[PATCH v2 16/20] arm64: percpu: Implement preemptible return RMW ops

Use the PCPU GPR infrastructure to implement all of the RMW ops which
return a value.

Test case:

| u64 outline_this_cpu_add_return_u64(u64 __percpu *p, u64 v)
| {
| 	return this_cpu_add_return(*p, v);
| }

Generated code before this patch (v7.2-rc4):

| <outline_this_cpu_add_return_u64>:
|        paciasp
|        stp     x29, x30, [sp, #-32]!
|        mrs     x3, sp_el0
|        mov     x29, sp
|        ldr     w2, [x3, #8]
|        add     w2, w2, #0x1
|        str     w2, [x3, #8]
|        mov     x2, x0
|        mrs     x4, tpidr_el1
|        add     x2, x2, x4
| 1:     ldxr    x0, [x2]
|        add     x0, x0, x1
|        stxr    w5, x0, [x2]
|        cbnz    w5, 1b
|        ldr     x1, [x3, #8]
|        sub     x1, x1, #0x1
|        str     w1, [x3, #8]
|        cbz     x1, 2f
|        ldr     x1, [x3, #8]
|        cbnz    x1, 3f
| 2:     str     x0, [sp, #24]
|        bl      0 <preempt_schedule_notrace>
|        ldr     x0, [sp, #24]
| 3:     ldp     x29, x30, [sp], #32
|        autiasp
|        ret

Generated code after this patch:

| <outline_this_cpu_add_return_u64>:
|        mrs     x3, sp_el0
|        mov     w5, #0x10a0
|        strh    w5, [x3, #20]
|        mrs     x5, tpidr_el1
|        add     x4, x0, x5
| 1:     ldxr    x2, [x4]
|        add     x2, x2, x1
|        stxr    w6, x2, [x4]
|        cbnz    w6, 1b
|        strh    wzr, [x3, #20]
|        mov     x0, x2
|        ret

Signed-off-by: Mark Rutland <[email protected]>
Cc: Ada Couprie Diaz <[email protected]>
Cc: Ard Biesheuvel <[email protected]>
Cc: Catalin Marinas <[email protected]>
Cc: James Morse <[email protected]>
Cc: Jinjie Ruan <[email protected]>
Cc: Marc Zyngier <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Vladimir Murzin <[email protected]>
Cc: Will Deacon <[email protected]>
Cc: Yang Shi <[email protected]>
---
 arch/arm64/include/asm/percpu.h | 36 ++++++++++++++++++++++-----------
 1 file changed, 24 insertions(+), 12 deletions(-)

diff --git a/arch/arm64/include/asm/percpu.h b/arch/arm64/include/asm/percpu.h
index 505926363d754..8e0862704095e 100644
--- a/arch/arm64/include/asm/percpu.h
+++ b/arch/arm64/include/asm/percpu.h
@@ -201,24 +201,36 @@ __percpu_##name##_case_##sz(void __percpu *pcp, unsigned long val)	\
 
 #define __PERCPU_RET_OP_CASE(w, sfx, name, sz, op_llsc, op_lse)		\
 static inline u##sz							\
-__percpu_##name##_return_case_##sz(void *ptr, unsigned long val)	\
+__percpu_##name##_return_case_##sz(void __percpu *pcp, unsigned long val)	\
 {									\
+	u16 *gprs = &current_thread_info()->pcpu_gprs;			\
+	unsigned long addr;						\
+	unsigned long off;						\
 	unsigned int loop;						\
 	u##sz ret;							\
 									\
-	asm volatile (ARM64_LSE_ATOMIC_INSN(				\
+	asm volatile (							\
+	__PCPU_GPRS_BEGIN("%[gprs]", "%[pcp]", "%[off]", "%[addr]")	\
+	ARM64_LSE_ATOMIC_INSN(						\
 	/* LL/SC */							\
-	"1:	ldxr" #sfx "\t%" #w "[ret], %[ptr]\n"			\
+	"1:	ldxr" #sfx "\t%" #w "[ret], [%[addr]]\n"		\
 		#op_llsc "\t%" #w "[ret], %" #w "[ret], %" #w "[val]\n"	\
-	"	stxr" #sfx "\t%w[loop], %" #w "[ret], %[ptr]\n"		\
+	"	stxr" #sfx "\t%w[loop], %" #w "[ret], [%[addr]]\n"	\
 	"	cbnz	%w[loop], 1b",					\
 	/* LSE atomics */						\
-		#op_lse "\t%" #w "[val], %" #w "[ret], %[ptr]\n"	\
+		#op_lse "\t%" #w "[val], %" #w "[ret], [%[addr]]\n"	\
 		#op_llsc "\t%" #w "[ret], %" #w "[ret], %" #w "[val]\n"	\
 		__nops(2))						\
-	: [loop] "=&r" (loop), [ret] "=&r" (ret),			\
-	  [ptr] "+Q"(*(u##sz *)ptr)					\
-	: [val] "r" ((u##sz)(val)));					\
+	__PCPU_GPRS_END("%[gprs]")					\
+	: [gprs] "=Qo" (*gprs),						\
+	  [addr] "=&r" (addr),						\
+	  [off] "=&r" (off),						\
+	  [loop] "=&r" (loop),						\
+	  [ret] "=&r" (ret)						\
+	: [pcp] "r" (pcp),						\
+	  [val] "r" ((u##sz)(val))					\
+	: "memory"							\
+	);								\
 									\
 	return ret;							\
 }
@@ -322,13 +334,13 @@ PERCPU_RET_OP(add, add, ldadd)
 	_pcp_wrap(__percpu_add_case_64, pcp, val)
 
 #define this_cpu_add_return_1(pcp, val)	\
-	_pcp_protect_return(__percpu_add_return_case_8, pcp, val)
+	_pcp_wrap_return(__percpu_add_return_case_8, pcp, val)
 #define this_cpu_add_return_2(pcp, val)	\
-	_pcp_protect_return(__percpu_add_return_case_16, pcp, val)
+	_pcp_wrap_return(__percpu_add_return_case_16, pcp, val)
 #define this_cpu_add_return_4(pcp, val)	\
-	_pcp_protect_return(__percpu_add_return_case_32, pcp, val)
+	_pcp_wrap_return(__percpu_add_return_case_32, pcp, val)
 #define this_cpu_add_return_8(pcp, val)	\
-	_pcp_protect_return(__percpu_add_return_case_64, pcp, val)
+	_pcp_wrap_return(__percpu_add_return_case_64, pcp, val)
 
 #define this_cpu_and_1(pcp, val)	\
 	_pcp_wrap(__percpu_andnot_case_8, pcp, ~(u8)(val))
-- 
2.30.2