commit ffeab18bc89961b48ef51d300af5a2296dec67a1
Author: Stacey Son <sson@FreeBSD.org>
Date:   Sat Feb 8 15:28:50 2014 -0600

    Use the 'rdhwr $29' instruction instead of sysarch(MIPS_GET_TLS,..).
    
    This change uses the 'rdhwr $29' instruction to retrieve the TLS pointer
    instead of using the sysarch(MIPS_GET_TLS,...) system call.  For hardware
    that supports this instruction this should be faster since it doesn't
    require the overhead of a system call.  For hardware that doesn't support
    this instruction the 'rdhwr $29' instruction is emulated by the trap()
    handler in sys/mips/mips/trap.c and the overhead should be about the
    same. (Actually, some quick micro benchmarks indicate that emulating the
    'rdhwr' instruction is slightly faster than the sysarch() system call.)

diff --git a/lib/libthr/arch/mips/include/pthread_md.h b/lib/libthr/arch/mips/include/pthread_md.h
index 3ba5818..b7c70db 100644
--- a/lib/libthr/arch/mips/include/pthread_md.h
+++ b/lib/libthr/arch/mips/include/pthread_md.h
@@ -67,6 +67,7 @@ _tcb_set(struct tcb *tcb)
 /*
  * Get the current tcb.
  */
+#ifdef TLS_USE_SYSARCH
 static __inline struct tcb *
 _tcb_get(void)
 {
@@ -76,6 +77,55 @@ _tcb_get(void)
 	return tcb;
 }
 
+#else /* ! TLS_USE_SYSARCH */
+
+#  if defined(__mips_n64)
+static __inline struct tcb *
+_tcb_get(void)
+{
+	uint64_t _rv;
+
+	__asm__ __volatile__ (
+	    ".set\tpush\n\t"
+	    ".set\tmips64r2\n\t"
+	    "rdhwr\t%0, $29\n\t"
+	    ".set\tpop"
+	    : "=v" (_rv));
+
+	/*
+	 * XXXSS See 'git show c6be4f4d2d1b71c04de5d3bbb6933ce2dbcdb317'
+	 *
+	 * Remove the offset since this really a request to get the TLS
+	 * pointer via sysarch() (in theory).  Of course, this may go away
+	 * once the TLS code is rewritten.
+	 */
+	return (struct tcb *)(_rv - TLS_TP_OFFSET - TLS_TCB_SIZE);
+}
+#  else /* mips 32 */
+static __inline struct tcb *
+_tcb_get(void)
+{
+	uint32_t _rv;
+
+	__asm__ __volatile__ (
+	    ".set\tpush\n\t"
+	    ".set\tmips32r2\n\t"
+	    "rdhwr\t%0, $29\n\t"
+	    ".set\tpop"
+	    : "=v" (_rv));
+
+	/*
+	 * XXXSS See 'git show c6be4f4d2d1b71c04de5d3bbb6933ce2dbcdb317'
+	 *
+	 * Remove the offset since this really a request to get the TLS
+	 * pointer via sysarch() (in theory).  Of course, this may go away
+	 * once the TLS code is rewritten.
+	 */
+	return (struct tcb *)(_rv - TLS_TP_OFFSET - TLS_TCB_SIZE32);
+}
+#  endif /* ! __mips_n64 */
+#endif /* ! TLS_USE_SYSARCH */
+
 extern struct pthread *_thr_initial;
 
 static __inline struct pthread *
diff --git a/libexec/rtld-elf/mips/reloc.c b/libexec/rtld-elf/mips/reloc.c
index 02291ce..e64ee63 100644
--- a/libexec/rtld-elf/mips/reloc.c
+++ b/libexec/rtld-elf/mips/reloc.c
@@ -630,13 +630,67 @@ allocate_initial_tls(Obj_Entry *objs)
 	sysarch(MIPS_SET_TLS, tls);
 }
 
+#ifdef __mips_n64
+void *
+_mips_get_tls(void)
+{
+	uint64_t _rv;
+
+	__asm__ __volatile__ (
+	    ".set\tpush\n\t"
+	    ".set\tmips64r2\n\t"
+	    "rdhwr\t%0, $29\n\t"
+	    ".set\tpop"
+	    : "=v" (_rv));
+	/*
+	 * XXXSS See 'git show c6be4f4d2d1b71c04de5d3bbb6933ce2dbcdb317'
+	 *
+	 * Remove the offset since this really a request to get the TLS
+	 * pointer via sysarch() (in theory).  Of course, this may go away
+	 * once the TLS code is rewritten.
+	 */
+	_rv = _rv - TLS_TP_OFFSET - TLS_TCB_SIZE;
+
+	return (void *)_rv;
+}
+
+#else /* mips 32 */
+
+void *
+_mips_get_tls(void)
+{
+	uint32_t _rv;
+
+	__asm__ __volatile__ (
+	    ".set\tpush\n\t"
+	    ".set\tmips32r2\n\t"
+	    "rdhwr\t%0, $29\n\t"
+	    ".set\tpop"
+	    : "=v" (_rv));
+	/*
+	 * XXXSS See 'git show c6be4f4d2d1b71c04de5d3bbb6933ce2dbcdb317'
+	 *
+	 * Remove the offset since this really a request to get the TLS
+	 * pointer via sysarch() (in theory).  Of course, this may go away
+	 * once the TLS code is rewritten.
+	 */
+	_rv = _rv - TLS_TP_OFFSET - TLS_TCB_SIZE32;
+
+	return (void *)_rv;
+}
+#endif /* ! __mips_n64 */
+
 void *
 __tls_get_addr(tls_index* ti)
 {
 	Elf_Addr** tls;
 	char *p;
 
+#ifdef TLS_USE_SYSARCH
 	sysarch(MIPS_GET_TLS, &tls);
+#else
+	tls = _mips_get_tls();
+#endif
 
 	p = tls_get_addr_common(tls, ti->ti_module, ti->ti_offset + TLS_DTP_OFFSET);