qemu/accel/tcg/softmmu_template.h
Emilio G. Cota 403f290c06 cputlb: read CPUTLBEntry.addr_write atomically
Updates can come from other threads, so readers that do not
take tlb_lock must use atomic_read to avoid undefined
behaviour (UB).

This completes the conversion to tlb_lock. This conversion results
on average in no performance loss, as the following experiments
(run on an Intel i7-6700K CPU @ 4.00GHz) show.

1. aarch64 bootup+shutdown test:

- Before:
 Performance counter stats for 'taskset -c 0 ../img/aarch64/die.sh' (10 runs):

       7487.087786      task-clock (msec)         #    0.998 CPUs utilized            ( +-  0.12% )
    31,574,905,303      cycles                    #    4.217 GHz                      ( +-  0.12% )
    57,097,908,812      instructions              #    1.81  insns per cycle          ( +-  0.08% )
    10,255,415,367      branches                  # 1369.747 M/sec                    ( +-  0.08% )
       173,278,962      branch-misses             #    1.69% of all branches          ( +-  0.18% )

       7.504481349 seconds time elapsed                                          ( +-  0.14% )

- After:
 Performance counter stats for 'taskset -c 0 ../img/aarch64/die.sh' (10 runs):

       7462.441328      task-clock (msec)         #    0.998 CPUs utilized            ( +-  0.07% )
    31,478,476,520      cycles                    #    4.218 GHz                      ( +-  0.07% )
    57,017,330,084      instructions              #    1.81  insns per cycle          ( +-  0.05% )
    10,251,929,667      branches                  # 1373.804 M/sec                    ( +-  0.05% )
       173,023,787      branch-misses             #    1.69% of all branches          ( +-  0.11% )

       7.474970463 seconds time elapsed                                          ( +-  0.07% )

2. SPEC06int:
                                              SPEC06int (test set)
                                           [Y axis: Speedup over master]
  1.15 +-+----+------+------+------+------+------+-------+------+------+------+------+------+------+----+-+
       |                                                                                                  |
   1.1 +-+.................................+++.............................+  tlb-lock-v2 (m+++x)       +-+
       |                                +++ |                   +++        tlb-lock-v3 (spinl|ck)         |
       |                    +++          |  |     +++    +++     |                           |            |
  1.05 +-+....+++...........####.........|####.+++.|......|.....###....+++...........+++....###.........+-+
       |      ###         ++#| #         |# |# ***### +++### +++#+#     |     +++     |     #|#    ###    |
     1 +-+++***+#++++####+++#++#++++++++++#++#+*+*++#++++#+#+****+#++++###++++###++++###++++#+#++++#+#+++-+
       |    *+* #    #++# ***  #   #### ***  # * *++# ****+# *| * # ****|#   |# #    #|#    #+#    # #    |
  0.95 +-+..*.*.#....#..#.*|*..#...#..#.*|*..#.*.*..#.*|.*.#.*++*.#.*++*+#.****.#....#+#....#.#..++#.#..+-+
       |    * * #    #  # *|*  #   #  # *|*  # * *  # *++* # *  * # *  * # * |* #  ++# #    # #  *** #    |
       |    * * #  ++#  # *+*  #   #  # *|*  # * *  # *  * # *  * # *  * # *++* # **** #  ++# #  * * #    |
   0.9 +-+..*.*.#...|#..#.*.*..#.++#..#.*|*..#.*.*..#.*..*.#.*..*.#.*..*.#.*..*.#.*.|*.#...|#.#..*.*.#..+-+
       |    * * #  ***  # * *  #  |#  # *+*  # * *  # *  * # *  * # *  * # *  * # *++* #   |# #  * * #    |
  0.85 +-+..*.*.#..*|*..#.*.*..#.***..#.*.*..#.*.*..#.*..*.#.*..*.#.*..*.#.*..*.#.*..*.#.****.#..*.*.#..+-+
       |    * * #  *+*  # * *  # *|*  # * *  # * *  # *  * # *  * # *  * # *  * # *  * # * |* #  * * #    |
       |    * * #  * *  # * *  # *+*  # * *  # * *  # *  * # *  * # *  * # *  * # *  * # * |* #  * * #    |
   0.8 +-+..*.*.#..*.*..#.*.*..#.*.*..#.*.*..#.*.*..#.*..*.#.*..*.#.*..*.#.*..*.#.*..*.#.*++*.#..*.*.#..+-+
       |    * * #  * *  # * *  # * *  # * *  # * *  # *  * # *  * # *  * # *  * # *  * # *  * #  * * #    |
  0.75 +-+--***##--***###-***###-***###-***###-***###-****##-****##-****##-****##-****##-****##--***##--+-+
 400.perlben401.bzip2403.gcc429.m445.gob456.hmme45462.libqua464.h26471.omnet473483.xalancbmkgeomean

  png: https://imgur.com/a/BHzpPTW

Notes:
- tlb-lock-v2 corresponds to an implementation with a mutex.
- tlb-lock-v3 corresponds to the current implementation, i.e.
  a spinlock and a single lock acquisition in tlb_set_page_with_attrs.

Signed-off-by: Emilio G. Cota <cota@braap.org>
Message-Id: <20181016153840.25877-1-cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
2018-10-18 19:46:53 -07:00

447 lines
15 KiB
C

/*
* Software MMU support
*
* Generate helpers used by TCG for qemu_ld/st ops and code load
* functions.
*
* Included from target op helpers and exec.c.
*
* Copyright (c) 2003 Fabrice Bellard
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2 of the License, or (at your option) any later version.
*
* This library is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with this library; if not, see <http://www.gnu.org/licenses/>.
*/
#if DATA_SIZE == 8
#define SUFFIX q
#define LSUFFIX q
#define SDATA_TYPE int64_t
#define DATA_TYPE uint64_t
#elif DATA_SIZE == 4
#define SUFFIX l
#define LSUFFIX l
#define SDATA_TYPE int32_t
#define DATA_TYPE uint32_t
#elif DATA_SIZE == 2
#define SUFFIX w
#define LSUFFIX uw
#define SDATA_TYPE int16_t
#define DATA_TYPE uint16_t
#elif DATA_SIZE == 1
#define SUFFIX b
#define LSUFFIX ub
#define SDATA_TYPE int8_t
#define DATA_TYPE uint8_t
#else
#error unsupported data size
#endif
/* For the benefit of TCG generated code, we want to avoid the complication
of ABI-specific return type promotion and always return a value extended
to the register size of the host. This is tcg_target_long, except in the
case of a 32-bit host and 64-bit data, and for that we always have
uint64_t. Don't bother with this widened value for SOFTMMU_CODE_ACCESS. */
#if defined(SOFTMMU_CODE_ACCESS) || DATA_SIZE == 8
# define WORD_TYPE DATA_TYPE
# define USUFFIX SUFFIX
#else
# define WORD_TYPE tcg_target_ulong
# define USUFFIX glue(u, SUFFIX)
# define SSUFFIX glue(s, SUFFIX)
#endif
#ifdef SOFTMMU_CODE_ACCESS
#define READ_ACCESS_TYPE MMU_INST_FETCH
#define ADDR_READ addr_code
#else
#define READ_ACCESS_TYPE MMU_DATA_LOAD
#define ADDR_READ addr_read
#endif
#if DATA_SIZE == 8
# define BSWAP(X) bswap64(X)
#elif DATA_SIZE == 4
# define BSWAP(X) bswap32(X)
#elif DATA_SIZE == 2
# define BSWAP(X) bswap16(X)
#else
# define BSWAP(X) (X)
#endif
#if DATA_SIZE == 1
# define helper_le_ld_name glue(glue(helper_ret_ld, USUFFIX), MMUSUFFIX)
# define helper_be_ld_name helper_le_ld_name
# define helper_le_lds_name glue(glue(helper_ret_ld, SSUFFIX), MMUSUFFIX)
# define helper_be_lds_name helper_le_lds_name
# define helper_le_st_name glue(glue(helper_ret_st, SUFFIX), MMUSUFFIX)
# define helper_be_st_name helper_le_st_name
#else
# define helper_le_ld_name glue(glue(helper_le_ld, USUFFIX), MMUSUFFIX)
# define helper_be_ld_name glue(glue(helper_be_ld, USUFFIX), MMUSUFFIX)
# define helper_le_lds_name glue(glue(helper_le_ld, SSUFFIX), MMUSUFFIX)
# define helper_be_lds_name glue(glue(helper_be_ld, SSUFFIX), MMUSUFFIX)
# define helper_le_st_name glue(glue(helper_le_st, SUFFIX), MMUSUFFIX)
# define helper_be_st_name glue(glue(helper_be_st, SUFFIX), MMUSUFFIX)
#endif
#ifndef SOFTMMU_CODE_ACCESS
static inline DATA_TYPE glue(io_read, SUFFIX)(CPUArchState *env,
size_t mmu_idx, size_t index,
target_ulong addr,
uintptr_t retaddr,
bool recheck,
MMUAccessType access_type)
{
CPUIOTLBEntry *iotlbentry = &env->iotlb[mmu_idx][index];
return io_readx(env, iotlbentry, mmu_idx, addr, retaddr, recheck,
access_type, DATA_SIZE);
}
#endif
WORD_TYPE helper_le_ld_name(CPUArchState *env, target_ulong addr,
TCGMemOpIdx oi, uintptr_t retaddr)
{
uintptr_t mmu_idx = get_mmuidx(oi);
uintptr_t index = tlb_index(env, mmu_idx, addr);
CPUTLBEntry *entry = tlb_entry(env, mmu_idx, addr);
target_ulong tlb_addr = entry->ADDR_READ;
unsigned a_bits = get_alignment_bits(get_memop(oi));
uintptr_t haddr;
DATA_TYPE res;
if (addr & ((1 << a_bits) - 1)) {
cpu_unaligned_access(ENV_GET_CPU(env), addr, READ_ACCESS_TYPE,
mmu_idx, retaddr);
}
/* If the TLB entry is for a different page, reload and try again. */
if (!tlb_hit(tlb_addr, addr)) {
if (!VICTIM_TLB_HIT(ADDR_READ, addr)) {
tlb_fill(ENV_GET_CPU(env), addr, DATA_SIZE, READ_ACCESS_TYPE,
mmu_idx, retaddr);
}
tlb_addr = entry->ADDR_READ;
}
/* Handle an IO access. */
if (unlikely(tlb_addr & ~TARGET_PAGE_MASK)) {
if ((addr & (DATA_SIZE - 1)) != 0) {
goto do_unaligned_access;
}
/* ??? Note that the io helpers always read data in the target
byte ordering. We should push the LE/BE request down into io. */
res = glue(io_read, SUFFIX)(env, mmu_idx, index, addr, retaddr,
tlb_addr & TLB_RECHECK,
READ_ACCESS_TYPE);
res = TGT_LE(res);
return res;
}
/* Handle slow unaligned access (it spans two pages or IO). */
if (DATA_SIZE > 1
&& unlikely((addr & ~TARGET_PAGE_MASK) + DATA_SIZE - 1
>= TARGET_PAGE_SIZE)) {
target_ulong addr1, addr2;
DATA_TYPE res1, res2;
unsigned shift;
do_unaligned_access:
addr1 = addr & ~(DATA_SIZE - 1);
addr2 = addr1 + DATA_SIZE;
res1 = helper_le_ld_name(env, addr1, oi, retaddr);
res2 = helper_le_ld_name(env, addr2, oi, retaddr);
shift = (addr & (DATA_SIZE - 1)) * 8;
/* Little-endian combine. */
res = (res1 >> shift) | (res2 << ((DATA_SIZE * 8) - shift));
return res;
}
haddr = addr + entry->addend;
#if DATA_SIZE == 1
res = glue(glue(ld, LSUFFIX), _p)((uint8_t *)haddr);
#else
res = glue(glue(ld, LSUFFIX), _le_p)((uint8_t *)haddr);
#endif
return res;
}
#if DATA_SIZE > 1
WORD_TYPE helper_be_ld_name(CPUArchState *env, target_ulong addr,
TCGMemOpIdx oi, uintptr_t retaddr)
{
uintptr_t mmu_idx = get_mmuidx(oi);
uintptr_t index = tlb_index(env, mmu_idx, addr);
CPUTLBEntry *entry = tlb_entry(env, mmu_idx, addr);
target_ulong tlb_addr = entry->ADDR_READ;
unsigned a_bits = get_alignment_bits(get_memop(oi));
uintptr_t haddr;
DATA_TYPE res;
if (addr & ((1 << a_bits) - 1)) {
cpu_unaligned_access(ENV_GET_CPU(env), addr, READ_ACCESS_TYPE,
mmu_idx, retaddr);
}
/* If the TLB entry is for a different page, reload and try again. */
if (!tlb_hit(tlb_addr, addr)) {
if (!VICTIM_TLB_HIT(ADDR_READ, addr)) {
tlb_fill(ENV_GET_CPU(env), addr, DATA_SIZE, READ_ACCESS_TYPE,
mmu_idx, retaddr);
}
tlb_addr = entry->ADDR_READ;
}
/* Handle an IO access. */
if (unlikely(tlb_addr & ~TARGET_PAGE_MASK)) {
if ((addr & (DATA_SIZE - 1)) != 0) {
goto do_unaligned_access;
}
/* ??? Note that the io helpers always read data in the target
byte ordering. We should push the LE/BE request down into io. */
res = glue(io_read, SUFFIX)(env, mmu_idx, index, addr, retaddr,
tlb_addr & TLB_RECHECK,
READ_ACCESS_TYPE);
res = TGT_BE(res);
return res;
}
/* Handle slow unaligned access (it spans two pages or IO). */
if (DATA_SIZE > 1
&& unlikely((addr & ~TARGET_PAGE_MASK) + DATA_SIZE - 1
>= TARGET_PAGE_SIZE)) {
target_ulong addr1, addr2;
DATA_TYPE res1, res2;
unsigned shift;
do_unaligned_access:
addr1 = addr & ~(DATA_SIZE - 1);
addr2 = addr1 + DATA_SIZE;
res1 = helper_be_ld_name(env, addr1, oi, retaddr);
res2 = helper_be_ld_name(env, addr2, oi, retaddr);
shift = (addr & (DATA_SIZE - 1)) * 8;
/* Big-endian combine. */
res = (res1 << shift) | (res2 >> ((DATA_SIZE * 8) - shift));
return res;
}
haddr = addr + entry->addend;
res = glue(glue(ld, LSUFFIX), _be_p)((uint8_t *)haddr);
return res;
}
#endif /* DATA_SIZE > 1 */
#ifndef SOFTMMU_CODE_ACCESS
/* Provide signed versions of the load routines as well. We can of course
avoid this for 64-bit data, or for 32-bit data on 32-bit host. */
#if DATA_SIZE * 8 < TCG_TARGET_REG_BITS
WORD_TYPE helper_le_lds_name(CPUArchState *env, target_ulong addr,
TCGMemOpIdx oi, uintptr_t retaddr)
{
return (SDATA_TYPE)helper_le_ld_name(env, addr, oi, retaddr);
}
# if DATA_SIZE > 1
WORD_TYPE helper_be_lds_name(CPUArchState *env, target_ulong addr,
TCGMemOpIdx oi, uintptr_t retaddr)
{
return (SDATA_TYPE)helper_be_ld_name(env, addr, oi, retaddr);
}
# endif
#endif
static inline void glue(io_write, SUFFIX)(CPUArchState *env,
size_t mmu_idx, size_t index,
DATA_TYPE val,
target_ulong addr,
uintptr_t retaddr,
bool recheck)
{
CPUIOTLBEntry *iotlbentry = &env->iotlb[mmu_idx][index];
return io_writex(env, iotlbentry, mmu_idx, val, addr, retaddr,
recheck, DATA_SIZE);
}
void helper_le_st_name(CPUArchState *env, target_ulong addr, DATA_TYPE val,
TCGMemOpIdx oi, uintptr_t retaddr)
{
uintptr_t mmu_idx = get_mmuidx(oi);
uintptr_t index = tlb_index(env, mmu_idx, addr);
CPUTLBEntry *entry = tlb_entry(env, mmu_idx, addr);
target_ulong tlb_addr = tlb_addr_write(entry);
unsigned a_bits = get_alignment_bits(get_memop(oi));
uintptr_t haddr;
if (addr & ((1 << a_bits) - 1)) {
cpu_unaligned_access(ENV_GET_CPU(env), addr, MMU_DATA_STORE,
mmu_idx, retaddr);
}
/* If the TLB entry is for a different page, reload and try again. */
if (!tlb_hit(tlb_addr, addr)) {
if (!VICTIM_TLB_HIT(addr_write, addr)) {
tlb_fill(ENV_GET_CPU(env), addr, DATA_SIZE, MMU_DATA_STORE,
mmu_idx, retaddr);
}
tlb_addr = tlb_addr_write(entry) & ~TLB_INVALID_MASK;
}
/* Handle an IO access. */
if (unlikely(tlb_addr & ~TARGET_PAGE_MASK)) {
if ((addr & (DATA_SIZE - 1)) != 0) {
goto do_unaligned_access;
}
/* ??? Note that the io helpers always read data in the target
byte ordering. We should push the LE/BE request down into io. */
val = TGT_LE(val);
glue(io_write, SUFFIX)(env, mmu_idx, index, val, addr,
retaddr, tlb_addr & TLB_RECHECK);
return;
}
/* Handle slow unaligned access (it spans two pages or IO). */
if (DATA_SIZE > 1
&& unlikely((addr & ~TARGET_PAGE_MASK) + DATA_SIZE - 1
>= TARGET_PAGE_SIZE)) {
int i;
target_ulong page2;
CPUTLBEntry *entry2;
do_unaligned_access:
/* Ensure the second page is in the TLB. Note that the first page
is already guaranteed to be filled, and that the second page
cannot evict the first. */
page2 = (addr + DATA_SIZE) & TARGET_PAGE_MASK;
entry2 = tlb_entry(env, mmu_idx, page2);
if (!tlb_hit_page(tlb_addr_write(entry2), page2)
&& !VICTIM_TLB_HIT(addr_write, page2)) {
tlb_fill(ENV_GET_CPU(env), page2, DATA_SIZE, MMU_DATA_STORE,
mmu_idx, retaddr);
}
/* XXX: not efficient, but simple. */
/* This loop must go in the forward direction to avoid issues
with self-modifying code in Windows 64-bit. */
for (i = 0; i < DATA_SIZE; ++i) {
/* Little-endian extract. */
uint8_t val8 = val >> (i * 8);
glue(helper_ret_stb, MMUSUFFIX)(env, addr + i, val8,
oi, retaddr);
}
return;
}
haddr = addr + entry->addend;
#if DATA_SIZE == 1
glue(glue(st, SUFFIX), _p)((uint8_t *)haddr, val);
#else
glue(glue(st, SUFFIX), _le_p)((uint8_t *)haddr, val);
#endif
}
#if DATA_SIZE > 1
void helper_be_st_name(CPUArchState *env, target_ulong addr, DATA_TYPE val,
TCGMemOpIdx oi, uintptr_t retaddr)
{
uintptr_t mmu_idx = get_mmuidx(oi);
uintptr_t index = tlb_index(env, mmu_idx, addr);
CPUTLBEntry *entry = tlb_entry(env, mmu_idx, addr);
target_ulong tlb_addr = tlb_addr_write(entry);
unsigned a_bits = get_alignment_bits(get_memop(oi));
uintptr_t haddr;
if (addr & ((1 << a_bits) - 1)) {
cpu_unaligned_access(ENV_GET_CPU(env), addr, MMU_DATA_STORE,
mmu_idx, retaddr);
}
/* If the TLB entry is for a different page, reload and try again. */
if (!tlb_hit(tlb_addr, addr)) {
if (!VICTIM_TLB_HIT(addr_write, addr)) {
tlb_fill(ENV_GET_CPU(env), addr, DATA_SIZE, MMU_DATA_STORE,
mmu_idx, retaddr);
}
tlb_addr = tlb_addr_write(entry) & ~TLB_INVALID_MASK;
}
/* Handle an IO access. */
if (unlikely(tlb_addr & ~TARGET_PAGE_MASK)) {
if ((addr & (DATA_SIZE - 1)) != 0) {
goto do_unaligned_access;
}
/* ??? Note that the io helpers always read data in the target
byte ordering. We should push the LE/BE request down into io. */
val = TGT_BE(val);
glue(io_write, SUFFIX)(env, mmu_idx, index, val, addr, retaddr,
tlb_addr & TLB_RECHECK);
return;
}
/* Handle slow unaligned access (it spans two pages or IO). */
if (DATA_SIZE > 1
&& unlikely((addr & ~TARGET_PAGE_MASK) + DATA_SIZE - 1
>= TARGET_PAGE_SIZE)) {
int i;
target_ulong page2;
CPUTLBEntry *entry2;
do_unaligned_access:
/* Ensure the second page is in the TLB. Note that the first page
is already guaranteed to be filled, and that the second page
cannot evict the first. */
page2 = (addr + DATA_SIZE) & TARGET_PAGE_MASK;
entry2 = tlb_entry(env, mmu_idx, page2);
if (!tlb_hit_page(tlb_addr_write(entry2), page2)
&& !VICTIM_TLB_HIT(addr_write, page2)) {
tlb_fill(ENV_GET_CPU(env), page2, DATA_SIZE, MMU_DATA_STORE,
mmu_idx, retaddr);
}
/* XXX: not efficient, but simple */
/* This loop must go in the forward direction to avoid issues
with self-modifying code. */
for (i = 0; i < DATA_SIZE; ++i) {
/* Big-endian extract. */
uint8_t val8 = val >> (((DATA_SIZE - 1) * 8) - (i * 8));
glue(helper_ret_stb, MMUSUFFIX)(env, addr + i, val8,
oi, retaddr);
}
return;
}
haddr = addr + entry->addend;
glue(glue(st, SUFFIX), _be_p)((uint8_t *)haddr, val);
}
#endif /* DATA_SIZE > 1 */
#endif /* !defined(SOFTMMU_CODE_ACCESS) */
#undef READ_ACCESS_TYPE
#undef DATA_TYPE
#undef SUFFIX
#undef LSUFFIX
#undef DATA_SIZE
#undef ADDR_READ
#undef WORD_TYPE
#undef SDATA_TYPE
#undef USUFFIX
#undef SSUFFIX
#undef BSWAP
#undef helper_le_ld_name
#undef helper_be_ld_name
#undef helper_le_lds_name
#undef helper_be_lds_name
#undef helper_le_st_name
#undef helper_be_st_name