mirror of https://gitee.com/openkylin/qemu.git
tcg: Reduce max TB opcode count
Also, assert that we don't overflow any of two different offsets into the TB. Both unwind and goto_tb both record a uint16_t for later use. This fixes an arm-softmmu test case utilizing NEON in which there is a TB generated that runs to 7800 opcodes, and compiles to 96k on an x86_64 host. This overflows the 16-bit offset in which we record the goto_tb reset offset. Because of that overflow, we install a jump destination that goes to neverland. Boom. With this reduced op count, the same TB compiles to about 48k for aarch64, ppc64le, and x86_64 hosts, and neither assertion fires. Cc: qemu-stable@nongnu.org Reported-by: "Jason A. Donenfeld" <Jason@zx2c4.com> Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
This commit is contained in:
parent
0ac20318ce
commit
9f75462065
|
@ -1733,7 +1733,7 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,
|
|||
tcg_out_insn(s, 3305, LDR, offset, TCG_REG_TMP);
|
||||
}
|
||||
tcg_out_insn(s, 3207, BR, TCG_REG_TMP);
|
||||
s->tb_jmp_reset_offset[a0] = tcg_current_code_size(s);
|
||||
set_jmp_reset_offset(s, a0);
|
||||
break;
|
||||
|
||||
case INDEX_op_goto_ptr:
|
||||
|
|
|
@ -1822,7 +1822,7 @@ static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
|
|||
tcg_out_movi32(s, COND_AL, base, ptr - dil);
|
||||
}
|
||||
tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, base, dil);
|
||||
s->tb_jmp_reset_offset[args[0]] = tcg_current_code_size(s);
|
||||
set_jmp_reset_offset(s, args[0]);
|
||||
}
|
||||
break;
|
||||
case INDEX_op_goto_ptr:
|
||||
|
|
|
@ -2245,7 +2245,7 @@ static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
|
|||
tcg_out_modrm_offset(s, OPC_GRP5, EXT5_JMPN_Ev, -1,
|
||||
(intptr_t)(s->tb_jmp_target_addr + a0));
|
||||
}
|
||||
s->tb_jmp_reset_offset[a0] = tcg_current_code_size(s);
|
||||
set_jmp_reset_offset(s, a0);
|
||||
break;
|
||||
case INDEX_op_goto_ptr:
|
||||
/* jmp to the given host address (could be epilogue) */
|
||||
|
|
|
@ -1744,7 +1744,7 @@ static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
|
|||
tcg_out_opc_reg(s, OPC_JR, 0, TCG_TMP0, 0);
|
||||
}
|
||||
tcg_out_nop(s);
|
||||
s->tb_jmp_reset_offset[a0] = tcg_current_code_size(s);
|
||||
set_jmp_reset_offset(s, a0);
|
||||
break;
|
||||
case INDEX_op_goto_ptr:
|
||||
/* jmp to the given host address (could be epilogue) */
|
||||
|
|
|
@ -2025,10 +2025,10 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc, const TCGArg *args,
|
|||
}
|
||||
tcg_out32(s, MTSPR | RS(TCG_REG_TB) | CTR);
|
||||
tcg_out32(s, BCCTR | BO_ALWAYS);
|
||||
s->tb_jmp_reset_offset[args[0]] = c = tcg_current_code_size(s);
|
||||
set_jmp_reset_offset(s, args[0]);
|
||||
if (USE_REG_TB) {
|
||||
/* For the unlinked case, need to reset TCG_REG_TB. */
|
||||
c = -c;
|
||||
c = -tcg_current_code_size(s);
|
||||
assert(c == (int16_t)c);
|
||||
tcg_out32(s, ADDI | TAI(TCG_REG_TB, TCG_REG_TB, c));
|
||||
}
|
||||
|
|
|
@ -1783,7 +1783,7 @@ static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
|
|||
/* and go there */
|
||||
tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, TCG_REG_TB);
|
||||
}
|
||||
s->tb_jmp_reset_offset[a0] = tcg_current_code_size(s);
|
||||
set_jmp_reset_offset(s, a0);
|
||||
|
||||
/* For the unlinked path of goto_tb, we need to reset
|
||||
TCG_REG_TB to the beginning of this TB. */
|
||||
|
|
|
@ -1388,12 +1388,12 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,
|
|||
tcg_out_arithi(s, TCG_REG_G0, TCG_REG_TB, 0, JMPL);
|
||||
tcg_out_nop(s);
|
||||
}
|
||||
s->tb_jmp_reset_offset[a0] = c = tcg_current_code_size(s);
|
||||
set_jmp_reset_offset(s, a0);
|
||||
|
||||
/* For the unlinked path of goto_tb, we need to reset
|
||||
TCG_REG_TB to the beginning of this TB. */
|
||||
if (USE_REG_TB) {
|
||||
c = -c;
|
||||
c = -tcg_current_code_size(s);
|
||||
if (check_fit_i32(c, 13)) {
|
||||
tcg_out_arithi(s, TCG_REG_TB, TCG_REG_TB, c, ARITH_ADD);
|
||||
} else {
|
||||
|
|
13
tcg/tcg.c
13
tcg/tcg.c
|
@ -306,6 +306,14 @@ TCGLabel *gen_new_label(void)
|
|||
return l;
|
||||
}
|
||||
|
||||
static void set_jmp_reset_offset(TCGContext *s, int which)
|
||||
{
|
||||
size_t off = tcg_current_code_size(s);
|
||||
s->tb_jmp_reset_offset[which] = off;
|
||||
/* Make sure that we didn't overflow the stored offset. */
|
||||
assert(s->tb_jmp_reset_offset[which] == off);
|
||||
}
|
||||
|
||||
#include "tcg-target.inc.c"
|
||||
|
||||
/* compare a pointer @ptr and a tb_tc @s */
|
||||
|
@ -3532,7 +3540,10 @@ int tcg_gen_code(TCGContext *s, TranslationBlock *tb)
|
|||
break;
|
||||
case INDEX_op_insn_start:
|
||||
if (num_insns >= 0) {
|
||||
s->gen_insn_end_off[num_insns] = tcg_current_code_size(s);
|
||||
size_t off = tcg_current_code_size(s);
|
||||
s->gen_insn_end_off[num_insns] = off;
|
||||
/* Assert that we do not overflow our stored offset. */
|
||||
assert(s->gen_insn_end_off[num_insns] == off);
|
||||
}
|
||||
num_insns++;
|
||||
for (i = 0; i < TARGET_INSN_START_WORDS; ++i) {
|
||||
|
|
|
@ -850,9 +850,11 @@ static inline bool tcg_op_buf_full(void)
|
|||
/* This is not a hard limit, it merely stops translation when
|
||||
* we have produced "enough" opcodes. We want to limit TB size
|
||||
* such that a RISC host can reasonably use a 16-bit signed
|
||||
* branch within the TB.
|
||||
* branch within the TB. We also need to be mindful of the
|
||||
* 16-bit unsigned offsets, TranslationBlock.jmp_reset_offset[]
|
||||
* and TCGContext.gen_insn_end_off[].
|
||||
*/
|
||||
return tcg_ctx->nb_ops >= 8000;
|
||||
return tcg_ctx->nb_ops >= 4000;
|
||||
}
|
||||
|
||||
/* pool based memory allocation */
|
||||
|
|
|
@ -574,7 +574,7 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc, const TCGArg *args,
|
|||
/* Indirect jump method. */
|
||||
TODO();
|
||||
}
|
||||
s->tb_jmp_reset_offset[args[0]] = tcg_current_code_size(s);
|
||||
set_jmp_reset_offset(s, args[0]);
|
||||
break;
|
||||
case INDEX_op_br:
|
||||
tci_out_label(s, arg_label(args[0]));
|
||||
|
|
Loading…
Reference in New Issue