CE5.0 - eboot汇编Startup.s中MMU设置流程详细分析

CE5.0 - eboot汇编Startup.s中MMU设置流程详细分析

以下为SMDK开发板startup.s部分启动代码.
;-------------------------------------------------------------------------------

MemoryMap EQU 0x2a4
BANK_SIZE EQU 0x00100000 ; 1MB per bank in MemoryMap array
BANK_SHIFT EQU 20

; Define RAM space for the Page Tables:
;
PHYBASE EQU 0x30000000 ; physical start
PTs EQU 0x30010000 ; 1st level page table address (PHYBASE + 0x10000)
; save room for interrupt vectors.
;------------------------------------------------------------------------------
; Copy boot loader to memory

ands r9, pc, #0xFF000000 ; see if we are in flash or in ram
//如果高8字节非0,那么说明pc已经在ram,否则说明现在pc位于reset之后的0x00000000地址区间[luther.gliethttp]

bne %f20 ; go ahead if we are already in ram//高8字节非0,说明eboot已在DDR中,跳到下面的20标号

; This is the loop that perform copying.
ldr r0, = 0x38000 ; offset into the RAM
add r0, r0, #PHYBASE ; add physical base //eboot最终的物理地址为0x30038000,这也就是PLATFORM/SMDK2440A/Src/Bootloader/Eboot_usb/boot.bib中定义的

//eboot链接地址EBOOT 8c038000 00016800 RAMIMAGE //一个.bib允许由一个RAMIMAGE类型段,生成一个对应的.bin文件,之后由romimage.exe生成.nb0同时填充eboot的pTOC全局变量指针[luther.gliethttp]

mov r1, r0 ; (r1) copy destination
//因为s3c2440在nand启动时,会将nand中前4k数据读取到s3c2440内部的4k ram中,同时ram映射成0地址,cpu

//跳转到ram开始执行从nand读取的前4k代码数据,于是读取位于ram中0地址之后的4k数据就等同于读取了nand的前4k数据[luther.gliethttp].

ldr r2, =0x0 ; (r2) flash started at physical address 0
ldr r3, =0x10000 ; counter (0x40000/4)//循环读取4k字节

10 ldr r4, [r2], #4
str r4, [r1], #4
subs r3, r3, #1
bne %b10

; Restart from the RAM position after copying.
mov pc, r0 //跳转到物理地址0x30038000重新执行上面的reset代码,再次执行时因为已经位于ram,所以将执行上面的bne %f20语句[luther.gliethttp]

nop //为流水线添加可能存在的指令最大空取次数

nop
nop

; Shouldn't get here.
b .

INCLUDE oemaddrtab_cfg.inc //这就是eboot的mmu映射表,位于PLATFORM/SMDK2440A/Src/Inc/oemaddrtab_cfg.inc
/*
EXPORT g_oalAddressTable[DATA]
g_oalAddressTable
DCD 0x80000000, 0x32000000, 32 ; 32 MB DRAM BANK 6
DCD 0x82000000, 0x08000000, 32 ; 32 MB SROM(SRAM/ROM) BANK 1
DCD 0x84000000, 0x10000000, 32 ; nGCS2: PCMCIA/PCCARD
DCD 0x86000000, 0x18000000, 32 ; 32 MB SROM(SRAM/ROM) BANK 3
DCD 0x88000000, 0x20000000, 32 ; 32 MB SROM(SRAM/ROM) BANK 4
DCD 0x8A000000, 0x28000000, 32 ; 32 MB SROM(SRAM/ROM) BANK 5
DCD 0x8C000000, 0x30000000, 32 ; 32 MB DRAM BANK 6
DCD 0x90800000, 0x48000000, 1 ; Memory control register
DCD 0x90900000, 0x49000000, 1 ; USB Host register
DCD 0x90A00000, 0x4A000000, 1 ; Interrupt Control register
DCD 0x90B00000, 0x4B000000, 1 ; DMA control register
DCD 0x90C00000, 0x4C000000, 1 ; Clock & Power register
DCD 0x90D00000, 0x4D000000, 1 ; LCD control register
DCD 0x90E00000, 0x4E000000, 1 ; NAND flash control register
DCD 0x90F00000, 0x4F000000, 1 ; Camera control register
DCD 0x91000000, 0x50000000, 1 ; UART control register
DCD 0x91100000, 0x51000000, 1 ; PWM timer register
DCD 0x91200000, 0x52000000, 1 ; USB device register
DCD 0x91300000, 0x53000000, 1 ; Watchdog Timer register
DCD 0x91400000, 0x54000000, 1 ; IIC control register
DCD 0x91500000, 0x55000000, 1 ; IIS control register
DCD 0x91600000, 0x56000000, 1 ; I/O Port register
DCD 0x91700000, 0x57000000, 1 ; RTC control register
DCD 0x91800000, 0x58000000, 1 ; A/D convert register
DCD 0x91900000, 0x59000000, 1 ; SPI register
DCD 0x91A00000, 0x5A000000, 1 ; SD Interface register
DCD 0x92000000, 0x00000000, 32 ; 32 MB SROM(SRAM/ROM) BANK 0
DCD 0x00000000, 0x00000000, 0 ; end of table
*/

; Compute physical address of the OEMAddressTable.
20 add r11, pc, #g_oalAddressTable - (. + 8)
ldr r10, =PTs ; (r10) = 1st level page table
0x20000000

; Setup 1st level page table (using section descriptor)
; Fill in first level page table entries to create "un-mapped" regions
; from the contents of the MemoryMap array.
;
; (r10) = 1st level page table
; (r11) = ptr to MemoryMap array
//PTR的地址大家都选在了DDR开始的offset处,ce为64k偏移,linux也是这样做的
//然后第1级MMU入口为pc地址的高14位(过滤掉低18位)[luther.gliethttp]
//比如0x80000000地址对应的第1级MMU入口偏移为0x80000000 >> 18 = 0x2000
add r10, r10, #0x2000 ; (r10) = ptr to 1st PTE for "unmapped space"
//r10存放0x80000000虚拟地址在PTR中的对应地址
mov r0, #0x0E ; (r0) = PTE for 0: 1MB cachable bufferable //有高速缓存C和写缓冲B
orr r0, r0, #0x400 ; set kernel r/w permission
25 mov r1, r11 ; (r1) = ptr to MemoryMap array

30 ldr r2, [r1], #4 ; (r2) = virtual address to map Bank at
ldr r3, [r1], #4 ; (r3) = physical address to map from
ldr r4, [r1], #4 ; (r4) = num MB to map

cmp r4, #0 ; End of table?
beq %f40 //表示cachable地址0x80000000~0xa0000000映射MMU创建完毕,表结尾,前跳到40标号,创建对应的uncached映射MMU[luther.gliethttp]

ldr r5, =0x1FF00000
and r2, r2, r5 ; VA needs 512MB, 1MB aligned. //对va虚拟地址进行512M对齐[luther.gliethttp]
//也就是r2的上限值不能超过0x20000000即512M,因为r10现在指向1级映射表的0x80000000虚拟地址,
//又因为0x80000000+0x20000000=0xa0000000开始地址存放的是uncached地址,所以这里要做512M数据限制[luther.gliethttp]

ldr r5, =0xFFF00000
and r3, r3, r5 ; PA needs 4GB, 1MB aligned. //对pa物理地址进行4G对齐

add r2, r10, r2, LSR #18
add r0, r0, r3 ; (r0) = PTE for next physical page //在MMU控制flag基础上追加映射到的pa物理地址[luther.gliethttp]

35 str r0, [r2], #4 //存到r2对应的PTR偏移中
add r0, r0, #0x00100000 ; (r0) = PTE for next physical page //1M递增,因为1<<18=256K,而每一个pte描述4k页,所以最终描述256k*4k=1M地址空间[luther.gliethttp]
sub r4, r4, #1 ; Decrement number of MB left
cmp r4, #0
bne %b35 ; Map next MB //遍历到该region结束

bic r0, r0, #0xF0000000 ; Clear Section Base Address Field
bic r0, r0, #0x0FF00000 ; Clear Section Base Address Field//清空r0中的地址信息[lutehr.gliethttp]
b %b30 ; Get next element //继续创建oemaddrtab_cfg.inc描述的下一region[luther.gliethttp]
//创建对应的0xa0000000开始的uncached映射MMU表
40 tst r0, #8
//比较C高速缓存是否仍然置位,如果仍然置位了,那么说明还没有执行uncached创建,如果C位已经清0,那么说明,uncached循环也执行完毕了,所以跳回到25标号继续创建[luther.gliethttp]
bic r0, r0, #0x0C ; clear cachable & bufferable bits in PTE//清除B写缓冲和C高速缓存[luther.gliethttp]
add r10, r10, #0x0800 ; (r10) = ptr to 1st PTE for "unmapped uncached space"//r10现在对应0x80000000虚拟地址的PTR起始地址,hex(0x20000000>>18)为0x800,
//所以r10 = r10 + 0x800;之后r10指向了0xa0000000虚拟地址对应的PTR起始地址[luther.gliethtp]
bne %b25 ; go setup PTEs for uncached space
//ok,现在0x80000000 cachable空间和0xa0000000 uncached空间都已经创建完毕了[luther.gliethttp]
sub r10, r10, #0x3000 ; (r10) = restore address of 1st level page table//该句好像有问题,不过还好后面用到r10之前,又对r10进行了重新赋值,所以也更说明,该句操作无意义[luther.gliethttp]

//接下来为虚拟0地址建立cachable和uncached映射[lutehr.gliethttp]
; Setup mmu to map (VA == 0) to (PA == 0x30000000).
ldr r0, =PTs ; PTE entry for VA = 0
ldr r1, =0x3000040E ; uncache/unbuffer/rw, PA base == 0x30000000//以C,B均开启的方式映射虚拟地址0到pa物理地址0x30000000
str r1, [r0]

; uncached area.
add r0, r0, #0x0800 ; PTE entry for VA = 0x0200.0000 , uncached//接下来的0x2000000虚拟地址被映射为虚拟0地址的uncached起始地址[luther.gliethttp]
ldr r1, =0x30000402 ; uncache/unbuffer/rw, base == 0x30000000//以C,B均不开启的方式映射虚拟地址0到pa物理地址0x30000000
str r1, [r0]
//经过上面设置可以看到
//虚拟地址0对应0x30000000的cachable
//虚拟地址0x2000000对应0x30000000的uncached.[luther.gliethttp]
; Comment:
; The following loop is to direct map RAM VA == PA. i.e.
; VA == 0x30XXXXXX => PA == 0x30XXXXXX for S3C2400
; Fill in 8 entries to have a direct mapping for DRAM
;
ldr r10, =PTs ; restore address of 1st level page table
ldr r0, =PHYBASE //PHYBASE为0x30000000,将虚拟地址映射到相同的物理地址

add r10, r10, #(0x3000 / 4) ; (r10) = ptr to 1st PTE for 0x30000000 //就是(0x30000000 >> 16) >> 2即(0x30000000 >> 16) / 4 [luther.gliethttp]
//现在r10指向了和DDR地址相等的PA地址对应的PTR地址处
add r0, r0, #0x1E ; 1MB cachable bufferable
orr r0, r0, #0x400 ; set kernel r/w permission
mov r1, #0
mov r3, #64 //映射64M空间,因为我们的物理DDR只有64M(见上面的oemaddrtab_cfg.inc表图),不能大于512,因为下面只支持到512次循环(r1为循环次数,每循环1次,创建1M空间映射表)[luther.gliethtp]
45 mov r2, r1 ; (r2) = virtual address to map Bank at
cmp r2, #0x20000000:SHR:BANK_SHIFT //0x20000000>>20=512M,
add r2, r10, r2, LSL #BANK_SHIFT-18 //add r2, r10, r2, 4即r2 = r10 + r2*4;正好为PTR操作时的4字节对齐边界[luther.gliethttp]
strlo r0, [r2] //如果还没有操作到512M,那么填充PTR,如果现在映射操作已经操作过了512M大小,那么忽略
add r0, r0, #0x00100000 ; (r0) = PTE for next physical page
subs r3, r3, #1
add r1, r1, #1
bgt %b45

ldr r10, =PTs ; (r10) = restore address of 1st level page table

; The page tables and exception vectors are setup.
; Initialize the MMU and turn it on.
mov r1, #1
mcr p15, 0, r1, c3, c0, 0 ; setup access to domain 0
mcr p15, 0, r10, c2, c0, 0

mcr p15, 0, r0, c8, c7, 0 ; flush I+D TLBs //flush所有MMU相关空间
mov r1, #0x0071 ; Enable: MMU //设置打开MMU标志到r1
orr r1, r1, #0x0004 ; Enable the cache //设置打开cache标志到r1

ldr r0, =VirtualStart //取出VirtualStart编译期间得到的虚拟地址值,在0x30038000~0x30038000+eboot_size之间[luther.gliethtp]

cmp r0, #0 ; make sure no stall on "mov pc,r0" below //执行该比较,使r0踏实的读入VirtualStart数值
mcr p15, 0, r1, c1, c0, 0 //将MMU设置标志写入p15协处理器,生效设置[luther.gliethttp]
mov pc, r0 ; & jump to new virtual address //跳转过去,此刻之后MMU已经打开,所有允许操作的虚拟地址都在PLATFORM/SMDK2440A/Src/Inc/oemaddrtab_cfg.inc中定义[luther.gliethttp]
nop

; MMU & caches now enabled.
; (r10) = physcial address of 1st level page table
;

VirtualStart

mov sp, #0x8C000000 //设置栈空间,从这里来看,sp栈顶对应的DDR物理地址为0x30030000,然后向下生长,从boot.bib中
/*
MEMORY
; Name Start Size Type
; ------- -------- -------- ----
ARGS 8c020800 00000800 RESERVED
BINFS 8c021000 00005000 RESERVED
RAM 8c026000 00006000 RAM
STACK 8c02c000 00004000 RESERVED //0x2c000+0x4000=0x30000 保留该16K地址空间用作eboot的sp栈空间
EBOOT 8c038000 00016800 RAMIMAGE
*/
可以看到sp栈空间由.bib保留,位于0x30030000~0x3002c000,栈sp从0x30030000开始向下生长[luther.gliethttp]
add sp, sp, #0x30000 ; arbitrary initial super-page stack pointer
b main //跳转到main主函数,startup.s并没有对c函数定义的全局变量或者static静态变量进行DDR释放性初始化,之所以没有做可以参考我的另一篇文章《CE5.0 - romimage.exe如何填充eboot.bin中的pTOC特殊指针来生成eboot.nb0》