Re: [PATCH 2/6] x86/alternative: Walk all replacements in debug builds

[Jan Beulich <jbeulich@xxxxxxxx>]

On 25.09.2024 23:15, Andrew Cooper wrote:
> On 23/04/2024 3:44 pm, Jan Beulich wrote:
>> On 22.04.2024 20:14, Andrew Cooper wrote:
>>> +                if ( res.rel_type == REL_TYPE_d8 )
>>> +                {
>>> +                    int8_t *d8 = res.rel;
>>> +                    void *target = ip + res.len + *d8;
>>> +
>>> +                    if ( target < repl || target > end )
>>> +                    {
>>> +                        printk("Alternative for %ps [%*ph]\n",
>>> +                               ALT_ORIG_PTR(a), a->repl_len, repl);
>>> +                        panic("'JMP/Jcc disp8' at +%u leaves alternative 
>>> block\n",
>>> +                              (unsigned int)(unsigned long)(ip - repl));
>>> +                    }
>>> +                }
>> Why's Disp8 more important to check than Disp32? A bad CALL in a
>> replacement can't possibly be encoded with Disp8, and both JMP and Jcc
>> are also more likely to be encoded with Disp32 when their target isn't
>> in the same blob (but e.g. in a different section).
> 
> Whatever the likelihood of them existing, Disp8's cannot possibly be
> correct if they cross the boundary of the replacement.  Checking for
> them has the side effect of running decode_lite() over all replacements.
> 
> Disp32's do exist in both external and internal forms (retpoline), and
> the point of this series is to make all external cases usable. 

Okay, fine then.

> Therefore, there are no invalid cases.

There definitely are: Any pointing outside of the present replacement
block, into another replacement block. Which can in principle happen if
a label was used wrongly. Anything pointing outside the block really
needs to be covered by logic adjusting the displacement when the
alternative is being put in place.

Jan