Initialize RAM in assembly

Cargo.toml

@@ -17,7 +17,6 @@ autoexamples = true
 links = "cortex-m-rt" # Prevent multiple versions of cortex-m-rt being linked
 
 [dependencies]
-r0 = "1.0"
 cortex-m-rt-macros = { path = "macros", version = "=0.6.11" }
 # Note: Do not depend on `cortex-m` here. This crate is used for testing `cortex-m`, so we need to
 # avoid pulling in multiple versions of `cortex-m`.

asm.S

@@ -0,0 +1,105 @@
+  .cfi_sections .debug_frame
+
+  # Notes for function attributes:
+  # .type and .thumb_func are _both_ required, otherwise the Thumb mode bit
+  # will not be set and an invalid vector table is generated.
+  # LLD requires that section flags are set explicitly.
+
+  .section .HardFaultTrampoline, "ax"
+  .global HardFaultTrampoline
+  .type HardFaultTrampoline,%function
+  .thumb_func
+  .cfi_startproc
+  # HardFault exceptions are bounced through this trampoline which grabs the
+  # stack pointer at the time of the exception and passes it to the user's
+  # HardFault handler in r0.
+HardFaultTrampoline:
+  # Depending on the stack mode in EXC_RETURN, fetch stack pointer from
+  # PSP or MSP.
+  mov r0, lr
+  mov r1, #4
+  tst r0, r1
+  bne 0f
+  mrs r0, MSP
+  b HardFault
+0:
+  mrs r0, PSP
+  b HardFault
+  .cfi_endproc
+  .size HardFaultTrampoline, . - HardFaultTrampoline
+
+  .section .Reset, "ax"
+  .global Reset
+  .type Reset,%function
+  .thumb_func
+  .cfi_startproc
+  # Main entry point after reset. This jumps to the user __pre_init function,
+  # which cannot be called from Rust code without invoking UB, then
+  # initialises RAM. If the target has an FPU, it is enabled. Finally, jumps
+  # to the user main function.
+Reset:
+  # ARMv6-M does not initialise LR, but many tools expect it to be 0xFFFF_FFFF
+  # when reaching the first call frame, so we set it at startup.
+  # ARMv7-M and above initialise LR to 0xFFFF_FFFF at reset.
+  ldr r4,=0xffffffff
+  mov lr,r4
+
+  # Run user pre-init code, which must be executed immediately after startup,
+  # before the potentially time-consuming memory initialisation takes place.
+  # Example use cases include disabling default watchdogs or enabling RAM.
+  bl __pre_init
+
+  # Restore LR after calling __pre_init (r4 is preserved by subroutines).
+  mov lr,r4
+
+  # Initialise .bss memory. `__sbss` and `__ebss` come from the linker script.
+  ldr r0,=__sbss
+  ldr r1,=__ebss
+  mov r2,#0
+0:
+  cmp r1, r0
+  beq 1f
+  stm r0!, {r2}
+  b 0b
+1:
+
+  # Initialise .data memory. `__sdata`, `__sidata`, and `__edata` come from the
+  # linker script. Copy from r2 into r0 until r0 reaches r1.
+  ldr r0,=__sdata
+  ldr r1,=__edata
+  ldr r2,=__sidata
+2:
+  cmp r1, r0
+  beq 3f
+  # load 1 word from r2 to r3, inc r2
+  ldm r2!, {r3}
+  # store 1 word from r3 to r0, inc r0
+  stm r0!, {r3}
+  b 2b
+3:
+
+#ifdef HAS_FPU
+  # Conditionally enable the FPU.
+  # Address of SCB.CPACR.
+  ldr r0, =0xE000ED88
+  # Enable access to CP10 and CP11 from both privileged and unprivileged mode.
+  ldr r1, =(0b1111 << 20)
+  # RMW.
+  ldr r2, [r0]
+  orr r2, r2, r1
+  str r2, [r0]
+  # Barrier is required on some processors.
+  dsb
+  isb
+#endif
+
+4:
+  # Jump to user main function. We use bl for the extended range, but the
+  # user main function may not return.
+  bl main
+
+  # Trap on return.
+  udf
+
+  .cfi_endproc
+  .size Reset, . - Reset

assemble.sh

@@ -9,21 +9,25 @@ crate=cortex-m-rt
 # remove existing blobs because otherwise this will append object files to the old blobs
 rm -f bin/*.a
 
-arm-none-eabi-as -march=armv6s-m asm.s -o bin/$crate.o
+arm-none-eabi-gcc -c -march=armv6s-m asm.S -o bin/$crate.o
 ar crs bin/thumbv6m-none-eabi.a bin/$crate.o
 
-arm-none-eabi-as -march=armv7-m asm.s -o bin/$crate.o
+arm-none-eabi-gcc -c -march=armv7-m asm.S -o bin/$crate.o
 ar crs bin/thumbv7m-none-eabi.a bin/$crate.o
 
-arm-none-eabi-as -march=armv7e-m asm.s -o bin/$crate.o
+arm-none-eabi-gcc -c -march=armv7e-m asm.S -o bin/$crate.o
 ar crs bin/thumbv7em-none-eabi.a bin/$crate.o
+
+arm-none-eabi-gcc -c -march=armv7e-m asm.S -DHAS_FPU -o bin/$crate.o
 ar crs bin/thumbv7em-none-eabihf.a bin/$crate.o
 
-arm-none-eabi-as -march=armv8-m.base asm.s -o bin/$crate.o
+arm-none-eabi-gcc -c -march=armv8-m.base asm.S -o bin/$crate.o
 ar crs bin/thumbv8m.base-none-eabi.a bin/$crate.o
 
-arm-none-eabi-as -march=armv8-m.main asm.s -o bin/$crate.o
+arm-none-eabi-gcc -c -march=armv8-m.main asm.S -o bin/$crate.o
 ar crs bin/thumbv8m.main-none-eabi.a bin/$crate.o
+
+arm-none-eabi-gcc -c -march=armv8-m.main -DHAS_FPU asm.S -o bin/$crate.o
 ar crs bin/thumbv8m.main-none-eabihf.a bin/$crate.o
 
 rm bin/$crate.o

build.rs

@@ -7,8 +7,6 @@ fn main() {
     let target = env::var("TARGET").unwrap();
     let out_dir = PathBuf::from(env::var("OUT_DIR").unwrap());
 
-    has_fpu(&target);
-
     if target.starts_with("thumbv") {
         fs::copy(
             format!("bin/{}.a", target),
@@ -81,9 +79,3 @@ handlers.");
     println!("cargo:rerun-if-changed=build.rs");
     println!("cargo:rerun-if-changed=link.x.in");
 }
-
-fn has_fpu(target: &str) {
-    if target.ends_with("eabihf") {
-        println!("cargo:rustc-cfg=has_fpu");
-    }
-}

link.x.in

@@ -23,8 +23,9 @@
 INCLUDE memory.x
 
 /* # Entry point = reset vector */
+EXTERN(__RESET_VECTOR);
+EXTERN(Reset);
 ENTRY(Reset);
-EXTERN(__RESET_VECTOR); /* depends on the `Reset` symbol */
 
 /* # Exception vectors */
 /* This is effectively weak aliasing at the linker level */
@@ -85,13 +86,15 @@ SECTIONS
   /* ### .text */
   .text _stext :
   {
-    /* place these 2 close to each other or the `b` instruction will fail to link */
-    *(.PreResetTrampoline);
     *(.Reset);
 
     *(.text .text.*);
+
+    /* The HardFaultTrampoline uses the `b` instruction to enter `HardFault`,
+       so must be placed close to it. */
     *(.HardFaultTrampoline);
     *(.HardFault.*);
+
     . = ALIGN(4); /* Pad .text to the alignment to workaround overlapping load section bug in old lld */
   } > FLASH
   . = ALIGN(4); /* Ensure __etext is aligned if something unaligned is inserted after .text */