1/* 2 * Transactional memory support routines to reclaim and recheckpoint 3 * transactional process state. 4 * 5 * Copyright 2012 Matt Evans & Michael Neuling, IBM Corporation. 6 */ 7 8#include <asm/asm-offsets.h> 9#include <asm/ppc_asm.h> 10#include <asm/ppc-opcode.h> 11#include <asm/ptrace.h> 12#include <asm/reg.h> 13#include <asm/bug.h> 14 15#ifdef CONFIG_VSX 16/* See fpu.S, this is borrowed from there */ 17#define __SAVE_32FPRS_VSRS(n,c,base) \ 18BEGIN_FTR_SECTION \ 19 b 2f; \ 20END_FTR_SECTION_IFSET(CPU_FTR_VSX); \ 21 SAVE_32FPRS(n,base); \ 22 b 3f; \ 232: SAVE_32VSRS(n,c,base); \ 243: 25#define __REST_32FPRS_VSRS(n,c,base) \ 26BEGIN_FTR_SECTION \ 27 b 2f; \ 28END_FTR_SECTION_IFSET(CPU_FTR_VSX); \ 29 REST_32FPRS(n,base); \ 30 b 3f; \ 312: REST_32VSRS(n,c,base); \ 323: 33#else 34#define __SAVE_32FPRS_VSRS(n,c,base) SAVE_32FPRS(n, base) 35#define __REST_32FPRS_VSRS(n,c,base) REST_32FPRS(n, base) 36#endif 37#define SAVE_32FPRS_VSRS(n,c,base) \ 38 __SAVE_32FPRS_VSRS(n,__REG_##c,__REG_##base) 39#define REST_32FPRS_VSRS(n,c,base) \ 40 __REST_32FPRS_VSRS(n,__REG_##c,__REG_##base) 41 42/* Stack frame offsets for local variables. */ 43#define TM_FRAME_L0 TM_FRAME_SIZE-16 44#define TM_FRAME_L1 TM_FRAME_SIZE-8 45 46 47/* In order to access the TM SPRs, TM must be enabled. So, do so: */ 48_GLOBAL(tm_enable) 49 mfmsr r4 50 li r3, MSR_TM >> 32 51 sldi r3, r3, 32 52 and. r0, r4, r3 53 bne 1f 54 or r4, r4, r3 55 mtmsrd r4 561: blr 57 58_GLOBAL(tm_save_sprs) 59 mfspr r0, SPRN_TFHAR 60 std r0, THREAD_TM_TFHAR(r3) 61 mfspr r0, SPRN_TEXASR 62 std r0, THREAD_TM_TEXASR(r3) 63 mfspr r0, SPRN_TFIAR 64 std r0, THREAD_TM_TFIAR(r3) 65 blr 66 67_GLOBAL(tm_restore_sprs) 68 ld r0, THREAD_TM_TFHAR(r3) 69 mtspr SPRN_TFHAR, r0 70 ld r0, THREAD_TM_TEXASR(r3) 71 mtspr SPRN_TEXASR, r0 72 ld r0, THREAD_TM_TFIAR(r3) 73 mtspr SPRN_TFIAR, r0 74 blr 75 76 /* Passed an 8-bit failure cause as first argument. */ 77_GLOBAL(tm_abort) 78 TABORT(R3) 79 blr 80 81/* void tm_reclaim(struct thread_struct *thread, 82 * unsigned long orig_msr, 83 * uint8_t cause) 84 * 85 * - Performs a full reclaim. This destroys outstanding 86 * transactions and updates thread->regs.tm_ckpt_* with the 87 * original checkpointed state. Note that thread->regs is 88 * unchanged. 89 * - FP regs are written back to thread->transact_fpr before 90 * reclaiming. These are the transactional (current) versions. 91 * 92 * Purpose is to both abort transactions of, and preserve the state of, 93 * a transactions at a context switch. We preserve/restore both sets of process 94 * state to restore them when the thread's scheduled again. We continue in 95 * userland as though nothing happened, but when the transaction is resumed 96 * they will abort back to the checkpointed state we save out here. 97 * 98 * Call with IRQs off, stacks get all out of sync for some periods in here! 99 */ 100_GLOBAL(tm_reclaim) 101 mfcr r6 102 mflr r0 103 stw r6, 8(r1) 104 std r0, 16(r1) 105 std r2, STK_GOT(r1) 106 stdu r1, -TM_FRAME_SIZE(r1) 107 108 /* We've a struct pt_regs at [r1+STACK_FRAME_OVERHEAD]. */ 109 110 std r3, STK_PARAM(R3)(r1) 111 std r4, STK_PARAM(R4)(r1) 112 SAVE_NVGPRS(r1) 113 114 /* We need to setup MSR for VSX register save instructions. */ 115 mfmsr r14 116 mr r15, r14 117 ori r15, r15, MSR_FP 118 li r16, 0 119 ori r16, r16, MSR_EE /* IRQs hard off */ 120 andc r15, r15, r16 121 oris r15, r15, MSR_VEC@h 122#ifdef CONFIG_VSX 123 BEGIN_FTR_SECTION 124 oris r15,r15, MSR_VSX@h 125 END_FTR_SECTION_IFSET(CPU_FTR_VSX) 126#endif 127 mtmsrd r15 128 std r14, TM_FRAME_L0(r1) 129 130 /* Do sanity check on MSR to make sure we are suspended */ 131 li r7, (MSR_TS_S)@higher 132 srdi r6, r14, 32 133 and r6, r6, r7 1341: tdeqi r6, 0 135 EMIT_BUG_ENTRY 1b,__FILE__,__LINE__,0 136 137 /* Stash the stack pointer away for use after reclaim */ 138 std r1, PACAR1(r13) 139 140 /* Clear MSR RI since we are about to change r1, EE is already off. */ 141 li r4, 0 142 mtmsrd r4, 1 143 144 /* 145 * BE CAREFUL HERE: 146 * At this point we can't take an SLB miss since we have MSR_RI 147 * off. Load only to/from the stack/paca which are in SLB bolted regions 148 * until we turn MSR RI back on. 149 * 150 * The moment we treclaim, ALL of our GPRs will switch 151 * to user register state. (FPRs, CCR etc. also!) 152 * Use an sprg and a tm_scratch in the PACA to shuffle. 153 */ 154 TRECLAIM(R5) /* Cause in r5 */ 155 156 /* ******************** GPRs ******************** */ 157 /* Stash the checkpointed r13 away in the scratch SPR and get the real 158 * paca 159 */ 160 SET_SCRATCH0(r13) 161 GET_PACA(r13) 162 163 /* Stash the checkpointed r1 away in paca tm_scratch and get the real 164 * stack pointer back 165 */ 166 std r1, PACATMSCRATCH(r13) 167 ld r1, PACAR1(r13) 168 169 /* Store the PPR in r11 and reset to decent value */ 170 std r11, GPR11(r1) /* Temporary stash */ 171 172 /* Reset MSR RI so we can take SLB faults again */ 173 li r11, MSR_RI 174 mtmsrd r11, 1 175 176 mfspr r11, SPRN_PPR 177 HMT_MEDIUM 178 179 /* Now get some more GPRS free */ 180 std r7, GPR7(r1) /* Temporary stash */ 181 std r12, GPR12(r1) /* '' '' '' */ 182 ld r12, STK_PARAM(R3)(r1) /* Param 0, thread_struct * */ 183 184 std r11, THREAD_TM_PPR(r12) /* Store PPR and free r11 */ 185 186 addi r7, r12, PT_CKPT_REGS /* Thread's ckpt_regs */ 187 188 /* Make r7 look like an exception frame so that we 189 * can use the neat GPRx(n) macros. r7 is NOT a pt_regs ptr! 190 */ 191 subi r7, r7, STACK_FRAME_OVERHEAD 192 193 /* Sync the userland GPRs 2-12, 14-31 to thread->regs: */ 194 SAVE_GPR(0, r7) /* user r0 */ 195 SAVE_GPR(2, r7) /* user r2 */ 196 SAVE_4GPRS(3, r7) /* user r3-r6 */ 197 SAVE_GPR(8, r7) /* user r8 */ 198 SAVE_GPR(9, r7) /* user r9 */ 199 SAVE_GPR(10, r7) /* user r10 */ 200 ld r3, PACATMSCRATCH(r13) /* user r1 */ 201 ld r4, GPR7(r1) /* user r7 */ 202 ld r5, GPR11(r1) /* user r11 */ 203 ld r6, GPR12(r1) /* user r12 */ 204 GET_SCRATCH0(8) /* user r13 */ 205 std r3, GPR1(r7) 206 std r4, GPR7(r7) 207 std r5, GPR11(r7) 208 std r6, GPR12(r7) 209 std r8, GPR13(r7) 210 211 SAVE_NVGPRS(r7) /* user r14-r31 */ 212 213 /* ******************** NIP ******************** */ 214 mfspr r3, SPRN_TFHAR 215 std r3, _NIP(r7) /* Returns to failhandler */ 216 /* The checkpointed NIP is ignored when rescheduling/rechkpting, 217 * but is used in signal return to 'wind back' to the abort handler. 218 */ 219 220 /* ******************** CR,LR,CCR,MSR ********** */ 221 mfctr r3 222 mflr r4 223 mfcr r5 224 mfxer r6 225 226 std r3, _CTR(r7) 227 std r4, _LINK(r7) 228 std r5, _CCR(r7) 229 std r6, _XER(r7) 230 231 232 /* ******************** TAR, DSCR ********** */ 233 mfspr r3, SPRN_TAR 234 mfspr r4, SPRN_DSCR 235 236 std r3, THREAD_TM_TAR(r12) 237 std r4, THREAD_TM_DSCR(r12) 238 239 /* MSR and flags: We don't change CRs, and we don't need to alter 240 * MSR. 241 */ 242 243 244 /* ******************** FPR/VR/VSRs ************ 245 * After reclaiming, capture the checkpointed FPRs/VRs /if used/. 246 * 247 * (If VSX used, FP and VMX are implied. Or, we don't need to look 248 * at MSR.VSX as copying FP regs if .FP, vector regs if .VMX covers it.) 249 * 250 * We're passed the thread's MSR as the second parameter 251 * 252 * We enabled VEC/FP/VSX in the msr above, so we can execute these 253 * instructions! 254 */ 255 ld r4, STK_PARAM(R4)(r1) /* Second parameter, MSR * */ 256 mr r3, r12 257 andis. r0, r4, MSR_VEC@h 258 beq dont_backup_vec 259 260 addi r7, r3, THREAD_CKVRSTATE 261 SAVE_32VRS(0, r6, r7) /* r6 scratch, r7 transact vr state */ 262 mfvscr v0 263 li r6, VRSTATE_VSCR 264 stvx v0, r7, r6 265dont_backup_vec: 266 mfspr r0, SPRN_VRSAVE 267 std r0, THREAD_CKVRSAVE(r3) 268 269 andi. r0, r4, MSR_FP 270 beq dont_backup_fp 271 272 addi r7, r3, THREAD_CKFPSTATE 273 SAVE_32FPRS_VSRS(0, R6, R7) /* r6 scratch, r7 transact fp state */ 274 275 mffs fr0 276 stfd fr0,FPSTATE_FPSCR(r7) 277 278dont_backup_fp: 279 280 /* TM regs, incl TEXASR -- these live in thread_struct. Note they've 281 * been updated by the treclaim, to explain to userland the failure 282 * cause (aborted). 283 */ 284 mfspr r0, SPRN_TEXASR 285 mfspr r3, SPRN_TFHAR 286 mfspr r4, SPRN_TFIAR 287 std r0, THREAD_TM_TEXASR(r12) 288 std r3, THREAD_TM_TFHAR(r12) 289 std r4, THREAD_TM_TFIAR(r12) 290 291 /* AMR is checkpointed too, but is unsupported by Linux. */ 292 293 /* Restore original MSR/IRQ state & clear TM mode */ 294 ld r14, TM_FRAME_L0(r1) /* Orig MSR */ 295 296 li r15, 0 297 rldimi r14, r15, MSR_TS_LG, (63-MSR_TS_LG)-1 298 mtmsrd r14 299 300 REST_NVGPRS(r1) 301 302 addi r1, r1, TM_FRAME_SIZE 303 lwz r4, 8(r1) 304 ld r0, 16(r1) 305 mtcr r4 306 mtlr r0 307 ld r2, STK_GOT(r1) 308 309 /* Load CPU's default DSCR */ 310 ld r0, PACA_DSCR_DEFAULT(r13) 311 mtspr SPRN_DSCR, r0 312 313 blr 314 315 316 /* void __tm_recheckpoint(struct thread_struct *thread, 317 * unsigned long orig_msr) 318 * - Restore the checkpointed register state saved by tm_reclaim 319 * when we switch_to a process. 320 * 321 * Call with IRQs off, stacks get all out of sync for 322 * some periods in here! 323 */ 324_GLOBAL(__tm_recheckpoint) 325 mfcr r5 326 mflr r0 327 stw r5, 8(r1) 328 std r0, 16(r1) 329 std r2, STK_GOT(r1) 330 stdu r1, -TM_FRAME_SIZE(r1) 331 332 /* We've a struct pt_regs at [r1+STACK_FRAME_OVERHEAD]. 333 * This is used for backing up the NVGPRs: 334 */ 335 SAVE_NVGPRS(r1) 336 337 /* Load complete register state from ts_ckpt* registers */ 338 339 addi r7, r3, PT_CKPT_REGS /* Thread's ckpt_regs */ 340 341 /* Make r7 look like an exception frame so that we 342 * can use the neat GPRx(n) macros. r7 is now NOT a pt_regs ptr! 343 */ 344 subi r7, r7, STACK_FRAME_OVERHEAD 345 346 mfmsr r6 347 /* R4 = original MSR to indicate whether thread used FP/Vector etc. */ 348 349 /* Enable FP/vec in MSR if necessary! */ 350 lis r5, MSR_VEC@h 351 ori r5, r5, MSR_FP 352 and. r5, r4, r5 353 beq restore_gprs /* if neither, skip both */ 354 355#ifdef CONFIG_VSX 356 BEGIN_FTR_SECTION 357 oris r5, r5, MSR_VSX@h 358 END_FTR_SECTION_IFSET(CPU_FTR_VSX) 359#endif 360 or r5, r6, r5 /* Set MSR.FP+.VSX/.VEC */ 361 mtmsr r5 362 363#ifdef CONFIG_ALTIVEC 364 /* 365 * FP and VEC registers: These are recheckpointed from 366 * thread.ckfp_state and thread.ckvr_state respectively. The 367 * thread.fp_state[] version holds the 'live' (transactional) 368 * and will be loaded subsequently by any FPUnavailable trap. 369 */ 370 andis. r0, r4, MSR_VEC@h 371 beq dont_restore_vec 372 373 addi r8, r3, THREAD_CKVRSTATE 374 li r5, VRSTATE_VSCR 375 lvx v0, r8, r5 376 mtvscr v0 377 REST_32VRS(0, r5, r8) /* r5 scratch, r8 ptr */ 378dont_restore_vec: 379 ld r5, THREAD_CKVRSAVE(r3) 380 mtspr SPRN_VRSAVE, r5 381#endif 382 383 andi. r0, r4, MSR_FP 384 beq dont_restore_fp 385 386 addi r8, r3, THREAD_CKFPSTATE 387 lfd fr0, FPSTATE_FPSCR(r8) 388 MTFSF_L(fr0) 389 REST_32FPRS_VSRS(0, R4, R8) 390 391dont_restore_fp: 392 mtmsr r6 /* FP/Vec off again! */ 393 394restore_gprs: 395 396 /* ******************** CR,LR,CCR,MSR ********** */ 397 ld r4, _CTR(r7) 398 ld r5, _LINK(r7) 399 ld r8, _XER(r7) 400 401 mtctr r4 402 mtlr r5 403 mtxer r8 404 405 /* ******************** TAR ******************** */ 406 ld r4, THREAD_TM_TAR(r3) 407 mtspr SPRN_TAR, r4 408 409 /* Load up the PPR and DSCR in GPRs only at this stage */ 410 ld r5, THREAD_TM_DSCR(r3) 411 ld r6, THREAD_TM_PPR(r3) 412 413 REST_GPR(0, r7) /* GPR0 */ 414 REST_2GPRS(2, r7) /* GPR2-3 */ 415 REST_GPR(4, r7) /* GPR4 */ 416 REST_4GPRS(8, r7) /* GPR8-11 */ 417 REST_2GPRS(12, r7) /* GPR12-13 */ 418 419 REST_NVGPRS(r7) /* GPR14-31 */ 420 421 /* Load up PPR and DSCR here so we don't run with user values for long 422 */ 423 mtspr SPRN_DSCR, r5 424 mtspr SPRN_PPR, r6 425 426 /* Do final sanity check on TEXASR to make sure FS is set. Do this 427 * here before we load up the userspace r1 so any bugs we hit will get 428 * a call chain */ 429 mfspr r5, SPRN_TEXASR 430 srdi r5, r5, 16 431 li r6, (TEXASR_FS)@h 432 and r6, r6, r5 4331: tdeqi r6, 0 434 EMIT_BUG_ENTRY 1b,__FILE__,__LINE__,0 435 436 /* Do final sanity check on MSR to make sure we are not transactional 437 * or suspended 438 */ 439 mfmsr r6 440 li r5, (MSR_TS_MASK)@higher 441 srdi r6, r6, 32 442 and r6, r6, r5 4431: tdnei r6, 0 444 EMIT_BUG_ENTRY 1b,__FILE__,__LINE__,0 445 446 /* Restore CR */ 447 ld r6, _CCR(r7) 448 mtcr r6 449 450 REST_GPR(6, r7) 451 452 /* 453 * Store r1 and r5 on the stack so that we can access them 454 * after we clear MSR RI. 455 */ 456 457 REST_GPR(5, r7) 458 std r5, -8(r1) 459 ld r5, GPR1(r7) 460 std r5, -16(r1) 461 462 REST_GPR(7, r7) 463 464 /* Clear MSR RI since we are about to change r1. EE is already off */ 465 li r5, 0 466 mtmsrd r5, 1 467 468 /* 469 * BE CAREFUL HERE: 470 * At this point we can't take an SLB miss since we have MSR_RI 471 * off. Load only to/from the stack/paca which are in SLB bolted regions 472 * until we turn MSR RI back on. 473 */ 474 475 SET_SCRATCH0(r1) 476 ld r5, -8(r1) 477 ld r1, -16(r1) 478 479 /* Commit register state as checkpointed state: */ 480 TRECHKPT 481 482 HMT_MEDIUM 483 484 /* Our transactional state has now changed. 485 * 486 * Now just get out of here. Transactional (current) state will be 487 * updated once restore is called on the return path in the _switch-ed 488 * -to process. 489 */ 490 491 GET_PACA(r13) 492 GET_SCRATCH0(r1) 493 494 /* R1 is restored, so we are recoverable again. EE is still off */ 495 li r4, MSR_RI 496 mtmsrd r4, 1 497 498 REST_NVGPRS(r1) 499 500 addi r1, r1, TM_FRAME_SIZE 501 lwz r4, 8(r1) 502 ld r0, 16(r1) 503 mtcr r4 504 mtlr r0 505 ld r2, STK_GOT(r1) 506 507 /* Load CPU's default DSCR */ 508 ld r0, PACA_DSCR_DEFAULT(r13) 509 mtspr SPRN_DSCR, r0 510 511 blr 512 513 /* ****************************************************************** */ 514