Re: [Qemu-devel] [PATCH 0/8] target-i386: Implement debug extensions

[Eduardo Habkost]

On Mon, Sep 21, 2015 at 02:05:52PM +0200, Paolo Bonzini wrote:
> On 15/09/2015 20:45, Richard Henderson wrote:
> > Best guess, since I can't find any code that actually uses them.
> > Linux actively turns them off at boot...
> 
> I've sent a kvm-unit-tests patch to test debug extensions.  It shows
> that debug extensions work, but the following needs to be squashed in
> patch 4:
> 
> diff --git a/target-i386/bpt_helper.c b/target-i386/bpt_helper.c
> index c258598..b24e446 100644
> --- a/target-i386/bpt_helper.c
> +++ b/target-i386/bpt_helper.c
> @@ -134,14 +134,14 @@ void cpu_x86_update_dr7(CPUX86State *env, uint32_t 
> new_dr7)


(I've expanded the diff context below, to make review easier)

>      /* If nothing is changing except the global/local enable bits,
>         then we can make the change more efficient.  */
>      if (((old_dr7 ^ new_dr7) & ~0xff) == 0) {
>          /* Fold the global and local enable bits together into the
>             global fields, then xor to show which registers have
>             changed collective enable state.  */
>          int mod = ((old_dr7 | old_dr7 * 2) ^ (new_dr7 | new_dr7 * 2)) & 0xff;
>  
>          for (i = 0; i < DR7_MAX_BP; i++) {
> -            if (mod & (2 << i * 2)) {
> -                /* We know that register i has changed enable state;
> -                   recheck what that state should be and apply.  */
> -                if (hw_breakpoint_enabled(new_dr7, i)) {
> -                    iobpt |= hw_breakpoint_insert(env, i);
> -                } else {
> -                    hw_breakpoint_remove(env, i);
> -                }
> +            if ((mod & (2 << i * 2)) && !hw_breakpoint_enabled(new_dr7, i)) {
> +                hw_breakpoint_remove(env, i);
> +            }
> +        }
> +        env->dr[7] = new_dr7 | DR7_FIXED_1;

Why is DR7_FIXED_1 forced here, but not at the other branch below?

Also, the previous code didn't force DR7_FIXED_1, so shouldn't we implement
this fix in a separate patch instead of squashing it into patch 4/8?

> +        for (i = 0; i < DR7_MAX_BP; i++) {
> +            if (mod & (2 << i * 2) && hw_breakpoint_enabled(new_dr7, i)) {
> +                iobpt |= hw_breakpoint_insert(env, i);
>              } else if (hw_breakpoint_type(new_dr7, i) == DR7_TYPE_IO_RW
>                         && hw_breakpoint_enabled(new_dr7, i)) {
>                  iobpt |= HF_IOBPT_MASK;
>              }
>          }
>      } else {
>          for (i = 0; i < DR7_MAX_BP; i++) {
>              hw_breakpoint_remove(env, i);
>          }
>          env->dr[7] = new_dr7;
>          for (i = 0; i < DR7_MAX_BP; i++) {
>              iobpt |= hw_breakpoint_insert(env, i);
>          }
>      }

-- 
Eduardo