[PATCH v2 1/2] ntdll: Add emulation for UMIP instructions.

[Alexandre Julliard]

Brendan Shanks <bshanks at codeweavers.com> writes:

>> On Nov 13, 2019, at 7:51 PM, Andrew Wesie <awesie at gmail.com> wrote:
>> 
>> On Thu, Nov 14, 2019 at 7:29 AM Brendan Shanks <bshanks at codeweavers.com> wrote:
>>> ...
>>> +/***********************************************************************
>>> + *           INSTR_GetOperandAddr
>>> + *
>>> + * Return the address of an instruction operand (from the mod/rm byte).
>>> + */
>>> +static int INSTR_GetOperandAddr( CONTEXT *context, BYTE *instr, unsigned int instr_len,
>>> +                                 int long_addr, int segprefix, int *len, void **addr )
>>> +{
>>> ...
>>> +    /* FIXME: we assume that all segments have a base of 0 */
>>> +    *addr = (void *)(base + (index << ss));
>>> +    return 1;
>> 
>> Does this FIXME need to be resolved? I don't have an easy way to test
>> UMIP but would the code example below give the correct output with
>> this patch? (This is not based on any real program, so it may be a
>> moot point.)
>> 
>> #include <stdio.h>
>> #include <stdlib.h>
>> 
>> int main()
>> {
>>    unsigned int ldt = 0x4141;
>>    unsigned short ldt2 = 0x4141;
>>    unsigned short ldt3 = 0x4141;
>>    asm volatile ("sldt %0" : "=a" (ldt));
>>    asm volatile ("sldt (%0)" : : "r" (&ldt2));
>> #if defined(__x86_64__) || defined(__amd64__)
>>    asm volatile ("sldt %%gs:(%0)" :: "r" (0x1250));
>>    asm volatile ("movw %%gs:0x1250, %0" : "=r" (ldt3));
>> #elif defined(__i386__)
>>    asm volatile ("sldt %%fs:(%0)" :: "r" (0xBF4));
>>    asm volatile ("movw %%fs:0xBF4, %0" : "=r" (ldt3));
>> #endif
>>    printf("ldt = %x\n", ldt);
>>    printf("ldt2 = %x\n", ldt2);
>>    printf("ldt3 = %x\n", ldt3);
>>    return 0;
>> }
>
> Thanks for spotting that, I tried out your test app and it does crash
> when run with the emulation. I believe you’re right that this usage
> won’t show up in real programs though, especially on x86_64. I’m
> inclined to leave the FIXME unresolved.
>
> I also investigated the history: I copied INSTR_GetOperandAddr() from
> ntoskrnl.exe/instr.c, and the FIXME has been present there since it
> was copied out of krnl386.exe16/instr.c 10 years ago. The original
> copy in krnl386.exe/instr.c still contains the segment code which
> could probably be re-used if necessary.

Outside of 16-bit code, these would in general be GDT selectors, and
there's no easy way of getting their base address. Hopefully it won't
become necessary.

-- 
Alexandre Julliard
julliard at winehq.org