aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorBryan Chan <bryanpkc@gmail.com>2016-01-22 12:26:47 -0500
committerBryan Chan <bryanpkc@gmail.com>2016-02-04 20:15:26 -0500
commit644a6bdbdb50b26a25e4428c43556467a6e8b5cc (patch)
treeea74ed406c13717aae9c661a4408c5fc70cc4912
parentbab7753aad44b3395a063966f32c23f632fee174 (diff)
downloadgperftools-644a6bdbdb50b26a25e4428c43556467a6e8b5cc.tar.gz
Add support for Linux s390x
This resolves gperftools/gperftools#761.
-rw-r--r--configure.ac9
-rw-r--r--m4/pc_from_ucontext.m41
-rw-r--r--src/base/basictypes.h2
-rw-r--r--src/base/linux_syscall_support.h287
-rw-r--r--src/base/linuxthreads.h3
-rwxr-xr-xsrc/malloc_hook_mmap_linux.h14
6 files changed, 306 insertions, 10 deletions
diff --git a/configure.ac b/configure.ac
index 6e25cf1..3f8a5e0 100644
--- a/configure.ac
+++ b/configure.ac
@@ -51,6 +51,13 @@ case "$host" in
*-darwin*) default_enable_heap_checker=no;;
esac
+# Currently only backtrace works on s390x.
+AC_COMPILE_IFELSE([AC_LANG_PROGRAM(, [return __s390x__])],
+ [default_enable_libunwind=no
+ default_enable_backtrace=yes],
+ [default_enable_libunwind=yes
+ default_enable_backtrace=no])
+
# Disable libunwind linking on ppc64 by default.
AC_COMPILE_IFELSE([AC_LANG_PROGRAM(, [return __PPC64__])],
[default_enable_libunwind=no
@@ -92,7 +99,7 @@ AC_ARG_ENABLE([stacktrace-via-backtrace],
[AS_HELP_STRING([--enable-stacktrace-via-backtrace],
[enable use of backtrace() for stacktrace capturing (may deadlock)])],
[enable_backtrace=yes],
- [])
+ [enable_backtrace="$default_enable_backtrace"])
AC_ARG_ENABLE([libunwind],
[AS_HELP_STRING([--enable-libunwind],
[enable libunwind linking])],
diff --git a/m4/pc_from_ucontext.m4 b/m4/pc_from_ucontext.m4
index b4fd0d0..d743fa4 100644
--- a/m4/pc_from_ucontext.m4
+++ b/m4/pc_from_ucontext.m4
@@ -27,6 +27,7 @@ AC_DEFUN([AC_PC_FROM_UCONTEXT],
pc_fields="$pc_fields uc_mcontext.sc_ip" # Linux (ia64)
pc_fields="$pc_fields uc_mcontext.pc" # Linux (mips)
pc_fields="$pc_fields uc_mcontext.uc_regs->gregs[[PT_NIP]]" # Linux (ppc)
+ pc_fields="$pc_fields uc_mcontext.psw.addr" # Linux (s390x)
pc_fields="$pc_fields uc_mcontext.gregs[[R15]]" # Linux (arm old [untested])
pc_fields="$pc_fields uc_mcontext.arm_pc" # Linux (arm arch 5)
pc_fields="$pc_fields uc_mcontext.gp_regs[[PT_NIP]]" # Suse SLES 11 (ppc64)
diff --git a/src/base/basictypes.h b/src/base/basictypes.h
index 9c930a0..f0e25da 100644
--- a/src/base/basictypes.h
+++ b/src/base/basictypes.h
@@ -357,6 +357,8 @@ class AssignAttributeStartEnd {
# elif (defined(__aarch64__))
# define CACHELINE_ALIGNED __attribute__((aligned(64)))
// implementation specific, Cortex-A53 and 57 should have 64 bytes
+# elif (defined(__s390x__))
+# define CACHELINE_ALIGNED __attribute__((aligned(256)))
# else
# error Could not determine cache line length - unknown architecture
# endif
diff --git a/src/base/linux_syscall_support.h b/src/base/linux_syscall_support.h
index 90add93..5d578cd 100644
--- a/src/base/linux_syscall_support.h
+++ b/src/base/linux_syscall_support.h
@@ -130,11 +130,13 @@
#ifndef SYS_LINUX_SYSCALL_SUPPORT_H
#define SYS_LINUX_SYSCALL_SUPPORT_H
-/* We currently only support x86-32, x86-64, ARM, MIPS, PPC/PPC64 and Aarch64 on Linux.
+/* We currently only support x86-32, x86-64, ARM, MIPS, PPC/PPC64, Aarch64 and s390x on Linux.
* Porting to other related platforms should not be difficult.
*/
#if (defined(__i386__) || defined(__x86_64__) || defined(__arm__) || \
- defined(__mips__) || defined(__PPC__) || defined(__aarch64__)) && defined(__linux)
+ defined(__mips__) || defined(__PPC__) || \
+ defined(__aarch64__) || defined(__s390x__)) \
+ && (defined(__linux))
#ifndef SYS_CPLUSPLUS
#ifdef __cplusplus
@@ -259,6 +261,8 @@ struct kernel_old_sigaction {
} __attribute__((packed,aligned(4)));
#elif (defined(__mips__) && _MIPS_SIM == _MIPS_SIM_ABI32)
#define kernel_old_sigaction kernel_sigaction
+#elif defined(__aarch64__) || defined(__s390x__)
+ // No kernel_old_sigaction defined for arm64 or s390x.
#endif
/* Some kernel functions (e.g. sigaction() in 2.6.23) require that the
@@ -373,7 +377,7 @@ struct kernel_stat64 {
};
#endif
-/* include/asm-{arm,generic,i386,mips,x86_64,ppc}/stat.h */
+/* include/asm-{arm,generic,i386,mips,x86_64,ppc,s390}/stat.h */
#if defined(__i386__) || defined(__arm__)
struct kernel_stat {
/* The kernel headers suggest that st_dev and st_rdev should be 32bit
@@ -489,6 +493,27 @@ struct kernel_stat {
unsigned int __unused4;
unsigned int __unused5;
};
+#elif defined(__s390x__)
+struct kernel_stat {
+ unsigned long st_dev;
+ unsigned long st_ino;
+ unsigned long st_nlink;
+ unsigned int st_mode;
+ unsigned int st_uid;
+ unsigned int st_gid;
+ unsigned int __pad1;
+ unsigned long st_rdev;
+ unsigned long st_size;
+ unsigned long st_atime_;
+ unsigned long st_atime_nsec_;
+ unsigned long st_mtime_;
+ unsigned long st_mtime_nsec_;
+ unsigned long st_ctime_;
+ unsigned long st_ctime_nsec_;
+ unsigned long st_blksize;
+ long st_blocks;
+ unsigned long __unused[3];
+};
#endif
@@ -703,6 +728,134 @@ struct kernel_stat {
#define __NR_fstatat 79
#endif
/* End of aarch64 defininitions */
+#elif defined(__s390x__)
+#ifndef __NR_quotactl
+#define __NR_quotactl 131
+#endif
+#ifndef __NR_rt_sigreturn
+#define __NR_rt_sigreturn 173
+#endif
+#ifndef __NR_rt_sigaction
+#define __NR_rt_sigaction 174
+#endif
+#ifndef __NR_rt_sigprocmask
+#define __NR_rt_sigprocmask 175
+#endif
+#ifndef __NR_rt_sigpending
+#define __NR_rt_sigpending 176
+#endif
+#ifndef __NR_rt_sigsuspend
+#define __NR_rt_sigsuspend 179
+#endif
+#ifndef __NR_pread64
+#define __NR_pread64 180
+#endif
+#ifndef __NR_pwrite64
+#define __NR_pwrite64 181
+#endif
+#ifndef __NR_getrlimit
+#define __NR_getrlimit 191
+#endif
+#ifndef __NR_setresuid
+#define __NR_setresuid 208
+#endif
+#ifndef __NR_getresuid
+#define __NR_getresuid 209
+#endif
+#ifndef __NR_setresgid
+#define __NR_setresgid 210
+#endif
+#ifndef __NR_getresgid
+#define __NR_getresgid 211
+#endif
+#ifndef __NR_setfsuid
+#define __NR_setfsuid 215
+#endif
+#ifndef __NR_setfsgid
+#define __NR_setfsgid 216
+#endif
+#ifndef __NR_getdents64
+#define __NR_getdents64 220
+#endif
+#ifndef __NR_readahead
+#define __NR_readahead 222
+#endif
+#ifndef __NR_setxattr
+#define __NR_setxattr 224
+#endif
+#ifndef __NR_lsetxattr
+#define __NR_lsetxattr 225
+#endif
+#ifndef __NR_getxattr
+#define __NR_getxattr 227
+#endif
+#ifndef __NR_lgetxattr
+#define __NR_lgetxattr 228
+#endif
+#ifndef __NR_listxattr
+#define __NR_listxattr 230
+#endif
+#ifndef __NR_llistxattr
+#define __NR_llistxattr 231
+#endif
+#ifndef __NR_gettid
+#define __NR_gettid 236
+#endif
+#ifndef __NR_tkill
+#define __NR_tkill 237
+#endif
+#ifndef __NR_futex
+#define __NR_futex 238
+#endif
+#ifndef __NR_sched_setaffinity
+#define __NR_sched_setaffinity 239
+#endif
+#ifndef __NR_sched_getaffinity
+#define __NR_sched_getaffinity 240
+#endif
+#ifndef __NR_set_tid_address
+#define __NR_set_tid_address 252
+#endif
+#ifndef __NR_fadvise64
+#define __NR_fadvise64 253
+#endif
+#ifndef __NR_clock_gettime
+#define __NR_clock_gettime 260
+#endif
+#ifndef __NR_clock_getres
+#define __NR_clock_getres 261
+#endif
+#ifndef __NR_statfs64
+#define __NR_statfs64 265
+#endif
+#ifndef __NR_fstatfs64
+#define __NR_fstatfs64 266
+#endif
+#ifndef __NR_ioprio_set
+#define __NR_ioprio_set 282
+#endif
+#ifndef __NR_ioprio_get
+#define __NR_ioprio_get 283
+#endif
+#ifndef __NR_openat
+#define __NR_openat 288
+#endif
+#ifndef __NR_newfstatat
+#define __NR_newfstatat 293
+#endif
+#ifndef __NR_unlinkat
+#define __NR_unlinkat 294
+#endif
+#ifndef __NR_move_pages
+#define __NR_move_pages 310
+#endif
+#ifndef __NR_getcpu
+#define __NR_getcpu 311
+#endif
+#ifndef __NR_fallocate
+#define __NR_fallocate 314
+#endif
+/* End of s390x definitions */
#endif
@@ -766,7 +919,7 @@ struct kernel_stat {
#undef LSS_RETURN
#if (defined(__i386__) || defined(__x86_64__) || defined(__arm__) || \
- defined(__aarch64__))
+ defined(__aarch64__) || defined(__s390x__))
/* Failing system calls return a negative result in the range of
* -1..-4095. These are "errno" values with the sign inverted.
*/
@@ -2083,6 +2236,116 @@ struct kernel_stat {
}
LSS_RETURN(int, __res);
}
+ #elif defined(__s390x__)
+ #undef LSS_REG
+ #define LSS_REG(r, a) register unsigned long __r##r __asm__("r"#r) = (unsigned long) a
+ #undef LSS_BODY
+ #define LSS_BODY(type, name, args...) \
+ register long __res_r2 __asm__("r2"); \
+ long __res; \
+ __asm__ __volatile__ \
+ ("lgfi %%r1, %1\n\t" \
+ "svc 0\n\t" \
+ : "=&r"(__res_r2) \
+ : "i"(__NR_##name), ## args \
+ : "r1", "memory"); \
+ __res = __res_r2; \
+ LSS_RETURN(type, __res)
+ #undef _syscall0
+ #define _syscall0(type, name) \
+ type LSS_NAME(name)(void) { \
+ LSS_BODY(type, name); \
+ }
+ #undef _syscall1
+ #define _syscall1(type, name, type1, arg1) \
+ type LSS_NAME(name)(type1 arg1) { \
+ LSS_REG(2, arg1); \
+ LSS_BODY(type, name, "0"(__r2)); \
+ }
+ #undef _syscall2
+ #define _syscall2(type, name, type1, arg1, type2, arg2) \
+ type LSS_NAME(name)(type1 arg1, type2 arg2) { \
+ LSS_REG(2, arg1); LSS_REG(3, arg2); \
+ LSS_BODY(type, name, "0"(__r2), "r"(__r3)); \
+ }
+ #undef _syscall3
+ #define _syscall3(type, name, type1, arg1, type2, arg2, type3, arg3) \
+ type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3) { \
+ LSS_REG(2, arg1); LSS_REG(3, arg2); LSS_REG(4, arg3); \
+ LSS_BODY(type, name, "0"(__r2), "r"(__r3), "r"(__r4)); \
+ }
+ #undef _syscall4
+ #define _syscall4(type, name, type1, arg1, type2, arg2, type3, arg3, \
+ type4, arg4) \
+ type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3, \
+ type4 arg4) { \
+ LSS_REG(2, arg1); LSS_REG(3, arg2); LSS_REG(4, arg3); \
+ LSS_REG(5, arg4); \
+ LSS_BODY(type, name, "0"(__r2), "r"(__r3), "r"(__r4), \
+ "r"(__r5)); \
+ }
+ #undef _syscall5
+ #define _syscall5(type, name, type1, arg1, type2, arg2, type3, arg3, \
+ type4, arg4, type5, arg5) \
+ type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3, \
+ type4 arg4, type5 arg5) { \
+ LSS_REG(2, arg1); LSS_REG(3, arg2); LSS_REG(4, arg3); \
+ LSS_REG(5, arg4); LSS_REG(6, arg5); \
+ LSS_BODY(type, name, "0"(__r2), "r"(__r3), "r"(__r4), \
+ "r"(__r5), "r"(__r6)); \
+ }
+ #undef _syscall6
+ #define _syscall6(type, name, type1, arg1, type2, arg2, type3, arg3, \
+ type4, arg4, type5, arg5, type6, arg6) \
+ type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3, \
+ type4 arg4, type5 arg5, type6 arg6) { \
+ LSS_REG(2, arg1); LSS_REG(3, arg2); LSS_REG(4, arg3); \
+ LSS_REG(5, arg4); LSS_REG(6, arg5); LSS_REG(7, arg6); \
+ LSS_BODY(type, name, "0"(__r2), "r"(__r3), "r"(__r4), \
+ "r"(__r5), "r"(__r6), "r"(__r7)); \
+ }
+ LSS_INLINE int LSS_NAME(clone)(int (*fn)(void *), void *child_stack,
+ int flags, void *arg, int *parent_tidptr,
+ void *newtls, int *child_tidptr) {
+ long __ret;
+ {
+ register int (*__fn)(void *) __asm__ ("r1") = fn;
+ register void *__cstack __asm__ ("r2") = child_stack;
+ register int __flags __asm__ ("r3") = flags;
+ register void *__arg __asm__ ("r0") = arg;
+ register int *__ptidptr __asm__ ("r4") = parent_tidptr;
+ register void *__newtls __asm__ ("r6") = newtls;
+ register int *__ctidptr __asm__ ("r5") = child_tidptr;
+ __asm__ __volatile__ (
+ /* arg already in r0 */
+ "ltgr %4, %4\n\t" /* check fn, which is already in r1 */
+ "jz 1f\n\t" /* NULL function pointer, return -EINVAL */
+ "ltgr %5, %5\n\t" /* check child_stack, which is already in r2 */
+ "jz 1f\n\t" /* NULL stack pointer, return -EINVAL */
+ /* flags already in r3 */
+ /* parent_tidptr already in r4 */
+ /* child_tidptr already in r5 */
+ /* newtls already in r6 */
+ "svc %2\n\t" /* invoke clone syscall */
+ "ltgr %0, %%r2\n\t" /* load return code into __ret and test */
+ "jnz 1f\n\t" /* return to parent if non-zero */
+ /* start child thread */
+ "lgr %%r2, %7\n\t" /* set first parameter to void *arg */
+ "aghi %%r15, -160\n\t" /* make room on the stack for the save area */
+ "xc 0(8,%%r15), 0(%%r15)\n\t"
+ "basr %%r14, %4\n\t" /* jump to fn */
+ "svc %3\n" /* invoke exit syscall */
+
+ "1:\n"
+ : "=r" (__ret)
+ : "0" (-EINVAL), "i" (__NR_clone), "i" (__NR_exit),
+ "r" (__fn), "r" (__cstack), "r" (__flags), "r" (__arg),
+ "r" (__ptidptr), "r" (__newtls), "r" (__ctidptr)
+ : "cc", "r14", "memory"
+ );
+ }
+ LSS_RETURN(int, __ret);
+ }
#endif
#define __NR__exit __NR_exit
#define __NR__gettid __NR_gettid
@@ -2165,7 +2428,10 @@ struct kernel_stat {
LSS_INLINE _syscall3(int, socket, int, d,
int, t, int, p)
#endif
- #if defined(__x86_64__)
+ #if defined(__x86_64__) || defined(__s390x__)
+ #if defined(__s390x__)
+ LSS_INLINE _syscall1(void*, mmap, void*, a)
+ #else
/* Need to make sure __off64_t isn't truncated to 32-bits under x32. */
LSS_INLINE void* LSS_NAME(mmap)(void *s, size_t l, int p, int f, int d,
__off64_t o) {
@@ -2173,10 +2439,12 @@ struct kernel_stat {
LSS_SYSCALL_ARG(p), LSS_SYSCALL_ARG(f),
LSS_SYSCALL_ARG(d), (uint64_t)(o));
}
+ #endif
LSS_INLINE int LSS_NAME(sigaction)(int signum,
const struct kernel_sigaction *act,
struct kernel_sigaction *oldact) {
+ #if defined(__x86_64__)
/* On x86_64, the kernel requires us to always set our own
* SA_RESTORER in order to be able to return from a signal handler.
* This function must have a "magic" signature that the "gdb"
@@ -2188,7 +2456,9 @@ struct kernel_stat {
a.sa_restorer = LSS_NAME(restore_rt)();
return LSS_NAME(rt_sigaction)(signum, &a, oldact,
(KERNEL_NSIG+7)/8);
- } else {
+ } else
+ #endif
+ {
return LSS_NAME(rt_sigaction)(signum, act, oldact,
(KERNEL_NSIG+7)/8);
}
@@ -2387,13 +2657,14 @@ struct kernel_stat {
#if defined(__i386__) || \
defined(__PPC__) || \
(defined(__arm__) && !defined(__ARM_EABI__)) || \
- (defined(__mips__) && _MIPS_SIM == _MIPS_SIM_ABI32)
+ (defined(__mips__) && _MIPS_SIM == _MIPS_SIM_ABI32) || \
+ defined(__s390x__)
/* See sys_socketcall in net/socket.c in kernel source.
* It de-multiplexes on its first arg and unpacks the arglist
* array in its second arg.
*/
- LSS_INLINE _syscall2(long, socketcall, int, c, unsigned long*, a)
+ LSS_INLINE _syscall2(int, socketcall, int, c, unsigned long*, a)
LSS_INLINE int LSS_NAME(socket)(int domain, int type, int protocol) {
unsigned long args[3] = {
diff --git a/src/base/linuxthreads.h b/src/base/linuxthreads.h
index 16bc8c6..82965af 100644
--- a/src/base/linuxthreads.h
+++ b/src/base/linuxthreads.h
@@ -41,7 +41,8 @@
* related platforms should not be difficult.
*/
#if (defined(__i386__) || defined(__x86_64__) || defined(__ARM_ARCH_3__) || \
- defined(__mips__) || defined(__PPC__) || defined(__aarch64__)) && defined(__linux)
+ defined(__mips__) || defined(__PPC__) || defined(__aarch64__) || \
+ defined(__s390x__)) && defined(__linux)
/* Define the THREADS symbol to make sure that there is exactly one core dumper
* built into the library.
diff --git a/src/malloc_hook_mmap_linux.h b/src/malloc_hook_mmap_linux.h
index 0f531db..1c4c766 100755
--- a/src/malloc_hook_mmap_linux.h
+++ b/src/malloc_hook_mmap_linux.h
@@ -119,6 +119,20 @@ static inline void* do_mmap64(void *start, size_t length,
#define MALLOC_HOOK_HAVE_DO_MMAP64 1
+#elif defined(__s390x__)
+
+static inline void* do_mmap64(void *start, size_t length,
+ int prot, int flags,
+ int fd, __off64_t offset) __THROW {
+ // mmap on s390x uses the old syscall interface
+ unsigned long args[6] = { (unsigned long) start, (unsigned long) length,
+ (unsigned long) prot, (unsigned long) flags,
+ (unsigned long) fd, (unsigned long) offset };
+ return sys_mmap(args);
+}
+
+#define MALLOC_HOOK_HAVE_DO_MMAP64 1
+
#endif // #if defined(__x86_64__)