Vorago VA416x0: simplify iram_write/iram_fill (single RMW path)

[dgarske]

Continuation of #778

Round-2 cleanup on the IRAM shadow helpers:

• Drop the word-aligned fast path. The volatile uint32_t * RMW path already minimizes word traffic (one ldr + one str per output word through grouped byte merges), and the fast path required BOTH src AND dst aligned -- ext_flash_write/read pass through the caller's source buffer, so an unaligned caller dropped the whole copy to the slow path anyway.

• Drop the d/s alignment snapshot locals (only fed the fast-path test).

• Drop the (uintptr_t)3u casts. uintptr_t is 32-bit on Cortex-M4 (same as unsigned int); the casts were defensive portability noise on a hardware-specific helper.

The volatile uint32_t store guarantee from round 1 is retained -- it is the load-bearing part (one STR per store, which the IRAM ECC requires).

Code-size win on wolfboot.elf (ECC384/SHA384 config):
.text 42348 -> 42292 (-56 B)
iram_write 94 -> 64 B (-30 B)
ext_flash_erase (inlined iram_fill) 96 -> 72 B (-24 B)

Verified on EVK hardware (vorago_va416x0.config):
build_test.sh clean -- v1 boots, signature OK, state NEW
build_test.sh update -- v1->v2 swap, signature OK, state TESTING,
"Booting new firmware, marking successful boot"
(5-byte pBOOT confirmation write -- exercises
byte-tail of the RMW loop)

[Copilot]

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

Simplifies the VA416x0 IRAM shadow write/fill helpers by removing the word-aligned fast path and relying solely on the volatile RMW loop, reducing code size.

Changes:

• Remove word-aligned fast path from iram_write and iram_fill.
• Drop alignment snapshot locals and (uintptr_t)3u casts.
• Update the explanatory comment to reflect the single-path approach.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.