Assert that op is known and that cc_op_live_ is populated.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
LFENCE and SFENCE were introduced with the original SSE instruction set;
marking them incorrectly as cpuid(SSE2) causes failures for CPU models
that lack SSE2, for example pentium3.
Reported-by: Guenter Roeck <linux@roeck-us.net>
Tested-by: Guenter Roeck <linux@roeck-us.net>
Cc: qemu-stable@nongnu.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Group them so that it is easier to figure out which two-byte opcodes to
tackle together.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The gen_cmpxchg8b and gen_cmpxchg16b functions even have the correct
prototype already; the only thing that needs to be done is removing the
gen_lea_modrm() call.
This moves the last LOCK-enabled instructions to the new decoder. It is
now possible to assume that gen_multi0F is called only after checking
that PREFIX_LOCK was not specified.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
There are now relatively few unconverted opcodes in translate.c (there
are 13 of them including 8 for x87), and all of them have the same
format with a mod/rm byte and no immediate. A good next step is
to remove the early bail out to disas_insn_x87/disas_insn_old,
instead giving these legacy translator functions the same prototype
as the other gen_* functions.
To do this, the X86DecodeInsn can be passed down to the places that
used to fetch address bytes from the instruction stream. To make
sure that everything is done cleanly, the CPUX86State* argument is
removed.
As part of the unification, the gen_lea_modrm() name is now free,
so rename gen_load_ea() to gen_lea_modrm(). This is as good a name
and it makes the changes to translate.c easier to review.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Code generation was rewritten; it reuses the same trick to use the
CC_OP_SAR values for cc_op, but it tries to use CC_OP_ADCX or CC_OP_ADCOX
instead of CC_OP_EFLAGS. This is a tiny bit more efficient in the
common case where only CF is checked in the resulting flags.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Rather that enumerating the types that can produce
MMX operands, examine the unit. No functional change.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Link: https://lore.kernel.org/r/20240812025844.58956-3-richard.henderson@linaro.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
In long mode, POP to memory will write a full 64-bit value. However,
the call to gen_writeback() in gen_POP will use MO_32 because the
decoding table is incorrect.
The bug was latent until commit aea49fbb01 ("target/i386: use gen_writeback()
within gen_POP()", 2024-06-08), and then became visible because gen_op_st_v
now receives op->ot instead of the "ot" returned by gen_pop_T0.
Analyzed-by: Clément Chigot <chigot@adacore.com>
Fixes: 5e9e21bcc4 ("target/i386: move 60-BF opcodes to new decoder", 2024-05-07)
Tested-by: Clément Chigot <chigot@adacore.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Use the same flag generation code as SHL and SHR, but use
the existing gen_shiftd_rm_T1 function to compute the result
as well as CC_SRC.
Decoding-wise, SHLD/SHRD by immediate count as a 4 operand
instruction because s->T0 and s->T1 actually occupy three op
slots. The infrastructure used by opcodes in the 0F 3A table
works fine.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
These have very simple generators and no need for complex group
decoding. Apart from LAR/LSL which are simplified to use
gen_op_deposit_reg_v and movcond, the code is generally lifted
from translate.c into the generators.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
SYSENTER is allowed in VM86 mode, but not in real mode. Split the check
so that PE and !VM86 are covered by separate bits.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This is already partly implemented due to VLDMXCSR and VSTMXCSR; finish
the job.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This is a bit more generic, as it can be applied to MPX as well.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Just like X86_ENTRYr, X86_ENTRYwr is easily changed to use only T0.
In this case, the motivation is to use it for the MOV instruction
family. The case when you need to preserve the input value is the
odd one, as it is used basically only for BLS* instructions.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
I am not sure why I made it use T1. It is a bit more symmetric with
respect to X86_ENTRYwr (which uses T0 for the "w"ritten operand
and T1 for the "r"ead operand), but it is also less flexible because it
does not let you apply zextT0/sextT0.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This makes for easier cpu_cc_* setup, and not using set_cc_op()
should come in handy if QEMU ever implements APX.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
CPUX86State argument would only be used to fetch bytes, but that has to be
done before the generator function is called. So remove it, and all
temptation together with it.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Use decode.c's support for intercepts, doing the check in TCG-generated
code rather than the helper. This is cleaner because it allows removing
the eip_addend argument to helper_pause(), even though it adds a bit of
bloat for opcode 0x90's new decoding function.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Use decode.c's support for intercepts, doing the check in TCG-generated
code rather than the helper. This is cleaner because it allows removing
the eip_addend argument to helper_hlt().
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This commit fixes an issue with MOV instructions (0x8C and 0x8E)
involving segment registers; MOV to segment register's source is
16-bit, while MOV from segment register has to explicitly set the
memory operand size to 16 bits. Introduce a new flag
X86_SPECIAL_Op0_Mw to handle this specification correctly.
Signed-off-by: Xinyu Li <lixinyu20s@ict.ac.cn>
Message-ID: <20240602100528.2135717-1-lixinyu20s@ict.ac.cn>
Fixes: 5e9e21bcc4 ("target/i386: move 60-BF opcodes to new decoder", 2024-05-07)
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Features check of CPUID_SSE and CPUID_SSE2 should use cpuid_features,
rather than cpuid_ext_features.
Signed-off-by: Xinyu Li <lixinyu20s@ict.ac.cn>
Reviewed-by: Zhao Liu <zhao1.liu@intel.com>
Message-ID: <20240602100904.2137939-1-lixinyu20s@ict.ac.cn>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The 32-bit AAM/AAD opcodes are using helpers that read and write flags and
env->regs[R_EAX]. Clean them up so that the table correctly includes AX
as a 16-bit input and output.
No real reason to do it to be honest, but they are nice one-output helpers
and it removes the masking of env->regs[R_EAX] that generic load/writeback
code already does.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-ID: <20240522123912.608497-1-pbonzini@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
These are trivial to add, and moving them to the new decoder fixes some
corner cases: raising #UD instead of an instruction fetch page fault for
the undefined opcodes, and incorrectly rejecting 0F 18 prefetches with
register operands (which are treated as reserved NOPs).
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Zhao Liu <zhao1.liu@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Now that a bulk of opcodes go through the new decoder, it is sensible
to do some cleanup. Go immediately through disas_insn_new and only jump
back after parsing the prefixes.
disas_insn() now only contains the three sigsetjmp cases, and they
are more easily managed if they are inlined into i386_tr_translate_insn.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Send all converted opcodes to disas_insn_new() directly from the big
decoding switch statement; once more, the debugging/bisecting logic
disappears.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
A few two-byte opcodes are simple extensions of existing one-byte opcodes;
they are easy to decode and need no change to emit.c.inc. Port them to
the new decoder.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Move long-displacement Jcc, SETcc and CMOVcc to the new decoder.
While filling in the tables makes the code seem longer, the new
emitters are all just one line of code.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The shift instructions are rewritten instead of reusing code from the old
decoder. Rotates use CC_OP_ADCOX more extensively and generally rely
more on the optimizer, so that the code generators are shared between
the immediate-count and variable-count cases.
In particular, this makes gen_RCL and gen_RCR pretty efficient for the
count == 1 case, which becomes (apart from a few extra movs) something like:
(compute_cc_all if needed)
// save old value for OF calculation
mov cc_src2, T0
// the bulk of RCL is just this!
deposit T0, cc_src, T0, 1, TARGET_LONG_BITS - 1
// compute carry
shr cc_dst, cc_src2, length - 1
and cc_dst, cc_dst, 1
// compute overflow
xor cc_src2, cc_src2, T0
extract cc_src2, cc_src2, length - 1, 1
32-bit MUL and IMUL are also slightly more efficient on 64-bit hosts.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Compared to the old decoder, the main differences in translation
are for the little-used ARPL instruction. IMUL is adjusted a bit
to share more code to produce flags, but is otherwise very similar.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
While keeping decode->immediate for convenience and for 4-operand instructions,
store the immediate in X86DecodedOp as well. This enables instructions
with more than one immediate such as ENTER. It can also be used for far
calls and jumps.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The main difficulty here is that a page fault when writing to the destination
must not overwrite the flags. Therefore, the flags computation must be
inlined instead of using gen_jcc1*.
For simplicity, I am using an unconditional cmpxchg operation, that becomes
a NOP if the comparison fails.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
ALU instructions can write to both memory and flags. If the CC_SRC*
and CC_DST locations have been written already when a memory access
causes a fault, the value in CC_SRC* and CC_DST might be interpreted
with the wrong CC_OP (the one that is in effect before the instruction.
Besides just using the wrong result for the flags, something like
subtracting -1 can have disastrous effects if the current CC_OP is
CC_OP_EFLAGS: this is because QEMU does not expect bits outside the ALU
flags to be set in CC_SRC, and env->eflags can end up set to all-ones.
In the case of the attached testcase, this sets IOPL to 3 and would
cause an assertion failure if SUB is moved to the new decoder.
This mechanism is not really needed for BMI instructions, which can
only write to a register, but put it to use anyway for cleanliness.
In the case of BZHI, the code has to be modified slightly to ensure
that decode->cc_src is written, otherwise the new assertions trigger.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
gen_jcc() has been changed to accept a relative offset since the
new decoder was written. Adjust the J operand, which is meant
to be used with jump instructions such as gen_jcc(), to not
include the program counter and to not truncate the result, as
both operations are now performed by common code.
The result is that J is now the same as the I operand.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
decode->mem is only used if one operand has has_ea == true. String
operations will not use decode->mem and will load A0 on their own, because
they are the only case of two memory operands in a single instruction.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Usually the registers are just moved into s->T0 without much care for
their operand size. However, in some cases we can get more efficient
code if the operand fetching logic syncs with the emission function
on what is nicer.
All the current uses are mostly demonstrative and only reduce the code
in the emission functions, because the instructions do not support
memory operands. However the logic is generic and applies to several
more instructions such as MOVSXD (aka movslq), one-byte shift
instructions, multiplications, XLAT, and indirect calls/jumps.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
X86_SPECIAL_ZExtOp0 and X86_SPECIAL_ZExtOp2 are poorly named; they are a hack
that is needed by scalar insertion and extraction instructions, and not really
related to zero extension: for PEXTR the zero extension is done by the generation
functions, for PINSR the high bits are not used at all and in fact are *not*
filled with zeroes when loaded into s->T1.
Rename the values to match the effect described in the manual, and explain
better in the comments.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The previous check erroneously allowed CMP to be modified with LOCK.
Instead, tag explicitly the instructions that do support LOCK.
Acked-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
