diff options
Diffstat (limited to 'accel')
-rw-r--r-- | accel/tcg/atomic_template.h | 112 | ||||
-rw-r--r-- | accel/tcg/tcg-runtime.h | 8 | ||||
-rw-r--r-- | accel/tcg/translator.c | 21 |
3 files changed, 92 insertions, 49 deletions
diff --git a/accel/tcg/atomic_template.h b/accel/tcg/atomic_template.h index e022df4571..3f41ef2782 100644 --- a/accel/tcg/atomic_template.h +++ b/accel/tcg/atomic_template.h @@ -25,18 +25,22 @@ #elif DATA_SIZE == 8 # define SUFFIX q # define DATA_TYPE uint64_t +# define SDATA_TYPE int64_t # define BSWAP bswap64 #elif DATA_SIZE == 4 # define SUFFIX l # define DATA_TYPE uint32_t +# define SDATA_TYPE int32_t # define BSWAP bswap32 #elif DATA_SIZE == 2 # define SUFFIX w # define DATA_TYPE uint16_t +# define SDATA_TYPE int16_t # define BSWAP bswap16 #elif DATA_SIZE == 1 # define SUFFIX b # define DATA_TYPE uint8_t +# define SDATA_TYPE int8_t # define BSWAP #else # error unsupported data size @@ -118,6 +122,39 @@ GEN_ATOMIC_HELPER(or_fetch) GEN_ATOMIC_HELPER(xor_fetch) #undef GEN_ATOMIC_HELPER + +/* These helpers are, as a whole, full barriers. Within the helper, + * the leading barrier is explicit and the trailing barrier is within + * cmpxchg primitive. + */ +#define GEN_ATOMIC_HELPER_FN(X, FN, XDATA_TYPE, RET) \ +ABI_TYPE ATOMIC_NAME(X)(CPUArchState *env, target_ulong addr, \ + ABI_TYPE xval EXTRA_ARGS) \ +{ \ + ATOMIC_MMU_DECLS; \ + XDATA_TYPE *haddr = ATOMIC_MMU_LOOKUP; \ + XDATA_TYPE cmp, old, new, val = xval; \ + smp_mb(); \ + cmp = atomic_read__nocheck(haddr); \ + do { \ + old = cmp; new = FN(old, val); \ + cmp = atomic_cmpxchg__nocheck(haddr, old, new); \ + } while (cmp != old); \ + ATOMIC_MMU_CLEANUP; \ + return RET; \ +} + +GEN_ATOMIC_HELPER_FN(fetch_smin, MIN, SDATA_TYPE, old) +GEN_ATOMIC_HELPER_FN(fetch_umin, MIN, DATA_TYPE, old) +GEN_ATOMIC_HELPER_FN(fetch_smax, MAX, SDATA_TYPE, old) +GEN_ATOMIC_HELPER_FN(fetch_umax, MAX, DATA_TYPE, old) + +GEN_ATOMIC_HELPER_FN(smin_fetch, MIN, SDATA_TYPE, new) +GEN_ATOMIC_HELPER_FN(umin_fetch, MIN, DATA_TYPE, new) +GEN_ATOMIC_HELPER_FN(smax_fetch, MAX, SDATA_TYPE, new) +GEN_ATOMIC_HELPER_FN(umax_fetch, MAX, DATA_TYPE, new) + +#undef GEN_ATOMIC_HELPER_FN #endif /* DATA SIZE >= 16 */ #undef END @@ -192,47 +229,45 @@ GEN_ATOMIC_HELPER(xor_fetch) #undef GEN_ATOMIC_HELPER +/* These helpers are, as a whole, full barriers. Within the helper, + * the leading barrier is explicit and the trailing barrier is within + * cmpxchg primitive. + */ +#define GEN_ATOMIC_HELPER_FN(X, FN, XDATA_TYPE, RET) \ +ABI_TYPE ATOMIC_NAME(X)(CPUArchState *env, target_ulong addr, \ + ABI_TYPE xval EXTRA_ARGS) \ +{ \ + ATOMIC_MMU_DECLS; \ + XDATA_TYPE *haddr = ATOMIC_MMU_LOOKUP; \ + XDATA_TYPE ldo, ldn, old, new, val = xval; \ + smp_mb(); \ + ldn = atomic_read__nocheck(haddr); \ + do { \ + ldo = ldn; old = BSWAP(ldo); new = FN(old, val); \ + ldn = atomic_cmpxchg__nocheck(haddr, ldo, BSWAP(new)); \ + } while (ldo != ldn); \ + ATOMIC_MMU_CLEANUP; \ + return RET; \ +} + +GEN_ATOMIC_HELPER_FN(fetch_smin, MIN, SDATA_TYPE, old) +GEN_ATOMIC_HELPER_FN(fetch_umin, MIN, DATA_TYPE, old) +GEN_ATOMIC_HELPER_FN(fetch_smax, MAX, SDATA_TYPE, old) +GEN_ATOMIC_HELPER_FN(fetch_umax, MAX, DATA_TYPE, old) + +GEN_ATOMIC_HELPER_FN(smin_fetch, MIN, SDATA_TYPE, new) +GEN_ATOMIC_HELPER_FN(umin_fetch, MIN, DATA_TYPE, new) +GEN_ATOMIC_HELPER_FN(smax_fetch, MAX, SDATA_TYPE, new) +GEN_ATOMIC_HELPER_FN(umax_fetch, MAX, DATA_TYPE, new) + /* Note that for addition, we need to use a separate cmpxchg loop instead of bswaps for the reverse-host-endian helpers. */ -ABI_TYPE ATOMIC_NAME(fetch_add)(CPUArchState *env, target_ulong addr, - ABI_TYPE val EXTRA_ARGS) -{ - ATOMIC_MMU_DECLS; - DATA_TYPE *haddr = ATOMIC_MMU_LOOKUP; - DATA_TYPE ldo, ldn, ret, sto; - - ldo = atomic_read__nocheck(haddr); - while (1) { - ret = BSWAP(ldo); - sto = BSWAP(ret + val); - ldn = atomic_cmpxchg__nocheck(haddr, ldo, sto); - if (ldn == ldo) { - ATOMIC_MMU_CLEANUP; - return ret; - } - ldo = ldn; - } -} +#define ADD(X, Y) (X + Y) +GEN_ATOMIC_HELPER_FN(fetch_add, ADD, DATA_TYPE, old) +GEN_ATOMIC_HELPER_FN(add_fetch, ADD, DATA_TYPE, new) +#undef ADD -ABI_TYPE ATOMIC_NAME(add_fetch)(CPUArchState *env, target_ulong addr, - ABI_TYPE val EXTRA_ARGS) -{ - ATOMIC_MMU_DECLS; - DATA_TYPE *haddr = ATOMIC_MMU_LOOKUP; - DATA_TYPE ldo, ldn, ret, sto; - - ldo = atomic_read__nocheck(haddr); - while (1) { - ret = BSWAP(ldo) + val; - sto = BSWAP(ret); - ldn = atomic_cmpxchg__nocheck(haddr, ldo, sto); - if (ldn == ldo) { - ATOMIC_MMU_CLEANUP; - return ret; - } - ldo = ldn; - } -} +#undef GEN_ATOMIC_HELPER_FN #endif /* DATA_SIZE >= 16 */ #undef END @@ -241,5 +276,6 @@ ABI_TYPE ATOMIC_NAME(add_fetch)(CPUArchState *env, target_ulong addr, #undef BSWAP #undef ABI_TYPE #undef DATA_TYPE +#undef SDATA_TYPE #undef SUFFIX #undef DATA_SIZE diff --git a/accel/tcg/tcg-runtime.h b/accel/tcg/tcg-runtime.h index 2536959a18..1bd39d136d 100644 --- a/accel/tcg/tcg-runtime.h +++ b/accel/tcg/tcg-runtime.h @@ -125,11 +125,19 @@ GEN_ATOMIC_HELPERS(fetch_add) GEN_ATOMIC_HELPERS(fetch_and) GEN_ATOMIC_HELPERS(fetch_or) GEN_ATOMIC_HELPERS(fetch_xor) +GEN_ATOMIC_HELPERS(fetch_smin) +GEN_ATOMIC_HELPERS(fetch_umin) +GEN_ATOMIC_HELPERS(fetch_smax) +GEN_ATOMIC_HELPERS(fetch_umax) GEN_ATOMIC_HELPERS(add_fetch) GEN_ATOMIC_HELPERS(and_fetch) GEN_ATOMIC_HELPERS(or_fetch) GEN_ATOMIC_HELPERS(xor_fetch) +GEN_ATOMIC_HELPERS(smin_fetch) +GEN_ATOMIC_HELPERS(umin_fetch) +GEN_ATOMIC_HELPERS(smax_fetch) +GEN_ATOMIC_HELPERS(umax_fetch) GEN_ATOMIC_HELPERS(xchg) diff --git a/accel/tcg/translator.c b/accel/tcg/translator.c index 23c6602cd9..0f9dca9113 100644 --- a/accel/tcg/translator.c +++ b/accel/tcg/translator.c @@ -34,8 +34,6 @@ void translator_loop_temp_check(DisasContextBase *db) void translator_loop(const TranslatorOps *ops, DisasContextBase *db, CPUState *cpu, TranslationBlock *tb) { - int max_insns; - /* Initialize DisasContext */ db->tb = tb; db->pc_first = tb->pc; @@ -45,18 +43,18 @@ void translator_loop(const TranslatorOps *ops, DisasContextBase *db, db->singlestep_enabled = cpu->singlestep_enabled; /* Instruction counting */ - max_insns = tb_cflags(db->tb) & CF_COUNT_MASK; - if (max_insns == 0) { - max_insns = CF_COUNT_MASK; + db->max_insns = tb_cflags(db->tb) & CF_COUNT_MASK; + if (db->max_insns == 0) { + db->max_insns = CF_COUNT_MASK; } - if (max_insns > TCG_MAX_INSNS) { - max_insns = TCG_MAX_INSNS; + if (db->max_insns > TCG_MAX_INSNS) { + db->max_insns = TCG_MAX_INSNS; } if (db->singlestep_enabled || singlestep) { - max_insns = 1; + db->max_insns = 1; } - max_insns = ops->init_disas_context(db, cpu, max_insns); + ops->init_disas_context(db, cpu); tcg_debug_assert(db->is_jmp == DISAS_NEXT); /* no early exit */ /* Reset the temp count so that we can identify leaks */ @@ -95,7 +93,8 @@ void translator_loop(const TranslatorOps *ops, DisasContextBase *db, update db->pc_next and db->is_jmp to indicate what should be done next -- either exiting this loop or locate the start of the next instruction. */ - if (db->num_insns == max_insns && (tb_cflags(db->tb) & CF_LAST_IO)) { + if (db->num_insns == db->max_insns + && (tb_cflags(db->tb) & CF_LAST_IO)) { /* Accept I/O on the last instruction. */ gen_io_start(); ops->translate_insn(db, cpu); @@ -111,7 +110,7 @@ void translator_loop(const TranslatorOps *ops, DisasContextBase *db, /* Stop translation if the output buffer is full, or we have executed all of the allowed instructions. */ - if (tcg_op_buf_full() || db->num_insns >= max_insns) { + if (tcg_op_buf_full() || db->num_insns >= db->max_insns) { db->is_jmp = DISAS_TOO_MANY; break; } |