Baremetal: Add Cortex-M33 MPS3-AN524 platform support

.github/workflows/baremetal.yml

@@ -25,6 +25,16 @@ jobs:
            alloc: true
            bench: true
            opt: no_opt
+         - runner: ubuntu-latest
+           name: 'M33-AN524'
+           makefile: test/baremetal/platform/m33-an524/platform.mk
+           nix-shell: cross-arm-embedded
+           func: true
+           kat: true
+           acvp: true
+           alloc: true
+           bench: true
+           opt: no_opt
     runs-on: ${{ matrix.target.runner }}
     steps:
       - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2

flake.nix

@@ -152,7 +152,7 @@
           devShells.cross-arm-embedded = util.mkShell {
             packages = builtins.attrValues
               {
-                inherit (util) m55-an547;
+                inherit (util) pqmx;
                 inherit (config.packages) linters;
                 inherit (pkgs) gcc-arm-embedded qemu coreutils python3 git;
               };

nix/m55-an547-arm-none-eabi/default.nix → nix/pqmx/default.nix

@@ -7,32 +7,36 @@
 }:
 
 stdenvNoCC.mkDerivation {
-  pname = "mldsa-native-m55-an547";
-  version = "main-2025-10-02";
+  pname = "mldsa-native-pqmx";
+  version = "main-2026-02-10";
 
-
-  # Fetch platform files from pqmx (envs/m55-an547)
+  # Fetch platform files from pqmx
   src = fetchFromGitHub {
     owner = "slothy-optimizer";
     repo = "pqmx";
-    rev = "4ed493d3cf2af62a08fd9fe36c3472a0dc50ad9f";
-    hash = "sha256-jLIqwknjRwcoDeEAETlMhRqZQ5a3QGCDZX9DENelGeQ=";
+    rev = "904451a615dc7926eba07b4f3d1a4137c368bb4a";
+    hash = "sha256-BjsToEWGlykKIKRfPom84BkD5RfetUKKwRAw3PecebU=";
   };
 
   dontBuild = true;
 
   installPhase = ''
+    mkdir -p $out/platform/m33-an524/src/platform/
+    cp -r envs/m33-an524/src/platform/. $out/platform/m33-an524/src/platform/
+    cp integration/*.c $out/platform/m33-an524/src/platform/
+
     mkdir -p $out/platform/m55-an547/src/platform/
     cp -r envs/m55-an547/src/platform/. $out/platform/m55-an547/src/platform/
     cp integration/*.c $out/platform/m55-an547/src/platform/
   '';
 
   setupHook = writeText "setup-hook.sh" ''
+    export M33_AN524_PATH="$1/platform/m33-an524/src/platform/"
     export M55_AN547_PATH="$1/platform/m55-an547/src/platform/"
   '';
 
   meta = {
-    description = "Platform files for the Cortex-M55 (AN547)";
+    description = "Platform files from pqmx for baremetal targets";
     homepage = "https://github.com/slothy-optimizer/pqmx";
   };
 }

nix/util.nix

@@ -103,7 +103,7 @@ rec {
   hol_server = pkgs.callPackage ./hol_light/hol_server.nix { inherit hol_light'; };
   s2n_bignum = pkgs.callPackage ./s2n_bignum { };
   slothy = pkgs.callPackage ./slothy { python3 = python3-for-slothy; };
-  m55-an547 = pkgs.callPackage ./m55-an547-arm-none-eabi { };
+  pqmx = pkgs.callPackage ./pqmx { };
 
   # Helper function to build individual cross toolchains
   _individual_toolchain = { name, cross_compilers }:

test/baremetal/platform/m33-an524/exec_wrapper.py

@@ -0,0 +1,54 @@
+#!/usr/bin/env python3
+# Copyright (c) The mldsa-native project authors
+# SPDX-License-Identifier: Apache-2.0 OR ISC OR MIT
+
+"""QEMU wrapper for executing Cortex-M33 bare-metal ELF binaries (mps3-an524)."""
+
+import struct as st
+import sys
+import subprocess
+import tempfile
+import os
+
+
+def err(msg, **kwargs):
+    print(msg, file=sys.stderr, **kwargs)
+
+
+binpath = sys.argv[1]
+args = sys.argv[1:]
+
+# Memory layout: [argc] [offset1] [offset2] ... [string1\0] [string2\0] ...
+# M33-AN524 RAM: 0x20000000-0x2001FFFF (128KB)
+# Heap ends at: ~0x20000b20
+# Stack: 0x20008000-0x2001FFFF (96KB, grows downward)
+# Use address after heap but before stack
+# cmdline.c CMDLINE_ADDR must match this value
+cmdline_offset = 0x20007000
+arg0_offset = cmdline_offset + 4 + len(args) * 4
+arg_offsets = [sum(map(len, args[:i])) + i + arg0_offset for i in range(len(args))]
+
+binargs = st.pack(
+    f"<{1+len(args)}I" + "".join(f"{len(a)+1}s" for a in args),
+    len(args),
+    *arg_offsets,
+    *map(lambda x: x.encode("utf-8"), args),
+)
+
+with tempfile.NamedTemporaryFile(mode="wb", delete=False, suffix=".bin") as fd:
+    args_file = fd.name
+    fd.write(binargs)
+
+try:
+    qemu_cmd = f"qemu-system-arm -M mps3-an524 -cpu cortex-m33 -nographic -semihosting -kernel {binpath} -device loader,file={args_file},addr=0x{cmdline_offset:x}".split()
+    result = subprocess.run(qemu_cmd, encoding="utf-8", capture_output=True)
+finally:
+    os.unlink(args_file)
+if result.returncode != 0:
+    err("FAIL!")
+    err(f"{qemu_cmd} failed with error code {result.returncode}")
+    err(result.stderr)
+    exit(1)
+
+for line in result.stdout.splitlines():
+    print(line)

test/baremetal/platform/m33-an524/platform.mk

@@ -0,0 +1,76 @@
+# Copyright (c) The mldsa-native project authors
+# Copyright (c) The mlkem-native project authors
+# SPDX-License-Identifier: Apache-2.0 OR ISC OR MIT
+
+PLATFORM_PATH:=test/baremetal/platform/m33-an524
+
+CROSS_PREFIX=arm-none-eabi-
+CC=gcc
+
+# Use PMU cycle counting by default
+CYCLES ?= PMU
+
+# Reduce iterations for benchmarking
+CFLAGS += -DMLD_BENCHMARK_NTESTS=3 -DMLD_BENCHMARK_NITERATIONS=2 -DMLD_BENCHMARK_NWARMUP=3
+
+CFLAGS += -DMLD_CONFIG_REDUCE_RAM -DMLD_BUMP_ALLOC_SIZE=65536
+
+CFLAGS += \
+	-O3 \
+	-Wall -Wextra -Wshadow \
+	-Wno-pedantic \
+	-Wno-redundant-decls \
+	-Wno-missing-prototypes \
+	-Wno-conversion \
+	-Wno-sign-conversion \
+	-fno-common \
+	-ffunction-sections \
+	-fdata-sections \
+	--sysroot=$(SYSROOT) \
+	-DDEVICE=an524 \
+	-I$(M33_AN524_PATH) \
+	-I$(M33_AN524_PATH)/m-profile \
+	-DARMCM33 \
+	-DSEMIHOSTING \
+	-DCMDLINE_BASE_ADDR=0x20007000
+
+ARCH_FLAGS += \
+	-mcpu=cortex-m33 \
+	-mthumb \
+	-mfloat-abi=soft
+
+CFLAGS += \
+	$(ARCH_FLAGS) \
+	--specs=nosys.specs
+
+CFLAGS += $(CFLAGS_EXTRA)
+
+LDSCRIPT = $(M33_AN524_PATH)/m33-an524.ld
+
+LDFLAGS += \
+	-Wl,--gc-sections \
+	-Wl,--no-warn-rwx-segments \
+	-L.
+
+LDFLAGS += \
+	--specs=nosys.specs \
+	-Wl,--wrap=_open \
+	-Wl,--wrap=_close \
+	-Wl,--wrap=_read \
+	-Wl,--wrap=_write \
+	-Wl,--wrap=_fstat \
+	-Wl,--wrap=_getpid \
+	-Wl,--wrap=_isatty \
+	-Wl,--wrap=_kill \
+	-Wl,--wrap=_lseek \
+	-Wl,--wrap=main \
+	-ffreestanding \
+	-T$(LDSCRIPT) \
+	$(ARCH_FLAGS)
+
+# Extra sources to be included in test binaries
+EXTRA_SOURCES = $(wildcard $(M33_AN524_PATH)/*.c)
+# The CMSIS files fail compilation if conversion warnings are enabled
+EXTRA_SOURCES_CFLAGS = -Wno-conversion -Wno-sign-conversion
+
+EXEC_WRAPPER := $(realpath $(PLATFORM_PATH)/exec_wrapper.py)

test/hal/hal.c

@@ -96,6 +96,25 @@ uint64_t get_cyclecounter(void)
 }
 
 #elif defined(__ARM_ARCH_8M_MAIN__) || defined(__ARM_ARCH_8_1M_MAIN__)
+/* ARMv8-M cycle counting support */
+#if defined(ARMCM33)
+/* Cortex-M33: Use DWT cycle counter (no dedicated PMU) */
+#include <ARMCM33.h>
+#include <system_ARMCM33.h>
+
+void enable_cyclecounter(void)
+{
+  CoreDebug->DEMCR |= CoreDebug_DEMCR_TRCENA_Msk;
+  DWT->CYCCNT = 0;
+  DWT->CTRL |= DWT_CTRL_CYCCNTENA_Msk;
+}
+
+void disable_cyclecounter(void) { DWT->CTRL &= ~DWT_CTRL_CYCCNTENA_Msk; }
+
+uint64_t get_cyclecounter(void) { return DWT->CYCCNT; }
+
+#elif defined(ARMCM55)
+/* Cortex-M55: Use dedicated PMU */
 #include <ARMCM55.h>
 #include <system_ARMCM55.h>
 #include "pmu_armv8.h"
@@ -116,6 +135,10 @@ void disable_cyclecounter(void)
 
 uint64_t get_cyclecounter(void) { return ARM_PMU_Get_CCNTR(); }
 
+#else /* !ARMCM33 && ARMCM55 */
+#error "PMU_CYCLES on ARMv8-M requires ARMCM33 or ARMCM55 to be defined"
+#endif /* !ARMCM33 && !ARMCM55 */
+
 #elif defined(__riscv)
 
 void enable_cyclecounter(void) {}

test/src/test_alloc.c

@@ -70,7 +70,9 @@ typedef struct
 } alloc_info_t;
 
 #define MLD_MAX_IN_FLIGHT_ALLOCS 100
+#ifndef MLD_BUMP_ALLOC_SIZE
 #define MLD_BUMP_ALLOC_SIZE (128 * 1024) /* 128KB buffer */
+#endif
 
 struct test_ctx_t
 {