG_CLR(addr, mask)   (REG32(addr) &= ~(mask))
#define REG_TOGGLE(addr, mask) (REG32(addr) ^= (mask))
#define REG_WRITE(addr, val)  (REG32(addr) = (val))
#define REG_READ(addr)        (REG32(addr))

Good — atomic intent for single-bit updates when register supports it:

#define GPIOA_BSRR  REG32(0x40020018U)
GPIOA_BSRR = (1U << 5);        /* set PA5 */
GPIOA_BSRR = (1U << (5+16));   /* reset PA5 — STM32 BSRR pattern */

Bad — non-atomic RMW on interrupt-shared registers:

uint32_t v = REG_READ(GPIOA_MODER);
v |= (1U << 10);
REG_WRITE(GPIOA_MODER, v);  /* ISR may interleave — lost update */

Fix: disable IRQ briefly, use hardware set/clear registers, or LL atomic bitband if available.

3. Bitfield pitfalls

/* Bad — layout is implementation-defined, not portable */
typedef struct {
    uint32_t mode  : 2;
    uint32_t type  : 1;
    uint32_t speed : 2;
} gpio_moder_bits_t;

Prefer explicit masks:

#define GPIO_MODER_MODE0_SHIFT   0
#define GPIO_MODER_MODE0_MASK    (3U << GPIO_MODER_MODE0_SHIFT)
#define GPIO_MODER_MODE0_VAL(n)  ((n) << GPIO_MODER_MODE0_SHIFT)

REG32(GPIOA_MODER) = (REG32(GPIOA_MODER) & ~GPIO_MODER_MODE0_MASK)
                   | GPIO_MODER_MODE0_VAL(1);  /* output */

4. Endianness and alignment

• Cortex-M and most MCUs: little-endian — uint32_t MMIO at word-aligned addresses
• Unaligned uint32_t access may fault on ARMv7-M+
• 8-bit registers: use volatile uint8_t with correct byte lane address

#define REG8(addr)  (*(volatile uint8_t *)(addr))

5. Memory barriers (when needed)

/* After configuring peripheral before first use */
__DSB();
__ISB();

/* After DMA setup, before enabling channel */
__DMB();

Use CMSIS barriers (core_cm4.h) on Cortex-M.

6. Agent usage examples

/mmio-and-bit-manipulation Safe pattern to set bit 3 without affecting other bits in ISR context
/mmio-and-bit-manipulation Why must peripheral pointers be volatile?

Common Problems

Symptom	Cause	Fix
Register write ignored	Wrong address/clock gated	Enable peripheral clock first
Random bit flips	RMW race with ISR	BSRR-style atomic regs or critical section
HardFault on access	Unaligned or protected bus	Match access width to datasheet
Optimized-away read	Missing volatile	Use volatile uint32_t
Bitfield wrong value	Compiler packs unexpectedly	Use shift/mask macros

Related Skills

• skills/baremetal/peripherals-from-datasheet — extracting register maps
• skills/baremetal/gpio-baremetal — GPIO register patterns
• skills/low-level-programming/assembly-arm — inline asm barriers
• skills/embedded/linker-scripts — peripheral memory map regions