[dgarske]

Description

Follow-up to #191 from the work on #192: one driver change and two documentation entries for STM32H7R/S.

Transmit from a static buffer

wolfIP builds each outgoing frame on the stack of the task that sends it, and hal_eth_send() passed that pointer straight to the Ethernet DMA. The DMA cannot read core-coupled memory, so with a stack in DTCM (or CCM on other families) nothing was transmitted. With STM32_HAL_ETH_TX_BOUNCE, hal_eth_send() copies the frame into a static, cache-line-aligned buffer and sends from there. It is on by default for STM32H7R/S, whose vendor linker scripts put the main stack in DTCM, and STM32_HAL_ETH_NO_TX_BOUNCE turns it off there. Other families are unchanged unless they define it.

Documentation

IDE/STM32Cube/README.md gains two troubleshooting entries:

• A BusFault in atomic operations when the AXI SRAM region is marked Shareable, and the fact that Shareable memory is not D-cached.
• Task stacks in DTCM or CCM, the define above, and that the driver's static buffers must be linked into DMA-reachable memory.

Testing

NUCLEO-H7S3L8, FreeRTOS port. With every task stack in DTCM the driver without the copy transmits nothing and with it serves HTTP normally; with stacks in AXI SRAM both behave the same. The atomics BusFault was reproduced on the same board. Not tested on other STM32 families.

[wolfSSL-Fenrir-bot]

Fenrir Automated Review — PR #194

Scan targets checked: wolfip-src, wolfip-bugs
Not reviewed (no scan target covers them): IDE/STM32Cube/README.md
Coverage: 1 of 1 in-scope changed file(s) opened by the reviewer

Fenrir result: Approved ✅

No new issues found in the changed files.

Advisory only — this automated result does not count as a GitHub approval.

Review tier: Max

[Copilot]

🟡 Changes recommended

The README incorrectly generalizes the speculative-read hazard to peripheral registers, so the explanation should be scoped to the GFXMMU window in the default map’s Normal-memory range.

1 open issue

• Limit the speculation warning to GFXMMU’s Normal-memory range · New

What changed in this PR

This documentation-only PR explains an STM32H7R/S freeze relevant to issue #192, caused by speculative reads into the unclocked GFXMMU window. It places the mitigation in the application’s MPU configuration rather than the wolfIP Ethernet driver.

Changes:

• Adds freeze symptoms, a debugger confirmation step, and an MPU no-access configuration example.
• Adds a pointer to the troubleshooting guidance in the Ethernet driver’s implementation notes.

File	Description
src/​port/​stm32_hal/​stm32_hal_eth.h	Adds a documentation pointer for the GFXMMU freeze and mitigation.
IDE/​STM32Cube/​README.md	Documents the freeze, diagnostic check, and application-level MPU mitigation.

_{Review effort: Balanced}

[wolfSSL-Fenrir-bot]

Fenrir Automated Review — PR #194

Scan targets checked: wolfip-src, wolfip-bugs
Not reviewed (no scan target covers them): IDE/STM32Cube/README.md
Coverage: 1 of 2 in-scope changed file(s) opened by the reviewer; not opened: src/port/stm32_hal/stm32_hal_eth.h

Fenrir result: Approved ✅

No new issues found in the changed files.

Advisory only — this automated result does not count as a GitHub approval.

Review tier: Lite

[Copilot]

🟢 Approval recommended

No active review issues remain for the STM32 Ethernet TX bounce-buffer and memory-configuration documentation changes.

No open issues 🎉

1 resolved since last review

• Limit the speculation warning to GFXMMU’s Normal-memory range

_{Review effort: Balanced}