lkml.org 
[lkml]   [2008]   [Oct]   [7]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [RFC Patch 2/9] x86 architecture implementation of Hardware Breakpoint interfaces
On Tue, Oct 07, 2008 at 11:36:30AM -0400, Alan Stern wrote:
> On Tue, 7 Oct 2008, K.Prasad wrote:
>
> > This patch introduces two new files named hw_breakpoint.[ch] inside x86 specific
> > directories. They contain functions which help validate and serve requests for
> > using Hardware Breakpoint registers on x86 processors.
>
> > --- /dev/null
> > +++ linux-bkpt-lkml-27-rc9/arch/x86/kernel/hw_breakpoint.c
> > @@ -0,0 +1,684 @@
>
> ...
> > +int pre_handler_allowed(unsigned type)
> > +{
> > + if (type == HW_BREAKPOINT_EXECUTE)
> > + return 1;
> > + else
> > + return -EINVAL;
> > +}
>
> The routine's name should match the name in the header file. "allowed"
> isn't right: You're _allowed_ to have pre_handlers -- they just won't
> get invoked. "supported" would be better.

pre_handler_supported() definitely sounds better. I will change them.

>
> Also, the comment in the header file should explain the meaning of the
> return value -- you should return 0 if a pre_handler is not supported,
> not -EINVAL. Better yet, define the function (both here and in the
> header file) as returning bool rather than int.
>

I will change them to boolean.

> > +
> > +int post_handler_allowed(unsigned type)
> > +{
> > + /* We can have a post handler for all types of breakpoints */
> > + return 1;
> > +}
>
> Same comments as above.
>
> Also, in this initial version I would prefer to avoid the complications
> of single-stepping. It can always be added later. So for now, the x86
> implementation should not support post_handlers for execution
> breakpoints.
>

There's been a perceivable inclination to let the user learn the
limitations/features of the underlying processor's breakpointing ability
(since the previous email mail thread on this topic) and the routines
pre_ and post_handler_allowed() are just a step towards that.

I can nullify the post_handler for x86-instruction breakpoint for now,
but it wouldn't simplify things very extensively (but for a few lines of
code in hw_breakpoint_handler() and the flag 'sstep_reason'). It also
benefits the code by bringing an understanding that there can be
multiple users of processor single-stepping (and therefore the need to
de-multiplex the exception and invoke the appropriate handler).

Left to me, I would like to retain the post_handler routine, unless you
strongly feel otherwise.

> ...
> > +/*
> > + * Validate the arch-specific HW Breakpoint register settings
> > + */
> > +static int arch_validate_hwbkpt_settings(struct hw_breakpoint *bp,
> > + unsigned long address, unsigned len, unsigned int type,
> > + unsigned int *align)
>
> Why did you move this routine into the arch-specific code?
>
> ...
> > +/*
> > + * Handle debug exception notifications.
> > + */
> > +
> > +static void switch_to_none_hw_breakpoint(void);
> > +struct hw_breakpoint *last_hit_bp;
> > +struct thread_hw_breakpoint *last_hit_thbi;
>
> Shouldn't these variables be static? Although if they're needed only for
> single-stepping, they can be removed entirely for now...
>
Agreed. Will make them static.

Thanks,
K.Prasad



\
 
 \ /
  Last update: 2008-10-07 19:25    [from the cache]
©2003-2011 Jasper Spaans