Document source: e47887d3a11f; code reference: e47887d3a11f. 中文. Source language: en; requested: en; actual: en; translation: original.

Source documentation does not establish software or physical hardware qualification. See the support and qualification records for their exact scope.

Selected I/O and clock extension delivery

This page records development observations for the selected H743, U575 and G474 software extensions and the checks actually executed for them. The current working source has not reached its final freeze. Focused model results below do not qualify a later commit, an ARM image, an electrical connection or a software candidate.

The result index identifies the exact fixtures, raw command receipts and fresh XML. The I/O architecture contract and its FIG-16 explain static assembly and direct runtime dispatch; the request lifecycle diagram defines borrow and withdrawal transitions. These existing diagrams do not establish qualification of the new selections.

Selected implementation

Slice	Exact selection	Ownership and present boundary
STM32H743	hsi64-pll384-cache-off, reviewed V revision, bare-metal reference	Explicit cold clock plan, nominal CPU 384 MHz/HCLK 192 MHz/APB 96 MHz with authored HSI bounds; peripheral kernel and timer domains remain distinct. This is a selected reference, not DVFS or a measured clock.
STM32U575	USART1 dma-tx, TZEN=0, MSIS 4 MHz, 8N1	One selected state and const ops table; GPDMA1/channel0/request25, finite 1–65535-byte TX wholly inside nonsecure SRAM1 [0x20000000,0x20030000). IRQ byte RX remains available; RX DMA, linked lists and a TrustZone broker remain outside this mode.
STM32G474	can_fd / iso-fd-normal or iso-fd-loopback, FDCAN1	One fixed 212-word message-RAM slice and three physical TX elements; nominal 500 kbit/s and data 2 Mbit/s at HSE24. Normal and internal-loopback model tests do not qualify a transceiver, terminated bus or peer.
STM32G474	pwm_adc / finite-regular-dma	One whole TIM1/ADC1/ADC12/DMA1-channel1/DMAMUX/COMP1 reference owner; 1 MHz counter, 100-tick period, 50-tick trigger and 170 kernel-tick deadtime. Finite halfword samples and the PWM frame remain borrowed through proven withdrawal.

The authored fixtures choose exact SoC facts and Board routes. Applications, control policies, network protocols, worker tasks, private PCB data and recovery policy remain in external repositories. The platform does not infer electrical thresholds or acceptable actuator behavior from a software fixture.

Typed factory lookup returns existing immutable faces without hardware access, allocation or a hidden worker. Optional state and private official headers are selected by the assembly. A source selection alone is not a measured Flash/RAM or latency result; final ARM costs require the matching ELF/BIN/map and checks.

Completion and failure

U575 DMA transfer completion does not imply USART wire completion. The selected mode retains DMAT across requests and cancellation, in accordance with the reviewed U5 erratum. Final settlement requires fresh channel withdrawal/reset proof, independent USART TC and reference drain. Failed drain retains the request in QUARANTINED; a timeout does not revoke its source-buffer loan.

G474 CAN FD separates pending TX add/request bits from fresh per-element sent or cancellation outcomes. Receive queue capacity does not hide the latched protocol state. G474 PWM/ADC joins timer, ADC, DMA and IRQ references and latches late faults before publishing its final request state. Products must continue service and preserve storage whenever withdrawal fails. See the public API and finite PWM/ADC contract for the selected call contexts and physical reference assumptions.

Executed generated-factory scopes

These five programs resolve the maintained TOML and run generated C together with actual providers on the host. GoogleTest/GoogleMock replaces only external CPU, MMIO, NVIC and explicit bus/event leaves. It does not replace provider admission, drain, factory lookup or platform rollback algorithms.

Program	Executed cases	Failures/errors/skips	What the result establishes
nexus_h743_high_factory_test	10	0/0/0	Selected high-profile cold assembly and software rollback behavior
nexus_g474_fdcan_factory_test	5	0/0/0	Selected internal-loopback factory, domain and modeled hardware behavior
nexus_g474_fdcan_normal_factory_test	5	0/0/0	Selected normal-mode factory and modeled controller behavior
nexus_u575_uart_gpdma_factory_test	8	0/0/0	Exact generated channel/vector binding, byte RX, finite TX and lifecycle behavior
nexus_g474_pwm_adc_factory_test	11	0/0/0	Coupled generated ownership and finite-window software behavior

Counts belong to their named scopes and are not a complete-project test count. Later provider changes require a matching rerun; the result index keeps each original receipt. Earlier RED attempts, fixture failures and corrected runs remain preserved rather than overwritten.

Recorded ARM development builds

Six normal preset configurations and builds completed with actual exit 0. The following values come from their resource reports and matching ELF files; Flash LOAD, Flash footprint and reserved RAM are separate byte quantities. Reserved RAM includes linker reservations and is not a measured stack high-water mark. These are recorded development builds, not final-source cost comparisons.

Preset suffix	Flash LOAD (B)	Flash footprint (B)	Reserved RAM (B)	Provider retention
h743-high-empty	5403	5404	2104	Not generated for the empty standard firmware target
h743-high-uart	12016	12016	2336	Passed
h743-high-peripherals	23660	23660	2628	Passed
u575-uart-gpdma	8964	8964	2312	Passed
g474-can-fd	10184	10184	2788	Passed
g474-pwm-adc	13620	13620	2264	Passed

Each suffix belongs to fixture-io-evolution-SUFFIX. The empty H743 result uses nexus_firmware, not a leftover io_evolution_fixture from an earlier failed strong-retention attempt. The other five use io_evolution_fixture and have actual passing provider-retention reports. The CAN FD build here is the loopback selection; no normal-bus ARM build is claimed by this table. No cycles, latency, electrical accuracy or dynamic high-water value was measured.

Build and dependency delivery

Private official inputs add exactly two U5 LL headers (dma, usart) and three G4 FDCAN-definition headers, including their transitive definitions. Pins, licenses and CMSIS pairing remain checked. The selected closure does not link HAL handles, a vendor tick, vendor IRQ callbacks or a second transfer owner.

The source SDK now requires the new public/private I/O modules, coupled configuration code and explicit family payload. verify_consumers --parallel N accepts a strict positive job count for every consumer build. Omitting it preserves the previous 4-job base/2-job modern defaults. Its Python regression checks all 13 consumer roots and 16 image targets with external subprocess test doubles; it does not constitute execution of a new real SDK consumer package.

Source and remaining qualification

Historical GitHub run 39028022087 reports 30 completed jobs with scalar conclusion="success" for clean source 2963c34aa399c43bea05022eb8846ad2370f0f76. This is the previous committed source. It does not validate the current working extensions. Job-level documentation success alone also does not establish a newly inspected whole documentation package or prove that every documentation step ran.

Before final delivery, bind the frozen source to complete normal hooks, focused final model reruns, exact ARM resource/retention checks, fresh source SDK and independent consumers, and current API/English/Chinese documentation checks. Report each actual result separately. Offline reproduction and a software candidate require their own executed evidence. Physical HIL remains not_executed; IO-01–42 ledger status is maintained separately and is not advanced by this page.