xref: /openbmc/linux/tools/perf/arch/x86/util/perf_regs.c (revision 9a29f5fc)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <errno.h>
3 #include <string.h>
4 #include <regex.h>
5 #include <linux/kernel.h>
6 #include <linux/zalloc.h>
7 
8 #include "../../../perf-sys.h"
9 #include "../../../util/perf_regs.h"
10 #include "../../../util/debug.h"
11 #include "../../../util/event.h"
12 #include "../../../util/pmu.h"
13 #include "../../../util/pmu-hybrid.h"
14 
15 const struct sample_reg sample_reg_masks[] = {
16 	SMPL_REG(AX, PERF_REG_X86_AX),
17 	SMPL_REG(BX, PERF_REG_X86_BX),
18 	SMPL_REG(CX, PERF_REG_X86_CX),
19 	SMPL_REG(DX, PERF_REG_X86_DX),
20 	SMPL_REG(SI, PERF_REG_X86_SI),
21 	SMPL_REG(DI, PERF_REG_X86_DI),
22 	SMPL_REG(BP, PERF_REG_X86_BP),
23 	SMPL_REG(SP, PERF_REG_X86_SP),
24 	SMPL_REG(IP, PERF_REG_X86_IP),
25 	SMPL_REG(FLAGS, PERF_REG_X86_FLAGS),
26 	SMPL_REG(CS, PERF_REG_X86_CS),
27 	SMPL_REG(SS, PERF_REG_X86_SS),
28 #ifdef HAVE_ARCH_X86_64_SUPPORT
29 	SMPL_REG(R8, PERF_REG_X86_R8),
30 	SMPL_REG(R9, PERF_REG_X86_R9),
31 	SMPL_REG(R10, PERF_REG_X86_R10),
32 	SMPL_REG(R11, PERF_REG_X86_R11),
33 	SMPL_REG(R12, PERF_REG_X86_R12),
34 	SMPL_REG(R13, PERF_REG_X86_R13),
35 	SMPL_REG(R14, PERF_REG_X86_R14),
36 	SMPL_REG(R15, PERF_REG_X86_R15),
37 #endif
38 	SMPL_REG2(XMM0, PERF_REG_X86_XMM0),
39 	SMPL_REG2(XMM1, PERF_REG_X86_XMM1),
40 	SMPL_REG2(XMM2, PERF_REG_X86_XMM2),
41 	SMPL_REG2(XMM3, PERF_REG_X86_XMM3),
42 	SMPL_REG2(XMM4, PERF_REG_X86_XMM4),
43 	SMPL_REG2(XMM5, PERF_REG_X86_XMM5),
44 	SMPL_REG2(XMM6, PERF_REG_X86_XMM6),
45 	SMPL_REG2(XMM7, PERF_REG_X86_XMM7),
46 	SMPL_REG2(XMM8, PERF_REG_X86_XMM8),
47 	SMPL_REG2(XMM9, PERF_REG_X86_XMM9),
48 	SMPL_REG2(XMM10, PERF_REG_X86_XMM10),
49 	SMPL_REG2(XMM11, PERF_REG_X86_XMM11),
50 	SMPL_REG2(XMM12, PERF_REG_X86_XMM12),
51 	SMPL_REG2(XMM13, PERF_REG_X86_XMM13),
52 	SMPL_REG2(XMM14, PERF_REG_X86_XMM14),
53 	SMPL_REG2(XMM15, PERF_REG_X86_XMM15),
54 	SMPL_REG_END
55 };
56 
57 struct sdt_name_reg {
58 	const char *sdt_name;
59 	const char *uprobe_name;
60 };
61 #define SDT_NAME_REG(n, m) {.sdt_name = "%" #n, .uprobe_name = "%" #m}
62 #define SDT_NAME_REG_END {.sdt_name = NULL, .uprobe_name = NULL}
63 
64 static const struct sdt_name_reg sdt_reg_tbl[] = {
65 	SDT_NAME_REG(eax, ax),
66 	SDT_NAME_REG(rax, ax),
67 	SDT_NAME_REG(al,  ax),
68 	SDT_NAME_REG(ah,  ax),
69 	SDT_NAME_REG(ebx, bx),
70 	SDT_NAME_REG(rbx, bx),
71 	SDT_NAME_REG(bl,  bx),
72 	SDT_NAME_REG(bh,  bx),
73 	SDT_NAME_REG(ecx, cx),
74 	SDT_NAME_REG(rcx, cx),
75 	SDT_NAME_REG(cl,  cx),
76 	SDT_NAME_REG(ch,  cx),
77 	SDT_NAME_REG(edx, dx),
78 	SDT_NAME_REG(rdx, dx),
79 	SDT_NAME_REG(dl,  dx),
80 	SDT_NAME_REG(dh,  dx),
81 	SDT_NAME_REG(esi, si),
82 	SDT_NAME_REG(rsi, si),
83 	SDT_NAME_REG(sil, si),
84 	SDT_NAME_REG(edi, di),
85 	SDT_NAME_REG(rdi, di),
86 	SDT_NAME_REG(dil, di),
87 	SDT_NAME_REG(ebp, bp),
88 	SDT_NAME_REG(rbp, bp),
89 	SDT_NAME_REG(bpl, bp),
90 	SDT_NAME_REG(rsp, sp),
91 	SDT_NAME_REG(esp, sp),
92 	SDT_NAME_REG(spl, sp),
93 
94 	/* rNN registers */
95 	SDT_NAME_REG(r8b,  r8),
96 	SDT_NAME_REG(r8w,  r8),
97 	SDT_NAME_REG(r8d,  r8),
98 	SDT_NAME_REG(r9b,  r9),
99 	SDT_NAME_REG(r9w,  r9),
100 	SDT_NAME_REG(r9d,  r9),
101 	SDT_NAME_REG(r10b, r10),
102 	SDT_NAME_REG(r10w, r10),
103 	SDT_NAME_REG(r10d, r10),
104 	SDT_NAME_REG(r11b, r11),
105 	SDT_NAME_REG(r11w, r11),
106 	SDT_NAME_REG(r11d, r11),
107 	SDT_NAME_REG(r12b, r12),
108 	SDT_NAME_REG(r12w, r12),
109 	SDT_NAME_REG(r12d, r12),
110 	SDT_NAME_REG(r13b, r13),
111 	SDT_NAME_REG(r13w, r13),
112 	SDT_NAME_REG(r13d, r13),
113 	SDT_NAME_REG(r14b, r14),
114 	SDT_NAME_REG(r14w, r14),
115 	SDT_NAME_REG(r14d, r14),
116 	SDT_NAME_REG(r15b, r15),
117 	SDT_NAME_REG(r15w, r15),
118 	SDT_NAME_REG(r15d, r15),
119 	SDT_NAME_REG_END,
120 };
121 
122 /*
123  * Perf only supports OP which is in  +/-NUM(REG)  form.
124  * Here plus-minus sign, NUM and parenthesis are optional,
125  * only REG is mandatory.
126  *
127  * SDT events also supports indirect addressing mode with a
128  * symbol as offset, scaled mode and constants in OP. But
129  * perf does not support them yet. Below are few examples.
130  *
131  * OP with scaled mode:
132  *     (%rax,%rsi,8)
133  *     10(%ras,%rsi,8)
134  *
135  * OP with indirect addressing mode:
136  *     check_action(%rip)
137  *     mp_+52(%rip)
138  *     44+mp_(%rip)
139  *
140  * OP with constant values:
141  *     $0
142  *     $123
143  *     $-1
144  */
145 #define SDT_OP_REGEX  "^([+\\-]?)([0-9]*)(\\(?)(%[a-z][a-z0-9]+)(\\)?)$"
146 
147 static regex_t sdt_op_regex;
148 
149 static int sdt_init_op_regex(void)
150 {
151 	static int initialized;
152 	int ret = 0;
153 
154 	if (initialized)
155 		return 0;
156 
157 	ret = regcomp(&sdt_op_regex, SDT_OP_REGEX, REG_EXTENDED);
158 	if (ret < 0) {
159 		pr_debug4("Regex compilation error.\n");
160 		return ret;
161 	}
162 
163 	initialized = 1;
164 	return 0;
165 }
166 
167 /*
168  * Max x86 register name length is 5(ex: %r15d). So, 6th char
169  * should always contain NULL. This helps to find register name
170  * length using strlen, instead of maintaining one more variable.
171  */
172 #define SDT_REG_NAME_SIZE  6
173 
174 /*
175  * The uprobe parser does not support all gas register names;
176  * so, we have to replace them (ex. for x86_64: %rax -> %ax).
177  * Note: If register does not require renaming, just copy
178  * paste as it is, but don't leave it empty.
179  */
180 static void sdt_rename_register(char *sdt_reg, int sdt_len, char *uprobe_reg)
181 {
182 	int i = 0;
183 
184 	for (i = 0; sdt_reg_tbl[i].sdt_name != NULL; i++) {
185 		if (!strncmp(sdt_reg_tbl[i].sdt_name, sdt_reg, sdt_len)) {
186 			strcpy(uprobe_reg, sdt_reg_tbl[i].uprobe_name);
187 			return;
188 		}
189 	}
190 
191 	strncpy(uprobe_reg, sdt_reg, sdt_len);
192 }
193 
194 int arch_sdt_arg_parse_op(char *old_op, char **new_op)
195 {
196 	char new_reg[SDT_REG_NAME_SIZE] = {0};
197 	int new_len = 0, ret;
198 	/*
199 	 * rm[0]:  +/-NUM(REG)
200 	 * rm[1]:  +/-
201 	 * rm[2]:  NUM
202 	 * rm[3]:  (
203 	 * rm[4]:  REG
204 	 * rm[5]:  )
205 	 */
206 	regmatch_t rm[6];
207 	/*
208 	 * Max prefix length is 2 as it may contains sign(+/-)
209 	 * and displacement 0 (Both sign and displacement 0 are
210 	 * optional so it may be empty). Use one more character
211 	 * to hold last NULL so that strlen can be used to find
212 	 * prefix length, instead of maintaining one more variable.
213 	 */
214 	char prefix[3] = {0};
215 
216 	ret = sdt_init_op_regex();
217 	if (ret < 0)
218 		return ret;
219 
220 	/*
221 	 * If unsupported OR does not match with regex OR
222 	 * register name too long, skip it.
223 	 */
224 	if (strchr(old_op, ',') || strchr(old_op, '$') ||
225 	    regexec(&sdt_op_regex, old_op, 6, rm, 0)   ||
226 	    rm[4].rm_eo - rm[4].rm_so > SDT_REG_NAME_SIZE) {
227 		pr_debug4("Skipping unsupported SDT argument: %s\n", old_op);
228 		return SDT_ARG_SKIP;
229 	}
230 
231 	/*
232 	 * Prepare prefix.
233 	 * If SDT OP has parenthesis but does not provide
234 	 * displacement, add 0 for displacement.
235 	 *     SDT         Uprobe     Prefix
236 	 *     -----------------------------
237 	 *     +24(%rdi)   +24(%di)   +
238 	 *     24(%rdi)    +24(%di)   +
239 	 *     %rdi        %di
240 	 *     (%rdi)      +0(%di)    +0
241 	 *     -80(%rbx)   -80(%bx)   -
242 	 */
243 	if (rm[3].rm_so != rm[3].rm_eo) {
244 		if (rm[1].rm_so != rm[1].rm_eo)
245 			prefix[0] = *(old_op + rm[1].rm_so);
246 		else if (rm[2].rm_so != rm[2].rm_eo)
247 			prefix[0] = '+';
248 		else
249 			scnprintf(prefix, sizeof(prefix), "+0");
250 	}
251 
252 	/* Rename register */
253 	sdt_rename_register(old_op + rm[4].rm_so, rm[4].rm_eo - rm[4].rm_so,
254 			    new_reg);
255 
256 	/* Prepare final OP which should be valid for uprobe_events */
257 	new_len = strlen(prefix)              +
258 		  (rm[2].rm_eo - rm[2].rm_so) +
259 		  (rm[3].rm_eo - rm[3].rm_so) +
260 		  strlen(new_reg)             +
261 		  (rm[5].rm_eo - rm[5].rm_so) +
262 		  1;					/* NULL */
263 
264 	*new_op = zalloc(new_len);
265 	if (!*new_op)
266 		return -ENOMEM;
267 
268 	scnprintf(*new_op, new_len, "%.*s%.*s%.*s%.*s%.*s",
269 		  strlen(prefix), prefix,
270 		  (int)(rm[2].rm_eo - rm[2].rm_so), old_op + rm[2].rm_so,
271 		  (int)(rm[3].rm_eo - rm[3].rm_so), old_op + rm[3].rm_so,
272 		  strlen(new_reg), new_reg,
273 		  (int)(rm[5].rm_eo - rm[5].rm_so), old_op + rm[5].rm_so);
274 
275 	return SDT_ARG_VALID;
276 }
277 
278 uint64_t arch__intr_reg_mask(void)
279 {
280 	struct perf_event_attr attr = {
281 		.type			= PERF_TYPE_HARDWARE,
282 		.config			= PERF_COUNT_HW_CPU_CYCLES,
283 		.sample_type		= PERF_SAMPLE_REGS_INTR,
284 		.sample_regs_intr	= PERF_REG_EXTENDED_MASK,
285 		.precise_ip		= 1,
286 		.disabled 		= 1,
287 		.exclude_kernel		= 1,
288 	};
289 	struct perf_pmu *pmu;
290 	int fd;
291 	/*
292 	 * In an unnamed union, init it here to build on older gcc versions
293 	 */
294 	attr.sample_period = 1;
295 
296 	if (perf_pmu__has_hybrid()) {
297 		/*
298 		 * The same register set is supported among different hybrid PMUs.
299 		 * Only check the first available one.
300 		 */
301 		pmu = list_first_entry(&perf_pmu__hybrid_pmus, typeof(*pmu), hybrid_list);
302 		attr.config |= (__u64)pmu->type << PERF_PMU_TYPE_SHIFT;
303 	}
304 
305 	event_attr_init(&attr);
306 
307 	fd = sys_perf_event_open(&attr, 0, -1, -1, 0);
308 	if (fd != -1) {
309 		close(fd);
310 		return (PERF_REG_EXTENDED_MASK | PERF_REGS_MASK);
311 	}
312 
313 	return PERF_REGS_MASK;
314 }
315