The addrsize should be ignored for PUSHF/POPF instructions. This was potentially used before the LONGMODE_ON context bit was introduced to distinguish the 64-bit behavior from the 32-bit behavior. Since addrsize can be manipulated using the address size override prefix, it can cause the wrong PUSHF/POPF size to be decoded. The LONGMODE_ON constraint is now sufficient, so this PR simply removes the unnecessary addrsize constraint.
The LONGMODE_OFF constructors have been left alone in this PR, although they technically should also ignore address size overrides. With LONGMODE_OFF, the address size is used to determine whether the segment pcodeop is applied during address translation, however segmentation should still be applied in 32-bit protected mode, but since most users are likely to be working with 32-bit programs where segment bases are all zero, the current behavior probably behaves better than injecting segment ops everywhere.
(It could make sense to introduce pseudo registers like CS.base/DS.base/etc... that are assumed to be zero but are user overridable, like the way the direction flag is handled)
The
addrsizeshould be ignored for PUSHF/POPF instructions. This was potentially used before theLONGMODE_ONcontext bit was introduced to distinguish the 64-bit behavior from the 32-bit behavior. Sinceaddrsizecan be manipulated using the address size override prefix, it can cause the wrong PUSHF/POPF size to be decoded. TheLONGMODE_ONconstraint is now sufficient, so this PR simply removes the unnecessaryaddrsizeconstraint.e.g.,
PUSHFwithRSP=0x1002,ZF=1mem[0x1000] = 4201,RSP=0x1000}x86:LE:64:default(Existing):"PUSHFQ"{mem[0xffa] = 4000000000000000,RSP=0x0ffa}x86:LE:64:default(This patch):"PUSHF"{mem[0x1000] = 4000,RSP=0x1000}(Note: on the hardware reference, extra bits that are set correspond to
EFLAGS.res1(bit 1) andEFLAGS.TF(bit 8))POPFwithRSP=0x1000,mem[0x1000] = 4000ZF=1,RSP=0x1002}x86:LE:64:default(Existing): Invalid Instructionx86:LE:64:default(This patch):"PUSHF"{ZF=1,RSP=0x1002}The
LONGMODE_OFFconstructors have been left alone in this PR, although they technically should also ignore address size overrides. WithLONGMODE_OFF, the address size is used to determine whether thesegmentpcodeop is applied during address translation, however segmentation should still be applied in 32-bit protected mode, but since most users are likely to be working with 32-bit programs where segment bases are all zero, the current behavior probably behaves better than injectingsegmentops everywhere.(It could make sense to introduce pseudo registers like
CS.base/DS.base/etc... that are assumed to be zero but are user overridable, like the way the direction flag is handled)