Add compressed firmware images: gzip-compressed A/B slots executed from RAM

.github/workflows/test-compressed-simulator.yml

@@ -0,0 +1,165 @@
+name: Compressed images - test with simulator target
+
+on:
+  push:
+    branches: [ 'master', 'main', 'release/**' ]
+  pull_request:
+    branches: [ '*' ]
+
+jobs:
+  compressed_simulator_tests:
+    runs-on: ubuntu-latest
+    container:
+      image: ghcr.io/wolfssl/wolfboot-ci-sim:v1.0
+    timeout-minutes: 20
+    strategy:
+      fail-fast: false
+      matrix:
+        # GZIP_SMALL selects a different Huffman decoder in src/gzip.c, so
+        # both have to be exercised and not merely built.
+        sign: [ ED25519, ECC256 ]
+        gzip_small: [ 0, 1 ]
+
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          submodules: true
+
+      - name: Trust workspace
+        run: git config --global --add safe.directory "$GITHUB_WORKSPACE"
+
+      - name: make clean
+        run: |
+          make distclean
+
+      - name: Select config
+        run: |
+          cp config/examples/sim-compressed-ram.config .config
+          sed -i "s/^SIGN?=.*/SIGN?=${{ matrix.sign }}/" .config
+          sed -i "s/^GZIP_SMALL=.*//" .config
+          echo "GZIP_SMALL=${{ matrix.gzip_small }}" >> .config
+
+      - name: Build tools
+        run: |
+          make -C tools/keytools && make -C tools/bin-assemble
+
+      - name: Build wolfboot.elf and test-app/image.elf
+        run: |
+          make clean && make
+
+      - name: Assemble the compressed A/B slots
+        run: |
+          make test-sim-compressed-ram-with-update
+
+      - name: Run the compressed image simulator test
+        run: |
+          ./tools/scripts/sim-compressed-ram.sh
+
+  compressed_encrypted_simulator_tests:
+    runs-on: ubuntu-latest
+    container:
+      image: ghcr.io/wolfssl/wolfboot-ci-sim:v1.0
+    timeout-minutes: 20
+    strategy:
+      fail-fast: false
+      matrix:
+        # Each cipher has its own block size, which the per-slot keystream
+        # stride is derived from, so all three have to run.
+        cipher: [ chacha, aes128, aes256 ]
+
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          submodules: true
+
+      - name: Trust workspace
+        run: git config --global --add safe.directory "$GITHUB_WORKSPACE"
+
+      - name: make clean
+        run: |
+          make distclean
+
+      - name: Select config
+        run: |
+          cp config/examples/sim-compressed-ram-enc.config .config
+          case "${{ matrix.cipher }}" in
+            aes128) echo "ENCRYPT_WITH_AES128=1" >> .config ;;
+            aes256) echo "ENCRYPT_WITH_AES256=1" >> .config ;;
+          esac
+
+      - name: Build tools
+        run: |
+          make -C tools/keytools && make -C tools/bin-assemble
+
+      - name: Build wolfboot.elf and test-app/image.elf
+        run: |
+          make clean && make
+
+      - name: Assemble the encrypted compressed A/B slots
+        run: |
+          make test-sim-compressed-ram-enc-with-update
+
+      - name: Run the encrypted compressed image simulator test
+        run: |
+          ./tools/scripts/sim-compressed-ram-enc.sh
+
+  compressed_cross_build:
+    runs-on: ubuntu-latest
+    container:
+      image: ghcr.io/wolfssl/wolfboot-ci-arm:v1.0
+    timeout-minutes: 20
+    strategy:
+      fail-fast: false
+      matrix:
+        config: [ stm32h7-compressed-ram, stm32f407-compressed-ram ]
+
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          submodules: true
+
+      - name: Trust workspace
+        run: git config --global --add safe.directory "$GITHUB_WORKSPACE"
+
+      - name: Select config
+        run: |
+          cp config/examples/${{ matrix.config }}.config .config
+
+      - name: Build
+        run: |
+          make clean && make keysclean && make
+
+      # A link that succeeds but places the vector table somewhere other
+      # than the load address boots into garbage, and nothing catches that
+      # without hardware. Assert the layout from the ELF instead.
+      - name: Check the application is linked for the load address
+        run: |
+          LOAD=$(grep -E '^WOLFBOOT_LOAD_ADDRESS\??=' .config | tail -1 | cut -d= -f2)
+          SIZE=$(grep -E '^WOLFBOOT_RAMBOOT_MAX_SIZE\??=' .config | tail -1 | cut -d= -f2)
+          echo "load address $LOAD, region size $SIZE"
+          # .text must start exactly at the load address, with LMA == VMA
+          arm-none-eabi-objdump -h test-app/image.elf | \
+            awk -v want=$(printf "%08x" $((LOAD))) '
+              $2==".text" { if ($4!=want || $5!=want) {
+                  printf "`.text` at vma %s lma %s, expected %s\n",$4,$5,want; exit 1 }
+                print "ok: .text at " $4 " (lma " $5 ")" }'
+          # word 0 of the image is the initial stack pointer and must be the
+          # top of the load region
+          SP=$(od -An -t x4 -N 4 test-app/image.bin | tr -d ' ')
+          WANT_SP=$(printf "%08x" $((LOAD + SIZE)))
+          if [ "$SP" != "$WANT_SP" ]; then
+            echo "initial SP is 0x$SP, expected 0x$WANT_SP"; exit 1
+          fi
+          echo "ok: initial SP 0x$SP"
+
+      - name: Check wolfBoot does not overlap the load region
+        run: |
+          LOAD=$(grep -E '^WOLFBOOT_LOAD_ADDRESS\??=' .config | tail -1 | cut -d= -f2)
+          END=$(arm-none-eabi-nm wolfboot.elf | awk '$3=="_end" {print $1}')
+          # Only meaningful when wolfBoot's RAM and the load region share an
+          # address space (STM32F4); on the H7 they are different SRAMs.
+          if [ $((0x$END)) -ge $((LOAD)) ] && \
+             [ $((0x$END)) -lt $((LOAD + 0x100000)) ]; then
+            echo "wolfBoot _end 0x$END is inside the load region at $LOAD"; exit 1
+          fi
+          echo "ok: wolfBoot _end 0x$END clear of the load region at $LOAD"

.gitignore

@@ -453,6 +453,7 @@ bootmode1/
 
 # Emulator test: generated linker scripts
 test-app/emu-test-apps/*/target.ld
+test-app/*.gz
 
 # AMD/Xilinx Versal test artifacts
 uart_log*.txt

arch.mk

@@ -3177,9 +3177,22 @@ else
   endif
 endif
 else
-  ifeq ($(DUALBANK_SWAP),1)
+  ifeq ($(COMPRESSED_RAM_SLOTS),1)
+    # Compressed A/B slots, execute from RAM. Both slots hold a signed,
+    # gzip-compressed image; the RAM loader picks the highest valid version,
+    # verifies the signature over the compressed bytes and inflates to
+    # WOLFBOOT_LOAD_ADDRESS. There is no swap or install step, so the slot
+    # roles are decided per boot by version, as for the hardware bank swap.
+    ifeq ($(DUALBANK_SWAP),1)
+      $(error COMPRESSED_RAM_SLOTS=1 is incompatible with DUALBANK_SWAP=1)
+    endif
     CFLAGS+=-DWOLFBOOT_DUALBOOT
-    UPDATE_OBJS:=src/update_flash_hwswap.o
+    UPDATE_OBJS:=src/update_ram.o
+  else
+    ifeq ($(DUALBANK_SWAP),1)
+      CFLAGS+=-DWOLFBOOT_DUALBOOT
+      UPDATE_OBJS:=src/update_flash_hwswap.o
+    endif
   endif
 endif
 

config/examples/sim-compressed-ram-enc.config

@@ -0,0 +1,31 @@
+ARCH=sim
+TARGET=sim
+SIGN?=ED25519
+HASH?=SHA256
+WOLFBOOT_SMALL_STACK?=0
+SPI_FLASH=0
+DEBUG=1
+
+# Compressed A/B slots, executed from RAM, encrypted at rest.
+COMPRESSED_RAM_SLOTS=1
+COMPRESSED_ENCRYPT=1
+GZIP=1
+
+# sizes should be multiple of system page size
+WOLFBOOT_PARTITION_SIZE=0x40000
+WOLFBOOT_SECTOR_SIZE=0x1000
+WOLFBOOT_PARTITION_BOOT_ADDRESS=0x80000
+WOLFBOOT_PARTITION_UPDATE_ADDRESS=0x100000
+WOLFBOOT_PARTITION_SWAP_ADDRESS=0x180000
+
+# See sim-compressed-ram.config for why the load region sits above wolfboot.elf.
+WOLFBOOT_LOAD_ADDRESS=0x1000000
+WOLFBOOT_RAMBOOT_MAX_SIZE=0x40000
+
+# An encrypted slot cannot be verified in place, so it is decrypted into a
+# staging region first. Must be disjoint from the load region, or the
+# decompressor would overwrite its own input.
+WOLFBOOT_COMPRESSED_STAGE_ADDRESS=0x1100000
+
+# required for keytools
+WOLFBOOT_FIXED_PARTITIONS=1

config/examples/sim-compressed-ram.config

@@ -0,0 +1,34 @@
+ARCH=sim
+TARGET=sim
+SIGN?=ED25519
+HASH?=SHA256
+WOLFBOOT_SMALL_STACK?=0
+SPI_FLASH=0
+DEBUG=1
+
+# Compressed A/B slots, executed from RAM. Both slots hold a signed gzip
+# stream; wolfBoot verifies the compressed bytes in place, inflates to
+# WOLFBOOT_LOAD_ADDRESS and boots there.
+COMPRESSED_RAM_SLOTS=1
+GZIP=1
+
+# sizes should be multiple of system page size
+WOLFBOOT_PARTITION_SIZE=0x40000
+WOLFBOOT_SECTOR_SIZE=0x1000
+WOLFBOOT_PARTITION_BOOT_ADDRESS=0x80000
+# if on external flash, it should be multiple of system page size
+WOLFBOOT_PARTITION_UPDATE_ADDRESS=0x100000
+WOLFBOOT_PARTITION_SWAP_ADDRESS=0x180000
+
+# The simulator's do_boot() writes a whole partition from the load address,
+# so the load region must be at least WOLFBOOT_PARTITION_SIZE.
+#
+# WOLFBOOT_ORIGIN is 0 here (simulator flash lives at the runtime
+# sim_ram_base), so the loader's bootloader-overlap guard degenerates to
+# "below _end is suspect". wolfboot.elf is a non-PIE host binary loaded at
+# 0x400000, so keep the load region above it.
+WOLFBOOT_LOAD_ADDRESS=0x1000000
+WOLFBOOT_RAMBOOT_MAX_SIZE=0x40000
+
+# required for keytools
+WOLFBOOT_FIXED_PARTITIONS=1

config/examples/stm32f407-compressed-ram.config

@@ -0,0 +1,44 @@
+# STM32F407-DISCOVERY with compressed A/B images executed from RAM.
+#
+# The F407 has one contiguous 128KB SRAM shared between wolfBoot and the
+# inflated application, so hal/stm32f4_ramboot.ld caps wolfBoot at the
+# bottom 16KB and the application occupies the remaining 112KB. That is the
+# ceiling on this part: the CCM RAM at 0x10000000 cannot be used, because
+# on the F4 it is reachable only by the data bus and instruction fetch from
+# it faults.
+#
+# GZIP_SMALL=1 selects the bit-serial Huffman decoder: a few KB less code
+# and no static tables, which matters more than throughput on a 1MB part.
+#
+# Note that hal/stm32f4.c has no debug UART, so DEBUG=1 only buys symbols
+# here: there is no console for the "Inflating ... done" progress line.
+# Inspect the load region over SWD instead, or add a UART to the HAL.
+ARCH?=ARM
+TARGET?=stm32f4
+SIGN?=ED25519
+HASH?=SHA256
+DEBUG?=1
+VTOR?=1
+SPMATH?=1
+
+COMPRESSED_RAM_SLOTS?=1
+GZIP?=1
+GZIP_SMALL?=1
+
+WOLFBOOT_PARTITION_SIZE?=0x20000
+WOLFBOOT_SECTOR_SIZE?=0x20000
+WOLFBOOT_PARTITION_BOOT_ADDRESS?=0x08020000
+WOLFBOOT_PARTITION_UPDATE_ADDRESS?=0x08040000
+# Unused in this mode (there is no swap or install step), but still
+# required by the keytools and the partition trailer layout.
+WOLFBOOT_PARTITION_SWAP_ADDRESS?=0x08060000
+
+# SRAM above wolfBoot's 16KB region. 0x400-aligned for the SCB->VTOR write.
+WOLFBOOT_LOAD_ADDRESS?=0x20004000
+WOLFBOOT_RAMBOOT_MAX_SIZE?=0x1C000
+
+CLOCK_SPEED?=160000000
+STM32_PLLM?=8
+STM32_PLLN?=336
+STM32_PLLP?=2
+STM32_PLLQ?=7

config/examples/stm32f439zi-compressed-ram.config

@@ -0,0 +1,66 @@
+# NUCLEO-F439ZI with compressed A/B images executed from RAM.
+#
+# The F439ZI has 2MB of flash and 256KB of SRAM, of which 192KB is
+# contiguous at 0x20000000 (SRAM1 112KB + SRAM2 16KB + SRAM3 64KB).
+# hal/stm32f4_ramboot.ld caps wolfBoot at the bottom 16KB, leaving 176KB
+# for the inflated application. The 64KB CCM at 0x10000000 cannot be used:
+# on the F4 it is reachable only by the data bus, so an instruction fetch
+# from it faults.
+#
+# Console is USART3 on PD8/PD9, the ST-LINK virtual COM port on the
+# NUCLEO-144 boards, via hal/uart/uart_drv_stm32f4.c.
+ARCH?=ARM
+TARGET?=stm32f4
+SIGN?=ED25519
+HASH?=SHA256
+DEBUG?=1
+DEBUG_UART?=1
+CFLAGS_EXTRA+=-DUSE_UART3
+VTOR?=1
+SPMATH?=1
+
+COMPRESSED_RAM_SLOTS?=1
+GZIP?=1
+# Required at a 16KB bootloader RAM region: the table-driven decoder needs
+# about 19KB of .bss for its Huffman tables and fails to link.
+GZIP_SMALL?=1
+
+WOLFBOOT_PARTITION_SIZE?=0x20000
+WOLFBOOT_SECTOR_SIZE?=0x20000
+WOLFBOOT_PARTITION_BOOT_ADDRESS?=0x08020000
+WOLFBOOT_PARTITION_UPDATE_ADDRESS?=0x08040000
+# Unused in this mode (there is no swap or install step), but still
+# required by the keytools and the partition trailer layout.
+WOLFBOOT_PARTITION_SWAP_ADDRESS?=0x08060000
+
+# SRAM above wolfBoot's 16KB region. 0x400-aligned for the SCB->VTOR write.
+# This uses all the remaining SRAM, which is the right choice when the slots
+# are not encrypted; the encrypted variant below has to give some back.
+WOLFBOOT_LOAD_ADDRESS?=0x20004000
+WOLFBOOT_RAMBOOT_MAX_SIZE?=0x2C000
+
+# Encryption of both slots at rest. Commented out because it needs a key
+# provisioned on the device first: hal_get_image_encrypt_key() reads
+# WOLFBOOT_COMPRESSED_KEY_ADDRESS, expecting the key followed by the nonce,
+# and rejects an erased or all-zero region. Override that hook in the HAL to
+# take the key from OTP or a secure element rather than flash.
+#
+# An encrypted slot cannot be verified in place, so it is decrypted into a
+# staging region first, and that region must be disjoint from the load
+# region. On this part the 176KB above wolfBoot therefore has to be split:
+# 112KB of load region and a 64KB staging region at the top of SRAM, which
+# bounds the compressed image to 64KB.
+#
+# Sign slot B with --encrypt-iv-offset $((WOLFBOOT_PARTITION_SIZE / 64)) for
+# ChaCha20, or / 16 for AES-CTR, so the two slots never share a keystream.
+#COMPRESSED_ENCRYPT?=1
+#WOLFBOOT_RAMBOOT_MAX_SIZE?=0x1C000
+#WOLFBOOT_COMPRESSED_STAGE_ADDRESS?=0x20020000
+#WOLFBOOT_COMPRESSED_KEY_ADDRESS?=0x080E0000
+
+# HSE 8MHz from the ST-LINK MCO -> 168MHz SYSCLK, PCLK1 42MHz
+CLOCK_SPEED?=168000000
+STM32_PLLM?=8
+STM32_PLLN?=336
+STM32_PLLP?=2
+STM32_PLLQ?=7