x86/percpu: Introduce %rip-relative addressing to PER_CPU_VAR()

Introduce x86_64 %rip-relative addressing to the PER_CPU_VAR() macro.
Instructions using %rip-relative address operand are one byte shorter
than their absolute address counterparts and are also compatible with
position independent executable (-fpie) builds. The patch reduces
code size of a test kernel build by 150 bytes.

The PER_CPU_VAR() macro is intended to be applied to a symbol and should
not be used with register operands. Introduce the new __percpu macro and
use it in cmpxchg{8,16}b_emu.S instead.

Also add a missing function comment to this_cpu_cmpxchg8b_emu().

No functional changes intended.

Signed-off-by: Uros Bizjak <ubizjak@gmail.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: linux-kernel@vger.kernel.org
Cc: Brian Gerst <brgerst@gmail.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Sean Christopherson <seanjc@google.com>

Author: ubizjak

arch/x86/include/asm/percpu.h

@@ -4,17 +4,21 @@
 
 #ifdef CONFIG_X86_64
 #define __percpu_seg		gs
+#define __percpu_rel		(%rip)
 #else
 #define __percpu_seg		fs
+#define __percpu_rel
 #endif
 
 #ifdef __ASSEMBLY__
 
 #ifdef CONFIG_SMP
-#define PER_CPU_VAR(var)	%__percpu_seg:var
-#else /* ! SMP */
-#define PER_CPU_VAR(var)	var
-#endif	/* SMP */
+#define __percpu		%__percpu_seg:
+#else
+#define __percpu
+#endif
+
+#define PER_CPU_VAR(var)	__percpu(var)__percpu_rel
 
 #ifdef CONFIG_X86_64_SMP
 #define INIT_PER_CPU_VAR(var)  init_per_cpu__##var

arch/x86/lib/cmpxchg16b_emu.S

@@ -23,14 +23,14 @@ SYM_FUNC_START(this_cpu_cmpxchg16b_emu)
 	cli
 
 	/* if (*ptr == old) */
-	cmpq	PER_CPU_VAR(0(%rsi)), %rax
+	cmpq	__percpu (%rsi), %rax
 	jne	.Lnot_same
-	cmpq	PER_CPU_VAR(8(%rsi)), %rdx
+	cmpq	__percpu 8(%rsi), %rdx
 	jne	.Lnot_same
 
 	/* *ptr = new */
-	movq	%rbx, PER_CPU_VAR(0(%rsi))
-	movq	%rcx, PER_CPU_VAR(8(%rsi))
+	movq	%rbx, __percpu (%rsi)
+	movq	%rcx, __percpu 8(%rsi)
 
 	/* set ZF in EFLAGS to indicate success */
 	orl	$X86_EFLAGS_ZF, (%rsp)
@@ -42,8 +42,8 @@ SYM_FUNC_START(this_cpu_cmpxchg16b_emu)
 	/* *ptr != old */
 
 	/* old = *ptr */
-	movq	PER_CPU_VAR(0(%rsi)), %rax
-	movq	PER_CPU_VAR(8(%rsi)), %rdx
+	movq	__percpu (%rsi), %rax
+	movq	__percpu 8(%rsi), %rdx
 
 	/* clear ZF in EFLAGS to indicate failure */
 	andl	$(~X86_EFLAGS_ZF), (%rsp)

arch/x86/lib/cmpxchg8b_emu.S

@@ -24,12 +24,12 @@ SYM_FUNC_START(cmpxchg8b_emu)
 	pushfl
 	cli
 
-	cmpl	0(%esi), %eax
+	cmpl	(%esi), %eax
 	jne	.Lnot_same
 	cmpl	4(%esi), %edx
 	jne	.Lnot_same
 
-	movl	%ebx, 0(%esi)
+	movl	%ebx, (%esi)
 	movl	%ecx, 4(%esi)
 
 	orl	$X86_EFLAGS_ZF, (%esp)
@@ -38,7 +38,7 @@ SYM_FUNC_START(cmpxchg8b_emu)
 	RET
 
 .Lnot_same:
-	movl	0(%esi), %eax
+	movl	(%esi), %eax
 	movl	4(%esi), %edx
 
 	andl	$(~X86_EFLAGS_ZF), (%esp)
@@ -53,27 +53,39 @@ EXPORT_SYMBOL(cmpxchg8b_emu)
 
 #ifndef CONFIG_UML
 
+/*
+ * Emulate 'cmpxchg8b %fs:(%rsi)'
+ *
+ * Inputs:
+ * %esi : memory location to compare
+ * %eax : low 32 bits of old value
+ * %edx : high 32 bits of old value
+ * %ebx : low 32 bits of new value
+ * %ecx : high 32 bits of new value
+ *
+ * Notably this is not LOCK prefixed and is not safe against NMIs
+ */
 SYM_FUNC_START(this_cpu_cmpxchg8b_emu)
 
 	pushfl
 	cli
 
-	cmpl	PER_CPU_VAR(0(%esi)), %eax
+	cmpl	__percpu (%esi), %eax
 	jne	.Lnot_same2
-	cmpl	PER_CPU_VAR(4(%esi)), %edx
+	cmpl	__percpu 4(%esi), %edx
 	jne	.Lnot_same2
 
-	movl	%ebx, PER_CPU_VAR(0(%esi))
-	movl	%ecx, PER_CPU_VAR(4(%esi))
+	movl	%ebx, __percpu (%esi)
+	movl	%ecx, __percpu 4(%esi)
 
 	orl	$X86_EFLAGS_ZF, (%esp)
 
 	popfl
 	RET
 
 .Lnot_same2:
-	movl	PER_CPU_VAR(0(%esi)), %eax
-	movl	PER_CPU_VAR(4(%esi)), %edx
+	movl	__percpu (%esi), %eax
+	movl	__percpu 4(%esi), %edx
 
 	andl	$(~X86_EFLAGS_ZF), (%esp)