Rationalise the directive order for inline asm.

I went with:

* .section
* .arm or .thumb
* .global
* .type
* label
* code

This fixes the smp_test built for thumb mode, which was missing a .arm
directive (and the start-up code doesn't work in thumb mode).

Author: thejpster

aarch32-rt/src/arch_v4/abort.rs

@@ -5,13 +5,13 @@ core::arch::global_asm!(
     // Work around https://github.com/rust-lang/rust/issues/127269
     .fpu vfp2
 
-    .section .text._asm_default_data_abort_handler
 
     // Called from the vector table when we have an undefined exception.
     // Saves state and calls a C-compatible handler like
     // `extern "C" fn _data_abort_handler(addr: usize);`
-    .global _asm_default_data_abort_handler
+    .section .text._asm_default_data_abort_handler
     .arm
+    .global _asm_default_data_abort_handler
     .type _asm_default_data_abort_handler, %function
     _asm_default_data_abort_handler:
         // Subtract 8 from the stored LR, see p.1214 of the ARMv7-A architecture manual.
@@ -39,13 +39,13 @@ core::arch::global_asm!(
         movs    pc, lr
     .size _asm_default_data_abort_handler, . - _asm_default_data_abort_handler
 
-    .section .text._asm_default_prefetch_abort_handler
 
     // Called from the vector table when we have a prefetch abort.
     // Saves state and calls a C-compatible handler like
     // `extern "C" fn _prefetch_abort_handler(addr: usize);`
-    .global _asm_default_prefetch_abort_handler
+    .section .text._asm_default_prefetch_abort_handler
     .arm
+    .global _asm_default_prefetch_abort_handler
     .type _asm_default_prefetch_abort_handler, %function
     _asm_default_prefetch_abort_handler:
         // Subtract 4 from the stored LR, see p.1212 of the ARMv7-A architecture manual.

aarch32-rt/src/arch_v4/interrupt.rs

@@ -8,13 +8,12 @@ core::arch::global_asm!(
     // Work around https://github.com/rust-lang/rust/issues/127269
     .fpu vfp2
 
-    .section .text._asm_default_irq_handler
-
     // Called from the vector table when we have an interrupt.
     // Saves state and calls a C-compatible handler like
     // `extern "C" fn _irq_handler();`
-    .global _asm_default_irq_handler
+    .section .text._asm_default_irq_handler
     .arm
+    .global _asm_default_irq_handler
     .type _asm_default_irq_handler, %function
     _asm_default_irq_handler:
         // make sure we jump back to the right place

aarch32-rt/src/arch_v4/svc.rs

@@ -6,13 +6,13 @@ core::arch::global_asm!(
     // Work around https://github.com/rust-lang/rust/issues/127269
     .fpu vfp2
 
-    .section .text._asm_default_svc_handler
 
     // Called from the vector table when we have an software interrupt.
     // Saves state and calls a C-compatible handler like
     // `extern "C" fn _svc_handler(svc: u32);`
-    .global _asm_default_svc_handler
+    .section .text._asm_default_svc_handler
     .arm
+    .global _asm_default_svc_handler
     .type _asm_default_svc_handler, %function
     _asm_default_svc_handler:
         stmfd   sp!, {{ r0, lr }}

aarch32-rt/src/lib.rs

@@ -804,8 +804,8 @@ core::arch::global_asm!(
 
     // Initialises stacks, .data and .bss
     .section .text._init_segments
-    .global _init_segments
     .arm
+    .global _init_segments
     .type _init_segments, %function
     _init_segments:
         // Initialise .bss

examples/mps3-an536/src/bin/smp_test.rs

@@ -264,6 +264,7 @@ core::arch::global_asm!(
     //
     // Pass in stack top in r0.
     .section .text._stack_setup
+    .arm
     .global _stack_setup
     .type _stack_setup, %function
     _stack_setup: