Nommu linux

.github/workflows/ci.yml

@@ -342,3 +342,202 @@ jobs:
         run: |
           docker run --rm -v ${{ github.workspace }}:/workspace frost-dev:latest \
             pre-commit run --all-files
+
+  # ---------------------------------------------------------------------------
+  # FROST no-MMU M-mode Linux: build the image from source, then boot it two
+  # ways off the SAME artifact -- on the FROST RTL in cocotb (the gremlin
+  # regression) and under QEMU (a fast full boot-to-shell reference).
+  # ---------------------------------------------------------------------------
+
+  # Build the kernel + busybox initramfs + FROST memory images from the vendored
+  # Buildroot submodule (linux/buildroot @ pinned SHA) driven by the FROST
+  # BR2_EXTERNAL tree. Runs inside the frost-dev image, which ships Buildroot's
+  # host deps + QEMU, so the host deps are single-sourced in the Dockerfile. The
+  # first build compiles a full rv32 uClibc cross toolchain from source; the dl/
+  # and ccache caches make later runs much faster.
+  build-frost-linux:
+    name: Build FROST Linux Image (Buildroot)
+    runs-on: ubuntu-24.04
+    needs: build-docker
+    timeout-minutes: 120
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+        with:
+          # Recursive so the pinned linux/buildroot submodule is fetched.
+          submodules: recursive
+
+      - name: Download Docker image
+        uses: actions/download-artifact@v4
+        with:
+          name: frost-docker-image
+          path: /tmp
+
+      - name: Load Docker image
+        run: docker load --input /tmp/frost-dev.tar
+
+      - name: Cache Buildroot downloads (source tarballs)
+        uses: actions/cache@v4
+        with:
+          path: linux/dl
+          # dl/ only changes when a package version (kernel/toolchain) changes.
+          key: br2-dl-${{ hashFiles('linux/buildroot-external/configs/frost_nommu_rv32_defconfig', 'linux/buildroot-external/board/frost/**') }}
+          restore-keys: |
+            br2-dl-
+
+      - name: Cache Buildroot ccache (toolchain + kernel object cache)
+        uses: actions/cache@v4
+        with:
+          path: linux/ccache
+          key: br2-ccache-${{ github.sha }}
+          restore-keys: |
+            br2-ccache-
+
+      - name: Build kernel + initramfs + FROST memory images
+        run: |
+          docker run --rm \
+            -e BR2_DL_DIR=/workspace/linux/dl \
+            -e BR2_CCACHE_DIR=/workspace/linux/ccache \
+            -v ${{ github.workspace }}:/workspace frost-dev:latest \
+            bash -c '
+              set -euo pipefail
+              make -C linux/buildroot O=/workspace/linux/build \
+                BR2_EXTERNAL=/workspace/linux/buildroot-external \
+                frost_nommu_rv32_defconfig
+              # Enable ccache for CI only (kept out of the committed defconfig).
+              echo "BR2_CCACHE=y" >> /workspace/linux/build/.config
+              make -C linux/buildroot O=/workspace/linux/build olddefconfig
+              make -C linux/buildroot O=/workspace/linux/build
+              # Stage the memory images where the cocotb linux_boot test resolves
+              # them (sw/apps/linux_boot/{sw,sw_ddr}.mem).
+              mkdir -p sw/apps/linux_boot
+              cp linux/build/images/sw.mem     sw/apps/linux_boot/sw.mem
+              cp linux/build/images/sw_ddr.mem sw/apps/linux_boot/sw_ddr.mem
+              ls -l linux/build/images/
+            '
+
+      - name: Upload FROST Linux boot images
+        uses: actions/upload-artifact@v4
+        with:
+          name: frost-linux-boot-images
+          path: |
+            linux/build/images/Image
+            linux/build/images/rootfs.cpio.gz
+            linux/build/images/frost-nommu-fpga.dtb
+            linux/build/images/sw.mem
+            linux/build/images/sw_ddr.mem
+          retention-days: 7
+          if-no-files-found: error
+
+  # Boot the freshly built image on the FROST RTL in cocotb, bounded to ~22M
+  # cycles in the genesys2-faithful cache shape (128 KiB L1I, no L2). This is the
+  # "gremlin" regression: that window is silent mem_init after devtmpfs (no deep
+  # console marker), so the run captures the full boot (FROST_LINUX_RUN_FULL) and
+  # check_linux_boot_regression.py then asserts boot health -- banner + devtmpfs,
+  # no panic, the periodic CLINT timer tick serviced (mtimecmp re-armed, the
+  # thing the gremlin hung), and forward progress (retire) all the way to the
+  # cap. CACHED_HAS_L2=0 must be an env/make var (the test's own -GCACHED_HAS_L2=0
+  # is otherwise overridden by the tests/Makefile default).
+  linux-boot-cocotb:
+    name: Cocotb Linux Boot (22M, genesys2 shape)
+    runs-on: ubuntu-24.04
+    needs: [build-docker, build-frost-linux]
+    timeout-minutes: 180
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+        with:
+          submodules: recursive
+
+      - name: Download Docker image
+        uses: actions/download-artifact@v4
+        with:
+          name: frost-docker-image
+          path: /tmp
+
+      - name: Load Docker image
+        run: docker load --input /tmp/frost-dev.tar
+
+      - name: Download FROST Linux boot images
+        uses: actions/download-artifact@v4
+        with:
+          name: frost-linux-boot-images
+          path: /tmp/frost-linux-images
+
+      - name: Stage boot images for the cocotb linux_boot test
+        run: |
+          mkdir -p sw/apps/linux_boot
+          cp /tmp/frost-linux-images/sw.mem     sw/apps/linux_boot/sw.mem
+          cp /tmp/frost-linux-images/sw_ddr.mem sw/apps/linux_boot/sw_ddr.mem
+
+      - name: Run cocotb linux_boot (22M-cycle gremlin regression)
+        run: |
+          docker run --rm \
+            -e CACHED_HAS_L2=0 \
+            -e FROST_LINUX_RUN_FULL=1 \
+            -e COCOTB_NUM_RUNS=1 \
+            -e COCOTB_LINUX_MAX_CYCLES=22000000 \
+            -e COCOTB_PROGRESS_INTERVAL=500000 \
+            -v ${{ github.workspace }}:/workspace frost-dev:latest \
+            bash -c '
+              set -o pipefail
+              cd tests && make clean
+              # RUN_FULL capture always trips the cocotb never-match assertion by
+              # design; the boot-health verdict comes from the checker below.
+              ./test_run_cocotb.py linux_boot_128k 2>&1 \
+                | tee /workspace/linux_boot_cocotb.log || true
+              python3 check_linux_boot_regression.py /workspace/linux_boot_cocotb.log
+            '
+
+      - name: Upload cocotb boot log
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: linux-boot-cocotb-log
+          path: linux_boot_cocotb.log
+          if-no-files-found: ignore
+
+  # Boot the SAME kernel + rootfs to a login prompt under QEMU: a fast (seconds)
+  # full-userspace reference that the RTL boot is bounded well short of. Confirms
+  # the built image itself reaches a shell, independent of the FROST core.
+  linux-boot-qemu:
+    name: QEMU Linux Boot to Shell
+    runs-on: ubuntu-24.04
+    needs: [build-docker, build-frost-linux]
+    timeout-minutes: 20
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Download Docker image
+        uses: actions/download-artifact@v4
+        with:
+          name: frost-docker-image
+          path: /tmp
+
+      - name: Load Docker image
+        run: docker load --input /tmp/frost-dev.tar
+
+      - name: Download FROST Linux boot images
+        uses: actions/download-artifact@v4
+        with:
+          name: frost-linux-boot-images
+          path: /tmp/frost-linux-images
+
+      - name: Boot to a shell under QEMU (assert login prompt)
+        run: |
+          docker run --rm -v /tmp/frost-linux-images:/img frost-dev:latest \
+            bash -c '
+              set -euo pipefail
+              # QEMU sits at the login prompt forever, so time-box it and assert
+              # the marker from the captured log (file redirect, never a pipe --
+              # piping QEMUs stdout to grep deadlocks).
+              timeout -k5 120 qemu-system-riscv32 -M virt -bios none \
+                -kernel /img/Image \
+                -append "earlycon=sbi console=ttyS0 rdinit=/sbin/init" \
+                -initrd /img/rootfs.cpio.gz -nographic -cpu rv32,mmu=off \
+                </dev/null >/tmp/qemu-boot.log 2>&1 || true
+              echo "===== QEMU boot log (tail) ====="
+              tail -n 20 /tmp/qemu-boot.log
+              grep -q "buildroot login:" /tmp/qemu-boot.log
+            '

.gitignore

@@ -65,6 +65,13 @@ sw_ddr.bin
 sw_imem_*.mem
 sw/apps/*/sw.S
 
+# FROST no-MMU Linux: out-of-tree Buildroot build output + download/ccache
+# caches. Regenerated from the linux/buildroot submodule + linux/buildroot-external
+# tree by `make -C sw/apps/linux_boot` (and the CI build-frost-linux job).
+/linux/build/
+/linux/dl/
+/linux/ccache/
+
 # mypy
 .mypy_cache/
 

.gitmodules

@@ -15,3 +15,6 @@
 [submodule "sw/apps/coremark_pro/coremark-pro"]
 	path = sw/apps/coremark_pro/coremark-pro
 	url = https://github.com/eembc/coremark-pro.git
+[submodule "linux/buildroot"]
+	path = linux/buildroot
+	url = https://github.com/buildroot/buildroot.git

CONTRIBUTING.md

@@ -20,7 +20,7 @@ This document provides guidelines for contributors. The detailed style sections
 
 ## Project Overview
 
-FROST is an out-of-order RISC-V processor implementing **RV32GCB** (G = IMAFD) with a Tomasulo back-end and full machine-mode privilege support. Understanding the architecture helps you contribute effectively:
+FROST is an out-of-order RISC-V processor implementing **RV32GCB** (G = IMAFD) with a Tomasulo back-end and Machine + User (M/U) privilege modes. Understanding the architecture helps you contribute effectively:
 
 ### Architecture Outline
 
@@ -577,7 +577,7 @@ We welcome contributions in these areas:
 |------|----------|
 | Bug fixes | OOO ordering, instruction encoding, timing issues |
 | ISA extensions | Additional standard or custom extensions |
-| Privilege modes | S-mode (supervisor), U-mode (user) support |
+| Privilege modes | S-mode (supervisor), PMP, virtual memory (M and U modes already supported) |
 | Board support | New FPGA boards, SoC integrations |
 | Performance | Branch predictor, scheduler, memory-system, or cache improvements |
 | Peripherals | SPI, I2C, GPIO, timers |

Dockerfile

@@ -151,6 +151,33 @@ ENV RISCV_PREFIX=riscv-none-elf-
 # Fix git "dubious ownership" error when mounting repo as volume
 RUN git config --global --add safe.directory /workspace
 
+# Buildroot host dependencies + QEMU. Used by the FROST no-MMU Linux CI jobs:
+#   * build-frost-linux  - builds the kernel + initramfs + FROST memory images
+#     from the linux/buildroot-external tree (Buildroot compiles its own rv32
+#     uClibc cross toolchain from source, so it needs a full host build env).
+#   * qemu-linux-boot     - boots the same Image + rootfs to a shell under
+#     qemu-system-riscv32 (qemu-system-misc provides the riscv32 target).
+# `load_software.py <board> linux_boot` self-builds via the same path, so these
+# are the single source of truth for the Linux build's host deps. Kept as a late
+# layer so the expensive Verilator/Yosys/SMT source builds above stay cached.
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    build-essential \
+    patch \
+    cpio \
+    rsync \
+    bc \
+    file \
+    unzip \
+    wget \
+    bzip2 \
+    ccache \
+    libssl-dev \
+    libelf-dev \
+    libncurses-dev \
+    device-tree-compiler \
+    qemu-system-misc \
+    && rm -rf /var/lib/apt/lists/*
+
 # Install Python dependencies (cocotb, pytest, pre-commit, etc.)
 RUN pip install --no-cache-dir --break-system-packages \
     "cocotb==2.0.1" \

README.md

@@ -2,7 +2,7 @@
 
 **F**PGA **R**ISC-V **O**pen-sourced in **S**ystemVerilog by **T**woSigma
 
-An out-of-order RISC-V processor implementing **RV32GCB** (G = IMAFD) with a Tomasulo back-end and full machine-mode privilege support for RTOS operation. Achieves 300 MHz on UltraScale+. Designed for FPGA deployment with clean, portable SystemVerilog.
+An out-of-order RISC-V processor implementing **RV32GCB** (G = IMAFD) with a Tomasulo back-end and Machine + User (M/U) privilege modes for RTOS operation. Achieves 300 MHz on UltraScale+. Designed for FPGA deployment with clean, portable SystemVerilog.
 
 ## Why FROST?
 
@@ -55,7 +55,7 @@ There are many RISC-V cores. Here's what makes FROST different:
 │                                                                              │
 │   ┌──────────────────────────┐    ┌─────────────────────────────────────┐    │
 │   │ Trap Unit                │    │ Peripherals                         │    │
-│   │ (M-mode, mret, wfi,      │    │ UART, mtime/mtimecmp, FIFO0/1       │    │
+│   │ (M/U traps, mret, wfi,   │    │ UART, mtime/mtimecmp, FIFO0/1       │    │
 │   │  interrupts, exceptions) │    │                                     │    │
 │   └──────────────────────────┘    └─────────────────────────────────────┘    │
 │                                                                              │
@@ -82,6 +82,7 @@ There are many RISC-V cores. Here's what makes FROST different:
 | **Zbkb**         | Bit manipulation for crypto                    |
 | **Zihintpause**  | Pause hint for spin-wait loops                 |
 | **Machine Mode** | M-mode privilege (mret, wfi, ecall, ebreak)    |
+| **User Mode**    | U-mode privilege (ecall traps to M-mode)       |
 
 ### Architecture Highlights
 
@@ -96,7 +97,7 @@ There are many RISC-V cores. Here's what makes FROST different:
 - **Two-tier branch recovery** — conditional-branch mispredictions use a fast ~2-cycle path (front-end redirect + RAT restore in the same cycle); JALR and exceptions take the slower commit-time path
 - **Branch prediction** with a 256-entry 2-bit BTB (trained for conditional branches and JAL, with slot-2 lookup support), 1024-entry bimodal direction predictor, 8-entry return address stack, and PD-stage computed-target redirects for conditional BTB misses predicted taken
 - **L0 cache** in front of the load queue reduces load-use latency (direct-mapped, write-through)
-- **M-mode trap handling** for RTOS support (interrupts and exceptions)
+- **Machine + User (M/U) privilege modes** for RTOS support — traps from both modes are taken in M-mode (interrupts and exceptions)
 - **CLINT-compatible timer** (mtime/mtimecmp) for preemptive scheduling
 - **Harvard architecture** with separate instruction and data memory ports
 - **Write-back cache hierarchy over DDR** — a 1 GiB cached region at `0x8000_0000` served by recursive line-port caches (`frost_cache`: direct-mapped, 32 B lines, write-back/write-allocate). Both instruction fetch (a 16 KiB read-only L1I) and data (a 128 KiB L1D) run through it on every board — so code can execute from DDR, not just from low BRAM — sharing a 2:1 line-port arbiter (data-side priority), plus a 2 MiB UltraRAM L2 spliced in on UltraScale+, over the board's DDR (DDR3 on Genesys2, DDR4 on X3) through a single-beat AXI bridge
@@ -352,12 +353,12 @@ controller calibrates, so software never observes an uninitialized main memory.
 
 | Resource | Used | Available | Util% |
 |----------|-----:|----------:|------:|
-| Slice LUTs | 129,281 | 203,800 | 63.4% |
-|   LUT as Logic | 120,714 | 203,800 | 59.2% |
-|   LUT as Distributed RAM | 7,722 | — | — |
+| Slice LUTs | 130,622 | 203,800 | 64.1% |
+|   LUT as Logic | 122,015 | 203,800 | 59.9% |
+|   LUT as Distributed RAM | 7,762 | — | — |
 |   LUT as Shift Register | 845 | — | — |
-| Slice Registers | 86,734 | 407,600 | 21.3% |
-| Block RAM Tile | 189.5 | 445 | 42.6% |
+| Slice Registers | 87,375 | 407,600 | 21.4% |
+| Block RAM Tile | 219 | 445 | 49.2% |
 | DSPs | 36 | 840 | 4.3% |
 | F7 Muxes | 98 | 101,900 | 0.1% |
 | F8 Muxes | 33 | 50,950 | 0.1% |

__init__.py

@@ -15,7 +15,7 @@
 """FROST - RISC-V processor package.
 
 This package contains a complete RV32GCB (G = IMAFD) RISC-V processor
-implementation with full machine-mode support and additional extensions
+implementation with Machine (M) and User (U) privilege modes and additional extensions
 (Zicsr, Zicntr, Zifencei, Zicond, Zbkb, and Zihintpause), along with
 verification infrastructure, build tools, and software libraries.
 

boards/genesys2/genesys2_frost.sv

@@ -196,7 +196,10 @@ module genesys2_frost (
       // backed by the DDR3 controller through the AXI port below.
       .ENABLE_CACHED_TIER(1),
       .CACHED_HAS_L2(0),
-      .USE_BEHAVIORAL_DDR(0)
+      .USE_BEHAVIORAL_DDR(0),
+      // Bump L1I 16 KiB -> 128 KiB: hold the kernel tick/softirq/scheduler
+      // working set to defeat the periodic-tick catch-up livelock (no L2 here).
+      .L1I_CACHE_BYTES(128 * 1024)
   ) subsystem (
       .i_clk(main_clock),
       .i_clk_div4(divided_clock_by_4),

boards/xilinx_frost_subsystem.sv

@@ -32,7 +32,13 @@ module xilinx_frost_subsystem #(
     // 1 = the cached tier ends in the simulation-only behavioral DDR model;
     // 0 = it ends at the o_ddr_axi_*/i_ddr_axi_* ports below, wired to the
     // board's DDR controller subsystem (both boards drive 0).
-    parameter int unsigned USE_BEHAVIORAL_DDR = 1
+    parameter int unsigned USE_BEHAVIORAL_DDR = 1,
+    // L1 instruction-cache size in bytes. genesys2 (L1-only, no L2) bumps this
+    // above the 16 KiB default so the kernel periodic-tick/softirq/scheduler
+    // working set stays resident, addressing the tick-livelock I$ thrash.
+    parameter int unsigned L1I_CACHE_BYTES = 16 * 1024,
+    // Optional boot-hang UART classifier. Leave off for interactive testing.
+    parameter int unsigned ENABLE_HANG_TRIAGE = 0
 ) (
     input logic i_clk,       // Main CPU clock
     input logic i_clk_div4,  // Divided clock for JTAG/UART (1/4 of main clock)
@@ -217,7 +223,9 @@ module xilinx_frost_subsystem #(
       .CLK_FREQ_HZ(CLK_FREQ_HZ),
       .ENABLE_CACHED_TIER(ENABLE_CACHED_TIER),
       .CACHED_HAS_L2(CACHED_HAS_L2),
-      .USE_BEHAVIORAL_DDR(USE_BEHAVIORAL_DDR)
+      .USE_BEHAVIORAL_DDR(USE_BEHAVIORAL_DDR),
+      .L1I_CACHE_BYTES(L1I_CACHE_BYTES),
+      .ENABLE_HANG_TRIAGE(ENABLE_HANG_TRIAGE)
   ) frost_processor (
       .i_clk(i_clk),
       .i_clk_div4(i_clk_div4),

formal/store_queue.sby

@@ -14,7 +14,7 @@ smtbmc boolector
 [script]
 read -formal -sv riscv_pkg.sv
 read -sv sdp_dist_ram.sv
-read -sv sq_forwarding_unit.sv
+read -formal -sv sq_forwarding_unit.sv
 read -formal -sv store_queue.sv
 prep -top store_queue