Add support for TLS in the libc

6 files changed, 182 insertions, 0 deletions

diff --git a/lib/libc/internal/arch/x86_64/Kbuild b/lib/libc/internal/arch/x86_64/Kbuild
new file mode 100644
index 00000000..830a1c5b
--- /dev/null
+++ b/lib/libc/internal/arch/x86_64/Kbuild
@@ -0,0 +1,5 @@
+obj-y += _start.o
+obj-y += fenv.o
+obj-y += longjmp.o
+obj-y += setjmp.o
+obj-y += sigsetjmp.o
diff --git a/lib/libc/internal/arch/x86_64/_start.c b/lib/libc/internal/arch/x86_64/_start.c
new file mode 100644
index 00000000..f3f4fef2
--- /dev/null
+++ b/lib/libc/internal/arch/x86_64/_start.c
@@ -0,0 +1,8 @@
+#include <libc.h>
+#include <sys/cdefs.h>
+
+__dead __naked void _start(void)
+{
+	__asm__ __volatile__("mov %rsp, %rdi\n"
+			     "call __init\n");
+}
diff --git a/lib/libc/internal/arch/x86_64/fenv.s b/lib/libc/internal/arch/x86_64/fenv.s
new file mode 100644
index 00000000..b9a46581
--- /dev/null
+++ b/lib/libc/internal/arch/x86_64/fenv.s
@@ -0,0 +1,97 @@
+.global feclearexcept
+.type feclearexcept,@function
+feclearexcept:
+	mov %edi,%ecx
+	and $0x3f,%ecx
+	fnstsw %ax
+	test %eax,%ecx
+	jz 1f
+	fnclex
+1:	stmxcsr -8(%rsp)
+	and $0x3f,%eax
+	or %eax,-8(%rsp)
+	test %ecx,-8(%rsp)
+	jz 1f
+	not %ecx
+	and %ecx,-8(%rsp)
+	ldmxcsr -8(%rsp)
+1:	xor %eax,%eax
+	ret
+
+.global feraiseexcept
+.type feraiseexcept,@function
+feraiseexcept:
+	and $0x3f,%edi
+	stmxcsr -8(%rsp)
+	or %edi,-8(%rsp)
+	ldmxcsr -8(%rsp)
+	xor %eax,%eax
+	ret
+
+.global __fesetround
+.hidden __fesetround
+.type __fesetround,@function
+__fesetround:
+	push %rax
+	xor %eax,%eax
+	mov %edi,%ecx
+	fnstcw (%rsp)
+	andb $0xf3,1(%rsp)
+	or %ch,1(%rsp)
+	fldcw (%rsp)
+	stmxcsr (%rsp)
+	shl $3,%ch
+	andb $0x9f,1(%rsp)
+	or %ch,1(%rsp)
+	ldmxcsr (%rsp)
+	pop %rcx
+	ret
+
+.global fegetround
+.type fegetround,@function
+fegetround:
+	push %rax
+	stmxcsr (%rsp)
+	pop %rax
+	shr $3,%eax
+	and $0xc00,%eax
+	ret
+
+.global fegetenv
+.type fegetenv,@function
+fegetenv:
+	xor %eax,%eax
+	fnstenv (%rdi)
+	stmxcsr 28(%rdi)
+	ret
+
+.global fesetenv
+.type fesetenv,@function
+fesetenv:
+	xor %eax,%eax
+	inc %rdi
+	jz 1f
+	fldenv -1(%rdi)
+	ldmxcsr 27(%rdi)
+	ret
+1:	push %rax
+	push %rax
+	pushq $0xffff
+	pushq $0x37f
+	fldenv (%rsp)
+	pushq $0x1f80
+	ldmxcsr (%rsp)
+	add $40,%rsp
+	ret
+
+.global fetestexcept
+.type fetestexcept,@function
+fetestexcept:
+	and $0x3f,%edi
+	push %rax
+	stmxcsr (%rsp)
+	pop %rsi
+	fnstsw %ax
+	or %esi,%eax
+	and %edi,%eax
+	ret
diff --git a/lib/libc/internal/arch/x86_64/longjmp.c b/lib/libc/internal/arch/x86_64/longjmp.c
new file mode 100644
index 00000000..dbd18632
--- /dev/null
+++ b/lib/libc/internal/arch/x86_64/longjmp.c
@@ -0,0 +1,20 @@
+#include "bits/setjmp.h" // for jmp_buf
+
+#include <setjmp.h> // for longjmp
+
+__attribute__((noreturn, naked, returns_twice)) void longjmp(jmp_buf env,
+							     int val)
+{
+	__asm__ __volatile__("mov 0(%rdi), %rbx\n"
+			     "mov 8(%rdi), %rbp\n"
+			     "mov 16(%rdi), %r12\n"
+			     "mov 24(%rdi), %r13\n"
+			     "mov 32(%rdi), %r14\n"
+			     "mov 40(%rdi), %r15\n"
+			     "mov 48(%rdi), %rsp\n"
+			     "xor %rax, %rax\n"
+			     "cmp $0, %esi\n"
+			     "setne %al\n"
+			     "mov %eax, %eax\n"
+			     "jmp *56(%rdi)\n");
+}
diff --git a/lib/libc/internal/arch/x86_64/setjmp.c b/lib/libc/internal/arch/x86_64/setjmp.c
new file mode 100644
index 00000000..ea205e16
--- /dev/null
+++ b/lib/libc/internal/arch/x86_64/setjmp.c
@@ -0,0 +1,19 @@
+#include "bits/setjmp.h" // for jmp_buf
+
+#include <setjmp.h> // for setjmp
+
+__attribute__((naked, returns_twice)) int setjmp(jmp_buf env)
+{
+	__asm__ __volatile__("mov %rbx,(%rdi)\n"
+			     "mov %rbp,8(%rdi)\n"
+			     "mov %r12,16(%rdi)\n"
+			     "mov %r13,24(%rdi)\n"
+			     "mov %r14,32(%rdi)\n"
+			     "mov %r15,40(%rdi)\n"
+			     "lea 8(%rsp),%rdx\n"
+			     "mov %rdx,48(%rdi)\n"
+			     "mov (%rsp),%rdx\n"
+			     "mov %rdx,56(%rdi)\n"
+			     "xor %eax,%eax\n"
+			     "ret\n");
+}
diff --git a/lib/libc/internal/arch/x86_64/sigsetjmp.c b/lib/libc/internal/arch/x86_64/sigsetjmp.c
new file mode 100644
index 00000000..2d900ce7
--- /dev/null
+++ b/lib/libc/internal/arch/x86_64/sigsetjmp.c
@@ -0,0 +1,33 @@
+#include <setjmp.h> // for sigsetjmp, sigjmp_buf
+
+int sigsetjmp(sigjmp_buf env, int savemask)
+{
+	__asm__ __volatile__("movq %%rbx, 0(%0)\n"
+			     "movq %%rbp, 8(%0)\n"
+			     "movq %%r12, 16(%0)\n"
+			     "movq %%r13, 24(%0)\n"
+			     "movq %%r14, 32(%0)\n"
+			     "movq %%r15, 40(%0)\n"
+
+			     "leaq 8(%%rsp), %%rax\n"
+			     "movq %%rax, 48(%0)\n"
+			     "movq (%%rsp), %%rax\n"
+			     "movq %%rax, 56(%0)\n"
+
+			     "testl %%esi, %%esi\n"
+			     "jz 1f\n"
+			     "movq $2, %%rdi\n"
+			     "movq $0, %%rsi\n"
+			     "leaq 64(%0), %%rdx\n"
+			     "movq $128, %%r10\n"
+			     "movq $14, %%rax\n"
+			     "syscall\n"
+
+			     "1:\n"
+			     "xor %%eax, %%eax\n"
+			     :
+			     : "r"(env), "S"(savemask)
+			     : "rax", "rdi", "rdx", "r10", "memory");
+
+	return 0;
+}