Re: [Qemu-devel] [PATCH] fcntl64 fix

[Thiemo Seufer]

Kirill A. Shutemov wrote:
> TARGET_F_*64 should be used instead of F_*64, because on 64-bit host
> systems F_GETLK == F_GETLK64(same for SETLK and SETLKW), so we cannot
> determinate if it's a long lock or not on a target 32-bit system.
> Patch in the attachment.
> 
> P.S. Please, review my privious patches, which I have added description
> recently. Or should I repost it?
> 

> diff -uNr qemu-0.9.0.cvs20070304.orig/linux-user/syscall.c 
> qemu-0.9.0.cvs20070304/linux-user/syscall.c
> --- qemu-0.9.0.cvs20070304.orig/linux-user/syscall.c  2007-03-09 20:08:59 
> +0200
> +++ qemu-0.9.0.cvs20070304/linux-user/syscall.c       2007-03-09 20:09:54 
> +0200
> @@ -4063,7 +4063,7 @@
>  #endif
>  
>          switch(arg2) {
> -        case F_GETLK64:
> +        case TARGET_F_GETLK64:
>              ret = get_errno(fcntl(arg1, arg2, &fl));

This changes the bug from checking the wrong flag to (potentially)
passing down the wrong flag...

>           if (ret == 0) {
>  #ifdef TARGET_ARM
> @@ -4089,8 +4089,8 @@
>           }
>           break;
>  
> -        case F_SETLK64:
> -        case F_SETLKW64:
> +        case TARGET_F_SETLK64:
> +        case TARGET_F_SETLKW64:
>  #ifdef TARGET_ARM
>              if (((CPUARMState *)cpu_env)->eabi) {
>                  lock_user_struct(target_efl, arg3, 1);

Likewise here. We should always check TARGET_* flags and pass down the
corresponding host flag.


Thiemo