Re: [PATCH 0/1] (Was: Linux 3.11-rc4)

[Frederic Weisbecker <fweisbec@gmail.com>]

On Thu, Aug 08, 2013 at 08:15:21PM +0200, Oleg Nesterov wrote:
> On 08/08, Frederic Weisbecker wrote:
> >
> > I'm all for fixing this. May be we can start by backporting a patch that
> > ignores the value of gen_len for instruction breakpoints in x86?
> 
> Or perhaps we can start with the something like below.

(commented on the diff below)

> 
> But probably we should move "attr.bp_len == HW_BREAKPOINT_LEN_1" check
> from arch_build_bp_info() to its caller, arch_validate_hwbkpt_settings().
> 
> Because:
> 
> > But this bp_len
> > should rather be used for range breakpoints on archs that support it.
> 
> Yes, exactly, and we already have the patches for amd, so bp->len can
> be actually != 1 but currently we can't support because it is checked
> in arch_build_bp_info().

Hmm, but how moving that to arch_validate_hwbkpt_seetings() would solve
the issue?

> 
> Oleg.
> 
> --- x/arch/x86/kernel/hw_breakpoint.c
> +++ x/arch/x86/kernel/hw_breakpoint.c
> @@ -208,19 +208,16 @@ int arch_bp_generic_fields(int x86_len, 
>  {
>  	/* Type */
>  	switch (x86_type) {
> -	case X86_BREAKPOINT_EXECUTE:
> -		if (x86_len != X86_BREAKPOINT_LEN_X)
> -			return -EINVAL;
> -
> -		*gen_type = HW_BREAKPOINT_X;
> -		*gen_len = sizeof(long);
> -		return 0;
>  	case X86_BREAKPOINT_WRITE:
>  		*gen_type = HW_BREAKPOINT_W;
>  		break;
>  	case X86_BREAKPOINT_RW:
>  		*gen_type = HW_BREAKPOINT_W | HW_BREAKPOINT_R;
>  		break;
> +	case X86_BREAKPOINT_EXECUTE:
> +		*gen_type = HW_BREAKPOINT_X;
> +		if (x86_len == X86_BREAKPOINT_LEN_1)
> +
>			break;
>  	default:
>  		return -EINVAL;
>  	}
> @@ -265,15 +262,11 @@ static int arch_build_bp_info(struct per
>  		break;
>  	case HW_BREAKPOINT_X:
>  		info->type = X86_BREAKPOINT_EXECUTE;
> -		/*
> -		 * x86 inst breakpoints need to have a specific undefined len.
> -		 * But we still need to check userspace is not trying to setup
> -		 * an unsupported length, to get a range breakpoint for example.
> -		 */
> -		if (bp->attr.bp_len == sizeof(long)) {
> -			info->len = X86_BREAKPOINT_LEN_X;
> -			return 0;
> -		}
> +		/* until we change tools/perf */
> +		if (bp->attr.bp_len == sizeof(long))
> +			bp->attr.bp_len = HW_BREAKPOINT_LEN_1;

Too bad we need to keep that compatibility around. Do you think this could be
a problem for AMD range breakpoints?

We can also fix the tools, then may be we'll be able to remove the kernel hack
compatibility in a few years.

Oh I need to check other archs as well.

thanks.

> +		if (bp->attr.bp_len == HW_BREAKPOINT_LEN_1)
> +			break;
>  	default:
>  		return -EINVAL;
>  	}
>