1/* SPDX-License-Identifier: LGPL-2.1 OR MIT */
2/*
3 * rseq-ppc.h
4 *
5 * (C) Copyright 2016-2022 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
6 * (C) Copyright 2016-2018 - Boqun Feng <boqun.feng@gmail.com>
7 */
8
9/*
10 * RSEQ_SIG is used with the following trap instruction:
11 *
12 * powerpc-be:    0f e5 00 0b           twui   r5,11
13 * powerpc64-le:  0b 00 e5 0f           twui   r5,11
14 * powerpc64-be:  0f e5 00 0b           twui   r5,11
15 */
16
17#define RSEQ_SIG	0x0fe5000b
18
19#define rseq_smp_mb()		__asm__ __volatile__ ("sync"	::: "memory", "cc")
20#define rseq_smp_lwsync()	__asm__ __volatile__ ("lwsync"	::: "memory", "cc")
21#define rseq_smp_rmb()		rseq_smp_lwsync()
22#define rseq_smp_wmb()		rseq_smp_lwsync()
23
24#define rseq_smp_load_acquire(p)					\
25__extension__ ({							\
26	rseq_unqual_scalar_typeof(*(p)) ____p1 = RSEQ_READ_ONCE(*(p));	\
27	rseq_smp_lwsync();						\
28	____p1;								\
29})
30
31#define rseq_smp_acquire__after_ctrl_dep()	rseq_smp_lwsync()
32
33#define rseq_smp_store_release(p, v)					\
34do {									\
35	rseq_smp_lwsync();						\
36	RSEQ_WRITE_ONCE(*(p), v);					\
37} while (0)
38
39/*
40 * The __rseq_cs_ptr_array and __rseq_cs sections can be used by debuggers to
41 * better handle single-stepping through the restartable critical sections.
42 */
43
44#ifdef __PPC64__
45
46#define RSEQ_STORE_LONG(arg)	"std%U[" __rseq_str(arg) "]%X[" __rseq_str(arg) "] "	/* To memory ("m" constraint) */
47#define RSEQ_STORE_INT(arg)	"stw%U[" __rseq_str(arg) "]%X[" __rseq_str(arg) "] "	/* To memory ("m" constraint) */
48#define RSEQ_LOAD_LONG(arg)	"ld%U[" __rseq_str(arg) "]%X[" __rseq_str(arg) "] "	/* From memory ("m" constraint) */
49#define RSEQ_LOAD_INT(arg)	"lwz%U[" __rseq_str(arg) "]%X[" __rseq_str(arg) "] "	/* From memory ("m" constraint) */
50#define RSEQ_LOADX_LONG		"ldx "							/* From base register ("b" constraint) */
51#define RSEQ_CMP_LONG		"cmpd "
52#define RSEQ_CMP_LONG_INT	"cmpdi "
53
54#define __RSEQ_ASM_DEFINE_TABLE(label, version, flags,				\
55			start_ip, post_commit_offset, abort_ip)			\
56		".pushsection __rseq_cs, \"aw\"\n\t"				\
57		".balign 32\n\t"						\
58		__rseq_str(label) ":\n\t"					\
59		".long " __rseq_str(version) ", " __rseq_str(flags) "\n\t"	\
60		".quad " __rseq_str(start_ip) ", " __rseq_str(post_commit_offset) ", " __rseq_str(abort_ip) "\n\t" \
61		".popsection\n\t"						\
62		".pushsection __rseq_cs_ptr_array, \"aw\"\n\t"			\
63		".quad " __rseq_str(label) "b\n\t"				\
64		".popsection\n\t"
65
66#define RSEQ_ASM_STORE_RSEQ_CS(label, cs_label, rseq_cs)			\
67		RSEQ_INJECT_ASM(1)						\
68		"lis %%r17, (" __rseq_str(cs_label) ")@highest\n\t"		\
69		"ori %%r17, %%r17, (" __rseq_str(cs_label) ")@higher\n\t"	\
70		"rldicr %%r17, %%r17, 32, 31\n\t"				\
71		"oris %%r17, %%r17, (" __rseq_str(cs_label) ")@high\n\t"	\
72		"ori %%r17, %%r17, (" __rseq_str(cs_label) ")@l\n\t"		\
73		"std %%r17, %[" __rseq_str(rseq_cs) "]\n\t"			\
74		__rseq_str(label) ":\n\t"
75
76/*
77 * Exit points of a rseq critical section consist of all instructions outside
78 * of the critical section where a critical section can either branch to or
79 * reach through the normal course of its execution. The abort IP and the
80 * post-commit IP are already part of the __rseq_cs section and should not be
81 * explicitly defined as additional exit points. Knowing all exit points is
82 * useful to assist debuggers stepping over the critical section.
83 */
84#define RSEQ_ASM_DEFINE_EXIT_POINT(start_ip, exit_ip)			\
85		".pushsection __rseq_exit_point_array, \"aw\"\n\t"	\
86		".quad " __rseq_str(start_ip) ", " __rseq_str(exit_ip) "\n\t" \
87		".popsection\n\t"
88
89#else /* #ifdef __PPC64__ */
90
91#define RSEQ_STORE_LONG(arg)	"stw%U[" __rseq_str(arg) "]%X[" __rseq_str(arg) "] "	/* To memory ("m" constraint) */
92#define RSEQ_STORE_INT(arg)	RSEQ_STORE_LONG(arg)					/* To memory ("m" constraint) */
93#define RSEQ_LOAD_LONG(arg)	"lwz%U[" __rseq_str(arg) "]%X[" __rseq_str(arg) "] "	/* From memory ("m" constraint) */
94#define RSEQ_LOAD_INT(arg)	RSEQ_LOAD_LONG(arg)					/* From memory ("m" constraint) */
95#define RSEQ_LOADX_LONG		"lwzx "							/* From base register ("b" constraint) */
96#define RSEQ_CMP_LONG		"cmpw "
97#define RSEQ_CMP_LONG_INT	"cmpwi "
98
99#define __RSEQ_ASM_DEFINE_TABLE(label, version, flags,				\
100			start_ip, post_commit_offset, abort_ip)			\
101		".pushsection __rseq_cs, \"aw\"\n\t"				\
102		".balign 32\n\t"						\
103		__rseq_str(label) ":\n\t"					\
104		".long " __rseq_str(version) ", " __rseq_str(flags) "\n\t"	\
105		/* 32-bit only supported on BE */				\
106		".long 0x0, " __rseq_str(start_ip) ", 0x0, " __rseq_str(post_commit_offset) ", 0x0, " __rseq_str(abort_ip) "\n\t" \
107		".popsection\n\t"					\
108		".pushsection __rseq_cs_ptr_array, \"aw\"\n\t"		\
109		".long 0x0, " __rseq_str(label) "b\n\t"			\
110		".popsection\n\t"
111
112/*
113 * Exit points of a rseq critical section consist of all instructions outside
114 * of the critical section where a critical section can either branch to or
115 * reach through the normal course of its execution. The abort IP and the
116 * post-commit IP are already part of the __rseq_cs section and should not be
117 * explicitly defined as additional exit points. Knowing all exit points is
118 * useful to assist debuggers stepping over the critical section.
119 */
120#define RSEQ_ASM_DEFINE_EXIT_POINT(start_ip, exit_ip)				\
121		".pushsection __rseq_exit_point_array, \"aw\"\n\t"		\
122		/* 32-bit only supported on BE */				\
123		".long 0x0, " __rseq_str(start_ip) ", 0x0, " __rseq_str(exit_ip) "\n\t"	\
124		".popsection\n\t"
125
126#define RSEQ_ASM_STORE_RSEQ_CS(label, cs_label, rseq_cs)			\
127		RSEQ_INJECT_ASM(1)						\
128		"lis %%r17, (" __rseq_str(cs_label) ")@ha\n\t"			\
129		"addi %%r17, %%r17, (" __rseq_str(cs_label) ")@l\n\t"		\
130		RSEQ_STORE_INT(rseq_cs) "%%r17, %[" __rseq_str(rseq_cs) "]\n\t"	\
131		__rseq_str(label) ":\n\t"
132
133#endif /* #ifdef __PPC64__ */
134
135#define RSEQ_ASM_DEFINE_TABLE(label, start_ip, post_commit_ip, abort_ip)	\
136		__RSEQ_ASM_DEFINE_TABLE(label, 0x0, 0x0, start_ip,		\
137					(post_commit_ip - start_ip), abort_ip)
138
139#define RSEQ_ASM_CMP_CPU_ID(cpu_id, current_cpu_id, label)			\
140		RSEQ_INJECT_ASM(2)						\
141		RSEQ_LOAD_INT(current_cpu_id) "%%r17, %[" __rseq_str(current_cpu_id) "]\n\t" \
142		"cmpw cr7, %[" __rseq_str(cpu_id) "], %%r17\n\t"		\
143		"bne- cr7, " __rseq_str(label) "\n\t"
144
145#define RSEQ_ASM_DEFINE_ABORT(label, abort_label)				\
146		".pushsection __rseq_failure, \"ax\"\n\t"			\
147		".long " __rseq_str(RSEQ_SIG) "\n\t"				\
148		__rseq_str(label) ":\n\t"					\
149		"b %l[" __rseq_str(abort_label) "]\n\t"				\
150		".popsection\n\t"
151
152/*
153 * RSEQ_ASM_OPs: asm operations for rseq
154 * 	RSEQ_ASM_OP_R_*: has hard-code registers in it
155 * 	RSEQ_ASM_OP_* (else): doesn't have hard-code registers(unless cr7)
156 */
157#define RSEQ_ASM_OP_CMPEQ(var, expect, label)					\
158		RSEQ_LOAD_LONG(var) "%%r17, %[" __rseq_str(var) "]\n\t"		\
159		RSEQ_CMP_LONG "cr7, %%r17, %[" __rseq_str(expect) "]\n\t"		\
160		"bne- cr7, " __rseq_str(label) "\n\t"
161
162#define RSEQ_ASM_OP_CMPNE(var, expectnot, label)				\
163		RSEQ_LOAD_LONG(var) "%%r17, %[" __rseq_str(var) "]\n\t"		\
164		RSEQ_CMP_LONG "cr7, %%r17, %[" __rseq_str(expectnot) "]\n\t"		\
165		"beq- cr7, " __rseq_str(label) "\n\t"
166
167#define RSEQ_ASM_OP_STORE(value, var)						\
168		RSEQ_STORE_LONG(var) "%[" __rseq_str(value) "], %[" __rseq_str(var) "]\n\t"
169
170/* Load @var to r17 */
171#define RSEQ_ASM_OP_R_LOAD(var)							\
172		RSEQ_LOAD_LONG(var) "%%r17, %[" __rseq_str(var) "]\n\t"
173
174/* Store r17 to @var */
175#define RSEQ_ASM_OP_R_STORE(var)						\
176		RSEQ_STORE_LONG(var) "%%r17, %[" __rseq_str(var) "]\n\t"
177
178/* Add @count to r17 */
179#define RSEQ_ASM_OP_R_ADD(count)						\
180		"add %%r17, %[" __rseq_str(count) "], %%r17\n\t"
181
182/* Load (r17 + voffp) to r17 */
183#define RSEQ_ASM_OP_R_LOADX(voffp)						\
184		RSEQ_LOADX_LONG "%%r17, %[" __rseq_str(voffp) "], %%r17\n\t"
185
186/* TODO: implement a faster memcpy. */
187#define RSEQ_ASM_OP_R_MEMCPY() \
188		RSEQ_CMP_LONG_INT "%%r19, 0\n\t" \
189		"beq 333f\n\t" \
190		"addi %%r20, %%r20, -1\n\t" \
191		"addi %%r21, %%r21, -1\n\t" \
192		"222:\n\t" \
193		"lbzu %%r18, 1(%%r20)\n\t" \
194		"stbu %%r18, 1(%%r21)\n\t" \
195		"addi %%r19, %%r19, -1\n\t" \
196		RSEQ_CMP_LONG_INT "%%r19, 0\n\t" \
197		"bne 222b\n\t" \
198		"333:\n\t" \
199
200#define RSEQ_ASM_OP_R_FINAL_STORE(var, post_commit_label)			\
201		RSEQ_STORE_LONG(var) "%%r17, %[" __rseq_str(var) "]\n\t"			\
202		__rseq_str(post_commit_label) ":\n\t"
203
204#define RSEQ_ASM_OP_FINAL_STORE(value, var, post_commit_label)			\
205		RSEQ_STORE_LONG(var) "%[" __rseq_str(value) "], %[" __rseq_str(var) "]\n\t" \
206		__rseq_str(post_commit_label) ":\n\t"
207
208/* Per-cpu-id indexing. */
209
210#define RSEQ_TEMPLATE_CPU_ID
211#define RSEQ_TEMPLATE_MO_RELAXED
212#include "rseq-ppc-bits.h"
213#undef RSEQ_TEMPLATE_MO_RELAXED
214
215#define RSEQ_TEMPLATE_MO_RELEASE
216#include "rseq-ppc-bits.h"
217#undef RSEQ_TEMPLATE_MO_RELEASE
218#undef RSEQ_TEMPLATE_CPU_ID
219
220/* Per-mm-cid indexing. */
221
222#define RSEQ_TEMPLATE_MM_CID
223#define RSEQ_TEMPLATE_MO_RELAXED
224#include "rseq-ppc-bits.h"
225#undef RSEQ_TEMPLATE_MO_RELAXED
226
227#define RSEQ_TEMPLATE_MO_RELEASE
228#include "rseq-ppc-bits.h"
229#undef RSEQ_TEMPLATE_MO_RELEASE
230#undef RSEQ_TEMPLATE_MM_CID
231
232/* APIs which are not based on cpu ids. */
233
234#define RSEQ_TEMPLATE_CPU_ID_NONE
235#define RSEQ_TEMPLATE_MO_RELAXED
236#include "rseq-ppc-bits.h"
237#undef RSEQ_TEMPLATE_MO_RELAXED
238#undef RSEQ_TEMPLATE_CPU_ID_NONE
239