xref: /openbmc/linux/arch/arm64/kernel/entry-ftrace.S (revision ffcdf473) 
1/* SPDX-License-Identifier: GPL-2.0-only */
2/*
3 * arch/arm64/kernel/entry-ftrace.S
4 *
5 * Copyright (C) 2013 Linaro Limited
6 * Author: AKASHI Takahiro <takahiro.akashi@linaro.org>
7 */
8
9#include <linux/linkage.h>
10#include <linux/cfi_types.h>
11#include <asm/asm-offsets.h>
12#include <asm/assembler.h>
13#include <asm/ftrace.h>
14#include <asm/insn.h>
15
16#ifdef CONFIG_DYNAMIC_FTRACE_WITH_ARGS
17/*
18 * Due to -fpatchable-function-entry=2, the compiler has placed two NOPs before
19 * the regular function prologue. For an enabled callsite, ftrace_init_nop() and
20 * ftrace_make_call() have patched those NOPs to:
21 *
22 * 	MOV	X9, LR
23 * 	BL	ftrace_caller
24 *
25 * Each instrumented function follows the AAPCS, so here x0-x8 and x18-x30 are
26 * live (x18 holds the Shadow Call Stack pointer), and x9-x17 are safe to
27 * clobber.
28 *
29 * We save the callsite's context into a struct ftrace_regs before invoking any
30 * ftrace callbacks. So that we can get a sensible backtrace, we create frame
31 * records for the callsite and the ftrace entry assembly. This is not
32 * sufficient for reliable stacktrace: until we create the callsite stack
33 * record, its caller is missing from the LR and existing chain of frame
34 * records.
35 */
36SYM_CODE_START(ftrace_caller)
37	bti	c
38
39#ifdef CONFIG_DYNAMIC_FTRACE_WITH_CALL_OPS
40	/*
41	 * The literal pointer to the ops is at an 8-byte aligned boundary
42	 * which is either 12 or 16 bytes before the BL instruction in the call
43	 * site. See ftrace_call_adjust() for details.
44	 *
45	 * Therefore here the LR points at `literal + 16` or `literal + 20`,
46	 * and we can find the address of the literal in either case by
47	 * aligning to an 8-byte boundary and subtracting 16. We do the
48	 * alignment first as this allows us to fold the subtraction into the
49	 * LDR.
50	 */
51	bic	x11, x30, 0x7
52	ldr	x11, [x11, #-(4 * AARCH64_INSN_SIZE)]		// op
53
54#ifdef CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS
55	/*
56	 * If the op has a direct call, handle it immediately without
57	 * saving/restoring registers.
58	 */
59	ldr	x17, [x11, #FTRACE_OPS_DIRECT_CALL]		// op->direct_call
60	cbnz	x17, ftrace_caller_direct
61#endif
62#endif
63
64	/* Save original SP */
65	mov	x10, sp
66
67	/* Make room for ftrace regs, plus two frame records */
68	sub	sp, sp, #(FREGS_SIZE + 32)
69
70	/* Save function arguments */
71	stp	x0, x1, [sp, #FREGS_X0]
72	stp	x2, x3, [sp, #FREGS_X2]
73	stp	x4, x5, [sp, #FREGS_X4]
74	stp	x6, x7, [sp, #FREGS_X6]
75	str	x8,     [sp, #FREGS_X8]
76
77#ifdef CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS
78	str	xzr, [sp, #FREGS_DIRECT_TRAMP]
79#endif
80
81	/* Save the callsite's FP, LR, SP */
82	str	x29, [sp, #FREGS_FP]
83	str	x9,  [sp, #FREGS_LR]
84	str	x10, [sp, #FREGS_SP]
85
86	/* Save the PC after the ftrace callsite */
87	str	x30, [sp, #FREGS_PC]
88
89	/* Create a frame record for the callsite above the ftrace regs */
90	stp	x29, x9, [sp, #FREGS_SIZE + 16]
91	add	x29, sp, #FREGS_SIZE + 16
92
93	/* Create our frame record above the ftrace regs */
94	stp	x29, x30, [sp, #FREGS_SIZE]
95	add	x29, sp, #FREGS_SIZE
96
97	/* Prepare arguments for the the tracer func */
98	sub	x0, x30, #AARCH64_INSN_SIZE		// ip (callsite's BL insn)
99	mov	x1, x9					// parent_ip (callsite's LR)
100	mov	x3, sp					// regs
101
102#ifdef CONFIG_DYNAMIC_FTRACE_WITH_CALL_OPS
103	mov	x2, x11					// op
104	ldr	x4, [x2, #FTRACE_OPS_FUNC]		// op->func
105	blr	x4					// op->func(ip, parent_ip, op, regs)
106
107#else
108	ldr_l   x2, function_trace_op			// op
109
110SYM_INNER_LABEL(ftrace_call, SYM_L_GLOBAL)
111	bl      ftrace_stub				// func(ip, parent_ip, op, regs)
112#endif
113
114/*
115 * At the callsite x0-x8 and x19-x30 were live. Any C code will have preserved
116 * x19-x29 per the AAPCS, and we created frame records upon entry, so we need
117 * to restore x0-x8, x29, and x30.
118 */
119	/* Restore function arguments */
120	ldp	x0, x1, [sp, #FREGS_X0]
121	ldp	x2, x3, [sp, #FREGS_X2]
122	ldp	x4, x5, [sp, #FREGS_X4]
123	ldp	x6, x7, [sp, #FREGS_X6]
124	ldr	x8,     [sp, #FREGS_X8]
125
126	/* Restore the callsite's FP */
127	ldr	x29, [sp, #FREGS_FP]
128
129#ifdef CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS
130	ldr	x17, [sp, #FREGS_DIRECT_TRAMP]
131	cbnz	x17, ftrace_caller_direct_late
132#endif
133
134	/* Restore the callsite's LR and PC */
135	ldr	x30, [sp, #FREGS_LR]
136	ldr	x9,  [sp, #FREGS_PC]
137
138	/* Restore the callsite's SP */
139	add	sp, sp, #FREGS_SIZE + 32
140
141	ret	x9
142
143#ifdef CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS
144SYM_INNER_LABEL(ftrace_caller_direct_late, SYM_L_LOCAL)
145	/*
146	 * Head to a direct trampoline in x17 after having run other tracers.
147	 * The ftrace_regs are live, and x0-x8 and FP have been restored. The
148	 * LR, PC, and SP have not been restored.
149	 */
150
151	/*
152	 * Restore the callsite's LR and PC matching the trampoline calling
153	 * convention.
154	 */
155	ldr	x9,  [sp, #FREGS_LR]
156	ldr	x30, [sp, #FREGS_PC]
157
158	/* Restore the callsite's SP */
159	add	sp, sp, #FREGS_SIZE + 32
160
161SYM_INNER_LABEL(ftrace_caller_direct, SYM_L_LOCAL)
162	/*
163	 * Head to a direct trampoline in x17.
164	 *
165	 * We use `BR X17` as this can safely land on a `BTI C` or `PACIASP` in
166	 * the trampoline, and will not unbalance any return stack.
167	 */
168	br	x17
169#endif /* CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS */
170SYM_CODE_END(ftrace_caller)
171
172#ifdef CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS
173SYM_CODE_START(ftrace_stub_direct_tramp)
174	bti	c
175	mov	x10, x30
176	mov	x30, x9
177	ret	x10
178SYM_CODE_END(ftrace_stub_direct_tramp)
179#endif /* CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS */
180
181#else /* CONFIG_DYNAMIC_FTRACE_WITH_ARGS */
182
183/*
184 * Gcc with -pg will put the following code in the beginning of each function:
185 *      mov x0, x30
186 *      bl _mcount
187 *	[function's body ...]
188 * "bl _mcount" may be replaced to "bl ftrace_caller" or NOP if dynamic
189 * ftrace is enabled.
190 *
191 * Please note that x0 as an argument will not be used here because we can
192 * get lr(x30) of instrumented function at any time by winding up call stack
193 * as long as the kernel is compiled without -fomit-frame-pointer.
194 * (or CONFIG_FRAME_POINTER, this is forced on arm64)
195 *
196 * stack layout after mcount_enter in _mcount():
197 *
198 * current sp/fp =>  0:+-----+
199 * in _mcount()        | x29 | -> instrumented function's fp
200 *                     +-----+
201 *                     | x30 | -> _mcount()'s lr (= instrumented function's pc)
202 * old sp       => +16:+-----+
203 * when instrumented   |     |
204 * function calls      | ... |
205 * _mcount()           |     |
206 *                     |     |
207 * instrumented => +xx:+-----+
208 * function's fp       | x29 | -> parent's fp
209 *                     +-----+
210 *                     | x30 | -> instrumented function's lr (= parent's pc)
211 *                     +-----+
212 *                     | ... |
213 */
214
215	.macro mcount_enter
216	stp	x29, x30, [sp, #-16]!
217	mov	x29, sp
218	.endm
219
220	.macro mcount_exit
221	ldp	x29, x30, [sp], #16
222	ret
223	.endm
224
225	.macro mcount_adjust_addr rd, rn
226	sub	\rd, \rn, #AARCH64_INSN_SIZE
227	.endm
228
229	/* for instrumented function's parent */
230	.macro mcount_get_parent_fp reg
231	ldr	\reg, [x29]
232	ldr	\reg, [\reg]
233	.endm
234
235	/* for instrumented function */
236	.macro mcount_get_pc0 reg
237	mcount_adjust_addr	\reg, x30
238	.endm
239
240	.macro mcount_get_pc reg
241	ldr	\reg, [x29, #8]
242	mcount_adjust_addr	\reg, \reg
243	.endm
244
245	.macro mcount_get_lr reg
246	ldr	\reg, [x29]
247	ldr	\reg, [\reg, #8]
248	.endm
249
250	.macro mcount_get_lr_addr reg
251	ldr	\reg, [x29]
252	add	\reg, \reg, #8
253	.endm
254
255/*
256 * _mcount() is used to build the kernel with -pg option, but all the branch
257 * instructions to _mcount() are replaced to NOP initially at kernel start up,
258 * and later on, NOP to branch to ftrace_caller() when enabled or branch to
259 * NOP when disabled per-function base.
260 */
261SYM_FUNC_START(_mcount)
262	ret
263SYM_FUNC_END(_mcount)
264EXPORT_SYMBOL(_mcount)
265NOKPROBE(_mcount)
266
267/*
268 * void ftrace_caller(unsigned long return_address)
269 * @return_address: return address to instrumented function
270 *
271 * This function is a counterpart of _mcount() in 'static' ftrace, and
272 * makes calls to:
273 *     - tracer function to probe instrumented function's entry,
274 *     - ftrace_graph_caller to set up an exit hook
275 */
276SYM_FUNC_START(ftrace_caller)
277	mcount_enter
278
279	mcount_get_pc0	x0		//     function's pc
280	mcount_get_lr	x1		//     function's lr
281
282SYM_INNER_LABEL(ftrace_call, SYM_L_GLOBAL)	// tracer(pc, lr);
283	nop				// This will be replaced with "bl xxx"
284					// where xxx can be any kind of tracer.
285
286#ifdef CONFIG_FUNCTION_GRAPH_TRACER
287SYM_INNER_LABEL(ftrace_graph_call, SYM_L_GLOBAL) // ftrace_graph_caller();
288	nop				// If enabled, this will be replaced
289					// "b ftrace_graph_caller"
290#endif
291
292	mcount_exit
293SYM_FUNC_END(ftrace_caller)
294
295#ifdef CONFIG_FUNCTION_GRAPH_TRACER
296/*
297 * void ftrace_graph_caller(void)
298 *
299 * Called from _mcount() or ftrace_caller() when function_graph tracer is
300 * selected.
301 * This function w/ prepare_ftrace_return() fakes link register's value on
302 * the call stack in order to intercept instrumented function's return path
303 * and run return_to_handler() later on its exit.
304 */
305SYM_FUNC_START(ftrace_graph_caller)
306	mcount_get_pc		  x0	//     function's pc
307	mcount_get_lr_addr	  x1	//     pointer to function's saved lr
308	mcount_get_parent_fp	  x2	//     parent's fp
309	bl	prepare_ftrace_return	// prepare_ftrace_return(pc, &lr, fp)
310
311	mcount_exit
312SYM_FUNC_END(ftrace_graph_caller)
313#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
314#endif /* CONFIG_DYNAMIC_FTRACE_WITH_ARGS */
315
316SYM_TYPED_FUNC_START(ftrace_stub)
317	ret
318SYM_FUNC_END(ftrace_stub)
319
320#ifdef CONFIG_FUNCTION_GRAPH_TRACER
321SYM_TYPED_FUNC_START(ftrace_stub_graph)
322	ret
323SYM_FUNC_END(ftrace_stub_graph)
324
325/*
326 * void return_to_handler(void)
327 *
328 * Run ftrace_return_to_handler() before going back to parent.
329 * @fp is checked against the value passed by ftrace_graph_caller().
330 */
331SYM_CODE_START(return_to_handler)
332	/* save return value regs */
333	sub sp, sp, #64
334	stp x0, x1, [sp]
335	stp x2, x3, [sp, #16]
336	stp x4, x5, [sp, #32]
337	stp x6, x7, [sp, #48]
338
339	mov	x0, x29			//     parent's fp
340	bl	ftrace_return_to_handler// addr = ftrace_return_to_hander(fp);
341	mov	x30, x0			// restore the original return address
342
343	/* restore return value regs */
344	ldp x0, x1, [sp]
345	ldp x2, x3, [sp, #16]
346	ldp x4, x5, [sp, #32]
347	ldp x6, x7, [sp, #48]
348	add sp, sp, #64
349
350	ret
351SYM_CODE_END(return_to_handler)
352#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
353