1 | // SPDX-License-Identifier: GPL-2.0-or-later |
2 | /* |
3 | * Code for Kernel probes Jump optimization. |
4 | * |
5 | * Copyright 2017, Anju T, IBM Corp. |
6 | */ |
7 | |
8 | #include <linux/kprobes.h> |
9 | #include <linux/jump_label.h> |
10 | #include <linux/types.h> |
11 | #include <linux/slab.h> |
12 | #include <linux/list.h> |
13 | #include <asm/kprobes.h> |
14 | #include <asm/ptrace.h> |
15 | #include <asm/cacheflush.h> |
16 | #include <asm/code-patching.h> |
17 | #include <asm/sstep.h> |
18 | #include <asm/ppc-opcode.h> |
19 | #include <asm/inst.h> |
20 | |
21 | #define TMPL_CALL_HDLR_IDX (optprobe_template_call_handler - optprobe_template_entry) |
22 | #define TMPL_EMULATE_IDX (optprobe_template_call_emulate - optprobe_template_entry) |
23 | #define TMPL_RET_IDX (optprobe_template_ret - optprobe_template_entry) |
24 | #define TMPL_OP_IDX (optprobe_template_op_address - optprobe_template_entry) |
25 | #define TMPL_INSN_IDX (optprobe_template_insn - optprobe_template_entry) |
26 | #define TMPL_END_IDX (optprobe_template_end - optprobe_template_entry) |
27 | |
28 | static bool insn_page_in_use; |
29 | |
30 | void *alloc_optinsn_page(void) |
31 | { |
32 | if (insn_page_in_use) |
33 | return NULL; |
34 | insn_page_in_use = true; |
35 | return &optinsn_slot; |
36 | } |
37 | |
38 | void free_optinsn_page(void *page) |
39 | { |
40 | insn_page_in_use = false; |
41 | } |
42 | |
43 | /* |
44 | * Check if we can optimize this probe. Returns NIP post-emulation if this can |
45 | * be optimized and 0 otherwise. |
46 | */ |
47 | static unsigned long can_optimize(struct kprobe *p) |
48 | { |
49 | struct pt_regs regs; |
50 | struct instruction_op op; |
51 | unsigned long nip = 0; |
52 | unsigned long addr = (unsigned long)p->addr; |
53 | |
54 | /* |
55 | * kprobe placed for kretprobe during boot time |
56 | * has a 'nop' instruction, which can be emulated. |
57 | * So further checks can be skipped. |
58 | */ |
59 | if (p->addr == (kprobe_opcode_t *)&__kretprobe_trampoline) |
60 | return addr + sizeof(kprobe_opcode_t); |
61 | |
62 | /* |
63 | * We only support optimizing kernel addresses, but not |
64 | * module addresses. |
65 | * |
66 | * FIXME: Optimize kprobes placed in module addresses. |
67 | */ |
68 | if (!is_kernel_addr(addr)) |
69 | return 0; |
70 | |
71 | memset(®s, 0, sizeof(struct pt_regs)); |
72 | regs.nip = addr; |
73 | regs.trap = 0x0; |
74 | regs.msr = MSR_KERNEL; |
75 | |
76 | /* |
77 | * Kprobe placed in conditional branch instructions are |
78 | * not optimized, as we can't predict the nip prior with |
79 | * dummy pt_regs and can not ensure that the return branch |
80 | * from detour buffer falls in the range of address (i.e 32MB). |
81 | * A branch back from trampoline is set up in the detour buffer |
82 | * to the nip returned by the analyse_instr() here. |
83 | * |
84 | * Ensure that the instruction is not a conditional branch, |
85 | * and that can be emulated. |
86 | */ |
87 | if (!is_conditional_branch(ppc_inst_read(p->ainsn.insn)) && |
88 | analyse_instr(&op, ®s, ppc_inst_read(p->ainsn.insn)) == 1) { |
89 | emulate_update_regs(®s, &op); |
90 | nip = regs.nip; |
91 | } |
92 | |
93 | return nip; |
94 | } |
95 | |
96 | static void optimized_callback(struct optimized_kprobe *op, |
97 | struct pt_regs *regs) |
98 | { |
99 | /* This is possible if op is under delayed unoptimizing */ |
100 | if (kprobe_disabled(p: &op->kp)) |
101 | return; |
102 | |
103 | preempt_disable(); |
104 | |
105 | if (kprobe_running()) { |
106 | kprobes_inc_nmissed_count(p: &op->kp); |
107 | } else { |
108 | __this_cpu_write(current_kprobe, &op->kp); |
109 | regs_set_return_ip(regs, (unsigned long)op->kp.addr); |
110 | get_kprobe_ctlblk()->kprobe_status = KPROBE_HIT_ACTIVE; |
111 | opt_pre_handler(p: &op->kp, regs); |
112 | __this_cpu_write(current_kprobe, NULL); |
113 | } |
114 | |
115 | preempt_enable(); |
116 | } |
117 | NOKPROBE_SYMBOL(optimized_callback); |
118 | |
119 | void arch_remove_optimized_kprobe(struct optimized_kprobe *op) |
120 | { |
121 | if (op->optinsn.insn) { |
122 | free_optinsn_slot(slot: op->optinsn.insn, dirty: 1); |
123 | op->optinsn.insn = NULL; |
124 | } |
125 | } |
126 | |
127 | static void patch_imm32_load_insns(unsigned long val, int reg, kprobe_opcode_t *addr) |
128 | { |
129 | patch_instruction(addr++, ppc_inst(PPC_RAW_LIS(reg, PPC_HI(val)))); |
130 | patch_instruction(addr, ppc_inst(PPC_RAW_ORI(reg, reg, PPC_LO(val)))); |
131 | } |
132 | |
133 | /* |
134 | * Generate instructions to load provided immediate 64-bit value |
135 | * to register 'reg' and patch these instructions at 'addr'. |
136 | */ |
137 | static void patch_imm64_load_insns(unsigned long long val, int reg, kprobe_opcode_t *addr) |
138 | { |
139 | patch_instruction(addr++, ppc_inst(PPC_RAW_LIS(reg, PPC_HIGHEST(val)))); |
140 | patch_instruction(addr++, ppc_inst(PPC_RAW_ORI(reg, reg, PPC_HIGHER(val)))); |
141 | patch_instruction(addr++, ppc_inst(PPC_RAW_SLDI(reg, reg, 32))); |
142 | patch_instruction(addr++, ppc_inst(PPC_RAW_ORIS(reg, reg, PPC_HI(val)))); |
143 | patch_instruction(addr, ppc_inst(PPC_RAW_ORI(reg, reg, PPC_LO(val)))); |
144 | } |
145 | |
146 | static void patch_imm_load_insns(unsigned long val, int reg, kprobe_opcode_t *addr) |
147 | { |
148 | if (IS_ENABLED(CONFIG_PPC64)) |
149 | patch_imm64_load_insns(val, reg, addr); |
150 | else |
151 | patch_imm32_load_insns(val, reg, addr); |
152 | } |
153 | |
154 | int arch_prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p) |
155 | { |
156 | ppc_inst_t branch_op_callback, branch_emulate_step, temp; |
157 | unsigned long op_callback_addr, emulate_step_addr; |
158 | kprobe_opcode_t *buff; |
159 | long b_offset; |
160 | unsigned long nip, size; |
161 | int rc, i; |
162 | |
163 | nip = can_optimize(p); |
164 | if (!nip) |
165 | return -EILSEQ; |
166 | |
167 | /* Allocate instruction slot for detour buffer */ |
168 | buff = get_optinsn_slot(); |
169 | if (!buff) |
170 | return -ENOMEM; |
171 | |
172 | /* |
173 | * OPTPROBE uses 'b' instruction to branch to optinsn.insn. |
174 | * |
175 | * The target address has to be relatively nearby, to permit use |
176 | * of branch instruction in powerpc, because the address is specified |
177 | * in an immediate field in the instruction opcode itself, ie 24 bits |
178 | * in the opcode specify the address. Therefore the address should |
179 | * be within 32MB on either side of the current instruction. |
180 | */ |
181 | b_offset = (unsigned long)buff - (unsigned long)p->addr; |
182 | if (!is_offset_in_branch_range(b_offset)) |
183 | goto error; |
184 | |
185 | /* Check if the return address is also within 32MB range */ |
186 | b_offset = (unsigned long)(buff + TMPL_RET_IDX) - nip; |
187 | if (!is_offset_in_branch_range(b_offset)) |
188 | goto error; |
189 | |
190 | /* Setup template */ |
191 | /* We can optimize this via patch_instruction_window later */ |
192 | size = (TMPL_END_IDX * sizeof(kprobe_opcode_t)) / sizeof(int); |
193 | pr_devel("Copying template to %p, size %lu\n" , buff, size); |
194 | for (i = 0; i < size; i++) { |
195 | rc = patch_instruction(buff + i, ppc_inst(*(optprobe_template_entry + i))); |
196 | if (rc < 0) |
197 | goto error; |
198 | } |
199 | |
200 | /* |
201 | * Fixup the template with instructions to: |
202 | * 1. load the address of the actual probepoint |
203 | */ |
204 | patch_imm_load_insns((unsigned long)op, 3, buff + TMPL_OP_IDX); |
205 | |
206 | /* |
207 | * 2. branch to optimized_callback() and emulate_step() |
208 | */ |
209 | op_callback_addr = ppc_kallsyms_lookup_name("optimized_callback" ); |
210 | emulate_step_addr = ppc_kallsyms_lookup_name("emulate_step" ); |
211 | if (!op_callback_addr || !emulate_step_addr) { |
212 | WARN(1, "Unable to lookup optimized_callback()/emulate_step()\n" ); |
213 | goto error; |
214 | } |
215 | |
216 | rc = create_branch(&branch_op_callback, buff + TMPL_CALL_HDLR_IDX, |
217 | op_callback_addr, BRANCH_SET_LINK); |
218 | |
219 | rc |= create_branch(&branch_emulate_step, buff + TMPL_EMULATE_IDX, |
220 | emulate_step_addr, BRANCH_SET_LINK); |
221 | |
222 | if (rc) |
223 | goto error; |
224 | |
225 | patch_instruction(buff + TMPL_CALL_HDLR_IDX, branch_op_callback); |
226 | patch_instruction(buff + TMPL_EMULATE_IDX, branch_emulate_step); |
227 | |
228 | /* |
229 | * 3. load instruction to be emulated into relevant register, and |
230 | */ |
231 | temp = ppc_inst_read(p->ainsn.insn); |
232 | patch_imm_load_insns(val: ppc_inst_as_ulong(temp), reg: 4, addr: buff + TMPL_INSN_IDX); |
233 | |
234 | /* |
235 | * 4. branch back from trampoline |
236 | */ |
237 | patch_branch(buff + TMPL_RET_IDX, nip, 0); |
238 | |
239 | flush_icache_range(start: (unsigned long)buff, end: (unsigned long)(&buff[TMPL_END_IDX])); |
240 | |
241 | op->optinsn.insn = buff; |
242 | |
243 | return 0; |
244 | |
245 | error: |
246 | free_optinsn_slot(slot: buff, dirty: 0); |
247 | return -ERANGE; |
248 | |
249 | } |
250 | |
251 | int arch_prepared_optinsn(struct arch_optimized_insn *optinsn) |
252 | { |
253 | return optinsn->insn != NULL; |
254 | } |
255 | |
256 | /* |
257 | * On powerpc, Optprobes always replaces one instruction (4 bytes |
258 | * aligned and 4 bytes long). It is impossible to encounter another |
259 | * kprobe in this address range. So always return 0. |
260 | */ |
261 | int arch_check_optimized_kprobe(struct optimized_kprobe *op) |
262 | { |
263 | return 0; |
264 | } |
265 | |
266 | void arch_optimize_kprobes(struct list_head *oplist) |
267 | { |
268 | ppc_inst_t instr; |
269 | struct optimized_kprobe *op; |
270 | struct optimized_kprobe *tmp; |
271 | |
272 | list_for_each_entry_safe(op, tmp, oplist, list) { |
273 | /* |
274 | * Backup instructions which will be replaced |
275 | * by jump address |
276 | */ |
277 | memcpy(op->optinsn.copied_insn, op->kp.addr, RELATIVEJUMP_SIZE); |
278 | create_branch(&instr, op->kp.addr, (unsigned long)op->optinsn.insn, 0); |
279 | patch_instruction(op->kp.addr, instr); |
280 | list_del_init(entry: &op->list); |
281 | } |
282 | } |
283 | |
284 | void arch_unoptimize_kprobe(struct optimized_kprobe *op) |
285 | { |
286 | arch_arm_kprobe(p: &op->kp); |
287 | } |
288 | |
289 | void arch_unoptimize_kprobes(struct list_head *oplist, struct list_head *done_list) |
290 | { |
291 | struct optimized_kprobe *op; |
292 | struct optimized_kprobe *tmp; |
293 | |
294 | list_for_each_entry_safe(op, tmp, oplist, list) { |
295 | arch_unoptimize_kprobe(op); |
296 | list_move(list: &op->list, head: done_list); |
297 | } |
298 | } |
299 | |
300 | int arch_within_optimized_kprobe(struct optimized_kprobe *op, kprobe_opcode_t *addr) |
301 | { |
302 | return (op->kp.addr <= addr && |
303 | op->kp.addr + (RELATIVEJUMP_SIZE / sizeof(kprobe_opcode_t)) > addr); |
304 | } |
305 | |