供head.S调用，其中__mmap_switched的b start_kernel跳转到C执行，且永不返回。
跳转到start_kernel时寄存器值：
R0 = cp#15 control register
R1 = machine id
R2 = atags/dtb pointer
R9 = processor ID
stext->__enable_mmu->__turn_mmu_on->ldr pc, __mmap_switched
->__mmap_switched(head-common.S)
->b start_kernel(init/main.c)
http://blog.chinaunix.net/uid-20451980-id-1945242.html
Linux kernel分析（二）
注：本文为Stephen Du原创，转载请注明
这里开始讲解head-common.S的内容。另，我的讲解顺序是按照源码的顺序来进行的而不是按照函数调用的顺序进行，所以读者要注意函数的入口以及返回地址。
这里定义了atag数据的存放地址
14 #define ATAG_CORE 0x54410001
15 #define ATAG_CORE_SIZE ((2*4 + 3*4) >> 2)
这个部分head.S里面开启MMU后执行。此处开始准备执行C代码做准备：主要是text段，初始化数据段的定位！起始kernel本身的段结构跟普通进程大致相同，这也是为什么它被成文宏内核的原因！
__data_loc是kernel的text段的开始
__bss_start是__data_loc的结束也是bss段的开始
从上图看得出，kernel必须将text段跟初始化数据段准备好，对于未初始化数据段跟stack段，heap段都是不需要进行准备的！只需要设置stack指针以及heap起始地址就好！因此到了22行后未出现stack跟heap段的定义！
17 .type __switch_data, %object
18 __switch_data:
19 .long __mmap_switched
20 .long __data_loc @ r4
21 .long __data_start @ r5
22 .long __bss_start @ r6
23 .long _end @ r7
该内存位置存储了处理器的id，也就是r4寄存器的内容，以防后面的代码丢弃r4的内容
24 .long processor_id @ r4
该内存位置存储了machine type的信息也是放置r5的内容被丢弃
25 .long __machine_arch_type @ r5
该内存存储了atag的地址指针，同理防止r6被丢弃了
26 .long __atags_pointer @ r6
27 .long cr_alignment @ r7
这个内存位置放置了kernel的stack & heap的内存位置信息以后，只要将pc pointer指向这里就能执行C代码了！
28 .long init_thread_union + THREAD_START_SP @ sp
29
此前的代码可能在NOR flash中以XIP方式运行（其实Nand flash也可以XIP，只是有点技术障碍），但是kernel的代码不能总在flash内运行，该函数会进行kernel段的搬移以及处理！
39 .type __mmap_switched, %function
40 __mmap_switched:
41 adr r3, __switch_data + 4 // r3 point to __mmap_switched本处的注释错误！r3指向__data_loc！
将代码段，初始化数据段以及未初始化的数据段地址分别加载进入r4~r7
43 ldmia r3!, {r4, r5, r6, r7} // load the function's addr into r4~r7;r3 point to processor_id
此处比较难以理解，r4=__data_loc是物理上段的存储位置（可能在flash中而不是在RAM中）；r5=_data_start是数据在内存的地址，如果二者相等说明已经在RAM中不必做copy，如果不在RAM中则执行copy使之在内存中运行！
44 cmp r4, r5 // Copy data segment if needed __data_start is the destination and __data_loc is the src!
copy相关操作
45 1: cmpne r5, r6 // if __data_start and __data_loc is not the same start the transfer session
46 ldrne fp, [r4], #4
47 strne fp, [r5], #4
48 bne 1b
清空未初始化数据段，为执行C代码扫清障碍
50 mov fp, #0 @ Clear BSS (and zero fp)
51 1: cmp r6, r7
52 strcc fp, [r6],#4
53 bcc 1b
执行保存操作。此处最重要的是sp指针的加载！该语句后sp已经指向了kernel的stack上，执行C代码的条件就绪了（代码段ok，初始化数据段ok，未初始化数据段ok，stack ok）
55 ldmia r3, {r4, r5, r6, r7, sp}
56 str r9, [r4] @ Save processor ID
57 str r1, [r5] @ Save machine type
58 str r2, [r6] @ Save atags pointer
59 bic r4, r0, #CR_A @ Clear 'A' bit
60 stmia r7, {r0, r4} @ Save control register values
61 @Now we will enter the world of C code and the MMU is on now!
62 b entry_for_C_call
63 //b start_kernel @For ARM start_kernel is defined in the init/main.c we will never return from start_kernel at all
144
145 /*
146 * Read processor ID register (CP#15, CR0), and look up in the linker-built
147 * supported processor list. Note that we can't use the absolute addresses
148 * for the __proc_info lists since we aren't running with the MMU on
149 * (and therefore, we are not in the correct address space). We have to
150 * calculate the offset.
151 *
152 * r9 = cpuid
153 * Returns:
154 * r3, r4, r6 corrupted
155 * r5 = proc_info pointer in physical address space
156 * r9 = cpuid (preserved)
157 */
158 .type __lookup_processor_type, %function
159 __lookup_processor_type:
160 adr r3, 3f
161 ldmda r3, {r5 - r7}
162 sub r3, r3, r7 @ get offset between virt&phys
163 add r5, r5, r3 @ convert virt addresses to
164 add r6, r6, r3 @ physical address space
165 1: ldmia r5, {r3, r4} @ value, mask
166 and r4, r4, r9 @ mask wanted bits
167 teq r3, r4
168 beq 2f
169 add r5, r5, #PROC_INFO_SZ @ sizeof(proc_info_list)
170 cmp r5, r6
171 blo 1b
172 mov r5, #0 @ unknown processor
173 2: mov pc, lr
174
175 /*
176 * This provides a C-API version of the above function.
177 */
178 ENTRY(lookup_processor_type)
179 stmfd sp!, {r4 - r7, r9, lr}
180 mov r9, r0
181 bl __lookup_processor_type
182 mov r0, r5
183 ldmfd sp!, {r4 - r7, r9, pc}
184
185 /*
186 * Look in and arch/arm/kernel/arch.[ch] for
187 * more information about the __proc_info and __arch_info structures.
188 */
189 .long __proc_info_begin
190 .long __proc_info_end
191 3: .long .
192 .long __arch_info_begin
193 .long __arch_info_end
194
195 /*
196 * Lookup machine architecture in the linker-build list of architectures.
197 * Note that we can't use the absolute addresses for the __arch_info
198 * lists since we aren't running with the MMU on (and therefore, we are
199 * not in the correct address space). We have to calculate the offset.
200 *
201 * r1 = machine architecture number
202 * Returns:
203 * r3, r4, r6 corrupted
204 * r5 = mach_info pointer in physical address space
205 */
206 .type __lookup_machine_type, %function
207 __lookup_machine_type:
208 adr r3, 3b
209 ldmia r3, {r4, r5, r6}
210 sub r3, r3, r4 @ get offset between virt&phys
211 add r5, r5, r3 @ convert virt addresses to
212 add r6, r6, r3 @ physical address space
213 1: ldr r3, [r5, #MACHINFO_TYPE] @ get machine type
214 teq r3, r1 @ matches loader number?
215 beq 2f @ found
216 add r5, r5, #SIZEOF_MACHINE_DESC @ next machine_desc
217 cmp r5, r6
218 blo 1b
219 mov r5, #0 @ unknown machine
220 2: mov pc, lr
221
222 /*
223 * This provides a C-API version of the above function.
224 */
225 ENTRY(lookup_machine_type)
226 stmfd sp!, {r4 - r6, lr}
227 mov r1, r0
228 bl __lookup_machine_type
229 mov r0, r5
230 ldmfd sp!, {r4 - r6, pc}
231
232 /* Determine validity of the r2 atags pointer. The heuristic requires
233 * that the pointer be aligned, in the first 16k of physical RAM and
234 * that the ATAG_CORE marker is first and present. Future revisions
235 * of this function may be more lenient with the physical address and
236 * may also be able to move the ATAGS block if necessary.
237 *
238 * r8 = machinfo
239 *
240 * Returns:
241 * r2 either valid atags pointer, or zero
242 * r5, r6 corrupted
243 */
244
245 .type __vet_atags, %function
246 __vet_atags:
247 tst r2, #0x3 @ aligned?
248 bne 1f
249
250 ldr r5, [r2, #0] @ is first tag ATAG_CORE?
251 subs r5, r5, #ATAG_CORE_SIZE
252 bne 1f
253 ldr r5, [r2, #4]
254 ldr r6, =ATAG_CORE
255 cmp r5, r6
256 bne 1f
257
258 mov pc, lr @ atag pointer is ok
259
260 1: mov r2, #0
261 mov pc, lr