qemu/tcg/tcg.h

1485 lines
47 KiB
C
Raw Normal View History

/*
* Tiny Code Generator for QEMU
*
* Copyright (c) 2008 Fabrice Bellard
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
* THE SOFTWARE.
*/
#ifndef TCG_H
#define TCG_H
#include "qemu-common.h"
#include "cpu.h"
#include "exec/tb-context.h"
#include "qemu/bitops.h"
#include "qemu/queue.h"
#include "tcg-mo.h"
#include "tcg-target.h"
#include "qemu/int128.h"
/* XXX: make safe guess about sizes */
#define MAX_OP_PER_INSTR 266
#if HOST_LONG_BITS == 32
#define MAX_OPC_PARAM_PER_ARG 2
#else
#define MAX_OPC_PARAM_PER_ARG 1
#endif
#define MAX_OPC_PARAM_IARGS 6
#define MAX_OPC_PARAM_OARGS 1
#define MAX_OPC_PARAM_ARGS (MAX_OPC_PARAM_IARGS + MAX_OPC_PARAM_OARGS)
/* A Call op needs up to 4 + 2N parameters on 32-bit archs,
* and up to 4 + N parameters on 64-bit archs
* (N = number of input arguments + output arguments). */
#define MAX_OPC_PARAM (4 + (MAX_OPC_PARAM_PER_ARG * MAX_OPC_PARAM_ARGS))
#define CPU_TEMP_BUF_NLONGS 128
/* Default target word size to pointer size. */
#ifndef TCG_TARGET_REG_BITS
# if UINTPTR_MAX == UINT32_MAX
# define TCG_TARGET_REG_BITS 32
# elif UINTPTR_MAX == UINT64_MAX
# define TCG_TARGET_REG_BITS 64
# else
# error Unknown pointer size for tcg target
# endif
#endif
#if TCG_TARGET_REG_BITS == 32
typedef int32_t tcg_target_long;
typedef uint32_t tcg_target_ulong;
#define TCG_PRIlx PRIx32
#define TCG_PRIld PRId32
#elif TCG_TARGET_REG_BITS == 64
typedef int64_t tcg_target_long;
typedef uint64_t tcg_target_ulong;
#define TCG_PRIlx PRIx64
#define TCG_PRIld PRId64
#else
#error unsupported
#endif
/* Oversized TCG guests make things like MTTCG hard
* as we can't use atomics for cputlb updates.
*/
#if TARGET_LONG_BITS > TCG_TARGET_REG_BITS
#define TCG_OVERSIZED_GUEST 1
#else
#define TCG_OVERSIZED_GUEST 0
#endif
#if TCG_TARGET_NB_REGS <= 32
typedef uint32_t TCGRegSet;
#elif TCG_TARGET_NB_REGS <= 64
typedef uint64_t TCGRegSet;
#else
#error unsupported
#endif
#if TCG_TARGET_REG_BITS == 32
/* Turn some undef macros into false macros. */
#define TCG_TARGET_HAS_extrl_i64_i32 0
#define TCG_TARGET_HAS_extrh_i64_i32 0
#define TCG_TARGET_HAS_div_i64 0
#define TCG_TARGET_HAS_rem_i64 0
#define TCG_TARGET_HAS_div2_i64 0
#define TCG_TARGET_HAS_rot_i64 0
#define TCG_TARGET_HAS_ext8s_i64 0
#define TCG_TARGET_HAS_ext16s_i64 0
#define TCG_TARGET_HAS_ext32s_i64 0
#define TCG_TARGET_HAS_ext8u_i64 0
#define TCG_TARGET_HAS_ext16u_i64 0
#define TCG_TARGET_HAS_ext32u_i64 0
#define TCG_TARGET_HAS_bswap16_i64 0
#define TCG_TARGET_HAS_bswap32_i64 0
#define TCG_TARGET_HAS_bswap64_i64 0
#define TCG_TARGET_HAS_neg_i64 0
#define TCG_TARGET_HAS_not_i64 0
#define TCG_TARGET_HAS_andc_i64 0
#define TCG_TARGET_HAS_orc_i64 0
#define TCG_TARGET_HAS_eqv_i64 0
#define TCG_TARGET_HAS_nand_i64 0
#define TCG_TARGET_HAS_nor_i64 0
#define TCG_TARGET_HAS_clz_i64 0
#define TCG_TARGET_HAS_ctz_i64 0
#define TCG_TARGET_HAS_ctpop_i64 0
#define TCG_TARGET_HAS_deposit_i64 0
#define TCG_TARGET_HAS_extract_i64 0
#define TCG_TARGET_HAS_sextract_i64 0
#define TCG_TARGET_HAS_movcond_i64 0
#define TCG_TARGET_HAS_add2_i64 0
#define TCG_TARGET_HAS_sub2_i64 0
#define TCG_TARGET_HAS_mulu2_i64 0
#define TCG_TARGET_HAS_muls2_i64 0
#define TCG_TARGET_HAS_muluh_i64 0
#define TCG_TARGET_HAS_mulsh_i64 0
/* Turn some undef macros into true macros. */
#define TCG_TARGET_HAS_add2_i32 1
#define TCG_TARGET_HAS_sub2_i32 1
#endif
#ifndef TCG_TARGET_deposit_i32_valid
#define TCG_TARGET_deposit_i32_valid(ofs, len) 1
#endif
#ifndef TCG_TARGET_deposit_i64_valid
#define TCG_TARGET_deposit_i64_valid(ofs, len) 1
#endif
#ifndef TCG_TARGET_extract_i32_valid
#define TCG_TARGET_extract_i32_valid(ofs, len) 1
#endif
#ifndef TCG_TARGET_extract_i64_valid
#define TCG_TARGET_extract_i64_valid(ofs, len) 1
#endif
/* Only one of DIV or DIV2 should be defined. */
#if defined(TCG_TARGET_HAS_div_i32)
#define TCG_TARGET_HAS_div2_i32 0
#elif defined(TCG_TARGET_HAS_div2_i32)
#define TCG_TARGET_HAS_div_i32 0
#define TCG_TARGET_HAS_rem_i32 0
#endif
#if defined(TCG_TARGET_HAS_div_i64)
#define TCG_TARGET_HAS_div2_i64 0
#elif defined(TCG_TARGET_HAS_div2_i64)
#define TCG_TARGET_HAS_div_i64 0
#define TCG_TARGET_HAS_rem_i64 0
#endif
/* For 32-bit targets, some sort of unsigned widening multiply is required. */
#if TCG_TARGET_REG_BITS == 32 \
&& !(defined(TCG_TARGET_HAS_mulu2_i32) \
|| defined(TCG_TARGET_HAS_muluh_i32))
# error "Missing unsigned widening multiply"
#endif
#if !defined(TCG_TARGET_HAS_v64) \
&& !defined(TCG_TARGET_HAS_v128) \
&& !defined(TCG_TARGET_HAS_v256)
#define TCG_TARGET_MAYBE_vec 0
#define TCG_TARGET_HAS_neg_vec 0
#define TCG_TARGET_HAS_not_vec 0
#define TCG_TARGET_HAS_andc_vec 0
#define TCG_TARGET_HAS_orc_vec 0
#define TCG_TARGET_HAS_shi_vec 0
#define TCG_TARGET_HAS_shs_vec 0
#define TCG_TARGET_HAS_shv_vec 0
#define TCG_TARGET_HAS_mul_vec 0
#else
#define TCG_TARGET_MAYBE_vec 1
#endif
#ifndef TCG_TARGET_HAS_v64
#define TCG_TARGET_HAS_v64 0
#endif
#ifndef TCG_TARGET_HAS_v128
#define TCG_TARGET_HAS_v128 0
#endif
#ifndef TCG_TARGET_HAS_v256
#define TCG_TARGET_HAS_v256 0
#endif
#ifndef TARGET_INSN_START_EXTRA_WORDS
# define TARGET_INSN_START_WORDS 1
#else
# define TARGET_INSN_START_WORDS (1 + TARGET_INSN_START_EXTRA_WORDS)
#endif
typedef enum TCGOpcode {
#define DEF(name, oargs, iargs, cargs, flags) INDEX_op_ ## name,
#include "tcg-opc.h"
#undef DEF
NB_OPS,
} TCGOpcode;
#define tcg_regset_set_reg(d, r) ((d) |= (TCGRegSet)1 << (r))
#define tcg_regset_reset_reg(d, r) ((d) &= ~((TCGRegSet)1 << (r)))
#define tcg_regset_test_reg(d, r) (((d) >> (r)) & 1)
#ifndef TCG_TARGET_INSN_UNIT_SIZE
# error "Missing TCG_TARGET_INSN_UNIT_SIZE"
#elif TCG_TARGET_INSN_UNIT_SIZE == 1
typedef uint8_t tcg_insn_unit;
#elif TCG_TARGET_INSN_UNIT_SIZE == 2
typedef uint16_t tcg_insn_unit;
#elif TCG_TARGET_INSN_UNIT_SIZE == 4
typedef uint32_t tcg_insn_unit;
#elif TCG_TARGET_INSN_UNIT_SIZE == 8
typedef uint64_t tcg_insn_unit;
#else
/* The port better have done this. */
#endif
#if defined CONFIG_DEBUG_TCG || defined QEMU_STATIC_ANALYSIS
# define tcg_debug_assert(X) do { assert(X); } while (0)
#elif QEMU_GNUC_PREREQ(4, 5)
# define tcg_debug_assert(X) \
do { if (!(X)) { __builtin_unreachable(); } } while (0)
#else
# define tcg_debug_assert(X) do { (void)(X); } while (0)
#endif
typedef struct TCGRelocation {
struct TCGRelocation *next;
int type;
tcg_insn_unit *ptr;
intptr_t addend;
} TCGRelocation;
typedef struct TCGLabel {
unsigned has_value : 1;
unsigned id : 31;
union {
uintptr_t value;
tcg_insn_unit *value_ptr;
TCGRelocation *first_reloc;
} u;
} TCGLabel;
typedef struct TCGPool {
struct TCGPool *next;
int size;
uint8_t data[0] __attribute__ ((aligned));
} TCGPool;
#define TCG_POOL_CHUNK_SIZE 32768
#define TCG_MAX_TEMPS 512
#define TCG_MAX_INSNS 512
/* when the size of the arguments of a called function is smaller than
this value, they are statically allocated in the TB stack frame */
#define TCG_STATIC_CALL_ARGS_SIZE 128
typedef enum TCGType {
TCG_TYPE_I32,
TCG_TYPE_I64,
TCG_TYPE_V64,
TCG_TYPE_V128,
TCG_TYPE_V256,
TCG_TYPE_COUNT, /* number of different types */
/* An alias for the size of the host register. */
#if TCG_TARGET_REG_BITS == 32
TCG_TYPE_REG = TCG_TYPE_I32,
#else
TCG_TYPE_REG = TCG_TYPE_I64,
#endif
/* An alias for the size of the native pointer. */
#if UINTPTR_MAX == UINT32_MAX
TCG_TYPE_PTR = TCG_TYPE_I32,
#else
TCG_TYPE_PTR = TCG_TYPE_I64,
#endif
/* An alias for the size of the target "long", aka register. */
#if TARGET_LONG_BITS == 64
TCG_TYPE_TL = TCG_TYPE_I64,
#else
TCG_TYPE_TL = TCG_TYPE_I32,
#endif
} TCGType;
/* Constants for qemu_ld and qemu_st for the Memory Operation field. */
typedef enum TCGMemOp {
MO_8 = 0,
MO_16 = 1,
MO_32 = 2,
MO_64 = 3,
MO_SIZE = 3, /* Mask for the above. */
MO_SIGN = 4, /* Sign-extended, otherwise zero-extended. */
MO_BSWAP = 8, /* Host reverse endian. */
#ifdef HOST_WORDS_BIGENDIAN
MO_LE = MO_BSWAP,
MO_BE = 0,
#else
MO_LE = 0,
MO_BE = MO_BSWAP,
#endif
#ifdef TARGET_WORDS_BIGENDIAN
MO_TE = MO_BE,
#else
MO_TE = MO_LE,
#endif
/* MO_UNALN accesses are never checked for alignment.
* MO_ALIGN accesses will result in a call to the CPU's
* do_unaligned_access hook if the guest address is not aligned.
* The default depends on whether the target CPU defines ALIGNED_ONLY.
*
* Some architectures (e.g. ARMv8) need the address which is aligned
* to a size more than the size of the memory access.
* Some architectures (e.g. SPARCv9) need an address which is aligned,
* but less strictly than the natural alignment.
*
* MO_ALIGN supposes the alignment size is the size of a memory access.
*
* There are three options:
* - unaligned access permitted (MO_UNALN).
* - an alignment to the size of an access (MO_ALIGN);
* - an alignment to a specified size, which may be more or less than
* the access size (MO_ALIGN_x where 'x' is a size in bytes);
*/
MO_ASHIFT = 4,
MO_AMASK = 7 << MO_ASHIFT,
#ifdef ALIGNED_ONLY
MO_ALIGN = 0,
MO_UNALN = MO_AMASK,
#else
MO_ALIGN = MO_AMASK,
MO_UNALN = 0,
#endif
MO_ALIGN_2 = 1 << MO_ASHIFT,
MO_ALIGN_4 = 2 << MO_ASHIFT,
MO_ALIGN_8 = 3 << MO_ASHIFT,
MO_ALIGN_16 = 4 << MO_ASHIFT,
MO_ALIGN_32 = 5 << MO_ASHIFT,
MO_ALIGN_64 = 6 << MO_ASHIFT,
/* Combinations of the above, for ease of use. */
MO_UB = MO_8,
MO_UW = MO_16,
MO_UL = MO_32,
MO_SB = MO_SIGN | MO_8,
MO_SW = MO_SIGN | MO_16,
MO_SL = MO_SIGN | MO_32,
MO_Q = MO_64,
MO_LEUW = MO_LE | MO_UW,
MO_LEUL = MO_LE | MO_UL,
MO_LESW = MO_LE | MO_SW,
MO_LESL = MO_LE | MO_SL,
MO_LEQ = MO_LE | MO_Q,
MO_BEUW = MO_BE | MO_UW,
MO_BEUL = MO_BE | MO_UL,
MO_BESW = MO_BE | MO_SW,
MO_BESL = MO_BE | MO_SL,
MO_BEQ = MO_BE | MO_Q,
MO_TEUW = MO_TE | MO_UW,
MO_TEUL = MO_TE | MO_UL,
MO_TESW = MO_TE | MO_SW,
MO_TESL = MO_TE | MO_SL,
MO_TEQ = MO_TE | MO_Q,
MO_SSIZE = MO_SIZE | MO_SIGN,
} TCGMemOp;
/**
* get_alignment_bits
* @memop: TCGMemOp value
*
* Extract the alignment size from the memop.
*/
static inline unsigned get_alignment_bits(TCGMemOp memop)
{
unsigned a = memop & MO_AMASK;
if (a == MO_UNALN) {
/* No alignment required. */
a = 0;
} else if (a == MO_ALIGN) {
/* A natural alignment requirement. */
a = memop & MO_SIZE;
} else {
/* A specific alignment requirement. */
a = a >> MO_ASHIFT;
}
#if defined(CONFIG_SOFTMMU)
/* The requested alignment cannot overlap the TLB flags. */
tcg_debug_assert((TLB_FLAGS_MASK & ((1 << a) - 1)) == 0);
#endif
return a;
}
typedef tcg_target_ulong TCGArg;
/* Define type and accessor macros for TCG variables.
TCG variables are the inputs and outputs of TCG ops, as described
in tcg/README. Target CPU front-end code uses these types to deal
with TCG variables as it emits TCG code via the tcg_gen_* functions.
They come in several flavours:
* TCGv_i32 : 32 bit integer type
* TCGv_i64 : 64 bit integer type
* TCGv_ptr : a host pointer type
* TCGv_vec : a host vector type; the exact size is not exposed
to the CPU front-end code.
* TCGv : an integer type the same size as target_ulong
(an alias for either TCGv_i32 or TCGv_i64)
The compiler's type checking will complain if you mix them
up and pass the wrong sized TCGv to a function.
Users of tcg_gen_* don't need to know about any of the internal
details of these, and should treat them as opaque types.
You won't be able to look inside them in a debugger either.
Internal implementation details follow:
Note that there is no definition of the structs TCGv_i32_d etc anywhere.
This is deliberate, because the values we store in variables of type
TCGv_i32 are not really pointers-to-structures. They're just small
integers, but keeping them in pointer types like this means that the
compiler will complain if you accidentally pass a TCGv_i32 to a
function which takes a TCGv_i64, and so on. Only the internals of
TCG need to care about the actual contents of the types. */
typedef struct TCGv_i32_d *TCGv_i32;
typedef struct TCGv_i64_d *TCGv_i64;
typedef struct TCGv_ptr_d *TCGv_ptr;
typedef struct TCGv_vec_d *TCGv_vec;
typedef TCGv_ptr TCGv_env;
#if TARGET_LONG_BITS == 32
#define TCGv TCGv_i32
#elif TARGET_LONG_BITS == 64
#define TCGv TCGv_i64
#else
#error Unhandled TARGET_LONG_BITS value
#endif
/* call flags */
/* Helper does not read globals (either directly or through an exception). It
implies TCG_CALL_NO_WRITE_GLOBALS. */
#define TCG_CALL_NO_READ_GLOBALS 0x0010
/* Helper does not write globals */
#define TCG_CALL_NO_WRITE_GLOBALS 0x0020
/* Helper can be safely suppressed if the return value is not used. */
#define TCG_CALL_NO_SIDE_EFFECTS 0x0040
/* convenience version of most used call flags */
#define TCG_CALL_NO_RWG TCG_CALL_NO_READ_GLOBALS
#define TCG_CALL_NO_WG TCG_CALL_NO_WRITE_GLOBALS
#define TCG_CALL_NO_SE TCG_CALL_NO_SIDE_EFFECTS
#define TCG_CALL_NO_RWG_SE (TCG_CALL_NO_RWG | TCG_CALL_NO_SE)
#define TCG_CALL_NO_WG_SE (TCG_CALL_NO_WG | TCG_CALL_NO_SE)
/* Used to align parameters. See the comment before tcgv_i32_temp. */
#define TCG_CALL_DUMMY_ARG ((TCGArg)0)
/* Conditions. Note that these are laid out for easy manipulation by
the functions below:
bit 0 is used for inverting;
bit 1 is signed,
bit 2 is unsigned,
bit 3 is used with bit 0 for swapping signed/unsigned. */
typedef enum {
/* non-signed */
TCG_COND_NEVER = 0 | 0 | 0 | 0,
TCG_COND_ALWAYS = 0 | 0 | 0 | 1,
TCG_COND_EQ = 8 | 0 | 0 | 0,
TCG_COND_NE = 8 | 0 | 0 | 1,
/* signed */
TCG_COND_LT = 0 | 0 | 2 | 0,
TCG_COND_GE = 0 | 0 | 2 | 1,
TCG_COND_LE = 8 | 0 | 2 | 0,
TCG_COND_GT = 8 | 0 | 2 | 1,
/* unsigned */
TCG_COND_LTU = 0 | 4 | 0 | 0,
TCG_COND_GEU = 0 | 4 | 0 | 1,
TCG_COND_LEU = 8 | 4 | 0 | 0,
TCG_COND_GTU = 8 | 4 | 0 | 1,
} TCGCond;
/* Invert the sense of the comparison. */
static inline TCGCond tcg_invert_cond(TCGCond c)
{
return (TCGCond)(c ^ 1);
}
/* Swap the operands in a comparison. */
static inline TCGCond tcg_swap_cond(TCGCond c)
{
return c & 6 ? (TCGCond)(c ^ 9) : c;
}
/* Create an "unsigned" version of a "signed" comparison. */
static inline TCGCond tcg_unsigned_cond(TCGCond c)
{
return c & 2 ? (TCGCond)(c ^ 6) : c;
}
/* Create a "signed" version of an "unsigned" comparison. */
static inline TCGCond tcg_signed_cond(TCGCond c)
{
return c & 4 ? (TCGCond)(c ^ 6) : c;
}
/* Must a comparison be considered unsigned? */
static inline bool is_unsigned_cond(TCGCond c)
{
return (c & 4) != 0;
}
/* Create a "high" version of a double-word comparison.
This removes equality from a LTE or GTE comparison. */
static inline TCGCond tcg_high_cond(TCGCond c)
{
switch (c) {
case TCG_COND_GE:
case TCG_COND_LE:
case TCG_COND_GEU:
case TCG_COND_LEU:
return (TCGCond)(c ^ 8);
default:
return c;
}
}
tcg: optimise memory layout of TCGTemp This brings down the size of the struct from 56 to 32 bytes on 64-bit, and to 20 bytes on 32-bit. This leads to memory savings: Before: $ find . -name 'tcg.o' | xargs size text data bss dec hex filename 41131 29800 88 71019 1156b ./aarch64-softmmu/tcg/tcg.o 37969 29416 96 67481 10799 ./x86_64-linux-user/tcg/tcg.o 39354 28816 96 68266 10aaa ./arm-linux-user/tcg/tcg.o 40802 29096 88 69986 11162 ./arm-softmmu/tcg/tcg.o 39417 29672 88 69177 10e39 ./x86_64-softmmu/tcg/tcg.o After: $ find . -name 'tcg.o' | xargs size text data bss dec hex filename 40883 29800 88 70771 11473 ./aarch64-softmmu/tcg/tcg.o 37473 29416 96 66985 105a9 ./x86_64-linux-user/tcg/tcg.o 38858 28816 96 67770 108ba ./arm-linux-user/tcg/tcg.o 40554 29096 88 69738 1106a ./arm-softmmu/tcg/tcg.o 39169 29672 88 68929 10d41 ./x86_64-softmmu/tcg/tcg.o Note that using an entire byte for some enums that need less than that wastes a few bits (noticeable in 32 bits, where we use 20 bytes instead of 16) but avoids extraction code, which overall is a win--I've tested several variations of the patch, and the appended is the best performer for OpenSSL's bntest by a very small margin: Before: $ taskset -c 0 perf stat -r 15 -- x86_64-linux-user/qemu-x86_64 img/bntest-x86_64 >/dev/null [...] Performance counter stats for 'x86_64-linux-user/qemu-x86_64 img/bntest-x86_64' (15 runs): 10538.479833 task-clock (msec) # 0.999 CPUs utilized ( +- 0.38% ) 772 context-switches # 0.073 K/sec ( +- 2.03% ) 0 cpu-migrations # 0.000 K/sec ( +-100.00% ) 2,207 page-faults # 0.209 K/sec ( +- 0.08% ) 10.552871687 seconds time elapsed ( +- 0.39% ) After: $ taskset -c 0 perf stat -r 15 -- x86_64-linux-user/qemu-x86_64 img/bntest-x86_64 >/dev/null Performance counter stats for 'x86_64-linux-user/qemu-x86_64 img/bntest-x86_64' (15 runs): 10459.968847 task-clock (msec) # 0.999 CPUs utilized ( +- 0.30% ) 739 context-switches # 0.071 K/sec ( +- 1.71% ) 0 cpu-migrations # 0.000 K/sec ( +- 68.14% ) 2,204 page-faults # 0.211 K/sec ( +- 0.10% ) 10.473900411 seconds time elapsed ( +- 0.30% ) Suggested-by: Stefan Weil <sw@weilnetz.de> Suggested-by: Richard Henderson <rth@twiddle.net> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Signed-off-by: Emilio G. Cota <cota@braap.org> Signed-off-by: Richard Henderson <rth@twiddle.net>
2015-04-03 03:07:53 +03:00
typedef enum TCGTempVal {
TEMP_VAL_DEAD,
TEMP_VAL_REG,
TEMP_VAL_MEM,
TEMP_VAL_CONST,
} TCGTempVal;
typedef struct TCGTemp {
TCGReg reg:8;
tcg: optimise memory layout of TCGTemp This brings down the size of the struct from 56 to 32 bytes on 64-bit, and to 20 bytes on 32-bit. This leads to memory savings: Before: $ find . -name 'tcg.o' | xargs size text data bss dec hex filename 41131 29800 88 71019 1156b ./aarch64-softmmu/tcg/tcg.o 37969 29416 96 67481 10799 ./x86_64-linux-user/tcg/tcg.o 39354 28816 96 68266 10aaa ./arm-linux-user/tcg/tcg.o 40802 29096 88 69986 11162 ./arm-softmmu/tcg/tcg.o 39417 29672 88 69177 10e39 ./x86_64-softmmu/tcg/tcg.o After: $ find . -name 'tcg.o' | xargs size text data bss dec hex filename 40883 29800 88 70771 11473 ./aarch64-softmmu/tcg/tcg.o 37473 29416 96 66985 105a9 ./x86_64-linux-user/tcg/tcg.o 38858 28816 96 67770 108ba ./arm-linux-user/tcg/tcg.o 40554 29096 88 69738 1106a ./arm-softmmu/tcg/tcg.o 39169 29672 88 68929 10d41 ./x86_64-softmmu/tcg/tcg.o Note that using an entire byte for some enums that need less than that wastes a few bits (noticeable in 32 bits, where we use 20 bytes instead of 16) but avoids extraction code, which overall is a win--I've tested several variations of the patch, and the appended is the best performer for OpenSSL's bntest by a very small margin: Before: $ taskset -c 0 perf stat -r 15 -- x86_64-linux-user/qemu-x86_64 img/bntest-x86_64 >/dev/null [...] Performance counter stats for 'x86_64-linux-user/qemu-x86_64 img/bntest-x86_64' (15 runs): 10538.479833 task-clock (msec) # 0.999 CPUs utilized ( +- 0.38% ) 772 context-switches # 0.073 K/sec ( +- 2.03% ) 0 cpu-migrations # 0.000 K/sec ( +-100.00% ) 2,207 page-faults # 0.209 K/sec ( +- 0.08% ) 10.552871687 seconds time elapsed ( +- 0.39% ) After: $ taskset -c 0 perf stat -r 15 -- x86_64-linux-user/qemu-x86_64 img/bntest-x86_64 >/dev/null Performance counter stats for 'x86_64-linux-user/qemu-x86_64 img/bntest-x86_64' (15 runs): 10459.968847 task-clock (msec) # 0.999 CPUs utilized ( +- 0.30% ) 739 context-switches # 0.071 K/sec ( +- 1.71% ) 0 cpu-migrations # 0.000 K/sec ( +- 68.14% ) 2,204 page-faults # 0.211 K/sec ( +- 0.10% ) 10.473900411 seconds time elapsed ( +- 0.30% ) Suggested-by: Stefan Weil <sw@weilnetz.de> Suggested-by: Richard Henderson <rth@twiddle.net> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Signed-off-by: Emilio G. Cota <cota@braap.org> Signed-off-by: Richard Henderson <rth@twiddle.net>
2015-04-03 03:07:53 +03:00
TCGTempVal val_type:8;
TCGType base_type:8;
TCGType type:8;
unsigned int fixed_reg:1;
unsigned int indirect_reg:1;
unsigned int indirect_base:1;
unsigned int mem_coherent:1;
unsigned int mem_allocated:1;
/* If true, the temp is saved across both basic blocks and
translation blocks. */
unsigned int temp_global:1;
/* If true, the temp is saved across basic blocks but dead
at the end of translation blocks. If false, the temp is
dead at the end of basic blocks. */
unsigned int temp_local:1;
unsigned int temp_allocated:1;
tcg: optimise memory layout of TCGTemp This brings down the size of the struct from 56 to 32 bytes on 64-bit, and to 20 bytes on 32-bit. This leads to memory savings: Before: $ find . -name 'tcg.o' | xargs size text data bss dec hex filename 41131 29800 88 71019 1156b ./aarch64-softmmu/tcg/tcg.o 37969 29416 96 67481 10799 ./x86_64-linux-user/tcg/tcg.o 39354 28816 96 68266 10aaa ./arm-linux-user/tcg/tcg.o 40802 29096 88 69986 11162 ./arm-softmmu/tcg/tcg.o 39417 29672 88 69177 10e39 ./x86_64-softmmu/tcg/tcg.o After: $ find . -name 'tcg.o' | xargs size text data bss dec hex filename 40883 29800 88 70771 11473 ./aarch64-softmmu/tcg/tcg.o 37473 29416 96 66985 105a9 ./x86_64-linux-user/tcg/tcg.o 38858 28816 96 67770 108ba ./arm-linux-user/tcg/tcg.o 40554 29096 88 69738 1106a ./arm-softmmu/tcg/tcg.o 39169 29672 88 68929 10d41 ./x86_64-softmmu/tcg/tcg.o Note that using an entire byte for some enums that need less than that wastes a few bits (noticeable in 32 bits, where we use 20 bytes instead of 16) but avoids extraction code, which overall is a win--I've tested several variations of the patch, and the appended is the best performer for OpenSSL's bntest by a very small margin: Before: $ taskset -c 0 perf stat -r 15 -- x86_64-linux-user/qemu-x86_64 img/bntest-x86_64 >/dev/null [...] Performance counter stats for 'x86_64-linux-user/qemu-x86_64 img/bntest-x86_64' (15 runs): 10538.479833 task-clock (msec) # 0.999 CPUs utilized ( +- 0.38% ) 772 context-switches # 0.073 K/sec ( +- 2.03% ) 0 cpu-migrations # 0.000 K/sec ( +-100.00% ) 2,207 page-faults # 0.209 K/sec ( +- 0.08% ) 10.552871687 seconds time elapsed ( +- 0.39% ) After: $ taskset -c 0 perf stat -r 15 -- x86_64-linux-user/qemu-x86_64 img/bntest-x86_64 >/dev/null Performance counter stats for 'x86_64-linux-user/qemu-x86_64 img/bntest-x86_64' (15 runs): 10459.968847 task-clock (msec) # 0.999 CPUs utilized ( +- 0.30% ) 739 context-switches # 0.071 K/sec ( +- 1.71% ) 0 cpu-migrations # 0.000 K/sec ( +- 68.14% ) 2,204 page-faults # 0.211 K/sec ( +- 0.10% ) 10.473900411 seconds time elapsed ( +- 0.30% ) Suggested-by: Stefan Weil <sw@weilnetz.de> Suggested-by: Richard Henderson <rth@twiddle.net> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Signed-off-by: Emilio G. Cota <cota@braap.org> Signed-off-by: Richard Henderson <rth@twiddle.net>
2015-04-03 03:07:53 +03:00
tcg_target_long val;
struct TCGTemp *mem_base;
tcg: optimise memory layout of TCGTemp This brings down the size of the struct from 56 to 32 bytes on 64-bit, and to 20 bytes on 32-bit. This leads to memory savings: Before: $ find . -name 'tcg.o' | xargs size text data bss dec hex filename 41131 29800 88 71019 1156b ./aarch64-softmmu/tcg/tcg.o 37969 29416 96 67481 10799 ./x86_64-linux-user/tcg/tcg.o 39354 28816 96 68266 10aaa ./arm-linux-user/tcg/tcg.o 40802 29096 88 69986 11162 ./arm-softmmu/tcg/tcg.o 39417 29672 88 69177 10e39 ./x86_64-softmmu/tcg/tcg.o After: $ find . -name 'tcg.o' | xargs size text data bss dec hex filename 40883 29800 88 70771 11473 ./aarch64-softmmu/tcg/tcg.o 37473 29416 96 66985 105a9 ./x86_64-linux-user/tcg/tcg.o 38858 28816 96 67770 108ba ./arm-linux-user/tcg/tcg.o 40554 29096 88 69738 1106a ./arm-softmmu/tcg/tcg.o 39169 29672 88 68929 10d41 ./x86_64-softmmu/tcg/tcg.o Note that using an entire byte for some enums that need less than that wastes a few bits (noticeable in 32 bits, where we use 20 bytes instead of 16) but avoids extraction code, which overall is a win--I've tested several variations of the patch, and the appended is the best performer for OpenSSL's bntest by a very small margin: Before: $ taskset -c 0 perf stat -r 15 -- x86_64-linux-user/qemu-x86_64 img/bntest-x86_64 >/dev/null [...] Performance counter stats for 'x86_64-linux-user/qemu-x86_64 img/bntest-x86_64' (15 runs): 10538.479833 task-clock (msec) # 0.999 CPUs utilized ( +- 0.38% ) 772 context-switches # 0.073 K/sec ( +- 2.03% ) 0 cpu-migrations # 0.000 K/sec ( +-100.00% ) 2,207 page-faults # 0.209 K/sec ( +- 0.08% ) 10.552871687 seconds time elapsed ( +- 0.39% ) After: $ taskset -c 0 perf stat -r 15 -- x86_64-linux-user/qemu-x86_64 img/bntest-x86_64 >/dev/null Performance counter stats for 'x86_64-linux-user/qemu-x86_64 img/bntest-x86_64' (15 runs): 10459.968847 task-clock (msec) # 0.999 CPUs utilized ( +- 0.30% ) 739 context-switches # 0.071 K/sec ( +- 1.71% ) 0 cpu-migrations # 0.000 K/sec ( +- 68.14% ) 2,204 page-faults # 0.211 K/sec ( +- 0.10% ) 10.473900411 seconds time elapsed ( +- 0.30% ) Suggested-by: Stefan Weil <sw@weilnetz.de> Suggested-by: Richard Henderson <rth@twiddle.net> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Signed-off-by: Emilio G. Cota <cota@braap.org> Signed-off-by: Richard Henderson <rth@twiddle.net>
2015-04-03 03:07:53 +03:00
intptr_t mem_offset;
const char *name;
/* Pass-specific information that can be stored for a temporary.
One word worth of integer data, and one pointer to data
allocated separately. */
uintptr_t state;
void *state_ptr;
} TCGTemp;
typedef struct TCGContext TCGContext;
typedef struct TCGTempSet {
unsigned long l[BITS_TO_LONGS(TCG_MAX_TEMPS)];
} TCGTempSet;
/* While we limit helpers to 6 arguments, for 32-bit hosts, with padding,
this imples a max of 6*2 (64-bit in) + 2 (64-bit out) = 14 operands.
There are never more than 2 outputs, which means that we can store all
dead + sync data within 16 bits. */
#define DEAD_ARG 4
#define SYNC_ARG 1
typedef uint16_t TCGLifeData;
/* The layout here is designed to avoid a bitfield crossing of
a 32-bit boundary, which would cause GCC to add extra padding. */
typedef struct TCGOp {
TCGOpcode opc : 8; /* 8 */
/* Parameters for this opcode. See below. */
unsigned param1 : 4; /* 12 */
unsigned param2 : 4; /* 16 */
/* Lifetime data of the operands. */
unsigned life : 16; /* 32 */
/* Next and previous opcodes. */
QTAILQ_ENTRY(TCGOp) link;
/* Arguments for the opcode. */
TCGArg args[MAX_OPC_PARAM];
} TCGOp;
#define TCGOP_CALLI(X) (X)->param1
#define TCGOP_CALLO(X) (X)->param2
#define TCGOP_VECL(X) (X)->param1
#define TCGOP_VECE(X) (X)->param2
/* Make sure operands fit in the bitfields above. */
QEMU_BUILD_BUG_ON(NB_OPS > (1 << 8));
typedef struct TCGProfile {
int64_t cpu_exec_time;
int64_t tb_count1;
int64_t tb_count;
int64_t op_count; /* total insn count */
int op_count_max; /* max insn per TB */
int temp_count_max;
int64_t temp_count;
int64_t del_op_count;
int64_t code_in_len;
int64_t code_out_len;
int64_t search_out_len;
int64_t interm_time;
int64_t code_time;
int64_t la_time;
int64_t opt_time;
int64_t restore_count;
int64_t restore_time;
int64_t table_op_count[NB_OPS];
} TCGProfile;
struct TCGContext {
uint8_t *pool_cur, *pool_end;
TCGPool *pool_first, *pool_current, *pool_first_large;
int nb_labels;
int nb_globals;
int nb_temps;
int nb_indirects;
int nb_ops;
/* goto_tb support */
tcg_insn_unit *code_buf;
uint16_t *tb_jmp_reset_offset; /* tb->jmp_reset_offset */
uintptr_t *tb_jmp_insn_offset; /* tb->jmp_target_arg if direct_jump */
uintptr_t *tb_jmp_target_addr; /* tb->jmp_target_arg if !direct_jump */
TCGRegSet reserved_regs;
uint32_t tb_cflags; /* cflags of the current TB */
intptr_t current_frame_offset;
intptr_t frame_start;
intptr_t frame_end;
TCGTemp *frame_temp;
tcg_insn_unit *code_ptr;
#ifdef CONFIG_PROFILER
TCGProfile prof;
#endif
#ifdef CONFIG_DEBUG_TCG
int temps_in_use;
int goto_tb_issue_mask;
#endif
/* Code generation. Note that we specifically do not use tcg_insn_unit
here, because there's too much arithmetic throughout that relies
on addition and subtraction working on bytes. Rely on the GCC
extension that allows arithmetic on void*. */
void *code_gen_prologue;
void *code_gen_epilogue;
void *code_gen_buffer;
size_t code_gen_buffer_size;
void *code_gen_ptr;
void *data_gen_ptr;
/* Threshold to flush the translated code buffer. */
void *code_gen_highwater;
size_t tb_phys_invalidate_count;
/* Track which vCPU triggers events */
CPUState *cpu; /* *_trans */
/* These structures are private to tcg-target.inc.c. */
#ifdef TCG_TARGET_NEED_LDST_LABELS
QSIMPLEQ_HEAD(ldst_labels, TCGLabelQemuLdst) ldst_labels;
#endif
#ifdef TCG_TARGET_NEED_POOL_LABELS
struct TCGLabelPoolData *pool_labels;
#endif
TCGLabel *exitreq_label;
TCGTempSet free_temps[TCG_TYPE_COUNT * 2];
TCGTemp temps[TCG_MAX_TEMPS]; /* globals first, temps after */
QTAILQ_HEAD(TCGOpHead, TCGOp) ops, free_ops;
/* Tells which temporary holds a given register.
It does not take into account fixed registers */
TCGTemp *reg_to_temp[TCG_TARGET_NB_REGS];
uint16_t gen_insn_end_off[TCG_MAX_INSNS];
target_ulong gen_insn_data[TCG_MAX_INSNS][TARGET_INSN_START_WORDS];
};
extern TCGContext tcg_init_ctx;
tcg: enable multiple TCG contexts in softmmu This enables parallel TCG code generation. However, we do not take advantage of it yet since tb_lock is still held during tb_gen_code. In user-mode we use a single TCG context; see the documentation added to tcg_region_init for the rationale. Note that targets do not need any conversion: targets initialize a TCGContext (e.g. defining TCG globals), and after this initialization has finished, the context is cloned by the vCPU threads, each of them keeping a separate copy. TCG threads claim one entry in tcg_ctxs[] by atomically increasing n_tcg_ctxs. Do not be too annoyed by the subsequent atomic_read's of that variable and tcg_ctxs; they are there just to play nice with analysis tools such as thread sanitizer. Note that we do not allocate an array of contexts (we allocate an array of pointers instead) because when tcg_context_init is called, we do not know yet how many contexts we'll use since the bool behind qemu_tcg_mttcg_enabled() isn't set yet. Previous patches folded some TCG globals into TCGContext. The non-const globals remaining are only set at init time, i.e. before the TCG threads are spawned. Here is a list of these set-at-init-time globals under tcg/: Only written by tcg_context_init: - indirect_reg_alloc_order - tcg_op_defs Only written by tcg_target_init (called from tcg_context_init): - tcg_target_available_regs - tcg_target_call_clobber_regs - arm: arm_arch, use_idiv_instructions - i386: have_cmov, have_bmi1, have_bmi2, have_lzcnt, have_movbe, have_popcnt - mips: use_movnz_instructions, use_mips32_instructions, use_mips32r2_instructions, got_sigill (tcg_target_detect_isa) - ppc: have_isa_2_06, have_isa_3_00, tb_ret_addr - s390: tb_ret_addr, s390_facilities - sparc: qemu_ld_trampoline, qemu_st_trampoline (build_trampolines), use_vis3_instructions Only written by tcg_prologue_init: - 'struct jit_code_entry one_entry' - aarch64: tb_ret_addr - arm: tb_ret_addr - i386: tb_ret_addr, guest_base_flags - ia64: tb_ret_addr - mips: tb_ret_addr, bswap32_addr, bswap32u_addr, bswap64_addr Reviewed-by: Richard Henderson <rth@twiddle.net> Signed-off-by: Emilio G. Cota <cota@braap.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
2017-07-20 01:57:58 +03:00
extern __thread TCGContext *tcg_ctx;
extern TCGv_env cpu_env;
static inline size_t temp_idx(TCGTemp *ts)
{
ptrdiff_t n = ts - tcg_ctx->temps;
tcg_debug_assert(n >= 0 && n < tcg_ctx->nb_temps);
return n;
}
static inline TCGArg temp_arg(TCGTemp *ts)
{
return (uintptr_t)ts;
}
static inline TCGTemp *arg_temp(TCGArg a)
{
return (TCGTemp *)(uintptr_t)a;
}
/* Using the offset of a temporary, relative to TCGContext, rather than
its index means that we don't use 0. That leaves offset 0 free for
a NULL representation without having to leave index 0 unused. */
static inline TCGTemp *tcgv_i32_temp(TCGv_i32 v)
{
uintptr_t o = (uintptr_t)v;
TCGTemp *t = (void *)tcg_ctx + o;
tcg_debug_assert(offsetof(TCGContext, temps[temp_idx(t)]) == o);
return t;
}
static inline TCGTemp *tcgv_i64_temp(TCGv_i64 v)
{
return tcgv_i32_temp((TCGv_i32)v);
}
static inline TCGTemp *tcgv_ptr_temp(TCGv_ptr v)
{
return tcgv_i32_temp((TCGv_i32)v);
}
static inline TCGTemp *tcgv_vec_temp(TCGv_vec v)
{
return tcgv_i32_temp((TCGv_i32)v);
}
static inline TCGArg tcgv_i32_arg(TCGv_i32 v)
{
return temp_arg(tcgv_i32_temp(v));
}
static inline TCGArg tcgv_i64_arg(TCGv_i64 v)
{
return temp_arg(tcgv_i64_temp(v));
}
static inline TCGArg tcgv_ptr_arg(TCGv_ptr v)
{
return temp_arg(tcgv_ptr_temp(v));
}
static inline TCGArg tcgv_vec_arg(TCGv_vec v)
{
return temp_arg(tcgv_vec_temp(v));
}
static inline TCGv_i32 temp_tcgv_i32(TCGTemp *t)
{
(void)temp_idx(t); /* trigger embedded assert */
return (TCGv_i32)((void *)t - (void *)tcg_ctx);
}
static inline TCGv_i64 temp_tcgv_i64(TCGTemp *t)
{
return (TCGv_i64)temp_tcgv_i32(t);
}
static inline TCGv_ptr temp_tcgv_ptr(TCGTemp *t)
{
return (TCGv_ptr)temp_tcgv_i32(t);
}
static inline TCGv_vec temp_tcgv_vec(TCGTemp *t)
{
return (TCGv_vec)temp_tcgv_i32(t);
}
#if TCG_TARGET_REG_BITS == 32
static inline TCGv_i32 TCGV_LOW(TCGv_i64 t)
{
return temp_tcgv_i32(tcgv_i64_temp(t));
}
static inline TCGv_i32 TCGV_HIGH(TCGv_i64 t)
{
return temp_tcgv_i32(tcgv_i64_temp(t) + 1);
}
#endif
static inline void tcg_set_insn_param(TCGOp *op, int arg, TCGArg v)
{
op->args[arg] = v;
}
static inline void tcg_set_insn_start_param(TCGOp *op, int arg, target_ulong v)
{
#if TARGET_LONG_BITS <= TCG_TARGET_REG_BITS
tcg_set_insn_param(op, arg, v);
#else
tcg_set_insn_param(op, arg * 2, v);
tcg_set_insn_param(op, arg * 2 + 1, v >> 32);
#endif
}
/* The last op that was emitted. */
static inline TCGOp *tcg_last_op(void)
{
return QTAILQ_LAST(&tcg_ctx->ops, TCGOpHead);
}
/* Test for whether to terminate the TB for using too many opcodes. */
static inline bool tcg_op_buf_full(void)
{
/* This is not a hard limit, it merely stops translation when
* we have produced "enough" opcodes. We want to limit TB size
* such that a RISC host can reasonably use a 16-bit signed
* branch within the TB. We also need to be mindful of the
* 16-bit unsigned offsets, TranslationBlock.jmp_reset_offset[]
* and TCGContext.gen_insn_end_off[].
*/
return tcg_ctx->nb_ops >= 4000;
}
/* pool based memory allocation */
tcg: remove tb_lock Use mmap_lock in user-mode to protect TCG state and the page descriptors. In !user-mode, each vCPU has its own TCG state, so no locks needed. Per-page locks are used to protect the page descriptors. Per-TB locks are used in both modes to protect TB jumps. Some notes: - tb_lock is removed from notdirty_mem_write by passing a locked page_collection to tb_invalidate_phys_page_fast. - tcg_tb_lookup/remove/insert/etc have their own internal lock(s), so there is no need to further serialize access to them. - do_tb_flush is run in a safe async context, meaning no other vCPU threads are running. Therefore acquiring mmap_lock there is just to please tools such as thread sanitizer. - Not visible in the diff, but tb_invalidate_phys_page already has an assert_memory_lock. - cpu_io_recompile is !user-only, so no mmap_lock there. - Added mmap_unlock()'s before all siglongjmp's that could be called in user-mode while mmap_lock is held. + Added an assert for !have_mmap_lock() after returning from the longjmp in cpu_exec, just like we do in cpu_exec_step_atomic. Performance numbers before/after: Host: AMD Opteron(tm) Processor 6376 ubuntu 17.04 ppc64 bootup+shutdown time 700 +-+--+----+------+------------+-----------+------------*--+-+ | + + + + + *B | | before ***B*** ** * | |tb lock removal ###D### *** | 600 +-+ *** +-+ | ** # | | *B* #D | | *** * ## | 500 +-+ *** ### +-+ | * *** ### | | *B* # ## | | ** * #D# | 400 +-+ ** ## +-+ | ** ### | | ** ## | | ** # ## | 300 +-+ * B* #D# +-+ | B *** ### | | * ** #### | | * *** ### | 200 +-+ B *B #D# +-+ | #B* * ## # | | #* ## | | + D##D# + + + + | 100 +-+--+----+------+------------+-----------+------------+--+-+ 1 8 16 Guest CPUs 48 64 png: https://imgur.com/HwmBHXe debian jessie aarch64 bootup+shutdown time 90 +-+--+-----+-----+------------+------------+------------+--+-+ | + + + + + + | | before ***B*** B | 80 +tb lock removal ###D### **D +-+ | **### | | **## | 70 +-+ ** # +-+ | ** ## | | ** # | 60 +-+ *B ## +-+ | ** ## | | *** #D | 50 +-+ *** ## +-+ | * ** ### | | **B* ### | 40 +-+ **** # ## +-+ | **** #D# | | ***B** ### | 30 +-+ B***B** #### +-+ | B * * # ### | | B ###D# | 20 +-+ D ##D## +-+ | D# | | + + + + + + | 10 +-+--+-----+-----+------------+------------+------------+--+-+ 1 8 16 Guest CPUs 48 64 png: https://imgur.com/iGpGFtv The gains are high for 4-8 CPUs. Beyond that point, however, unrelated lock contention significantly hurts scalability. Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Signed-off-by: Emilio G. Cota <cota@braap.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
2017-08-05 06:46:31 +03:00
/* user-mode: mmap_lock must be held for tcg_malloc_internal. */
void *tcg_malloc_internal(TCGContext *s, int size);
void tcg_pool_reset(TCGContext *s);
tcg: allocate TB structs before the corresponding translated code Allocating an arbitrarily-sized array of tbs results in either (a) a lot of memory wasted or (b) unnecessary flushes of the code cache when we run out of TB structs in the array. An obvious solution would be to just malloc a TB struct when needed, and keep the TB array as an array of pointers (recall that tb_find_pc() needs the TB array to run in O(log n)). Perhaps a better solution, which is implemented in this patch, is to allocate TB's right before the translated code they describe. This results in some memory waste due to padding to have code and TBs in separate cache lines--for instance, I measured 4.7% of padding in the used portion of code_gen_buffer when booting aarch64 Linux on a host with 64-byte cache lines. However, it can allow for optimizations in some host architectures, since TCG backends could safely assume that the TB and the corresponding translated code are very close to each other in memory. See this message by rth for a detailed explanation: https://lists.gnu.org/archive/html/qemu-devel/2017-03/msg05172.html Subject: Re: GSoC 2017 Proposal: TCG performance enhancements Message-ID: <1e67644b-4b30-887e-d329-1848e94c9484@twiddle.net> Suggested-by: Richard Henderson <rth@twiddle.net> Reviewed-by: Pranith Kumar <bobby.prani@gmail.com> Signed-off-by: Emilio G. Cota <cota@braap.org> Message-Id: <1496790745-314-3-git-send-email-cota@braap.org> [rth: Simplify the arithmetic in tcg_tb_alloc] Signed-off-by: Richard Henderson <rth@twiddle.net>
2017-06-07 02:12:25 +03:00
TranslationBlock *tcg_tb_alloc(TCGContext *s);
tcg: introduce regions to split code_gen_buffer This is groundwork for supporting multiple TCG contexts. The naive solution here is to split code_gen_buffer statically among the TCG threads; this however results in poor utilization if translation needs are different across TCG threads. What we do here is to add an extra layer of indirection, assigning regions that act just like pages do in virtual memory allocation. (BTW if you are wondering about the chosen naming, I did not want to use blocks or pages because those are already heavily used in QEMU). We use a global lock to serialize allocations as well as statistics reporting (we now export the size of the used code_gen_buffer with tcg_code_size()). Note that for the allocator we could just use a counter and atomic_inc; however, that would complicate the gathering of tcg_code_size()-like stats. So given that the region operations are not a fast path, a lock seems the most reasonable choice. The effectiveness of this approach is clear after seeing some numbers. I used the bootup+shutdown of debian-arm with '-tb-size 80' as a benchmark. Note that I'm evaluating this after enabling per-thread TCG (which is done by a subsequent commit). * -smp 1, 1 region (entire buffer): qemu: flush code_size=83885014 nb_tbs=154739 avg_tb_size=357 qemu: flush code_size=83884902 nb_tbs=153136 avg_tb_size=363 qemu: flush code_size=83885014 nb_tbs=152777 avg_tb_size=364 qemu: flush code_size=83884950 nb_tbs=150057 avg_tb_size=373 qemu: flush code_size=83884998 nb_tbs=150234 avg_tb_size=373 qemu: flush code_size=83885014 nb_tbs=154009 avg_tb_size=360 qemu: flush code_size=83885014 nb_tbs=151007 avg_tb_size=370 qemu: flush code_size=83885014 nb_tbs=151816 avg_tb_size=367 That is, 8 flushes. * -smp 8, 32 regions (80/32 MB per region) [i.e. this patch]: qemu: flush code_size=76328008 nb_tbs=141040 avg_tb_size=356 qemu: flush code_size=75366534 nb_tbs=138000 avg_tb_size=361 qemu: flush code_size=76864546 nb_tbs=140653 avg_tb_size=361 qemu: flush code_size=76309084 nb_tbs=135945 avg_tb_size=375 qemu: flush code_size=74581856 nb_tbs=132909 avg_tb_size=375 qemu: flush code_size=73927256 nb_tbs=135616 avg_tb_size=360 qemu: flush code_size=78629426 nb_tbs=142896 avg_tb_size=365 qemu: flush code_size=76667052 nb_tbs=138508 avg_tb_size=368 Again, 8 flushes. Note how buffer utilization is not 100%, but it is close. Smaller region sizes would yield higher utilization, but we want region allocation to be rare (it acquires a lock), so we do not want to go too small. * -smp 8, static partitioning of 8 regions (10 MB per region): qemu: flush code_size=21936504 nb_tbs=40570 avg_tb_size=354 qemu: flush code_size=11472174 nb_tbs=20633 avg_tb_size=370 qemu: flush code_size=11603976 nb_tbs=21059 avg_tb_size=365 qemu: flush code_size=23254872 nb_tbs=41243 avg_tb_size=377 qemu: flush code_size=28289496 nb_tbs=52057 avg_tb_size=358 qemu: flush code_size=43605160 nb_tbs=78896 avg_tb_size=367 qemu: flush code_size=45166552 nb_tbs=82158 avg_tb_size=364 qemu: flush code_size=63289640 nb_tbs=116494 avg_tb_size=358 qemu: flush code_size=51389960 nb_tbs=93937 avg_tb_size=362 qemu: flush code_size=59665928 nb_tbs=107063 avg_tb_size=372 qemu: flush code_size=38380824 nb_tbs=68597 avg_tb_size=374 qemu: flush code_size=44884568 nb_tbs=79901 avg_tb_size=376 qemu: flush code_size=50782632 nb_tbs=90681 avg_tb_size=374 qemu: flush code_size=39848888 nb_tbs=71433 avg_tb_size=372 qemu: flush code_size=64708840 nb_tbs=119052 avg_tb_size=359 qemu: flush code_size=49830008 nb_tbs=90992 avg_tb_size=362 qemu: flush code_size=68372408 nb_tbs=123442 avg_tb_size=368 qemu: flush code_size=33555560 nb_tbs=59514 avg_tb_size=378 qemu: flush code_size=44748344 nb_tbs=80974 avg_tb_size=367 qemu: flush code_size=37104248 nb_tbs=67609 avg_tb_size=364 That is, 20 flushes. Note how a static partitioning approach uses the code buffer poorly, leading to many unnecessary flushes. Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Signed-off-by: Emilio G. Cota <cota@braap.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
2017-07-08 02:24:20 +03:00
void tcg_region_init(void);
void tcg_region_reset_all(void);
size_t tcg_code_size(void);
size_t tcg_code_capacity(void);
tcg: track TBs with per-region BST's This paves the way for enabling scalable parallel generation of TCG code. Instead of tracking TBs with a single binary search tree (BST), use a BST for each TCG region, protecting it with a lock. This is as scalable as it gets, since each TCG thread operates on a separate region. The core of this change is the introduction of struct tcg_region_tree, which contains a pointer to a GTree and an associated lock to serialize accesses to it. We then allocate an array of tcg_region_tree's, adding the appropriate padding to avoid false sharing based on qemu_dcache_linesize. Given a tc_ptr, we first find the corresponding region_tree. This is done by special-casing the first and last regions first, since they might be of size != region.size; otherwise we just divide the offset by region.stride. I was worried about this division (several dozen cycles of latency), but profiling shows that this is not a fast path. Note that region.stride is not required to be a power of two; it is only required to be a multiple of the host's page size. Note that with this design we can also provide consistent snapshots about all region trees at once; for instance, tcg_tb_foreach acquires/releases all region_tree locks before/after iterating over them. For this reason we now drop tb_lock in dump_exec_info(). As an alternative I considered implementing a concurrent BST, but this can be tricky to get right, offers no consistent snapshots of the BST, and performance and scalability-wise I don't think it could ever beat having separate GTrees, given that our workload is insert-mostly (all concurrent BST designs I've seen focus, understandably, on making lookups fast, which comes at the expense of convoluted, non-wait-free insertions/removals). Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Signed-off-by: Emilio G. Cota <cota@braap.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
2017-07-26 23:58:05 +03:00
void tcg_tb_insert(TranslationBlock *tb);
void tcg_tb_remove(TranslationBlock *tb);
size_t tcg_tb_phys_invalidate_count(void);
tcg: track TBs with per-region BST's This paves the way for enabling scalable parallel generation of TCG code. Instead of tracking TBs with a single binary search tree (BST), use a BST for each TCG region, protecting it with a lock. This is as scalable as it gets, since each TCG thread operates on a separate region. The core of this change is the introduction of struct tcg_region_tree, which contains a pointer to a GTree and an associated lock to serialize accesses to it. We then allocate an array of tcg_region_tree's, adding the appropriate padding to avoid false sharing based on qemu_dcache_linesize. Given a tc_ptr, we first find the corresponding region_tree. This is done by special-casing the first and last regions first, since they might be of size != region.size; otherwise we just divide the offset by region.stride. I was worried about this division (several dozen cycles of latency), but profiling shows that this is not a fast path. Note that region.stride is not required to be a power of two; it is only required to be a multiple of the host's page size. Note that with this design we can also provide consistent snapshots about all region trees at once; for instance, tcg_tb_foreach acquires/releases all region_tree locks before/after iterating over them. For this reason we now drop tb_lock in dump_exec_info(). As an alternative I considered implementing a concurrent BST, but this can be tricky to get right, offers no consistent snapshots of the BST, and performance and scalability-wise I don't think it could ever beat having separate GTrees, given that our workload is insert-mostly (all concurrent BST designs I've seen focus, understandably, on making lookups fast, which comes at the expense of convoluted, non-wait-free insertions/removals). Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Signed-off-by: Emilio G. Cota <cota@braap.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
2017-07-26 23:58:05 +03:00
TranslationBlock *tcg_tb_lookup(uintptr_t tc_ptr);
void tcg_tb_foreach(GTraverseFunc func, gpointer user_data);
size_t tcg_nb_tbs(void);
tcg: remove tb_lock Use mmap_lock in user-mode to protect TCG state and the page descriptors. In !user-mode, each vCPU has its own TCG state, so no locks needed. Per-page locks are used to protect the page descriptors. Per-TB locks are used in both modes to protect TB jumps. Some notes: - tb_lock is removed from notdirty_mem_write by passing a locked page_collection to tb_invalidate_phys_page_fast. - tcg_tb_lookup/remove/insert/etc have their own internal lock(s), so there is no need to further serialize access to them. - do_tb_flush is run in a safe async context, meaning no other vCPU threads are running. Therefore acquiring mmap_lock there is just to please tools such as thread sanitizer. - Not visible in the diff, but tb_invalidate_phys_page already has an assert_memory_lock. - cpu_io_recompile is !user-only, so no mmap_lock there. - Added mmap_unlock()'s before all siglongjmp's that could be called in user-mode while mmap_lock is held. + Added an assert for !have_mmap_lock() after returning from the longjmp in cpu_exec, just like we do in cpu_exec_step_atomic. Performance numbers before/after: Host: AMD Opteron(tm) Processor 6376 ubuntu 17.04 ppc64 bootup+shutdown time 700 +-+--+----+------+------------+-----------+------------*--+-+ | + + + + + *B | | before ***B*** ** * | |tb lock removal ###D### *** | 600 +-+ *** +-+ | ** # | | *B* #D | | *** * ## | 500 +-+ *** ### +-+ | * *** ### | | *B* # ## | | ** * #D# | 400 +-+ ** ## +-+ | ** ### | | ** ## | | ** # ## | 300 +-+ * B* #D# +-+ | B *** ### | | * ** #### | | * *** ### | 200 +-+ B *B #D# +-+ | #B* * ## # | | #* ## | | + D##D# + + + + | 100 +-+--+----+------+------------+-----------+------------+--+-+ 1 8 16 Guest CPUs 48 64 png: https://imgur.com/HwmBHXe debian jessie aarch64 bootup+shutdown time 90 +-+--+-----+-----+------------+------------+------------+--+-+ | + + + + + + | | before ***B*** B | 80 +tb lock removal ###D### **D +-+ | **### | | **## | 70 +-+ ** # +-+ | ** ## | | ** # | 60 +-+ *B ## +-+ | ** ## | | *** #D | 50 +-+ *** ## +-+ | * ** ### | | **B* ### | 40 +-+ **** # ## +-+ | **** #D# | | ***B** ### | 30 +-+ B***B** #### +-+ | B * * # ### | | B ###D# | 20 +-+ D ##D## +-+ | D# | | + + + + + + | 10 +-+--+-----+-----+------------+------------+------------+--+-+ 1 8 16 Guest CPUs 48 64 png: https://imgur.com/iGpGFtv The gains are high for 4-8 CPUs. Beyond that point, however, unrelated lock contention significantly hurts scalability. Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Signed-off-by: Emilio G. Cota <cota@braap.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
2017-08-05 06:46:31 +03:00
/* user-mode: Called with mmap_lock held. */
static inline void *tcg_malloc(int size)
{
TCGContext *s = tcg_ctx;
uint8_t *ptr, *ptr_end;
/* ??? This is a weak placeholder for minimum malloc alignment. */
size = QEMU_ALIGN_UP(size, 8);
ptr = s->pool_cur;
ptr_end = ptr + size;
if (unlikely(ptr_end > s->pool_end)) {
return tcg_malloc_internal(tcg_ctx, size);
} else {
s->pool_cur = ptr_end;
return ptr;
}
}
void tcg_context_init(TCGContext *s);
tcg: enable multiple TCG contexts in softmmu This enables parallel TCG code generation. However, we do not take advantage of it yet since tb_lock is still held during tb_gen_code. In user-mode we use a single TCG context; see the documentation added to tcg_region_init for the rationale. Note that targets do not need any conversion: targets initialize a TCGContext (e.g. defining TCG globals), and after this initialization has finished, the context is cloned by the vCPU threads, each of them keeping a separate copy. TCG threads claim one entry in tcg_ctxs[] by atomically increasing n_tcg_ctxs. Do not be too annoyed by the subsequent atomic_read's of that variable and tcg_ctxs; they are there just to play nice with analysis tools such as thread sanitizer. Note that we do not allocate an array of contexts (we allocate an array of pointers instead) because when tcg_context_init is called, we do not know yet how many contexts we'll use since the bool behind qemu_tcg_mttcg_enabled() isn't set yet. Previous patches folded some TCG globals into TCGContext. The non-const globals remaining are only set at init time, i.e. before the TCG threads are spawned. Here is a list of these set-at-init-time globals under tcg/: Only written by tcg_context_init: - indirect_reg_alloc_order - tcg_op_defs Only written by tcg_target_init (called from tcg_context_init): - tcg_target_available_regs - tcg_target_call_clobber_regs - arm: arm_arch, use_idiv_instructions - i386: have_cmov, have_bmi1, have_bmi2, have_lzcnt, have_movbe, have_popcnt - mips: use_movnz_instructions, use_mips32_instructions, use_mips32r2_instructions, got_sigill (tcg_target_detect_isa) - ppc: have_isa_2_06, have_isa_3_00, tb_ret_addr - s390: tb_ret_addr, s390_facilities - sparc: qemu_ld_trampoline, qemu_st_trampoline (build_trampolines), use_vis3_instructions Only written by tcg_prologue_init: - 'struct jit_code_entry one_entry' - aarch64: tb_ret_addr - arm: tb_ret_addr - i386: tb_ret_addr, guest_base_flags - ia64: tb_ret_addr - mips: tb_ret_addr, bswap32_addr, bswap32u_addr, bswap64_addr Reviewed-by: Richard Henderson <rth@twiddle.net> Signed-off-by: Emilio G. Cota <cota@braap.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
2017-07-20 01:57:58 +03:00
void tcg_register_thread(void);
void tcg_prologue_init(TCGContext *s);
void tcg_func_start(TCGContext *s);
int tcg_gen_code(TCGContext *s, TranslationBlock *tb);
void tcg_set_frame(TCGContext *s, TCGReg reg, intptr_t start, intptr_t size);
TCGTemp *tcg_global_mem_new_internal(TCGType, TCGv_ptr,
intptr_t, const char *);
TCGTemp *tcg_temp_new_internal(TCGType, bool);
void tcg_temp_free_internal(TCGTemp *);
TCGv_vec tcg_temp_new_vec(TCGType type);
TCGv_vec tcg_temp_new_vec_matching(TCGv_vec match);
static inline void tcg_temp_free_i32(TCGv_i32 arg)
{
tcg_temp_free_internal(tcgv_i32_temp(arg));
}
static inline void tcg_temp_free_i64(TCGv_i64 arg)
{
tcg_temp_free_internal(tcgv_i64_temp(arg));
}
static inline void tcg_temp_free_ptr(TCGv_ptr arg)
{
tcg_temp_free_internal(tcgv_ptr_temp(arg));
}
static inline void tcg_temp_free_vec(TCGv_vec arg)
{
tcg_temp_free_internal(tcgv_vec_temp(arg));
}
static inline TCGv_i32 tcg_global_mem_new_i32(TCGv_ptr reg, intptr_t offset,
const char *name)
{
TCGTemp *t = tcg_global_mem_new_internal(TCG_TYPE_I32, reg, offset, name);
return temp_tcgv_i32(t);
}
static inline TCGv_i32 tcg_temp_new_i32(void)
{
TCGTemp *t = tcg_temp_new_internal(TCG_TYPE_I32, false);
return temp_tcgv_i32(t);
}
static inline TCGv_i32 tcg_temp_local_new_i32(void)
{
TCGTemp *t = tcg_temp_new_internal(TCG_TYPE_I32, true);
return temp_tcgv_i32(t);
}
static inline TCGv_i64 tcg_global_mem_new_i64(TCGv_ptr reg, intptr_t offset,
const char *name)
{
TCGTemp *t = tcg_global_mem_new_internal(TCG_TYPE_I64, reg, offset, name);
return temp_tcgv_i64(t);
}
static inline TCGv_i64 tcg_temp_new_i64(void)
{
TCGTemp *t = tcg_temp_new_internal(TCG_TYPE_I64, false);
return temp_tcgv_i64(t);
}
static inline TCGv_i64 tcg_temp_local_new_i64(void)
{
TCGTemp *t = tcg_temp_new_internal(TCG_TYPE_I64, true);
return temp_tcgv_i64(t);
}
static inline TCGv_ptr tcg_global_mem_new_ptr(TCGv_ptr reg, intptr_t offset,
const char *name)
{
TCGTemp *t = tcg_global_mem_new_internal(TCG_TYPE_PTR, reg, offset, name);
return temp_tcgv_ptr(t);
}
static inline TCGv_ptr tcg_temp_new_ptr(void)
{
TCGTemp *t = tcg_temp_new_internal(TCG_TYPE_PTR, false);
return temp_tcgv_ptr(t);
}
static inline TCGv_ptr tcg_temp_local_new_ptr(void)
{
TCGTemp *t = tcg_temp_new_internal(TCG_TYPE_PTR, true);
return temp_tcgv_ptr(t);
}
#if defined(CONFIG_DEBUG_TCG)
/* If you call tcg_clear_temp_count() at the start of a section of
* code which is not supposed to leak any TCG temporaries, then
* calling tcg_check_temp_count() at the end of the section will
* return 1 if the section did in fact leak a temporary.
*/
void tcg_clear_temp_count(void);
int tcg_check_temp_count(void);
#else
#define tcg_clear_temp_count() do { } while (0)
#define tcg_check_temp_count() 0
#endif
int64_t tcg_cpu_exec_time(void);
void tcg_dump_info(FILE *f, fprintf_function cpu_fprintf);
void tcg_dump_op_count(FILE *f, fprintf_function cpu_fprintf);
#define TCG_CT_ALIAS 0x80
#define TCG_CT_IALIAS 0x40
#define TCG_CT_NEWREG 0x20 /* output requires a new register */
#define TCG_CT_REG 0x01
#define TCG_CT_CONST 0x02 /* any constant of register size */
typedef struct TCGArgConstraint {
uint16_t ct;
uint8_t alias_index;
union {
TCGRegSet regs;
} u;
} TCGArgConstraint;
#define TCG_MAX_OP_ARGS 16
/* Bits for TCGOpDef->flags, 8 bits available. */
enum {
/* Instruction defines the end of a basic block. */
TCG_OPF_BB_END = 0x01,
/* Instruction clobbers call registers and potentially update globals. */
TCG_OPF_CALL_CLOBBER = 0x02,
/* Instruction has side effects: it cannot be removed if its outputs
are not used, and might trigger exceptions. */
TCG_OPF_SIDE_EFFECTS = 0x04,
/* Instruction operands are 64-bits (otherwise 32-bits). */
TCG_OPF_64BIT = 0x08,
/* Instruction is optional and not implemented by the host, or insn
is generic and should not be implemened by the host. */
TCG_OPF_NOT_PRESENT = 0x10,
/* Instruction operands are vectors. */
TCG_OPF_VECTOR = 0x20,
};
typedef struct TCGOpDef {
const char *name;
uint8_t nb_oargs, nb_iargs, nb_cargs, nb_args;
uint8_t flags;
TCGArgConstraint *args_ct;
int *sorted_args;
#if defined(CONFIG_DEBUG_TCG)
int used;
#endif
} TCGOpDef;
extern TCGOpDef tcg_op_defs[];
extern const size_t tcg_op_defs_max;
typedef struct TCGTargetOpDef {
TCGOpcode op;
const char *args_ct_str[TCG_MAX_OP_ARGS];
} TCGTargetOpDef;
#define tcg_abort() \
do {\
fprintf(stderr, "%s:%d: tcg fatal error\n", __FILE__, __LINE__);\
abort();\
} while (0)
bool tcg_op_supported(TCGOpcode op);
void tcg_gen_callN(void *func, TCGTemp *ret, int nargs, TCGTemp **args);
TCGOp *tcg_emit_op(TCGOpcode opc);
void tcg_op_remove(TCGContext *s, TCGOp *op);
TCGOp *tcg_op_insert_before(TCGContext *s, TCGOp *op, TCGOpcode opc, int narg);
TCGOp *tcg_op_insert_after(TCGContext *s, TCGOp *op, TCGOpcode opc, int narg);
void tcg_optimize(TCGContext *s);
/* only used for debugging purposes */
void tcg_dump_ops(TCGContext *s);
TCGv_i32 tcg_const_i32(int32_t val);
TCGv_i64 tcg_const_i64(int64_t val);
TCGv_i32 tcg_const_local_i32(int32_t val);
TCGv_i64 tcg_const_local_i64(int64_t val);
TCGv_vec tcg_const_zeros_vec(TCGType);
TCGv_vec tcg_const_ones_vec(TCGType);
TCGv_vec tcg_const_zeros_vec_matching(TCGv_vec);
TCGv_vec tcg_const_ones_vec_matching(TCGv_vec);
#if UINTPTR_MAX == UINT32_MAX
# define tcg_const_ptr(x) ((TCGv_ptr)tcg_const_i32((intptr_t)(x)))
# define tcg_const_local_ptr(x) ((TCGv_ptr)tcg_const_local_i32((intptr_t)(x)))
#else
# define tcg_const_ptr(x) ((TCGv_ptr)tcg_const_i64((intptr_t)(x)))
# define tcg_const_local_ptr(x) ((TCGv_ptr)tcg_const_local_i64((intptr_t)(x)))
#endif
TCGLabel *gen_new_label(void);
/**
* label_arg
* @l: label
*
* Encode a label for storage in the TCG opcode stream.
*/
static inline TCGArg label_arg(TCGLabel *l)
{
return (uintptr_t)l;
}
/**
* arg_label
* @i: value
*
* The opposite of label_arg. Retrieve a label from the
* encoding of the TCG opcode stream.
*/
static inline TCGLabel *arg_label(TCGArg i)
{
return (TCGLabel *)(uintptr_t)i;
}
/**
* tcg_ptr_byte_diff
* @a, @b: addresses to be differenced
*
* There are many places within the TCG backends where we need a byte
* difference between two pointers. While this can be accomplished
* with local casting, it's easy to get wrong -- especially if one is
* concerned with the signedness of the result.
*
* This version relies on GCC's void pointer arithmetic to get the
* correct result.
*/
static inline ptrdiff_t tcg_ptr_byte_diff(void *a, void *b)
{
return a - b;
}
/**
* tcg_pcrel_diff
* @s: the tcg context
* @target: address of the target
*
* Produce a pc-relative difference, from the current code_ptr
* to the destination address.
*/
static inline ptrdiff_t tcg_pcrel_diff(TCGContext *s, void *target)
{
return tcg_ptr_byte_diff(target, s->code_ptr);
}
/**
* tcg_current_code_size
* @s: the tcg context
*
* Compute the current code size within the translation block.
* This is used to fill in qemu's data structures for goto_tb.
*/
static inline size_t tcg_current_code_size(TCGContext *s)
{
return tcg_ptr_byte_diff(s->code_ptr, s->code_buf);
}
/* Combine the TCGMemOp and mmu_idx parameters into a single value. */
typedef uint32_t TCGMemOpIdx;
/**
* make_memop_idx
* @op: memory operation
* @idx: mmu index
*
* Encode these values into a single parameter.
*/
static inline TCGMemOpIdx make_memop_idx(TCGMemOp op, unsigned idx)
{
tcg_debug_assert(idx <= 15);
return (op << 4) | idx;
}
/**
* get_memop
* @oi: combined op/idx parameter
*
* Extract the memory operation from the combined value.
*/
static inline TCGMemOp get_memop(TCGMemOpIdx oi)
{
return oi >> 4;
}
/**
* get_mmuidx
* @oi: combined op/idx parameter
*
* Extract the mmu index from the combined value.
*/
static inline unsigned get_mmuidx(TCGMemOpIdx oi)
{
return oi & 15;
}
/**
* tcg_qemu_tb_exec:
* @env: pointer to CPUArchState for the CPU
* @tb_ptr: address of generated code for the TB to execute
*
* Start executing code from a given translation block.
* Where translation blocks have been linked, execution
* may proceed from the given TB into successive ones.
* Control eventually returns only when some action is needed
* from the top-level loop: either control must pass to a TB
* which has not yet been directly linked, or an asynchronous
* event such as an interrupt needs handling.
*
* Return: The return value is the value passed to the corresponding
* tcg_gen_exit_tb() at translation time of the last TB attempted to execute.
* The value is either zero or a 4-byte aligned pointer to that TB combined
* with additional information in its two least significant bits. The
* additional information is encoded as follows:
* 0, 1: the link between this TB and the next is via the specified
* TB index (0 or 1). That is, we left the TB via (the equivalent
* of) "goto_tb <index>". The main loop uses this to determine
* how to link the TB just executed to the next.
* 2: we are using instruction counting code generation, and we
* did not start executing this TB because the instruction counter
* would hit zero midway through it. In this case the pointer
* returned is the TB we were about to execute, and the caller must
* arrange to execute the remaining count of instructions.
* 3: we stopped because the CPU's exit_request flag was set
* (usually meaning that there is an interrupt that needs to be
* handled). The pointer returned is the TB we were about to execute
* when we noticed the pending exit request.
*
* If the bottom two bits indicate an exit-via-index then the CPU
* state is correctly synchronised and ready for execution of the next
* TB (and in particular the guest PC is the address to execute next).
* Otherwise, we gave up on execution of this TB before it started, and
* the caller must fix up the CPU state by calling the CPU's
* synchronize_from_tb() method with the TB pointer we return (falling
* back to calling the CPU's set_pc method with tb->pb if no
* synchronize_from_tb() method exists).
*
* Note that TCG targets may use a different definition of tcg_qemu_tb_exec
* to this default (which just calls the prologue.code emitted by
* tcg_target_qemu_prologue()).
*/
#define TB_EXIT_MASK 3
#define TB_EXIT_IDX0 0
#define TB_EXIT_IDX1 1
#define TB_EXIT_IDXMAX 1
#define TB_EXIT_REQUESTED 3
#ifdef HAVE_TCG_QEMU_TB_EXEC
uintptr_t tcg_qemu_tb_exec(CPUArchState *env, uint8_t *tb_ptr);
#else
# define tcg_qemu_tb_exec(env, tb_ptr) \
((uintptr_t (*)(void *, void *))tcg_ctx->code_gen_prologue)(env, tb_ptr)
#endif
void tcg_register_jit(void *buf, size_t buf_size);
#if TCG_TARGET_MAYBE_vec
/* Return zero if the tuple (opc, type, vece) is unsupportable;
return > 0 if it is directly supportable;
return < 0 if we must call tcg_expand_vec_op. */
int tcg_can_emit_vec_op(TCGOpcode, TCGType, unsigned);
#else
static inline int tcg_can_emit_vec_op(TCGOpcode o, TCGType t, unsigned ve)
{
return 0;
}
#endif
/* Expand the tuple (opc, type, vece) on the given arguments. */
void tcg_expand_vec_op(TCGOpcode, TCGType, unsigned, TCGArg, ...);
/* Replicate a constant C accoring to the log2 of the element size. */
uint64_t dup_const(unsigned vece, uint64_t c);
#define dup_const(VECE, C) \
(__builtin_constant_p(VECE) \
? ( (VECE) == MO_8 ? 0x0101010101010101ull * (uint8_t)(C) \
: (VECE) == MO_16 ? 0x0001000100010001ull * (uint16_t)(C) \
: (VECE) == MO_32 ? 0x0000000100000001ull * (uint32_t)(C) \
: dup_const(VECE, C)) \
: dup_const(VECE, C))
/*
* Memory helpers that will be used by TCG generated code.
*/
#ifdef CONFIG_SOFTMMU
/* Value zero-extended to tcg register size. */
tcg_target_ulong helper_ret_ldub_mmu(CPUArchState *env, target_ulong addr,
TCGMemOpIdx oi, uintptr_t retaddr);
tcg_target_ulong helper_le_lduw_mmu(CPUArchState *env, target_ulong addr,
TCGMemOpIdx oi, uintptr_t retaddr);
tcg_target_ulong helper_le_ldul_mmu(CPUArchState *env, target_ulong addr,
TCGMemOpIdx oi, uintptr_t retaddr);
uint64_t helper_le_ldq_mmu(CPUArchState *env, target_ulong addr,
TCGMemOpIdx oi, uintptr_t retaddr);
tcg_target_ulong helper_be_lduw_mmu(CPUArchState *env, target_ulong addr,
TCGMemOpIdx oi, uintptr_t retaddr);
tcg_target_ulong helper_be_ldul_mmu(CPUArchState *env, target_ulong addr,
TCGMemOpIdx oi, uintptr_t retaddr);
uint64_t helper_be_ldq_mmu(CPUArchState *env, target_ulong addr,
TCGMemOpIdx oi, uintptr_t retaddr);
/* Value sign-extended to tcg register size. */
tcg_target_ulong helper_ret_ldsb_mmu(CPUArchState *env, target_ulong addr,
TCGMemOpIdx oi, uintptr_t retaddr);
tcg_target_ulong helper_le_ldsw_mmu(CPUArchState *env, target_ulong addr,
TCGMemOpIdx oi, uintptr_t retaddr);
tcg_target_ulong helper_le_ldsl_mmu(CPUArchState *env, target_ulong addr,
TCGMemOpIdx oi, uintptr_t retaddr);
tcg_target_ulong helper_be_ldsw_mmu(CPUArchState *env, target_ulong addr,
TCGMemOpIdx oi, uintptr_t retaddr);
tcg_target_ulong helper_be_ldsl_mmu(CPUArchState *env, target_ulong addr,
TCGMemOpIdx oi, uintptr_t retaddr);
void helper_ret_stb_mmu(CPUArchState *env, target_ulong addr, uint8_t val,
TCGMemOpIdx oi, uintptr_t retaddr);
void helper_le_stw_mmu(CPUArchState *env, target_ulong addr, uint16_t val,
TCGMemOpIdx oi, uintptr_t retaddr);
void helper_le_stl_mmu(CPUArchState *env, target_ulong addr, uint32_t val,
TCGMemOpIdx oi, uintptr_t retaddr);
void helper_le_stq_mmu(CPUArchState *env, target_ulong addr, uint64_t val,
TCGMemOpIdx oi, uintptr_t retaddr);
void helper_be_stw_mmu(CPUArchState *env, target_ulong addr, uint16_t val,
TCGMemOpIdx oi, uintptr_t retaddr);
void helper_be_stl_mmu(CPUArchState *env, target_ulong addr, uint32_t val,
TCGMemOpIdx oi, uintptr_t retaddr);
void helper_be_stq_mmu(CPUArchState *env, target_ulong addr, uint64_t val,
TCGMemOpIdx oi, uintptr_t retaddr);
uint8_t helper_ret_ldb_cmmu(CPUArchState *env, target_ulong addr,
TCGMemOpIdx oi, uintptr_t retaddr);
uint16_t helper_le_ldw_cmmu(CPUArchState *env, target_ulong addr,
TCGMemOpIdx oi, uintptr_t retaddr);
uint32_t helper_le_ldl_cmmu(CPUArchState *env, target_ulong addr,
TCGMemOpIdx oi, uintptr_t retaddr);
uint64_t helper_le_ldq_cmmu(CPUArchState *env, target_ulong addr,
TCGMemOpIdx oi, uintptr_t retaddr);
uint16_t helper_be_ldw_cmmu(CPUArchState *env, target_ulong addr,
TCGMemOpIdx oi, uintptr_t retaddr);
uint32_t helper_be_ldl_cmmu(CPUArchState *env, target_ulong addr,
TCGMemOpIdx oi, uintptr_t retaddr);
uint64_t helper_be_ldq_cmmu(CPUArchState *env, target_ulong addr,
TCGMemOpIdx oi, uintptr_t retaddr);
/* Temporary aliases until backends are converted. */
#ifdef TARGET_WORDS_BIGENDIAN
# define helper_ret_ldsw_mmu helper_be_ldsw_mmu
# define helper_ret_lduw_mmu helper_be_lduw_mmu
# define helper_ret_ldsl_mmu helper_be_ldsl_mmu
# define helper_ret_ldul_mmu helper_be_ldul_mmu
# define helper_ret_ldl_mmu helper_be_ldul_mmu
# define helper_ret_ldq_mmu helper_be_ldq_mmu
# define helper_ret_stw_mmu helper_be_stw_mmu
# define helper_ret_stl_mmu helper_be_stl_mmu
# define helper_ret_stq_mmu helper_be_stq_mmu
# define helper_ret_ldw_cmmu helper_be_ldw_cmmu
# define helper_ret_ldl_cmmu helper_be_ldl_cmmu
# define helper_ret_ldq_cmmu helper_be_ldq_cmmu
#else
# define helper_ret_ldsw_mmu helper_le_ldsw_mmu
# define helper_ret_lduw_mmu helper_le_lduw_mmu
# define helper_ret_ldsl_mmu helper_le_ldsl_mmu
# define helper_ret_ldul_mmu helper_le_ldul_mmu
# define helper_ret_ldl_mmu helper_le_ldul_mmu
# define helper_ret_ldq_mmu helper_le_ldq_mmu
# define helper_ret_stw_mmu helper_le_stw_mmu
# define helper_ret_stl_mmu helper_le_stl_mmu
# define helper_ret_stq_mmu helper_le_stq_mmu
# define helper_ret_ldw_cmmu helper_le_ldw_cmmu
# define helper_ret_ldl_cmmu helper_le_ldl_cmmu
# define helper_ret_ldq_cmmu helper_le_ldq_cmmu
#endif
uint32_t helper_atomic_cmpxchgb_mmu(CPUArchState *env, target_ulong addr,
uint32_t cmpv, uint32_t newv,
TCGMemOpIdx oi, uintptr_t retaddr);
uint32_t helper_atomic_cmpxchgw_le_mmu(CPUArchState *env, target_ulong addr,
uint32_t cmpv, uint32_t newv,
TCGMemOpIdx oi, uintptr_t retaddr);
uint32_t helper_atomic_cmpxchgl_le_mmu(CPUArchState *env, target_ulong addr,
uint32_t cmpv, uint32_t newv,
TCGMemOpIdx oi, uintptr_t retaddr);
uint64_t helper_atomic_cmpxchgq_le_mmu(CPUArchState *env, target_ulong addr,
uint64_t cmpv, uint64_t newv,
TCGMemOpIdx oi, uintptr_t retaddr);
uint32_t helper_atomic_cmpxchgw_be_mmu(CPUArchState *env, target_ulong addr,
uint32_t cmpv, uint32_t newv,
TCGMemOpIdx oi, uintptr_t retaddr);
uint32_t helper_atomic_cmpxchgl_be_mmu(CPUArchState *env, target_ulong addr,
uint32_t cmpv, uint32_t newv,
TCGMemOpIdx oi, uintptr_t retaddr);
uint64_t helper_atomic_cmpxchgq_be_mmu(CPUArchState *env, target_ulong addr,
uint64_t cmpv, uint64_t newv,
TCGMemOpIdx oi, uintptr_t retaddr);
#define GEN_ATOMIC_HELPER(NAME, TYPE, SUFFIX) \
TYPE helper_atomic_ ## NAME ## SUFFIX ## _mmu \
(CPUArchState *env, target_ulong addr, TYPE val, \
TCGMemOpIdx oi, uintptr_t retaddr);
#ifdef CONFIG_ATOMIC64
#define GEN_ATOMIC_HELPER_ALL(NAME) \
GEN_ATOMIC_HELPER(NAME, uint32_t, b) \
GEN_ATOMIC_HELPER(NAME, uint32_t, w_le) \
GEN_ATOMIC_HELPER(NAME, uint32_t, w_be) \
GEN_ATOMIC_HELPER(NAME, uint32_t, l_le) \
GEN_ATOMIC_HELPER(NAME, uint32_t, l_be) \
GEN_ATOMIC_HELPER(NAME, uint64_t, q_le) \
GEN_ATOMIC_HELPER(NAME, uint64_t, q_be)
#else
#define GEN_ATOMIC_HELPER_ALL(NAME) \
GEN_ATOMIC_HELPER(NAME, uint32_t, b) \
GEN_ATOMIC_HELPER(NAME, uint32_t, w_le) \
GEN_ATOMIC_HELPER(NAME, uint32_t, w_be) \
GEN_ATOMIC_HELPER(NAME, uint32_t, l_le) \
GEN_ATOMIC_HELPER(NAME, uint32_t, l_be)
#endif
GEN_ATOMIC_HELPER_ALL(fetch_add)
GEN_ATOMIC_HELPER_ALL(fetch_sub)
GEN_ATOMIC_HELPER_ALL(fetch_and)
GEN_ATOMIC_HELPER_ALL(fetch_or)
GEN_ATOMIC_HELPER_ALL(fetch_xor)
GEN_ATOMIC_HELPER_ALL(fetch_smin)
GEN_ATOMIC_HELPER_ALL(fetch_umin)
GEN_ATOMIC_HELPER_ALL(fetch_smax)
GEN_ATOMIC_HELPER_ALL(fetch_umax)
GEN_ATOMIC_HELPER_ALL(add_fetch)
GEN_ATOMIC_HELPER_ALL(sub_fetch)
GEN_ATOMIC_HELPER_ALL(and_fetch)
GEN_ATOMIC_HELPER_ALL(or_fetch)
GEN_ATOMIC_HELPER_ALL(xor_fetch)
GEN_ATOMIC_HELPER_ALL(smin_fetch)
GEN_ATOMIC_HELPER_ALL(umin_fetch)
GEN_ATOMIC_HELPER_ALL(smax_fetch)
GEN_ATOMIC_HELPER_ALL(umax_fetch)
GEN_ATOMIC_HELPER_ALL(xchg)
#undef GEN_ATOMIC_HELPER_ALL
#undef GEN_ATOMIC_HELPER
#endif /* CONFIG_SOFTMMU */
/*
* These aren't really a "proper" helpers because TCG cannot manage Int128.
* However, use the same format as the others, for use by the backends.
*
* The cmpxchg functions are only defined if HAVE_CMPXCHG128;
* the ld/st functions are only defined if HAVE_ATOMIC128,
* as defined by <qemu/atomic128.h>.
*/
Int128 helper_atomic_cmpxchgo_le_mmu(CPUArchState *env, target_ulong addr,
Int128 cmpv, Int128 newv,
TCGMemOpIdx oi, uintptr_t retaddr);
Int128 helper_atomic_cmpxchgo_be_mmu(CPUArchState *env, target_ulong addr,
Int128 cmpv, Int128 newv,
TCGMemOpIdx oi, uintptr_t retaddr);
Int128 helper_atomic_ldo_le_mmu(CPUArchState *env, target_ulong addr,
TCGMemOpIdx oi, uintptr_t retaddr);
Int128 helper_atomic_ldo_be_mmu(CPUArchState *env, target_ulong addr,
TCGMemOpIdx oi, uintptr_t retaddr);
void helper_atomic_sto_le_mmu(CPUArchState *env, target_ulong addr, Int128 val,
TCGMemOpIdx oi, uintptr_t retaddr);
void helper_atomic_sto_be_mmu(CPUArchState *env, target_ulong addr, Int128 val,
TCGMemOpIdx oi, uintptr_t retaddr);
#endif /* TCG_H */