Merge remote-tracking branch 'remotes/rth/tags/pull-tcg-20170619' into staging

Queued TCG patches

# gpg: Signature made Mon 19 Jun 2017 19:12:06 BST
# gpg:                using RSA key 0xAD1270CC4DD0279B
# gpg: Good signature from "Richard Henderson <rth7680@gmail.com>"
# gpg:                 aka "Richard Henderson <rth@redhat.com>"
# gpg:                 aka "Richard Henderson <rth@twiddle.net>"
# Primary key fingerprint: 9CB1 8DDA F8E8 49AD 2AFC  16A4 AD12 70CC 4DD0 279B

* remotes/rth/tags/pull-tcg-20170619:
  target/arm: Exit after clearing aarch64 interrupt mask
  target/s390x: Exit after changing PSW mask
  target/alpha: Use tcg_gen_lookup_and_goto_ptr
  tcg: Increase hit rate of lookup_tb_ptr
  tcg/arm: Use ldr (literal) for goto_tb
  tcg/arm: Try pc-relative addresses for movi
  tcg/arm: Remove limit on code buffer size
  tcg/arm: Use indirect branch for goto_tb
  tcg/aarch64: Use ADR in tcg_out_movi
  translate-all: consolidate tb init in tb_gen_code
  tcg: allocate TB structs before the corresponding translated code
  util: add cacheinfo

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

15 files changed, 345 insertions, 152 deletions

diff --git a/accel/tcg/translate-all.c b/accel/tcg/translate-all.c
index 7b25a16244..f6ad46b613 100644
--- a/accel/tcg/translate-all.c
+++ b/accel/tcg/translate-all.c
@@ -523,8 +523,6 @@ static inline PageDesc *page_find(tb_page_addr_t index)
 # define MAX_CODE_GEN_BUFFER_SIZE  (32u * 1024 * 1024)
 #elif defined(__aarch64__)
 # define MAX_CODE_GEN_BUFFER_SIZE  (128ul * 1024 * 1024)
-#elif defined(__arm__)
-# define MAX_CODE_GEN_BUFFER_SIZE  (16u * 1024 * 1024)
 #elif defined(__s390x__)
   /* We have a +- 4GB range on the branches; leave some slop.  */
 # define MAX_CODE_GEN_BUFFER_SIZE  (3ul * 1024 * 1024 * 1024)
@@ -781,12 +779,13 @@ static inline void code_gen_alloc(size_t tb_size)
         exit(1);
     }
 
-    /* Estimate a good size for the number of TBs we can support.  We
-       still haven't deducted the prologue from the buffer size here,
-       but that's minimal and won't affect the estimate much.  */
-    tcg_ctx.code_gen_max_blocks
-        = tcg_ctx.code_gen_buffer_size / CODE_GEN_AVG_BLOCK_SIZE;
-    tcg_ctx.tb_ctx.tbs = g_new(TranslationBlock, tcg_ctx.code_gen_max_blocks);
+    /* size this conservatively -- realloc later if needed */
+    tcg_ctx.tb_ctx.tbs_size =
+        tcg_ctx.code_gen_buffer_size / CODE_GEN_AVG_BLOCK_SIZE / 8;
+    if (unlikely(!tcg_ctx.tb_ctx.tbs_size)) {
+        tcg_ctx.tb_ctx.tbs_size = 64 * 1024;
+    }
+    tcg_ctx.tb_ctx.tbs = g_new(TranslationBlock *, tcg_ctx.tb_ctx.tbs_size);
 
     qemu_mutex_init(&tcg_ctx.tb_ctx.tb_lock);
 }
@@ -828,16 +827,20 @@ bool tcg_enabled(void)
 static TranslationBlock *tb_alloc(target_ulong pc)
 {
     TranslationBlock *tb;
+    TBContext *ctx;
 
     assert_tb_locked();
 
-    if (tcg_ctx.tb_ctx.nb_tbs >= tcg_ctx.code_gen_max_blocks) {
+    tb = tcg_tb_alloc(&tcg_ctx);
+    if (unlikely(tb == NULL)) {
         return NULL;
     }
-    tb = &tcg_ctx.tb_ctx.tbs[tcg_ctx.tb_ctx.nb_tbs++];
-    tb->pc = pc;
-    tb->cflags = 0;
-    tb->invalid = false;
+    ctx = &tcg_ctx.tb_ctx;
+    if (unlikely(ctx->nb_tbs == ctx->tbs_size)) {
+        ctx->tbs_size *= 2;
+        ctx->tbs = g_renew(TranslationBlock *, ctx->tbs, ctx->tbs_size);
+    }
+    ctx->tbs[ctx->nb_tbs++] = tb;
     return tb;
 }
 
@@ -850,8 +853,10 @@ void tb_free(TranslationBlock *tb)
        Ignore the hard cases and just back up if this TB happens to
        be the last one generated.  */
     if (tcg_ctx.tb_ctx.nb_tbs > 0 &&
-            tb == &tcg_ctx.tb_ctx.tbs[tcg_ctx.tb_ctx.nb_tbs - 1]) {
-        tcg_ctx.code_gen_ptr = tb->tc_ptr;
+            tb == tcg_ctx.tb_ctx.tbs[tcg_ctx.tb_ctx.nb_tbs - 1]) {
+        size_t struct_size = ROUND_UP(sizeof(*tb), qemu_icache_linesize);
+
+        tcg_ctx.code_gen_ptr = tb->tc_ptr - struct_size;
         tcg_ctx.tb_ctx.nb_tbs--;
     }
 }
@@ -1279,9 +1284,11 @@ TranslationBlock *tb_gen_code(CPUState *cpu,
 
     gen_code_buf = tcg_ctx.code_gen_ptr;
     tb->tc_ptr = gen_code_buf;
+    tb->pc = pc;
     tb->cs_base = cs_base;
     tb->flags = flags;
     tb->cflags = cflags;
+    tb->invalid = false;
 
 #ifdef CONFIG_PROFILER
     tcg_ctx.tb_count1++; /* includes aborted translations because of
@@ -1666,7 +1673,7 @@ static TranslationBlock *tb_find_pc(uintptr_t tc_ptr)
     m_max = tcg_ctx.tb_ctx.nb_tbs - 1;
     while (m_min <= m_max) {
         m = (m_min + m_max) >> 1;
-        tb = &tcg_ctx.tb_ctx.tbs[m];
+        tb = tcg_ctx.tb_ctx.tbs[m];
         v = (uintptr_t)tb->tc_ptr;
         if (v == tc_ptr) {
             return tb;
@@ -1676,7 +1683,7 @@ static TranslationBlock *tb_find_pc(uintptr_t tc_ptr)
             m_min = m + 1;
         }
     }
-    return &tcg_ctx.tb_ctx.tbs[m_max];
+    return tcg_ctx.tb_ctx.tbs[m_max];
 }
 
 #if !defined(CONFIG_USER_ONLY)
@@ -1874,7 +1881,7 @@ void dump_exec_info(FILE *f, fprintf_function cpu_fprintf)
     direct_jmp_count = 0;
     direct_jmp2_count = 0;
     for (i = 0; i < tcg_ctx.tb_ctx.nb_tbs; i++) {
-        tb = &tcg_ctx.tb_ctx.tbs[i];
+        tb = tcg_ctx.tb_ctx.tbs[i];
         target_code_size += tb->size;
         if (tb->size > max_target_code_size) {
             max_target_code_size = tb->size;
@@ -1894,8 +1901,7 @@ void dump_exec_info(FILE *f, fprintf_function cpu_fprintf)
     cpu_fprintf(f, "gen code size       %td/%zd\n",
                 tcg_ctx.code_gen_ptr - tcg_ctx.code_gen_buffer,
                 tcg_ctx.code_gen_highwater - tcg_ctx.code_gen_buffer);
-    cpu_fprintf(f, "TB count            %d/%d\n",
-            tcg_ctx.tb_ctx.nb_tbs, tcg_ctx.code_gen_max_blocks);
+    cpu_fprintf(f, "TB count            %d\n", tcg_ctx.tb_ctx.nb_tbs);
     cpu_fprintf(f, "TB avg target size  %d max=%d bytes\n",
             tcg_ctx.tb_ctx.nb_tbs ? target_code_size /
                     tcg_ctx.tb_ctx.nb_tbs : 0,
diff --git a/include/exec/exec-all.h b/include/exec/exec-all.h
index 87ae10bcc9..724ec73dce 100644
--- a/include/exec/exec-all.h
+++ b/include/exec/exec-all.h
@@ -301,7 +301,7 @@ static inline void tlb_flush_by_mmuidx_all_cpus_synced(CPUState *cpu,
 #define CODE_GEN_AVG_BLOCK_SIZE 150
 #endif
 
-#if defined(__arm__) || defined(_ARCH_PPC) \
+#if defined(_ARCH_PPC) \
     || defined(__x86_64__) || defined(__i386__) \
     || defined(__sparc__) || defined(__aarch64__) \
     || defined(__s390x__) || defined(__mips__) \
@@ -401,9 +401,6 @@ static inline void tb_set_jmp_target1(uintptr_t jmp_addr, uintptr_t addr)
 #elif defined(__aarch64__)
 void aarch64_tb_set_jmp_target(uintptr_t jmp_addr, uintptr_t addr);
 #define tb_set_jmp_target1 aarch64_tb_set_jmp_target
-#elif defined(__arm__)
-void arm_tb_set_jmp_target(uintptr_t jmp_addr, uintptr_t addr);
-#define tb_set_jmp_target1 arm_tb_set_jmp_target
 #elif defined(__sparc__) || defined(__mips__)
 void tb_set_jmp_target1(uintptr_t jmp_addr, uintptr_t addr);
 #else
diff --git a/include/exec/tb-context.h b/include/exec/tb-context.h
index c7f17f26e0..25c2afe753 100644
--- a/include/exec/tb-context.h
+++ b/include/exec/tb-context.h
@@ -31,8 +31,9 @@ typedef struct TBContext TBContext;
 
 struct TBContext {
 
-    TranslationBlock *tbs;
+    TranslationBlock **tbs;
     struct qht htable;
+    size_t tbs_size;
     int nb_tbs;
     /* any access to the tbs or the page table must use this lock */
     QemuMutex tb_lock;
diff --git a/include/qemu/osdep.h b/include/qemu/osdep.h
index fb008a2e65..85596341fa 100644
--- a/include/qemu/osdep.h
+++ b/include/qemu/osdep.h
@@ -483,4 +483,7 @@ char *qemu_get_pid_name(pid_t pid);
  */
 pid_t qemu_fork(Error **errp);
 
+extern int qemu_icache_linesize;
+extern int qemu_dcache_linesize;
+
 #endif
diff --git a/target/alpha/translate.c b/target/alpha/translate.c
index 7c45ae360c..232af9e177 100644
--- a/target/alpha/translate.c
+++ b/target/alpha/translate.c
@@ -84,6 +84,7 @@ typedef enum {
        the PC (for whatever reason), so there's no need to do it again on
        exiting the TB.  */
     EXIT_PC_UPDATED,
+    EXIT_PC_UPDATED_NOCHAIN,
 
     /* We are exiting the TB, but have neither emitted a goto_tb, nor
        updated the PC for the next instruction to be executed.  */
@@ -458,11 +459,17 @@ static bool in_superpage(DisasContext *ctx, int64_t addr)
 #endif
 }
 
+static bool use_exit_tb(DisasContext *ctx)
+{
+    return ((ctx->tb->cflags & CF_LAST_IO)
+            || ctx->singlestep_enabled
+            || singlestep);
+}
+
 static bool use_goto_tb(DisasContext *ctx, uint64_t dest)
 {
     /* Suppress goto_tb in the case of single-steping and IO.  */
-    if ((ctx->tb->cflags & CF_LAST_IO)
-        || ctx->singlestep_enabled || singlestep) {
+    if (unlikely(use_exit_tb(ctx))) {
         return false;
     }
 #ifndef CONFIG_USER_ONLY
@@ -1198,7 +1205,10 @@ static ExitStatus gen_call_pal(DisasContext *ctx, int palcode)
             tcg_gen_andi_i64(tmp, ctx->ir[IR_A0], PS_INT_MASK);
             tcg_gen_st8_i64(tmp, cpu_env, offsetof(CPUAlphaState, ps));
             tcg_temp_free(tmp);
-            break;
+
+            /* Allow interrupts to be recognized right away.  */
+            tcg_gen_movi_i64(cpu_pc, ctx->pc);
+            return EXIT_PC_UPDATED_NOCHAIN;
 
         case 0x36:
             /* RDPS */
@@ -1266,7 +1276,7 @@ static ExitStatus gen_call_pal(DisasContext *ctx, int palcode)
            need the page permissions check.  We'll see the existence of
            the page when we create the TB, and we'll flush all TBs if
            we change the PAL base register.  */
-        if (!ctx->singlestep_enabled && !(ctx->tb->cflags & CF_LAST_IO)) {
+        if (!use_exit_tb(ctx)) {
             tcg_gen_goto_tb(0);
             tcg_gen_movi_i64(cpu_pc, entry);
             tcg_gen_exit_tb((uintptr_t)ctx->tb);
@@ -2686,7 +2696,8 @@ static ExitStatus translate_one(DisasContext *ctx, uint32_t insn)
         tcg_gen_andi_i64(tmp, vb, 1);
         tcg_gen_st8_i64(tmp, cpu_env, offsetof(CPUAlphaState, pal_mode));
         tcg_gen_andi_i64(cpu_pc, vb, ~3);
-        ret = EXIT_PC_UPDATED;
+        /* Allow interrupts to be recognized right away.  */
+        ret = EXIT_PC_UPDATED_NOCHAIN;
         break;
 #else
         goto invalid_opc;
@@ -3010,6 +3021,12 @@ void gen_intermediate_code(CPUAlphaState *env, struct TranslationBlock *tb)
         tcg_gen_movi_i64(cpu_pc, ctx.pc);
         /* FALLTHRU */
     case EXIT_PC_UPDATED:
+        if (!use_exit_tb(&ctx)) {
+            tcg_gen_lookup_and_goto_ptr(cpu_pc);
+            break;
+        }
+        /* FALLTHRU */
+    case EXIT_PC_UPDATED_NOCHAIN:
         if (ctx.singlestep_enabled) {
             gen_excp_1(EXCP_DEBUG, 0);
         } else {
diff --git a/target/arm/translate-a64.c b/target/arm/translate-a64.c
index 860e279658..e55547d95d 100644
--- a/target/arm/translate-a64.c
+++ b/target/arm/translate-a64.c
@@ -1422,7 +1422,9 @@ static void handle_msr_i(DisasContext *s, uint32_t insn,
         gen_helper_msr_i_pstate(cpu_env, tcg_op, tcg_imm);
         tcg_temp_free_i32(tcg_imm);
         tcg_temp_free_i32(tcg_op);
-        s->is_jmp = DISAS_UPDATE;
+        /* For DAIFClear, exit the cpu loop to re-evaluate pending IRQs.  */
+        gen_a64_set_pc_im(s->pc);
+        s->is_jmp = (op == 0x1f ? DISAS_EXIT : DISAS_JUMP);
         break;
     }
     default:
@@ -11369,6 +11371,9 @@ void gen_intermediate_code_a64(ARMCPU *cpu, TranslationBlock *tb)
         case DISAS_JUMP:
             tcg_gen_lookup_and_goto_ptr(cpu_pc);
             break;
+        case DISAS_EXIT:
+            tcg_gen_exit_tb(0);
+            break;
         case DISAS_TB_JUMP:
         case DISAS_EXC:
         case DISAS_SWI:
diff --git a/target/s390x/translate.c b/target/s390x/translate.c
index 8c055b7bb7..640354271c 100644
--- a/target/s390x/translate.c
+++ b/target/s390x/translate.c
@@ -1173,6 +1173,8 @@ typedef enum {
     /* We are exiting the TB, but have neither emitted a goto_tb, nor
        updated the PC for the next instruction to be executed.  */
     EXIT_PC_STALE,
+    /* We are exiting the TB to the main loop.  */
+    EXIT_PC_STALE_NOCHAIN,
     /* We are ending the TB with a noreturn function call, e.g. longjmp.
        No following code will be executed.  */
     EXIT_NORETURN,
@@ -3795,7 +3797,8 @@ static ExitStatus op_ssm(DisasContext *s, DisasOps *o)
 {
     check_privileged(s);
     tcg_gen_deposit_i64(psw_mask, psw_mask, o->in2, 56, 8);
-    return NO_EXIT;
+    /* Exit to main loop to reevaluate s390_cpu_exec_interrupt.  */
+    return EXIT_PC_STALE_NOCHAIN;
 }
 
 static ExitStatus op_stap(DisasContext *s, DisasOps *o)
@@ -4038,7 +4041,9 @@ static ExitStatus op_stnosm(DisasContext *s, DisasOps *o)
     } else {
         tcg_gen_ori_i64(psw_mask, psw_mask, i2 << 56);
     }
-    return NO_EXIT;
+
+    /* Exit to main loop to reevaluate s390_cpu_exec_interrupt.  */
+    return EXIT_PC_STALE_NOCHAIN;
 }
 
 static ExitStatus op_stura(DisasContext *s, DisasOps *o)
@@ -5788,6 +5793,7 @@ void gen_intermediate_code(CPUS390XState *env, struct TranslationBlock *tb)
     case EXIT_NORETURN:
         break;
     case EXIT_PC_STALE:
+    case EXIT_PC_STALE_NOCHAIN:
         update_psw_addr(&dc);
         /* FALLTHRU */
     case EXIT_PC_UPDATED:
@@ -5799,14 +5805,14 @@ void gen_intermediate_code(CPUS390XState *env, struct TranslationBlock *tb)
         /* Exit the TB, either by raising a debug exception or by return.  */
         if (do_debug) {
             gen_exception(EXCP_DEBUG);
-        } else if (use_exit_tb(&dc)) {
+        } else if (use_exit_tb(&dc) || status == EXIT_PC_STALE_NOCHAIN) {
             tcg_gen_exit_tb(0);
         } else {
             tcg_gen_lookup_and_goto_ptr(psw_addr);
         }
         break;
     default:
-        abort();
+        g_assert_not_reached();
     }
 
     gen_tb_end(tb, num_insns);
diff --git a/tcg/aarch64/tcg-target.inc.c b/tcg/aarch64/tcg-target.inc.c
index 5f185458f1..1fa3bccc89 100644
--- a/tcg/aarch64/tcg-target.inc.c
+++ b/tcg/aarch64/tcg-target.inc.c
@@ -616,7 +616,12 @@ static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg rd,
     /* Look for host pointer values within 4G of the PC.  This happens
        often when loading pointers to QEMU's own data structures.  */
     if (type == TCG_TYPE_I64) {
-        tcg_target_long disp = (value >> 12) - ((intptr_t)s->code_ptr >> 12);
+        tcg_target_long disp = value - (intptr_t)s->code_ptr;
+        if (disp == sextract64(disp, 0, 21)) {
+            tcg_out_insn(s, 3406, ADR, rd, disp);
+            return;
+        }
+        disp = (value >> 12) - ((intptr_t)s->code_ptr >> 12);
         if (disp == sextract64(disp, 0, 21)) {
             tcg_out_insn(s, 3406, ADRP, rd, disp);
             if (value & 0xfff) {
diff --git a/tcg/arm/tcg-target.inc.c b/tcg/arm/tcg-target.inc.c
index 9f5cb66718..d1793ec77d 100644
--- a/tcg/arm/tcg-target.inc.c
+++ b/tcg/arm/tcg-target.inc.c
@@ -418,23 +418,37 @@ static inline void tcg_out_dat_imm(TCGContext *s,
 
 static void tcg_out_movi32(TCGContext *s, int cond, int rd, uint32_t arg)
 {
-    int rot, opc, rn;
-
-    /* For armv7, make sure not to use movw+movt when mov/mvn would do.
-       Speed things up by only checking when movt would be required.
-       Prior to armv7, have one go at fully rotated immediates before
-       doing the decomposition thing below.  */
-    if (!use_armv7_instructions || (arg & 0xffff0000)) {
-        rot = encode_imm(arg);
+    int rot, opc, rn, diff;
+
+    /* Check a single MOV/MVN before anything else.  */
+    rot = encode_imm(arg);
+    if (rot >= 0) {
+        tcg_out_dat_imm(s, cond, ARITH_MOV, rd, 0,
+                        rotl(arg, rot) | (rot << 7));
+        return;
+    }
+    rot = encode_imm(~arg);
+    if (rot >= 0) {
+        tcg_out_dat_imm(s, cond, ARITH_MVN, rd, 0,
+                        rotl(~arg, rot) | (rot << 7));
+        return;
+    }
+
+    /* Check for a pc-relative address.  This will usually be the TB,
+       or within the TB, which is immediately before the code block.  */
+    diff = arg - ((intptr_t)s->code_ptr + 8);
+    if (diff >= 0) {
+        rot = encode_imm(diff);
         if (rot >= 0) {
-            tcg_out_dat_imm(s, cond, ARITH_MOV, rd, 0,
-                            rotl(arg, rot) | (rot << 7));
+            tcg_out_dat_imm(s, cond, ARITH_ADD, rd, TCG_REG_PC,
+                            rotl(diff, rot) | (rot << 7));
             return;
         }
-        rot = encode_imm(~arg);
+    } else {
+        rot = encode_imm(-diff);
         if (rot >= 0) {
-            tcg_out_dat_imm(s, cond, ARITH_MVN, rd, 0,
-                            rotl(~arg, rot) | (rot << 7));
+            tcg_out_dat_imm(s, cond, ARITH_SUB, rd, TCG_REG_PC,
+                            rotl(-diff, rot) | (rot << 7));
             return;
         }
     }
@@ -1026,16 +1040,6 @@ static void tcg_out_call(TCGContext *s, tcg_insn_unit *addr)
     }
 }
 
-void arm_tb_set_jmp_target(uintptr_t jmp_addr, uintptr_t addr)
-{
-    tcg_insn_unit *code_ptr = (tcg_insn_unit *)jmp_addr;
-    tcg_insn_unit *target = (tcg_insn_unit *)addr;
-
-    /* we could use a ldr pc, [pc, #-4] kind of branch and avoid the flush */
-    reloc_pc24_atomic(code_ptr, target);
-    flush_icache_range(jmp_addr, jmp_addr + 4);
-}
-
 static inline void tcg_out_goto_label(TCGContext *s, int cond, TCGLabel *l)
 {
     if (l->has_value) {
@@ -1665,17 +1669,27 @@ static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
         }
         break;
     case INDEX_op_goto_tb:
-        if (s->tb_jmp_insn_offset) {
-            /* Direct jump method */
-            s->tb_jmp_insn_offset[args[0]] = tcg_current_code_size(s);
-            tcg_out_b_noaddr(s, COND_AL);
-        } else {
+        {
             /* Indirect jump method */
-            intptr_t ptr = (intptr_t)(s->tb_jmp_target_addr + args[0]);
-            tcg_out_movi32(s, COND_AL, TCG_REG_R0, ptr & ~0xfff);
-            tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_R0, ptr & 0xfff);
+            intptr_t ptr, dif, dil;
+            TCGReg base = TCG_REG_PC;
+
+            tcg_debug_assert(s->tb_jmp_insn_offset == 0);
+            ptr = (intptr_t)(s->tb_jmp_target_addr + args[0]);
+            dif = ptr - ((intptr_t)s->code_ptr + 8);
+            dil = sextract32(dif, 0, 12);
+            if (dif != dil) {
+                /* The TB is close, but outside the 12 bits addressable by
+                   the load.  We can extend this to 20 bits with a sub of a
+                   shifted immediate from pc.  In the vastly unlikely event
+                   the code requires more than 1MB, we'll use 2 insns and
+                   be no worse off.  */
+                base = TCG_REG_R0;
+                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);
         }
-        s->tb_jmp_reset_offset[args[0]] = tcg_current_code_size(s);
         break;
     case INDEX_op_goto_ptr:
         tcg_out_bx(s, COND_AL, args[0]);
diff --git a/tcg/ppc/tcg-target.inc.c b/tcg/ppc/tcg-target.inc.c
index 8d50f18328..1f690df20d 100644
--- a/tcg/ppc/tcg-target.inc.c
+++ b/tcg/ppc/tcg-target.inc.c
@@ -2820,14 +2820,11 @@ void tcg_register_jit(void *buf, size_t buf_size)
 }
 #endif /* __ELF__ */
 
-static size_t dcache_bsize = 16;
-static size_t icache_bsize = 16;
-
 void flush_icache_range(uintptr_t start, uintptr_t stop)
 {
     uintptr_t p, start1, stop1;
-    size_t dsize = dcache_bsize;
-    size_t isize = icache_bsize;
+    size_t dsize = qemu_dcache_linesize;
+    size_t isize = qemu_icache_linesize;
 
     start1 = start & ~(dsize - 1);
     stop1 = (stop + dsize - 1) & ~(dsize - 1);
@@ -2844,67 +2841,3 @@ void flush_icache_range(uintptr_t start, uintptr_t stop)
     asm volatile ("sync" : : : "memory");
     asm volatile ("isync" : : : "memory");
 }
-
-#if defined _AIX
-#include <sys/systemcfg.h>
-
-static void __attribute__((constructor)) tcg_cache_init(void)
-{
-    icache_bsize = _system_configuration.icache_line;
-    dcache_bsize = _system_configuration.dcache_line;
-}
-
-#elif defined __linux__
-static void __attribute__((constructor)) tcg_cache_init(void)
-{
-    unsigned long dsize = qemu_getauxval(AT_DCACHEBSIZE);
-    unsigned long isize = qemu_getauxval(AT_ICACHEBSIZE);
-
-    if (dsize == 0 || isize == 0) {
-        if (dsize == 0) {
-            fprintf(stderr, "getauxval AT_DCACHEBSIZE failed\n");
-        }
-        if (isize == 0) {
-            fprintf(stderr, "getauxval AT_ICACHEBSIZE failed\n");
-        }
-        exit(1);
-    }
-    dcache_bsize = dsize;
-    icache_bsize = isize;
-}
-
-#elif defined __APPLE__
-#include <sys/sysctl.h>
-
-static void __attribute__((constructor)) tcg_cache_init(void)
-{
-    size_t len;
-    unsigned cacheline;
-    int name[2] = { CTL_HW, HW_CACHELINE };
-
-    len = sizeof(cacheline);
-    if (sysctl(name, 2, &cacheline, &len, NULL, 0)) {
-        perror("sysctl CTL_HW HW_CACHELINE failed");
-        exit(1);
-    }
-    dcache_bsize = cacheline;
-    icache_bsize = cacheline;
-}
-
-#elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
-#include <sys/sysctl.h>
-
-static void __attribute__((constructor)) tcg_cache_init(void)
-{
-    size_t len = 4;
-    unsigned cacheline;
-
-    if (sysctlbyname ("machdep.cacheline_size", &cacheline, &len, NULL, 0)) {
-        fprintf(stderr, "sysctlbyname machdep.cacheline_size failed: %s\n",
-                strerror(errno));
-        exit(1);
-    }
-    dcache_bsize = cacheline;
-    icache_bsize = cacheline;
-}
-#endif
diff --git a/tcg/tcg-runtime.c b/tcg/tcg-runtime.c
index 7fa90ce508..ec3a34e461 100644
--- a/tcg/tcg-runtime.c
+++ b/tcg/tcg-runtime.c
@@ -149,23 +149,23 @@ void *HELPER(lookup_tb_ptr)(CPUArchState *env, target_ulong addr)
     CPUState *cpu = ENV_GET_CPU(env);
     TranslationBlock *tb;
     target_ulong cs_base, pc;
-    uint32_t flags;
-
-    tb = atomic_rcu_read(&cpu->tb_jmp_cache[tb_jmp_cache_hash_func(addr)]);
-    if (likely(tb)) {
-        cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);
-        if (likely(tb->pc == addr && tb->cs_base == cs_base &&
-                   tb->flags == flags)) {
-            goto found;
-        }
+    uint32_t flags, addr_hash;
+
+    addr_hash = tb_jmp_cache_hash_func(addr);
+    tb = atomic_rcu_read(&cpu->tb_jmp_cache[addr_hash]);
+    cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);
+
+    if (unlikely(!(tb
+                   && tb->pc == addr
+                   && tb->cs_base == cs_base
+                   && tb->flags == flags))) {
         tb = tb_htable_lookup(cpu, addr, cs_base, flags);
-        if (likely(tb)) {
-            atomic_set(&cpu->tb_jmp_cache[tb_jmp_cache_hash_func(addr)], tb);
-            goto found;
+        if (!tb) {
+            return tcg_ctx.code_gen_epilogue;
         }
+        atomic_set(&cpu->tb_jmp_cache[addr_hash], tb);
     }
-    return tcg_ctx.code_gen_epilogue;
- found:
+
     qemu_log_mask_and_addr(CPU_LOG_EXEC, addr,
                            "Chain %p [%d: " TARGET_FMT_lx "] %s\n",
                            tb->tc_ptr, cpu->cpu_index, addr,
diff --git a/tcg/tcg.c b/tcg/tcg.c
index 564292f54d..35598296c5 100644
--- a/tcg/tcg.c
+++ b/tcg/tcg.c
@@ -383,6 +383,26 @@ void tcg_context_init(TCGContext *s)
     }
 }
 
+/*
+ * Allocate TBs right before their corresponding translated code, making
+ * sure that TBs and code are on different cache lines.
+ */
+TranslationBlock *tcg_tb_alloc(TCGContext *s)
+{
+    uintptr_t align = qemu_icache_linesize;
+    TranslationBlock *tb;
+    void *next;
+
+    tb = (void *)ROUND_UP((uintptr_t)s->code_gen_ptr, align);
+    next = (void *)ROUND_UP((uintptr_t)(tb + 1), align);
+
+    if (unlikely(next > s->code_gen_highwater)) {
+        return NULL;
+    }
+    s->code_gen_ptr = next;
+    return tb;
+}
+
 void tcg_prologue_init(TCGContext *s)
 {
     size_t prologue_size, total_size;
diff --git a/tcg/tcg.h b/tcg/tcg.h
index 5ec48d1787..9e37722799 100644
--- a/tcg/tcg.h
+++ b/tcg/tcg.h
@@ -697,7 +697,6 @@ struct TCGContext {
        here, because there's too much arithmetic throughout that relies
        on addition and subtraction working on bytes.  Rely on the GCC
        extension that allows arithmetic on void*.  */
-    int code_gen_max_blocks;
     void *code_gen_prologue;
     void *code_gen_epilogue;
     void *code_gen_buffer;
@@ -756,6 +755,7 @@ static inline bool tcg_op_buf_full(void)
 /* tb_lock must be held for tcg_malloc_internal. */
 void *tcg_malloc_internal(TCGContext *s, int size);
 void tcg_pool_reset(TCGContext *s);
+TranslationBlock *tcg_tb_alloc(TCGContext *s);
 
 void tb_lock(void);
 void tb_unlock(void);
diff --git a/util/Makefile.objs b/util/Makefile.objs
index 8a333d3dd7..50a55ecc75 100644
--- a/util/Makefile.objs
+++ b/util/Makefile.objs
@@ -20,6 +20,7 @@ util-obj-y += host-utils.o
 util-obj-y += bitmap.o bitops.o hbitmap.o
 util-obj-y += fifo8.o
 util-obj-y += acl.o
+util-obj-y += cacheinfo.o
 util-obj-y += error.o qemu-error.o
 util-obj-y += id.o
 util-obj-y += iov.o qemu-config.o qemu-sockets.o uri.o notify.o
diff --git a/util/cacheinfo.c b/util/cacheinfo.c
new file mode 100644
index 0000000000..f987522df4
--- /dev/null
+++ b/util/cacheinfo.c
@@ -0,0 +1,185 @@
+/*
+ * cacheinfo.c - helpers to query the host about its caches
+ *
+ * Copyright (C) 2017, Emilio G. Cota <cota@braap.org>
+ * License: GNU GPL, version 2 or later.
+ *   See the COPYING file in the top-level directory.
+ */
+
+#include "qemu/osdep.h"
+
+int qemu_icache_linesize = 0;
+int qemu_dcache_linesize = 0;
+
+/*
+ * Operating system specific detection mechanisms.
+ */
+
+#if defined(_AIX)
+# include <sys/systemcfg.h>
+
+static void sys_cache_info(int *isize, int *dsize)
+{
+    *isize = _system_configuration.icache_line;
+    *dsize = _system_configuration.dcache_line;
+}
+
+#elif defined(_WIN32)
+
+static void sys_cache_info(int *isize, int *dsize)
+{
+    SYSTEM_LOGICAL_PROCESSOR_INFORMATION *buf;
+    DWORD size = 0;
+    BOOL success;
+    size_t i, n;
+
+    /* Check for the required buffer size first.  Note that if the zero
+       size we use for the probe results in success, then there is no
+       data available; fail in that case.  */
+    success = GetLogicalProcessorInformation(0, &size);
+    if (success || GetLastError() != ERROR_INSUFFICIENT_BUFFER) {
+        return;
+    }
+
+    n = size / sizeof(SYSTEM_LOGICAL_PROCESSOR_INFORMATION);
+    size = n * sizeof(SYSTEM_LOGICAL_PROCESSOR_INFORMATION);
+    buf = g_new0(SYSTEM_LOGICAL_PROCESSOR_INFORMATION, n);
+    if (!GetLogicalProcessorInformation(buf, &size)) {
+        goto fail;
+    }
+
+    for (i = 0; i < n; i++) {
+        if (buf[i].Relationship == RelationCache
+            && buf[i].Cache.Level == 1) {
+            switch (buf[i].Cache.Type) {
+            case CacheUnified:
+                *isize = *dsize = buf[i].Cache.LineSize;
+                break;
+            case CacheInstruction:
+                *isize = buf[i].Cache.LineSize;
+                break;
+            case CacheData:
+                *dsize = buf[i].Cache.LineSize;
+                break;
+            default:
+                break;
+            }
+        }
+    }
+ fail:
+    g_free(buf);
+}
+
+#elif defined(__APPLE__) \
+      || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
+# include <sys/sysctl.h>
+# if defined(__APPLE__)
+#  define SYSCTL_CACHELINE_NAME "hw.cachelinesize"
+# else
+#  define SYSCTL_CACHELINE_NAME "machdep.cacheline_size"
+# endif
+
+static void sys_cache_info(int *isize, int *dsize)
+{
+    /* There's only a single sysctl for both I/D cache line sizes.  */
+    long size;
+    size_t len = sizeof(size);
+    if (!sysctlbyname(SYSCTL_CACHELINE_NAME, &size, &len, NULL, 0)) {
+        *isize = *dsize = size;
+    }
+}
+
+#else
+/* POSIX */
+
+static void sys_cache_info(int *isize, int *dsize)
+{
+# ifdef _SC_LEVEL1_ICACHE_LINESIZE
+    *isize = sysconf(_SC_LEVEL1_ICACHE_LINESIZE);
+# endif
+# ifdef _SC_LEVEL1_DCACHE_LINESIZE
+    *dsize = sysconf(_SC_LEVEL1_DCACHE_LINESIZE);
+# endif
+}
+#endif /* sys_cache_info */
+
+/*
+ * Architecture (+ OS) specific detection mechanisms.
+ */
+
+#if defined(__aarch64__)
+
+static void arch_cache_info(int *isize, int *dsize)
+{
+    if (*isize == 0 || *dsize == 0) {
+        unsigned ctr;
+
+        /* The real cache geometry is in CCSIDR_EL1/CLIDR_EL1/CSSELR_EL1,
+           but (at least under Linux) these are marked protected by the
+           kernel.  However, CTR_EL0 contains the minimum linesize in the
+           entire hierarchy, and is used by userspace cache flushing.  */
+        asm volatile("mrs\t%0, ctr_el0" : "=r"(ctr));
+        if (*isize == 0) {
+            *isize = 4 << (ctr & 0xf);
+        }
+        if (*dsize == 0) {
+            *dsize = 4 << ((ctr >> 16) & 0xf);
+        }
+    }
+}
+
+#elif defined(_ARCH_PPC) && defined(__linux__)
+
+static void arch_cache_info(int *isize, int *dsize)
+{
+    if (*isize == 0) {
+        *isize = qemu_getauxval(AT_ICACHEBSIZE);
+    }
+    if (*dsize == 0) {
+        *dsize = qemu_getauxval(AT_DCACHEBSIZE);
+    }
+}
+
+#else
+static void arch_cache_info(int *isize, int *dsize) { }
+#endif /* arch_cache_info */
+
+/*
+ * ... and if all else fails ...
+ */
+
+static void fallback_cache_info(int *isize, int *dsize)
+{
+    /* If we can only find one of the two, assume they're the same.  */
+    if (*isize) {
+        if (*dsize) {
+            /* Success! */
+        } else {
+            *dsize = *isize;
+        }
+    } else if (*dsize) {
+        *isize = *dsize;
+    } else {
+#if defined(_ARCH_PPC)
+        /* For PPC, we're going to use the icache size computed for
+           flush_icache_range.  Which means that we must use the
+           architecture minimum.  */
+        *isize = *dsize = 16;
+#else
+        /* Otherwise, 64 bytes is not uncommon.  */
+        *isize = *dsize = 64;
+#endif
+    }
+}
+
+static void __attribute__((constructor)) init_cache_info(void)
+{
+    int isize = 0, dsize = 0;
+
+    sys_cache_info(&isize, &dsize);
+    arch_cache_info(&isize, &dsize);
+    fallback_cache_info(&isize, &dsize);
+
+    qemu_icache_linesize = isize;
+    qemu_dcache_linesize = dsize;
+}