kernel loader issue in bochs

issue: 使用 bochs 2.6, kernel loader 無法載入 kernel

使用自動變數
u16 offset, next_cluster, cur_cluster;
在 bochs 2.6 下 kernel loader 無法載入 kernel

使用 static 變數 + gcc 4.4
static u16 offset, next_cluster, cur_cluster;
就沒問題 ,但若使用 gcc 4.7 compile, 還是一樣的問題。
很值得看看, 應該是 stack 這邊出了狀況, 看起來又不太像, 慘 ...
每一次的問題都是一次很好的學習機會。

git@github.com:descent/simple_os.git

issue:
git commit: c7229045fd1fb93edff72f2e816eedce6c408c30

fixed only under gcc 4.4
git commit: 7138421dd7c683734cdf774322821b943cc09b78


fixed under gcc 4.7

把 volatile u8 *buff 改成 u16 buff, 讓 "b"(buff)

 __asm__ __volatile__ 
 (
 @@ -349,6 +354,10 @@ int __REGPARM read_sector(volatile u8 *buff, u8 sector_no, u8 track_no, u8 head_
  :"a"(0x0200|blocks), "b"(buff), "c"(track_no << 8 | sector_no), "d"(head_no << 8 | disk_no)
 ); 

直接吃 u16 的值, 可以修復這問題, 可是還沒找到 root cause。

git diff f78742eea8d145fcbb8b302bd1f7491d16d783b8 ff8d23e70841af5cc48b7131aca6a04d5e2233ce

 diff --git a/kernel_loader/kernel_loader.c b/kernel_loader/kernel_loader.c
 index 3a071ca..464da43 100644
 --- a/kernel_loader/kernel_loader.c
 +++ b/kernel_loader/kernel_loader.c
 @@ -314,7 +314,7 @@ int __REGPARM __NOINLINE get_drive_params(drive_params_t    *p, unsigned char
  // not 0: fail
  // if using the function read floppy fail, suggest reset floppy disk,
  // than try twice again.
 -int __REGPARM read_sector(volatile u8 *buff, u8 sector_no, u8 track_no, u8 head_no, u8 disk_no, u8 blocks)
 +int __REGPARM read_sector(u16 buff, u8 sector_no, u8 track_no, u8 head_no, u8 disk_no, u8 blocks)
  {
    //bios_wait_key();
  #if 0
 @@ -333,12 +333,17 @@ int __REGPARM read_sector(volatile u8 *buff, u8 sector_no, u8 track_no, u8 head_
  //  __asm__ __volatile__("movb $2, %ah\n"); 
  //  __asm__ __volatile__("movb $1, %al\n"); 
  #if 1
 -    //BOCHS_MB
    // read sector to %es:%bx, if %bx is more than 64k, need change %es
    // to next 64k beginning address
    // ref: http://dc0d32.blogspot.tw/2010/06/real-mode-in-c-with-gcc-writing.html
    u16 num_blocks_transferred = 0;
    u8 failed = 0;
 +  //BOCHS_MB
 +  //__asm__ __volatile__("xchg %bx, %bx");
 +  __asm__ __volatile__("push %ax");
 +  __asm__ __volatile__("push %bx");
 +  __asm__ __volatile__("push %cx");
 +  __asm__ __volatile__("push %dx");
  
    __asm__ __volatile__ 
      (
 @@ -349,6 +354,10 @@ int __REGPARM read_sector(volatile u8 *buff, u8 sector_no, u8 track_no, u8 head_
        :"a"(0x0200|blocks), "b"(buff), "c"(track_no << 8 | sector_no), "d"(head_no << 8 | disk_no)
      ); 
  #endif
 +  __asm__ __volatile__("pop %dx");
 +  __asm__ __volatile__("pop %cx");
 +  __asm__ __volatile__("pop %bx");
 +  __asm__ __volatile__("pop %ax");
    u8 ret_status = (num_blocks_transferred >> 8);
    #ifdef MORE_ERR_MSG
    NAME_VALUE(num_blocks_transferred)
 @@ -524,7 +533,7 @@ int load_file_to_ram(int begin_cluster, int fat, u16 org_es, u16 es)
  {
    int r;
    int r_sec = begin_cluster - 2 + bpb.root_dir_occupy_sector + bpb.root_dir_start_sector;
 -  volatile u8 *buff = (u8*)LOAD_KERNEL_OFFSET;
 +  u16 buff = LOAD_KERNEL_OFFSET;
  
    print_num(begin_cluster, "begin_cluster");
    print_num(r_sec, "cluster sector no");
 @@ -568,7 +577,7 @@ int load_file_to_ram(int begin_cluster, int fat, u16 org_es, u16 es)
  #endif
          if (read_sector_count == 65536/512)
          {
 -          buff = (u8*)LOAD_KERNEL_OFFSET;
 +          buff = LOAD_KERNEL_OFFSET;
  #ifdef MORE_ERR_MSG
            print("\r\nmore than 64Kb\r\n");
  #endif
 @@ -664,7 +673,7 @@ void start_c()
    head_no=0; // dh
    disk_no=0; // dl
  #endif
 -  *buff = 0x1;
 +  //*buff = 0x1;
  
    int r = read_sector(buff, sector_no, track_no, head_no, disk_no, 1);
    init_bpb(buff, &bpb);

 +  __asm__ __volatile__("push %eax");
 +  __asm__ __volatile__("push %ebx");
 +  __asm__ __volatile__("push %ecx");
 +  __asm__ __volatile__("push %edx");
  
    __asm__ __volatile__ 
      (
 @@ -349,6 +354,10 @@ int __REGPARM read_sector(volatile u8 *buff, u8 sector_no, u8 track_no, u8 head_
        :"a"(0x0200|blocks), "b"(buff), "c"(track_no << 8 | sector_no), "d"(head_no << 8 | disk_no)
      ); 
  #endif
 +  __asm__ __volatile__("pop %edx");
 +  __asm__ __volatile__("pop %ecx");
 +  __asm__ __volatile__("pop %ebx");
 +  __asm__ __volatile__("pop %eax");

似乎和 static 變數無關, 改成 push/pop 32 bit register 比較正常了。

git commit 03532548e7d57d660033e648c793a657ebe596de

似乎還得補上 push/pop %esi, %edi

 +  __asm__ __volatile__("push %esi");
 +  __asm__ __volatile__("push %edi");
    __asm__ __volatile__("push %eax");
    __asm__ __volatile__("push %ebx");
    __asm__ __volatile__("push %ecx");
 @@ -366,12 +369,15 @@ int __REGPARM read_sector(u16 buff, u8 sector_no, u8 track_no, u8 head_n
    __asm__ __volatile__("pop %ecx");
    __asm__ __volatile__("pop %ebx");
    __asm__ __volatile__("pop %eax");
 +  __asm__ __volatile__("pop %edi");
 +  __asm__ __volatile__("pop %esi");

git commit 38a3df774b25348aceffb479d731a5579641f595

我對在 c 語言使用 inline assembly 還不熟悉, 可能踩到什麼陷阱了。會加上 push/pop 這些暫存器只是我反組譯看到 c compiler 會用到這些暫存器, 所以加上。