Arm-linux memory management (4)

The previous article mainly focused on data structures related to specific systems, and left a question at the end:

So, how do these data structures work? In particular, how does the ARM-Linux kernel establish address mappings based on these data structures?

1. Let's start with the initialization of the system. After the system is booted, the CPU enters the main entrance of the kernel, which is the starting point of the code segment stext. In stext, the CPU first reads the CPU model and the machine model from itself, stores the relevant information in the two global variables processor_id and machine_arch_type, and then goes to start_kernel().

asmlinkage void __init start_kernel(void)

{

......

setup_arch(&command_line);

The code of this function depends on the specific CPU model, and it mainly completes the following operations during the system initialization phase:

......

mem_init();

kmem_cache_sizes_init();

.......

｝

For the MMU, a specific page directory is a specific mapping, but for the kernel some additional information is needed. From the kernel's perspective, a specific mapping is described and managed by an mm_struct data structure, among which the kernel's mm_struct data structure init_mm has special importance:

struct mm_struct init_mm = INIT_MM(init_mm);

#define INIT_MM(name)

{

mm_rb:  RB_ROOT,

pgd: swapper_pg_dir,

mm_users:  ATOMIC_INIT(2),

mm_count:  ATOMIC_INIT(1),

mmap_sem:  __RWSEM_INITIALIZER(name.mmap_sem),

page_table_lock: SPIN_LOCK_UNLOCKED, 

mmlist:  LIST_HEAD_INIT(name.mmlist),

}

The correct term should be the mm_struct data structure of the kernel thread, or the mapping of the kernel thread, because this mapping does not contain any user space mappings. All kernel threads in the system share the same mapping, which is init_mm. In particular, the mm_struct data structure of the kernel thread init is init_mm.

Each process or thread has a process control block, which contains a pointer mm pointing to its own mm_struct data structure. In the mm_struct structure, there is a pointer pgd pointing to its first-level page directory.

/*

 * We place the page tables 16K below TEXTADDR.  Therefore, we must make sure

 * that TEXTADDR is correctly set.  Currently, we expect the least significant

 * "short" to be 0x8000, but we could probably relax this restriction to

 * TEXTADDR > PAGE_OFFSET + 0x4000

 *

 * Note that swapper_pg_dir is the virtual address of the page tables, and

 * pgtbl gives us a position-independent reference to these tables.  We can

 * do this because stext == TEXTADDR

 *

 * swapper_pg_dir, pgtbl and krnladr are all closely related.

 */

#if (TEXTADDR & 0xffff) != 0x8000

#error TEXTADDR must start at 0xXXXX8000

#endif

.globl  SYMBOL_NAME(swapper_pg_dir)

.equ SYMBOL_NAME(swapper_pg_dir), TEXTADDR - 0x4000

.macro  pgtbl, reg, rambase

adr reg, stext

sub reg, reg, #0x4000

.endm

/*

/*