[PATCH v3 0/6] tci and macosx fixes

[PATCH v3 0/6] tci and macosx fixes

[Richard Henderson]

Changes for v3:
  - Skip v2, re-posted by phil,
  - Drop any changes related to maybe_extend_addr64,
    since the atomics are using helpers already.
  - Drop some changes related to plugins.  This does
    duplicate some code, but simplifies temp management.
  - Don't extend for 32-bit tci hosts.
  - New fixes for other failures affecting macosx.


r~


Richard Henderson (6):
  tcg: Zero extend 32-bit addresses for TCI
  tcg/tci: Introduce INDEX_op_tci_qemu_{ld,st}_rrr
  tcg: Remove duplicate test from plugin_gen_mem_callbacks
  tcg/tci: Disable -Wundef FFI_GO_CLOSURES warning
  include/generic/host: Fix atomic128-cas.h.inc for Int128 structure
  include/aarch64/host: Fix atomic16_fetch_{and,or}

 include/tcg/helper-info.h                     | 12 +++
 tcg/tcg-op-ldst.c                             | 96 ++++++++++++++-----
 tcg/tci.c                                     | 19 ++++
 host/include/aarch64/host/atomic128-cas.h.inc | 12 +--
 host/include/generic/host/atomic128-cas.h.inc | 24 +++--
 tcg/tci/tcg-target-opc.h.inc                  |  2 +
 tcg/tci/tcg-target.c.inc                      | 14 ++-
 7 files changed, 138 insertions(+), 41 deletions(-)

-- 
2.43.0

[PATCH v3 1/6] tcg: Zero extend 32-bit addresses for TCI

[Richard Henderson]

For native code generation, zero-extending 32-bit addresses for
the slow path helpers happens in tcg_out_{ld,st}_helper_args,
but there isn't really a slow path for TCI, so that didn't happen.

Make the extension for TCI explicit in the opcode stream,
much like we already do for plugins and atomic helpers.

Cc: qemu-stable@nongnu.org
Fixes: 24e46e6c9d9 ("accel/tcg: Widen tcg-ldst.h addresses to uint64_t")
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/tcg-op-ldst.c | 72 +++++++++++++++++++++++++++++++++++++++--------
 1 file changed, 61 insertions(+), 11 deletions(-)

diff --git a/tcg/tcg-op-ldst.c b/tcg/tcg-op-ldst.c
index 67c15fd4d0..1c0b06dbc7 100644
--- a/tcg/tcg-op-ldst.c
+++ b/tcg/tcg-op-ldst.c
@@ -135,6 +135,30 @@ static void tcg_gen_req_mo(TCGBar type)
     }
 }
 
+static TCGTemp *tci_extend_addr(TCGTemp *addr)
+{
+#ifdef CONFIG_TCG_INTERPRETER
+    /*
+     * 64-bit interpreter requires 64-bit addresses.
+     * Compare to the extension performed by tcg_out_{ld,st}_helper_args
+     * for native code generation.
+     */
+    if (TCG_TARGET_REG_BITS == 64 && tcg_ctx->addr_type == TCG_TYPE_I32) {
+        TCGv_i64 temp = tcg_temp_ebb_new_i64();
+        tcg_gen_extu_i32_i64(temp, temp_tcgv_i32(addr));
+        return tcgv_i64_temp(temp);
+    }
+#endif
+    return addr;
+}
+
+static void maybe_free_addr(TCGTemp *addr, TCGTemp *copy)
+{
+    if (addr != copy) {
+        tcg_temp_free_internal(copy);
+    }
+}
+
 /* Only required for loads, where value might overlap addr. */
 static TCGv_i64 plugin_maybe_preserve_addr(TCGTemp *addr)
 {
@@ -234,6 +258,7 @@ static void tcg_gen_qemu_ld_i32_int(TCGv_i32 val, TCGTemp *addr,
     MemOp orig_memop;
     MemOpIdx orig_oi, oi;
     TCGv_i64 copy_addr;
+    TCGTemp *addr_new;
 
     tcg_gen_req_mo(TCG_MO_LD_LD | TCG_MO_ST_LD);
     orig_memop = memop = tcg_canonicalize_memop(memop, 0, 0);
@@ -248,10 +273,12 @@ static void tcg_gen_qemu_ld_i32_int(TCGv_i32 val, TCGTemp *addr,
         oi = make_memop_idx(memop, idx);
     }
 
+    addr_new = tci_extend_addr(addr);
     copy_addr = plugin_maybe_preserve_addr(addr);
-    gen_ldst1(INDEX_op_qemu_ld, TCG_TYPE_I32, tcgv_i32_temp(val), addr, oi);
+    gen_ldst1(INDEX_op_qemu_ld, TCG_TYPE_I32, tcgv_i32_temp(val), addr_new, oi);
     plugin_gen_mem_callbacks_i32(val, copy_addr, addr, orig_oi,
                                  QEMU_PLUGIN_MEM_R);
+    maybe_free_addr(addr, addr_new);
 
     if ((orig_memop ^ memop) & MO_BSWAP) {
         switch (orig_memop & MO_SIZE) {
@@ -282,6 +309,7 @@ static void tcg_gen_qemu_st_i32_int(TCGv_i32 val, TCGTemp *addr,
 {
     TCGv_i32 swap = NULL;
     MemOpIdx orig_oi, oi;
+    TCGTemp *addr_new;
 
     tcg_gen_req_mo(TCG_MO_LD_ST | TCG_MO_ST_ST);
     memop = tcg_canonicalize_memop(memop, 0, 1);
@@ -304,8 +332,10 @@ static void tcg_gen_qemu_st_i32_int(TCGv_i32 val, TCGTemp *addr,
         oi = make_memop_idx(memop, idx);
     }
 
-    gen_ldst1(INDEX_op_qemu_st, TCG_TYPE_I32, tcgv_i32_temp(val), addr, oi);
+    addr_new = tci_extend_addr(addr);
+    gen_ldst1(INDEX_op_qemu_st, TCG_TYPE_I32, tcgv_i32_temp(val), addr_new, oi);
     plugin_gen_mem_callbacks_i32(val, NULL, addr, orig_oi, QEMU_PLUGIN_MEM_W);
+    maybe_free_addr(addr, addr_new);
 
     if (swap) {
         tcg_temp_free_i32(swap);
@@ -326,6 +356,7 @@ static void tcg_gen_qemu_ld_i64_int(TCGv_i64 val, TCGTemp *addr,
     MemOp orig_memop;
     MemOpIdx orig_oi, oi;
     TCGv_i64 copy_addr;
+    TCGTemp *addr_new;
 
     if (TCG_TARGET_REG_BITS == 32 && (memop & MO_SIZE) < MO_64) {
         tcg_gen_qemu_ld_i32_int(TCGV_LOW(val), addr, idx, memop);
@@ -350,10 +381,12 @@ static void tcg_gen_qemu_ld_i64_int(TCGv_i64 val, TCGTemp *addr,
         oi = make_memop_idx(memop, idx);
     }
 
+    addr_new = tci_extend_addr(addr);
     copy_addr = plugin_maybe_preserve_addr(addr);
-    gen_ld_i64(val, addr, oi);
+    gen_ld_i64(val, addr_new, oi);
     plugin_gen_mem_callbacks_i64(val, copy_addr, addr, orig_oi,
                                  QEMU_PLUGIN_MEM_R);
+    maybe_free_addr(addr, addr_new);
 
     if ((orig_memop ^ memop) & MO_BSWAP) {
         int flags = (orig_memop & MO_SIGN
@@ -388,6 +421,7 @@ static void tcg_gen_qemu_st_i64_int(TCGv_i64 val, TCGTemp *addr,
 {
     TCGv_i64 swap = NULL;
     MemOpIdx orig_oi, oi;
+    TCGTemp *addr_new;
 
     if (TCG_TARGET_REG_BITS == 32 && (memop & MO_SIZE) < MO_64) {
         tcg_gen_qemu_st_i32_int(TCGV_LOW(val), addr, idx, memop);
@@ -418,8 +452,10 @@ static void tcg_gen_qemu_st_i64_int(TCGv_i64 val, TCGTemp *addr,
         oi = make_memop_idx(memop, idx);
     }
 
-    gen_st_i64(val, addr, oi);
+    addr_new = tci_extend_addr(addr);
+    gen_st_i64(val, addr_new, oi);
     plugin_gen_mem_callbacks_i64(val, NULL, addr, orig_oi, QEMU_PLUGIN_MEM_W);
+    maybe_free_addr(addr, addr_new);
 
     if (swap) {
         tcg_temp_free_i64(swap);
@@ -530,6 +566,7 @@ static void tcg_gen_qemu_ld_i128_int(TCGv_i128 val, TCGTemp *addr,
 {
     MemOpIdx orig_oi;
     TCGv_i64 ext_addr = NULL;
+    TCGTemp *addr_new;
 
     check_max_alignment(memop_alignment_bits(memop));
     tcg_gen_req_mo(TCG_MO_LD_LD | TCG_MO_ST_LD);
@@ -557,8 +594,10 @@ static void tcg_gen_qemu_ld_i128_int(TCGv_i128 val, TCGTemp *addr,
             hi = TCGV128_HIGH(val);
         }
 
+        addr_new = tci_extend_addr(addr);
         gen_ldst2(INDEX_op_qemu_ld2, TCG_TYPE_I128, tcgv_i64_temp(lo),
-                  tcgv_i64_temp(hi), addr, oi);
+                  tcgv_i64_temp(hi), addr_new, oi);
+        maybe_free_addr(addr, addr_new);
 
         if (need_bswap) {
             tcg_gen_bswap64_i64(lo, lo);
@@ -586,7 +625,9 @@ static void tcg_gen_qemu_ld_i128_int(TCGv_i128 val, TCGTemp *addr,
             y = TCGV128_LOW(val);
         }
 
-        gen_ld_i64(x, addr, make_memop_idx(mop[0], idx));
+        addr_new = tci_extend_addr(addr);
+        gen_ld_i64(x, addr_new, make_memop_idx(mop[0], idx));
+        maybe_free_addr(addr, addr_new);
 
         if (need_bswap) {
             tcg_gen_bswap64_i64(x, x);
@@ -602,7 +643,9 @@ static void tcg_gen_qemu_ld_i128_int(TCGv_i128 val, TCGTemp *addr,
             addr_p8 = tcgv_i64_temp(t);
         }
 
-        gen_ld_i64(y, addr_p8, make_memop_idx(mop[1], idx));
+        addr_new = tci_extend_addr(addr_p8);
+        gen_ld_i64(y, addr_new, make_memop_idx(mop[1], idx));
+        maybe_free_addr(addr_p8, addr_new);
         tcg_temp_free_internal(addr_p8);
 
         if (need_bswap) {
@@ -636,6 +679,7 @@ static void tcg_gen_qemu_st_i128_int(TCGv_i128 val, TCGTemp *addr,
 {
     MemOpIdx orig_oi;
     TCGv_i64 ext_addr = NULL;
+    TCGTemp *addr_new;
 
     check_max_alignment(memop_alignment_bits(memop));
     tcg_gen_req_mo(TCG_MO_ST_LD | TCG_MO_ST_ST);
@@ -666,8 +710,10 @@ static void tcg_gen_qemu_st_i128_int(TCGv_i128 val, TCGTemp *addr,
             hi = TCGV128_HIGH(val);
         }
 
+        addr_new = tci_extend_addr(addr);
         gen_ldst2(INDEX_op_qemu_st2, TCG_TYPE_I128,
-                  tcgv_i64_temp(lo), tcgv_i64_temp(hi), addr, oi);
+                  tcgv_i64_temp(lo), tcgv_i64_temp(hi), addr_new, oi);
+        maybe_free_addr(addr, addr_new);
 
         if (need_bswap) {
             tcg_temp_free_i64(lo);
@@ -694,7 +740,9 @@ static void tcg_gen_qemu_st_i128_int(TCGv_i128 val, TCGTemp *addr,
             x = b;
         }
 
-        gen_st_i64(x, addr, make_memop_idx(mop[0], idx));
+        addr_new = tci_extend_addr(addr);
+        gen_st_i64(x, addr_new, make_memop_idx(mop[0], idx));
+        maybe_free_addr(addr, addr_new);
 
         if (tcg_ctx->addr_type == TCG_TYPE_I32) {
             TCGv_i32 t = tcg_temp_ebb_new_i32();
@@ -706,13 +754,15 @@ static void tcg_gen_qemu_st_i128_int(TCGv_i128 val, TCGTemp *addr,
             addr_p8 = tcgv_i64_temp(t);
         }
 
+        addr_new = tci_extend_addr(addr_p8);
         if (b) {
             tcg_gen_bswap64_i64(b, y);
-            gen_st_i64(b, addr_p8, make_memop_idx(mop[1], idx));
+            gen_st_i64(b, addr_new, make_memop_idx(mop[1], idx));
             tcg_temp_free_i64(b);
         } else {
-            gen_st_i64(y, addr_p8, make_memop_idx(mop[1], idx));
+            gen_st_i64(y, addr_new, make_memop_idx(mop[1], idx));
         }
+        maybe_free_addr(addr_p8, addr_new);
         tcg_temp_free_internal(addr_p8);
     } else {
         if (tcg_ctx->addr_type == TCG_TYPE_I32) {
-- 
2.43.0

[PATCH v3 2/6] tcg/tci: Introduce INDEX_op_tci_qemu_{ld,st}_rrr

[Richard Henderson]

Since d182123974c4, the number of bits in a MemOpIdx tops out at 17.
which won't fit in the TCI rrm format, thus an assertion failure.
Introduce new opcodes that take the MemOpIdx from a register, as
we already do for qemu_ld2 and qemu_st2.

Fixes: d182123974c4 ("include/exec/memopidx: Adjust for 32 mmu indexes")
Tested-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/tci.c                    | 19 +++++++++++++++++++
 tcg/tci/tcg-target-opc.h.inc |  2 ++
 tcg/tci/tcg-target.c.inc     | 14 ++++++++++++--
 3 files changed, 33 insertions(+), 2 deletions(-)

diff --git a/tcg/tci.c b/tcg/tci.c
index 700e672616..e15d4e8e08 100644
--- a/tcg/tci.c
+++ b/tcg/tci.c
@@ -794,12 +794,24 @@ uintptr_t QEMU_DISABLE_CFI tcg_qemu_tb_exec(CPUArchState *env,
             taddr = regs[r1];
             regs[r0] = tci_qemu_ld(env, taddr, oi, tb_ptr);
             break;
+        case INDEX_op_tci_qemu_ld_rrr:
+            tci_args_rrr(insn, &r0, &r1, &r2);
+            taddr = regs[r1];
+            oi = regs[r2];
+            regs[r0] = tci_qemu_ld(env, taddr, oi, tb_ptr);
+            break;
 
         case INDEX_op_qemu_st:
             tci_args_rrm(insn, &r0, &r1, &oi);
             taddr = regs[r1];
             tci_qemu_st(env, taddr, regs[r0], oi, tb_ptr);
             break;
+        case INDEX_op_tci_qemu_st_rrr:
+            tci_args_rrr(insn, &r0, &r1, &r2);
+            taddr = regs[r1];
+            oi = regs[r2];
+            tci_qemu_st(env, taddr, regs[r0], oi, tb_ptr);
+            break;
 
         case INDEX_op_qemu_ld2:
             tcg_debug_assert(TCG_TARGET_REG_BITS == 32);
@@ -1050,6 +1062,13 @@ int print_insn_tci(bfd_vma addr, disassemble_info *info)
                            op_name, str_r(r0), str_r(r1), oi);
         break;
 
+    case INDEX_op_tci_qemu_ld_rrr:
+    case INDEX_op_tci_qemu_st_rrr:
+        tci_args_rrr(insn, &r0, &r1, &r2);
+        info->fprintf_func(info->stream, "%-12s  %s, %s, %s",
+                           op_name, str_r(r0), str_r(r1), str_r(r2));
+        break;
+
     case INDEX_op_qemu_ld2:
     case INDEX_op_qemu_st2:
         tci_args_rrrr(insn, &r0, &r1, &r2, &r3);
diff --git a/tcg/tci/tcg-target-opc.h.inc b/tcg/tci/tcg-target-opc.h.inc
index 4eb32ed736..f8bfffc125 100644
--- a/tcg/tci/tcg-target-opc.h.inc
+++ b/tcg/tci/tcg-target-opc.h.inc
@@ -13,3 +13,5 @@ DEF(tci_rotl32, 1, 2, 0, TCG_OPF_NOT_PRESENT)
 DEF(tci_rotr32, 1, 2, 0, TCG_OPF_NOT_PRESENT)
 DEF(tci_setcond32, 1, 2, 1, TCG_OPF_NOT_PRESENT)
 DEF(tci_movcond32, 1, 2, 1, TCG_OPF_NOT_PRESENT)
+DEF(tci_qemu_ld_rrr, 1, 2, 0, TCG_OPF_NOT_PRESENT)
+DEF(tci_qemu_st_rrr, 0, 3, 0, TCG_OPF_NOT_PRESENT)
diff --git a/tcg/tci/tcg-target.c.inc b/tcg/tci/tcg-target.c.inc
index 35c66a4836..532f87262c 100644
--- a/tcg/tci/tcg-target.c.inc
+++ b/tcg/tci/tcg-target.c.inc
@@ -1188,7 +1188,12 @@ static const TCGOutOpStore outop_st = {
 static void tgen_qemu_ld(TCGContext *s, TCGType type, TCGReg data,
                          TCGReg addr, MemOpIdx oi)
 {
-    tcg_out_op_rrm(s, INDEX_op_qemu_ld, data, addr, oi);
+    if (oi & ~0xffff) {
+        tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_TMP, oi);
+        tcg_out_op_rrr(s, INDEX_op_tci_qemu_ld_rrr, data, addr, TCG_REG_TMP);
+    } else {
+        tcg_out_op_rrm(s, INDEX_op_qemu_ld, data, addr, oi);
+    }
 }
 
 static const TCGOutOpQemuLdSt outop_qemu_ld = {
@@ -1213,7 +1218,12 @@ static const TCGOutOpQemuLdSt2 outop_qemu_ld2 = {
 static void tgen_qemu_st(TCGContext *s, TCGType type, TCGReg data,
                          TCGReg addr, MemOpIdx oi)
 {
-    tcg_out_op_rrm(s, INDEX_op_qemu_st, data, addr, oi);
+    if (oi & ~0xffff) {
+        tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_TMP, oi);
+        tcg_out_op_rrr(s, INDEX_op_tci_qemu_st_rrr, data, addr, TCG_REG_TMP);
+    } else {
+        tcg_out_op_rrm(s, INDEX_op_qemu_st, data, addr, oi);
+    }
 }
 
 static const TCGOutOpQemuLdSt outop_qemu_st = {
-- 
2.43.0

Re: [PATCH v3 2/6] tcg/tci: Introduce INDEX_op_tci_qemu_{ld,st}_rrr

[Philippe Mathieu-Daudé]

On 3/12/25 17:14, Richard Henderson wrote:
> Since d182123974c4, the number of bits in a MemOpIdx tops out at 17.
> which won't fit in the TCI rrm format, thus an assertion failure.
> Introduce new opcodes that take the MemOpIdx from a register, as
> we already do for qemu_ld2 and qemu_st2.
> 
> Fixes: d182123974c4 ("include/exec/memopidx: Adjust for 32 mmu indexes")
> Tested-by: Alex Bennée <alex.bennee@linaro.org>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>   tcg/tci.c                    | 19 +++++++++++++++++++
>   tcg/tci/tcg-target-opc.h.inc |  2 ++
>   tcg/tci/tcg-target.c.inc     | 14 ++++++++++++--
>   3 files changed, 33 insertions(+), 2 deletions(-)

Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>

[PATCH v3 3/6] tcg: Remove duplicate test from plugin_gen_mem_callbacks

[Richard Henderson]

All callers have already tested tcg_ctx->plugin_insn.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/tcg-op-ldst.c | 24 +++++++++++-------------
 1 file changed, 11 insertions(+), 13 deletions(-)

diff --git a/tcg/tcg-op-ldst.c b/tcg/tcg-op-ldst.c
index 1c0b06dbc7..7716c3ad7c 100644
--- a/tcg/tcg-op-ldst.c
+++ b/tcg/tcg-op-ldst.c
@@ -182,23 +182,21 @@ static void
 plugin_gen_mem_callbacks(TCGv_i64 copy_addr, TCGTemp *orig_addr, MemOpIdx oi,
                          enum qemu_plugin_mem_rw rw)
 {
-    if (tcg_ctx->plugin_insn != NULL) {
-        qemu_plugin_meminfo_t info = make_plugin_meminfo(oi, rw);
+    qemu_plugin_meminfo_t info = make_plugin_meminfo(oi, rw);
 
-        if (tcg_ctx->addr_type == TCG_TYPE_I32) {
-            if (!copy_addr) {
-                copy_addr = tcg_temp_ebb_new_i64();
-                tcg_gen_extu_i32_i64(copy_addr, temp_tcgv_i32(orig_addr));
-            }
+    if (tcg_ctx->addr_type == TCG_TYPE_I32) {
+        if (!copy_addr) {
+            copy_addr = tcg_temp_ebb_new_i64();
+            tcg_gen_extu_i32_i64(copy_addr, temp_tcgv_i32(orig_addr));
+        }
+        tcg_gen_plugin_mem_cb(copy_addr, info);
+        tcg_temp_free_i64(copy_addr);
+    } else {
+        if (copy_addr) {
             tcg_gen_plugin_mem_cb(copy_addr, info);
             tcg_temp_free_i64(copy_addr);
         } else {
-            if (copy_addr) {
-                tcg_gen_plugin_mem_cb(copy_addr, info);
-                tcg_temp_free_i64(copy_addr);
-            } else {
-                tcg_gen_plugin_mem_cb(temp_tcgv_i64(orig_addr), info);
-            }
+            tcg_gen_plugin_mem_cb(temp_tcgv_i64(orig_addr), info);
         }
     }
 }
-- 
2.43.0

Re: [PATCH v3 3/6] tcg: Remove duplicate test from plugin_gen_mem_callbacks

[Philippe Mathieu-Daudé]

On 3/12/25 17:14, Richard Henderson wrote:
> All callers have already tested tcg_ctx->plugin_insn.
> 
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>   tcg/tcg-op-ldst.c | 24 +++++++++++-------------
>   1 file changed, 11 insertions(+), 13 deletions(-)

Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Tested-by: Philippe Mathieu-Daudé <philmd@linaro.org>

[PATCH v3 4/6] tcg/tci: Disable -Wundef FFI_GO_CLOSURES warning

[Richard Henderson]

Since we build TCI with FFI (commit 22f15579fa1 "tcg: Build ffi data
structures for helpers") we get on Darwin:

  In file included from ../../tcg/tci.c:22:
  In file included from include/tcg/helper-info.h:13:
  /Library/Developer/CommandLineTools/SDKs/MacOSX15.sdk/usr/include/ffi/ffi.h:483:5: warning: 'FFI_GO_CLOSURES' is not defined, evaluates to 0 [-Wundef]
    483 | #if FFI_GO_CLOSURES
        |     ^
  1 warning generated.

This was fixed in upstream libffi in 2023, but not backported to MacOSX.
Simply disable the warning locally.

Reported-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 include/tcg/helper-info.h | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/include/tcg/helper-info.h b/include/tcg/helper-info.h
index 909fe73afa..49a27e4eae 100644
--- a/include/tcg/helper-info.h
+++ b/include/tcg/helper-info.h
@@ -10,7 +10,19 @@
 #define TCG_HELPER_INFO_H
 
 #ifdef CONFIG_TCG_INTERPRETER
+/*
+ * MacOSX 15 uses an old version of libffi which contains
+ *   #if FFI_GO_CLOSURES
+ * but does not define that in <ffitarget.h>, included from <ffi.h>.
+ * This was fixed upstream with
+ *   https://github.com/libffi/libffi/commit/c23e9a1c
+ * We don't care about go closures one way or the other;
+ * just suppress the warning.
+ */
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wundef"
 #include <ffi.h>
+#pragma GCC diagnostic pop
 #endif
 #include "tcg-target-reg-bits.h"
 
-- 
2.43.0

Re: [PATCH v3 4/6] tcg/tci: Disable -Wundef FFI_GO_CLOSURES warning

[Philippe Mathieu-Daudé]

On 3/12/25 17:14, Richard Henderson wrote:
> Since we build TCI with FFI (commit 22f15579fa1 "tcg: Build ffi data
> structures for helpers") we get on Darwin:
> 
>    In file included from ../../tcg/tci.c:22:
>    In file included from include/tcg/helper-info.h:13:
>    /Library/Developer/CommandLineTools/SDKs/MacOSX15.sdk/usr/include/ffi/ffi.h:483:5: warning: 'FFI_GO_CLOSURES' is not defined, evaluates to 0 [-Wundef]
>      483 | #if FFI_GO_CLOSURES
>          |     ^
>    1 warning generated.
> 
> This was fixed in upstream libffi in 2023, but not backported to MacOSX.
> Simply disable the warning locally.
> 
> Reported-by: Philippe Mathieu-Daudé <philmd@linaro.org>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>   include/tcg/helper-info.h | 12 ++++++++++++
>   1 file changed, 12 insertions(+)

Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Tested-by: Philippe Mathieu-Daudé <philmd@linaro.org>

[PATCH v3 5/6] include/generic/host: Fix atomic128-cas.h.inc for Int128 structure

[Richard Henderson]

Use the Int128Alias structure more when we need to convert
between Int128 and __int128_t, when Int128 is a struct.

Fixes the build on aarch64 host with TCI, which forces
the use of the struct.

Reported-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 host/include/generic/host/atomic128-cas.h.inc | 24 ++++++++++++-------
 1 file changed, 15 insertions(+), 9 deletions(-)

diff --git a/host/include/generic/host/atomic128-cas.h.inc b/host/include/generic/host/atomic128-cas.h.inc
index 990162c56f..8bf5f47768 100644
--- a/host/include/generic/host/atomic128-cas.h.inc
+++ b/host/include/generic/host/atomic128-cas.h.inc
@@ -34,39 +34,45 @@ static inline Int128 ATTRIBUTE_ATOMIC128_OPT
 atomic16_xchg(Int128 *ptr, Int128 new)
 {
     __int128_t *ptr_align = __builtin_assume_aligned(ptr, 16);
-    Int128 old = *ptr_align;
+    Int128Alias o, n;
 
-    while (!__atomic_compare_exchange_n(ptr_align, &old, new, true,
+    n.s = new;
+    o.i = *ptr_align;
+    while (!__atomic_compare_exchange_n(ptr_align, &o.i, n.i, true,
                                         __ATOMIC_SEQ_CST, 0)) {
         continue;
     }
-    return old;
+    return o.s;
 }
 
 static inline Int128 ATTRIBUTE_ATOMIC128_OPT
 atomic16_fetch_and(Int128 *ptr, Int128 val)
 {
     __int128_t *ptr_align = __builtin_assume_aligned(ptr, 16);
-    Int128 old = *ptr_align;
+    Int128Alias o, v;
 
-    while (!__atomic_compare_exchange_n(ptr_align, &old, old & val, true,
+    v.s = val;
+    o.i = *ptr_align;
+    while (!__atomic_compare_exchange_n(ptr_align, &o.i, o.i & v.i, true,
                                         __ATOMIC_SEQ_CST, 0)) {
         continue;
     }
-    return old;
+    return o.s;
 }
 
 static inline Int128 ATTRIBUTE_ATOMIC128_OPT
 atomic16_fetch_or(Int128 *ptr, Int128 val)
 {
     __int128_t *ptr_align = __builtin_assume_aligned(ptr, 16);
-    Int128 old = *ptr_align;
+    Int128Alias o, v;
 
-    while (!__atomic_compare_exchange_n(ptr_align, &old, old | val, true,
+    v.s = val;
+    o.i = *ptr_align;
+    while (!__atomic_compare_exchange_n(ptr_align, &o.i, o.i | v.i, true,
                                         __ATOMIC_SEQ_CST, 0)) {
         continue;
     }
-    return old;
+    return o.s;
 }
 # define HAVE_CMPXCHG128 1
 #elif defined(CONFIG_CMPXCHG128)
-- 
2.43.0

Re: [PATCH v3 5/6] include/generic/host: Fix atomic128-cas.h.inc for Int128 structure

[Philippe Mathieu-Daudé]

On 3/12/25 17:14, Richard Henderson wrote:
> Use the Int128Alias structure more when we need to convert
> between Int128 and __int128_t, when Int128 is a struct.
> 
> Fixes the build on aarch64 host with TCI, which forces
> the use of the struct.
> 
> Reported-by: Philippe Mathieu-Daudé <philmd@linaro.org>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>   host/include/generic/host/atomic128-cas.h.inc | 24 ++++++++++++-------
>   1 file changed, 15 insertions(+), 9 deletions(-)

Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>

[PATCH v3 6/6] include/aarch64/host: Fix atomic16_fetch_{and,or}

[Richard Henderson]

The tmp[lh] variables were defined as inputs to the
asm rather than outputs, which meant that the compiler
rightly diagnosed uninitialized inputs.

Reported-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 host/include/aarch64/host/atomic128-cas.h.inc | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/host/include/aarch64/host/atomic128-cas.h.inc b/host/include/aarch64/host/atomic128-cas.h.inc
index aec27df182..52e98a0bdd 100644
--- a/host/include/aarch64/host/atomic128-cas.h.inc
+++ b/host/include/aarch64/host/atomic128-cas.h.inc
@@ -67,9 +67,9 @@ static inline Int128 atomic16_fetch_and(Int128 *ptr, Int128 new)
         "stlxp %w[tmp], %[tmpl], %[tmph], %[mem]\n\t"
         "cbnz %w[tmp], 0b"
         : [mem] "+m"(*ptr), [tmp] "=&r"(tmp),
-          [oldl] "=&r"(oldl), [oldh] "=&r"(oldh)
-        : [newl] "r"(newl), [newh] "r"(newh),
-          [tmpl] "r"(tmpl), [tmph] "r"(tmph)
+          [oldl] "=&r"(oldl), [oldh] "=&r"(oldh),
+          [tmpl] "=&r"(tmpl), [tmph] "=&r"(tmph)
+        : [newl] "r"(newl), [newh] "r"(newh)
         : "memory");
 
     return int128_make128(oldl, oldh);
@@ -87,9 +87,9 @@ static inline Int128 atomic16_fetch_or(Int128 *ptr, Int128 new)
         "stlxp %w[tmp], %[tmpl], %[tmph], %[mem]\n\t"
         "cbnz %w[tmp], 0b"
         : [mem] "+m"(*ptr), [tmp] "=&r"(tmp),
-          [oldl] "=&r"(oldl), [oldh] "=&r"(oldh)
-        : [newl] "r"(newl), [newh] "r"(newh),
-          [tmpl] "r"(tmpl), [tmph] "r"(tmph)
+          [oldl] "=&r"(oldl), [oldh] "=&r"(oldh),
+          [tmpl] "=&r"(tmpl), [tmph] "=&r"(tmph)
+        : [newl] "r"(newl), [newh] "r"(newh)
         : "memory");
 
     return int128_make128(oldl, oldh);
-- 
2.43.0

Re: [PATCH v3 6/6] include/aarch64/host: Fix atomic16_fetch_{and,or}

[Philippe Mathieu-Daudé]

On 3/12/25 17:14, Richard Henderson wrote:
> The tmp[lh] variables were defined as inputs to the
> asm rather than outputs, which meant that the compiler
> rightly diagnosed uninitialized inputs.
> 
> Reported-by: Philippe Mathieu-Daudé <philmd@linaro.org>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>   host/include/aarch64/host/atomic128-cas.h.inc | 12 ++++++------
>   1 file changed, 6 insertions(+), 6 deletions(-)

Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Tested-by: Philippe Mathieu-Daudé <philmd@linaro.org>