This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
PATCH: 4/6: Add AVX support (amd64 changes)
Hi,
Here are the amd64 changes to support AVX. OK to install?
Thanks.
H.J.
---
2010-03-04 H.J. Lu <hongjiu.lu@intel.com>
* amd64-linux-nat.c: Include "regset.h", "elf/common.h" and
<sys/uio.h>.
(xstate_size): New.
(xstate_size_n_of_int64): Likewise.
(amd64_linux_fetch_inferior_registers): Support PTRACE_GETFPREGS.
(amd64_linux_store_inferior_registers): Likewise.
(amd64_linux_read_description): Check and enable AVX target
descriptions.
* amd64-linux-tdep.c: Include "regset.h", "i386-linux-tdep.h"
and "features/i386/amd64-avx-linux.c".
(amd64_linux_regset_sections): New.
(amd64_linux_core_read_description): Check and enable AVX
target description.
(amd64_linux_init_abi): Set xsave_xcr0_offset. Call
set_gdbarch_core_regset_sections.
(_initialize_amd64_linux_tdep): Call
initialize_tdesc_amd64_avx_linux.
* amd64-linux-tdep.h (tdesc_amd64_avx_linux): New.
(amd64_linux_regset_sections): Likewise.
* amd64-tdep.c: Include "features/i386/amd64-avx.c".
(amd64_register_names): Renamed to ...
(amd64_sse_register_names): This.
(amd64_avx_register_names): New.
(amd64_xmm_names): Likewise.
(amd64_supply_xstateregset): Likewise.
(amd64_collect_xstateregset): Likewise.
(amd64_supply_xsave): Likewise.
(amd64_collect_xsave): Likewise.
(AMD64_NUM_REGS): Updated.
(amd64_pseudo_register_name): Support pseudo XMM registers.
(amd64_regset_from_core_section): Support .reg-xstate section.
(amd64_init_abi): Set num_xmm_regs, register_names and
num_vector_regs.
(amd64_init_abi): Call initialize_tdesc_amd64_avx.
* amd64-tdep.h (amd64_supply_xsave): New.
(amd64_collect_xsave): Likewise.
diff --git a/gdb/amd64-linux-nat.c b/gdb/amd64-linux-nat.c
index b9d5833..4a79891 100644
--- a/gdb/amd64-linux-nat.c
+++ b/gdb/amd64-linux-nat.c
@@ -23,11 +23,14 @@
#include "inferior.h"
#include "gdbcore.h"
#include "regcache.h"
+#include "regset.h"
#include "linux-nat.h"
#include "amd64-linux-tdep.h"
#include "gdb_assert.h"
#include "gdb_string.h"
+#include "elf/common.h"
+#include <sys/uio.h>
#include <sys/ptrace.h>
#include <sys/debugreg.h>
#include <sys/syscall.h>
@@ -52,6 +55,16 @@
#include "amd64-nat.h"
#include "i386-nat.h"
+/* The extended state size in bytes. */
+static unsigned int xstate_size;
+
+/* The extended state size in unit of int64. We use array of int64 for
+ better alignment. */
+static unsigned int xstate_size_n_of_int64;
+
+/* Does the current host support PTRACE_GETREGSET? */
+static int have_ptrace_getregset = -1;
+
/* Mapping between the general-purpose registers in GNU/Linux x86-64
`struct user' format and GDB's register cache layout. */
@@ -183,10 +196,26 @@ amd64_linux_fetch_inferior_registers (struct target_ops *ops,
{
elf_fpregset_t fpregs;
- if (ptrace (PTRACE_GETFPREGS, tid, 0, (long) &fpregs) < 0)
- perror_with_name (_("Couldn't get floating point status"));
+ if (have_ptrace_getregset)
+ {
+ unsigned long long xstateregs[xstate_size_n_of_int64];
+ struct iovec iov;
+
+ iov.iov_base = xstateregs;
+ iov.iov_len = xstate_size;
+ if (ptrace (PTRACE_GETREGSET, tid,
+ (unsigned int) NT_X86_XSTATE, (long) &iov) < 0)
+ perror_with_name (_("Couldn't get extended state status"));
+
+ amd64_supply_xsave (regcache, -1, xstateregs);
+ }
+ else
+ {
+ if (ptrace (PTRACE_GETFPREGS, tid, 0, (long) &fpregs) < 0)
+ perror_with_name (_("Couldn't get floating point status"));
- amd64_supply_fxsave (regcache, -1, &fpregs);
+ amd64_supply_fxsave (regcache, -1, &fpregs);
+ }
}
}
@@ -226,15 +255,33 @@ amd64_linux_store_inferior_registers (struct target_ops *ops,
{
elf_fpregset_t fpregs;
- if (ptrace (PTRACE_GETFPREGS, tid, 0, (long) &fpregs) < 0)
- perror_with_name (_("Couldn't get floating point status"));
+ if (have_ptrace_getregset)
+ {
+ unsigned long long xstateregs[xstate_size_n_of_int64];
+ struct iovec iov;
- amd64_collect_fxsave (regcache, regnum, &fpregs);
+ iov.iov_base = xstateregs;
+ iov.iov_len = xstate_size;
+ if (ptrace (PTRACE_GETREGSET, tid,
+ (unsigned int) NT_X86_XSTATE, (long) &iov) < 0)
+ perror_with_name (_("Couldn't get extended state status"));
- if (ptrace (PTRACE_SETFPREGS, tid, 0, (long) &fpregs) < 0)
- perror_with_name (_("Couldn't write floating point status"));
+ amd64_collect_xsave (regcache, regnum, xstateregs, 0);
- return;
+ if (ptrace (PTRACE_SETREGSET, tid,
+ (unsigned int) NT_X86_XSTATE, (long) &iov) < 0)
+ perror_with_name (_("Couldn't write extended state status"));
+ }
+ else
+ {
+ if (ptrace (PTRACE_GETFPREGS, tid, 0, (long) &fpregs) < 0)
+ perror_with_name (_("Couldn't get floating point status"));
+
+ amd64_collect_fxsave (regcache, regnum, &fpregs);
+
+ if (ptrace (PTRACE_SETFPREGS, tid, 0, (long) &fpregs) < 0)
+ perror_with_name (_("Couldn't write floating point status"));
+ }
}
}
@@ -688,6 +735,8 @@ amd64_linux_read_description (struct target_ops *ops)
{
unsigned long cs;
int tid;
+ int is_64bit;
+ static unsigned long long xcr0;
/* GNU/Linux LWP ID's are process ID's. */
tid = TIDGET (inferior_ptid);
@@ -701,10 +750,53 @@ amd64_linux_read_description (struct target_ops *ops)
if (errno != 0)
perror_with_name (_("Couldn't get CS register"));
- if (cs == AMD64_LINUX_USER64_CS)
- return tdesc_amd64_linux;
+ is_64bit = cs == AMD64_LINUX_USER64_CS;
+
+ if (have_ptrace_getregset == -1)
+ {
+ unsigned long long xstateregs[(I386_XSTATE_SSE_SIZE
+ / sizeof (long long))];
+ struct iovec iov;
+
+ iov.iov_base = xstateregs;
+ iov.iov_len = I386_XSTATE_SSE_SIZE;
+
+ /* Check if PTRACE_GETREGSET works. */
+ if (ptrace (PTRACE_GETREGSET, tid,
+ (unsigned int) NT_X86_XSTATE, (long) &iov) < 0)
+ have_ptrace_getregset = 0;
+ else
+ {
+ have_ptrace_getregset = 1;
+
+ /* Get XCR0 from XSAVE extended state. */
+ xcr0 = xstateregs[(I386_LINUX_XSAVE_XCR0_OFFSET
+ / sizeof (long long))];
+
+ xstate_size = I386_XSTATE_SIZE (xcr0);
+ xstate_size_n_of_int64 = xstate_size / sizeof (long long);
+ }
+
+ i386_linux_update_xstateregset (amd64_linux_regset_sections,
+ xstate_size);
+ }
+
+ /* Check the native XCR0 only if PTRACE_GETREGSET is available. */
+ if (have_ptrace_getregset
+ && (xcr0 & I386_XSTATE_AVX_MASK) == I386_XSTATE_AVX_MASK)
+ {
+ if (is_64bit)
+ return tdesc_amd64_avx_linux;
+ else
+ return tdesc_i386_avx_linux;
+ }
else
- return tdesc_i386_linux;
+ {
+ if (is_64bit)
+ return tdesc_amd64_linux;
+ else
+ return tdesc_i386_linux;
+ }
}
/* Provide a prototype to silence -Wmissing-prototypes. */
diff --git a/gdb/amd64-linux-tdep.c b/gdb/amd64-linux-tdep.c
index 4ad6dc9..51722bf 100644
--- a/gdb/amd64-linux-tdep.c
+++ b/gdb/amd64-linux-tdep.c
@@ -28,7 +28,9 @@
#include "symtab.h"
#include "gdbtypes.h"
#include "reggroups.h"
+#include "regset.h"
#include "amd64-linux-tdep.h"
+#include "i386-linux-tdep.h"
#include "linux-tdep.h"
#include "gdb_string.h"
@@ -38,6 +40,7 @@
#include "xml-syscall.h"
#include "features/i386/amd64-linux.c"
+#include "features/i386/amd64-avx-linux.c"
/* The syscall's XML filename for i386. */
#define XML_SYSCALL_FILENAME_AMD64 "syscalls/amd64-linux.xml"
@@ -45,6 +48,15 @@
#include "record.h"
#include "linux-record.h"
+/* Supported register note sections. */
+struct core_regset_section amd64_linux_regset_sections[] =
+{
+ { ".reg", 144, "general-purpose" },
+ { ".reg2", 512, "floating-point" },
+ { ".reg-xstate", 0, "XSAVE extended state" },
+ { NULL, 0 }
+};
+
/* Mapping between the general-purpose registers in `struct user'
format and GDB's register cache layout. */
@@ -1250,12 +1262,17 @@ amd64_linux_core_read_description (struct gdbarch *gdbarch,
bfd *abfd)
{
asection *section = bfd_get_section_by_name (abfd, ".reg2");
+ unsigned long long xcr0;
if (section == NULL)
return NULL;
/* Linux/x86-64. */
- return tdesc_amd64_linux;
+ xcr0 = i386_linux_core_read_xcr0 (gdbarch, target, abfd);
+ if ((xcr0 & I386_XSTATE_AVX_MASK) == I386_XSTATE_AVX_MASK)
+ return tdesc_amd64_avx_linux;
+ else
+ return tdesc_amd64_linux;
}
static void
@@ -1297,6 +1314,8 @@ amd64_linux_init_abi (struct gdbarch_info info, struct gdbarch *gdbarch)
tdep->sc_reg_offset = amd64_linux_sc_reg_offset;
tdep->sc_num_regs = ARRAY_SIZE (amd64_linux_sc_reg_offset);
+ tdep->xsave_xcr0_offset = I386_LINUX_XSAVE_XCR0_OFFSET;
+
/* GNU/Linux uses SVR4-style shared libraries. */
set_solib_svr4_fetch_link_map_offsets
(gdbarch, svr4_lp64_fetch_link_map_offsets);
@@ -1318,6 +1337,9 @@ amd64_linux_init_abi (struct gdbarch_info info, struct gdbarch *gdbarch)
/* GNU/Linux uses SVR4-style shared libraries. */
set_gdbarch_skip_trampoline_code (gdbarch, find_solib_trampoline_target);
+ /* Install supported register note sections. */
+ set_gdbarch_core_regset_sections (gdbarch, amd64_linux_regset_sections);
+
set_gdbarch_core_read_description (gdbarch,
amd64_linux_core_read_description);
@@ -1517,4 +1539,5 @@ _initialize_amd64_linux_tdep (void)
/* Initialize the Linux target description */
initialize_tdesc_amd64_linux ();
+ initialize_tdesc_amd64_avx_linux ();
}
diff --git a/gdb/amd64-linux-tdep.h b/gdb/amd64-linux-tdep.h
index 33316fb..78d9744 100644
--- a/gdb/amd64-linux-tdep.h
+++ b/gdb/amd64-linux-tdep.h
@@ -33,6 +33,10 @@
/* Linux target description. */
extern struct target_desc *tdesc_amd64_linux;
+extern struct target_desc *tdesc_amd64_avx_linux;
+
+/* Supported register note sections. */
+extern struct core_regset_section amd64_linux_regset_sections[];
/* Enum that defines the syscall identifiers for amd64 linux.
Used for process record/replay, these will be translated into
diff --git a/gdb/amd64-tdep.c b/gdb/amd64-tdep.c
index 8c41a8a..1a6cd80 100644
--- a/gdb/amd64-tdep.c
+++ b/gdb/amd64-tdep.c
@@ -43,6 +43,7 @@
#include "i387-tdep.h"
#include "features/i386/amd64.c"
+#include "features/i386/amd64-avx.c"
/* Note that the AMD64 architecture was previously known as x86-64.
The latter is (forever) engraved into the canonical system name as
@@ -53,7 +54,7 @@
/* Register information. */
-static const char *amd64_register_names[] =
+static const char *amd64_sse_register_names[] =
{
"rax", "rbx", "rcx", "rdx", "rsi", "rdi", "rbp", "rsp",
@@ -71,8 +72,26 @@ static const char *amd64_register_names[] =
"mxcsr",
};
+static const char *amd64_avx_register_names[] =
+{
+ "rax", "rbx", "rcx", "rdx", "rsi", "rdi", "rbp", "rsp",
+
+ /* %r8 is indeed register number 8. */
+ "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15",
+ "rip", "eflags", "cs", "ss", "ds", "es", "fs", "gs",
+
+ /* %st0 is register number 24. */
+ "st0", "st1", "st2", "st3", "st4", "st5", "st6", "st7",
+ "fctrl", "fstat", "ftag", "fiseg", "fioff", "foseg", "fooff", "fop",
+
+ /* %ymm0 is register number 40. */
+ "ymm0", "ymm1", "ymm2", "ymm3", "ymm4", "ymm5", "ymm6", "ymm7",
+ "ymm8", "ymm9", "ymm10", "ymm11", "ymm12", "ymm13", "ymm14", "ymm15",
+ "mxcsr"
+};
+
/* Total number of registers. */
-#define AMD64_NUM_REGS ARRAY_SIZE (amd64_register_names)
+#define AMD64_NUM_REGS ARRAY_SIZE (amd64_sse_register_names)
/* The registers used to pass integer arguments during a function call. */
static int amd64_dummy_call_integer_regs[] =
@@ -234,6 +253,14 @@ static const char *amd64_dword_names[] =
"r8d", "r9d", "r10d", "r11d", "r12d", "r13d", "r14d", "r15d"
};
+/* Register names for XMMM pseudo-registers. */
+
+static const char *amd64_xmm_names[] =
+{
+ "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7",
+ "xmm8", "xmm9", "xmm10", "xmm11", "xmm12", "xmm13", "xmm14", "xmm15"
+};
+
/* Return the name of register REGNUM. */
static const char *
@@ -242,6 +269,8 @@ amd64_pseudo_register_name (struct gdbarch *gdbarch, int regnum)
struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
if (i386_byte_regnum_p (gdbarch, regnum))
return amd64_byte_names[regnum - tdep->al_regnum];
+ else if (i386_xmm_regnum_p (gdbarch, regnum))
+ return amd64_xmm_names[regnum - tdep->xmm0_regnum];
else if (i386_word_regnum_p (gdbarch, regnum))
return amd64_word_names[regnum - tdep->ax_regnum];
else if (i386_dword_regnum_p (gdbarch, regnum))
@@ -2148,6 +2177,28 @@ amd64_collect_fpregset (const struct regset *regset,
amd64_collect_fxsave (regcache, regnum, fpregs);
}
+/* Similar to amd64_supply_fpregset, but use XSAVE extended state. */
+
+static void
+amd64_supply_xstateregset (const struct regset *regset,
+ struct regcache *regcache, int regnum,
+ const void *xstateregs, size_t len)
+{
+ const struct gdbarch_tdep *tdep = gdbarch_tdep (regset->arch);
+ amd64_supply_xsave (regcache, regnum, xstateregs);
+}
+
+/* Similar to amd64_collect_fpregset, but use XSAVE extended state. */
+
+static void
+amd64_collect_xstateregset (const struct regset *regset,
+ const struct regcache *regcache,
+ int regnum, void *xstateregs, size_t len)
+{
+ const struct gdbarch_tdep *tdep = gdbarch_tdep (regset->arch);
+ amd64_collect_xsave (regcache, regnum, xstateregs, 1);
+}
+
/* Return the appropriate register set for the core section identified
by SECT_NAME and SECT_SIZE. */
@@ -2166,6 +2217,16 @@ amd64_regset_from_core_section (struct gdbarch *gdbarch,
return tdep->fpregset;
}
+ if (strcmp (sect_name, ".reg-xstate") == 0)
+ {
+ if (tdep->xstateregset == NULL)
+ tdep->xstateregset = regset_alloc (gdbarch,
+ amd64_supply_xstateregset,
+ amd64_collect_xstateregset);
+
+ return tdep->xstateregset;
+ }
+
return i386_regset_from_core_section (gdbarch, sect_name, sect_size);
}
@@ -2226,7 +2287,17 @@ amd64_init_abi (struct gdbarch_info info, struct gdbarch *gdbarch)
tdep->tdesc = tdesc;
tdep->num_core_regs = AMD64_NUM_GREGS + I387_NUM_REGS;
- tdep->register_names = amd64_register_names;
+
+ if (tdesc_find_feature (tdesc, "org.gnu.gdb.i386.sse") != NULL)
+ {
+ tdep->register_names = amd64_sse_register_names;
+ tdep->num_xmm_regs = 0;
+ }
+ else
+ {
+ tdep->register_names = amd64_avx_register_names;
+ tdep->num_xmm_regs = 16;
+ }
tdep->num_byte_regs = 16;
tdep->num_word_regs = 16;
@@ -2243,7 +2314,7 @@ amd64_init_abi (struct gdbarch_info info, struct gdbarch *gdbarch)
/* AMD64 has an FPU and 16 SSE registers. */
tdep->st0_regnum = AMD64_ST0_REGNUM;
- tdep->num_xmm_regs = 16;
+ tdep->num_vector_regs = 16;
/* This is what all the fuss is about. */
set_gdbarch_long_bit (gdbarch, 64);
@@ -2321,6 +2392,7 @@ void
_initialize_amd64_tdep (void)
{
initialize_tdesc_amd64 ();
+ initialize_tdesc_amd64_avx ();
}
@@ -2356,6 +2428,30 @@ amd64_supply_fxsave (struct regcache *regcache, int regnum,
}
}
+/* Similar to amd64_supply_fxsave, but use XSAVE extended state. */
+
+void
+amd64_supply_xsave (struct regcache *regcache, int regnum,
+ const void *xsave)
+{
+ struct gdbarch *gdbarch = get_regcache_arch (regcache);
+ struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
+
+ i387_supply_xsave (regcache, regnum, xsave);
+
+ if (xsave && gdbarch_ptr_bit (gdbarch) == 64)
+ {
+ const gdb_byte *regs = xsave;
+
+ if (regnum == -1 || regnum == I387_FISEG_REGNUM (tdep))
+ regcache_raw_supply (regcache, I387_FISEG_REGNUM (tdep),
+ regs + 12);
+ if (regnum == -1 || regnum == I387_FOSEG_REGNUM (tdep))
+ regcache_raw_supply (regcache, I387_FOSEG_REGNUM (tdep),
+ regs + 20);
+ }
+}
+
/* Fill register REGNUM (if it is a floating-point or SSE register) in
*FXSAVE with the value from REGCACHE. If REGNUM is -1, do this for
all registers. This function doesn't touch any of the reserved
@@ -2379,3 +2475,26 @@ amd64_collect_fxsave (const struct regcache *regcache, int regnum,
regcache_raw_collect (regcache, I387_FOSEG_REGNUM (tdep), regs + 20);
}
}
+
+/* Similar to amd64_collect_fxsave, but but use XSAVE extended state. */
+
+void
+amd64_collect_xsave (const struct regcache *regcache, int regnum,
+ void *xsave, int gcore)
+{
+ struct gdbarch *gdbarch = get_regcache_arch (regcache);
+ struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
+ gdb_byte *regs = xsave;
+
+ i387_collect_xsave (regcache, regnum, xsave, gcore);
+
+ if (gdbarch_ptr_bit (gdbarch) == 64)
+ {
+ if (regnum == -1 || regnum == I387_FISEG_REGNUM (tdep))
+ regcache_raw_collect (regcache, I387_FISEG_REGNUM (tdep),
+ regs + 12);
+ if (regnum == -1 || regnum == I387_FOSEG_REGNUM (tdep))
+ regcache_raw_collect (regcache, I387_FOSEG_REGNUM (tdep),
+ regs + 20);
+ }
+}
diff --git a/gdb/amd64-tdep.h b/gdb/amd64-tdep.h
index 363479c..4dccb4f 100644
--- a/gdb/amd64-tdep.h
+++ b/gdb/amd64-tdep.h
@@ -91,6 +91,10 @@ extern struct type *amd64_register_type (struct gdbarch *gdbarch, int regnum);
extern void amd64_supply_fxsave (struct regcache *regcache, int regnum,
const void *fxsave);
+/* Similar to amd64_supply_fxsave, but use XSAVE extended state. */
+extern void amd64_supply_xsave (struct regcache *regcache, int regnum,
+ const void *xsave);
+
/* Fill register REGNUM (if it is a floating-point or SSE register) in
*FXSAVE with the value from REGCACHE. If REGNUM is -1, do this for
all registers. This function doesn't touch any of the reserved
@@ -99,6 +103,10 @@ extern void amd64_supply_fxsave (struct regcache *regcache, int regnum,
extern void amd64_collect_fxsave (const struct regcache *regcache, int regnum,
void *fxsave);
+/* Similar to amd64_collect_fxsave, but but use XSAVE extended state. */
+extern void amd64_collect_xsave (const struct regcache *regcache,
+ int regnum, void *xsave, int gcore);
+
void amd64_classify (struct type *type, enum amd64_reg_class class[2]);