[kernel][assembler] Add .cfi_startproc, .cfi_endproc directives

... and switch to using .debug_frame from .eh_frame.

Some functions don't really need .cfi directives, e.g. during early boot.
These functions are marked with FUNCTION_LABEL instead of FUNCTION
as now all uses of FUNCTION require a corresponding END_FUNCTION.

MG-870 #comment baby steps

Change-Id: I3c7d182ca91b2f705100880282477bfe930c6633

kernel/arch/arm64/exceptions.S

@@ -145,7 +145,7 @@ add sp, sp, #(4*8) + (10*8)
     b  arm64_exc_shared_restore_long
 .endm
 
-FUNCTION(arm64_exception_base)
+FUNCTION_LABEL(arm64_exception_base)
 
 /* exceptions from current EL, using SP0 */
 LOCAL_FUNCTION(arm64_sync_exc_current_el_SP0)

kernel/arch/arm64/mexec.S

@@ -91,8 +91,9 @@ FUNCTION(mexec_asm)
 
     // Branch to the next kernel.
     br      new_kernel_addr
+END_FUNCTION(mexec_asm)
 
-memmove_mexec:
+LOCAL_FUNCTION(memmove_mexec)
         // x6 contains the stride (1 word if we're copying forward
         // -1 word if we're copying backwards)
         mov     x6, 1
@@ -142,6 +143,7 @@ memmove_mexec:
         bne     .copy_loop
 .done:
         ret
+END_FUNCTION(memmove_mexec)
 
 // Perform a bulk clean/invalidate across the whole cache
 // Normally on ARM we can use the CIVAC, CVAC, CVAU and IVAC instructions to
@@ -155,7 +157,7 @@ memmove_mexec:
 // cache.
 // The original implementation can be found in the ARMv8-A TRM or at the
 // following URL: http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.den0024a/BABJDBHI.html
-mexec_arch_clean_invalidate_cache_all:
+LOCAL_FUNCTION(mexec_arch_clean_invalidate_cache_all)
          mrs x0, clidr_el1
          and w3, w0, #0x07000000  // get 2 x level of coherence
          lsr w3, w3, #23
@@ -199,6 +201,7 @@ mexec_arch_clean_invalidate_cache_all:
          dsb sy
 
          ret
+END_FUNCTION(mexec_arch_clean_invalidate_cache_all)
 
 
 /* This .ltorg emits any immediate constants here. We need to put this before

kernel/arch/arm64/start.S

@@ -42,7 +42,7 @@ size                    .req x26
 attr                    .req x27
 
 .section .text.boot, "ax", @progbits
-FUNCTION(_start)
+FUNCTION_LABEL(_start)
 #if FASTBOOT_HEADER
 #include "fastboot-header.S"
 #endif

kernel/arch/x86/64/asm.S

@@ -31,6 +31,7 @@ FUNCTION(x86_64_context_switch)
     popf
 
     retq
+END_FUNCTION(x86_64_context_switch)
 
 #include <arch/x86/mp.h>
 
@@ -51,6 +52,7 @@ FUNCTION(arch_spin_lock)
     cmp $0, %rax // make sure we actually got the lock
     jne .Lspin // if we lost the race, resume waiting
     ret
+END_FUNCTION(arch_spin_lock)
 
 /* int arch_spin_trylock(unsigned long *lock) */
 FUNCTION(arch_spin_trylock)
@@ -60,12 +62,14 @@ FUNCTION(arch_spin_trylock)
 
     xchg %rax, (%rdi)
     ret // return 0 if we got the lock
+END_FUNCTION(arch_spin_trylock)
 
 /* void arch_spin_unlock(spin_lock_t *lock) */
 FUNCTION(arch_spin_unlock)
     mov $0, %rax
     xchg %rax, (%rdi)
     ret
+END_FUNCTION(arch_spin_unlock)
 
 /* rep stos version of page zero */
 FUNCTION(arch_zero_page)
@@ -76,3 +80,4 @@ FUNCTION(arch_zero_page)
     rep     stosq
 
     ret
+END_FUNCTION(arch_zero_page)

kernel/arch/x86/64/exceptions.S

@@ -42,6 +42,7 @@ FUNCTION(_isr_\@)
     pushq $\@                                /* interrupt number */
     jmp interrupt_common
 .endif
+END_FUNCTION(_isr_\@)
 .popsection
 .pushsection .rodata
 .quad _isr_\@
@@ -122,3 +123,4 @@ interrupt_common_iframe_set_up_for_debugger:
     addq $16, %rsp
 
     iretq
+END_FUNCTION(interrupt_common)

kernel/arch/x86/64/hypervisor.S

@@ -79,6 +79,7 @@ failure:
     // Return MX_ERR_INTERNAL.
     mov $MX_ERR_INTERNAL, %eax
     ret
+END_FUNCTION(vmx_enter)
 
 /* This is effectively the second-half of vmx_enter. When we return from a
  * VM exit, vmx_state argument is stored in RSP. We use this to restore the
@@ -138,3 +139,4 @@ FUNCTION(vmx_exit_entry)
     // Return MX_OK, using the return address of vmx_enter pushed above.
     mov $MX_OK, %eax
     ret
+END_FUNCTION(vmx_exit_entry)

kernel/arch/x86/64/ops.S

@@ -32,6 +32,7 @@ FUNCTION(arch_idle)
     hlt
 1:
     ret
+END_FUNCTION(arch_idle)
 
 
 /* status_t read_msr_safe(uint32_t msr_id, uint64_t *val); */
@@ -67,3 +68,4 @@ FUNCTION(read_msr_safe)
 
     mov $MX_ERR_NOT_SUPPORTED, %rax
     ret
+END_FUNCTION(read_msr_safe)

kernel/arch/x86/64/start.S

@@ -86,7 +86,7 @@
 
 .section .text.boot, "ax", @progbits
 .code32
-FUNCTION(_multiboot_start)
+FUNCTION_LABEL(_multiboot_start)
     cmpl $MULTIBOOT_BOOTLOADER_MAGIC, %eax
     jne .Lcommon_boot
     movl %ebx, PHYS(_multiboot_info)
@@ -214,7 +214,7 @@ high_entry:
 
     /* 64bit entry point from a secondary loader */
 .align 8
-FUNCTION(_entry64)
+FUNCTION_LABEL(_entry64)
     mov %esi, PHYS(_bootdata_base)
     /* ensure the stack pointer is sane */
 

kernel/arch/x86/64/start16.S

@@ -17,10 +17,10 @@
 .align PAGE_SIZE
 
 .section .text
-FUNCTION(x86_bootstrap16_start)
+FUNCTION_LABEL(x86_bootstrap16_start)
 
 .code16
-FUNCTION(x86_bootstrap16_entry)
+FUNCTION_LABEL(x86_bootstrap16_entry)
     # Enter no-fill cache mode (allegedly this is the initial state
     # according to Intel 3A, but on at least one Broadwell the APs can
     # come up with caching enabled)
@@ -84,7 +84,7 @@ FUNCTION(x86_bootstrap16_entry)
 
 # Get the secondary cpu into 64-bit mode with interrupts disabled and no TSS
 .code64
-FUNCTION(_x86_secondary_cpu_long_mode_entry)
+FUNCTION_LABEL(_x86_secondary_cpu_long_mode_entry)
     # When we get here, %rsi should contain the absolute address of our data
     # page.
     mov $1, %rdi
@@ -144,7 +144,7 @@ FUNCTION(_x86_secondary_cpu_long_mode_entry)
     jmp 0b
 
 # Get the cpu into 64-bit mode with interrupts disabled and no TSS
-FUNCTION(_x86_suspend_wakeup)
+FUNCTION_LABEL(_x86_suspend_wakeup)
     # Retrieve the new PML4 address before our data page becomes unreachable
     mov BCD_PHYS_KERNEL_PML4_OFFSET(%esi), %ecx
 

kernel/arch/x86/64/user_copy.S

@@ -90,6 +90,7 @@ FUNCTION(_x86_copy_from_user)
 
     end_usercopy
     ret
+END_FUNCTION(_x86_copy_from_user)
 
 # status_t _x86_copy_to_user(void *dst, const void *src, size_t len, bool smap, void **fault_return)
 FUNCTION(_x86_copy_to_user)
@@ -130,3 +131,4 @@ FUNCTION(_x86_copy_to_user)
 
     end_usercopy
     ret
+END_FUNCTION(_x86_copy_to_user)

kernel/arch/x86/64/uspace_entry.S

@@ -44,3 +44,4 @@ FUNCTION(x86_uspace_entry)
     mov %ax, %gs
 
     iretq
+END_FUNCTION(x86_uspace_entry)

kernel/arch/x86/mexec.S

@@ -8,7 +8,10 @@
 #include <asm.h>
 
 .section .text
+
 FUNCTION(mexec_asm)
 	// Unimplemented for now, emit an undefined instruction.
 	UD2
+END_FUNCTION(mexec_asm)
+
 DATA(mexec_asm_end)

kernel/dev/psci/psci.S

@@ -12,9 +12,11 @@
 FUNCTION(psci_smc_call)
     smc     #0
     ret
+END_FUNCTION(psci_smc_call)
 
 FUNCTION(psci_hvc_call)
     hvc     #0
     ret
+END_FUNCTION(psci_hvc_call)
 
 .ltorg

kernel/include/asm.h

@@ -8,14 +8,28 @@
 #ifndef __ASM_H
 #define __ASM_H
 
-#define FUNCTION(x) .global x; .type x,STT_FUNC; x:
-#define DATA(x) .global x; .type x,STT_OBJECT; x:
+#ifndef ASSEMBLY
+#error for assembly files only
+#endif
+
+// for functions that don't have an "end" or don't want .cfi_startproc
+#define LOCAL_FUNCTION_LABEL(x) .type x,STT_FUNC; x:
+#define FUNCTION_LABEL(x) .global x; LOCAL_FUNCTION_LABEL(x)
+
+#define LOCAL_FUNCTION(x) LOCAL_FUNCTION_LABEL(x) .cfi_startproc
+#define FUNCTION(x) .global x; LOCAL_FUNCTION(x)
+
+// for local or global functions
+#define END_FUNCTION(x) .cfi_endproc; .size x, . - x
 
-#define LOCAL_FUNCTION(x) .type x,STT_FUNC; x:
 #define LOCAL_DATA(x) .type x,STT_OBJECT; x:
+#define DATA(x) .global x; LOCAL_DATA(x)
 
-#define END_FUNCTION(x) .size x, . - x
+// for local or global data
 #define END_DATA(x) .size x, . - x
 
-#endif
+// We want .debug_frame not .eh_frame.
+// WARNING: This is a subtle side-effect of including this file. Heads up!
+.cfi_sections .debug_frame
 
+#endif