diff mbox series

[v2,4/8] disas: Support the Capstone disassembler library

Message ID 20170919150313.10833-5-richard.henderson@linaro.org
State Superseded
Headers show
Series Support the Capstone disassembler | expand

Commit Message

Richard Henderson Sept. 19, 2017, 3:03 p.m. UTC
If configured, prefer this over our rather dated copy of the
GPLv2-only binutils.  This will be especially apparent with
the proposed vector extensions to TCG, as disas/i386.c does
not handle AVX.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

---
 include/disas/bfd.h      |   4 +
 include/disas/capstone.h |  38 ++++++++
 disas.c                  | 219 ++++++++++++++++++++++++++++++++++++++++++++---
 configure                |  26 ++++++
 4 files changed, 274 insertions(+), 13 deletions(-)
 create mode 100644 include/disas/capstone.h

-- 
2.13.5

Comments

Alex Bennée Sept. 20, 2017, 11:17 a.m. UTC | #1
Richard Henderson <richard.henderson@linaro.org> writes:

> If configured, prefer this over our rather dated copy of the

> GPLv2-only binutils.  This will be especially apparent with

> the proposed vector extensions to TCG, as disas/i386.c does

> not handle AVX.

>

> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>


Hmm I get a link failure:

  LINK    tilegx-linux-user/qemu-tilegx
disas.o: In function `cap_disas_start':
/home/alex/lsrc/qemu/qemu.git/disas.c:196: undefined reference to `cs_open'
/home/alex/lsrc/qemu/qemu.git/disas.c:210: undefined reference to `cs_option'
/home/alex/lsrc/qemu/qemu.git/disas.c:214: undefined reference to `cs_malloc'
/home/alex/lsrc/qemu/qemu.git/disas.c:216: undefined reference to `cs_close'
/home/alex/lsrc/qemu/qemu.git/disas.c:206: undefined reference to `cs_option'
disas.o: In function `cap_disas_target':
/home/alex/lsrc/qemu/qemu.git/disas.c:244: undefined reference to `cs_disasm_iter'
/home/alex/lsrc/qemu/qemu.git/disas.c:272: undefined reference to `cs_close'
disas.o: In function `cap_disas_host':
/home/alex/lsrc/qemu/qemu.git/disas.c:292: undefined reference to `cs_disasm_iter'
/home/alex/lsrc/qemu/qemu.git/disas.c:304: undefined reference to `cs_close'
collect2: error: ld returned 1 exit status
Makefile:199: recipe for target 'qemu-tilegx' failed
make[1]: *** [qemu-tilegx] Error 1
make[1]: Target 'all' not remade because of errors.
Makefile:326: recipe for target 'subdir-tilegx-linux-user' failed
make: *** [subdir-tilegx-linux-user] Error 2
  LINK    x86_64-linux-user/qemu-x86_64

I wonder if this is a version thing?

    Package: libcapstone-dev
    Status: install ok installed
    Priority: optional
    Section: libdevel
    Installed-Size: 4079
    Maintainer: Ubuntu Developers <ubuntu-devel-discuss@lists.ubuntu.com>
    Architecture: amd64
    Source: capstone
    Version: 3.0.4-0.2
    Depends: libcapstone3 (= 3.0.4-0.2)
    Description: lightweight multi-architecture disassembly framework - devel files
     Capstone is a lightweight multi-platform, multi-architecture disassembly
     framework.


--
Alex Bennée
Richard Henderson Sept. 20, 2017, 1:04 p.m. UTC | #2
On 09/20/2017 06:17 AM, Alex Bennée wrote:
> 

> Richard Henderson <richard.henderson@linaro.org> writes:

> 

>> If configured, prefer this over our rather dated copy of the

>> GPLv2-only binutils.  This will be especially apparent with

>> the proposed vector extensions to TCG, as disas/i386.c does

>> not handle AVX.

>>

>> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

> 

> Hmm I get a link failure:

> 

>   LINK    tilegx-linux-user/qemu-tilegx

> disas.o: In function `cap_disas_start':

> /home/alex/lsrc/qemu/qemu.git/disas.c:196: undefined reference to `cs_open'


I'm going to assume that this isn't the first executable to be linked, and
therefore it's weirder than what I imagine.

> I wonder if this is a version thing?

> 

>     Package: libcapstone-dev

>     Status: install ok installed

>     Priority: optional

>     Section: libdevel

>     Installed-Size: 4079

>     Maintainer: Ubuntu Developers <ubuntu-devel-discuss@lists.ubuntu.com>

>     Architecture: amd64

>     Source: capstone

>     Version: 3.0.4-0.2


No, I've got capstone-devel-3.0.4-8.fc26.x86_64.

The only thing I can imagine is that you didn't start from a fresh build tree
and something went wrong during re-configure.


r~
Richard Henderson Sept. 20, 2017, 9:06 p.m. UTC | #3
On 09/20/2017 06:17 AM, Alex Bennée wrote:
> Hmm I get a link failure:

> 

>   LINK    tilegx-linux-user/qemu-tilegx

> disas.o: In function `cap_disas_start':

> /home/alex/lsrc/qemu/qemu.git/disas.c:196: undefined reference to `cs_open'


As discussed on IRC, this turned out to be wrong link ordering due to me
placing -lcapstone in LDFLAGS instead of LIBS.  Fixed locally.

Any other comments on the patch set?


r~
Richard Henderson Sept. 26, 2017, 6:20 p.m. UTC | #4
On 09/19/2017 08:03 AM, Richard Henderson wrote:
> +    while (1) {

> +        /* We want to read memory for one insn, but generically we do not

> +           know how much memory that is.  We have a small buffer which is

> +           known to be sufficient for all supported targets.  Try to not

> +           read beyond the page, Just In Case.  For even more simplicity,

> +           ignore the actual target page size and use a 1k boundary.  If

> +           that turns out to be insufficient, we'll come back around the

> +           loop and read more.  */

> +        uint64_t epc = QEMU_ALIGN_UP(pc + csize, 1024);

> +        size_t tsize = MIN(sizeof(cap_buf) - csize, epc - pc);

...
> +        assert(tsize != 0);


By inspection, epc must be computed from pc + csize + 1.
If csize = 0 and pc % 1024 == 0, epc == pc and tsize == 0,
which will assert.


r~
diff mbox series

Patch

diff --git a/include/disas/bfd.h b/include/disas/bfd.h
index b01e002b4c..0f4ecdeb88 100644
--- a/include/disas/bfd.h
+++ b/include/disas/bfd.h
@@ -377,6 +377,10 @@  typedef struct disassemble_info {
   /* Command line options specific to the target disassembler.  */
   char * disassembler_options;
 
+  /* Options for Capstone disassembly.  */
+  int cap_arch;
+  int cap_mode;
+
 } disassemble_info;
 
 
diff --git a/include/disas/capstone.h b/include/disas/capstone.h
new file mode 100644
index 0000000000..84e214956d
--- /dev/null
+++ b/include/disas/capstone.h
@@ -0,0 +1,38 @@ 
+#ifndef QEMU_CAPSTONE_H
+#define QEMU_CAPSTONE_H 1
+
+#ifdef CONFIG_CAPSTONE
+
+#include <capstone.h>
+
+#else
+
+/* Just enough to allow backends to init without ifdefs.  */
+
+#define CS_ARCH_ARM     -1
+#define CS_ARCH_ARM64   -1
+#define CS_ARCH_MIPS    -1
+#define CS_ARCH_X86     -1
+#define CS_ARCH_PPC     -1
+#define CS_ARCH_SPARC   -1
+#define CS_ARCH_SYSZ    -1
+
+#define CS_MODE_LITTLE_ENDIAN    0
+#define CS_MODE_BIG_ENDIAN       0
+#define CS_MODE_ARM              0
+#define CS_MODE_16               0
+#define CS_MODE_32               0
+#define CS_MODE_64               0
+#define CS_MODE_THUMB            0
+#define CS_MODE_MCLASS           0
+#define CS_MODE_V8               0
+#define CS_MODE_MICRO            0
+#define CS_MODE_MIPS3            0
+#define CS_MODE_MIPS32R6         0
+#define CS_MODE_MIPSGP64         0
+#define CS_MODE_V9               0
+#define CS_MODE_MIPS32           0
+#define CS_MODE_MIPS64           0
+
+#endif /* CONFIG_CAPSTONE */
+#endif /* QEMU_CAPSTONE_H */
diff --git a/disas.c b/disas.c
index ad675dc361..4dbb492219 100644
--- a/disas.c
+++ b/disas.c
@@ -6,6 +6,7 @@ 
 
 #include "cpu.h"
 #include "disas/disas.h"
+#include "disas/capstone.h"
 
 typedef struct CPUDebug {
     struct disassemble_info info;
@@ -171,6 +172,192 @@  static int print_insn_od_target(bfd_vma pc, disassemble_info *info)
     return print_insn_objdump(pc, info, "OBJD-T");
 }
 
+#ifdef CONFIG_CAPSTONE
+/* Temporary storage for the capstone library.  This will be alloced via
+   malloc with a size private to the library; thus there's no reason not
+   to share this across calls and across host vs target disassembly.  */
+static __thread cs_insn *cap_insn;
+
+/* Initialize the Capstone library.  */
+/* ??? It would be nice to cache this.  We would need one handle for the
+   host and one for the target.  For most targets we can reset specific
+   parameters via cs_option(CS_OPT_MODE, new_mode), but we cannot change
+   CS_ARCH_* in this way.  Thus We would need to be able to close and
+   re-open the target handle with a different arch for the target in order
+   to handle AArch64 vs AArch32 mode switching.  */
+static cs_err cap_disas_start(disassemble_info *info, csh *handle)
+{
+    cs_mode cap_mode = info->cap_mode;
+    cs_err err;
+
+    cap_mode += (info->endian == BFD_ENDIAN_BIG ? CS_MODE_BIG_ENDIAN
+                 : CS_MODE_LITTLE_ENDIAN);
+
+    err = cs_open(info->cap_arch, cap_mode, handle);
+    if (err != CS_ERR_OK) {
+        return err;
+    }
+
+    /* ??? There probably ought to be a better place to put this.  */
+    if (info->cap_arch == CS_ARCH_X86) {
+        /* We don't care about errors (if for some reason the library
+           is compiled without AT&T syntax); the user will just have
+           to deal with the Intel syntax.  */
+        cs_option(*handle, CS_OPT_SYNTAX, CS_OPT_SYNTAX_ATT);
+    }
+
+    /* "Disassemble" unknown insns as ".byte W,X,Y,Z".  */
+    cs_option(*handle, CS_OPT_SKIPDATA, CS_OPT_ON);
+
+    /* Allocate temp space for cs_disasm_iter.  */
+    if (cap_insn == NULL) {
+        cap_insn = cs_malloc(*handle);
+        if (cap_insn == NULL) {
+            cs_close(handle);
+            return CS_ERR_MEM;
+        }
+    }
+    return CS_ERR_OK;
+}
+
+/* Disassemble SIZE bytes at PC for the target.  */
+static bool cap_disas_target(disassemble_info *info, uint64_t pc, size_t size)
+{
+    uint8_t cap_buf[1024];
+    csh handle;
+    cs_insn *insn;
+    size_t csize = 0;
+
+    if (cap_disas_start(info, &handle) != CS_ERR_OK) {
+        return false;
+    }
+    insn = cap_insn;
+
+    while (1) {
+        size_t tsize = MIN(sizeof(cap_buf) - csize, size);
+        const uint8_t *cbuf = cap_buf;
+
+        target_read_memory(pc + csize, cap_buf + csize, tsize, info);
+        csize += tsize;
+        size -= tsize;
+
+        while (cs_disasm_iter(handle, &cbuf, &csize, &pc, insn)) {
+            (*info->fprintf_func)(info->stream,
+                                  "0x%08" PRIx64 ":  %-12s %s\n",
+                                  insn->address, insn->mnemonic,
+                                  insn->op_str);
+        }
+
+        /* If the target memory is not consumed, go back for more... */
+        if (size != 0) {
+            /* ... taking care to move any remaining fractional insn
+               to the beginning of the buffer.  */
+            if (csize != 0) {
+                memmove(cap_buf, cbuf, csize);
+            }
+            continue;
+        }
+
+        /* Since the target memory is consumed, we should not have
+           a remaining fractional insn.  */
+        if (csize != 0) {
+            (*info->fprintf_func)(info->stream,
+                "Disassembler disagrees with translator "
+                "over instruction decoding\n"
+                "Please report this to qemu-devel@nongnu.org\n");
+        }
+        break;
+    }
+
+    cs_close(&handle);
+    return true;
+}
+
+/* Disassemble SIZE bytes at CODE for the host.  */
+static bool cap_disas_host(disassemble_info *info, void *code, size_t size)
+{
+    csh handle;
+    const uint8_t *cbuf;
+    cs_insn *insn;
+    uint64_t pc;
+
+    if (cap_disas_start(info, &handle) != CS_ERR_OK) {
+        return false;
+    }
+    insn = cap_insn;
+
+    cbuf = code;
+    pc = (uintptr_t)code;
+
+    while (cs_disasm_iter(handle, &cbuf, &size, &pc, insn)) {
+        (*info->fprintf_func)(info->stream,
+                              "0x%08" PRIx64 ":  %-12s %s\n",
+                              insn->address, insn->mnemonic,
+                              insn->op_str);
+    }
+    if (size != 0) {
+        (*info->fprintf_func)(info->stream,
+            "Disassembler disagrees with TCG over instruction encoding\n"
+            "Please report this to qemu-devel@nongnu.org\n");
+    }
+
+    cs_close(&handle);
+    return true;
+}
+
+#if !defined(CONFIG_USER_ONLY)
+/* Disassemble COUNT insns at PC for the target.  */
+static bool cap_disas_monitor(disassemble_info *info, uint64_t pc, int count)
+{
+    uint8_t cap_buf[32];
+    csh handle;
+    cs_insn *insn;
+    size_t csize = 0;
+
+    if (cap_disas_start(info, &handle) != CS_ERR_OK) {
+        return false;
+    }
+    insn = cap_insn;
+
+    while (1) {
+        /* We want to read memory for one insn, but generically we do not
+           know how much memory that is.  We have a small buffer which is
+           known to be sufficient for all supported targets.  Try to not
+           read beyond the page, Just In Case.  For even more simplicity,
+           ignore the actual target page size and use a 1k boundary.  If
+           that turns out to be insufficient, we'll come back around the
+           loop and read more.  */
+        uint64_t epc = QEMU_ALIGN_UP(pc + csize, 1024);
+        size_t tsize = MIN(sizeof(cap_buf) - csize, epc - pc);
+        const uint8_t *cbuf = cap_buf;
+
+        /* Make certain that we can make progress.  */
+        assert(tsize != 0);
+        info->read_memory_func(pc, cap_buf + csize, tsize, info);
+        csize += tsize;
+
+        if (cs_disasm_iter(handle, &cbuf, &csize, &pc, insn)) {
+            (*info->fprintf_func)(info->stream,
+                                  "0x%08" PRIx64 ":  %-12s %s\n",
+                                  insn->address, insn->mnemonic,
+                                  insn->op_str);
+            if (--count <= 0) {
+                break;
+            }
+        }
+        memmove(cap_buf, cbuf, csize);
+    }
+
+    cs_close(&handle);
+    return true;
+}
+#endif /* !CONFIG_USER_ONLY */
+#else
+# define cap_disas_target(i, p, s)  false
+# define cap_disas_host(i, p, s)  false
+# define cap_disas_monitor(i, p, c)  false
+#endif /* CONFIG_CAPSTONE */
+
 /* Disassemble this for me please... (debugging).  */
 void target_disas(FILE *out, CPUState *cpu, target_ulong code,
                   target_ulong size)
@@ -188,6 +375,8 @@  void target_disas(FILE *out, CPUState *cpu, target_ulong code,
     s.info.buffer_vma = code;
     s.info.buffer_length = size;
     s.info.print_address_func = generic_print_address;
+    s.info.cap_arch = -1;
+    s.info.cap_mode = 0;
 
 #ifdef TARGET_WORDS_BIGENDIAN
     s.info.endian = BFD_ENDIAN_BIG;
@@ -199,6 +388,10 @@  void target_disas(FILE *out, CPUState *cpu, target_ulong code,
         cc->disas_set_info(cpu, &s.info);
     }
 
+    if (s.info.cap_arch >= 0 && cap_disas_target(&s.info, code, size)) {
+        return;
+    }
+
     if (s.info.print_insn == NULL) {
         s.info.print_insn = print_insn_od_target;
     }
@@ -206,18 +399,6 @@  void target_disas(FILE *out, CPUState *cpu, target_ulong code,
     for (pc = code; size > 0; pc += count, size -= count) {
 	fprintf(out, "0x" TARGET_FMT_lx ":  ", pc);
 	count = s.info.print_insn(pc, &s.info);
-#if 0
-        {
-            int i;
-            uint8_t b;
-            fprintf(out, " {");
-            for(i = 0; i < count; i++) {
-                target_read_memory(pc + i, &b, 1, &s.info);
-                fprintf(out, " %02x", b);
-            }
-            fprintf(out, " }");
-        }
-#endif
 	fprintf(out, "\n");
 	if (count < 0)
 	    break;
@@ -245,6 +426,8 @@  void disas(FILE *out, void *code, unsigned long size)
     s.info.buffer = code;
     s.info.buffer_vma = (uintptr_t)code;
     s.info.buffer_length = size;
+    s.info.cap_arch = -1;
+    s.info.cap_mode = 0;
 
 #ifdef HOST_WORDS_BIGENDIAN
     s.info.endian = BFD_ENDIAN_BIG;
@@ -282,6 +465,11 @@  void disas(FILE *out, void *code, unsigned long size)
 #elif defined(__hppa__)
     print_insn = print_insn_hppa;
 #endif
+
+    if (s.info.cap_arch >= 0 && cap_disas_host(&s.info, code, size)) {
+        return;
+    }
+
     if (print_insn == NULL) {
         print_insn = print_insn_od_host;
     }
@@ -344,8 +532,9 @@  void monitor_disas(Monitor *mon, CPUState *cpu,
     monitor_disas_is_physical = is_physical;
     s.info.read_memory_func = monitor_read_memory;
     s.info.print_address_func = generic_print_address;
-
     s.info.buffer_vma = pc;
+    s.info.cap_arch = -1;
+    s.info.cap_mode = 0;
 
 #ifdef TARGET_WORDS_BIGENDIAN
     s.info.endian = BFD_ENDIAN_BIG;
@@ -357,6 +546,10 @@  void monitor_disas(Monitor *mon, CPUState *cpu,
         cc->disas_set_info(cpu, &s.info);
     }
 
+    if (s.info.cap_arch >= 0 && cap_disas_monitor(&s.info, pc, nb_insn)) {
+        return;
+    }
+
     if (!s.info.print_insn) {
         monitor_printf(mon, "0x" TARGET_FMT_lx
                        ": Asm output not supported on this arch\n", pc);
diff --git a/configure b/configure
index 94db2d103e..388bd2ad43 100755
--- a/configure
+++ b/configure
@@ -365,6 +365,7 @@  opengl_dmabuf="no"
 cpuid_h="no"
 avx2_opt="no"
 zlib="yes"
+capstone=""
 lzo=""
 snappy=""
 bzip2=""
@@ -1279,6 +1280,10 @@  for opt do
           error_exit "vhost-user isn't available on win32"
       fi
   ;;
+  --disable-capstone) capstone="no"
+  ;;
+  --enable-capstone) capstone="yes"
+  ;;
   *)
       echo "ERROR: unknown option $opt"
       echo "Try '$0 --help' for more information"
@@ -1524,6 +1529,7 @@  disabled with --disable-FEATURE, default is enabled if available:
   vxhs            Veritas HyperScale vDisk backend support
   crypto-afalg    Linux AF_ALG crypto backend driver
   vhost-user      vhost-user support
+  capstone        capstone disassembler support
 
 NOTE: The object files are built at the place where configure is launched
 EOF
@@ -4350,6 +4356,22 @@  EOF
 fi
 
 ##########################################
+# capstone
+
+if test "$capstone" != no; then
+  if $pkg_config capstone; then
+    capstone=yes
+    QEMU_CFLAGS="$QEMU_CFLAGS $($pkg_config --cflags capstone)"
+    LDFLAGS="$LDFLAGS $($pkg_config --libs capstone)"
+  else
+    if test "$capstone" = yes; then
+      feature_not_found capstone
+    fi
+    capstone=no
+  fi
+fi
+
+##########################################
 # check if we have fdatasync
 
 fdatasync=no
@@ -5370,6 +5392,7 @@  echo "jemalloc support  $jemalloc"
 echo "avx2 optimization $avx2_opt"
 echo "replication support $replication"
 echo "VxHS block device $vxhs"
+echo "capstone          $capstone"
 
 if test "$sdl_too_old" = "yes"; then
 echo "-> Your SDL version is too old - please upgrade to have SDL support"
@@ -6013,6 +6036,9 @@  fi
 if test "$ivshmem" = "yes" ; then
   echo "CONFIG_IVSHMEM=y" >> $config_host_mak
 fi
+if test "$capstone" = "yes" ; then
+  echo "CONFIG_CAPSTONE=y" >> $config_host_mak
+fi
 
 # Hold two types of flag:
 #   CONFIG_THREAD_SETNAME_BYTHREAD  - we've got a way of setting the name on