treatment of SEG prefixes by ndisasm

[ludloff]

nasm/disasm/prefix.c

Line 368 in fbdc885

pf->segover = R_ES;

this tries to do "last SEG wins", right?

cute... but ndisasm should emit what it actually sees, i.e. all SEGs

low priority

more thought needed

[tremalrik]

Has there ever existed an x86 processor that implements anything else than "last SEG wins" for conflicting segment-override prefixes? (from what I've seen in my testing, last-SEG-wins has been consistent not just on the Intel/AMD processors I've tested, but also on a NEC V20, a Zhaoxin KX-U6580 and a DM&P M6117D; if I were to guess at a CPU where this might not be the case, NEC V55 might be a candidate, but for now I don't have one to test on)

[ludloff]

see my recent LKML+binutils msgs for the case of multiple segment prefixes on WBINVD – that is what lead me to file this issue here, because that is when I saw that ndisasm tries to be smart but silently eats the prefixes

again, the key is "emit what you see"

[ludloff]

also see these LKML+binutils msgs for the case of FS+GS in the evolved part of the x86 world