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>
Instructions in VEX exception class 6 generally look at the value of
VEX.W. Note that the manual places some instructions incorrectly in
class 4, for example VPERMQ which has no non-VEX encoding and no legacy
SSE analogue. AMD does a mess of its own, as documented in the comment
that this patch adds.
Most of them are checked for VEX.W=0, and are listed in the manual
(though with an omission) in table 2-16; VPERMQ and VPERMPD check for
VEX.W=1, which is only listed in the instruction description. Others,
such as VPSRLV, VPSLLV and the FMA3 instructions, use VEX.W to switch
between a 32-bit and 64-bit operation.
Fix more of the class 4/class 6 mismatches, and implement the check for
VEX.W in TCG.
Acked-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
In preparation for adding more similar checks, move the VEX.L=0 check
and several X86_SPECIAL_* checks to a new field, where each bit represent
a common check on unused bits, or a restriction on the processor mode.
Likewise, many SVM intercepts can be checked during the decoding phase,
the main exception being the selective CR0 write, MSR and IOIO intercepts.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The implementation was validated with OpenSSL and with the test vectors in
https://github.com/rust-lang/stdarch/blob/master/crates/core_arch/src/x86/sha.rs.
The instructions provide a ~25% improvement on hashing a 64 MiB file:
runtime goes down from 1.8 seconds to 1.4 seconds; instruction count on
the host goes down from 5.8 billion to 4.8 billion with slightly better
IPC too. Good job Intel. ;)
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Allow the name 'cpu_env' to be used for something else.
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
CVTPS2PD only loads a half-register for memory, unlike the other
operations under 0x0F 0x5A. "Unpack" the group into separate
emission functions instead of using gen_unary_fp_sse.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
CVTPS2PD only loads a half-register for memory, like CVTPH2PS. It can
reuse the "ph" packed half-precision size to load a half-register,
but rename it to "xh" because it is now a variation of "x" (it is not
used only for half-precision values).
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
When CR0.TS=1, execution of x87 FPU, MMX, and some SSE instructions will
cause a Device Not Available (DNA) exception (#NM). System software uses
this exception event to lazily context switch FPU state.
Before this patch, enter_mmx helpers may be generated just before #NM
generation, prematurely resetting FPU state before the guest has a
chance to save it.
Signed-off-by: Matt Borgerson <contact@mborgerson.com>
Message-ID: <CADc=-s5F10muEhLs4f3mxqsEPAHWj0XFfOC2sfFMVHrk9fcpMg@mail.gmail.com>
Cc: qemu-stable@nongnu.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Before this change, MOVNTPS and MOVNTPD were labeled as Exception Class
4 (only requiring alignment for legacy SSE instructions). This changes
them to Exception Class 1 (always requiring memory alignment), as
documented in the Intel manual.
Message-Id: <20230501111428.95998-3-ricky@rzhou.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Fix the exception classes for some SSE/AVX instructions to match what is
documented in the Intel manual.
These changes are expected to have no functional effect on the behavior
that qemu implements (primarily >= 16-byte memory alignment checks). For
instance, since qemu does not implement the AC flag, there is no
difference in behavior between Exception Classes 4 and 5 for
instructions where the SSE version only takes <16 byte memory operands.
Message-Id: <20230501111428.95998-2-ricky@rzhou.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Adds some comments describing what instructions correspond to decoding
table entries and fixes some existing comments which named the wrong
instruction.
Message-Id: <20230501111428.95998-1-ricky@rzhou.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Compared to other SSE instructions, VUCOMISx and VCOMISx are different:
the single and double precision versions are distinguished through a
prefix, however they use no-prefix and 0x66 for SS and SD respectively.
Scalar values usually are associated with 0xF2 and 0xF3.
Because of these, they incorrectly perform a 128-bit memory load instead
of a 32- or 64-bit load. Fix this by writing a custom decoding function.
I tested that the reproducer is fixed and the test-avx output does not
change.
Reported-by: Gabriele Svelto <gsvelto@mozilla.com>
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1637
Fixes: f8d19eec0d ("target/i386: reimplement 0x0f 0x28-0x2f, add AVX", 2022-10-18)
Cc: qemu-stable@nongnu.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Translators are no longer required to free tcg temporaries.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
VRCPSS, VRSQRTSS and VCVTSx2Sx have a 32-bit or 64-bit memory operand,
which is represented in the decoding tables by X86_VEX_REPScalar. Add it
to the tables, and make validate_vex() handle the case of an instruction
that is in exception type 4 without the REP prefix and exception type 5
with it; this is the cas of VRCP and VRSQRT.
Reported-by: yongwoo <https://gitlab.com/yongwoo36>
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1377
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
MMX state is saved/restored by FSAVE/FRSTOR so the instructions are
not illegal opcodes even if CR4.OSFXSR=0. Make sure that validate_vex
takes into account the prefix and only checks HF_OSFXSR_MASK in the
presence of an SSE instruction.
Fixes: 20581aadec ("target/i386: validate VEX prefixes via the instructions' exception classes", 2022-10-18)
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1350
Reported-by: Helge Konetzka (@hejko on gitlab.com)
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The only issue with FMA instructions is that there are _a lot_ of them (30
opcodes, each of which comes in up to 4 versions depending on VEX.W and
VEX.L; a total of 96 possibilities). However, they can be implement with
only 6 helpers, two for scalar operations and four for packed operations.
(Scalar versions do not do any merging; they only affect the bottom 32
or 64 bits of the output operand. Therefore, there is no separate XMM
and YMM of the scalar helpers).
First, we can reduce the number of helpers to one third by passing four
operands (one output and three inputs); the reordering of which operands
go to the multiply and which go to the add is done in emit.c.
Second, the different instructions also dispatch to the same softfloat
function, so the flags for float32_muladd and float64_muladd are passed
in the helper as int arguments, with a little extra complication to
handle FMADDSUB and FMSUBADD.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
F16C only consists of two instructions, which are a bit peculiar
nevertheless.
First, they access only the low half of an YMM or XMM register for the
packed-half operand; the exact size still depends on the VEX.L flag.
This is similar to the existing avx_movx flag, but not exactly because
avx_movx is hardcoded to affect operand 2. To this end I added a "ph"
format name; it's possible to reuse this approach for the VPMOVSX and
VPMOVZX instructions, though that would also require adding two more
formats for the low-quarter and low-eighth of an operand.
Second, VCVTPS2PH is somewhat weird because it *stores* the result of
the instruction into memory rather than loading it.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
With all SSE (and AVX!) instructions now implemented in disas_insn_new,
it's possible to remove gen_sse, as well as the helpers for instructions
that now use gvec.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This adds another kind of weirdness when you thought you had seen it all:
an opcode byte that comes _after_ the address, not before. It's not
worth adding a new X86_SPECIAL_* constant for it, but it's actually
not unlike VCMP; so, forgive me for exploiting the similarity and just
deciding to dispatch to the right gen_helper_* call in a single code
generation function.
In fact, the old decoder had a bug where s->rip_offset should have
been set to 1 for 3DNow! instructions, and it's fixed now.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
These are exactly the same as the non-VEX version, but one has to be careful
that only VEX.L=0 is allowed.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Here the code is a bit uglier due to the truncation and extension
of registers to and from 32-bit. There is also a mistake in the
manual with respect to the size of the memory operand of CVTPS2PI
and CVTTPS2PI, reported by Ricky Zhou.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
These are mostly moves, and yet are a total pain. The main issue
is that:
1) some instructions are selected by mod==11 (register operand)
vs. mod=00/01/10 (memory operand)
2) stores to memory are two-operand operations, while the 3-register
and load-from-memory versions operate on the entire contents of the
destination; this makes it easier to separate the gen_* function for
the store case
3) it's inefficient to load into xmm_T0 only to move the value out
again, so the gen_* function for the load case is separated too
The manual also has various mistakes in the operands here, for example
the store case of MOVHPS operates on a 128-bit source (albeit discarding
the bottom 64 bits) and therefore should be Mq,Vdq rather than Mq,Vq.
Likewise for the destination and source of MOVHLPS.
VUNPCK?PS and VUNPCK?PD are the same as VUNPCK?DQ and VUNPCK?QDQ,
but encoded as prefixes rather than separate operands. The helpers
can be reused however.
For MOVSLDUP, MOVSHDUP and MOVDDUP I chose to reimplement them as
helpers. I named the helper for MOVDDUP "movdldup" in preparation
for possible future introduction of MOVDHDUP and to clarify the
similarity with MOVSLDUP.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Nothing special going on here, for once.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
There are several special cases here:
1) extending moves have different widths for the helpers vs. for the
memory loads, and the width for memory loads depends on VEX.L too.
This is represented by X86_SPECIAL_AVXExtMov.
2) some instructions, such as variable-width shifts, select the vector element
size via REX.W.
3) VSIB instructions (VGATHERxPy, VPGATHERxy) are also part of this group,
and they have (among other things) two output operands.
3) the macros for 4-operand blends (which are under 0x0f 0x3a) have to be
extended to support 2-operand blends. The 2-operand variant actually
came a few years earlier, but it is clearer to implement them in the
opposite order.
X86_TYPE_WM, introduced earlier for unaligned loads, is reused for helpers
that accept a Reg* but have a M argument.
These three-byte opcodes also include AVX new instructions, for which
the helpers were originally implemented by Paul Brook <paul@nowt.org>.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The more complicated operations here are insertions and extractions.
Otherwise, there are just more entries than usual because the PS/PD/SS/SD
variations are encoded in the opcode rater than in the prefixes.
These three-byte opcodes also include AVX new instructions, whose
implementation in the helpers was originally done by Paul Brook
<paul@nowt.org>.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The more complicated ones here are d6-d7, e6-e7, f7. The others
are trivial.
For LDDQU, using gen_load_sse directly might corrupt the register if
the second part of the load fails. Therefore, add a custom X86_TYPE_WM
value; like X86_TYPE_W it does call gen_load(), but it also rejects a
value of 11 in the ModRM field like X86_TYPE_M.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This includes shifts by immediate, which use bits 3-5 of the ModRM byte
as an opcode extension. With the exception of 128-bit shifts, they are
implemented using gvec.
This also covers VZEROALL and VZEROUPPER, which use the same opcode
as EMMS. If we were wanting to optimize out gen_clear_ymmh then this
would be one of the starting points. The implementation of the VZEROALL
and VZEROUPPER helpers is by Paul Brook.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
These are a mixed batch, including the first two horizontal
(66 and F2 only) operations, more moves, and SSE4a extract/insert.
Because SSE4a is pretty rare, I chose to leave the helper as they are,
but it is possible to unify them by loading index and length from the
source XMM register and generating deposit or extract TCG ops.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
These are mostly floating-point SSE operations. The odd ones out
are MOVMSK and CVTxx2yy, the others are straightforward.
Unary operations are a bit special in AVX because they have 2 operands
for PD/PS operands (VEX.vvvv must be 1111b), and 3 operands for SD/SS.
They are handled using X86_OP_GROUP3 for compactness.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
These are more simple integer instructions present in both MMX and SSE/AVX,
with no holes that were later occupied by newer instructions.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
These are both MMX and SSE/AVX instructions, except for vmovdqu. In both
cases the inputs and output is in s->ptr{0,1,2}, so the only difference
between MMX, SSE, and AVX is which helper to call.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Because these are the only VEX instructions that QEMU supports, the
new decoder is entered on the first byte of a valid VEX prefix, and VEX
decoding only needs to be done in decode-new.c.inc.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Many SSE and AVX instructions are only valid with specific prefixes
(none, 66, F3, F2). Introduce a direct way to encode this in the
decoding table to avoid using decode groups too much.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Add generic code generation that takes care of preparing operands
around calls to decode.e.gen in a table-driven manner, so that ALU
operations need not take care of that.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The new decoder is based on three principles:
- use mostly table-driven decoding, using tables derived as much as possible
from the Intel manual. Centralizing the decode the operands makes it
more homogeneous, for example all immediates are signed. All modrm
handling is in one function, and can be shared between SSE and ALU
instructions (including XMM<->GPR instructions). The SSE/AVX decoder
will also not have duplicated code between the 0F, 0F38 and 0F3A tables.
- keep the code as "non-branchy" as possible. Generally, the code for
the new decoder is more verbose, but the control flow is simpler.
Conditionals are not nested and have small bodies. All instruction
groups are resolved even before operands are decoded, and code
generation is separated as much as possible within small functions
that only handle one instruction each.
- keep address generation and (for ALU operands) memory loads and writeback
as much in common code as possible. All ALU operations for example
are implemented as T0=f(T0,T1). For non-ALU instructions,
read-modify-write memory operations are rare, but registers do not
have TCGv equivalents: therefore, the common logic sets up pointer
temporaries with the operands, while load and writeback are handled
by gvec or by helpers.
These principles make future code review and extensibility simpler, at
the cost of having a relatively large amount of code in the form of this
patch. Even EVEX should not be _too_ hard to implement (it's just a crazy
large amount of possibilities).
This patch introduces the main decoder flow, and integrates the old
decoder with the new one. The old decoder takes care of parsing
prefixes and then optionally drops to the new one. The changes to the
old decoder are minimal and allow it to be replaced incrementally with
the new one.
There is a debugging mechanism through a "LIMIT" environment variable.
In user-mode emulation, the variable is the number of instructions
decoded by the new decoder before permanently switching to the old one.
In system emulation, the variable is the highest opcode that is decoded
by the new decoder (this is less friendly, but it's the best that can
be done without requiring deterministic execution).
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
