401 lines
11 KiB
ArmAsm
401 lines
11 KiB
ArmAsm
/*
|
|
* Copyright (c) 2010-2015 Wind River Systems, Inc.
|
|
*
|
|
* SPDX-License-Identifier: Apache-2.0
|
|
*/
|
|
/**
|
|
* @file
|
|
* @brief Crt0 module for the IA-32 boards
|
|
*
|
|
* This module contains the initial code executed by the Zephyr Kernel ELF image
|
|
* after having been loaded into RAM.
|
|
*/
|
|
|
|
#include <arch/x86/asm.h>
|
|
|
|
/* exports (private APIs) */
|
|
|
|
GTEXT(__start)
|
|
|
|
/* externs */
|
|
GTEXT(_Cstart)
|
|
|
|
GDATA(_idt_base_address)
|
|
GDATA(_interrupt_stack)
|
|
GDATA(_Idt)
|
|
#ifndef CONFIG_GDT_DYNAMIC
|
|
GDATA(_gdt)
|
|
#endif
|
|
|
|
|
|
#if defined(CONFIG_SSE)
|
|
GDATA(_sse_mxcsr_default_value)
|
|
#endif
|
|
|
|
#if defined(CONFIG_BOOT_TIME_MEASUREMENT)
|
|
GDATA(__start_tsc)
|
|
#endif
|
|
|
|
#ifdef CONFIG_SYS_POWER_DEEP_SLEEP
|
|
GTEXT(_sys_soc_resume_from_deep_sleep)
|
|
#endif
|
|
|
|
|
|
/* processor is executing in 32-bit protected mode */
|
|
|
|
.balign 16,0x90
|
|
|
|
SECTION_FUNC(TEXT_START, __start)
|
|
|
|
#ifdef CONFIG_BOOT_TIME_MEASUREMENT
|
|
/*
|
|
* Record BootTime from start of Kernel.
|
|
* Store value temporarily in Register edi & esi and
|
|
* write to memory once memory access is allowed.
|
|
* That is, once the data segment register has been setup to access
|
|
* the .data/.rodata/.bss section of the linked image.
|
|
*/
|
|
rdtsc
|
|
mov %eax, %esi /* low value */
|
|
mov %edx, %edi /* high value */
|
|
#endif
|
|
|
|
/* Enable write-back caching by clearing the NW and CD bits */
|
|
movl %cr0, %eax
|
|
andl $0x9fffffff, %eax
|
|
movl %eax, %cr0
|
|
|
|
/*
|
|
* Ensure interrupts are disabled. Interrupts are enabled when
|
|
* the first context switch occurs.
|
|
*/
|
|
|
|
cli
|
|
|
|
/*
|
|
* Although the bootloader sets up an Interrupt Descriptor Table (IDT)
|
|
* and a Global Descriptor Table (GDT), the specification encourages
|
|
* booted operating systems to setup their own IDT and GDT.
|
|
*/
|
|
#if CONFIG_SET_GDT
|
|
lgdt _gdt_rom /* load 32-bit operand size GDT */
|
|
#endif
|
|
lidt _Idt /* load 32-bit operand size IDT */
|
|
|
|
|
|
#ifdef CONFIG_SET_GDT
|
|
/* If we set our own GDT, update the segment registers as well.
|
|
*/
|
|
movw $0x10, %ax /* data segment selector (entry = 3) */
|
|
movw %ax, %ds /* set DS */
|
|
movw %ax, %es /* set ES */
|
|
movw %ax, %fs /* set FS */
|
|
movw %ax, %gs /* set GS */
|
|
movw %ax, %ss /* set SS */
|
|
|
|
ljmp $0x08, $__csSet /* set CS = 0x08 */
|
|
|
|
__csSet:
|
|
#endif /* CONFIG_SET_GDT */
|
|
|
|
|
|
#ifdef CONFIG_BOOT_TIME_MEASUREMENT
|
|
/*
|
|
* Store rdtsc result from temporary regiter ESI & EDI into memory.
|
|
*/
|
|
mov %esi, __start_tsc /* low value */
|
|
mov %edi, __start_tsc+4 /* high value */
|
|
#endif
|
|
|
|
#if !defined(CONFIG_FLOAT)
|
|
/*
|
|
* Force an #NM exception for floating point instructions
|
|
* since FP support hasn't been configured
|
|
*/
|
|
|
|
movl %cr0, %eax /* move CR0 to EAX */
|
|
orl $0x2e, %eax /* CR0[NE+TS+EM+MP]=1 */
|
|
movl %eax, %cr0 /* move EAX to CR0 */
|
|
#else
|
|
/*
|
|
* Permit use of x87 FPU instructions
|
|
*
|
|
* Note that all floating point exceptions are masked by default,
|
|
* and that _no_ handler for x87 FPU exceptions (#MF) is provided.
|
|
*/
|
|
|
|
movl %cr0, %eax /* move CR0 to EAX */
|
|
orl $0x22, %eax /* CR0[NE+MP]=1 */
|
|
andl $~0xc, %eax /* CR0[TS+EM]=0 */
|
|
movl %eax, %cr0 /* move EAX to CR0 */
|
|
|
|
fninit /* set x87 FPU to its default state */
|
|
|
|
#if defined(CONFIG_SSE)
|
|
/*
|
|
* Permit use of SSE instructions
|
|
*
|
|
* Note that all SSE exceptions are masked by default,
|
|
* and that _no_ handler for SSE exceptions (#XM) is provided.
|
|
*/
|
|
|
|
movl %cr4, %eax /* move CR4 to EAX */
|
|
orl $0x200, %eax /* CR4[OSFXSR] = 1 */
|
|
andl $~0x400, %eax /* CR4[OSXMMEXCPT] = 0 */
|
|
movl %eax, %cr4 /* move EAX to CR4 */
|
|
|
|
ldmxcsr _sse_mxcsr_default_value /* initialize SSE control/status reg */
|
|
|
|
#endif /* CONFIG_SSE */
|
|
|
|
#endif /* !CONFIG_FLOAT */
|
|
|
|
/*
|
|
* Set the stack pointer to the area used for the interrupt stack.
|
|
* Note this stack is used during the execution of __start() and
|
|
* _Cstart() until the multi-tasking kernel is initialized. The
|
|
* dual-purposing of this area of memory is safe since
|
|
* interrupts are disabled until the first context switch.
|
|
*
|
|
* nano_init.c enforces that the _interrupt_stack pointer and
|
|
* the ISR stack size are some multiple of STACK_ALIGN, which
|
|
* is at least 4.
|
|
*
|
|
* This is also used to call the _sys_soc_resume_from_deep_sleep()
|
|
* routine to avoid memory corruption if the system is resuming from
|
|
* deep sleep. It is important that _sys_soc_resume_from_deep_sleep()
|
|
* restores the stack pointer to what it was at deep sleep before
|
|
* enabling interrupts. This is necessary to avoid
|
|
* interfering with interrupt handler use of this stack.
|
|
* If it is a cold boot then _sys_soc_resume_from_deep_sleep() should
|
|
* not do anything and must return immediately.
|
|
*/
|
|
#ifdef CONFIG_INIT_STACKS
|
|
movl $0xAAAAAAAA, %eax
|
|
leal _interrupt_stack, %edi
|
|
stack_size_dwords = (CONFIG_ISR_STACK_SIZE / 4)
|
|
movl $stack_size_dwords, %ecx
|
|
rep stosl
|
|
#endif
|
|
|
|
movl $_interrupt_stack, %esp
|
|
addl $CONFIG_ISR_STACK_SIZE, %esp
|
|
|
|
#if defined(CONFIG_SYS_POWER_DEEP_SLEEP) && \
|
|
!defined(CONFIG_BOOTLOADER_CONTEXT_RESTORE)
|
|
/*
|
|
* Invoke _sys_soc_resume_from_deep_sleep() hook to handle resume from
|
|
* deep sleep. It should first check whether system is recovering from
|
|
* deep sleep state. If it is, then this function should restore
|
|
* states and resume at the point system went to deep sleep.
|
|
* In this case this function will never return.
|
|
*
|
|
* If system is not recovering from deep sleep then it is a
|
|
* cold boot. In this case, this function would immediately
|
|
* return and execution falls through to cold boot path.
|
|
*/
|
|
|
|
call _sys_soc_resume_from_deep_sleep
|
|
|
|
#endif
|
|
|
|
#ifdef CONFIG_XIP
|
|
/*
|
|
* copy DATA section from ROM to RAM region
|
|
* DATA is followed by BSS section.
|
|
*/
|
|
|
|
movl $__data_ram_start, %edi /* DATA in RAM (dest) */
|
|
movl $__data_rom_start, %esi /* DATA in ROM (src) */
|
|
movl $__data_num_words, %ecx /* Size of DATA in quad bytes */
|
|
|
|
#ifdef CONFIG_SSE
|
|
/* copy 16 bytes at a time using XMM until < 16 bytes remain */
|
|
|
|
movl %ecx ,%edx /* save number of quad bytes */
|
|
shrl $2, %ecx /* How many 16 bytes? */
|
|
je dataWords
|
|
|
|
dataDQ:
|
|
movdqu (%esi), %xmm0
|
|
movdqu %xmm0, (%edi)
|
|
addl $16, %esi
|
|
addl $16, %edi
|
|
loop dataDQ
|
|
|
|
dataWords:
|
|
movl %edx, %ecx /* restore # quad bytes */
|
|
andl $0x3, %ecx /* only need to copy at most 3 quad bytes */
|
|
#endif /* CONFIG_SSE */
|
|
|
|
rep
|
|
movsl /* copy data 4 bytes at a time */
|
|
|
|
#endif /* CONFIG_XIP */
|
|
|
|
/*
|
|
* Clear BSS: bzero (__bss_start, __bss_num_words*4)
|
|
*
|
|
* It's assumed that BSS size will be a multiple of a long (4 bytes),
|
|
* and aligned on a double word (32-bit) boundary
|
|
*/
|
|
|
|
#ifdef CONFIG_SSE
|
|
|
|
/* use XMM register to clear 16 bytes at a time */
|
|
|
|
pxor %xmm0, %xmm0 /* zero out xmm0 register */
|
|
movl $__bss_start, %edi /* load BSS start address */
|
|
movl $__bss_num_words, %ecx /* number of quad bytes in .bss */
|
|
movl %ecx, %edx /* make a copy of # quad bytes */
|
|
shrl $2, %ecx /* How many multiples of 16 byte ? */
|
|
je bssWords
|
|
bssDQ:
|
|
movdqu %xmm0, (%edi) /* zero 16 bytes... */
|
|
addl $16, %edi
|
|
loop bssDQ
|
|
|
|
/* fall through to handle the remaining double words (32-bit chunks) */
|
|
|
|
bssWords:
|
|
xorl %eax, %eax /* fill memory with 0 */
|
|
movl %edx, %ecx /* move # quad bytes into ECX (for rep) */
|
|
andl $0x3, %ecx /* only need to zero at most 3 quad bytes */
|
|
cld
|
|
rep
|
|
stosl /* zero memory per 4 bytes */
|
|
|
|
#else /* !CONFIG_SSE */
|
|
|
|
/* clear out BSS double words (32-bits at a time) */
|
|
|
|
xorl %eax, %eax /* fill memory with 0 */
|
|
movl $__bss_start, %edi /* load BSS start address */
|
|
movl $__bss_num_words, %ecx /* number of quad bytes */
|
|
cld
|
|
rep
|
|
stosl /* zero memory per 4 bytes */
|
|
|
|
#endif /* CONFIG_SSE */
|
|
|
|
#ifdef CONFIG_GDT_DYNAMIC
|
|
/* activate RAM-based Global Descriptor Table (GDT) */
|
|
lgdt %ds:_gdt
|
|
#endif
|
|
|
|
/* Jump to C portion of kernel initialization and never return */
|
|
|
|
jmp _Cstart
|
|
|
|
#if defined(CONFIG_SSE)
|
|
|
|
/* SSE control & status register initial value */
|
|
|
|
_sse_mxcsr_default_value:
|
|
.long 0x1f80 /* all SSE exceptions clear & masked */
|
|
|
|
#endif /* CONFIG_SSE */
|
|
|
|
/* Interrupt Descriptor Table (IDT) definition */
|
|
|
|
_Idt:
|
|
.word (CONFIG_IDT_NUM_VECTORS * 8) - 1 /* limit: size of IDT-1 */
|
|
|
|
/*
|
|
* Physical start address = 0. When executing natively, this
|
|
* will be placed at the same location as the interrupt vector table
|
|
* setup by the BIOS (or GRUB?).
|
|
*/
|
|
|
|
.long _idt_base_address /* physical start address */
|
|
|
|
|
|
#ifdef CONFIG_BOOTLOADER_UNKNOWN
|
|
/* Multiboot header definition is needed for some bootloaders */
|
|
|
|
/*
|
|
* The multiboot header must be in the first 8 Kb of the kernel image
|
|
* (not including the ELF section header(s)) and be aligned on a
|
|
* 4 byte boundary.
|
|
*/
|
|
|
|
.balign 4,0x90
|
|
|
|
.long 0x1BADB002 /* multiboot magic number */
|
|
|
|
/*
|
|
* Flags = no bits are being set, specifically bit 16 is not being
|
|
* set since the supplied kernel image is an ELF file, and the
|
|
* multiboot loader shall use the information from the program and
|
|
* section header to load and boot the kernel image.
|
|
*/
|
|
|
|
.long 0x00000000
|
|
|
|
/*
|
|
* checksum = 32-bit unsigned value which, when added to the other
|
|
* magic fields (i.e. "magic" and "flags"), must have a 32-bit
|
|
* unsigned sum of zero.
|
|
*/
|
|
|
|
.long -(0x1BADB002 + 0)
|
|
#endif /* CONFIG_BOOTLOADER_UNKNOWN */
|
|
|
|
#ifdef CONFIG_SET_GDT
|
|
|
|
/* GDT should be aligned on 8-byte boundary for best processor
|
|
* performance, see Section 3.5.1 of IA architecture SW developer
|
|
* manual, Vol 3.
|
|
*/
|
|
|
|
.balign 8
|
|
|
|
/*
|
|
* The following 3 GDT entries implement the so-called "basic
|
|
* flat model", i.e. a single code segment descriptor and a single
|
|
* data segment descriptor, giving the kernel access to a continuous,
|
|
* unsegmented address space. Both segment descriptors map the entire
|
|
* linear address space (i.e. 0 to 4 GB-1), thus the segmentation
|
|
* mechanism will never generate "out of limit memory reference"
|
|
* exceptions even if physical memory does not reside at the referenced
|
|
* address.
|
|
*
|
|
* The 'A' (accessed) bit in the type field is set for all the
|
|
* data/code segment descriptors to accommodate placing these entries
|
|
* in ROM, to prevent the processor from freaking out when it tries
|
|
* and fails to set it.
|
|
*/
|
|
|
|
#ifndef CONFIG_GDT_DYNAMIC
|
|
_gdt:
|
|
#endif
|
|
_gdt_rom:
|
|
|
|
/* Entry 0 (selector=0x0000): The "NULL descriptor". The CPU never
|
|
* actually looks at this entry, so we stuff 6-byte the pseudo
|
|
* descriptor here */
|
|
.word _gdt_rom_end - _gdt_rom - 1 /* Limit on GDT */
|
|
.long _gdt_rom /* table address: _gdt_rom */
|
|
.word 0x0000
|
|
|
|
/* Entry 1 (selector=0x0008): Code descriptor: DPL0 */
|
|
|
|
.word 0xffff /* limit: xffff */
|
|
.word 0x0000 /* base : xxxx0000 */
|
|
.byte 0x00 /* base : xx00xxxx */
|
|
.byte 0x9b /* Accessed, Code e/r, Present, DPL0 */
|
|
.byte 0xcf /* limit: fxxxx, Page Gra, 32bit */
|
|
.byte 0x00 /* base : 00xxxxxx */
|
|
|
|
/* Entry 2 (selector=0x0010): Data descriptor: DPL0 */
|
|
|
|
.word 0xffff /* limit: xffff */
|
|
.word 0x0000 /* base : xxxx0000 */
|
|
.byte 0x00 /* base : xx00xxxx */
|
|
.byte 0x93 /* Accessed, Data r/w, Present, DPL0 */
|
|
.byte 0xcf /* limit: fxxxx, Page Gra, 32bit */
|
|
.byte 0x00 /* base : 00xxxxxx */
|
|
|
|
_gdt_rom_end:
|
|
#endif
|