arch/powerpc/kernel/optprobes.c

   1 /*
   2  * Code for Kernel probes Jump optimization.
   3  *
   4  * Copyright 2017, Anju T, IBM Corp.
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License
   8  * as published by the Free Software Foundation; either version
   9  * 2 of the License, or (at your option) any later version.
  10  */
  11
  12 #include <linux/kprobes.h>
  13 #include <linux/jump_label.h>
  14 #include <linux/types.h>
  15 #include <linux/slab.h>
  16 #include <linux/list.h>
  17 #include <asm/kprobes.h>
  18 #include <asm/ptrace.h>
  19 #include <asm/cacheflush.h>
  20 #include <asm/code-patching.h>
  21 #include <asm/sstep.h>
  22 #include <asm/ppc-opcode.h>
  23
  24 #define TMPL_CALL_HDLR_IDX      \
  25         (optprobe_template_call_handler - optprobe_template_entry)
  26 #define TMPL_EMULATE_IDX        \
  27         (optprobe_template_call_emulate - optprobe_template_entry)
  28 #define TMPL_RET_IDX            \
  29         (optprobe_template_ret - optprobe_template_entry)
  30 #define TMPL_OP_IDX             \
  31         (optprobe_template_op_address - optprobe_template_entry)
  32 #define TMPL_INSN_IDX           \
  33         (optprobe_template_insn - optprobe_template_entry)
  34 #define TMPL_END_IDX            \
  35         (optprobe_template_end - optprobe_template_entry)
  36
  37 DEFINE_INSN_CACHE_OPS(ppc_optinsn);
  38
  39 static bool insn_page_in_use;
  40
  41 static void *__ppc_alloc_insn_page(void)
  42 {
  43         if (insn_page_in_use)
  44                 return NULL;
  45         insn_page_in_use = true;
  46         return &optinsn_slot;
  47 }
  48
  49 static void __ppc_free_insn_page(void *page __maybe_unused)
  50 {
  51         insn_page_in_use = false;
  52 }
  53
  54 struct kprobe_insn_cache kprobe_ppc_optinsn_slots = {
  55         .mutex = __MUTEX_INITIALIZER(kprobe_ppc_optinsn_slots.mutex),
  56         .pages = LIST_HEAD_INIT(kprobe_ppc_optinsn_slots.pages),
  57         /* insn_size initialized later */
  58         .alloc = __ppc_alloc_insn_page,
  59         .free = __ppc_free_insn_page,
  60         .nr_garbage = 0,
  61 };
  62
  63 /*
  64  * Check if we can optimize this probe. Returns NIP post-emulation if this can
  65  * be optimized and 0 otherwise.
  66  */
  67 static unsigned long can_optimize(struct kprobe *p)
  68 {
  69         struct pt_regs regs;
  70         struct instruction_op op;
  71         unsigned long nip = 0;
  72
  73         /*
  74          * kprobe placed for kretprobe during boot time
  75          * is not optimizing now.
  76          *
  77          * TODO: Optimize kprobe in kretprobe_trampoline
  78          */
  79         if (p->addr == (kprobe_opcode_t *)&kretprobe_trampoline)
  80                 return 0;
  81
  82         /*
  83          * We only support optimizing kernel addresses, but not
  84          * module addresses.
  85          *
  86          * FIXME: Optimize kprobes placed in module addresses.
  87          */
  88         if (!is_kernel_addr((unsigned long)p->addr))
  89                 return 0;
  90
  91         memset(&regs, 0, sizeof(struct pt_regs));
  92         regs.nip = (unsigned long)p->addr;
  93         regs.trap = 0x0;
  94         regs.msr = MSR_KERNEL;
  95
  96         /*
  97          * Kprobe placed in conditional branch instructions are
  98          * not optimized, as we can't predict the nip prior with
  99          * dummy pt_regs and can not ensure that the return branch
 100          * from detour buffer falls in the range of address (i.e 32MB).
 101          * A branch back from trampoline is set up in the detour buffer
 102          * to the nip returned by the analyse_instr() here.
 103          *
 104          * Ensure that the instruction is not a conditional branch,
 105          * and that can be emulated.
 106          */
 107         if (!is_conditional_branch(*p->ainsn.insn) &&
 108                         analyse_instr(&op, &regs, *p->ainsn.insn))
 109                 nip = regs.nip;
 110
 111         return nip;
 112 }
 113
 114 static void optimized_callback(struct optimized_kprobe *op,
 115                                struct pt_regs *regs)
 116 {
 117         struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
 118         unsigned long flags;
 119
 120         /* This is possible if op is under delayed unoptimizing */
 121         if (kprobe_disabled(&op->kp))
 122                 return;
 123
 124         local_irq_save(flags);
 125         hard_irq_disable();
 126
 127         if (kprobe_running()) {
 128                 kprobes_inc_nmissed_count(&op->kp);
 129         } else {
 130                 __this_cpu_write(current_kprobe, &op->kp);
 131                 regs->nip = (unsigned long)op->kp.addr;
 132                 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
 133                 opt_pre_handler(&op->kp, regs);
 134                 __this_cpu_write(current_kprobe, NULL);
 135         }
 136
 137         /*
 138          * No need for an explicit __hard_irq_enable() here.
 139          * local_irq_restore() will re-enable interrupts,
 140          * if they were hard disabled.
 141          */
 142         local_irq_restore(flags);
 143 }
 144 NOKPROBE_SYMBOL(optimized_callback);
 145
 146 void arch_remove_optimized_kprobe(struct optimized_kprobe *op)
 147 {
 148         if (op->optinsn.insn) {
 149                 free_ppc_optinsn_slot(op->optinsn.insn, 1);
 150                 op->optinsn.insn = NULL;
 151         }
 152 }
 153
 154 /*
 155  * emulate_step() requires insn to be emulated as
 156  * second parameter. Load register 'r4' with the
 157  * instruction.
 158  */
 159 void patch_imm32_load_insns(unsigned int val, kprobe_opcode_t *addr)
 160 {
 161         /* addis r4,0,(insn)@h */
 162         *addr++ = PPC_INST_ADDIS | ___PPC_RT(4) |
 163                   ((val >> 16) & 0xffff);
 164
 165         /* ori r4,r4,(insn)@l */
 166         *addr = PPC_INST_ORI | ___PPC_RA(4) | ___PPC_RS(4) |
 167                 (val & 0xffff);
 168 }
 169
 170 /*
 171  * Generate instructions to load provided immediate 64-bit value
 172  * to register 'r3' and patch these instructions at 'addr'.
 173  */
 174 void patch_imm64_load_insns(unsigned long val, kprobe_opcode_t *addr)
 175 {
 176         /* lis r3,(op)@highest */
 177         *addr++ = PPC_INST_ADDIS | ___PPC_RT(3) |
 178                   ((val >> 48) & 0xffff);
 179
 180         /* ori r3,r3,(op)@higher */
 181         *addr++ = PPC_INST_ORI | ___PPC_RA(3) | ___PPC_RS(3) |
 182                   ((val >> 32) & 0xffff);
 183
 184         /* rldicr r3,r3,32,31 */
 185         *addr++ = PPC_INST_RLDICR | ___PPC_RA(3) | ___PPC_RS(3) |
 186                   __PPC_SH64(32) | __PPC_ME64(31);
 187
 188         /* oris r3,r3,(op)@h */
 189         *addr++ = PPC_INST_ORIS | ___PPC_RA(3) | ___PPC_RS(3) |
 190                   ((val >> 16) & 0xffff);
 191
 192         /* ori r3,r3,(op)@l */
 193         *addr = PPC_INST_ORI | ___PPC_RA(3) | ___PPC_RS(3) |
 194                 (val & 0xffff);
 195 }
 196
 197 int arch_prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p)
 198 {
 199         kprobe_opcode_t *buff, branch_op_callback, branch_emulate_step;
 200         kprobe_opcode_t *op_callback_addr, *emulate_step_addr;
 201         long b_offset;
 202         unsigned long nip;
 203
 204         kprobe_ppc_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
 205
 206         nip = can_optimize(p);
 207         if (!nip)
 208                 return -EILSEQ;
 209
 210         /* Allocate instruction slot for detour buffer */
 211         buff = get_ppc_optinsn_slot();
 212         if (!buff)
 213                 return -ENOMEM;
 214
 215         /*
 216          * OPTPROBE uses 'b' instruction to branch to optinsn.insn.
 217          *
 218          * The target address has to be relatively nearby, to permit use
 219          * of branch instruction in powerpc, because the address is specified
 220          * in an immediate field in the instruction opcode itself, ie 24 bits
 221          * in the opcode specify the address. Therefore the address should
 222          * be within 32MB on either side of the current instruction.
 223          */
 224         b_offset = (unsigned long)buff - (unsigned long)p->addr;
 225         if (!is_offset_in_branch_range(b_offset))
 226                 goto error;
 227
 228         /* Check if the return address is also within 32MB range */
 229         b_offset = (unsigned long)(buff + TMPL_RET_IDX) -
 230                         (unsigned long)nip;
 231         if (!is_offset_in_branch_range(b_offset))
 232                 goto error;
 233
 234         /* Setup template */
 235         memcpy(buff, optprobe_template_entry,
 236                         TMPL_END_IDX * sizeof(kprobe_opcode_t));
 237
 238         /*
 239          * Fixup the template with instructions to:
 240          * 1. load the address of the actual probepoint
 241          */
 242         patch_imm64_load_insns((unsigned long)op, buff + TMPL_OP_IDX);
 243
 244         /*
 245          * 2. branch to optimized_callback() and emulate_step()
 246          */
 247         kprobe_lookup_name("optimized_callback", op_callback_addr);
 248         kprobe_lookup_name("emulate_step", emulate_step_addr);
 249         if (!op_callback_addr || !emulate_step_addr) {
 250                 WARN(1, "kprobe_lookup_name() failed\n");
 251                 goto error;
 252         }
 253
 254         branch_op_callback = create_branch((unsigned int *)buff + TMPL_CALL_HDLR_IDX,
 255                                 (unsigned long)op_callback_addr,
 256                                 BRANCH_SET_LINK);
 257
 258         branch_emulate_step = create_branch((unsigned int *)buff + TMPL_EMULATE_IDX,
 259                                 (unsigned long)emulate_step_addr,
 260                                 BRANCH_SET_LINK);
 261
 262         if (!branch_op_callback || !branch_emulate_step)
 263                 goto error;
 264
 265         buff[TMPL_CALL_HDLR_IDX] = branch_op_callback;
 266         buff[TMPL_EMULATE_IDX] = branch_emulate_step;
 267
 268         /*
 269          * 3. load instruction to be emulated into relevant register, and
 270          */
 271         patch_imm32_load_insns(*p->ainsn.insn, buff + TMPL_INSN_IDX);
 272
 273         /*
 274          * 4. branch back from trampoline
 275          */
 276         buff[TMPL_RET_IDX] = create_branch((unsigned int *)buff + TMPL_RET_IDX,
 277                                 (unsigned long)nip, 0);
 278
 279         flush_icache_range((unsigned long)buff,
 280                            (unsigned long)(&buff[TMPL_END_IDX]));
 281
 282         op->optinsn.insn = buff;
 283
 284         return 0;
 285
 286 error:
 287         free_ppc_optinsn_slot(buff, 0);
 288         return -ERANGE;
 289
 290 }
 291
 292 int arch_prepared_optinsn(struct arch_optimized_insn *optinsn)
 293 {
 294         return optinsn->insn != NULL;
 295 }
 296
 297 /*
 298  * On powerpc, Optprobes always replaces one instruction (4 bytes
 299  * aligned and 4 bytes long). It is impossible to encounter another
 300  * kprobe in this address range. So always return 0.
 301  */
 302 int arch_check_optimized_kprobe(struct optimized_kprobe *op)
 303 {
 304         return 0;
 305 }
 306
 307 void arch_optimize_kprobes(struct list_head *oplist)
 308 {
 309         struct optimized_kprobe *op;
 310         struct optimized_kprobe *tmp;
 311
 312         list_for_each_entry_safe(op, tmp, oplist, list) {
 313                 /*
 314                  * Backup instructions which will be replaced
 315                  * by jump address
 316                  */
 317                 memcpy(op->optinsn.copied_insn, op->kp.addr,
 318                                                RELATIVEJUMP_SIZE);
 319                 patch_instruction(op->kp.addr,
 320                         create_branch((unsigned int *)op->kp.addr,
 321                                       (unsigned long)op->optinsn.insn, 0));
 322                 list_del_init(&op->list);
 323         }
 324 }
 325
 326 void arch_unoptimize_kprobe(struct optimized_kprobe *op)
 327 {
 328         arch_arm_kprobe(&op->kp);
 329 }
 330
 331 void arch_unoptimize_kprobes(struct list_head *oplist,
 332                              struct list_head *done_list)
 333 {
 334         struct optimized_kprobe *op;
 335         struct optimized_kprobe *tmp;
 336
 337         list_for_each_entry_safe(op, tmp, oplist, list) {
 338                 arch_unoptimize_kprobe(op);
 339                 list_move(&op->list, done_list);
 340         }
 341 }
 342
 343 int arch_within_optimized_kprobe(struct optimized_kprobe *op,
 344                                  unsigned long addr)
 345 {
 346         return ((unsigned long)op->kp.addr <= addr &&
 347                 (unsigned long)op->kp.addr + RELATIVEJUMP_SIZE > addr);
 348 }