[lkml]   [2017]   [Sep]   [28]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [PATCH v8 12/28] x86/insn-eval: Add utility functions to get segment selector
On Wed, Sep 27, 2017 at 03:32:26PM -0700, Ricardo Neri wrote:
> The idea is that get_overridden_seg_reg() would implement the logic you
> just described. It would return return INAT_SEG_REG_DEFAULT/IGNORE when
> segment override prefixes are not allowed (i.e., valid insn with
> operand rDI and string instruction; and rIP) or needed (i.e., long
> mode, except if there are override prefixes for FS or GS); or
> INAT_SEG_REG_[CSDEFG]S otherwise.

Ok, lemme see if we're talking the same thing. Your diff is linewrapped
so parsing that is hard.

Do this

if (regoff == offsetof(struct pt_regs, ip)) {
if (user_64bit_mode(regs))

and all the other checking *before* you do insn_init(). Because you have
crazy stuff like:

if (seg_reg == INAT_SEG_REG_IGNORE)
return seg_reg;

which shortcuts those functions and is simply clumsy and complicates
following the code. The mere fact that you have to call the function
"get_overridden_seg_reg_if_any_or_needed()" already tells you that that
function is doing too many things at once.

When the function is called get_segment_register() then it should do
only that. And all the checking is done before or in wrappers.

IOW, all the rIP checking and early return down the
insn_get_seg_base() -> resolve_seg_register() -> .. should be done

*Then* you do insn_init() and hand it down to insn_get_seg_base() and
from now on you have a proper insn pointer which you hand around and
check for NULL only once, on function entry.

Then your code flow is much simpler: first you take care of the case
where rIP doesn't do segment overrides and all the other cases are
handled by the normal path, with a proper struct insn.

Makes more sense?


SUSE Linux GmbH, GF: Felix Imendörffer, Jane Smithard, Graham Norton, HRB 21284 (AG Nürnberg)

 \ /
  Last update: 2017-09-28 11:37    [W:0.070 / U:12.816 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site