Re: [PATCH 0/2] Update use_goto_tb() in hppa and rx targets

[Ahmed Karaman]

On Tue, May 19, 2020 at 8:01 PM Richard Henderson
<address@hidden> wrote:
>
> On 5/19/20 9:21 AM, Ahmed Karaman wrote:
> > The issue arose because the page crossings check in use_goto_tb()
> > function is required only in the system mode. Checking it in both
> > modes causes an unnecessary overhead in the user mode.
>
> It is not only required in system mode.
>
> You can see failures in user-mode if you modify executable pages, or
> change their permissions with mmap.  Such as if the guest program
> contains a JIT.
>
>
> r~

Hi Mr. Richard,

I've checked how the use_goto_tb() function is implemented in other
targets, and it appears that they do the page crossings check in the
system mode only.

Below is an example from the arm64 target:
-----------------------------------------------------------------------
static inline bool use_goto_tb(DisasContext *s, int n, uint64_t dest)
{
  /* No direct tb linking with singlestep (either QEMU's or the ARM
   * debug architecture kind) or deterministic io
   */
  if (s->base.singlestep_enabled || s->ss_active ||
    (tb_cflags(s->base.tb) & CF_LAST_IO)) {
    return false;
  }

#ifndef CONFIG_USER_ONLY
  /* Only link tbs from inside the same guest page */
  if ((s->base.tb->pc & TARGET_PAGE_MASK)!=(dest & TARGET_PAGE_MASK)) {
    return false;
  }
#endif

  return true;
}
-----------------------------------------------------------------------
Please let me know what you think. Does this mean that there is a bug
in this function for the other targets?
That we have to do the page crossings check in both modes to avoid the
user-mode failures that you have mentioned above?

Regards,
Ahmed Karaman