]> git.proxmox.com Git - mirror_qemu.git/commitdiff
Merge remote-tracking branch 'remotes/rth/tags/tcg-next-20140422' into staging
authorPeter Maydell <peter.maydell@linaro.org>
Thu, 24 Apr 2014 14:24:51 +0000 (15:24 +0100)
committerPeter Maydell <peter.maydell@linaro.org>
Thu, 24 Apr 2014 14:24:52 +0000 (15:24 +0100)
Pull tcg 2014-04-22

# gpg: Signature made Tue 22 Apr 2014 22:00:04 BST using RSA key ID 4DD0279B
# gpg: Can't check signature: public key not found

* remotes/rth/tags/tcg-next-20140422:
  tcg: Use HOST_WORDS_BIGENDIAN
  tcg: Fix fallback from muls2_i64 to mulu2_i64
  tcg: Use tcg_gen_mulu2_i32 in tcg_gen_muls2_i32
  tcg: Relax requirement for mulu2_i32 on 32-bit hosts
  tcg-s390: Remove W constraint
  tcg-sparc: Use the type parameter to tcg_target_const_match
  tcg-ppc64: Use the type parameter to tcg_target_const_match
  tcg-aarch64: Remove w constraint
  tcg: Add TCGType parameter to tcg_target_const_match
  tcg: Fix out of range shift in deposit optimizations
  tci: Mask shift counts to avoid undefined behavior
  tcg: Mask shift quantities while folding
  tcg: Use "unspecified behavior" for shifts
  tcg: Fix warning (1 bit signed bitfield entry) and replace int by bool

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
14 files changed:
configure
hw/i2c/smbus_eeprom.c
hw/ide/ahci.c
include/qemu/int128.h
net/net.c
qemu-doc.texi
qemu-options.hx
qga/commands-posix.c
qga/qapi-schema.json
scripts/coverity-model.c [new file with mode: 0644]
tcg/ia64/tcg-target.c
tcg/ia64/tcg-target.h
vl.c
xbzrle.c

index 69b9f56847469f80cb3afc31d21f707ed64d0122..b08afc3fb8cbc3026e8491532ad02e7a773b8728 100755 (executable)
--- a/configure
+++ b/configure
@@ -1217,8 +1217,8 @@ Advanced options (experts only):
   --enable-modules         enable modules support
   --enable-debug-tcg       enable TCG debugging
   --disable-debug-tcg      disable TCG debugging (default)
-  --enable-debug-info       enable debugging information (default)
-  --disable-debug-info      disable debugging information
+  --enable-debug-info      enable debugging information (default)
+  --disable-debug-info     disable debugging information
   --enable-debug           enable common debug build options
   --enable-sparse          enable sparse checker
   --disable-sparse         disable sparse checker (default)
@@ -4095,7 +4095,6 @@ echo "sparse enabled    $sparse"
 echo "strip binaries    $strip_opt"
 echo "profiler          $profiler"
 echo "static build      $static"
-echo "-Werror enabled   $werror"
 if test "$darwin" = "yes" ; then
     echo "Cocoa support     $cocoa"
 fi
index 86f35c11de7c6426375a517d3b4eb6dcf8c6e86a..72c09cba6b657fd5681d58a8a550f67534779dd3 100644 (file)
@@ -71,7 +71,7 @@ static void eeprom_write_data(SMBusDevice *dev, uint8_t cmd, uint8_t *buf, int l
     printf("eeprom_write_byte: addr=0x%02x cmd=0x%02x val=0x%02x\n",
            dev->i2c.address, cmd, buf[0]);
 #endif
-    /* An page write operation is not a valid SMBus command.
+    /* A page write operation is not a valid SMBus command.
        It is a block write without a length byte.  Fortunately we
        get the full block anyway.  */
     /* TODO: Should this set the current location?  */
index bfe633f3a5e3bf6824f5c88cbdca5c571f513d01..50327ffdf127e943af5e70da33a5921c0e56714f 100644 (file)
@@ -438,9 +438,9 @@ static void check_cmd(AHCIState *s, int port)
 
     if ((pr->cmd & PORT_CMD_START) && pr->cmd_issue) {
         for (slot = 0; (slot < 32) && pr->cmd_issue; slot++) {
-            if ((pr->cmd_issue & (1 << slot)) &&
+            if ((pr->cmd_issue & (1U << slot)) &&
                 !handle_cmd(s, port, slot)) {
-                pr->cmd_issue &= ~(1 << slot);
+                pr->cmd_issue &= ~(1U << slot);
             }
         }
     }
index 9ed47aafd317911176ffdbfdb0f259c731d53517..f59703143ae6c45db90658c8fa103e89da63ad86 100644 (file)
@@ -53,7 +53,7 @@ static inline Int128 int128_rshift(Int128 a, int n)
     if (n >= 64) {
         return (Int128) { h, h >> 63 };
     } else {
-        return (Int128) { (a.lo >> n) | (a.hi << (64 - n)), h };
+        return (Int128) { (a.lo >> n) | ((uint64_t)a.hi << (64 - n)), h };
     }
 }
 
@@ -78,7 +78,7 @@ static inline Int128 int128_neg(Int128 a)
 
 static inline Int128 int128_sub(Int128 a, Int128 b)
 {
-    return (Int128){ a.lo - b.lo, a.hi - b.hi - (a.lo < b.lo) };
+    return (Int128){ a.lo - b.lo, (uint64_t)a.hi - b.hi - (a.lo < b.lo) };
 }
 
 static inline bool int128_nonneg(Int128 a)
index e3ef1e4f1d1c6d0cdc290182fd495cfbe4e0ff93..a4aadffc111553c22970d1d6337046694815856e 100644 (file)
--- a/net/net.c
+++ b/net/net.c
@@ -952,10 +952,12 @@ void net_host_device_remove(Monitor *mon, const QDict *qdict)
 
     nc = net_hub_find_client_by_name(vlan_id, device);
     if (!nc) {
+        error_report("Host network device '%s' on hub '%d' not found",
+                     device, vlan_id);
         return;
     }
     if (!net_host_check_device(nc->model)) {
-        monitor_printf(mon, "invalid host network device %s\n", device);
+        error_report("invalid host network device '%s'", device);
         return;
     }
     qemu_del_net_client(nc);
index e6e20ebbd315bd084b7cf05954abf583fe827d13..88ec9bb1331c2d311cc8dbb1583a517c9a92b7ec 100644 (file)
@@ -823,7 +823,7 @@ In this case, the block device must be exported using qemu-nbd:
 qemu-nbd --socket=/tmp/my_socket my_disk.qcow2
 @end example
 
-The use of qemu-nbd allows to share a disk between several guests:
+The use of qemu-nbd allows sharing of a disk between several guests:
 @example
 qemu-nbd --socket=/tmp/my_socket --share=2 my_disk.qcow2
 @end example
index 2d33815fb9b80653440372a4e6c7ea45f8966dc8..6457034b8c25aee253f2446c64735f3088b2241e 100644 (file)
@@ -444,7 +444,8 @@ This option defines the type of the media: disk or cdrom.
 @item cyls=@var{c},heads=@var{h},secs=@var{s}[,trans=@var{t}]
 These options have the same definition as they have in @option{-hdachs}.
 @item snapshot=@var{snapshot}
-@var{snapshot} is "on" or "off" and allows to enable snapshot for given drive (see @option{-snapshot}).
+@var{snapshot} is "on" or "off" and controls snapshot mode for the given drive
+(see @option{-snapshot}).
 @item cache=@var{cache}
 @var{cache} is "none", "writeback", "unsafe", "directsync" or "writethrough" and controls how the host cache is used to access block data.
 @item aio=@var{aio}
@@ -1242,7 +1243,7 @@ Disable adaptive encodings. Adaptive encodings are enabled by default.
 An adaptive encoding will try to detect frequently updated screen regions,
 and send updates in these regions using a lossy encoding (like JPEG).
 This can be really helpful to save bandwidth when playing videos. Disabling
-adaptive encodings allows to restore the original static behavior of encodings
+adaptive encodings restores the original static behavior of encodings
 like Tight.
 
 @item share=[allow-exclusive|force-shared|ignore]
@@ -2805,7 +2806,7 @@ UTC or local time, respectively. @code{localtime} is required for correct date i
 MS-DOS or Windows. To start at a specific point in time, provide @var{date} in the
 format @code{2006-06-17T16:01:21} or @code{2006-06-17}. The default base is UTC.
 
-By default the RTC is driven by the host system time. This allows to use the
+By default the RTC is driven by the host system time. This allows using of the
 RTC as accurate reference clock inside the guest, specifically if the host
 time is smoothly following an accurate external reference clock, e.g. via NTP.
 If you want to isolate the guest time from the host, you can set @option{clock}
index 6b5f11f83f9bcb140fc5f0d42f1f7ad477654f71..935a4ec14df3951494bf853fefd4c975c64d8d21 100644 (file)
@@ -171,7 +171,7 @@ void qmp_guest_set_time(bool has_time, int64_t time_ns, Error **errp)
     /* Now, if user has passed a time to set and the system time is set, we
      * just need to synchronize the hardware clock. However, if no time was
      * passed, user is requesting the opposite: set the system time from the
-     * hardware clock. */
+     * hardware clock (RTC). */
     pid = fork();
     if (pid == 0) {
         setsid();
index 80edca143aefa93cbdfaa10ad493a15e50703694..a8cdcb35c4f603e0645a202816aa8600e165d374 100644 (file)
@@ -96,8 +96,8 @@
 ##
 # @guest-get-time:
 #
-# Get the information about guest time relative to the Epoch
-# of 1970-01-01 in UTC.
+# Get the information about guest's System Time relative to
+# the Epoch of 1970-01-01 in UTC.
 #
 # Returns: Time in nanoseconds.
 #
 # gap was, NTP might not be able to resynchronize the
 # guest.
 #
-# This command tries to set guest time to the given value,
-# then sets the Hardware Clock to the current System Time.
-# This will make it easier for a guest to resynchronize
-# without waiting for NTP. If no @time is specified, then
-# the time to set is read from RTC.
+# This command tries to set guest's System Time to the
+# given value, then sets the Hardware Clock (RTC) to the
+# current System Time. This will make it easier for a guest
+# to resynchronize without waiting for NTP. If no @time is
+# specified, then the time to set is read from RTC.
 #
 # @time: #optional time of nanoseconds, relative to the Epoch
 #        of 1970-01-01 in UTC.
diff --git a/scripts/coverity-model.c b/scripts/coverity-model.c
new file mode 100644 (file)
index 0000000..4c99a85
--- /dev/null
@@ -0,0 +1,183 @@
+/* Coverity Scan model
+ *
+ * Copyright (C) 2014 Red Hat, Inc.
+ *
+ * Authors:
+ *  Markus Armbruster <armbru@redhat.com>
+ *  Paolo Bonzini <pbonzini@redhat.com>
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or, at your
+ * option, any later version.  See the COPYING file in the top-level directory.
+ */
+
+
+/*
+ * This is the source code for our Coverity user model file.  The
+ * purpose of user models is to increase scanning accuracy by explaining
+ * code Coverity can't see (out of tree libraries) or doesn't
+ * sufficiently understand.  Better accuracy means both fewer false
+ * positives and more true defects.  Memory leaks in particular.
+ *
+ * - A model file can't import any header files.  Some built-in primitives are
+ *   available but not wchar_t, NULL etc.
+ * - Modeling doesn't need full structs and typedefs. Rudimentary structs
+ *   and similar types are sufficient.
+ * - An uninitialized local variable signifies that the variable could be
+ *   any value.
+ *
+ * The model file must be uploaded by an admin in the analysis settings of
+ * http://scan.coverity.com/projects/378
+ */
+
+#define NULL ((void *)0)
+
+typedef unsigned char uint8_t;
+typedef char int8_t;
+typedef unsigned int uint32_t;
+typedef int int32_t;
+typedef long ssize_t;
+typedef unsigned long long uint64_t;
+typedef long long int64_t;
+typedef _Bool bool;
+
+/* exec.c */
+
+typedef struct AddressSpace AddressSpace;
+typedef uint64_t hwaddr;
+
+static void __write(uint8_t *buf, ssize_t len)
+{
+    int first, last;
+    __coverity_negative_sink__(len);
+    if (len == 0) return;
+    buf[0] = first;
+    buf[len-1] = last;
+    __coverity_writeall__(buf);
+}
+
+static void __read(uint8_t *buf, ssize_t len)
+{
+    __coverity_negative_sink__(len);
+    if (len == 0) return;
+    int first = buf[0];
+    int last = buf[len-1];
+}
+
+bool address_space_rw(AddressSpace *as, hwaddr addr, uint8_t *buf,
+                      int len, bool is_write)
+{
+    bool result;
+
+    // TODO: investigate impact of treating reads as producing
+    // tainted data, with __coverity_tainted_data_argument__(buf).
+    if (is_write) __write(buf, len); else __read(buf, len);
+
+    return result;
+}
+
+/* Tainting */
+
+typedef struct {} name2keysym_t;
+static int get_keysym(const name2keysym_t *table,
+                      const char *name)
+{
+    int result;
+    if (result > 0) {
+        __coverity_tainted_string_sanitize_content__(name);
+        return result;
+    } else {
+        return 0;
+    }
+}
+
+/* glib memory allocation functions.
+ *
+ * Note that we ignore the fact that g_malloc of 0 bytes returns NULL,
+ * and g_realloc of 0 bytes frees the pointer.
+ *
+ * Modeling this would result in Coverity flagging a lot of memory
+ * allocations as potentially returning NULL, and asking us to check
+ * whether the result of the allocation is NULL or not.  However, the
+ * resulting pointer should never be dereferenced anyway, and in fact
+ * it is not in the vast majority of cases.
+ *
+ * If a dereference did happen, this would suppress a defect report
+ * for an actual null pointer dereference.  But it's too unlikely to
+ * be worth wading through the false positives, and with some luck
+ * we'll get a buffer overflow reported anyway.
+ */
+
+void *malloc(size_t);
+void *calloc(size_t, size_t);
+void *realloc(void *, size_t);
+void free(void *);
+
+void *
+g_malloc(size_t n_bytes)
+{
+    void *mem;
+    __coverity_negative_sink__(n_bytes);
+    mem = malloc(n_bytes == 0 ? 1 : n_bytes);
+    if (!mem) __coverity_panic__();
+    return mem;
+}
+
+void *
+g_malloc0(size_t n_bytes)
+{
+    void *mem;
+    __coverity_negative_sink__(n_bytes);
+    mem = calloc(1, n_bytes == 0 ? 1 : n_bytes);
+    if (!mem) __coverity_panic__();
+    return mem;
+}
+
+void g_free(void *mem)
+{
+    free(mem);
+}
+
+void *g_realloc(void * mem, size_t n_bytes)
+{
+    __coverity_negative_sink__(n_bytes);
+    mem = realloc(mem, n_bytes == 0 ? 1 : n_bytes);
+    if (!mem) __coverity_panic__();
+    return mem;
+}
+
+void *g_try_malloc(size_t n_bytes)
+{
+    __coverity_negative_sink__(n_bytes);
+    return malloc(n_bytes == 0 ? 1 : n_bytes);
+}
+
+void *g_try_malloc0(size_t n_bytes)
+{
+    __coverity_negative_sink__(n_bytes);
+    return calloc(1, n_bytes == 0 ? 1 : n_bytes);
+}
+
+void *g_try_realloc(void *mem, size_t n_bytes)
+{
+    __coverity_negative_sink__(n_bytes);
+    return realloc(mem, n_bytes == 0 ? 1 : n_bytes);
+}
+
+/* Other glib functions */
+
+typedef struct _GIOChannel GIOChannel;
+GIOChannel *g_io_channel_unix_new(int fd)
+{
+    GIOChannel *c = g_malloc0(sizeof(GIOChannel));
+    __coverity_escape__(fd);
+    return c;
+}
+
+void g_assertion_message_expr(const char     *domain,
+                              const char     *file,
+                              int             line,
+                              const char     *func,
+                              const char     *expr)
+{
+    __coverity_panic__();
+}
index c640184e86173f47dd4f6fa359a53c8643ae089f..1f523d64661ce128b47e034b0a8d1b57f0810a52 100644 (file)
@@ -23,8 +23,6 @@
  * THE SOFTWARE.
  */
 
-#include "tcg-be-null.h"
-
 /*
  * Register definitions
  */
@@ -221,10 +219,12 @@ enum {
     OPC_ALLOC_M34             = 0x02c00000000ull,
     OPC_BR_DPTK_FEW_B1        = 0x08400000000ull,
     OPC_BR_SPTK_MANY_B1       = 0x08000001000ull,
+    OPC_BR_CALL_SPNT_FEW_B3   = 0x0a200000000ull,
     OPC_BR_SPTK_MANY_B4       = 0x00100001000ull,
     OPC_BR_CALL_SPTK_MANY_B5  = 0x02100001000ull,
     OPC_BR_RET_SPTK_MANY_B4   = 0x00108001100ull,
     OPC_BRL_SPTK_MANY_X3      = 0x18000001000ull,
+    OPC_BRL_CALL_SPNT_MANY_X4 = 0x1a200001000ull,
     OPC_BRL_CALL_SPTK_MANY_X4 = 0x1a000001000ull,
     OPC_CMP_LT_A6             = 0x18000000000ull,
     OPC_CMP_LTU_A6            = 0x1a000000000ull,
@@ -356,6 +356,15 @@ static inline uint64_t tcg_opc_b1(int qp, uint64_t opc, uint64_t imm)
            | (qp & 0x3f);
 }
 
+static inline uint64_t tcg_opc_b3(int qp, uint64_t opc, int b1, uint64_t imm)
+{
+    return opc
+           | ((imm & 0x100000) << 16) /* s */
+           | ((imm & 0x0fffff) << 13) /* imm20b */
+           | ((b1 & 0x7) << 6)
+           | (qp & 0x3f);
+}
+
 static inline uint64_t tcg_opc_b4(int qp, uint64_t opc, int b2)
 {
     return opc
@@ -815,6 +824,7 @@ static int target_parse_constraint(TCGArgConstraint *ct, const char **pct_str)
 #if defined(CONFIG_SOFTMMU)
         tcg_regset_reset_reg(ct->u.regs, TCG_REG_R56);
         tcg_regset_reset_reg(ct->u.regs, TCG_REG_R57);
+        tcg_regset_reset_reg(ct->u.regs, TCG_REG_R58);
 #endif
         break;
     case 'Z':
@@ -950,15 +960,21 @@ static inline void tcg_out_callr(TCGContext *s, TCGReg addr)
 static void tcg_out_exit_tb(TCGContext *s, tcg_target_long arg)
 {
     int64_t disp;
-    uint64_t imm;
+    uint64_t imm, opc1;
 
-    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R8, arg);
+    /* At least arg == 0 is a common operation.  */
+    if (arg == sextract64(arg, 0, 22)) {
+        opc1 = tcg_opc_movi_a(TCG_REG_P0, TCG_REG_R8, arg);
+    } else {
+        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R8, arg);
+        opc1 = INSN_NOP_M;
+    }
 
     disp = tb_ret_addr - s->code_ptr;
     imm = (uint64_t)disp >> 4;
 
     tcg_out_bundle(s, mLX,
-                   INSN_NOP_M,
+                   opc1,
                    tcg_opc_l3 (imm),
                    tcg_opc_x3 (TCG_REG_P0, OPC_BRL_SPTK_MANY_X3, imm));
 }
@@ -1558,238 +1574,284 @@ static inline void tcg_out_movcond(TCGContext *s, TCGCond cond, TCGArg ret,
 }
 
 #if defined(CONFIG_SOFTMMU)
+/* We're expecting to use an signed 22-bit immediate add.  */
+QEMU_BUILD_BUG_ON(offsetof(CPUArchState, tlb_table[NB_MMU_MODES - 1][1])
+                  > 0x1fffff)
+
 /* Load and compare a TLB entry, and return the result in (p6, p7).
-   R2 is loaded with the address of the addend TLB entry.
-   R57 is loaded with the address, zero extented on 32-bit targets. */
-static inline void tcg_out_qemu_tlb(TCGContext *s, TCGArg addr_reg,
-                                    TCGMemOp s_bits, uint64_t offset_rw,
-                                    uint64_t offset_addend)
-{
-    tcg_out_bundle(s, mII,
-                   INSN_NOP_M,
-                   tcg_opc_i11(TCG_REG_P0, OPC_EXTR_U_I11, TCG_REG_R2,
+   R2 is loaded with the addend TLB entry.
+   R57 is loaded with the address, zero extented on 32-bit targets.
+   R1, R3 are clobbered, leaving R56 free for...
+   BSWAP_1, BSWAP_2 and I-slot insns for swapping data for store.  */
+static inline void tcg_out_qemu_tlb(TCGContext *s, TCGReg addr_reg,
+                                    TCGMemOp s_bits, int off_rw, int off_add,
+                                    uint64_t bswap1, uint64_t bswap2)
+{
+     /*
+        .mii
+        mov    r2 = off_rw
+        extr.u r3 = addr_reg, ...              # extract tlb page
+        zxt4   r57 = addr_reg                  # or mov for 64-bit guest
+        ;;
+        .mii
+        addl   r2 = r2, areg0
+        shl    r3 = r3, cteb                   # via dep.z
+        dep    r1 = 0, r57, ...                # zero page ofs, keep align
+        ;;
+        .mmi
+        add    r2 = r2, r3
+        ;;
+        ld4    r3 = [r2], off_add-off_rw       # or ld8 for 64-bit guest
+        nop
+        ;;
+        .mmi
+        nop
+        cmp.eq p6, p7 = r3, r58
+        nop
+        ;;
+      */
+    tcg_out_bundle(s, miI,
+                   tcg_opc_movi_a(TCG_REG_P0, TCG_REG_R2, off_rw),
+                   tcg_opc_i11(TCG_REG_P0, OPC_EXTR_U_I11, TCG_REG_R3,
                                addr_reg, TARGET_PAGE_BITS, CPU_TLB_BITS - 1),
-                   tcg_opc_i12(TCG_REG_P0, OPC_DEP_Z_I12, TCG_REG_R2,
-                               TCG_REG_R2, 63 - CPU_TLB_ENTRY_BITS,
-                               63 - CPU_TLB_ENTRY_BITS));
-    tcg_out_bundle(s, mII,
-                   tcg_opc_a5 (TCG_REG_P0, OPC_ADDL_A5, TCG_REG_R2,
-                               offset_rw, TCG_REG_R2),
                    tcg_opc_ext_i(TCG_REG_P0,
                                  TARGET_LONG_BITS == 32 ? MO_UL : MO_Q,
-                                 TCG_REG_R57, addr_reg),
+                                 TCG_REG_R57, addr_reg));
+    tcg_out_bundle(s, miI,
                    tcg_opc_a1 (TCG_REG_P0, OPC_ADD_A1, TCG_REG_R2,
-                               TCG_REG_R2, TCG_AREG0));
-    tcg_out_bundle(s, mII,
+                               TCG_REG_R2, TCG_AREG0),
+                   tcg_opc_i12(TCG_REG_P0, OPC_DEP_Z_I12, TCG_REG_R3,
+                               TCG_REG_R3, 63 - CPU_TLB_ENTRY_BITS,
+                               63 - CPU_TLB_ENTRY_BITS),
+                   tcg_opc_i14(TCG_REG_P0, OPC_DEP_I14, TCG_REG_R1, 0,
+                               TCG_REG_R57, 63 - s_bits,
+                               TARGET_PAGE_BITS - s_bits - 1));
+    tcg_out_bundle(s, MmI,
+                   tcg_opc_a1 (TCG_REG_P0, OPC_ADD_A1,
+                               TCG_REG_R2, TCG_REG_R2, TCG_REG_R3),
                    tcg_opc_m3 (TCG_REG_P0,
                                (TARGET_LONG_BITS == 32
-                                ? OPC_LD4_M3 : OPC_LD8_M3), TCG_REG_R56,
-                               TCG_REG_R2, offset_addend - offset_rw),
-                   tcg_opc_i14(TCG_REG_P0, OPC_DEP_I14, TCG_REG_R3, 0,
-                               TCG_REG_R57, 63 - s_bits,
-                               TARGET_PAGE_BITS - s_bits - 1),
+                                ? OPC_LD4_M3 : OPC_LD8_M3), TCG_REG_R3,
+                               TCG_REG_R2, off_add - off_rw),
+                   bswap1);
+    tcg_out_bundle(s, mmI,
+                   tcg_opc_m1 (TCG_REG_P0, OPC_LD8_M1, TCG_REG_R2, TCG_REG_R2),
                    tcg_opc_a6 (TCG_REG_P0, OPC_CMP_EQ_A6, TCG_REG_P6,
-                               TCG_REG_P7, TCG_REG_R3, TCG_REG_R56));
+                               TCG_REG_P7, TCG_REG_R1, TCG_REG_R3),
+                   bswap2);
 }
 
-/* helper signature: helper_ld_mmu(CPUState *env, target_ulong addr,
-   int mmu_idx) */
-static const void * const qemu_ld_helpers[4] = {
-    helper_ldb_mmu,
-    helper_ldw_mmu,
-    helper_ldl_mmu,
-    helper_ldq_mmu,
-};
+#define TCG_MAX_QEMU_LDST       640
 
-static inline void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args,
-                                   TCGMemOp opc)
+typedef struct TCGLabelQemuLdst {
+    bool is_ld;
+    TCGMemOp size;
+    uint8_t *label_ptr;     /* label pointers to be updated */
+} TCGLabelQemuLdst;
+
+typedef struct TCGBackendData {
+    int nb_ldst_labels;
+    TCGLabelQemuLdst ldst_labels[TCG_MAX_QEMU_LDST];
+} TCGBackendData;
+
+static inline void tcg_out_tb_init(TCGContext *s)
+{
+    s->be->nb_ldst_labels = 0;
+}
+
+static void add_qemu_ldst_label(TCGContext *s, bool is_ld, TCGMemOp opc,
+                                uint8_t *label_ptr)
+{
+    TCGBackendData *be = s->be;
+    TCGLabelQemuLdst *l = &be->ldst_labels[be->nb_ldst_labels++];
+
+    assert(be->nb_ldst_labels <= TCG_MAX_QEMU_LDST);
+    l->is_ld = is_ld;
+    l->size = opc & MO_SIZE;
+    l->label_ptr = label_ptr;
+}
+
+static void tcg_out_tb_finalize(TCGContext *s)
+{
+    static const void * const helpers[8] = {
+        helper_ret_stb_mmu,
+        helper_le_stw_mmu,
+        helper_le_stl_mmu,
+        helper_le_stq_mmu,
+        helper_ret_ldub_mmu,
+        helper_le_lduw_mmu,
+        helper_le_ldul_mmu,
+        helper_le_ldq_mmu,
+    };
+    uintptr_t thunks[8] = { };
+    TCGBackendData *be = s->be;
+    size_t i, n = be->nb_ldst_labels;
+
+    for (i = 0; i < n; i++) {
+        TCGLabelQemuLdst *l = &be->ldst_labels[i];
+        long x = l->is_ld * 4 + l->size;
+        uintptr_t dest = thunks[x];
+
+        /* The out-of-line thunks are all the same; load the return address
+           from B0, load the GP, and branch to the code.  Note that we are
+           always post-call, so the register window has rolled, so we're
+           using incomming parameter register numbers, not outgoing.  */
+        if (dest == 0) {
+            uintptr_t disp, *desc = (uintptr_t *)helpers[x];
+
+            thunks[x] = dest = (uintptr_t)s->code_ptr;
+
+            tcg_out_bundle(s, mlx,
+                           INSN_NOP_M,
+                           tcg_opc_l2 (desc[1]),
+                           tcg_opc_x2 (TCG_REG_P0, OPC_MOVL_X2,
+                                       TCG_REG_R1, desc[1]));
+            tcg_out_bundle(s, mii,
+                           INSN_NOP_M,
+                           INSN_NOP_I,
+                           tcg_opc_i22(TCG_REG_P0, OPC_MOV_I22,
+                                       l->is_ld ? TCG_REG_R35 : TCG_REG_R36,
+                                       TCG_REG_B0));
+            disp = (desc[0] - (uintptr_t)s->code_ptr) >> 4;
+            tcg_out_bundle(s, mLX,
+                           INSN_NOP_M,
+                           tcg_opc_l3 (disp),
+                           tcg_opc_x3 (TCG_REG_P0, OPC_BRL_SPTK_MANY_X3, disp));
+        }
+
+        reloc_pcrel21b(l->label_ptr, dest);
+    }
+}
+
+static inline void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args)
 {
     static const uint64_t opc_ld_m1[4] = {
         OPC_LD1_M1, OPC_LD2_M1, OPC_LD4_M1, OPC_LD8_M1
     };
     int addr_reg, data_reg, mem_index;
-    TCGMemOp s_bits, bswap;
-
-    data_reg = *args++;
-    addr_reg = *args++;
-    mem_index = *args;
+    TCGMemOp opc, s_bits;
+    uint64_t fin1, fin2;
+    uint8_t *label_ptr;
+
+    data_reg = args[0];
+    addr_reg = args[1];
+    opc = args[2];
+    mem_index = args[3];
     s_bits = opc & MO_SIZE;
-    bswap = opc & MO_BSWAP;
 
     /* Read the TLB entry */
     tcg_out_qemu_tlb(s, addr_reg, s_bits,
                      offsetof(CPUArchState, tlb_table[mem_index][0].addr_read),
-                     offsetof(CPUArchState, tlb_table[mem_index][0].addend));
+                     offsetof(CPUArchState, tlb_table[mem_index][0].addend),
+                     INSN_NOP_I, INSN_NOP_I);
 
     /* P6 is the fast path, and P7 the slow path */
-    tcg_out_bundle(s, mLX,
-                   tcg_opc_mov_a(TCG_REG_P7, TCG_REG_R56, TCG_AREG0),
-                   tcg_opc_l2 ((tcg_target_long) qemu_ld_helpers[s_bits]),
-                   tcg_opc_x2 (TCG_REG_P7, OPC_MOVL_X2, TCG_REG_R2,
-                               (tcg_target_long) qemu_ld_helpers[s_bits]));
-    tcg_out_bundle(s, MmI,
-                   tcg_opc_m3 (TCG_REG_P0, OPC_LD8_M3, TCG_REG_R3,
-                               TCG_REG_R2, 8),
-                   tcg_opc_a1 (TCG_REG_P6, OPC_ADD_A1, TCG_REG_R3,
-                               TCG_REG_R3, TCG_REG_R57),
-                   tcg_opc_i21(TCG_REG_P7, OPC_MOV_I21, TCG_REG_B6,
-                               TCG_REG_R3, 0));
-    if (bswap && s_bits == MO_16) {
-        tcg_out_bundle(s, MmI,
-                       tcg_opc_m1 (TCG_REG_P6, opc_ld_m1[s_bits],
-                                   TCG_REG_R8, TCG_REG_R3),
-                       tcg_opc_m1 (TCG_REG_P7, OPC_LD8_M1, TCG_REG_R1, TCG_REG_R2),
-                       tcg_opc_i12(TCG_REG_P6, OPC_DEP_Z_I12,
-                                   TCG_REG_R8, TCG_REG_R8, 15, 15));
-    } else if (bswap && s_bits == MO_32) {
-        tcg_out_bundle(s, MmI,
-                       tcg_opc_m1 (TCG_REG_P6, opc_ld_m1[s_bits],
-                                   TCG_REG_R8, TCG_REG_R3),
-                       tcg_opc_m1 (TCG_REG_P7, OPC_LD8_M1, TCG_REG_R1, TCG_REG_R2),
-                       tcg_opc_i12(TCG_REG_P6, OPC_DEP_Z_I12,
-                                   TCG_REG_R8, TCG_REG_R8, 31, 31));
-    } else {
-        tcg_out_bundle(s, mmI,
-                       tcg_opc_m1 (TCG_REG_P6, opc_ld_m1[s_bits],
-                                   TCG_REG_R8, TCG_REG_R3),
-                       tcg_opc_m1 (TCG_REG_P7, OPC_LD8_M1, TCG_REG_R1, TCG_REG_R2),
-                       INSN_NOP_I);
-    }
-    if (!bswap) {
-        tcg_out_bundle(s, miB,
-                       tcg_opc_movi_a(TCG_REG_P7, TCG_REG_R58, mem_index),
-                       INSN_NOP_I,
-                       tcg_opc_b5 (TCG_REG_P7, OPC_BR_CALL_SPTK_MANY_B5,
-                                   TCG_REG_B0, TCG_REG_B6));
+
+    fin2 = 0;
+    if (opc & MO_BSWAP) {
+        fin1 = tcg_opc_bswap64_i(TCG_REG_P0, data_reg, TCG_REG_R8);
+        if (s_bits < MO_64) {
+            int shift = 64 - (8 << s_bits);
+            fin2 = (opc & MO_SIGN ? OPC_EXTR_I11 : OPC_EXTR_U_I11);
+            fin2 = tcg_opc_i11(TCG_REG_P0, fin2,
+                               data_reg, data_reg, shift, 63 - shift);
+        }
     } else {
-        tcg_out_bundle(s, miB,
-                       tcg_opc_movi_a(TCG_REG_P7, TCG_REG_R58, mem_index),
-                       tcg_opc_bswap64_i(TCG_REG_P6, TCG_REG_R8, TCG_REG_R8),
-                       tcg_opc_b5 (TCG_REG_P7, OPC_BR_CALL_SPTK_MANY_B5,
-                                   TCG_REG_B0, TCG_REG_B6));
+        fin1 = tcg_opc_ext_i(TCG_REG_P0, opc, data_reg, TCG_REG_R8);
     }
 
-    tcg_out_bundle(s, miI,
-                   INSN_NOP_M,
+    tcg_out_bundle(s, mmI,
+                   tcg_opc_mov_a(TCG_REG_P7, TCG_REG_R56, TCG_AREG0),
+                   tcg_opc_a1 (TCG_REG_P6, OPC_ADD_A1, TCG_REG_R2,
+                               TCG_REG_R2, TCG_REG_R57),
+                   tcg_opc_movi_a(TCG_REG_P7, TCG_REG_R58, mem_index));
+    label_ptr = s->code_ptr + 2;
+    tcg_out_bundle(s, miB,
+                   tcg_opc_m1 (TCG_REG_P6, opc_ld_m1[s_bits],
+                               TCG_REG_R8, TCG_REG_R2),
                    INSN_NOP_I,
-                   tcg_opc_ext_i(TCG_REG_P0, opc, data_reg, TCG_REG_R8));
-}
+                   tcg_opc_b3 (TCG_REG_P7, OPC_BR_CALL_SPNT_FEW_B3, TCG_REG_B0,
+                               get_reloc_pcrel21b(label_ptr)));
 
-/* helper signature: helper_st_mmu(CPUState *env, target_ulong addr,
-   uintxx_t val, int mmu_idx) */
-static const void * const qemu_st_helpers[4] = {
-    helper_stb_mmu,
-    helper_stw_mmu,
-    helper_stl_mmu,
-    helper_stq_mmu,
-};
+    add_qemu_ldst_label(s, 1, opc, label_ptr);
 
-static inline void tcg_out_qemu_st(TCGContext *s, const TCGArg *args,
-                                   TCGMemOp opc)
+    /* Note that we always use LE helper functions, so the bswap insns
+       here for the fast path also apply to the slow path.  */
+    tcg_out_bundle(s, (fin2 ? mII : miI),
+                   INSN_NOP_M,
+                   fin1,
+                   fin2 ? fin2 : INSN_NOP_I);
+}
+
+static inline void tcg_out_qemu_st(TCGContext *s, const TCGArg *args)
 {
     static const uint64_t opc_st_m4[4] = {
         OPC_ST1_M4, OPC_ST2_M4, OPC_ST4_M4, OPC_ST8_M4
     };
-    int addr_reg, data_reg, mem_index;
-    TCGMemOp s_bits;
-
-    data_reg = *args++;
-    addr_reg = *args++;
-    mem_index = *args;
+    TCGReg addr_reg, data_reg;
+    int mem_index;
+    uint64_t pre1, pre2;
+    TCGMemOp opc, s_bits;
+    uint8_t *label_ptr;
+
+    data_reg = args[0];
+    addr_reg = args[1];
+    opc = args[2];
+    mem_index = args[3];
     s_bits = opc & MO_SIZE;
 
+    /* Note that we always use LE helper functions, so the bswap insns
+       that are here for the fast path also apply to the slow path,
+       and move the data into the argument register.  */
+    pre2 = INSN_NOP_I;
+    if (opc & MO_BSWAP) {
+        pre1 = tcg_opc_bswap64_i(TCG_REG_P0, TCG_REG_R58, data_reg);
+        if (s_bits < MO_64) {
+            int shift = 64 - (8 << s_bits);
+            pre2 = tcg_opc_i11(TCG_REG_P0, OPC_EXTR_U_I11,
+                               TCG_REG_R58, TCG_REG_R58, shift, 63 - shift);
+        }
+    } else {
+        /* Just move the data into place for the slow path.  */
+        pre1 = tcg_opc_ext_i(TCG_REG_P0, opc, TCG_REG_R58, data_reg);
+    }
+
     tcg_out_qemu_tlb(s, addr_reg, s_bits,
                      offsetof(CPUArchState, tlb_table[mem_index][0].addr_write),
-                     offsetof(CPUArchState, tlb_table[mem_index][0].addend));
+                     offsetof(CPUArchState, tlb_table[mem_index][0].addend),
+                     pre1, pre2);
 
     /* P6 is the fast path, and P7 the slow path */
-    tcg_out_bundle(s, mLX,
+    tcg_out_bundle(s, mmI,
                    tcg_opc_mov_a(TCG_REG_P7, TCG_REG_R56, TCG_AREG0),
-                   tcg_opc_l2 ((tcg_target_long) qemu_st_helpers[s_bits]),
-                   tcg_opc_x2 (TCG_REG_P7, OPC_MOVL_X2, TCG_REG_R2,
-                               (tcg_target_long) qemu_st_helpers[s_bits]));
-    tcg_out_bundle(s, MmI,
-                   tcg_opc_m3 (TCG_REG_P0, OPC_LD8_M3, TCG_REG_R3,
-                               TCG_REG_R2, 8),
-                   tcg_opc_a1 (TCG_REG_P6, OPC_ADD_A1, TCG_REG_R3,
-                               TCG_REG_R3, TCG_REG_R57),
-                   tcg_opc_i21(TCG_REG_P7, OPC_MOV_I21, TCG_REG_B6,
-                               TCG_REG_R3, 0));
-
-    switch (opc) {
-    case MO_8:
-    case MO_16:
-    case MO_32:
-    case MO_64:
-        tcg_out_bundle(s, mii,
-                       tcg_opc_m1 (TCG_REG_P7, OPC_LD8_M1,
-                                   TCG_REG_R1, TCG_REG_R2),
-                       tcg_opc_mov_a(TCG_REG_P7, TCG_REG_R58, data_reg),
-                       INSN_NOP_I);
-        break;
-
-    case MO_16 | MO_BSWAP:
-        tcg_out_bundle(s, miI,
-                       tcg_opc_m1 (TCG_REG_P7, OPC_LD8_M1,
-                                   TCG_REG_R1, TCG_REG_R2),
-                       INSN_NOP_I,
-                       tcg_opc_i12(TCG_REG_P6, OPC_DEP_Z_I12,
-                                   TCG_REG_R2, data_reg, 15, 15));
-        tcg_out_bundle(s, miI,
-                       tcg_opc_mov_a(TCG_REG_P7, TCG_REG_R58, data_reg),
-                       INSN_NOP_I,
-                       tcg_opc_bswap64_i(TCG_REG_P6, TCG_REG_R2, TCG_REG_R2));
-        data_reg = TCG_REG_R2;
-        break;
-
-    case MO_32 | MO_BSWAP:
-        tcg_out_bundle(s, miI,
-                       tcg_opc_m1 (TCG_REG_P7, OPC_LD8_M1,
-                                   TCG_REG_R1, TCG_REG_R2),
-                       INSN_NOP_I,
-                       tcg_opc_i12(TCG_REG_P6, OPC_DEP_Z_I12,
-                                   TCG_REG_R2, data_reg, 31, 31));
-        tcg_out_bundle(s, miI,
-                       tcg_opc_mov_a(TCG_REG_P7, TCG_REG_R58, data_reg),
-                       INSN_NOP_I,
-                       tcg_opc_bswap64_i(TCG_REG_P6, TCG_REG_R2, TCG_REG_R2));
-        data_reg = TCG_REG_R2;
-        break;
-
-    case MO_64 | MO_BSWAP:
-        tcg_out_bundle(s, miI,
-                       tcg_opc_m1 (TCG_REG_P7, OPC_LD8_M1,
-                                   TCG_REG_R1, TCG_REG_R2),
-                       tcg_opc_mov_a(TCG_REG_P7, TCG_REG_R58, data_reg),
-                       tcg_opc_bswap64_i(TCG_REG_P6, TCG_REG_R2, data_reg));
-        data_reg = TCG_REG_R2;
-        break;
-
-    default:
-        tcg_abort();
-    }
-
+                   tcg_opc_a1 (TCG_REG_P6, OPC_ADD_A1, TCG_REG_R2,
+                               TCG_REG_R2, TCG_REG_R57),
+                   tcg_opc_movi_a(TCG_REG_P7, TCG_REG_R59, mem_index));
+    label_ptr = s->code_ptr + 2;
     tcg_out_bundle(s, miB,
                    tcg_opc_m4 (TCG_REG_P6, opc_st_m4[s_bits],
-                               data_reg, TCG_REG_R3),
-                   tcg_opc_movi_a(TCG_REG_P7, TCG_REG_R59, mem_index),
-                   tcg_opc_b5 (TCG_REG_P7, OPC_BR_CALL_SPTK_MANY_B5,
-                               TCG_REG_B0, TCG_REG_B6));
+                               TCG_REG_R58, TCG_REG_R2),
+                   INSN_NOP_I,
+                   tcg_opc_b3 (TCG_REG_P7, OPC_BR_CALL_SPNT_FEW_B3, TCG_REG_B0,
+                               get_reloc_pcrel21b(label_ptr)));
+
+    add_qemu_ldst_label(s, 0, opc, label_ptr);
 }
 
 #else /* !CONFIG_SOFTMMU */
+# include "tcg-be-null.h"
 
-static inline void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args,
-                                   TCGMemOp opc)
+static inline void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args)
 {
     static uint64_t const opc_ld_m1[4] = {
         OPC_LD1_M1, OPC_LD2_M1, OPC_LD4_M1, OPC_LD8_M1
     };
     int addr_reg, data_reg;
-    TCGMemOp s_bits, bswap;
+    TCGMemOp opc, s_bits, bswap;
 
-    data_reg = *args++;
-    addr_reg = *args++;
+    data_reg = args[0];
+    addr_reg = args[1];
+    opc = args[2];
     s_bits = opc & MO_SIZE;
     bswap = opc & MO_BSWAP;
 
@@ -1900,8 +1962,7 @@ static inline void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args,
 #endif
 }
 
-static inline void tcg_out_qemu_st(TCGContext *s, const TCGArg *args,
-                                   TCGMemOp opc)
+static inline void tcg_out_qemu_st(TCGContext *s, const TCGArg *args)
 {
     static uint64_t const opc_st_m4[4] = {
         OPC_ST1_M4, OPC_ST2_M4, OPC_ST4_M4, OPC_ST8_M4
@@ -1910,10 +1971,11 @@ static inline void tcg_out_qemu_st(TCGContext *s, const TCGArg *args,
 #if TARGET_LONG_BITS == 64
     uint64_t add_guest_base;
 #endif
-    TCGMemOp s_bits, bswap;
+    TCGMemOp opc, s_bits, bswap;
 
-    data_reg = *args++;
-    addr_reg = *args++;
+    data_reg = args[0];
+    addr_reg = args[1];
+    opc = args[2];
     s_bits = opc & MO_SIZE;
     bswap = opc & MO_BSWAP;
 
@@ -2237,40 +2299,17 @@ static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
                         args[3], const_args[3], args[4], const_args[4], 0);
         break;
 
-    case INDEX_op_qemu_ld8u:
-        tcg_out_qemu_ld(s, args, MO_UB);
-        break;
-    case INDEX_op_qemu_ld8s:
-        tcg_out_qemu_ld(s, args, MO_SB);
-        break;
-    case INDEX_op_qemu_ld16u:
-        tcg_out_qemu_ld(s, args, MO_TEUW);
-        break;
-    case INDEX_op_qemu_ld16s:
-        tcg_out_qemu_ld(s, args, MO_TESW);
-        break;
-    case INDEX_op_qemu_ld32:
-    case INDEX_op_qemu_ld32u:
-        tcg_out_qemu_ld(s, args, MO_TEUL);
-        break;
-    case INDEX_op_qemu_ld32s:
-        tcg_out_qemu_ld(s, args, MO_TESL);
-        break;
-    case INDEX_op_qemu_ld64:
-        tcg_out_qemu_ld(s, args, MO_TEQ);
-        break;
-
-    case INDEX_op_qemu_st8:
-        tcg_out_qemu_st(s, args, MO_UB);
+    case INDEX_op_qemu_ld_i32:
+        tcg_out_qemu_ld(s, args);
         break;
-    case INDEX_op_qemu_st16:
-        tcg_out_qemu_st(s, args, MO_TEUW);
+    case INDEX_op_qemu_ld_i64:
+        tcg_out_qemu_ld(s, args);
         break;
-    case INDEX_op_qemu_st32:
-        tcg_out_qemu_st(s, args, MO_TEUL);
+    case INDEX_op_qemu_st_i32:
+        tcg_out_qemu_st(s, args);
         break;
-    case INDEX_op_qemu_st64:
-        tcg_out_qemu_st(s, args, MO_TEQ);
+    case INDEX_op_qemu_st_i64:
+        tcg_out_qemu_st(s, args);
         break;
 
     default:
@@ -2381,19 +2420,10 @@ static const TCGTargetOpDef ia64_op_defs[] = {
     { INDEX_op_deposit_i32, { "r", "rZ", "ri" } },
     { INDEX_op_deposit_i64, { "r", "rZ", "ri" } },
 
-    { INDEX_op_qemu_ld8u, { "r", "r" } },
-    { INDEX_op_qemu_ld8s, { "r", "r" } },
-    { INDEX_op_qemu_ld16u, { "r", "r" } },
-    { INDEX_op_qemu_ld16s, { "r", "r" } },
-    { INDEX_op_qemu_ld32, { "r", "r" } },
-    { INDEX_op_qemu_ld32u, { "r", "r" } },
-    { INDEX_op_qemu_ld32s, { "r", "r" } },
-    { INDEX_op_qemu_ld64, { "r", "r" } },
-
-    { INDEX_op_qemu_st8, { "SZ", "r" } },
-    { INDEX_op_qemu_st16, { "SZ", "r" } },
-    { INDEX_op_qemu_st32, { "SZ", "r" } },
-    { INDEX_op_qemu_st64, { "SZ", "r" } },
+    { INDEX_op_qemu_ld_i32, { "r", "r" } },
+    { INDEX_op_qemu_ld_i64, { "r", "r" } },
+    { INDEX_op_qemu_st_i32, { "SZ", "r" } },
+    { INDEX_op_qemu_st_i64, { "SZ", "r" } },
 
     { -1 },
 };
index 52a939c9466d63d6a1cb2e77e1e63ded73ac5007..09c3ba8fe3f08bb058e77a61d06bd7f12e3b4c05 100644 (file)
@@ -153,7 +153,7 @@ typedef enum {
 #define TCG_TARGET_HAS_mulsh_i32        0
 #define TCG_TARGET_HAS_mulsh_i64        0
 
-#define TCG_TARGET_HAS_new_ldst         0
+#define TCG_TARGET_HAS_new_ldst         1
 
 #define TCG_TARGET_deposit_i32_valid(ofs, len) ((len) <= 16)
 #define TCG_TARGET_deposit_i64_valid(ofs, len) ((len) <= 16)
diff --git a/vl.c b/vl.c
index 9975e5a4e3d5d9a89133816e02647be1a73bd7f1..db9ea90a1d128879f90400e65f14146f38d56648 100644 (file)
--- a/vl.c
+++ b/vl.c
@@ -2740,7 +2740,7 @@ static int configure_accelerator(QEMUMachine *machine)
                 if (!accel_list[i].available()) {
                     printf("%s not supported for this target\n",
                            accel_list[i].name);
-                    continue;
+                    break;
                 }
                 *(accel_list[i].allowed) = true;
                 ret = accel_list[i].init(machine);
index fbcb35d0e33ffaf9ea8aedb4c33a32ff5643dd25..8e220bf25b9a4002675e4532b48b25ec8b47e9d6 100644 (file)
--- a/xbzrle.c
+++ b/xbzrle.c
@@ -28,7 +28,7 @@ int xbzrle_encode_buffer(uint8_t *old_buf, uint8_t *new_buf, int slen,
 {
     uint32_t zrun_len = 0, nzrun_len = 0;
     int d = 0, i = 0;
-    long res, xor;
+    long res;
     uint8_t *nzrun_start = NULL;
 
     g_assert(!(((uintptr_t)old_buf | (uintptr_t)new_buf | slen) %
@@ -93,9 +93,11 @@ int xbzrle_encode_buffer(uint8_t *old_buf, uint8_t *new_buf, int slen,
         /* word at a time for speed, use of 32-bit long okay */
         if (!res) {
             /* truncation to 32-bit long okay */
-            long mask = (long)0x0101010101010101ULL;
+            unsigned long mask = (unsigned long)0x0101010101010101ULL;
             while (i < slen) {
-                xor = *(long *)(old_buf + i) ^ *(long *)(new_buf + i);
+                unsigned long xor;
+                xor = *(unsigned long *)(old_buf + i)
+                    ^ *(unsigned long *)(new_buf + i);
                 if ((xor - mask) & ~xor & (mask << 7)) {
                     /* found the end of an nzrun within the current long */
                     while (old_buf[i] != new_buf[i]) {