Page 1 of 1

long mode segment selectors

Posted: Fri Aug 13, 2010 2:41 am
by a5498828
long mode ignore base and limit of CS/DS/SS/ES.

why i cant load ss to 0 in ring != 0?
when cpu is in compatibility mode, it does matter. but when cpu is in long mode? does default bit somehow matters in long mode? i think long mode use always RSP (not ESP/SP like in compatibility mode).

so my question is why long modew disallow ss = 0?

Re: long mode segment selectors

Posted: Fri Aug 13, 2010 9:30 am
by Owen
Wrong descriptor privilege level bits perhaps?

In any case: My understanding is that the processor uses SS=0 to correctly nest/un-nest mode switches.

Re: long mode segment selectors

Posted: Sun Aug 15, 2010 7:10 am
by a5498828
i dont get it. retf allows ss null unless im returning to non64bit code (wich is obvious) and if im returning to ring3.

WHY?
i can load ss to null only in ring0

explain me it plz.

Re: long mode segment selectors

Posted: Sun Aug 15, 2010 7:23 am
by gerryg400
WHY?
i can load ss to null only in ring0

explain me it plz.
Are you asking why the AMD engineers chose to do it this way ?

Try loading a NULL selector that has an RPL that matches the DPL and CPL. That is for ring 1 use 0x01 as the NULL selector and for ring 2 use 0x02. For ring 3 you cannot load a NULL selector. Even NULL selectors have an RPL.

Re: long mode segment selectors

Posted: Sun Aug 15, 2010 9:37 am
by a5498828
Are you asking why the AMD engineers chose to do it this way ?
yes.
if im to understand it, i have to know what creators had in mind.

Re: long mode segment selectors

Posted: Tue Aug 17, 2010 6:56 am
by a5498828
when i return using iret/retf and ss is null, according to manual cpu wont load new ss.
instead it force it load null with any rpl (wich is impossible with non-null value). is this correct, or my emulator is broken? manual dont say what will be new ss after iret (and that frustrate me, it doesnt exactly say what happens, i have to test to know, everything.).

Re: long mode segment selectors

Posted: Tue Aug 17, 2010 7:48 am
by gerryg400
when i return using iret/retf and ss is null, according to manual cpu wont load new ss.
instead it force it load null with any rpl (wich is impossible with non-null value). is this correct, or my emulator is broken? manual dont say what will be new ss after iret (and that frustrate me, it doesnt exactly say what happens, i have to test to know, everything.).
It sounds correct to me. Read both AMD and Intel book to get more information. The AMD has a little bit more info.
manual dont say what will be new ss after iret
The SS after IRET pops from the stack. If the POPed value selector is NULL it means that the interrupt is nested.