fix(linker): correct several cross-toolchain memory map inconsistencies - #126

Not tied to a specific existing issue - found by systematically cross-checking each example's GCC (SW4STM32)/IAR (EWARM)/Keil (MDK-ARM) linker scripts against each other and against known-good Templates/Templates_LL sibling projects for the same chip, following the same methodology already used for STM32CubeG4, STM32CubeWB, and STM32CubeU5 in this repo family.

1. STM32F769I-Discovery and STM32F769I_EVAL (STM32F769NIH6, 2MB flash / 512KB RAM) - 18 examples, EWARM alone wrong

EWARM's .icf declared only 1MB flash (ROM_end=0x080FFFFF) and either 320K or 64K RAM instead of the chip's real 2MB / 512K, across 18 ordinary HAL peripheral examples on both boards. GCC and Keil both already correctly use the full 2MB/512K for the same projects - confirmed via _estack (computed as 0x20080000 = 0x20000000 + 512K) and each project's own GCC MEMORY block. Fixed all 18 EWARM files to ROM_end=0x081FFFFF, RAM_end=0x2007FFFF.

2. STM32F769I_EVAL FatFs example family (6 project identities, 18 files) - EWARM and GCC both wrong

Both EWARM (.icf) and GCC (.ld, across the USB-host-mode SW4STM32 variants) declared only 1MB flash / 320K RAM - only Keil (via its .uvprojx <Cpu> line, IRAM(0x20000000-0x2007FFFF) IROM(0x8000000-0x81FFFFF)) correctly used the full 2MB flash / 512K RAM. This is a "2-of-3-toolchains-wrong" pattern - a naive majority vote across toolchains would have picked the wrong answer here, similar to what was already found once in STM32CubeWB. Fixed EWARM's ROM/RAM end symbols and GCC's FLASH/RAM LENGTH to 2048K/512K in all 6 EWARM .icf + 12 GCC .ld files.

3. NUCLEO-F767ZI (STM32F767ZIT6, 2MB flash / 512KB RAM) - 25 examples, Keil alone wrong

Keil ships no explicit .sct scatter file for these projects - the project's own <Cpu> device-descriptor line in Project.uvprojx is what it actually links against - and that line declared only 320K RAM (IRAM(0x20000000-0x2004FFFF)) instead of the chip's real 512K. Confirmed against GCC/IAR agreement on 512K for the same projects, and cross-checked across two different Keil device-pack versions used by different examples on this board (2.3.3 and 2.7.5), ruling out a pack-specific fluke rather than a real content bug. Fixed the IRAM range to 0x20000000-0x2007FFFF in all 25 files.

The same board's plain Templates project (as opposed to Templates_LL, which already agrees on 512K across all 3 toolchains and served as the gold standard for this finding) had this exact same bug in both its GCC .ld and Keil .uvprojx, with only its own IAR .icf correct - fixed those 2 files too.

Test plan

No local ARM toolchain (arm-none-eabi-gcc/IAR/Keil) available to compile/link-test these, so verification relied on address-arithmetic cross-referencing for each finding against multiple independent references: Templates/Templates_LL gold standards, sibling toolchain files of the same project, and checking _estack/actual linker symbol usage rather than trusting MEMORY{}/Cpu-line declarations alone.

Note on scope

A 4th candidate was found (STM32F7308-DISCO / STM32F7508-DISCO, flash-less QSPI-execute-in-place parts where Keil ships with no scatter file at all and falls back to a non-functional 64KB internal-flash default) but is not included in this PR - fixing it would require authoring a new Keil scatter file for QSPI XIP from scratch rather than correcting an existing value, which is a larger, more design-heavy change that deserves its own separate scrutiny/PR rather than being bundled here.

Disclosure

Generative AI (Claude) was used to help investigate this (systematic cross-toolchain linker script comparison) and implement/verify the fixes. All changes were reviewed by me before submission.