1 | The following changes since commit 1316b1ddc8a05e418c8134243f8bff8cccbbccb1: | 1 | The following changes since commit 6eeea6725a70e6fcb5abba0764496bdab07ddfb3: |
---|---|---|---|
2 | 2 | ||
3 | Merge remote-tracking branch 'remotes/kevin/tags/for-upstream' into staging (2019-07-12 15:38:22 +0100) | 3 | Merge remote-tracking branch 'remotes/huth-gitlab/tags/pull-request-2020-10-06' into staging (2020-10-06 21:13:34 +0100) |
4 | 4 | ||
5 | are available in the Git repository at: | 5 | are available in the Git repository at: |
6 | 6 | ||
7 | https://github.com/rth7680/qemu.git tags/pull-tcg-20190714 | 7 | https://github.com/rth7680/qemu.git tags/pull-tcg-20201008 |
8 | 8 | ||
9 | for you to fetch changes up to 52ba13f042714c4086416973fb88e2465e0888a1: | 9 | for you to fetch changes up to 62475e9d007d83db4d0a6ccebcda8914f392e9c9: |
10 | 10 | ||
11 | tcg: Release mmap_lock on translation fault (2019-07-14 12:19:01 +0200) | 11 | accel/tcg: Fix computing of is_write for MIPS (2020-10-08 05:57:32 -0500) |
12 | 12 | ||
13 | ---------------------------------------------------------------- | 13 | ---------------------------------------------------------------- |
14 | Fixes for 3 tcg bugs | 14 | Extend maximum gvec vector size |
15 | Fix i386 avx2 dupi | ||
16 | Fix mips host user-only write detection | ||
17 | Misc cleanups. | ||
15 | 18 | ||
16 | ---------------------------------------------------------------- | 19 | ---------------------------------------------------------------- |
17 | Richard Henderson (7): | 20 | Kele Huang (1): |
18 | tcg: Fix constant folding of INDEX_op_extract2_i32 | 21 | accel/tcg: Fix computing of is_write for MIPS |
19 | tcg/aarch64: Fix output of extract2 opcodes | ||
20 | include/qemu/atomic.h: Add signal_barrier | ||
21 | tcg: Introduce set/clear_helper_retaddr | ||
22 | tcg: Remove cpu_ld*_code_ra | ||
23 | tcg: Remove duplicate #if !defined(CODE_ACCESS) | ||
24 | tcg: Release mmap_lock on translation fault | ||
25 | 22 | ||
26 | include/exec/cpu_ldst.h | 20 ++++++++ | 23 | Richard Henderson (10): |
27 | include/exec/cpu_ldst_useronly_template.h | 40 ++++++++++------ | 24 | tcg: Adjust simd_desc size encoding |
28 | include/qemu/atomic.h | 11 +++++ | 25 | tcg: Drop union from TCGArgConstraint |
29 | accel/tcg/user-exec.c | 77 +++++++++++++++++++++---------- | 26 | tcg: Move sorted_args into TCGArgConstraint.sort_index |
30 | target/arm/helper-a64.c | 8 ++-- | 27 | tcg: Remove TCG_CT_REG |
31 | target/arm/sve_helper.c | 43 +++++++++-------- | 28 | tcg: Move some TCG_CT_* bits to TCGArgConstraint bitfields |
32 | tcg/aarch64/tcg-target.inc.c | 2 +- | 29 | tcg: Remove TCGOpDef.used |
33 | tcg/optimize.c | 4 +- | 30 | tcg/i386: Fix dupi for avx2 32-bit hosts |
34 | 8 files changed, 139 insertions(+), 66 deletions(-) | 31 | tcg: Fix generation of dupi_vec for 32-bit host |
32 | tcg/optimize: Fold dup2_vec | ||
33 | tcg: Remove TCG_TARGET_HAS_cmp_vec | ||
35 | 34 | ||
35 | include/tcg/tcg-gvec-desc.h | 38 ++++++++++++------ | ||
36 | include/tcg/tcg.h | 22 ++++------ | ||
37 | tcg/aarch64/tcg-target.h | 1 - | ||
38 | tcg/i386/tcg-target.h | 1 - | ||
39 | tcg/ppc/tcg-target.h | 1 - | ||
40 | accel/tcg/user-exec.c | 43 ++++++++++++++++++-- | ||
41 | tcg/optimize.c | 15 +++++++ | ||
42 | tcg/tcg-op-gvec.c | 35 ++++++++++++---- | ||
43 | tcg/tcg-op-vec.c | 12 ++++-- | ||
44 | tcg/tcg.c | 96 +++++++++++++++++++------------------------- | ||
45 | tcg/aarch64/tcg-target.c.inc | 17 ++++---- | ||
46 | tcg/arm/tcg-target.c.inc | 29 ++++++------- | ||
47 | tcg/i386/tcg-target.c.inc | 39 +++++++----------- | ||
48 | tcg/mips/tcg-target.c.inc | 21 +++++----- | ||
49 | tcg/ppc/tcg-target.c.inc | 29 ++++++------- | ||
50 | tcg/riscv/tcg-target.c.inc | 16 ++++---- | ||
51 | tcg/s390/tcg-target.c.inc | 22 +++++----- | ||
52 | tcg/sparc/tcg-target.c.inc | 21 ++++------ | ||
53 | tcg/tci/tcg-target.c.inc | 3 +- | ||
54 | 19 files changed, 244 insertions(+), 217 deletions(-) | ||
55 | diff view generated by jsdifflib |
1 | This code block is already surrounded by #ifndef CODE_ACCESS. | 1 | With larger vector sizes, it turns out oprsz == maxsz, and we only |
---|---|---|---|
2 | need to represent mismatch for oprsz <= 32. We do, however, need | ||
3 | to represent larger oprsz and do so without reducing SIMD_DATA_BITS. | ||
2 | 4 | ||
3 | Reviewed-by: Alex Bennée <alex.bennee@linaro.org> | 5 | Reduce the size of the oprsz field and increase the maxsz field. |
6 | Steal the oprsz value of 24 to indicate equality with maxsz. | ||
7 | |||
8 | Tested-by: Frank Chang <frank.chang@sifive.com> | ||
9 | Reviewed-by: Frank Chang <frank.chang@sifive.com> | ||
10 | Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org> | ||
4 | Signed-off-by: Richard Henderson <richard.henderson@linaro.org> | 11 | Signed-off-by: Richard Henderson <richard.henderson@linaro.org> |
5 | --- | 12 | --- |
6 | include/exec/cpu_ldst_useronly_template.h | 2 -- | 13 | include/tcg/tcg-gvec-desc.h | 38 ++++++++++++++++++++++++------------- |
7 | 1 file changed, 2 deletions(-) | 14 | tcg/tcg-op-gvec.c | 35 ++++++++++++++++++++++++++-------- |
15 | 2 files changed, 52 insertions(+), 21 deletions(-) | ||
8 | 16 | ||
9 | diff --git a/include/exec/cpu_ldst_useronly_template.h b/include/exec/cpu_ldst_useronly_template.h | 17 | diff --git a/include/tcg/tcg-gvec-desc.h b/include/tcg/tcg-gvec-desc.h |
10 | index XXXXXXX..XXXXXXX 100644 | 18 | index XXXXXXX..XXXXXXX 100644 |
11 | --- a/include/exec/cpu_ldst_useronly_template.h | 19 | --- a/include/tcg/tcg-gvec-desc.h |
12 | +++ b/include/exec/cpu_ldst_useronly_template.h | 20 | +++ b/include/tcg/tcg-gvec-desc.h |
13 | @@ -XXX,XX +XXX,XX @@ static inline void | 21 | @@ -XXX,XX +XXX,XX @@ |
14 | glue(glue(cpu_st, SUFFIX), MEMSUFFIX)(CPUArchState *env, abi_ptr ptr, | 22 | #ifndef TCG_TCG_GVEC_DESC_H |
15 | RES_TYPE v) | 23 | #define TCG_TCG_GVEC_DESC_H |
24 | |||
25 | -/* ??? These bit widths are set for ARM SVE, maxing out at 256 byte vectors. */ | ||
26 | -#define SIMD_OPRSZ_SHIFT 0 | ||
27 | -#define SIMD_OPRSZ_BITS 5 | ||
28 | +/* | ||
29 | + * This configuration allows MAXSZ to represent 2048 bytes, and | ||
30 | + * OPRSZ to match MAXSZ, or represent the smaller values 8, 16, or 32. | ||
31 | + * | ||
32 | + * Encode this with: | ||
33 | + * 0, 1, 3 -> 8, 16, 32 | ||
34 | + * 2 -> maxsz | ||
35 | + * | ||
36 | + * This steals the input that would otherwise map to 24 to match maxsz. | ||
37 | + */ | ||
38 | +#define SIMD_MAXSZ_SHIFT 0 | ||
39 | +#define SIMD_MAXSZ_BITS 8 | ||
40 | |||
41 | -#define SIMD_MAXSZ_SHIFT (SIMD_OPRSZ_SHIFT + SIMD_OPRSZ_BITS) | ||
42 | -#define SIMD_MAXSZ_BITS 5 | ||
43 | +#define SIMD_OPRSZ_SHIFT (SIMD_MAXSZ_SHIFT + SIMD_MAXSZ_BITS) | ||
44 | +#define SIMD_OPRSZ_BITS 2 | ||
45 | |||
46 | -#define SIMD_DATA_SHIFT (SIMD_MAXSZ_SHIFT + SIMD_MAXSZ_BITS) | ||
47 | +#define SIMD_DATA_SHIFT (SIMD_OPRSZ_SHIFT + SIMD_OPRSZ_BITS) | ||
48 | #define SIMD_DATA_BITS (32 - SIMD_DATA_SHIFT) | ||
49 | |||
50 | /* Create a descriptor from components. */ | ||
51 | uint32_t simd_desc(uint32_t oprsz, uint32_t maxsz, int32_t data); | ||
52 | |||
53 | -/* Extract the operation size from a descriptor. */ | ||
54 | -static inline intptr_t simd_oprsz(uint32_t desc) | ||
55 | -{ | ||
56 | - return (extract32(desc, SIMD_OPRSZ_SHIFT, SIMD_OPRSZ_BITS) + 1) * 8; | ||
57 | -} | ||
58 | - | ||
59 | /* Extract the max vector size from a descriptor. */ | ||
60 | static inline intptr_t simd_maxsz(uint32_t desc) | ||
16 | { | 61 | { |
17 | -#if !defined(CODE_ACCESS) | 62 | - return (extract32(desc, SIMD_MAXSZ_SHIFT, SIMD_MAXSZ_BITS) + 1) * 8; |
18 | trace_guest_mem_before_exec( | 63 | + return extract32(desc, SIMD_MAXSZ_SHIFT, SIMD_MAXSZ_BITS) * 8 + 8; |
19 | env_cpu(env), ptr, | 64 | +} |
20 | trace_mem_build_info(SHIFT, false, MO_TE, true)); | 65 | + |
21 | -#endif | 66 | +/* Extract the operation size from a descriptor. */ |
22 | glue(glue(st, SUFFIX), _p)(g2h(ptr), v); | 67 | +static inline intptr_t simd_oprsz(uint32_t desc) |
68 | +{ | ||
69 | + uint32_t f = extract32(desc, SIMD_OPRSZ_SHIFT, SIMD_OPRSZ_BITS); | ||
70 | + intptr_t o = f * 8 + 8; | ||
71 | + intptr_t m = simd_maxsz(desc); | ||
72 | + return f == 2 ? m : o; | ||
23 | } | 73 | } |
24 | 74 | ||
75 | /* Extract the operation-specific data from a descriptor. */ | ||
76 | diff --git a/tcg/tcg-op-gvec.c b/tcg/tcg-op-gvec.c | ||
77 | index XXXXXXX..XXXXXXX 100644 | ||
78 | --- a/tcg/tcg-op-gvec.c | ||
79 | +++ b/tcg/tcg-op-gvec.c | ||
80 | @@ -XXX,XX +XXX,XX @@ static const TCGOpcode vecop_list_empty[1] = { 0 }; | ||
81 | of the operand offsets so that we can check them all at once. */ | ||
82 | static void check_size_align(uint32_t oprsz, uint32_t maxsz, uint32_t ofs) | ||
83 | { | ||
84 | - uint32_t opr_align = oprsz >= 16 ? 15 : 7; | ||
85 | - uint32_t max_align = maxsz >= 16 || oprsz >= 16 ? 15 : 7; | ||
86 | - tcg_debug_assert(oprsz > 0); | ||
87 | - tcg_debug_assert(oprsz <= maxsz); | ||
88 | - tcg_debug_assert((oprsz & opr_align) == 0); | ||
89 | + uint32_t max_align; | ||
90 | + | ||
91 | + switch (oprsz) { | ||
92 | + case 8: | ||
93 | + case 16: | ||
94 | + case 32: | ||
95 | + tcg_debug_assert(oprsz <= maxsz); | ||
96 | + break; | ||
97 | + default: | ||
98 | + tcg_debug_assert(oprsz == maxsz); | ||
99 | + break; | ||
100 | + } | ||
101 | + tcg_debug_assert(maxsz <= (8 << SIMD_MAXSZ_BITS)); | ||
102 | + | ||
103 | + max_align = maxsz >= 16 ? 15 : 7; | ||
104 | tcg_debug_assert((maxsz & max_align) == 0); | ||
105 | tcg_debug_assert((ofs & max_align) == 0); | ||
106 | } | ||
107 | @@ -XXX,XX +XXX,XX @@ uint32_t simd_desc(uint32_t oprsz, uint32_t maxsz, int32_t data) | ||
108 | { | ||
109 | uint32_t desc = 0; | ||
110 | |||
111 | - assert(oprsz % 8 == 0 && oprsz <= (8 << SIMD_OPRSZ_BITS)); | ||
112 | - assert(maxsz % 8 == 0 && maxsz <= (8 << SIMD_MAXSZ_BITS)); | ||
113 | - assert(data == sextract32(data, 0, SIMD_DATA_BITS)); | ||
114 | + check_size_align(oprsz, maxsz, 0); | ||
115 | + tcg_debug_assert(data == sextract32(data, 0, SIMD_DATA_BITS)); | ||
116 | |||
117 | oprsz = (oprsz / 8) - 1; | ||
118 | maxsz = (maxsz / 8) - 1; | ||
119 | + | ||
120 | + /* | ||
121 | + * We have just asserted in check_size_align that either | ||
122 | + * oprsz is {8,16,32} or matches maxsz. Encode the final | ||
123 | + * case with '2', as that would otherwise map to 24. | ||
124 | + */ | ||
125 | + if (oprsz == maxsz) { | ||
126 | + oprsz = 2; | ||
127 | + } | ||
128 | + | ||
129 | desc = deposit32(desc, SIMD_OPRSZ_SHIFT, SIMD_OPRSZ_BITS, oprsz); | ||
130 | desc = deposit32(desc, SIMD_MAXSZ_SHIFT, SIMD_MAXSZ_BITS, maxsz); | ||
131 | desc = deposit32(desc, SIMD_DATA_SHIFT, SIMD_DATA_BITS, data); | ||
25 | -- | 132 | -- |
26 | 2.17.1 | 133 | 2.25.1 |
27 | 134 | ||
28 | 135 | diff view generated by jsdifflib |
New patch | |||
---|---|---|---|
1 | The union is unused; let "regs" appear in the main structure | ||
2 | without the "u.regs" wrapping. | ||
1 | 3 | ||
4 | Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org> | ||
5 | Signed-off-by: Richard Henderson <richard.henderson@linaro.org> | ||
6 | --- | ||
7 | include/tcg/tcg.h | 4 +--- | ||
8 | tcg/tcg.c | 22 +++++++++++----------- | ||
9 | tcg/aarch64/tcg-target.c.inc | 14 +++++++------- | ||
10 | tcg/arm/tcg-target.c.inc | 26 +++++++++++++------------- | ||
11 | tcg/i386/tcg-target.c.inc | 26 +++++++++++++------------- | ||
12 | tcg/mips/tcg-target.c.inc | 18 +++++++++--------- | ||
13 | tcg/ppc/tcg-target.c.inc | 24 ++++++++++++------------ | ||
14 | tcg/riscv/tcg-target.c.inc | 14 +++++++------- | ||
15 | tcg/s390/tcg-target.c.inc | 18 +++++++++--------- | ||
16 | tcg/sparc/tcg-target.c.inc | 16 ++++++++-------- | ||
17 | tcg/tci/tcg-target.c.inc | 2 +- | ||
18 | 11 files changed, 91 insertions(+), 93 deletions(-) | ||
19 | |||
20 | diff --git a/include/tcg/tcg.h b/include/tcg/tcg.h | ||
21 | index XXXXXXX..XXXXXXX 100644 | ||
22 | --- a/include/tcg/tcg.h | ||
23 | +++ b/include/tcg/tcg.h | ||
24 | @@ -XXX,XX +XXX,XX @@ void tcg_dump_op_count(void); | ||
25 | typedef struct TCGArgConstraint { | ||
26 | uint16_t ct; | ||
27 | uint8_t alias_index; | ||
28 | - union { | ||
29 | - TCGRegSet regs; | ||
30 | - } u; | ||
31 | + TCGRegSet regs; | ||
32 | } TCGArgConstraint; | ||
33 | |||
34 | #define TCG_MAX_OP_ARGS 16 | ||
35 | diff --git a/tcg/tcg.c b/tcg/tcg.c | ||
36 | index XXXXXXX..XXXXXXX 100644 | ||
37 | --- a/tcg/tcg.c | ||
38 | +++ b/tcg/tcg.c | ||
39 | @@ -XXX,XX +XXX,XX @@ static int get_constraint_priority(const TCGOpDef *def, int k) | ||
40 | return 0; | ||
41 | n = 0; | ||
42 | for(i = 0; i < TCG_TARGET_NB_REGS; i++) { | ||
43 | - if (tcg_regset_test_reg(arg_ct->u.regs, i)) | ||
44 | + if (tcg_regset_test_reg(arg_ct->regs, i)) | ||
45 | n++; | ||
46 | } | ||
47 | } | ||
48 | @@ -XXX,XX +XXX,XX @@ static void process_op_defs(TCGContext *s) | ||
49 | /* Incomplete TCGTargetOpDef entry. */ | ||
50 | tcg_debug_assert(ct_str != NULL); | ||
51 | |||
52 | - def->args_ct[i].u.regs = 0; | ||
53 | + def->args_ct[i].regs = 0; | ||
54 | def->args_ct[i].ct = 0; | ||
55 | while (*ct_str != '\0') { | ||
56 | switch(*ct_str) { | ||
57 | @@ -XXX,XX +XXX,XX @@ static void liveness_pass_1(TCGContext *s) | ||
58 | pset = la_temp_pref(ts); | ||
59 | set = *pset; | ||
60 | |||
61 | - set &= ct->u.regs; | ||
62 | + set &= ct->regs; | ||
63 | if (ct->ct & TCG_CT_IALIAS) { | ||
64 | set &= op->output_pref[ct->alias_index]; | ||
65 | } | ||
66 | /* If the combination is not possible, restart. */ | ||
67 | if (set == 0) { | ||
68 | - set = ct->u.regs; | ||
69 | + set = ct->regs; | ||
70 | } | ||
71 | *pset = set; | ||
72 | } | ||
73 | @@ -XXX,XX +XXX,XX @@ static void tcg_reg_alloc_dup(TCGContext *s, const TCGOp *op) | ||
74 | return; | ||
75 | } | ||
76 | |||
77 | - dup_out_regs = tcg_op_defs[INDEX_op_dup_vec].args_ct[0].u.regs; | ||
78 | - dup_in_regs = tcg_op_defs[INDEX_op_dup_vec].args_ct[1].u.regs; | ||
79 | + dup_out_regs = tcg_op_defs[INDEX_op_dup_vec].args_ct[0].regs; | ||
80 | + dup_in_regs = tcg_op_defs[INDEX_op_dup_vec].args_ct[1].regs; | ||
81 | |||
82 | /* Allocate the output register now. */ | ||
83 | if (ots->val_type != TEMP_VAL_REG) { | ||
84 | @@ -XXX,XX +XXX,XX @@ static void tcg_reg_alloc_op(TCGContext *s, const TCGOp *op) | ||
85 | } | ||
86 | } | ||
87 | |||
88 | - temp_load(s, ts, arg_ct->u.regs, i_allocated_regs, i_preferred_regs); | ||
89 | + temp_load(s, ts, arg_ct->regs, i_allocated_regs, i_preferred_regs); | ||
90 | reg = ts->reg; | ||
91 | |||
92 | - if (tcg_regset_test_reg(arg_ct->u.regs, reg)) { | ||
93 | + if (tcg_regset_test_reg(arg_ct->regs, reg)) { | ||
94 | /* nothing to do : the constraint is satisfied */ | ||
95 | } else { | ||
96 | allocate_in_reg: | ||
97 | @@ -XXX,XX +XXX,XX @@ static void tcg_reg_alloc_op(TCGContext *s, const TCGOp *op) | ||
98 | and move the temporary register into it */ | ||
99 | temp_load(s, ts, tcg_target_available_regs[ts->type], | ||
100 | i_allocated_regs, 0); | ||
101 | - reg = tcg_reg_alloc(s, arg_ct->u.regs, i_allocated_regs, | ||
102 | + reg = tcg_reg_alloc(s, arg_ct->regs, i_allocated_regs, | ||
103 | o_preferred_regs, ts->indirect_base); | ||
104 | if (!tcg_out_mov(s, ts->type, reg, ts->reg)) { | ||
105 | /* | ||
106 | @@ -XXX,XX +XXX,XX @@ static void tcg_reg_alloc_op(TCGContext *s, const TCGOp *op) | ||
107 | && !const_args[arg_ct->alias_index]) { | ||
108 | reg = new_args[arg_ct->alias_index]; | ||
109 | } else if (arg_ct->ct & TCG_CT_NEWREG) { | ||
110 | - reg = tcg_reg_alloc(s, arg_ct->u.regs, | ||
111 | + reg = tcg_reg_alloc(s, arg_ct->regs, | ||
112 | i_allocated_regs | o_allocated_regs, | ||
113 | op->output_pref[k], ts->indirect_base); | ||
114 | } else { | ||
115 | - reg = tcg_reg_alloc(s, arg_ct->u.regs, o_allocated_regs, | ||
116 | + reg = tcg_reg_alloc(s, arg_ct->regs, o_allocated_regs, | ||
117 | op->output_pref[k], ts->indirect_base); | ||
118 | } | ||
119 | tcg_regset_set_reg(o_allocated_regs, reg); | ||
120 | diff --git a/tcg/aarch64/tcg-target.c.inc b/tcg/aarch64/tcg-target.c.inc | ||
121 | index XXXXXXX..XXXXXXX 100644 | ||
122 | --- a/tcg/aarch64/tcg-target.c.inc | ||
123 | +++ b/tcg/aarch64/tcg-target.c.inc | ||
124 | @@ -XXX,XX +XXX,XX @@ static const char *target_parse_constraint(TCGArgConstraint *ct, | ||
125 | switch (*ct_str++) { | ||
126 | case 'r': /* general registers */ | ||
127 | ct->ct |= TCG_CT_REG; | ||
128 | - ct->u.regs |= 0xffffffffu; | ||
129 | + ct->regs |= 0xffffffffu; | ||
130 | break; | ||
131 | case 'w': /* advsimd registers */ | ||
132 | ct->ct |= TCG_CT_REG; | ||
133 | - ct->u.regs |= 0xffffffff00000000ull; | ||
134 | + ct->regs |= 0xffffffff00000000ull; | ||
135 | break; | ||
136 | case 'l': /* qemu_ld / qemu_st address, data_reg */ | ||
137 | ct->ct |= TCG_CT_REG; | ||
138 | - ct->u.regs = 0xffffffffu; | ||
139 | + ct->regs = 0xffffffffu; | ||
140 | #ifdef CONFIG_SOFTMMU | ||
141 | /* x0 and x1 will be overwritten when reading the tlb entry, | ||
142 | and x2, and x3 for helper args, better to avoid using them. */ | ||
143 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_X0); | ||
144 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_X1); | ||
145 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_X2); | ||
146 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_X3); | ||
147 | + tcg_regset_reset_reg(ct->regs, TCG_REG_X0); | ||
148 | + tcg_regset_reset_reg(ct->regs, TCG_REG_X1); | ||
149 | + tcg_regset_reset_reg(ct->regs, TCG_REG_X2); | ||
150 | + tcg_regset_reset_reg(ct->regs, TCG_REG_X3); | ||
151 | #endif | ||
152 | break; | ||
153 | case 'A': /* Valid for arithmetic immediate (positive or negative). */ | ||
154 | diff --git a/tcg/arm/tcg-target.c.inc b/tcg/arm/tcg-target.c.inc | ||
155 | index XXXXXXX..XXXXXXX 100644 | ||
156 | --- a/tcg/arm/tcg-target.c.inc | ||
157 | +++ b/tcg/arm/tcg-target.c.inc | ||
158 | @@ -XXX,XX +XXX,XX @@ static const char *target_parse_constraint(TCGArgConstraint *ct, | ||
159 | |||
160 | case 'r': | ||
161 | ct->ct |= TCG_CT_REG; | ||
162 | - ct->u.regs = 0xffff; | ||
163 | + ct->regs = 0xffff; | ||
164 | break; | ||
165 | |||
166 | /* qemu_ld address */ | ||
167 | case 'l': | ||
168 | ct->ct |= TCG_CT_REG; | ||
169 | - ct->u.regs = 0xffff; | ||
170 | + ct->regs = 0xffff; | ||
171 | #ifdef CONFIG_SOFTMMU | ||
172 | /* r0-r2,lr will be overwritten when reading the tlb entry, | ||
173 | so don't use these. */ | ||
174 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_R0); | ||
175 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_R1); | ||
176 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_R2); | ||
177 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_R3); | ||
178 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_R14); | ||
179 | + tcg_regset_reset_reg(ct->regs, TCG_REG_R0); | ||
180 | + tcg_regset_reset_reg(ct->regs, TCG_REG_R1); | ||
181 | + tcg_regset_reset_reg(ct->regs, TCG_REG_R2); | ||
182 | + tcg_regset_reset_reg(ct->regs, TCG_REG_R3); | ||
183 | + tcg_regset_reset_reg(ct->regs, TCG_REG_R14); | ||
184 | #endif | ||
185 | break; | ||
186 | |||
187 | /* qemu_st address & data */ | ||
188 | case 's': | ||
189 | ct->ct |= TCG_CT_REG; | ||
190 | - ct->u.regs = 0xffff; | ||
191 | + ct->regs = 0xffff; | ||
192 | /* r0-r2 will be overwritten when reading the tlb entry (softmmu only) | ||
193 | and r0-r1 doing the byte swapping, so don't use these. */ | ||
194 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_R0); | ||
195 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_R1); | ||
196 | + tcg_regset_reset_reg(ct->regs, TCG_REG_R0); | ||
197 | + tcg_regset_reset_reg(ct->regs, TCG_REG_R1); | ||
198 | #if defined(CONFIG_SOFTMMU) | ||
199 | /* Avoid clashes with registers being used for helper args */ | ||
200 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_R2); | ||
201 | + tcg_regset_reset_reg(ct->regs, TCG_REG_R2); | ||
202 | #if TARGET_LONG_BITS == 64 | ||
203 | /* Avoid clashes with registers being used for helper args */ | ||
204 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_R3); | ||
205 | + tcg_regset_reset_reg(ct->regs, TCG_REG_R3); | ||
206 | #endif | ||
207 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_R14); | ||
208 | + tcg_regset_reset_reg(ct->regs, TCG_REG_R14); | ||
209 | #endif | ||
210 | break; | ||
211 | |||
212 | diff --git a/tcg/i386/tcg-target.c.inc b/tcg/i386/tcg-target.c.inc | ||
213 | index XXXXXXX..XXXXXXX 100644 | ||
214 | --- a/tcg/i386/tcg-target.c.inc | ||
215 | +++ b/tcg/i386/tcg-target.c.inc | ||
216 | @@ -XXX,XX +XXX,XX @@ static const char *target_parse_constraint(TCGArgConstraint *ct, | ||
217 | switch(*ct_str++) { | ||
218 | case 'a': | ||
219 | ct->ct |= TCG_CT_REG; | ||
220 | - tcg_regset_set_reg(ct->u.regs, TCG_REG_EAX); | ||
221 | + tcg_regset_set_reg(ct->regs, TCG_REG_EAX); | ||
222 | break; | ||
223 | case 'b': | ||
224 | ct->ct |= TCG_CT_REG; | ||
225 | - tcg_regset_set_reg(ct->u.regs, TCG_REG_EBX); | ||
226 | + tcg_regset_set_reg(ct->regs, TCG_REG_EBX); | ||
227 | break; | ||
228 | case 'c': | ||
229 | ct->ct |= TCG_CT_REG; | ||
230 | - tcg_regset_set_reg(ct->u.regs, TCG_REG_ECX); | ||
231 | + tcg_regset_set_reg(ct->regs, TCG_REG_ECX); | ||
232 | break; | ||
233 | case 'd': | ||
234 | ct->ct |= TCG_CT_REG; | ||
235 | - tcg_regset_set_reg(ct->u.regs, TCG_REG_EDX); | ||
236 | + tcg_regset_set_reg(ct->regs, TCG_REG_EDX); | ||
237 | break; | ||
238 | case 'S': | ||
239 | ct->ct |= TCG_CT_REG; | ||
240 | - tcg_regset_set_reg(ct->u.regs, TCG_REG_ESI); | ||
241 | + tcg_regset_set_reg(ct->regs, TCG_REG_ESI); | ||
242 | break; | ||
243 | case 'D': | ||
244 | ct->ct |= TCG_CT_REG; | ||
245 | - tcg_regset_set_reg(ct->u.regs, TCG_REG_EDI); | ||
246 | + tcg_regset_set_reg(ct->regs, TCG_REG_EDI); | ||
247 | break; | ||
248 | case 'q': | ||
249 | /* A register that can be used as a byte operand. */ | ||
250 | ct->ct |= TCG_CT_REG; | ||
251 | - ct->u.regs = TCG_TARGET_REG_BITS == 64 ? 0xffff : 0xf; | ||
252 | + ct->regs = TCG_TARGET_REG_BITS == 64 ? 0xffff : 0xf; | ||
253 | break; | ||
254 | case 'Q': | ||
255 | /* A register with an addressable second byte (e.g. %ah). */ | ||
256 | ct->ct |= TCG_CT_REG; | ||
257 | - ct->u.regs = 0xf; | ||
258 | + ct->regs = 0xf; | ||
259 | break; | ||
260 | case 'r': | ||
261 | /* A general register. */ | ||
262 | ct->ct |= TCG_CT_REG; | ||
263 | - ct->u.regs |= ALL_GENERAL_REGS; | ||
264 | + ct->regs |= ALL_GENERAL_REGS; | ||
265 | break; | ||
266 | case 'W': | ||
267 | /* With TZCNT/LZCNT, we can have operand-size as an input. */ | ||
268 | @@ -XXX,XX +XXX,XX @@ static const char *target_parse_constraint(TCGArgConstraint *ct, | ||
269 | case 'x': | ||
270 | /* A vector register. */ | ||
271 | ct->ct |= TCG_CT_REG; | ||
272 | - ct->u.regs |= ALL_VECTOR_REGS; | ||
273 | + ct->regs |= ALL_VECTOR_REGS; | ||
274 | break; | ||
275 | |||
276 | /* qemu_ld/st address constraint */ | ||
277 | case 'L': | ||
278 | ct->ct |= TCG_CT_REG; | ||
279 | - ct->u.regs = TCG_TARGET_REG_BITS == 64 ? 0xffff : 0xff; | ||
280 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_L0); | ||
281 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_L1); | ||
282 | + ct->regs = TCG_TARGET_REG_BITS == 64 ? 0xffff : 0xff; | ||
283 | + tcg_regset_reset_reg(ct->regs, TCG_REG_L0); | ||
284 | + tcg_regset_reset_reg(ct->regs, TCG_REG_L1); | ||
285 | break; | ||
286 | |||
287 | case 'e': | ||
288 | diff --git a/tcg/mips/tcg-target.c.inc b/tcg/mips/tcg-target.c.inc | ||
289 | index XXXXXXX..XXXXXXX 100644 | ||
290 | --- a/tcg/mips/tcg-target.c.inc | ||
291 | +++ b/tcg/mips/tcg-target.c.inc | ||
292 | @@ -XXX,XX +XXX,XX @@ static const char *target_parse_constraint(TCGArgConstraint *ct, | ||
293 | switch(*ct_str++) { | ||
294 | case 'r': | ||
295 | ct->ct |= TCG_CT_REG; | ||
296 | - ct->u.regs = 0xffffffff; | ||
297 | + ct->regs = 0xffffffff; | ||
298 | break; | ||
299 | case 'L': /* qemu_ld input arg constraint */ | ||
300 | ct->ct |= TCG_CT_REG; | ||
301 | - ct->u.regs = 0xffffffff; | ||
302 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_A0); | ||
303 | + ct->regs = 0xffffffff; | ||
304 | + tcg_regset_reset_reg(ct->regs, TCG_REG_A0); | ||
305 | #if defined(CONFIG_SOFTMMU) | ||
306 | if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) { | ||
307 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_A2); | ||
308 | + tcg_regset_reset_reg(ct->regs, TCG_REG_A2); | ||
309 | } | ||
310 | #endif | ||
311 | break; | ||
312 | case 'S': /* qemu_st constraint */ | ||
313 | ct->ct |= TCG_CT_REG; | ||
314 | - ct->u.regs = 0xffffffff; | ||
315 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_A0); | ||
316 | + ct->regs = 0xffffffff; | ||
317 | + tcg_regset_reset_reg(ct->regs, TCG_REG_A0); | ||
318 | #if defined(CONFIG_SOFTMMU) | ||
319 | if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) { | ||
320 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_A2); | ||
321 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_A3); | ||
322 | + tcg_regset_reset_reg(ct->regs, TCG_REG_A2); | ||
323 | + tcg_regset_reset_reg(ct->regs, TCG_REG_A3); | ||
324 | } else { | ||
325 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_A1); | ||
326 | + tcg_regset_reset_reg(ct->regs, TCG_REG_A1); | ||
327 | } | ||
328 | #endif | ||
329 | break; | ||
330 | diff --git a/tcg/ppc/tcg-target.c.inc b/tcg/ppc/tcg-target.c.inc | ||
331 | index XXXXXXX..XXXXXXX 100644 | ||
332 | --- a/tcg/ppc/tcg-target.c.inc | ||
333 | +++ b/tcg/ppc/tcg-target.c.inc | ||
334 | @@ -XXX,XX +XXX,XX @@ static const char *target_parse_constraint(TCGArgConstraint *ct, | ||
335 | switch (*ct_str++) { | ||
336 | case 'A': case 'B': case 'C': case 'D': | ||
337 | ct->ct |= TCG_CT_REG; | ||
338 | - tcg_regset_set_reg(ct->u.regs, 3 + ct_str[0] - 'A'); | ||
339 | + tcg_regset_set_reg(ct->regs, 3 + ct_str[0] - 'A'); | ||
340 | break; | ||
341 | case 'r': | ||
342 | ct->ct |= TCG_CT_REG; | ||
343 | - ct->u.regs = 0xffffffff; | ||
344 | + ct->regs = 0xffffffff; | ||
345 | break; | ||
346 | case 'v': | ||
347 | ct->ct |= TCG_CT_REG; | ||
348 | - ct->u.regs = 0xffffffff00000000ull; | ||
349 | + ct->regs = 0xffffffff00000000ull; | ||
350 | break; | ||
351 | case 'L': /* qemu_ld constraint */ | ||
352 | ct->ct |= TCG_CT_REG; | ||
353 | - ct->u.regs = 0xffffffff; | ||
354 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_R3); | ||
355 | + ct->regs = 0xffffffff; | ||
356 | + tcg_regset_reset_reg(ct->regs, TCG_REG_R3); | ||
357 | #ifdef CONFIG_SOFTMMU | ||
358 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_R4); | ||
359 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_R5); | ||
360 | + tcg_regset_reset_reg(ct->regs, TCG_REG_R4); | ||
361 | + tcg_regset_reset_reg(ct->regs, TCG_REG_R5); | ||
362 | #endif | ||
363 | break; | ||
364 | case 'S': /* qemu_st constraint */ | ||
365 | ct->ct |= TCG_CT_REG; | ||
366 | - ct->u.regs = 0xffffffff; | ||
367 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_R3); | ||
368 | + ct->regs = 0xffffffff; | ||
369 | + tcg_regset_reset_reg(ct->regs, TCG_REG_R3); | ||
370 | #ifdef CONFIG_SOFTMMU | ||
371 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_R4); | ||
372 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_R5); | ||
373 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_R6); | ||
374 | + tcg_regset_reset_reg(ct->regs, TCG_REG_R4); | ||
375 | + tcg_regset_reset_reg(ct->regs, TCG_REG_R5); | ||
376 | + tcg_regset_reset_reg(ct->regs, TCG_REG_R6); | ||
377 | #endif | ||
378 | break; | ||
379 | case 'I': | ||
380 | diff --git a/tcg/riscv/tcg-target.c.inc b/tcg/riscv/tcg-target.c.inc | ||
381 | index XXXXXXX..XXXXXXX 100644 | ||
382 | --- a/tcg/riscv/tcg-target.c.inc | ||
383 | +++ b/tcg/riscv/tcg-target.c.inc | ||
384 | @@ -XXX,XX +XXX,XX @@ static const char *target_parse_constraint(TCGArgConstraint *ct, | ||
385 | switch (*ct_str++) { | ||
386 | case 'r': | ||
387 | ct->ct |= TCG_CT_REG; | ||
388 | - ct->u.regs = 0xffffffff; | ||
389 | + ct->regs = 0xffffffff; | ||
390 | break; | ||
391 | case 'L': | ||
392 | /* qemu_ld/qemu_st constraint */ | ||
393 | ct->ct |= TCG_CT_REG; | ||
394 | - ct->u.regs = 0xffffffff; | ||
395 | + ct->regs = 0xffffffff; | ||
396 | /* qemu_ld/qemu_st uses TCG_REG_TMP0 */ | ||
397 | #if defined(CONFIG_SOFTMMU) | ||
398 | - tcg_regset_reset_reg(ct->u.regs, tcg_target_call_iarg_regs[0]); | ||
399 | - tcg_regset_reset_reg(ct->u.regs, tcg_target_call_iarg_regs[1]); | ||
400 | - tcg_regset_reset_reg(ct->u.regs, tcg_target_call_iarg_regs[2]); | ||
401 | - tcg_regset_reset_reg(ct->u.regs, tcg_target_call_iarg_regs[3]); | ||
402 | - tcg_regset_reset_reg(ct->u.regs, tcg_target_call_iarg_regs[4]); | ||
403 | + tcg_regset_reset_reg(ct->regs, tcg_target_call_iarg_regs[0]); | ||
404 | + tcg_regset_reset_reg(ct->regs, tcg_target_call_iarg_regs[1]); | ||
405 | + tcg_regset_reset_reg(ct->regs, tcg_target_call_iarg_regs[2]); | ||
406 | + tcg_regset_reset_reg(ct->regs, tcg_target_call_iarg_regs[3]); | ||
407 | + tcg_regset_reset_reg(ct->regs, tcg_target_call_iarg_regs[4]); | ||
408 | #endif | ||
409 | break; | ||
410 | case 'I': | ||
411 | diff --git a/tcg/s390/tcg-target.c.inc b/tcg/s390/tcg-target.c.inc | ||
412 | index XXXXXXX..XXXXXXX 100644 | ||
413 | --- a/tcg/s390/tcg-target.c.inc | ||
414 | +++ b/tcg/s390/tcg-target.c.inc | ||
415 | @@ -XXX,XX +XXX,XX @@ static const char *target_parse_constraint(TCGArgConstraint *ct, | ||
416 | switch (*ct_str++) { | ||
417 | case 'r': /* all registers */ | ||
418 | ct->ct |= TCG_CT_REG; | ||
419 | - ct->u.regs = 0xffff; | ||
420 | + ct->regs = 0xffff; | ||
421 | break; | ||
422 | case 'L': /* qemu_ld/st constraint */ | ||
423 | ct->ct |= TCG_CT_REG; | ||
424 | - ct->u.regs = 0xffff; | ||
425 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_R2); | ||
426 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_R3); | ||
427 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_R4); | ||
428 | + ct->regs = 0xffff; | ||
429 | + tcg_regset_reset_reg(ct->regs, TCG_REG_R2); | ||
430 | + tcg_regset_reset_reg(ct->regs, TCG_REG_R3); | ||
431 | + tcg_regset_reset_reg(ct->regs, TCG_REG_R4); | ||
432 | break; | ||
433 | case 'a': /* force R2 for division */ | ||
434 | ct->ct |= TCG_CT_REG; | ||
435 | - ct->u.regs = 0; | ||
436 | - tcg_regset_set_reg(ct->u.regs, TCG_REG_R2); | ||
437 | + ct->regs = 0; | ||
438 | + tcg_regset_set_reg(ct->regs, TCG_REG_R2); | ||
439 | break; | ||
440 | case 'b': /* force R3 for division */ | ||
441 | ct->ct |= TCG_CT_REG; | ||
442 | - ct->u.regs = 0; | ||
443 | - tcg_regset_set_reg(ct->u.regs, TCG_REG_R3); | ||
444 | + ct->regs = 0; | ||
445 | + tcg_regset_set_reg(ct->regs, TCG_REG_R3); | ||
446 | break; | ||
447 | case 'A': | ||
448 | ct->ct |= TCG_CT_CONST_S33; | ||
449 | diff --git a/tcg/sparc/tcg-target.c.inc b/tcg/sparc/tcg-target.c.inc | ||
450 | index XXXXXXX..XXXXXXX 100644 | ||
451 | --- a/tcg/sparc/tcg-target.c.inc | ||
452 | +++ b/tcg/sparc/tcg-target.c.inc | ||
453 | @@ -XXX,XX +XXX,XX @@ static const char *target_parse_constraint(TCGArgConstraint *ct, | ||
454 | switch (*ct_str++) { | ||
455 | case 'r': | ||
456 | ct->ct |= TCG_CT_REG; | ||
457 | - ct->u.regs = 0xffffffff; | ||
458 | + ct->regs = 0xffffffff; | ||
459 | break; | ||
460 | case 'R': | ||
461 | ct->ct |= TCG_CT_REG; | ||
462 | - ct->u.regs = ALL_64; | ||
463 | + ct->regs = ALL_64; | ||
464 | break; | ||
465 | case 'A': /* qemu_ld/st address constraint */ | ||
466 | ct->ct |= TCG_CT_REG; | ||
467 | - ct->u.regs = TARGET_LONG_BITS == 64 ? ALL_64 : 0xffffffff; | ||
468 | + ct->regs = TARGET_LONG_BITS == 64 ? ALL_64 : 0xffffffff; | ||
469 | reserve_helpers: | ||
470 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_O0); | ||
471 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_O1); | ||
472 | - tcg_regset_reset_reg(ct->u.regs, TCG_REG_O2); | ||
473 | + tcg_regset_reset_reg(ct->regs, TCG_REG_O0); | ||
474 | + tcg_regset_reset_reg(ct->regs, TCG_REG_O1); | ||
475 | + tcg_regset_reset_reg(ct->regs, TCG_REG_O2); | ||
476 | break; | ||
477 | case 's': /* qemu_st data 32-bit constraint */ | ||
478 | ct->ct |= TCG_CT_REG; | ||
479 | - ct->u.regs = 0xffffffff; | ||
480 | + ct->regs = 0xffffffff; | ||
481 | goto reserve_helpers; | ||
482 | case 'S': /* qemu_st data 64-bit constraint */ | ||
483 | ct->ct |= TCG_CT_REG; | ||
484 | - ct->u.regs = ALL_64; | ||
485 | + ct->regs = ALL_64; | ||
486 | goto reserve_helpers; | ||
487 | case 'I': | ||
488 | ct->ct |= TCG_CT_CONST_S11; | ||
489 | diff --git a/tcg/tci/tcg-target.c.inc b/tcg/tci/tcg-target.c.inc | ||
490 | index XXXXXXX..XXXXXXX 100644 | ||
491 | --- a/tcg/tci/tcg-target.c.inc | ||
492 | +++ b/tcg/tci/tcg-target.c.inc | ||
493 | @@ -XXX,XX +XXX,XX @@ static const char *target_parse_constraint(TCGArgConstraint *ct, | ||
494 | case 'L': /* qemu_ld constraint */ | ||
495 | case 'S': /* qemu_st constraint */ | ||
496 | ct->ct |= TCG_CT_REG; | ||
497 | - ct->u.regs = BIT(TCG_TARGET_NB_REGS) - 1; | ||
498 | + ct->regs = BIT(TCG_TARGET_NB_REGS) - 1; | ||
499 | break; | ||
500 | default: | ||
501 | return NULL; | ||
502 | -- | ||
503 | 2.25.1 | ||
504 | |||
505 | diff view generated by jsdifflib |
New patch | |||
---|---|---|---|
1 | This uses an existing hole in the TCGArgConstraint structure | ||
2 | and will be convenient for keeping the data in one place. | ||
1 | 3 | ||
4 | Signed-off-by: Richard Henderson <richard.henderson@linaro.org> | ||
5 | --- | ||
6 | include/tcg/tcg.h | 2 +- | ||
7 | tcg/tcg.c | 35 +++++++++++++++++------------------ | ||
8 | 2 files changed, 18 insertions(+), 19 deletions(-) | ||
9 | |||
10 | diff --git a/include/tcg/tcg.h b/include/tcg/tcg.h | ||
11 | index XXXXXXX..XXXXXXX 100644 | ||
12 | --- a/include/tcg/tcg.h | ||
13 | +++ b/include/tcg/tcg.h | ||
14 | @@ -XXX,XX +XXX,XX @@ void tcg_dump_op_count(void); | ||
15 | typedef struct TCGArgConstraint { | ||
16 | uint16_t ct; | ||
17 | uint8_t alias_index; | ||
18 | + uint8_t sort_index; | ||
19 | TCGRegSet regs; | ||
20 | } TCGArgConstraint; | ||
21 | |||
22 | @@ -XXX,XX +XXX,XX @@ typedef struct TCGOpDef { | ||
23 | uint8_t nb_oargs, nb_iargs, nb_cargs, nb_args; | ||
24 | uint8_t flags; | ||
25 | TCGArgConstraint *args_ct; | ||
26 | - int *sorted_args; | ||
27 | #if defined(CONFIG_DEBUG_TCG) | ||
28 | int used; | ||
29 | #endif | ||
30 | diff --git a/tcg/tcg.c b/tcg/tcg.c | ||
31 | index XXXXXXX..XXXXXXX 100644 | ||
32 | --- a/tcg/tcg.c | ||
33 | +++ b/tcg/tcg.c | ||
34 | @@ -XXX,XX +XXX,XX @@ void tcg_context_init(TCGContext *s) | ||
35 | int op, total_args, n, i; | ||
36 | TCGOpDef *def; | ||
37 | TCGArgConstraint *args_ct; | ||
38 | - int *sorted_args; | ||
39 | TCGTemp *ts; | ||
40 | |||
41 | memset(s, 0, sizeof(*s)); | ||
42 | @@ -XXX,XX +XXX,XX @@ void tcg_context_init(TCGContext *s) | ||
43 | } | ||
44 | |||
45 | args_ct = g_malloc(sizeof(TCGArgConstraint) * total_args); | ||
46 | - sorted_args = g_malloc(sizeof(int) * total_args); | ||
47 | |||
48 | for(op = 0; op < NB_OPS; op++) { | ||
49 | def = &tcg_op_defs[op]; | ||
50 | def->args_ct = args_ct; | ||
51 | - def->sorted_args = sorted_args; | ||
52 | n = def->nb_iargs + def->nb_oargs; | ||
53 | - sorted_args += n; | ||
54 | args_ct += n; | ||
55 | } | ||
56 | |||
57 | @@ -XXX,XX +XXX,XX @@ static int get_constraint_priority(const TCGOpDef *def, int k) | ||
58 | /* sort from highest priority to lowest */ | ||
59 | static void sort_constraints(TCGOpDef *def, int start, int n) | ||
60 | { | ||
61 | - int i, j, p1, p2, tmp; | ||
62 | + int i, j; | ||
63 | + TCGArgConstraint *a = def->args_ct; | ||
64 | |||
65 | - for(i = 0; i < n; i++) | ||
66 | - def->sorted_args[start + i] = start + i; | ||
67 | - if (n <= 1) | ||
68 | + for (i = 0; i < n; i++) { | ||
69 | + a[start + i].sort_index = start + i; | ||
70 | + } | ||
71 | + if (n <= 1) { | ||
72 | return; | ||
73 | - for(i = 0; i < n - 1; i++) { | ||
74 | - for(j = i + 1; j < n; j++) { | ||
75 | - p1 = get_constraint_priority(def, def->sorted_args[start + i]); | ||
76 | - p2 = get_constraint_priority(def, def->sorted_args[start + j]); | ||
77 | + } | ||
78 | + for (i = 0; i < n - 1; i++) { | ||
79 | + for (j = i + 1; j < n; j++) { | ||
80 | + int p1 = get_constraint_priority(def, a[start + i].sort_index); | ||
81 | + int p2 = get_constraint_priority(def, a[start + j].sort_index); | ||
82 | if (p1 < p2) { | ||
83 | - tmp = def->sorted_args[start + i]; | ||
84 | - def->sorted_args[start + i] = def->sorted_args[start + j]; | ||
85 | - def->sorted_args[start + j] = tmp; | ||
86 | + int tmp = a[start + i].sort_index; | ||
87 | + a[start + i].sort_index = a[start + j].sort_index; | ||
88 | + a[start + j].sort_index = tmp; | ||
89 | } | ||
90 | } | ||
91 | } | ||
92 | @@ -XXX,XX +XXX,XX @@ static void tcg_reg_alloc_op(TCGContext *s, const TCGOp *op) | ||
93 | for (k = 0; k < nb_iargs; k++) { | ||
94 | TCGRegSet i_preferred_regs, o_preferred_regs; | ||
95 | |||
96 | - i = def->sorted_args[nb_oargs + k]; | ||
97 | + i = def->args_ct[nb_oargs + k].sort_index; | ||
98 | arg = op->args[i]; | ||
99 | arg_ct = &def->args_ct[i]; | ||
100 | ts = arg_temp(arg); | ||
101 | @@ -XXX,XX +XXX,XX @@ static void tcg_reg_alloc_op(TCGContext *s, const TCGOp *op) | ||
102 | int k2, i2; | ||
103 | reg = ts->reg; | ||
104 | for (k2 = 0 ; k2 < k ; k2++) { | ||
105 | - i2 = def->sorted_args[nb_oargs + k2]; | ||
106 | + i2 = def->args_ct[nb_oargs + k2].sort_index; | ||
107 | if ((def->args_ct[i2].ct & TCG_CT_IALIAS) && | ||
108 | reg == new_args[i2]) { | ||
109 | goto allocate_in_reg; | ||
110 | @@ -XXX,XX +XXX,XX @@ static void tcg_reg_alloc_op(TCGContext *s, const TCGOp *op) | ||
111 | |||
112 | /* satisfy the output constraints */ | ||
113 | for(k = 0; k < nb_oargs; k++) { | ||
114 | - i = def->sorted_args[k]; | ||
115 | + i = def->args_ct[k].sort_index; | ||
116 | arg = op->args[i]; | ||
117 | arg_ct = &def->args_ct[i]; | ||
118 | ts = arg_temp(arg); | ||
119 | -- | ||
120 | 2.25.1 | ||
121 | |||
122 | diff view generated by jsdifflib |
New patch | |||
---|---|---|---|
1 | This wasn't actually used for anything, really. All variable | ||
2 | operands must accept registers, and which are indicated by the | ||
3 | set in TCGArgConstraint.regs. | ||
1 | 4 | ||
5 | Signed-off-by: Richard Henderson <richard.henderson@linaro.org> | ||
6 | --- | ||
7 | include/tcg/tcg.h | 1 - | ||
8 | tcg/tcg.c | 15 ++++----------- | ||
9 | tcg/aarch64/tcg-target.c.inc | 3 --- | ||
10 | tcg/arm/tcg-target.c.inc | 3 --- | ||
11 | tcg/i386/tcg-target.c.inc | 11 ----------- | ||
12 | tcg/mips/tcg-target.c.inc | 3 --- | ||
13 | tcg/ppc/tcg-target.c.inc | 5 ----- | ||
14 | tcg/riscv/tcg-target.c.inc | 2 -- | ||
15 | tcg/s390/tcg-target.c.inc | 4 ---- | ||
16 | tcg/sparc/tcg-target.c.inc | 5 ----- | ||
17 | tcg/tci/tcg-target.c.inc | 1 - | ||
18 | 11 files changed, 4 insertions(+), 49 deletions(-) | ||
19 | |||
20 | diff --git a/include/tcg/tcg.h b/include/tcg/tcg.h | ||
21 | index XXXXXXX..XXXXXXX 100644 | ||
22 | --- a/include/tcg/tcg.h | ||
23 | +++ b/include/tcg/tcg.h | ||
24 | @@ -XXX,XX +XXX,XX @@ void tcg_dump_op_count(void); | ||
25 | #define TCG_CT_ALIAS 0x80 | ||
26 | #define TCG_CT_IALIAS 0x40 | ||
27 | #define TCG_CT_NEWREG 0x20 /* output requires a new register */ | ||
28 | -#define TCG_CT_REG 0x01 | ||
29 | #define TCG_CT_CONST 0x02 /* any constant of register size */ | ||
30 | |||
31 | typedef struct TCGArgConstraint { | ||
32 | diff --git a/tcg/tcg.c b/tcg/tcg.c | ||
33 | index XXXXXXX..XXXXXXX 100644 | ||
34 | --- a/tcg/tcg.c | ||
35 | +++ b/tcg/tcg.c | ||
36 | @@ -XXX,XX +XXX,XX @@ static void tcg_dump_ops(TCGContext *s, bool have_prefs) | ||
37 | /* we give more priority to constraints with less registers */ | ||
38 | static int get_constraint_priority(const TCGOpDef *def, int k) | ||
39 | { | ||
40 | - const TCGArgConstraint *arg_ct; | ||
41 | + const TCGArgConstraint *arg_ct = &def->args_ct[k]; | ||
42 | + int n; | ||
43 | |||
44 | - int i, n; | ||
45 | - arg_ct = &def->args_ct[k]; | ||
46 | if (arg_ct->ct & TCG_CT_ALIAS) { | ||
47 | /* an alias is equivalent to a single register */ | ||
48 | n = 1; | ||
49 | } else { | ||
50 | - if (!(arg_ct->ct & TCG_CT_REG)) | ||
51 | - return 0; | ||
52 | - n = 0; | ||
53 | - for(i = 0; i < TCG_TARGET_NB_REGS; i++) { | ||
54 | - if (tcg_regset_test_reg(arg_ct->regs, i)) | ||
55 | - n++; | ||
56 | - } | ||
57 | + n = ctpop64(arg_ct->regs); | ||
58 | } | ||
59 | return TCG_TARGET_NB_REGS - n + 1; | ||
60 | } | ||
61 | @@ -XXX,XX +XXX,XX @@ static void process_op_defs(TCGContext *s) | ||
62 | int oarg = *ct_str - '0'; | ||
63 | tcg_debug_assert(ct_str == tdefs->args_ct_str[i]); | ||
64 | tcg_debug_assert(oarg < def->nb_oargs); | ||
65 | - tcg_debug_assert(def->args_ct[oarg].ct & TCG_CT_REG); | ||
66 | + tcg_debug_assert(def->args_ct[oarg].regs != 0); | ||
67 | /* TCG_CT_ALIAS is for the output arguments. | ||
68 | The input is tagged with TCG_CT_IALIAS. */ | ||
69 | def->args_ct[i] = def->args_ct[oarg]; | ||
70 | diff --git a/tcg/aarch64/tcg-target.c.inc b/tcg/aarch64/tcg-target.c.inc | ||
71 | index XXXXXXX..XXXXXXX 100644 | ||
72 | --- a/tcg/aarch64/tcg-target.c.inc | ||
73 | +++ b/tcg/aarch64/tcg-target.c.inc | ||
74 | @@ -XXX,XX +XXX,XX @@ static const char *target_parse_constraint(TCGArgConstraint *ct, | ||
75 | { | ||
76 | switch (*ct_str++) { | ||
77 | case 'r': /* general registers */ | ||
78 | - ct->ct |= TCG_CT_REG; | ||
79 | ct->regs |= 0xffffffffu; | ||
80 | break; | ||
81 | case 'w': /* advsimd registers */ | ||
82 | - ct->ct |= TCG_CT_REG; | ||
83 | ct->regs |= 0xffffffff00000000ull; | ||
84 | break; | ||
85 | case 'l': /* qemu_ld / qemu_st address, data_reg */ | ||
86 | - ct->ct |= TCG_CT_REG; | ||
87 | ct->regs = 0xffffffffu; | ||
88 | #ifdef CONFIG_SOFTMMU | ||
89 | /* x0 and x1 will be overwritten when reading the tlb entry, | ||
90 | diff --git a/tcg/arm/tcg-target.c.inc b/tcg/arm/tcg-target.c.inc | ||
91 | index XXXXXXX..XXXXXXX 100644 | ||
92 | --- a/tcg/arm/tcg-target.c.inc | ||
93 | +++ b/tcg/arm/tcg-target.c.inc | ||
94 | @@ -XXX,XX +XXX,XX @@ static const char *target_parse_constraint(TCGArgConstraint *ct, | ||
95 | break; | ||
96 | |||
97 | case 'r': | ||
98 | - ct->ct |= TCG_CT_REG; | ||
99 | ct->regs = 0xffff; | ||
100 | break; | ||
101 | |||
102 | /* qemu_ld address */ | ||
103 | case 'l': | ||
104 | - ct->ct |= TCG_CT_REG; | ||
105 | ct->regs = 0xffff; | ||
106 | #ifdef CONFIG_SOFTMMU | ||
107 | /* r0-r2,lr will be overwritten when reading the tlb entry, | ||
108 | @@ -XXX,XX +XXX,XX @@ static const char *target_parse_constraint(TCGArgConstraint *ct, | ||
109 | |||
110 | /* qemu_st address & data */ | ||
111 | case 's': | ||
112 | - ct->ct |= TCG_CT_REG; | ||
113 | ct->regs = 0xffff; | ||
114 | /* r0-r2 will be overwritten when reading the tlb entry (softmmu only) | ||
115 | and r0-r1 doing the byte swapping, so don't use these. */ | ||
116 | diff --git a/tcg/i386/tcg-target.c.inc b/tcg/i386/tcg-target.c.inc | ||
117 | index XXXXXXX..XXXXXXX 100644 | ||
118 | --- a/tcg/i386/tcg-target.c.inc | ||
119 | +++ b/tcg/i386/tcg-target.c.inc | ||
120 | @@ -XXX,XX +XXX,XX @@ static const char *target_parse_constraint(TCGArgConstraint *ct, | ||
121 | { | ||
122 | switch(*ct_str++) { | ||
123 | case 'a': | ||
124 | - ct->ct |= TCG_CT_REG; | ||
125 | tcg_regset_set_reg(ct->regs, TCG_REG_EAX); | ||
126 | break; | ||
127 | case 'b': | ||
128 | - ct->ct |= TCG_CT_REG; | ||
129 | tcg_regset_set_reg(ct->regs, TCG_REG_EBX); | ||
130 | break; | ||
131 | case 'c': | ||
132 | - ct->ct |= TCG_CT_REG; | ||
133 | tcg_regset_set_reg(ct->regs, TCG_REG_ECX); | ||
134 | break; | ||
135 | case 'd': | ||
136 | - ct->ct |= TCG_CT_REG; | ||
137 | tcg_regset_set_reg(ct->regs, TCG_REG_EDX); | ||
138 | break; | ||
139 | case 'S': | ||
140 | - ct->ct |= TCG_CT_REG; | ||
141 | tcg_regset_set_reg(ct->regs, TCG_REG_ESI); | ||
142 | break; | ||
143 | case 'D': | ||
144 | - ct->ct |= TCG_CT_REG; | ||
145 | tcg_regset_set_reg(ct->regs, TCG_REG_EDI); | ||
146 | break; | ||
147 | case 'q': | ||
148 | /* A register that can be used as a byte operand. */ | ||
149 | - ct->ct |= TCG_CT_REG; | ||
150 | ct->regs = TCG_TARGET_REG_BITS == 64 ? 0xffff : 0xf; | ||
151 | break; | ||
152 | case 'Q': | ||
153 | /* A register with an addressable second byte (e.g. %ah). */ | ||
154 | - ct->ct |= TCG_CT_REG; | ||
155 | ct->regs = 0xf; | ||
156 | break; | ||
157 | case 'r': | ||
158 | /* A general register. */ | ||
159 | - ct->ct |= TCG_CT_REG; | ||
160 | ct->regs |= ALL_GENERAL_REGS; | ||
161 | break; | ||
162 | case 'W': | ||
163 | @@ -XXX,XX +XXX,XX @@ static const char *target_parse_constraint(TCGArgConstraint *ct, | ||
164 | break; | ||
165 | case 'x': | ||
166 | /* A vector register. */ | ||
167 | - ct->ct |= TCG_CT_REG; | ||
168 | ct->regs |= ALL_VECTOR_REGS; | ||
169 | break; | ||
170 | |||
171 | /* qemu_ld/st address constraint */ | ||
172 | case 'L': | ||
173 | - ct->ct |= TCG_CT_REG; | ||
174 | ct->regs = TCG_TARGET_REG_BITS == 64 ? 0xffff : 0xff; | ||
175 | tcg_regset_reset_reg(ct->regs, TCG_REG_L0); | ||
176 | tcg_regset_reset_reg(ct->regs, TCG_REG_L1); | ||
177 | diff --git a/tcg/mips/tcg-target.c.inc b/tcg/mips/tcg-target.c.inc | ||
178 | index XXXXXXX..XXXXXXX 100644 | ||
179 | --- a/tcg/mips/tcg-target.c.inc | ||
180 | +++ b/tcg/mips/tcg-target.c.inc | ||
181 | @@ -XXX,XX +XXX,XX @@ static const char *target_parse_constraint(TCGArgConstraint *ct, | ||
182 | { | ||
183 | switch(*ct_str++) { | ||
184 | case 'r': | ||
185 | - ct->ct |= TCG_CT_REG; | ||
186 | ct->regs = 0xffffffff; | ||
187 | break; | ||
188 | case 'L': /* qemu_ld input arg constraint */ | ||
189 | - ct->ct |= TCG_CT_REG; | ||
190 | ct->regs = 0xffffffff; | ||
191 | tcg_regset_reset_reg(ct->regs, TCG_REG_A0); | ||
192 | #if defined(CONFIG_SOFTMMU) | ||
193 | @@ -XXX,XX +XXX,XX @@ static const char *target_parse_constraint(TCGArgConstraint *ct, | ||
194 | #endif | ||
195 | break; | ||
196 | case 'S': /* qemu_st constraint */ | ||
197 | - ct->ct |= TCG_CT_REG; | ||
198 | ct->regs = 0xffffffff; | ||
199 | tcg_regset_reset_reg(ct->regs, TCG_REG_A0); | ||
200 | #if defined(CONFIG_SOFTMMU) | ||
201 | diff --git a/tcg/ppc/tcg-target.c.inc b/tcg/ppc/tcg-target.c.inc | ||
202 | index XXXXXXX..XXXXXXX 100644 | ||
203 | --- a/tcg/ppc/tcg-target.c.inc | ||
204 | +++ b/tcg/ppc/tcg-target.c.inc | ||
205 | @@ -XXX,XX +XXX,XX @@ static const char *target_parse_constraint(TCGArgConstraint *ct, | ||
206 | { | ||
207 | switch (*ct_str++) { | ||
208 | case 'A': case 'B': case 'C': case 'D': | ||
209 | - ct->ct |= TCG_CT_REG; | ||
210 | tcg_regset_set_reg(ct->regs, 3 + ct_str[0] - 'A'); | ||
211 | break; | ||
212 | case 'r': | ||
213 | - ct->ct |= TCG_CT_REG; | ||
214 | ct->regs = 0xffffffff; | ||
215 | break; | ||
216 | case 'v': | ||
217 | - ct->ct |= TCG_CT_REG; | ||
218 | ct->regs = 0xffffffff00000000ull; | ||
219 | break; | ||
220 | case 'L': /* qemu_ld constraint */ | ||
221 | - ct->ct |= TCG_CT_REG; | ||
222 | ct->regs = 0xffffffff; | ||
223 | tcg_regset_reset_reg(ct->regs, TCG_REG_R3); | ||
224 | #ifdef CONFIG_SOFTMMU | ||
225 | @@ -XXX,XX +XXX,XX @@ static const char *target_parse_constraint(TCGArgConstraint *ct, | ||
226 | #endif | ||
227 | break; | ||
228 | case 'S': /* qemu_st constraint */ | ||
229 | - ct->ct |= TCG_CT_REG; | ||
230 | ct->regs = 0xffffffff; | ||
231 | tcg_regset_reset_reg(ct->regs, TCG_REG_R3); | ||
232 | #ifdef CONFIG_SOFTMMU | ||
233 | diff --git a/tcg/riscv/tcg-target.c.inc b/tcg/riscv/tcg-target.c.inc | ||
234 | index XXXXXXX..XXXXXXX 100644 | ||
235 | --- a/tcg/riscv/tcg-target.c.inc | ||
236 | +++ b/tcg/riscv/tcg-target.c.inc | ||
237 | @@ -XXX,XX +XXX,XX @@ static const char *target_parse_constraint(TCGArgConstraint *ct, | ||
238 | { | ||
239 | switch (*ct_str++) { | ||
240 | case 'r': | ||
241 | - ct->ct |= TCG_CT_REG; | ||
242 | ct->regs = 0xffffffff; | ||
243 | break; | ||
244 | case 'L': | ||
245 | /* qemu_ld/qemu_st constraint */ | ||
246 | - ct->ct |= TCG_CT_REG; | ||
247 | ct->regs = 0xffffffff; | ||
248 | /* qemu_ld/qemu_st uses TCG_REG_TMP0 */ | ||
249 | #if defined(CONFIG_SOFTMMU) | ||
250 | diff --git a/tcg/s390/tcg-target.c.inc b/tcg/s390/tcg-target.c.inc | ||
251 | index XXXXXXX..XXXXXXX 100644 | ||
252 | --- a/tcg/s390/tcg-target.c.inc | ||
253 | +++ b/tcg/s390/tcg-target.c.inc | ||
254 | @@ -XXX,XX +XXX,XX @@ static const char *target_parse_constraint(TCGArgConstraint *ct, | ||
255 | { | ||
256 | switch (*ct_str++) { | ||
257 | case 'r': /* all registers */ | ||
258 | - ct->ct |= TCG_CT_REG; | ||
259 | ct->regs = 0xffff; | ||
260 | break; | ||
261 | case 'L': /* qemu_ld/st constraint */ | ||
262 | - ct->ct |= TCG_CT_REG; | ||
263 | ct->regs = 0xffff; | ||
264 | tcg_regset_reset_reg(ct->regs, TCG_REG_R2); | ||
265 | tcg_regset_reset_reg(ct->regs, TCG_REG_R3); | ||
266 | tcg_regset_reset_reg(ct->regs, TCG_REG_R4); | ||
267 | break; | ||
268 | case 'a': /* force R2 for division */ | ||
269 | - ct->ct |= TCG_CT_REG; | ||
270 | ct->regs = 0; | ||
271 | tcg_regset_set_reg(ct->regs, TCG_REG_R2); | ||
272 | break; | ||
273 | case 'b': /* force R3 for division */ | ||
274 | - ct->ct |= TCG_CT_REG; | ||
275 | ct->regs = 0; | ||
276 | tcg_regset_set_reg(ct->regs, TCG_REG_R3); | ||
277 | break; | ||
278 | diff --git a/tcg/sparc/tcg-target.c.inc b/tcg/sparc/tcg-target.c.inc | ||
279 | index XXXXXXX..XXXXXXX 100644 | ||
280 | --- a/tcg/sparc/tcg-target.c.inc | ||
281 | +++ b/tcg/sparc/tcg-target.c.inc | ||
282 | @@ -XXX,XX +XXX,XX @@ static const char *target_parse_constraint(TCGArgConstraint *ct, | ||
283 | { | ||
284 | switch (*ct_str++) { | ||
285 | case 'r': | ||
286 | - ct->ct |= TCG_CT_REG; | ||
287 | ct->regs = 0xffffffff; | ||
288 | break; | ||
289 | case 'R': | ||
290 | - ct->ct |= TCG_CT_REG; | ||
291 | ct->regs = ALL_64; | ||
292 | break; | ||
293 | case 'A': /* qemu_ld/st address constraint */ | ||
294 | - ct->ct |= TCG_CT_REG; | ||
295 | ct->regs = TARGET_LONG_BITS == 64 ? ALL_64 : 0xffffffff; | ||
296 | reserve_helpers: | ||
297 | tcg_regset_reset_reg(ct->regs, TCG_REG_O0); | ||
298 | @@ -XXX,XX +XXX,XX @@ static const char *target_parse_constraint(TCGArgConstraint *ct, | ||
299 | tcg_regset_reset_reg(ct->regs, TCG_REG_O2); | ||
300 | break; | ||
301 | case 's': /* qemu_st data 32-bit constraint */ | ||
302 | - ct->ct |= TCG_CT_REG; | ||
303 | ct->regs = 0xffffffff; | ||
304 | goto reserve_helpers; | ||
305 | case 'S': /* qemu_st data 64-bit constraint */ | ||
306 | - ct->ct |= TCG_CT_REG; | ||
307 | ct->regs = ALL_64; | ||
308 | goto reserve_helpers; | ||
309 | case 'I': | ||
310 | diff --git a/tcg/tci/tcg-target.c.inc b/tcg/tci/tcg-target.c.inc | ||
311 | index XXXXXXX..XXXXXXX 100644 | ||
312 | --- a/tcg/tci/tcg-target.c.inc | ||
313 | +++ b/tcg/tci/tcg-target.c.inc | ||
314 | @@ -XXX,XX +XXX,XX @@ static const char *target_parse_constraint(TCGArgConstraint *ct, | ||
315 | case 'r': | ||
316 | case 'L': /* qemu_ld constraint */ | ||
317 | case 'S': /* qemu_st constraint */ | ||
318 | - ct->ct |= TCG_CT_REG; | ||
319 | ct->regs = BIT(TCG_TARGET_NB_REGS) - 1; | ||
320 | break; | ||
321 | default: | ||
322 | -- | ||
323 | 2.25.1 | ||
324 | |||
325 | diff view generated by jsdifflib |
1 | Turn helper_retaddr into a multi-state flag that may now also | 1 | These are easier to set and test when they have their own fields. |
---|---|---|---|
2 | indicate when we're performing a read on behalf of the translator. | 2 | Reduce the size of alias_index and sort_index to 4 bits, which is |
3 | In this case, release the mmap_lock before the longjmp back to | 3 | sufficient for TCG_MAX_OP_ARGS. This leaves only the bits indicating |
4 | the main cpu loop, and thereby avoid a failing assert therein. | 4 | constants within the ct field. |
5 | 5 | ||
6 | Fixes: https://bugs.launchpad.net/qemu/+bug/1832353 | 6 | Move all initialization to allocation time, rather than init |
7 | Tested-by: Alex Bennée <alex.bennee@linaro.org> | 7 | individual fields in process_op_defs. |
8 | Reviewed-by: Alex Bennée <alex.bennee@linaro.org> | 8 | |
9 | Signed-off-by: Richard Henderson <richard.henderson@linaro.org> | 9 | Signed-off-by: Richard Henderson <richard.henderson@linaro.org> |
10 | --- | 10 | --- |
11 | include/exec/cpu_ldst_useronly_template.h | 20 +++++-- | 11 | include/tcg/tcg.h | 14 +++++++------- |
12 | accel/tcg/user-exec.c | 66 ++++++++++++++++------- | 12 | tcg/tcg.c | 28 ++++++++++++---------------- |
13 | 2 files changed, 63 insertions(+), 23 deletions(-) | 13 | 2 files changed, 19 insertions(+), 23 deletions(-) |
14 | 14 | ||
15 | diff --git a/include/exec/cpu_ldst_useronly_template.h b/include/exec/cpu_ldst_useronly_template.h | 15 | diff --git a/include/tcg/tcg.h b/include/tcg/tcg.h |
16 | index XXXXXXX..XXXXXXX 100644 | 16 | index XXXXXXX..XXXXXXX 100644 |
17 | --- a/include/exec/cpu_ldst_useronly_template.h | 17 | --- a/include/tcg/tcg.h |
18 | +++ b/include/exec/cpu_ldst_useronly_template.h | 18 | +++ b/include/tcg/tcg.h |
19 | @@ -XXX,XX +XXX,XX @@ | 19 | @@ -XXX,XX +XXX,XX @@ int64_t tcg_cpu_exec_time(void); |
20 | static inline RES_TYPE | 20 | void tcg_dump_info(void); |
21 | glue(glue(cpu_ld, USUFFIX), MEMSUFFIX)(CPUArchState *env, abi_ptr ptr) | 21 | void tcg_dump_op_count(void); |
22 | { | 22 | |
23 | -#if !defined(CODE_ACCESS) | 23 | -#define TCG_CT_ALIAS 0x80 |
24 | +#ifdef CODE_ACCESS | 24 | -#define TCG_CT_IALIAS 0x40 |
25 | + RES_TYPE ret; | 25 | -#define TCG_CT_NEWREG 0x20 /* output requires a new register */ |
26 | + set_helper_retaddr(1); | 26 | -#define TCG_CT_CONST 0x02 /* any constant of register size */ |
27 | + ret = glue(glue(ld, USUFFIX), _p)(g2h(ptr)); | 27 | +#define TCG_CT_CONST 1 /* any constant of register size */ |
28 | + clear_helper_retaddr(); | 28 | |
29 | + return ret; | 29 | typedef struct TCGArgConstraint { |
30 | +#else | 30 | - uint16_t ct; |
31 | trace_guest_mem_before_exec( | 31 | - uint8_t alias_index; |
32 | env_cpu(env), ptr, | 32 | - uint8_t sort_index; |
33 | trace_mem_build_info(SHIFT, false, MO_TE, false)); | 33 | + unsigned ct : 16; |
34 | -#endif | 34 | + unsigned alias_index : 4; |
35 | return glue(glue(ld, USUFFIX), _p)(g2h(ptr)); | 35 | + unsigned sort_index : 4; |
36 | +#endif | 36 | + bool oalias : 1; |
37 | } | 37 | + bool ialias : 1; |
38 | 38 | + bool newreg : 1; | |
39 | #ifndef CODE_ACCESS | 39 | TCGRegSet regs; |
40 | @@ -XXX,XX +XXX,XX @@ glue(glue(glue(cpu_ld, USUFFIX), MEMSUFFIX), _ra)(CPUArchState *env, | 40 | } TCGArgConstraint; |
41 | static inline int | 41 | |
42 | glue(glue(cpu_lds, SUFFIX), MEMSUFFIX)(CPUArchState *env, abi_ptr ptr) | 42 | diff --git a/tcg/tcg.c b/tcg/tcg.c |
43 | { | ||
44 | -#if !defined(CODE_ACCESS) | ||
45 | +#ifdef CODE_ACCESS | ||
46 | + int ret; | ||
47 | + set_helper_retaddr(1); | ||
48 | + ret = glue(glue(lds, SUFFIX), _p)(g2h(ptr)); | ||
49 | + clear_helper_retaddr(); | ||
50 | + return ret; | ||
51 | +#else | ||
52 | trace_guest_mem_before_exec( | ||
53 | env_cpu(env), ptr, | ||
54 | trace_mem_build_info(SHIFT, true, MO_TE, false)); | ||
55 | -#endif | ||
56 | return glue(glue(lds, SUFFIX), _p)(g2h(ptr)); | ||
57 | +#endif | ||
58 | } | ||
59 | |||
60 | #ifndef CODE_ACCESS | ||
61 | diff --git a/accel/tcg/user-exec.c b/accel/tcg/user-exec.c | ||
62 | index XXXXXXX..XXXXXXX 100644 | 43 | index XXXXXXX..XXXXXXX 100644 |
63 | --- a/accel/tcg/user-exec.c | 44 | --- a/tcg/tcg.c |
64 | +++ b/accel/tcg/user-exec.c | 45 | +++ b/tcg/tcg.c |
65 | @@ -XXX,XX +XXX,XX @@ static inline int handle_cpu_signal(uintptr_t pc, siginfo_t *info, | 46 | @@ -XXX,XX +XXX,XX @@ void tcg_context_init(TCGContext *s) |
66 | CPUState *cpu = current_cpu; | 47 | total_args += n; |
67 | CPUClass *cc; | ||
68 | unsigned long address = (unsigned long)info->si_addr; | ||
69 | - MMUAccessType access_type; | ||
70 | + MMUAccessType access_type = is_write ? MMU_DATA_STORE : MMU_DATA_LOAD; | ||
71 | |||
72 | - /* We must handle PC addresses from two different sources: | ||
73 | - * a call return address and a signal frame address. | ||
74 | - * | ||
75 | - * Within cpu_restore_state_from_tb we assume the former and adjust | ||
76 | - * the address by -GETPC_ADJ so that the address is within the call | ||
77 | - * insn so that addr does not accidentally match the beginning of the | ||
78 | - * next guest insn. | ||
79 | - * | ||
80 | - * However, when the PC comes from the signal frame, it points to | ||
81 | - * the actual faulting host insn and not a call insn. Subtracting | ||
82 | - * GETPC_ADJ in that case may accidentally match the previous guest insn. | ||
83 | - * | ||
84 | - * So for the later case, adjust forward to compensate for what | ||
85 | - * will be done later by cpu_restore_state_from_tb. | ||
86 | - */ | ||
87 | - if (helper_retaddr) { | ||
88 | + switch (helper_retaddr) { | ||
89 | + default: | ||
90 | + /* | ||
91 | + * Fault during host memory operation within a helper function. | ||
92 | + * The helper's host return address, saved here, gives us a | ||
93 | + * pointer into the generated code that will unwind to the | ||
94 | + * correct guest pc. | ||
95 | + */ | ||
96 | pc = helper_retaddr; | ||
97 | - } else { | ||
98 | + break; | ||
99 | + | ||
100 | + case 0: | ||
101 | + /* | ||
102 | + * Fault during host memory operation within generated code. | ||
103 | + * (Or, a unrelated bug within qemu, but we can't tell from here). | ||
104 | + * | ||
105 | + * We take the host pc from the signal frame. However, we cannot | ||
106 | + * use that value directly. Within cpu_restore_state_from_tb, we | ||
107 | + * assume PC comes from GETPC(), as used by the helper functions, | ||
108 | + * so we adjust the address by -GETPC_ADJ to form an address that | ||
109 | + * is within the call insn, so that the address does not accidentially | ||
110 | + * match the beginning of the next guest insn. However, when the | ||
111 | + * pc comes from the signal frame it points to the actual faulting | ||
112 | + * host memory insn and not the return from a call insn. | ||
113 | + * | ||
114 | + * Therefore, adjust to compensate for what will be done later | ||
115 | + * by cpu_restore_state_from_tb. | ||
116 | + */ | ||
117 | pc += GETPC_ADJ; | ||
118 | + break; | ||
119 | + | ||
120 | + case 1: | ||
121 | + /* | ||
122 | + * Fault during host read for translation, or loosely, "execution". | ||
123 | + * | ||
124 | + * The guest pc is already pointing to the start of the TB for which | ||
125 | + * code is being generated. If the guest translator manages the | ||
126 | + * page crossings correctly, this is exactly the correct address | ||
127 | + * (and if the translator doesn't handle page boundaries correctly | ||
128 | + * there's little we can do about that here). Therefore, do not | ||
129 | + * trigger the unwinder. | ||
130 | + * | ||
131 | + * Like tb_gen_code, release the memory lock before cpu_loop_exit. | ||
132 | + */ | ||
133 | + pc = 0; | ||
134 | + access_type = MMU_INST_FETCH; | ||
135 | + mmap_unlock(); | ||
136 | + break; | ||
137 | } | 48 | } |
138 | 49 | ||
139 | /* For synchronous signals we expect to be coming from the vCPU | 50 | - args_ct = g_malloc(sizeof(TCGArgConstraint) * total_args); |
140 | @@ -XXX,XX +XXX,XX @@ static inline int handle_cpu_signal(uintptr_t pc, siginfo_t *info, | 51 | + args_ct = g_new0(TCGArgConstraint, total_args); |
141 | clear_helper_retaddr(); | 52 | |
142 | 53 | for(op = 0; op < NB_OPS; op++) { | |
143 | cc = CPU_GET_CLASS(cpu); | 54 | def = &tcg_op_defs[op]; |
144 | - access_type = is_write ? MMU_DATA_STORE : MMU_DATA_LOAD; | 55 | @@ -XXX,XX +XXX,XX @@ static int get_constraint_priority(const TCGOpDef *def, int k) |
145 | cc->tlb_fill(cpu, address, 0, access_type, MMU_USER_IDX, false, pc); | 56 | const TCGArgConstraint *arg_ct = &def->args_ct[k]; |
146 | g_assert_not_reached(); | 57 | int n; |
147 | } | 58 | |
59 | - if (arg_ct->ct & TCG_CT_ALIAS) { | ||
60 | + if (arg_ct->oalias) { | ||
61 | /* an alias is equivalent to a single register */ | ||
62 | n = 1; | ||
63 | } else { | ||
64 | @@ -XXX,XX +XXX,XX @@ static void process_op_defs(TCGContext *s) | ||
65 | /* Incomplete TCGTargetOpDef entry. */ | ||
66 | tcg_debug_assert(ct_str != NULL); | ||
67 | |||
68 | - def->args_ct[i].regs = 0; | ||
69 | - def->args_ct[i].ct = 0; | ||
70 | while (*ct_str != '\0') { | ||
71 | switch(*ct_str) { | ||
72 | case '0' ... '9': | ||
73 | @@ -XXX,XX +XXX,XX @@ static void process_op_defs(TCGContext *s) | ||
74 | tcg_debug_assert(ct_str == tdefs->args_ct_str[i]); | ||
75 | tcg_debug_assert(oarg < def->nb_oargs); | ||
76 | tcg_debug_assert(def->args_ct[oarg].regs != 0); | ||
77 | - /* TCG_CT_ALIAS is for the output arguments. | ||
78 | - The input is tagged with TCG_CT_IALIAS. */ | ||
79 | def->args_ct[i] = def->args_ct[oarg]; | ||
80 | - def->args_ct[oarg].ct |= TCG_CT_ALIAS; | ||
81 | + /* The output sets oalias. */ | ||
82 | + def->args_ct[oarg].oalias = true; | ||
83 | def->args_ct[oarg].alias_index = i; | ||
84 | - def->args_ct[i].ct |= TCG_CT_IALIAS; | ||
85 | + /* The input sets ialias. */ | ||
86 | + def->args_ct[i].ialias = true; | ||
87 | def->args_ct[i].alias_index = oarg; | ||
88 | } | ||
89 | ct_str++; | ||
90 | break; | ||
91 | case '&': | ||
92 | - def->args_ct[i].ct |= TCG_CT_NEWREG; | ||
93 | + def->args_ct[i].newreg = true; | ||
94 | ct_str++; | ||
95 | break; | ||
96 | case 'i': | ||
97 | @@ -XXX,XX +XXX,XX @@ static void liveness_pass_1(TCGContext *s) | ||
98 | set = *pset; | ||
99 | |||
100 | set &= ct->regs; | ||
101 | - if (ct->ct & TCG_CT_IALIAS) { | ||
102 | + if (ct->ialias) { | ||
103 | set &= op->output_pref[ct->alias_index]; | ||
104 | } | ||
105 | /* If the combination is not possible, restart. */ | ||
106 | @@ -XXX,XX +XXX,XX @@ static void tcg_reg_alloc_op(TCGContext *s, const TCGOp *op) | ||
107 | } | ||
108 | |||
109 | i_preferred_regs = o_preferred_regs = 0; | ||
110 | - if (arg_ct->ct & TCG_CT_IALIAS) { | ||
111 | + if (arg_ct->ialias) { | ||
112 | o_preferred_regs = op->output_pref[arg_ct->alias_index]; | ||
113 | if (ts->fixed_reg) { | ||
114 | /* if fixed register, we must allocate a new register | ||
115 | @@ -XXX,XX +XXX,XX @@ static void tcg_reg_alloc_op(TCGContext *s, const TCGOp *op) | ||
116 | reg = ts->reg; | ||
117 | for (k2 = 0 ; k2 < k ; k2++) { | ||
118 | i2 = def->args_ct[nb_oargs + k2].sort_index; | ||
119 | - if ((def->args_ct[i2].ct & TCG_CT_IALIAS) && | ||
120 | - reg == new_args[i2]) { | ||
121 | + if (def->args_ct[i2].ialias && reg == new_args[i2]) { | ||
122 | goto allocate_in_reg; | ||
123 | } | ||
124 | } | ||
125 | @@ -XXX,XX +XXX,XX @@ static void tcg_reg_alloc_op(TCGContext *s, const TCGOp *op) | ||
126 | /* ENV should not be modified. */ | ||
127 | tcg_debug_assert(!ts->fixed_reg); | ||
128 | |||
129 | - if ((arg_ct->ct & TCG_CT_ALIAS) | ||
130 | - && !const_args[arg_ct->alias_index]) { | ||
131 | + if (arg_ct->oalias && !const_args[arg_ct->alias_index]) { | ||
132 | reg = new_args[arg_ct->alias_index]; | ||
133 | - } else if (arg_ct->ct & TCG_CT_NEWREG) { | ||
134 | + } else if (arg_ct->newreg) { | ||
135 | reg = tcg_reg_alloc(s, arg_ct->regs, | ||
136 | i_allocated_regs | o_allocated_regs, | ||
137 | op->output_pref[k], ts->indirect_base); | ||
148 | -- | 138 | -- |
149 | 2.17.1 | 139 | 2.25.1 |
150 | 140 | ||
151 | 141 | diff view generated by jsdifflib |
New patch | |||
---|---|---|---|
1 | The last user of this field disappeared in f69d277ece4. | ||
1 | 2 | ||
3 | Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org> | ||
4 | Signed-off-by: Richard Henderson <richard.henderson@linaro.org> | ||
5 | --- | ||
6 | include/tcg/tcg.h | 3 --- | ||
7 | 1 file changed, 3 deletions(-) | ||
8 | |||
9 | diff --git a/include/tcg/tcg.h b/include/tcg/tcg.h | ||
10 | index XXXXXXX..XXXXXXX 100644 | ||
11 | --- a/include/tcg/tcg.h | ||
12 | +++ b/include/tcg/tcg.h | ||
13 | @@ -XXX,XX +XXX,XX @@ typedef struct TCGOpDef { | ||
14 | uint8_t nb_oargs, nb_iargs, nb_cargs, nb_args; | ||
15 | uint8_t flags; | ||
16 | TCGArgConstraint *args_ct; | ||
17 | -#if defined(CONFIG_DEBUG_TCG) | ||
18 | - int used; | ||
19 | -#endif | ||
20 | } TCGOpDef; | ||
21 | |||
22 | extern TCGOpDef tcg_op_defs[]; | ||
23 | -- | ||
24 | 2.25.1 | ||
25 | |||
26 | diff view generated by jsdifflib |
1 | This patch fixes two problems: | 1 | The previous change wrongly stated that 32-bit avx2 should have |
---|---|---|---|
2 | (1) The inputs to the EXTR insn were reversed, | 2 | used VPBROADCASTW. But that's a 16-bit broadcast and we want a |
3 | (2) The input constraints use rZ, which means that we need to use | 3 | 32-bit broadcast. |
4 | the REG0 macro in order to supply XZR for a constant 0 input. | ||
5 | 4 | ||
6 | Fixes: 464c2969d5d | 5 | Fixes: 7b60ef3264e |
7 | Reported-by: Peter Maydell <peter.maydell@linaro.org> | 6 | Cc: qemu-stable@nongnu.org |
8 | Tested-by: Alex Bennée <alex.bennee@linaro.org> | ||
9 | Reviewed-by: Alex Bennée <alex.bennee@linaro.org> | ||
10 | Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com> | ||
11 | Signed-off-by: Richard Henderson <richard.henderson@linaro.org> | 7 | Signed-off-by: Richard Henderson <richard.henderson@linaro.org> |
12 | --- | 8 | --- |
13 | tcg/aarch64/tcg-target.inc.c | 2 +- | 9 | tcg/i386/tcg-target.c.inc | 2 +- |
14 | 1 file changed, 1 insertion(+), 1 deletion(-) | 10 | 1 file changed, 1 insertion(+), 1 deletion(-) |
15 | 11 | ||
16 | diff --git a/tcg/aarch64/tcg-target.inc.c b/tcg/aarch64/tcg-target.inc.c | 12 | diff --git a/tcg/i386/tcg-target.c.inc b/tcg/i386/tcg-target.c.inc |
17 | index XXXXXXX..XXXXXXX 100644 | 13 | index XXXXXXX..XXXXXXX 100644 |
18 | --- a/tcg/aarch64/tcg-target.inc.c | 14 | --- a/tcg/i386/tcg-target.c.inc |
19 | +++ b/tcg/aarch64/tcg-target.inc.c | 15 | +++ b/tcg/i386/tcg-target.c.inc |
20 | @@ -XXX,XX +XXX,XX @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc, | 16 | @@ -XXX,XX +XXX,XX @@ static void tcg_out_dupi_vec(TCGContext *s, TCGType type, |
21 | 17 | new_pool_label(s, arg, R_386_PC32, s->code_ptr - 4, -4); | |
22 | case INDEX_op_extract2_i64: | 18 | } else { |
23 | case INDEX_op_extract2_i32: | 19 | if (have_avx2) { |
24 | - tcg_out_extr(s, ext, a0, a1, a2, args[3]); | 20 | - tcg_out_vex_modrm_pool(s, OPC_VPBROADCASTW + vex_l, ret); |
25 | + tcg_out_extr(s, ext, a0, REG0(2), REG0(1), args[3]); | 21 | + tcg_out_vex_modrm_pool(s, OPC_VPBROADCASTD + vex_l, ret); |
26 | break; | 22 | } else { |
27 | 23 | tcg_out_vex_modrm_pool(s, OPC_VBROADCASTSS, ret); | |
28 | case INDEX_op_add2_i32: | 24 | } |
29 | -- | 25 | -- |
30 | 2.17.1 | 26 | 2.25.1 |
31 | 27 | ||
32 | 28 | diff view generated by jsdifflib |
1 | These functions are not used, and are not usable in the | 1 | The definition of INDEX_op_dupi_vec is that it operates on |
---|---|---|---|
2 | context of code generation, because we never have a helper | 2 | units of tcg_target_ulong -- in this case 32 bits. It does |
3 | return address to pass in to them. | 3 | not work to use this for a uint64_t value that happens to be |
4 | small enough to fit in tcg_target_ulong. | ||
4 | 5 | ||
5 | Reviewed-by: Alex Bennée <alex.bennee@linaro.org> | 6 | Fixes: d2fd745fe8b |
7 | Fixes: db432672dc5 | ||
8 | Cc: qemu-stable@nongnu.org | ||
6 | Signed-off-by: Richard Henderson <richard.henderson@linaro.org> | 9 | Signed-off-by: Richard Henderson <richard.henderson@linaro.org> |
7 | --- | 10 | --- |
8 | include/exec/cpu_ldst_useronly_template.h | 6 +++++- | 11 | tcg/tcg-op-vec.c | 12 ++++++++---- |
9 | 1 file changed, 5 insertions(+), 1 deletion(-) | 12 | 1 file changed, 8 insertions(+), 4 deletions(-) |
10 | 13 | ||
11 | diff --git a/include/exec/cpu_ldst_useronly_template.h b/include/exec/cpu_ldst_useronly_template.h | 14 | diff --git a/tcg/tcg-op-vec.c b/tcg/tcg-op-vec.c |
12 | index XXXXXXX..XXXXXXX 100644 | 15 | index XXXXXXX..XXXXXXX 100644 |
13 | --- a/include/exec/cpu_ldst_useronly_template.h | 16 | --- a/tcg/tcg-op-vec.c |
14 | +++ b/include/exec/cpu_ldst_useronly_template.h | 17 | +++ b/tcg/tcg-op-vec.c |
15 | @@ -XXX,XX +XXX,XX @@ glue(glue(cpu_ld, USUFFIX), MEMSUFFIX)(CPUArchState *env, abi_ptr ptr) | 18 | @@ -XXX,XX +XXX,XX @@ TCGv_vec tcg_const_ones_vec_matching(TCGv_vec m) |
16 | return glue(glue(ld, USUFFIX), _p)(g2h(ptr)); | 19 | |
20 | void tcg_gen_dup64i_vec(TCGv_vec r, uint64_t a) | ||
21 | { | ||
22 | - if (TCG_TARGET_REG_BITS == 32 && a == deposit64(a, 32, 32, a)) { | ||
23 | - do_dupi_vec(r, MO_32, a); | ||
24 | - } else if (TCG_TARGET_REG_BITS == 64 || a == (uint64_t)(int32_t)a) { | ||
25 | + if (TCG_TARGET_REG_BITS == 64) { | ||
26 | do_dupi_vec(r, MO_64, a); | ||
27 | + } else if (a == dup_const(MO_32, a)) { | ||
28 | + do_dupi_vec(r, MO_32, a); | ||
29 | } else { | ||
30 | TCGv_i64 c = tcg_const_i64(a); | ||
31 | tcg_gen_dup_i64_vec(MO_64, r, c); | ||
32 | @@ -XXX,XX +XXX,XX @@ void tcg_gen_dup8i_vec(TCGv_vec r, uint32_t a) | ||
33 | |||
34 | void tcg_gen_dupi_vec(unsigned vece, TCGv_vec r, uint64_t a) | ||
35 | { | ||
36 | - do_dupi_vec(r, MO_REG, dup_const(vece, a)); | ||
37 | + if (vece == MO_64) { | ||
38 | + tcg_gen_dup64i_vec(r, a); | ||
39 | + } else { | ||
40 | + do_dupi_vec(r, MO_REG, dup_const(vece, a)); | ||
41 | + } | ||
17 | } | 42 | } |
18 | 43 | ||
19 | +#ifndef CODE_ACCESS | 44 | void tcg_gen_dup_i64_vec(unsigned vece, TCGv_vec r, TCGv_i64 a) |
20 | static inline RES_TYPE | ||
21 | glue(glue(glue(cpu_ld, USUFFIX), MEMSUFFIX), _ra)(CPUArchState *env, | ||
22 | abi_ptr ptr, | ||
23 | @@ -XXX,XX +XXX,XX @@ glue(glue(glue(cpu_ld, USUFFIX), MEMSUFFIX), _ra)(CPUArchState *env, | ||
24 | clear_helper_retaddr(); | ||
25 | return ret; | ||
26 | } | ||
27 | +#endif | ||
28 | |||
29 | #if DATA_SIZE <= 2 | ||
30 | static inline int | ||
31 | @@ -XXX,XX +XXX,XX @@ glue(glue(cpu_lds, SUFFIX), MEMSUFFIX)(CPUArchState *env, abi_ptr ptr) | ||
32 | return glue(glue(lds, SUFFIX), _p)(g2h(ptr)); | ||
33 | } | ||
34 | |||
35 | +#ifndef CODE_ACCESS | ||
36 | static inline int | ||
37 | glue(glue(glue(cpu_lds, SUFFIX), MEMSUFFIX), _ra)(CPUArchState *env, | ||
38 | abi_ptr ptr, | ||
39 | @@ -XXX,XX +XXX,XX @@ glue(glue(glue(cpu_lds, SUFFIX), MEMSUFFIX), _ra)(CPUArchState *env, | ||
40 | clear_helper_retaddr(); | ||
41 | return ret; | ||
42 | } | ||
43 | -#endif | ||
44 | +#endif /* CODE_ACCESS */ | ||
45 | +#endif /* DATA_SIZE <= 2 */ | ||
46 | |||
47 | #ifndef CODE_ACCESS | ||
48 | static inline void | ||
49 | -- | 45 | -- |
50 | 2.17.1 | 46 | 2.25.1 |
51 | 47 | ||
52 | 48 | diff view generated by jsdifflib |
1 | On a 64-bit host, discard any replications of the 32-bit | 1 | When the two arguments are identical, this can be reduced to |
---|---|---|---|
2 | sign bit when performing the shift and merge. | 2 | dup_vec or to mov_vec from a tcg_constant_vec. |
3 | 3 | ||
4 | Fixes: https://bugs.launchpad.net/bugs/1834496 | ||
5 | Tested-by: Christophe Lyon <christophe.lyon@linaro.org> | ||
6 | Tested-by: Alex Bennée <alex.bennee@linaro.org> | ||
7 | Reviewed-by: Alex Bennée <alex.bennee@linaro.org> | ||
8 | Signed-off-by: Richard Henderson <richard.henderson@linaro.org> | 4 | Signed-off-by: Richard Henderson <richard.henderson@linaro.org> |
9 | --- | 5 | --- |
10 | tcg/optimize.c | 4 ++-- | 6 | tcg/optimize.c | 15 +++++++++++++++ |
11 | 1 file changed, 2 insertions(+), 2 deletions(-) | 7 | 1 file changed, 15 insertions(+) |
12 | 8 | ||
13 | diff --git a/tcg/optimize.c b/tcg/optimize.c | 9 | diff --git a/tcg/optimize.c b/tcg/optimize.c |
14 | index XXXXXXX..XXXXXXX 100644 | 10 | index XXXXXXX..XXXXXXX 100644 |
15 | --- a/tcg/optimize.c | 11 | --- a/tcg/optimize.c |
16 | +++ b/tcg/optimize.c | 12 | +++ b/tcg/optimize.c |
17 | @@ -XXX,XX +XXX,XX @@ void tcg_optimize(TCGContext *s) | 13 | @@ -XXX,XX +XXX,XX @@ void tcg_optimize(TCGContext *s) |
18 | if (opc == INDEX_op_extract2_i64) { | 14 | } |
19 | tmp = (v1 >> op->args[3]) | (v2 << (64 - op->args[3])); | 15 | goto do_default; |
20 | } else { | 16 | |
21 | - tmp = (v1 >> op->args[3]) | (v2 << (32 - op->args[3])); | 17 | + case INDEX_op_dup2_vec: |
22 | - tmp = (int32_t)tmp; | 18 | + assert(TCG_TARGET_REG_BITS == 32); |
23 | + tmp = (int32_t)(((uint32_t)v1 >> op->args[3]) | | 19 | + if (arg_is_const(op->args[1]) && arg_is_const(op->args[2])) { |
24 | + ((uint32_t)v2 << (32 - op->args[3]))); | 20 | + tmp = arg_info(op->args[1])->val; |
25 | } | 21 | + if (tmp == arg_info(op->args[2])->val) { |
26 | tcg_opt_gen_movi(s, op, op->args[0], tmp); | 22 | + tcg_opt_gen_movi(s, op, op->args[0], tmp); |
27 | break; | 23 | + break; |
24 | + } | ||
25 | + } else if (args_are_copies(op->args[1], op->args[2])) { | ||
26 | + op->opc = INDEX_op_dup_vec; | ||
27 | + TCGOP_VECE(op) = MO_32; | ||
28 | + nb_iargs = 1; | ||
29 | + } | ||
30 | + goto do_default; | ||
31 | + | ||
32 | CASE_OP_32_64(not): | ||
33 | CASE_OP_32_64(neg): | ||
34 | CASE_OP_32_64(ext8s): | ||
28 | -- | 35 | -- |
29 | 2.17.1 | 36 | 2.25.1 |
30 | 37 | ||
31 | 38 | diff view generated by jsdifflib |
1 | We have some potential race conditions vs our user-exec signal | 1 | The cmp_vec opcode is mandatory; this symbol is unused. |
---|---|---|---|
2 | handler that will be solved with this barrier. | ||
3 | 2 | ||
4 | Reviewed-by: Alex Bennée <alex.bennee@linaro.org> | 3 | Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org> |
5 | Signed-off-by: Richard Henderson <richard.henderson@linaro.org> | 4 | Signed-off-by: Richard Henderson <richard.henderson@linaro.org> |
6 | --- | 5 | --- |
7 | include/qemu/atomic.h | 11 +++++++++++ | 6 | tcg/aarch64/tcg-target.h | 1 - |
8 | 1 file changed, 11 insertions(+) | 7 | tcg/i386/tcg-target.h | 1 - |
8 | tcg/ppc/tcg-target.h | 1 - | ||
9 | 3 files changed, 3 deletions(-) | ||
9 | 10 | ||
10 | diff --git a/include/qemu/atomic.h b/include/qemu/atomic.h | 11 | diff --git a/tcg/aarch64/tcg-target.h b/tcg/aarch64/tcg-target.h |
11 | index XXXXXXX..XXXXXXX 100644 | 12 | index XXXXXXX..XXXXXXX 100644 |
12 | --- a/include/qemu/atomic.h | 13 | --- a/tcg/aarch64/tcg-target.h |
13 | +++ b/include/qemu/atomic.h | 14 | +++ b/tcg/aarch64/tcg-target.h |
14 | @@ -XXX,XX +XXX,XX @@ | 15 | @@ -XXX,XX +XXX,XX @@ typedef enum { |
15 | #define smp_read_barrier_depends() barrier() | 16 | #define TCG_TARGET_HAS_shi_vec 1 |
16 | #endif | 17 | #define TCG_TARGET_HAS_shs_vec 0 |
17 | 18 | #define TCG_TARGET_HAS_shv_vec 1 | |
18 | +/* | 19 | -#define TCG_TARGET_HAS_cmp_vec 1 |
19 | + * A signal barrier forces all pending local memory ops to be observed before | 20 | #define TCG_TARGET_HAS_mul_vec 1 |
20 | + * a SIGSEGV is delivered to the *same* thread. In practice this is exactly | 21 | #define TCG_TARGET_HAS_sat_vec 1 |
21 | + * the same as barrier(), but since we have the correct builtin, use it. | 22 | #define TCG_TARGET_HAS_minmax_vec 1 |
22 | + */ | 23 | diff --git a/tcg/i386/tcg-target.h b/tcg/i386/tcg-target.h |
23 | +#define signal_barrier() __atomic_signal_fence(__ATOMIC_SEQ_CST) | 24 | index XXXXXXX..XXXXXXX 100644 |
24 | + | 25 | --- a/tcg/i386/tcg-target.h |
25 | /* Sanity check that the size of an atomic operation isn't "overly large". | 26 | +++ b/tcg/i386/tcg-target.h |
26 | * Despite the fact that e.g. i686 has 64-bit atomic operations, we do not | 27 | @@ -XXX,XX +XXX,XX @@ extern bool have_avx2; |
27 | * want to use them because we ought not need them, and this lets us do a | 28 | #define TCG_TARGET_HAS_shi_vec 1 |
28 | @@ -XXX,XX +XXX,XX @@ | 29 | #define TCG_TARGET_HAS_shs_vec 1 |
29 | #define smp_read_barrier_depends() barrier() | 30 | #define TCG_TARGET_HAS_shv_vec have_avx2 |
30 | #endif | 31 | -#define TCG_TARGET_HAS_cmp_vec 1 |
31 | 32 | #define TCG_TARGET_HAS_mul_vec 1 | |
32 | +#ifndef signal_barrier | 33 | #define TCG_TARGET_HAS_sat_vec 1 |
33 | +#define signal_barrier() barrier() | 34 | #define TCG_TARGET_HAS_minmax_vec 1 |
34 | +#endif | 35 | diff --git a/tcg/ppc/tcg-target.h b/tcg/ppc/tcg-target.h |
35 | + | 36 | index XXXXXXX..XXXXXXX 100644 |
36 | /* These will only be atomic if the processor does the fetch or store | 37 | --- a/tcg/ppc/tcg-target.h |
37 | * in a single issue memory operation | 38 | +++ b/tcg/ppc/tcg-target.h |
38 | */ | 39 | @@ -XXX,XX +XXX,XX @@ extern bool have_vsx; |
40 | #define TCG_TARGET_HAS_shi_vec 0 | ||
41 | #define TCG_TARGET_HAS_shs_vec 0 | ||
42 | #define TCG_TARGET_HAS_shv_vec 1 | ||
43 | -#define TCG_TARGET_HAS_cmp_vec 1 | ||
44 | #define TCG_TARGET_HAS_mul_vec 1 | ||
45 | #define TCG_TARGET_HAS_sat_vec 1 | ||
46 | #define TCG_TARGET_HAS_minmax_vec 1 | ||
39 | -- | 47 | -- |
40 | 2.17.1 | 48 | 2.25.1 |
41 | 49 | ||
42 | 50 | diff view generated by jsdifflib |
1 | At present we have a potential error in that helper_retaddr contains | 1 | From: Kele Huang <kele.hwang@gmail.com> |
---|---|---|---|
2 | data for handle_cpu_signal, but we have not ensured that those stores | ||
3 | will be scheduled properly before the operation that may fault. | ||
4 | 2 | ||
5 | It might be that these races are not in practice observable, due to | 3 | Detect all MIPS store instructions in cpu_signal_handler for all available |
6 | our use of -fno-strict-aliasing, but better safe than sorry. | 4 | MIPS versions, and set is_write if encountering such store instructions. |
7 | 5 | ||
8 | Adjust all of the setters of helper_retaddr. | 6 | This fixed the error while dealing with self-modified code for MIPS. |
9 | 7 | ||
10 | Reviewed-by: Alex Bennée <alex.bennee@linaro.org> | 8 | Reviewed-by: Richard Henderson <richard.henderson@linaro.org> |
9 | Signed-off-by: Kele Huang <kele.hwang@gmail.com> | ||
10 | Signed-off-by: Xu Zou <iwatchnima@gmail.com> | ||
11 | Message-Id: <20201002081420.10814-1-kele.hwang@gmail.com> | ||
12 | [rth: Use uintptr_t for pc to fix n32 build error.] | ||
11 | Signed-off-by: Richard Henderson <richard.henderson@linaro.org> | 13 | Signed-off-by: Richard Henderson <richard.henderson@linaro.org> |
12 | --- | 14 | --- |
13 | include/exec/cpu_ldst.h | 20 +++++++++++ | 15 | accel/tcg/user-exec.c | 43 +++++++++++++++++++++++++++++++++++++++---- |
14 | include/exec/cpu_ldst_useronly_template.h | 12 +++---- | 16 | 1 file changed, 39 insertions(+), 4 deletions(-) |
15 | accel/tcg/user-exec.c | 11 +++--- | ||
16 | target/arm/helper-a64.c | 8 ++--- | ||
17 | target/arm/sve_helper.c | 43 +++++++++++------------ | ||
18 | 5 files changed, 57 insertions(+), 37 deletions(-) | ||
19 | 17 | ||
20 | diff --git a/include/exec/cpu_ldst.h b/include/exec/cpu_ldst.h | ||
21 | index XXXXXXX..XXXXXXX 100644 | ||
22 | --- a/include/exec/cpu_ldst.h | ||
23 | +++ b/include/exec/cpu_ldst.h | ||
24 | @@ -XXX,XX +XXX,XX @@ typedef target_ulong abi_ptr; | ||
25 | |||
26 | extern __thread uintptr_t helper_retaddr; | ||
27 | |||
28 | +static inline void set_helper_retaddr(uintptr_t ra) | ||
29 | +{ | ||
30 | + helper_retaddr = ra; | ||
31 | + /* | ||
32 | + * Ensure that this write is visible to the SIGSEGV handler that | ||
33 | + * may be invoked due to a subsequent invalid memory operation. | ||
34 | + */ | ||
35 | + signal_barrier(); | ||
36 | +} | ||
37 | + | ||
38 | +static inline void clear_helper_retaddr(void) | ||
39 | +{ | ||
40 | + /* | ||
41 | + * Ensure that previous memory operations have succeeded before | ||
42 | + * removing the data visible to the signal handler. | ||
43 | + */ | ||
44 | + signal_barrier(); | ||
45 | + helper_retaddr = 0; | ||
46 | +} | ||
47 | + | ||
48 | /* In user-only mode we provide only the _code and _data accessors. */ | ||
49 | |||
50 | #define MEMSUFFIX _data | ||
51 | diff --git a/include/exec/cpu_ldst_useronly_template.h b/include/exec/cpu_ldst_useronly_template.h | ||
52 | index XXXXXXX..XXXXXXX 100644 | ||
53 | --- a/include/exec/cpu_ldst_useronly_template.h | ||
54 | +++ b/include/exec/cpu_ldst_useronly_template.h | ||
55 | @@ -XXX,XX +XXX,XX @@ glue(glue(glue(cpu_ld, USUFFIX), MEMSUFFIX), _ra)(CPUArchState *env, | ||
56 | uintptr_t retaddr) | ||
57 | { | ||
58 | RES_TYPE ret; | ||
59 | - helper_retaddr = retaddr; | ||
60 | + set_helper_retaddr(retaddr); | ||
61 | ret = glue(glue(cpu_ld, USUFFIX), MEMSUFFIX)(env, ptr); | ||
62 | - helper_retaddr = 0; | ||
63 | + clear_helper_retaddr(); | ||
64 | return ret; | ||
65 | } | ||
66 | |||
67 | @@ -XXX,XX +XXX,XX @@ glue(glue(glue(cpu_lds, SUFFIX), MEMSUFFIX), _ra)(CPUArchState *env, | ||
68 | uintptr_t retaddr) | ||
69 | { | ||
70 | int ret; | ||
71 | - helper_retaddr = retaddr; | ||
72 | + set_helper_retaddr(retaddr); | ||
73 | ret = glue(glue(cpu_lds, SUFFIX), MEMSUFFIX)(env, ptr); | ||
74 | - helper_retaddr = 0; | ||
75 | + clear_helper_retaddr(); | ||
76 | return ret; | ||
77 | } | ||
78 | #endif | ||
79 | @@ -XXX,XX +XXX,XX @@ glue(glue(glue(cpu_st, SUFFIX), MEMSUFFIX), _ra)(CPUArchState *env, | ||
80 | RES_TYPE v, | ||
81 | uintptr_t retaddr) | ||
82 | { | ||
83 | - helper_retaddr = retaddr; | ||
84 | + set_helper_retaddr(retaddr); | ||
85 | glue(glue(cpu_st, SUFFIX), MEMSUFFIX)(env, ptr, v); | ||
86 | - helper_retaddr = 0; | ||
87 | + clear_helper_retaddr(); | ||
88 | } | ||
89 | #endif | ||
90 | |||
91 | diff --git a/accel/tcg/user-exec.c b/accel/tcg/user-exec.c | 18 | diff --git a/accel/tcg/user-exec.c b/accel/tcg/user-exec.c |
92 | index XXXXXXX..XXXXXXX 100644 | 19 | index XXXXXXX..XXXXXXX 100644 |
93 | --- a/accel/tcg/user-exec.c | 20 | --- a/accel/tcg/user-exec.c |
94 | +++ b/accel/tcg/user-exec.c | 21 | +++ b/accel/tcg/user-exec.c |
95 | @@ -XXX,XX +XXX,XX @@ static inline int handle_cpu_signal(uintptr_t pc, siginfo_t *info, | 22 | @@ -XXX,XX +XXX,XX @@ int cpu_signal_handler(int host_signum, void *pinfo, |
96 | * currently executing TB was modified and must be exited | 23 | |
97 | * immediately. Clear helper_retaddr for next execution. | 24 | #elif defined(__mips__) |
98 | */ | 25 | |
99 | - helper_retaddr = 0; | 26 | +#if defined(__misp16) || defined(__mips_micromips) |
100 | + clear_helper_retaddr(); | 27 | +#error "Unsupported encoding" |
101 | cpu_exit_tb_from_sighandler(cpu, old_set); | 28 | +#endif |
102 | /* NORETURN */ | 29 | + |
103 | 30 | int cpu_signal_handler(int host_signum, void *pinfo, | |
104 | @@ -XXX,XX +XXX,XX @@ static inline int handle_cpu_signal(uintptr_t pc, siginfo_t *info, | 31 | void *puc) |
105 | * an exception. Undo signal and retaddr state prior to longjmp. | 32 | { |
106 | */ | 33 | siginfo_t *info = pinfo; |
107 | sigprocmask(SIG_SETMASK, old_set, NULL); | 34 | ucontext_t *uc = puc; |
108 | - helper_retaddr = 0; | 35 | - greg_t pc = uc->uc_mcontext.pc; |
109 | + clear_helper_retaddr(); | 36 | - int is_write; |
110 | 37 | + uintptr_t pc = uc->uc_mcontext.pc; | |
111 | cc = CPU_GET_CLASS(cpu); | 38 | + uint32_t insn = *(uint32_t *)pc; |
112 | access_type = is_write ? MMU_DATA_STORE : MMU_DATA_LOAD; | 39 | + int is_write = 0; |
113 | @@ -XXX,XX +XXX,XX @@ static void *atomic_mmu_lookup(CPUArchState *env, target_ulong addr, | 40 | + |
114 | if (unlikely(addr & (size - 1))) { | 41 | + /* Detect all store instructions at program counter. */ |
115 | cpu_loop_exit_atomic(env_cpu(env), retaddr); | 42 | + switch((insn >> 26) & 077) { |
116 | } | 43 | + case 050: /* SB */ |
117 | - helper_retaddr = retaddr; | 44 | + case 051: /* SH */ |
118 | - return g2h(addr); | 45 | + case 052: /* SWL */ |
119 | + void *ret = g2h(addr); | 46 | + case 053: /* SW */ |
120 | + set_helper_retaddr(retaddr); | 47 | + case 054: /* SDL */ |
121 | + return ret; | 48 | + case 055: /* SDR */ |
49 | + case 056: /* SWR */ | ||
50 | + case 070: /* SC */ | ||
51 | + case 071: /* SWC1 */ | ||
52 | + case 074: /* SCD */ | ||
53 | + case 075: /* SDC1 */ | ||
54 | + case 077: /* SD */ | ||
55 | +#if !defined(__mips_isa_rev) || __mips_isa_rev < 6 | ||
56 | + case 072: /* SWC2 */ | ||
57 | + case 076: /* SDC2 */ | ||
58 | +#endif | ||
59 | + is_write = 1; | ||
60 | + break; | ||
61 | + case 023: /* COP1X */ | ||
62 | + /* Required in all versions of MIPS64 since | ||
63 | + MIPS64r1 and subsequent versions of MIPS32r2. */ | ||
64 | + switch (insn & 077) { | ||
65 | + case 010: /* SWXC1 */ | ||
66 | + case 011: /* SDXC1 */ | ||
67 | + case 015: /* SUXC1 */ | ||
68 | + is_write = 1; | ||
69 | + } | ||
70 | + break; | ||
71 | + } | ||
72 | |||
73 | - /* XXX: compute is_write */ | ||
74 | - is_write = 0; | ||
75 | return handle_cpu_signal(pc, info, is_write, &uc->uc_sigmask); | ||
122 | } | 76 | } |
123 | 77 | ||
124 | /* Macro to call the above, with local variables from the use context. */ | ||
125 | #define ATOMIC_MMU_DECLS do {} while (0) | ||
126 | #define ATOMIC_MMU_LOOKUP atomic_mmu_lookup(env, addr, DATA_SIZE, GETPC()) | ||
127 | -#define ATOMIC_MMU_CLEANUP do { helper_retaddr = 0; } while (0) | ||
128 | +#define ATOMIC_MMU_CLEANUP do { clear_helper_retaddr(); } while (0) | ||
129 | |||
130 | #define ATOMIC_NAME(X) HELPER(glue(glue(atomic_ ## X, SUFFIX), END)) | ||
131 | #define EXTRA_ARGS | ||
132 | diff --git a/target/arm/helper-a64.c b/target/arm/helper-a64.c | ||
133 | index XXXXXXX..XXXXXXX 100644 | ||
134 | --- a/target/arm/helper-a64.c | ||
135 | +++ b/target/arm/helper-a64.c | ||
136 | @@ -XXX,XX +XXX,XX @@ uint64_t HELPER(paired_cmpxchg64_le)(CPUARMState *env, uint64_t addr, | ||
137 | /* ??? Enforce alignment. */ | ||
138 | uint64_t *haddr = g2h(addr); | ||
139 | |||
140 | - helper_retaddr = ra; | ||
141 | + set_helper_retaddr(ra); | ||
142 | o0 = ldq_le_p(haddr + 0); | ||
143 | o1 = ldq_le_p(haddr + 1); | ||
144 | oldv = int128_make128(o0, o1); | ||
145 | @@ -XXX,XX +XXX,XX @@ uint64_t HELPER(paired_cmpxchg64_le)(CPUARMState *env, uint64_t addr, | ||
146 | stq_le_p(haddr + 0, int128_getlo(newv)); | ||
147 | stq_le_p(haddr + 1, int128_gethi(newv)); | ||
148 | } | ||
149 | - helper_retaddr = 0; | ||
150 | + clear_helper_retaddr(); | ||
151 | #else | ||
152 | int mem_idx = cpu_mmu_index(env, false); | ||
153 | TCGMemOpIdx oi0 = make_memop_idx(MO_LEQ | MO_ALIGN_16, mem_idx); | ||
154 | @@ -XXX,XX +XXX,XX @@ uint64_t HELPER(paired_cmpxchg64_be)(CPUARMState *env, uint64_t addr, | ||
155 | /* ??? Enforce alignment. */ | ||
156 | uint64_t *haddr = g2h(addr); | ||
157 | |||
158 | - helper_retaddr = ra; | ||
159 | + set_helper_retaddr(ra); | ||
160 | o1 = ldq_be_p(haddr + 0); | ||
161 | o0 = ldq_be_p(haddr + 1); | ||
162 | oldv = int128_make128(o0, o1); | ||
163 | @@ -XXX,XX +XXX,XX @@ uint64_t HELPER(paired_cmpxchg64_be)(CPUARMState *env, uint64_t addr, | ||
164 | stq_be_p(haddr + 0, int128_gethi(newv)); | ||
165 | stq_be_p(haddr + 1, int128_getlo(newv)); | ||
166 | } | ||
167 | - helper_retaddr = 0; | ||
168 | + clear_helper_retaddr(); | ||
169 | #else | ||
170 | int mem_idx = cpu_mmu_index(env, false); | ||
171 | TCGMemOpIdx oi0 = make_memop_idx(MO_BEQ | MO_ALIGN_16, mem_idx); | ||
172 | diff --git a/target/arm/sve_helper.c b/target/arm/sve_helper.c | ||
173 | index XXXXXXX..XXXXXXX 100644 | ||
174 | --- a/target/arm/sve_helper.c | ||
175 | +++ b/target/arm/sve_helper.c | ||
176 | @@ -XXX,XX +XXX,XX @@ static intptr_t max_for_page(target_ulong base, intptr_t mem_off, | ||
177 | return MIN(split, mem_max - mem_off) + mem_off; | ||
178 | } | ||
179 | |||
180 | -static inline void set_helper_retaddr(uintptr_t ra) | ||
181 | -{ | ||
182 | -#ifdef CONFIG_USER_ONLY | ||
183 | - helper_retaddr = ra; | ||
184 | +#ifndef CONFIG_USER_ONLY | ||
185 | +/* These are normally defined only for CONFIG_USER_ONLY in <exec/cpu_ldst.h> */ | ||
186 | +static inline void set_helper_retaddr(uintptr_t ra) { } | ||
187 | +static inline void clear_helper_retaddr(void) { } | ||
188 | #endif | ||
189 | -} | ||
190 | |||
191 | /* | ||
192 | * The result of tlb_vaddr_to_host for user-only is just g2h(x), | ||
193 | @@ -XXX,XX +XXX,XX @@ static void sve_ld1_r(CPUARMState *env, void *vg, const target_ulong addr, | ||
194 | if (test_host_page(host)) { | ||
195 | mem_off = host_fn(vd, vg, host - mem_off, mem_off, mem_max); | ||
196 | tcg_debug_assert(mem_off == mem_max); | ||
197 | - set_helper_retaddr(0); | ||
198 | + clear_helper_retaddr(); | ||
199 | /* After having taken any fault, zero leading inactive elements. */ | ||
200 | swap_memzero(vd, reg_off); | ||
201 | return; | ||
202 | @@ -XXX,XX +XXX,XX @@ static void sve_ld1_r(CPUARMState *env, void *vg, const target_ulong addr, | ||
203 | } | ||
204 | #endif | ||
205 | |||
206 | - set_helper_retaddr(0); | ||
207 | + clear_helper_retaddr(); | ||
208 | memcpy(vd, &scratch, reg_max); | ||
209 | } | ||
210 | |||
211 | @@ -XXX,XX +XXX,XX @@ static void sve_ld2_r(CPUARMState *env, void *vg, target_ulong addr, | ||
212 | addr += 2 * size; | ||
213 | } while (i & 15); | ||
214 | } | ||
215 | - set_helper_retaddr(0); | ||
216 | + clear_helper_retaddr(); | ||
217 | |||
218 | /* Wait until all exceptions have been raised to write back. */ | ||
219 | memcpy(&env->vfp.zregs[rd], &scratch[0], oprsz); | ||
220 | @@ -XXX,XX +XXX,XX @@ static void sve_ld3_r(CPUARMState *env, void *vg, target_ulong addr, | ||
221 | addr += 3 * size; | ||
222 | } while (i & 15); | ||
223 | } | ||
224 | - set_helper_retaddr(0); | ||
225 | + clear_helper_retaddr(); | ||
226 | |||
227 | /* Wait until all exceptions have been raised to write back. */ | ||
228 | memcpy(&env->vfp.zregs[rd], &scratch[0], oprsz); | ||
229 | @@ -XXX,XX +XXX,XX @@ static void sve_ld4_r(CPUARMState *env, void *vg, target_ulong addr, | ||
230 | addr += 4 * size; | ||
231 | } while (i & 15); | ||
232 | } | ||
233 | - set_helper_retaddr(0); | ||
234 | + clear_helper_retaddr(); | ||
235 | |||
236 | /* Wait until all exceptions have been raised to write back. */ | ||
237 | memcpy(&env->vfp.zregs[rd], &scratch[0], oprsz); | ||
238 | @@ -XXX,XX +XXX,XX @@ static void sve_ldff1_r(CPUARMState *env, void *vg, const target_ulong addr, | ||
239 | if (test_host_page(host)) { | ||
240 | mem_off = host_fn(vd, vg, host - mem_off, mem_off, mem_max); | ||
241 | tcg_debug_assert(mem_off == mem_max); | ||
242 | - set_helper_retaddr(0); | ||
243 | + clear_helper_retaddr(); | ||
244 | /* After any fault, zero any leading inactive elements. */ | ||
245 | swap_memzero(vd, reg_off); | ||
246 | return; | ||
247 | @@ -XXX,XX +XXX,XX @@ static void sve_ldff1_r(CPUARMState *env, void *vg, const target_ulong addr, | ||
248 | } | ||
249 | #endif | ||
250 | |||
251 | - set_helper_retaddr(0); | ||
252 | + clear_helper_retaddr(); | ||
253 | record_fault(env, reg_off, reg_max); | ||
254 | } | ||
255 | |||
256 | @@ -XXX,XX +XXX,XX @@ static void sve_st1_r(CPUARMState *env, void *vg, target_ulong addr, | ||
257 | addr += msize; | ||
258 | } while (i & 15); | ||
259 | } | ||
260 | - set_helper_retaddr(0); | ||
261 | + clear_helper_retaddr(); | ||
262 | } | ||
263 | |||
264 | static void sve_st2_r(CPUARMState *env, void *vg, target_ulong addr, | ||
265 | @@ -XXX,XX +XXX,XX @@ static void sve_st2_r(CPUARMState *env, void *vg, target_ulong addr, | ||
266 | addr += 2 * msize; | ||
267 | } while (i & 15); | ||
268 | } | ||
269 | - set_helper_retaddr(0); | ||
270 | + clear_helper_retaddr(); | ||
271 | } | ||
272 | |||
273 | static void sve_st3_r(CPUARMState *env, void *vg, target_ulong addr, | ||
274 | @@ -XXX,XX +XXX,XX @@ static void sve_st3_r(CPUARMState *env, void *vg, target_ulong addr, | ||
275 | addr += 3 * msize; | ||
276 | } while (i & 15); | ||
277 | } | ||
278 | - set_helper_retaddr(0); | ||
279 | + clear_helper_retaddr(); | ||
280 | } | ||
281 | |||
282 | static void sve_st4_r(CPUARMState *env, void *vg, target_ulong addr, | ||
283 | @@ -XXX,XX +XXX,XX @@ static void sve_st4_r(CPUARMState *env, void *vg, target_ulong addr, | ||
284 | addr += 4 * msize; | ||
285 | } while (i & 15); | ||
286 | } | ||
287 | - set_helper_retaddr(0); | ||
288 | + clear_helper_retaddr(); | ||
289 | } | ||
290 | |||
291 | #define DO_STN_1(N, NAME, ESIZE) \ | ||
292 | @@ -XXX,XX +XXX,XX @@ static void sve_ld1_zs(CPUARMState *env, void *vd, void *vg, void *vm, | ||
293 | i += 4, pg >>= 4; | ||
294 | } while (i & 15); | ||
295 | } | ||
296 | - set_helper_retaddr(0); | ||
297 | + clear_helper_retaddr(); | ||
298 | |||
299 | /* Wait until all exceptions have been raised to write back. */ | ||
300 | memcpy(vd, &scratch, oprsz); | ||
301 | @@ -XXX,XX +XXX,XX @@ static void sve_ld1_zd(CPUARMState *env, void *vd, void *vg, void *vm, | ||
302 | tlb_fn(env, &scratch, i * 8, base + (off << scale), oi, ra); | ||
303 | } | ||
304 | } | ||
305 | - set_helper_retaddr(0); | ||
306 | + clear_helper_retaddr(); | ||
307 | |||
308 | /* Wait until all exceptions have been raised to write back. */ | ||
309 | memcpy(vd, &scratch, oprsz * 8); | ||
310 | @@ -XXX,XX +XXX,XX @@ static inline void sve_ldff1_zs(CPUARMState *env, void *vd, void *vg, void *vm, | ||
311 | tlb_fn(env, vd, reg_off, addr, oi, ra); | ||
312 | |||
313 | /* The rest of the reads will be non-faulting. */ | ||
314 | - set_helper_retaddr(0); | ||
315 | + clear_helper_retaddr(); | ||
316 | } | ||
317 | |||
318 | /* After any fault, zero the leading predicated false elements. */ | ||
319 | @@ -XXX,XX +XXX,XX @@ static inline void sve_ldff1_zd(CPUARMState *env, void *vd, void *vg, void *vm, | ||
320 | tlb_fn(env, vd, reg_off, addr, oi, ra); | ||
321 | |||
322 | /* The rest of the reads will be non-faulting. */ | ||
323 | - set_helper_retaddr(0); | ||
324 | + clear_helper_retaddr(); | ||
325 | } | ||
326 | |||
327 | /* After any fault, zero the leading predicated false elements. */ | ||
328 | @@ -XXX,XX +XXX,XX @@ static void sve_st1_zs(CPUARMState *env, void *vd, void *vg, void *vm, | ||
329 | i += 4, pg >>= 4; | ||
330 | } while (i & 15); | ||
331 | } | ||
332 | - set_helper_retaddr(0); | ||
333 | + clear_helper_retaddr(); | ||
334 | } | ||
335 | |||
336 | static void sve_st1_zd(CPUARMState *env, void *vd, void *vg, void *vm, | ||
337 | @@ -XXX,XX +XXX,XX @@ static void sve_st1_zd(CPUARMState *env, void *vd, void *vg, void *vm, | ||
338 | tlb_fn(env, vd, i * 8, base + (off << scale), oi, ra); | ||
339 | } | ||
340 | } | ||
341 | - set_helper_retaddr(0); | ||
342 | + clear_helper_retaddr(); | ||
343 | } | ||
344 | |||
345 | #define DO_ST1_ZPZ_S(MEM, OFS) \ | ||
346 | -- | 78 | -- |
347 | 2.17.1 | 79 | 2.25.1 |
348 | 80 | ||
349 | 81 | diff view generated by jsdifflib |