add powerpc64 port

src/fenv/powerpc64/fenv.c

@@ -0,0 +1,68 @@
+#define _GNU_SOURCE
+#include <fenv.h>
+
+static inline double get_fpscr_f(void)
+{
+	double d;
+	__asm__ __volatile__("mffs %0" : "=d"(d));
+	return d;
+}
+
+static inline long get_fpscr(void)
+{
+	return (union {double f; long i;}) {get_fpscr_f()}.i;
+}
+
+static inline void set_fpscr_f(double fpscr)
+{
+	__asm__ __volatile__("mtfsf 255, %0" : : "d"(fpscr));
+}
+
+static void set_fpscr(long fpscr)
+{
+	set_fpscr_f((union {long i; double f;}) {fpscr}.f);
+}
+
+int feclearexcept(int mask)
+{
+	mask &= FE_ALL_EXCEPT;
+	if (mask & FE_INVALID) mask |= FE_ALL_INVALID;
+	set_fpscr(get_fpscr() & ~mask);
+	return 0;
+}
+
+int feraiseexcept(int mask)
+{
+	mask &= FE_ALL_EXCEPT;
+	if (mask & FE_INVALID) mask |= FE_INVALID_SOFTWARE;
+	set_fpscr(get_fpscr() | mask);
+	return 0;
+}
+
+int fetestexcept(int mask)
+{
+	return get_fpscr() & mask & FE_ALL_EXCEPT;
+}
+
+int fegetround(void)
+{
+	return get_fpscr() & 3;
+}
+
+int __fesetround(int r)
+{
+	set_fpscr(get_fpscr() & ~3L | r);
+	return 0;
+}
+
+int fegetenv(fenv_t *envp)
+{
+	*envp = get_fpscr_f();
+	return 0;
+}
+
+int fesetenv(const fenv_t *envp)
+{
+	set_fpscr_f(envp != FE_DFL_ENV ? *envp : 0);
+	return 0;
+}

src/internal/powerpc64/syscall.s

@@ -0,0 +1,17 @@
+	.global __syscall
+	.hidden __syscall
+	.type   __syscall,@function
+__syscall:
+	mr      0, 3                  # Save the system call number
+	mr      3, 4                  # Shift the arguments: arg1
+	mr      4, 5                  # arg2
+	mr      5, 6                  # arg3
+	mr      6, 7                  # arg4
+	mr      7, 8                  # arg5
+	mr      8, 9                  # arg6
+	sc
+	bnslr+       # return if not summary overflow
+	neg     3, 3 # otherwise error: return negated value.
+	blr
+	.end    __syscall
+	.size   __syscall, .-__syscall

src/ldso/powerpc64/dlsym.s

@@ -0,0 +1,12 @@
+	.text
+	.global dlsym
+	.hidden __dlsym
+	.type   dlsym,@function
+dlsym:
+	addis   2, 12, .TOC.-dlsym@ha
+	addi    2,  2, .TOC.-dlsym@l
+	.localentry dlsym,.-dlsym
+	mflr    5                      # The return address is arg3.
+	b       __dlsym
+	.end    dlsym
+	.size   dlsym, .-dlsym

src/setjmp/powerpc64/longjmp.s

@@ -0,0 +1,77 @@
+	.global _longjmp
+	.global longjmp
+	.type   _longjmp,@function
+	.type   longjmp,@function
+_longjmp:
+longjmp:
+	# 0) move old return address into the link register
+	ld   0,  0*8(3)
+	mtlr 0
+	# 1) restore cr
+	ld   0,  1*8(3)
+	mtcr 0
+	# 2) restore r1-r2 (SP and TOC)
+	ld   1,  2*8(3)
+	ld   2,  3*8(3)
+	# 3) restore r14-r31
+	ld  14,  4*8(3)
+	ld  15,  5*8(3)
+	ld  16,  6*8(3)
+	ld  17,  7*8(3)
+	ld  18,  8*8(3)
+	ld  19,  9*8(3)
+	ld  20, 10*8(3)
+	ld  21, 11*8(3)
+	ld  22, 12*8(3)
+	ld  23, 13*8(3)
+	ld  24, 14*8(3)
+	ld  25, 15*8(3)
+	ld  26, 16*8(3)
+	ld  27, 17*8(3)
+	ld  28, 18*8(3)
+	ld  29, 19*8(3)
+	ld  30, 20*8(3)
+	ld  31, 21*8(3)
+	# 4) restore floating point registers f14-f31
+	lfd 14, 22*8(3)
+	lfd 15, 23*8(3)
+	lfd 16, 24*8(3)
+	lfd 17, 25*8(3)
+	lfd 18, 26*8(3)
+	lfd 19, 27*8(3)
+	lfd 20, 28*8(3)
+	lfd 21, 29*8(3)
+	lfd 22, 30*8(3)
+	lfd 23, 31*8(3)
+	lfd 24, 32*8(3)
+	lfd 25, 33*8(3)
+	lfd 26, 34*8(3)
+	lfd 27, 35*8(3)
+	lfd 28, 36*8(3)
+	lfd 29, 37*8(3)
+	lfd 30, 38*8(3)
+	lfd 31, 39*8(3)
+
+	# 5) restore vector registers v20-v31
+	addi 3, 3, 40*8
+	lvx 20, 0, 3 ; addi 3, 3, 16
+	lvx 21, 0, 3 ; addi 3, 3, 16
+	lvx 22, 0, 3 ; addi 3, 3, 16
+	lvx 23, 0, 3 ; addi 3, 3, 16
+	lvx 24, 0, 3 ; addi 3, 3, 16
+	lvx 25, 0, 3 ; addi 3, 3, 16
+	lvx 26, 0, 3 ; addi 3, 3, 16
+	lvx 27, 0, 3 ; addi 3, 3, 16
+	lvx 28, 0, 3 ; addi 3, 3, 16
+	lvx 29, 0, 3 ; addi 3, 3, 16
+	lvx 30, 0, 3 ; addi 3, 3, 16
+	lvx 31, 0, 3
+
+	# 6) return r4 ? r4 : 1
+	mr    3,   4
+	cmpwi cr7, 4, 0
+	bne   cr7, 1f
+	li    3,   1
+1:
+	blr
+

src/setjmp/powerpc64/setjmp.s

@@ -0,0 +1,78 @@
+	.global ___setjmp
+	.hidden ___setjmp
+	.global __setjmp
+	.global _setjmp
+	.global setjmp
+	.type   __setjmp,@function
+	.type   _setjmp,@function
+	.type   setjmp,@function
+___setjmp:
+__setjmp:
+_setjmp:
+setjmp:
+	# 0) store IP into 0, then into the jmpbuf pointed to by r3 (first arg)
+	mflr  0
+	std   0,  0*8(3)
+	# 1) store cr
+	mfcr  0
+	std   0,  1*8(3)
+	# 2) store r1-r2 (SP and TOC)
+	std   1,  2*8(3)
+	std   2,  3*8(3)
+	# 3) store r14-31
+	std  14,  4*8(3)
+	std  15,  5*8(3)
+	std  16,  6*8(3)
+	std  17,  7*8(3)
+	std  18,  8*8(3)
+	std  19,  9*8(3)
+	std  20, 10*8(3)
+	std  21, 11*8(3)
+	std  22, 12*8(3)
+	std  23, 13*8(3)
+	std  24, 14*8(3)
+	std  25, 15*8(3)
+	std  26, 16*8(3)
+	std  27, 17*8(3)
+	std  28, 18*8(3)
+	std  29, 19*8(3)
+	std  30, 20*8(3)
+	std  31, 21*8(3)
+	# 4) store floating point registers f14-f31
+	stfd 14, 22*8(3)
+	stfd 15, 23*8(3)
+	stfd 16, 24*8(3)
+	stfd 17, 25*8(3)
+	stfd 18, 26*8(3)
+	stfd 19, 27*8(3)
+	stfd 20, 28*8(3)
+	stfd 21, 29*8(3)
+	stfd 22, 30*8(3)
+	stfd 23, 31*8(3)
+	stfd 24, 32*8(3)
+	stfd 25, 33*8(3)
+	stfd 26, 34*8(3)
+	stfd 27, 35*8(3)
+	stfd 28, 36*8(3)
+	stfd 29, 37*8(3)
+	stfd 30, 38*8(3)
+	stfd 31, 39*8(3)
+
+	# 5) store vector registers v20-v31
+	addi  3, 3, 40*8
+	stvx 20, 0, 3 ; addi 3, 3, 16
+	stvx 21, 0, 3 ; addi 3, 3, 16
+	stvx 22, 0, 3 ; addi 3, 3, 16
+	stvx 23, 0, 3 ; addi 3, 3, 16
+	stvx 24, 0, 3 ; addi 3, 3, 16
+	stvx 25, 0, 3 ; addi 3, 3, 16
+	stvx 26, 0, 3 ; addi 3, 3, 16
+	stvx 27, 0, 3 ; addi 3, 3, 16
+	stvx 28, 0, 3 ; addi 3, 3, 16
+	stvx 29, 0, 3 ; addi 3, 3, 16
+	stvx 30, 0, 3 ; addi 3, 3, 16
+	stvx 31, 0, 3
+
+	# 6) return 0
+	li 3, 0
+	blr

src/signal/powerpc64/restore.s

@@ -0,0 +1,11 @@
+	.global __restore
+	.type __restore,%function
+__restore:
+	li      0, 119 #__NR_sigreturn
+	sc
+
+	.global __restore_rt
+	.type __restore_rt,%function
+__restore_rt:
+	li      0, 172 # __NR_rt_sigreturn
+	sc

src/signal/powerpc64/sigsetjmp.s

@@ -0,0 +1,30 @@
+	.global sigsetjmp
+	.global __sigsetjmp
+	.type sigsetjmp,%function
+	.type __sigsetjmp,%function
+	.hidden ___setjmp
+sigsetjmp:
+__sigsetjmp:
+	addis 2, 12, .TOC.-__sigsetjmp@ha
+	addi  2,  2, .TOC.-__sigsetjmp@l
+	.localentry sigsetjmp,.-sigsetjmp
+	.localentry __sigsetjmp,.-__sigsetjmp
+
+	cmpwi cr7, 4, 0
+	beq-  cr7, ___setjmp
+
+	mflr  5
+	std   5, 512(3)
+	std  16, 512+8+8(3)
+	mr   16, 3
+
+	bl ___setjmp
+
+	mr   4,  3
+	mr   3, 16
+	ld   5, 512(3)
+	mtlr 5
+	ld  16, 512+8+8(3)
+
+.hidden __sigsetjmp_tail
+	b __sigsetjmp_tail

src/thread/powerpc64/__set_thread_area.s

@@ -0,0 +1,8 @@
+.text
+.global __set_thread_area
+.type   __set_thread_area, %function
+__set_thread_area:
+	mr 13, 3
+	li  3, 0
+	blr
+

src/thread/powerpc64/__unmapself.s

@@ -0,0 +1,9 @@
+	.text
+	.global __unmapself
+	.type   __unmapself,%function
+__unmapself:
+	li      0, 91 # __NR_munmap
+	sc
+	li      0, 1 #__NR_exit
+	sc
+	blr

src/thread/powerpc64/clone.s

@@ -0,0 +1,47 @@
+.text
+.global __clone
+.type __clone, %function
+__clone:
+	# int clone(fn, stack, flags, arg, ptid, tls, ctid)
+	#            a  b       c     d     e    f    g
+	#            3  4       5     6     7    8    9
+	# pseudo C code:
+	# tid = syscall(SYS_clone,c,b,e,f,g);
+	# if (!tid) syscall(SYS_exit, a(d));
+	# return tid;
+
+	# create initial stack frame for new thread
+	clrrdi 4, 4, 4
+	li     0, 0
+	stdu   0,-32(4)
+
+	# save fn and arg to child stack
+	std    3,  8(4)
+	std    6, 16(4)
+
+	# shuffle args into correct registers and call SYS_clone
+	mr    3, 5
+	#mr   4, 4
+	mr    5, 7
+	mr    6, 8
+	mr    7, 9
+	li    0, 120  # SYS_clone = 120
+	sc
+
+	# if error, negate return (errno)
+	bns+  1f
+	neg   3, 3
+
+1:	# if we're the parent, return
+	cmpwi cr7, 3, 0
+	bnelr cr7
+
+	# we're the child. call fn(arg)
+	ld     3, 16(1)
+	ld    12,  8(1)
+	mtctr 12
+	bctrl
+
+	# call SYS_exit. exit code is already in r3 from fn return value
+	li    0, 1    # SYS_exit = 1
+	sc

src/thread/powerpc64/syscall_cp.s

@@ -0,0 +1,37 @@
+	.global __cp_begin
+	.hidden __cp_begin
+	.global __cp_end
+	.hidden __cp_end
+	.global __cp_cancel
+	.hidden __cp_cancel
+	.hidden __cancel
+	.global __syscall_cp_asm
+	.hidden __syscall_cp_asm
+	.text
+	.type   __syscall_cp_asm,%function
+__syscall_cp_asm:
+	# at enter: r3 = pointer to self->cancel, r4: syscall no, r5: first arg, r6: 2nd, r7: 3rd, r8: 4th, r9: 5th, r10: 6th
+__cp_begin:
+	# if (self->cancel) goto __cp_cancel
+	lwz   0, 0(3)
+	cmpwi cr7, 0, 0
+	bne-  cr7, __cp_cancel
+
+	# make syscall
+	mr    0,  4
+	mr    3,  5
+	mr    4,  6
+	mr    5,  7
+	mr    6,  8
+	mr    7,  9
+	mr    8, 10
+	sc
+
+__cp_end:
+	# return error ? -r3 : r3
+	bnslr+
+	neg 3, 3
+	blr
+
+__cp_cancel:
+	b __cancel