xref: /openbmc/linux/arch/arm64/include/asm/percpu.h (revision a44e4f3a)
1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3  * Copyright (C) 2013 ARM Ltd.
4  */
5 #ifndef __ASM_PERCPU_H
6 #define __ASM_PERCPU_H
7 
8 #include <linux/preempt.h>
9 
10 #include <asm/alternative.h>
11 #include <asm/cmpxchg.h>
12 #include <asm/stack_pointer.h>
13 
14 static inline void set_my_cpu_offset(unsigned long off)
15 {
16 	asm volatile(ALTERNATIVE("msr tpidr_el1, %0",
17 				 "msr tpidr_el2, %0",
18 				 ARM64_HAS_VIRT_HOST_EXTN)
19 			:: "r" (off) : "memory");
20 }
21 
22 static inline unsigned long __my_cpu_offset(void)
23 {
24 	unsigned long off;
25 
26 	/*
27 	 * We want to allow caching the value, so avoid using volatile and
28 	 * instead use a fake stack read to hazard against barrier().
29 	 */
30 	asm(ALTERNATIVE("mrs %0, tpidr_el1",
31 			"mrs %0, tpidr_el2",
32 			ARM64_HAS_VIRT_HOST_EXTN)
33 		: "=r" (off) :
34 		"Q" (*(const unsigned long *)current_stack_pointer));
35 
36 	return off;
37 }
38 #define __my_cpu_offset __my_cpu_offset()
39 
40 #define PERCPU_RW_OPS(sz)						\
41 static inline unsigned long __percpu_read_##sz(void *ptr)		\
42 {									\
43 	return READ_ONCE(*(u##sz *)ptr);				\
44 }									\
45 									\
46 static inline void __percpu_write_##sz(void *ptr, unsigned long val)	\
47 {									\
48 	WRITE_ONCE(*(u##sz *)ptr, (u##sz)val);				\
49 }
50 
51 #define __PERCPU_OP_CASE(w, sfx, name, sz, op_llsc, op_lse)		\
52 static inline void							\
53 __percpu_##name##_case_##sz(void *ptr, unsigned long val)		\
54 {									\
55 	unsigned int loop;						\
56 	u##sz tmp;							\
57 									\
58 	asm volatile (ARM64_LSE_ATOMIC_INSN(				\
59 	/* LL/SC */							\
60 	"1:	ldxr" #sfx "\t%" #w "[tmp], %[ptr]\n"			\
61 		#op_llsc "\t%" #w "[tmp], %" #w "[tmp], %" #w "[val]\n"	\
62 	"	stxr" #sfx "\t%w[loop], %" #w "[tmp], %[ptr]\n"		\
63 	"	cbnz	%w[loop], 1b",					\
64 	/* LSE atomics */						\
65 		#op_lse "\t%" #w "[val], %[ptr]\n"			\
66 		__nops(3))						\
67 	: [loop] "=&r" (loop), [tmp] "=&r" (tmp),			\
68 	  [ptr] "+Q"(*(u##sz *)ptr)					\
69 	: [val] "r" ((u##sz)(val)));					\
70 }
71 
72 #define __PERCPU_RET_OP_CASE(w, sfx, name, sz, op_llsc, op_lse)		\
73 static inline u##sz							\
74 __percpu_##name##_return_case_##sz(void *ptr, unsigned long val)	\
75 {									\
76 	unsigned int loop;						\
77 	u##sz ret;							\
78 									\
79 	asm volatile (ARM64_LSE_ATOMIC_INSN(				\
80 	/* LL/SC */							\
81 	"1:	ldxr" #sfx "\t%" #w "[ret], %[ptr]\n"			\
82 		#op_llsc "\t%" #w "[ret], %" #w "[ret], %" #w "[val]\n"	\
83 	"	stxr" #sfx "\t%w[loop], %" #w "[ret], %[ptr]\n"		\
84 	"	cbnz	%w[loop], 1b",					\
85 	/* LSE atomics */						\
86 		#op_lse "\t%" #w "[val], %" #w "[ret], %[ptr]\n"	\
87 		#op_llsc "\t%" #w "[ret], %" #w "[ret], %" #w "[val]\n"	\
88 		__nops(2))						\
89 	: [loop] "=&r" (loop), [ret] "=&r" (ret),			\
90 	  [ptr] "+Q"(*(u##sz *)ptr)					\
91 	: [val] "r" ((u##sz)(val)));					\
92 									\
93 	return ret;							\
94 }
95 
96 #define PERCPU_OP(name, op_llsc, op_lse)				\
97 	__PERCPU_OP_CASE(w, b, name,  8, op_llsc, op_lse)		\
98 	__PERCPU_OP_CASE(w, h, name, 16, op_llsc, op_lse)		\
99 	__PERCPU_OP_CASE(w,  , name, 32, op_llsc, op_lse)		\
100 	__PERCPU_OP_CASE( ,  , name, 64, op_llsc, op_lse)
101 
102 #define PERCPU_RET_OP(name, op_llsc, op_lse)				\
103 	__PERCPU_RET_OP_CASE(w, b, name,  8, op_llsc, op_lse)		\
104 	__PERCPU_RET_OP_CASE(w, h, name, 16, op_llsc, op_lse)		\
105 	__PERCPU_RET_OP_CASE(w,  , name, 32, op_llsc, op_lse)		\
106 	__PERCPU_RET_OP_CASE( ,  , name, 64, op_llsc, op_lse)
107 
108 PERCPU_RW_OPS(8)
109 PERCPU_RW_OPS(16)
110 PERCPU_RW_OPS(32)
111 PERCPU_RW_OPS(64)
112 PERCPU_OP(add, add, stadd)
113 PERCPU_OP(andnot, bic, stclr)
114 PERCPU_OP(or, orr, stset)
115 PERCPU_RET_OP(add, add, ldadd)
116 
117 #undef PERCPU_RW_OPS
118 #undef __PERCPU_OP_CASE
119 #undef __PERCPU_RET_OP_CASE
120 #undef PERCPU_OP
121 #undef PERCPU_RET_OP
122 
123 /*
124  * It would be nice to avoid the conditional call into the scheduler when
125  * re-enabling preemption for preemptible kernels, but doing that in a way
126  * which builds inside a module would mean messing directly with the preempt
127  * count. If you do this, peterz and tglx will hunt you down.
128  */
129 #define this_cpu_cmpxchg_double_8(ptr1, ptr2, o1, o2, n1, n2)		\
130 ({									\
131 	int __ret;							\
132 	preempt_disable_notrace();					\
133 	__ret = cmpxchg_double_local(	raw_cpu_ptr(&(ptr1)),		\
134 					raw_cpu_ptr(&(ptr2)),		\
135 					o1, o2, n1, n2);		\
136 	preempt_enable_notrace();					\
137 	__ret;								\
138 })
139 
140 #define _pcp_protect(op, pcp, ...)					\
141 ({									\
142 	preempt_disable_notrace();					\
143 	op(raw_cpu_ptr(&(pcp)), __VA_ARGS__);				\
144 	preempt_enable_notrace();					\
145 })
146 
147 #define _pcp_protect_return(op, pcp, args...)				\
148 ({									\
149 	typeof(pcp) __retval;						\
150 	preempt_disable_notrace();					\
151 	__retval = (typeof(pcp))op(raw_cpu_ptr(&(pcp)), ##args);	\
152 	preempt_enable_notrace();					\
153 	__retval;							\
154 })
155 
156 #define this_cpu_read_1(pcp)		\
157 	_pcp_protect_return(__percpu_read_8, pcp)
158 #define this_cpu_read_2(pcp)		\
159 	_pcp_protect_return(__percpu_read_16, pcp)
160 #define this_cpu_read_4(pcp)		\
161 	_pcp_protect_return(__percpu_read_32, pcp)
162 #define this_cpu_read_8(pcp)		\
163 	_pcp_protect_return(__percpu_read_64, pcp)
164 
165 #define this_cpu_write_1(pcp, val)	\
166 	_pcp_protect(__percpu_write_8, pcp, (unsigned long)val)
167 #define this_cpu_write_2(pcp, val)	\
168 	_pcp_protect(__percpu_write_16, pcp, (unsigned long)val)
169 #define this_cpu_write_4(pcp, val)	\
170 	_pcp_protect(__percpu_write_32, pcp, (unsigned long)val)
171 #define this_cpu_write_8(pcp, val)	\
172 	_pcp_protect(__percpu_write_64, pcp, (unsigned long)val)
173 
174 #define this_cpu_add_1(pcp, val)	\
175 	_pcp_protect(__percpu_add_case_8, pcp, val)
176 #define this_cpu_add_2(pcp, val)	\
177 	_pcp_protect(__percpu_add_case_16, pcp, val)
178 #define this_cpu_add_4(pcp, val)	\
179 	_pcp_protect(__percpu_add_case_32, pcp, val)
180 #define this_cpu_add_8(pcp, val)	\
181 	_pcp_protect(__percpu_add_case_64, pcp, val)
182 
183 #define this_cpu_add_return_1(pcp, val)	\
184 	_pcp_protect_return(__percpu_add_return_case_8, pcp, val)
185 #define this_cpu_add_return_2(pcp, val)	\
186 	_pcp_protect_return(__percpu_add_return_case_16, pcp, val)
187 #define this_cpu_add_return_4(pcp, val)	\
188 	_pcp_protect_return(__percpu_add_return_case_32, pcp, val)
189 #define this_cpu_add_return_8(pcp, val)	\
190 	_pcp_protect_return(__percpu_add_return_case_64, pcp, val)
191 
192 #define this_cpu_and_1(pcp, val)	\
193 	_pcp_protect(__percpu_andnot_case_8, pcp, ~val)
194 #define this_cpu_and_2(pcp, val)	\
195 	_pcp_protect(__percpu_andnot_case_16, pcp, ~val)
196 #define this_cpu_and_4(pcp, val)	\
197 	_pcp_protect(__percpu_andnot_case_32, pcp, ~val)
198 #define this_cpu_and_8(pcp, val)	\
199 	_pcp_protect(__percpu_andnot_case_64, pcp, ~val)
200 
201 #define this_cpu_or_1(pcp, val)		\
202 	_pcp_protect(__percpu_or_case_8, pcp, val)
203 #define this_cpu_or_2(pcp, val)		\
204 	_pcp_protect(__percpu_or_case_16, pcp, val)
205 #define this_cpu_or_4(pcp, val)		\
206 	_pcp_protect(__percpu_or_case_32, pcp, val)
207 #define this_cpu_or_8(pcp, val)		\
208 	_pcp_protect(__percpu_or_case_64, pcp, val)
209 
210 #define this_cpu_xchg_1(pcp, val)	\
211 	_pcp_protect_return(xchg_relaxed, pcp, val)
212 #define this_cpu_xchg_2(pcp, val)	\
213 	_pcp_protect_return(xchg_relaxed, pcp, val)
214 #define this_cpu_xchg_4(pcp, val)	\
215 	_pcp_protect_return(xchg_relaxed, pcp, val)
216 #define this_cpu_xchg_8(pcp, val)	\
217 	_pcp_protect_return(xchg_relaxed, pcp, val)
218 
219 #define this_cpu_cmpxchg_1(pcp, o, n)	\
220 	_pcp_protect_return(cmpxchg_relaxed, pcp, o, n)
221 #define this_cpu_cmpxchg_2(pcp, o, n)	\
222 	_pcp_protect_return(cmpxchg_relaxed, pcp, o, n)
223 #define this_cpu_cmpxchg_4(pcp, o, n)	\
224 	_pcp_protect_return(cmpxchg_relaxed, pcp, o, n)
225 #define this_cpu_cmpxchg_8(pcp, o, n)	\
226 	_pcp_protect_return(cmpxchg_relaxed, pcp, o, n)
227 
228 #include <asm-generic/percpu.h>
229 
230 #endif /* __ASM_PERCPU_H */
231