Linux Audio

Check our new training course

Loading...
v5.9
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 *  Derived from "arch/i386/kernel/process.c"
   4 *    Copyright (C) 1995  Linus Torvalds
   5 *
   6 *  Updated and modified by Cort Dougan (cort@cs.nmt.edu) and
   7 *  Paul Mackerras (paulus@cs.anu.edu.au)
   8 *
   9 *  PowerPC version
  10 *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
  11 */
  12
  13#include <linux/errno.h>
  14#include <linux/sched.h>
  15#include <linux/sched/debug.h>
  16#include <linux/sched/task.h>
  17#include <linux/sched/task_stack.h>
  18#include <linux/kernel.h>
  19#include <linux/mm.h>
  20#include <linux/smp.h>
  21#include <linux/stddef.h>
  22#include <linux/unistd.h>
  23#include <linux/ptrace.h>
  24#include <linux/slab.h>
  25#include <linux/user.h>
  26#include <linux/elf.h>
  27#include <linux/prctl.h>
  28#include <linux/init_task.h>
  29#include <linux/export.h>
  30#include <linux/kallsyms.h>
  31#include <linux/mqueue.h>
  32#include <linux/hardirq.h>
  33#include <linux/utsname.h>
  34#include <linux/ftrace.h>
  35#include <linux/kernel_stat.h>
  36#include <linux/personality.h>
  37#include <linux/random.h>
  38#include <linux/hw_breakpoint.h>
  39#include <linux/uaccess.h>
  40#include <linux/elf-randomize.h>
  41#include <linux/pkeys.h>
  42#include <linux/seq_buf.h>
  43
 
  44#include <asm/io.h>
  45#include <asm/processor.h>
  46#include <asm/mmu.h>
  47#include <asm/prom.h>
  48#include <asm/machdep.h>
  49#include <asm/time.h>
  50#include <asm/runlatch.h>
  51#include <asm/syscalls.h>
  52#include <asm/switch_to.h>
  53#include <asm/tm.h>
  54#include <asm/debug.h>
  55#ifdef CONFIG_PPC64
  56#include <asm/firmware.h>
  57#include <asm/hw_irq.h>
  58#endif
  59#include <asm/code-patching.h>
  60#include <asm/exec.h>
  61#include <asm/livepatch.h>
  62#include <asm/cpu_has_feature.h>
  63#include <asm/asm-prototypes.h>
  64#include <asm/stacktrace.h>
  65#include <asm/hw_breakpoint.h>
  66
  67#include <linux/kprobes.h>
  68#include <linux/kdebug.h>
  69
  70/* Transactional Memory debug */
  71#ifdef TM_DEBUG_SW
  72#define TM_DEBUG(x...) printk(KERN_INFO x)
  73#else
  74#define TM_DEBUG(x...) do { } while(0)
  75#endif
  76
  77extern unsigned long _get_SP(void);
  78
  79#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
  80/*
  81 * Are we running in "Suspend disabled" mode? If so we have to block any
  82 * sigreturn that would get us into suspended state, and we also warn in some
  83 * other paths that we should never reach with suspend disabled.
  84 */
  85bool tm_suspend_disabled __ro_after_init = false;
  86
  87static void check_if_tm_restore_required(struct task_struct *tsk)
  88{
  89	/*
  90	 * If we are saving the current thread's registers, and the
  91	 * thread is in a transactional state, set the TIF_RESTORE_TM
  92	 * bit so that we know to restore the registers before
  93	 * returning to userspace.
  94	 */
  95	if (tsk == current && tsk->thread.regs &&
  96	    MSR_TM_ACTIVE(tsk->thread.regs->msr) &&
  97	    !test_thread_flag(TIF_RESTORE_TM)) {
  98		tsk->thread.ckpt_regs.msr = tsk->thread.regs->msr;
  99		set_thread_flag(TIF_RESTORE_TM);
 100	}
 101}
 102
 103#else
 104static inline void check_if_tm_restore_required(struct task_struct *tsk) { }
 105#endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
 106
 107bool strict_msr_control;
 108EXPORT_SYMBOL(strict_msr_control);
 109
 110static int __init enable_strict_msr_control(char *str)
 111{
 112	strict_msr_control = true;
 113	pr_info("Enabling strict facility control\n");
 114
 115	return 0;
 116}
 117early_param("ppc_strict_facility_enable", enable_strict_msr_control);
 118
 119/* notrace because it's called by restore_math */
 120unsigned long notrace msr_check_and_set(unsigned long bits)
 121{
 122	unsigned long oldmsr = mfmsr();
 123	unsigned long newmsr;
 124
 125	newmsr = oldmsr | bits;
 126
 127#ifdef CONFIG_VSX
 128	if (cpu_has_feature(CPU_FTR_VSX) && (bits & MSR_FP))
 129		newmsr |= MSR_VSX;
 130#endif
 131
 132	if (oldmsr != newmsr)
 133		mtmsr_isync(newmsr);
 134
 135	return newmsr;
 136}
 137EXPORT_SYMBOL_GPL(msr_check_and_set);
 138
 139/* notrace because it's called by restore_math */
 140void notrace __msr_check_and_clear(unsigned long bits)
 141{
 142	unsigned long oldmsr = mfmsr();
 143	unsigned long newmsr;
 144
 145	newmsr = oldmsr & ~bits;
 146
 147#ifdef CONFIG_VSX
 148	if (cpu_has_feature(CPU_FTR_VSX) && (bits & MSR_FP))
 149		newmsr &= ~MSR_VSX;
 150#endif
 151
 152	if (oldmsr != newmsr)
 153		mtmsr_isync(newmsr);
 154}
 155EXPORT_SYMBOL(__msr_check_and_clear);
 156
 157#ifdef CONFIG_PPC_FPU
 158static void __giveup_fpu(struct task_struct *tsk)
 159{
 160	unsigned long msr;
 161
 162	save_fpu(tsk);
 163	msr = tsk->thread.regs->msr;
 164	msr &= ~(MSR_FP|MSR_FE0|MSR_FE1);
 165#ifdef CONFIG_VSX
 166	if (cpu_has_feature(CPU_FTR_VSX))
 167		msr &= ~MSR_VSX;
 168#endif
 169	tsk->thread.regs->msr = msr;
 170}
 171
 172void giveup_fpu(struct task_struct *tsk)
 173{
 174	check_if_tm_restore_required(tsk);
 175
 176	msr_check_and_set(MSR_FP);
 177	__giveup_fpu(tsk);
 178	msr_check_and_clear(MSR_FP);
 179}
 180EXPORT_SYMBOL(giveup_fpu);
 181
 182/*
 183 * Make sure the floating-point register state in the
 184 * the thread_struct is up to date for task tsk.
 185 */
 186void flush_fp_to_thread(struct task_struct *tsk)
 187{
 188	if (tsk->thread.regs) {
 189		/*
 190		 * We need to disable preemption here because if we didn't,
 191		 * another process could get scheduled after the regs->msr
 192		 * test but before we have finished saving the FP registers
 193		 * to the thread_struct.  That process could take over the
 194		 * FPU, and then when we get scheduled again we would store
 195		 * bogus values for the remaining FP registers.
 196		 */
 197		preempt_disable();
 198		if (tsk->thread.regs->msr & MSR_FP) {
 199			/*
 200			 * This should only ever be called for current or
 201			 * for a stopped child process.  Since we save away
 202			 * the FP register state on context switch,
 203			 * there is something wrong if a stopped child appears
 204			 * to still have its FP state in the CPU registers.
 205			 */
 206			BUG_ON(tsk != current);
 207			giveup_fpu(tsk);
 208		}
 209		preempt_enable();
 210	}
 211}
 212EXPORT_SYMBOL_GPL(flush_fp_to_thread);
 213
 214void enable_kernel_fp(void)
 215{
 216	unsigned long cpumsr;
 217
 218	WARN_ON(preemptible());
 219
 220	cpumsr = msr_check_and_set(MSR_FP);
 221
 222	if (current->thread.regs && (current->thread.regs->msr & MSR_FP)) {
 223		check_if_tm_restore_required(current);
 224		/*
 225		 * If a thread has already been reclaimed then the
 226		 * checkpointed registers are on the CPU but have definitely
 227		 * been saved by the reclaim code. Don't need to and *cannot*
 228		 * giveup as this would save  to the 'live' structure not the
 229		 * checkpointed structure.
 230		 */
 231		if (!MSR_TM_ACTIVE(cpumsr) &&
 232		     MSR_TM_ACTIVE(current->thread.regs->msr))
 233			return;
 234		__giveup_fpu(current);
 235	}
 236}
 237EXPORT_SYMBOL(enable_kernel_fp);
 
 
 
 
 
 
 
 
 
 
 
 
 238#endif /* CONFIG_PPC_FPU */
 239
 240#ifdef CONFIG_ALTIVEC
 
 
 241static void __giveup_altivec(struct task_struct *tsk)
 242{
 243	unsigned long msr;
 244
 245	save_altivec(tsk);
 246	msr = tsk->thread.regs->msr;
 247	msr &= ~MSR_VEC;
 248#ifdef CONFIG_VSX
 249	if (cpu_has_feature(CPU_FTR_VSX))
 250		msr &= ~MSR_VSX;
 251#endif
 252	tsk->thread.regs->msr = msr;
 253}
 254
 255void giveup_altivec(struct task_struct *tsk)
 256{
 257	check_if_tm_restore_required(tsk);
 258
 259	msr_check_and_set(MSR_VEC);
 260	__giveup_altivec(tsk);
 261	msr_check_and_clear(MSR_VEC);
 262}
 263EXPORT_SYMBOL(giveup_altivec);
 264
 265void enable_kernel_altivec(void)
 266{
 267	unsigned long cpumsr;
 268
 269	WARN_ON(preemptible());
 270
 271	cpumsr = msr_check_and_set(MSR_VEC);
 272
 273	if (current->thread.regs && (current->thread.regs->msr & MSR_VEC)) {
 274		check_if_tm_restore_required(current);
 275		/*
 276		 * If a thread has already been reclaimed then the
 277		 * checkpointed registers are on the CPU but have definitely
 278		 * been saved by the reclaim code. Don't need to and *cannot*
 279		 * giveup as this would save  to the 'live' structure not the
 280		 * checkpointed structure.
 281		 */
 282		if (!MSR_TM_ACTIVE(cpumsr) &&
 283		     MSR_TM_ACTIVE(current->thread.regs->msr))
 284			return;
 285		__giveup_altivec(current);
 286	}
 287}
 288EXPORT_SYMBOL(enable_kernel_altivec);
 289
 290/*
 291 * Make sure the VMX/Altivec register state in the
 292 * the thread_struct is up to date for task tsk.
 293 */
 294void flush_altivec_to_thread(struct task_struct *tsk)
 295{
 296	if (tsk->thread.regs) {
 297		preempt_disable();
 298		if (tsk->thread.regs->msr & MSR_VEC) {
 299			BUG_ON(tsk != current);
 300			giveup_altivec(tsk);
 301		}
 302		preempt_enable();
 303	}
 304}
 305EXPORT_SYMBOL_GPL(flush_altivec_to_thread);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 306#endif /* CONFIG_ALTIVEC */
 307
 308#ifdef CONFIG_VSX
 309static void __giveup_vsx(struct task_struct *tsk)
 310{
 311	unsigned long msr = tsk->thread.regs->msr;
 312
 313	/*
 314	 * We should never be ssetting MSR_VSX without also setting
 315	 * MSR_FP and MSR_VEC
 316	 */
 317	WARN_ON((msr & MSR_VSX) && !((msr & MSR_FP) && (msr & MSR_VEC)));
 318
 319	/* __giveup_fpu will clear MSR_VSX */
 320	if (msr & MSR_FP)
 321		__giveup_fpu(tsk);
 322	if (msr & MSR_VEC)
 323		__giveup_altivec(tsk);
 324}
 325
 326static void giveup_vsx(struct task_struct *tsk)
 327{
 328	check_if_tm_restore_required(tsk);
 329
 330	msr_check_and_set(MSR_FP|MSR_VEC|MSR_VSX);
 331	__giveup_vsx(tsk);
 332	msr_check_and_clear(MSR_FP|MSR_VEC|MSR_VSX);
 333}
 334
 335void enable_kernel_vsx(void)
 336{
 337	unsigned long cpumsr;
 338
 339	WARN_ON(preemptible());
 340
 341	cpumsr = msr_check_and_set(MSR_FP|MSR_VEC|MSR_VSX);
 342
 343	if (current->thread.regs &&
 344	    (current->thread.regs->msr & (MSR_VSX|MSR_VEC|MSR_FP))) {
 345		check_if_tm_restore_required(current);
 346		/*
 347		 * If a thread has already been reclaimed then the
 348		 * checkpointed registers are on the CPU but have definitely
 349		 * been saved by the reclaim code. Don't need to and *cannot*
 350		 * giveup as this would save  to the 'live' structure not the
 351		 * checkpointed structure.
 352		 */
 353		if (!MSR_TM_ACTIVE(cpumsr) &&
 354		     MSR_TM_ACTIVE(current->thread.regs->msr))
 355			return;
 356		__giveup_vsx(current);
 357	}
 358}
 359EXPORT_SYMBOL(enable_kernel_vsx);
 360
 361void flush_vsx_to_thread(struct task_struct *tsk)
 362{
 363	if (tsk->thread.regs) {
 364		preempt_disable();
 365		if (tsk->thread.regs->msr & (MSR_VSX|MSR_VEC|MSR_FP)) {
 366			BUG_ON(tsk != current);
 367			giveup_vsx(tsk);
 368		}
 369		preempt_enable();
 370	}
 371}
 372EXPORT_SYMBOL_GPL(flush_vsx_to_thread);
 
 
 
 
 
 
 
 
 
 
 
 
 373#endif /* CONFIG_VSX */
 374
 375#ifdef CONFIG_SPE
 376void giveup_spe(struct task_struct *tsk)
 377{
 378	check_if_tm_restore_required(tsk);
 379
 380	msr_check_and_set(MSR_SPE);
 381	__giveup_spe(tsk);
 382	msr_check_and_clear(MSR_SPE);
 383}
 384EXPORT_SYMBOL(giveup_spe);
 385
 386void enable_kernel_spe(void)
 387{
 388	WARN_ON(preemptible());
 389
 390	msr_check_and_set(MSR_SPE);
 391
 392	if (current->thread.regs && (current->thread.regs->msr & MSR_SPE)) {
 393		check_if_tm_restore_required(current);
 394		__giveup_spe(current);
 395	}
 396}
 397EXPORT_SYMBOL(enable_kernel_spe);
 398
 399void flush_spe_to_thread(struct task_struct *tsk)
 400{
 401	if (tsk->thread.regs) {
 402		preempt_disable();
 403		if (tsk->thread.regs->msr & MSR_SPE) {
 404			BUG_ON(tsk != current);
 405			tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
 406			giveup_spe(tsk);
 407		}
 408		preempt_enable();
 409	}
 410}
 411#endif /* CONFIG_SPE */
 412
 413static unsigned long msr_all_available;
 414
 415static int __init init_msr_all_available(void)
 416{
 417#ifdef CONFIG_PPC_FPU
 418	msr_all_available |= MSR_FP;
 419#endif
 420#ifdef CONFIG_ALTIVEC
 421	if (cpu_has_feature(CPU_FTR_ALTIVEC))
 422		msr_all_available |= MSR_VEC;
 423#endif
 424#ifdef CONFIG_VSX
 425	if (cpu_has_feature(CPU_FTR_VSX))
 426		msr_all_available |= MSR_VSX;
 427#endif
 428#ifdef CONFIG_SPE
 429	if (cpu_has_feature(CPU_FTR_SPE))
 430		msr_all_available |= MSR_SPE;
 431#endif
 432
 433	return 0;
 434}
 435early_initcall(init_msr_all_available);
 436
 437void giveup_all(struct task_struct *tsk)
 438{
 439	unsigned long usermsr;
 440
 441	if (!tsk->thread.regs)
 442		return;
 443
 444	check_if_tm_restore_required(tsk);
 445
 446	usermsr = tsk->thread.regs->msr;
 447
 448	if ((usermsr & msr_all_available) == 0)
 449		return;
 450
 451	msr_check_and_set(msr_all_available);
 452
 453	WARN_ON((usermsr & MSR_VSX) && !((usermsr & MSR_FP) && (usermsr & MSR_VEC)));
 454
 455#ifdef CONFIG_PPC_FPU
 456	if (usermsr & MSR_FP)
 457		__giveup_fpu(tsk);
 458#endif
 459#ifdef CONFIG_ALTIVEC
 460	if (usermsr & MSR_VEC)
 461		__giveup_altivec(tsk);
 462#endif
 463#ifdef CONFIG_SPE
 464	if (usermsr & MSR_SPE)
 465		__giveup_spe(tsk);
 466#endif
 467
 468	msr_check_and_clear(msr_all_available);
 469}
 470EXPORT_SYMBOL(giveup_all);
 471
 472#ifdef CONFIG_PPC_BOOK3S_64
 473#ifdef CONFIG_PPC_FPU
 474static bool should_restore_fp(void)
 475{
 476	if (current->thread.load_fp) {
 477		current->thread.load_fp++;
 478		return true;
 479	}
 480	return false;
 481}
 482
 483static void do_restore_fp(void)
 484{
 485	load_fp_state(&current->thread.fp_state);
 486}
 487#else
 488static bool should_restore_fp(void) { return false; }
 489static void do_restore_fp(void) { }
 490#endif /* CONFIG_PPC_FPU */
 491
 492#ifdef CONFIG_ALTIVEC
 493static bool should_restore_altivec(void)
 494{
 495	if (cpu_has_feature(CPU_FTR_ALTIVEC) && (current->thread.load_vec)) {
 496		current->thread.load_vec++;
 497		return true;
 498	}
 499	return false;
 500}
 501
 502static void do_restore_altivec(void)
 503{
 504	load_vr_state(&current->thread.vr_state);
 505	current->thread.used_vr = 1;
 506}
 507#else
 508static bool should_restore_altivec(void) { return false; }
 509static void do_restore_altivec(void) { }
 510#endif /* CONFIG_ALTIVEC */
 511
 512#ifdef CONFIG_VSX
 513static bool should_restore_vsx(void)
 514{
 515	if (cpu_has_feature(CPU_FTR_VSX))
 516		return true;
 517	return false;
 518}
 519static void do_restore_vsx(void)
 520{
 521	current->thread.used_vsr = 1;
 522}
 523#else
 524static bool should_restore_vsx(void) { return false; }
 525static void do_restore_vsx(void) { }
 526#endif /* CONFIG_VSX */
 527
 528/*
 529 * The exception exit path calls restore_math() with interrupts hard disabled
 530 * but the soft irq state not "reconciled". ftrace code that calls
 531 * local_irq_save/restore causes warnings.
 532 *
 533 * Rather than complicate the exit path, just don't trace restore_math. This
 534 * could be done by having ftrace entry code check for this un-reconciled
 535 * condition where MSR[EE]=0 and PACA_IRQ_HARD_DIS is not set, and
 536 * temporarily fix it up for the duration of the ftrace call.
 537 */
 538void notrace restore_math(struct pt_regs *regs)
 539{
 540	unsigned long msr;
 541	unsigned long new_msr = 0;
 
 
 
 542
 543	msr = regs->msr;
 
 544
 545	/*
 546	 * new_msr tracks the facilities that are to be restored. Only reload
 547	 * if the bit is not set in the user MSR (if it is set, the registers
 548	 * are live for the user thread).
 549	 */
 550	if ((!(msr & MSR_FP)) && should_restore_fp())
 551		new_msr |= MSR_FP;
 552
 553	if ((!(msr & MSR_VEC)) && should_restore_altivec())
 554		new_msr |= MSR_VEC;
 555
 556	if ((!(msr & MSR_VSX)) && should_restore_vsx()) {
 557		if (((msr | new_msr) & (MSR_FP | MSR_VEC)) == (MSR_FP | MSR_VEC))
 558			new_msr |= MSR_VSX;
 559	}
 560
 561	if (new_msr) {
 562		unsigned long fpexc_mode = 0;
 563
 564		msr_check_and_set(new_msr);
 565
 566		if (new_msr & MSR_FP) {
 567			do_restore_fp();
 568
 569			// This also covers VSX, because VSX implies FP
 570			fpexc_mode = current->thread.fpexc_mode;
 571		}
 572
 573		if (new_msr & MSR_VEC)
 574			do_restore_altivec();
 575
 576		if (new_msr & MSR_VSX)
 577			do_restore_vsx();
 578
 579		msr_check_and_clear(new_msr);
 580
 581		regs->msr |= new_msr | fpexc_mode;
 582	}
 583}
 584#endif
 585
 586static void save_all(struct task_struct *tsk)
 587{
 588	unsigned long usermsr;
 589
 590	if (!tsk->thread.regs)
 591		return;
 592
 593	usermsr = tsk->thread.regs->msr;
 594
 595	if ((usermsr & msr_all_available) == 0)
 596		return;
 597
 598	msr_check_and_set(msr_all_available);
 599
 600	WARN_ON((usermsr & MSR_VSX) && !((usermsr & MSR_FP) && (usermsr & MSR_VEC)));
 601
 602	if (usermsr & MSR_FP)
 603		save_fpu(tsk);
 604
 605	if (usermsr & MSR_VEC)
 606		save_altivec(tsk);
 607
 608	if (usermsr & MSR_SPE)
 609		__giveup_spe(tsk);
 610
 611	msr_check_and_clear(msr_all_available);
 612	thread_pkey_regs_save(&tsk->thread);
 613}
 614
 615void flush_all_to_thread(struct task_struct *tsk)
 616{
 617	if (tsk->thread.regs) {
 618		preempt_disable();
 619		BUG_ON(tsk != current);
 620#ifdef CONFIG_SPE
 621		if (tsk->thread.regs->msr & MSR_SPE)
 622			tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
 623#endif
 624		save_all(tsk);
 625
 626		preempt_enable();
 627	}
 628}
 629EXPORT_SYMBOL(flush_all_to_thread);
 630
 631#ifdef CONFIG_PPC_ADV_DEBUG_REGS
 632void do_send_trap(struct pt_regs *regs, unsigned long address,
 633		  unsigned long error_code, int breakpt)
 634{
 635	current->thread.trap_nr = TRAP_HWBKPT;
 636	if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
 637			11, SIGSEGV) == NOTIFY_STOP)
 638		return;
 639
 640	/* Deliver the signal to userspace */
 641	force_sig_ptrace_errno_trap(breakpt, /* breakpoint or watchpoint id */
 642				    (void __user *)address);
 643}
 644#else	/* !CONFIG_PPC_ADV_DEBUG_REGS */
 645void do_break (struct pt_regs *regs, unsigned long address,
 646		    unsigned long error_code)
 647{
 648	current->thread.trap_nr = TRAP_HWBKPT;
 649	if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
 650			11, SIGSEGV) == NOTIFY_STOP)
 651		return;
 652
 653	if (debugger_break_match(regs))
 654		return;
 655
 
 
 
 656	/* Deliver the signal to userspace */
 657	force_sig_fault(SIGTRAP, TRAP_HWBKPT, (void __user *)address);
 658}
 659#endif	/* CONFIG_PPC_ADV_DEBUG_REGS */
 660
 661static DEFINE_PER_CPU(struct arch_hw_breakpoint, current_brk[HBP_NUM_MAX]);
 662
 663#ifdef CONFIG_PPC_ADV_DEBUG_REGS
 664/*
 665 * Set the debug registers back to their default "safe" values.
 666 */
 667static void set_debug_reg_defaults(struct thread_struct *thread)
 668{
 669	thread->debug.iac1 = thread->debug.iac2 = 0;
 670#if CONFIG_PPC_ADV_DEBUG_IACS > 2
 671	thread->debug.iac3 = thread->debug.iac4 = 0;
 672#endif
 673	thread->debug.dac1 = thread->debug.dac2 = 0;
 674#if CONFIG_PPC_ADV_DEBUG_DVCS > 0
 675	thread->debug.dvc1 = thread->debug.dvc2 = 0;
 676#endif
 677	thread->debug.dbcr0 = 0;
 678#ifdef CONFIG_BOOKE
 679	/*
 680	 * Force User/Supervisor bits to b11 (user-only MSR[PR]=1)
 681	 */
 682	thread->debug.dbcr1 = DBCR1_IAC1US | DBCR1_IAC2US |
 683			DBCR1_IAC3US | DBCR1_IAC4US;
 684	/*
 685	 * Force Data Address Compare User/Supervisor bits to be User-only
 686	 * (0b11 MSR[PR]=1) and set all other bits in DBCR2 register to be 0.
 687	 */
 688	thread->debug.dbcr2 = DBCR2_DAC1US | DBCR2_DAC2US;
 689#else
 690	thread->debug.dbcr1 = 0;
 691#endif
 692}
 693
 694static void prime_debug_regs(struct debug_reg *debug)
 695{
 696	/*
 697	 * We could have inherited MSR_DE from userspace, since
 698	 * it doesn't get cleared on exception entry.  Make sure
 699	 * MSR_DE is clear before we enable any debug events.
 700	 */
 701	mtmsr(mfmsr() & ~MSR_DE);
 702
 703	mtspr(SPRN_IAC1, debug->iac1);
 704	mtspr(SPRN_IAC2, debug->iac2);
 705#if CONFIG_PPC_ADV_DEBUG_IACS > 2
 706	mtspr(SPRN_IAC3, debug->iac3);
 707	mtspr(SPRN_IAC4, debug->iac4);
 708#endif
 709	mtspr(SPRN_DAC1, debug->dac1);
 710	mtspr(SPRN_DAC2, debug->dac2);
 711#if CONFIG_PPC_ADV_DEBUG_DVCS > 0
 712	mtspr(SPRN_DVC1, debug->dvc1);
 713	mtspr(SPRN_DVC2, debug->dvc2);
 714#endif
 715	mtspr(SPRN_DBCR0, debug->dbcr0);
 716	mtspr(SPRN_DBCR1, debug->dbcr1);
 717#ifdef CONFIG_BOOKE
 718	mtspr(SPRN_DBCR2, debug->dbcr2);
 719#endif
 720}
 721/*
 722 * Unless neither the old or new thread are making use of the
 723 * debug registers, set the debug registers from the values
 724 * stored in the new thread.
 725 */
 726void switch_booke_debug_regs(struct debug_reg *new_debug)
 727{
 728	if ((current->thread.debug.dbcr0 & DBCR0_IDM)
 729		|| (new_debug->dbcr0 & DBCR0_IDM))
 730			prime_debug_regs(new_debug);
 731}
 732EXPORT_SYMBOL_GPL(switch_booke_debug_regs);
 733#else	/* !CONFIG_PPC_ADV_DEBUG_REGS */
 734#ifndef CONFIG_HAVE_HW_BREAKPOINT
 735static void set_breakpoint(int i, struct arch_hw_breakpoint *brk)
 736{
 737	preempt_disable();
 738	__set_breakpoint(i, brk);
 739	preempt_enable();
 740}
 741
 742static void set_debug_reg_defaults(struct thread_struct *thread)
 743{
 744	int i;
 745	struct arch_hw_breakpoint null_brk = {0};
 746
 747	for (i = 0; i < nr_wp_slots(); i++) {
 748		thread->hw_brk[i] = null_brk;
 749		if (ppc_breakpoint_available())
 750			set_breakpoint(i, &thread->hw_brk[i]);
 751	}
 752}
 753
 754static inline bool hw_brk_match(struct arch_hw_breakpoint *a,
 755				struct arch_hw_breakpoint *b)
 756{
 757	if (a->address != b->address)
 758		return false;
 759	if (a->type != b->type)
 760		return false;
 761	if (a->len != b->len)
 762		return false;
 763	/* no need to check hw_len. it's calculated from address and len */
 764	return true;
 765}
 766
 767static void switch_hw_breakpoint(struct task_struct *new)
 768{
 769	int i;
 770
 771	for (i = 0; i < nr_wp_slots(); i++) {
 772		if (likely(hw_brk_match(this_cpu_ptr(&current_brk[i]),
 773					&new->thread.hw_brk[i])))
 774			continue;
 775
 776		__set_breakpoint(i, &new->thread.hw_brk[i]);
 777	}
 778}
 779#endif /* !CONFIG_HAVE_HW_BREAKPOINT */
 780#endif	/* CONFIG_PPC_ADV_DEBUG_REGS */
 781
 782#ifdef CONFIG_PPC_ADV_DEBUG_REGS
 783static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
 784{
 785	mtspr(SPRN_DAC1, dabr);
 786#ifdef CONFIG_PPC_47x
 787	isync();
 788#endif
 789	return 0;
 790}
 791#elif defined(CONFIG_PPC_BOOK3S)
 792static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
 793{
 794	mtspr(SPRN_DABR, dabr);
 795	if (cpu_has_feature(CPU_FTR_DABRX))
 796		mtspr(SPRN_DABRX, dabrx);
 797	return 0;
 798}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 799#else
 800static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
 801{
 802	return -EINVAL;
 803}
 804#endif
 805
 806static inline int set_dabr(struct arch_hw_breakpoint *brk)
 807{
 808	unsigned long dabr, dabrx;
 809
 810	dabr = brk->address | (brk->type & HW_BRK_TYPE_DABR);
 811	dabrx = ((brk->type >> 3) & 0x7);
 812
 813	if (ppc_md.set_dabr)
 814		return ppc_md.set_dabr(dabr, dabrx);
 815
 816	return __set_dabr(dabr, dabrx);
 817}
 818
 819static inline int set_breakpoint_8xx(struct arch_hw_breakpoint *brk)
 820{
 821	unsigned long lctrl1 = LCTRL1_CTE_GT | LCTRL1_CTF_LT | LCTRL1_CRWE_RW |
 822			       LCTRL1_CRWF_RW;
 823	unsigned long lctrl2 = LCTRL2_LW0EN | LCTRL2_LW0LADC | LCTRL2_SLW0EN;
 824	unsigned long start_addr = ALIGN_DOWN(brk->address, HW_BREAKPOINT_SIZE);
 825	unsigned long end_addr = ALIGN(brk->address + brk->len, HW_BREAKPOINT_SIZE);
 826
 827	if (start_addr == 0)
 828		lctrl2 |= LCTRL2_LW0LA_F;
 829	else if (end_addr == 0)
 830		lctrl2 |= LCTRL2_LW0LA_E;
 831	else
 832		lctrl2 |= LCTRL2_LW0LA_EandF;
 833
 834	mtspr(SPRN_LCTRL2, 0);
 835
 836	if ((brk->type & HW_BRK_TYPE_RDWR) == 0)
 837		return 0;
 838
 839	if ((brk->type & HW_BRK_TYPE_RDWR) == HW_BRK_TYPE_READ)
 840		lctrl1 |= LCTRL1_CRWE_RO | LCTRL1_CRWF_RO;
 841	if ((brk->type & HW_BRK_TYPE_RDWR) == HW_BRK_TYPE_WRITE)
 842		lctrl1 |= LCTRL1_CRWE_WO | LCTRL1_CRWF_WO;
 843
 844	mtspr(SPRN_CMPE, start_addr - 1);
 845	mtspr(SPRN_CMPF, end_addr);
 846	mtspr(SPRN_LCTRL1, lctrl1);
 847	mtspr(SPRN_LCTRL2, lctrl2);
 848
 849	return 0;
 850}
 851
 852void __set_breakpoint(int nr, struct arch_hw_breakpoint *brk)
 853{
 854	memcpy(this_cpu_ptr(&current_brk[nr]), brk, sizeof(*brk));
 855
 856	if (dawr_enabled())
 857		// Power8 or later
 858		set_dawr(nr, brk);
 859	else if (IS_ENABLED(CONFIG_PPC_8xx))
 860		set_breakpoint_8xx(brk);
 861	else if (!cpu_has_feature(CPU_FTR_ARCH_207S))
 862		// Power7 or earlier
 863		set_dabr(brk);
 864	else
 865		// Shouldn't happen due to higher level checks
 866		WARN_ON_ONCE(1);
 867}
 868
 869/* Check if we have DAWR or DABR hardware */
 870bool ppc_breakpoint_available(void)
 871{
 872	if (dawr_enabled())
 873		return true; /* POWER8 DAWR or POWER9 forced DAWR */
 874	if (cpu_has_feature(CPU_FTR_ARCH_207S))
 875		return false; /* POWER9 with DAWR disabled */
 876	/* DABR: Everything but POWER8 and POWER9 */
 877	return true;
 878}
 879EXPORT_SYMBOL_GPL(ppc_breakpoint_available);
 880
 
 
 
 
 
 
 
 
 
 
 
 
 881#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
 882
 883static inline bool tm_enabled(struct task_struct *tsk)
 884{
 885	return tsk && tsk->thread.regs && (tsk->thread.regs->msr & MSR_TM);
 886}
 887
 888static void tm_reclaim_thread(struct thread_struct *thr, uint8_t cause)
 889{
 890	/*
 891	 * Use the current MSR TM suspended bit to track if we have
 892	 * checkpointed state outstanding.
 893	 * On signal delivery, we'd normally reclaim the checkpointed
 894	 * state to obtain stack pointer (see:get_tm_stackpointer()).
 895	 * This will then directly return to userspace without going
 896	 * through __switch_to(). However, if the stack frame is bad,
 897	 * we need to exit this thread which calls __switch_to() which
 898	 * will again attempt to reclaim the already saved tm state.
 899	 * Hence we need to check that we've not already reclaimed
 900	 * this state.
 901	 * We do this using the current MSR, rather tracking it in
 902	 * some specific thread_struct bit, as it has the additional
 903	 * benefit of checking for a potential TM bad thing exception.
 904	 */
 905	if (!MSR_TM_SUSPENDED(mfmsr()))
 906		return;
 907
 908	giveup_all(container_of(thr, struct task_struct, thread));
 909
 910	tm_reclaim(thr, cause);
 911
 912	/*
 913	 * If we are in a transaction and FP is off then we can't have
 914	 * used FP inside that transaction. Hence the checkpointed
 915	 * state is the same as the live state. We need to copy the
 916	 * live state to the checkpointed state so that when the
 917	 * transaction is restored, the checkpointed state is correct
 918	 * and the aborted transaction sees the correct state. We use
 919	 * ckpt_regs.msr here as that's what tm_reclaim will use to
 920	 * determine if it's going to write the checkpointed state or
 921	 * not. So either this will write the checkpointed registers,
 922	 * or reclaim will. Similarly for VMX.
 923	 */
 924	if ((thr->ckpt_regs.msr & MSR_FP) == 0)
 925		memcpy(&thr->ckfp_state, &thr->fp_state,
 926		       sizeof(struct thread_fp_state));
 927	if ((thr->ckpt_regs.msr & MSR_VEC) == 0)
 928		memcpy(&thr->ckvr_state, &thr->vr_state,
 929		       sizeof(struct thread_vr_state));
 930}
 931
 932void tm_reclaim_current(uint8_t cause)
 933{
 934	tm_enable();
 935	tm_reclaim_thread(&current->thread, cause);
 936}
 937
 938static inline void tm_reclaim_task(struct task_struct *tsk)
 939{
 940	/* We have to work out if we're switching from/to a task that's in the
 941	 * middle of a transaction.
 942	 *
 943	 * In switching we need to maintain a 2nd register state as
 944	 * oldtask->thread.ckpt_regs.  We tm_reclaim(oldproc); this saves the
 945	 * checkpointed (tbegin) state in ckpt_regs, ckfp_state and
 946	 * ckvr_state
 947	 *
 948	 * We also context switch (save) TFHAR/TEXASR/TFIAR in here.
 949	 */
 950	struct thread_struct *thr = &tsk->thread;
 951
 952	if (!thr->regs)
 953		return;
 954
 955	if (!MSR_TM_ACTIVE(thr->regs->msr))
 956		goto out_and_saveregs;
 957
 958	WARN_ON(tm_suspend_disabled);
 959
 960	TM_DEBUG("--- tm_reclaim on pid %d (NIP=%lx, "
 961		 "ccr=%lx, msr=%lx, trap=%lx)\n",
 962		 tsk->pid, thr->regs->nip,
 963		 thr->regs->ccr, thr->regs->msr,
 964		 thr->regs->trap);
 965
 966	tm_reclaim_thread(thr, TM_CAUSE_RESCHED);
 967
 968	TM_DEBUG("--- tm_reclaim on pid %d complete\n",
 969		 tsk->pid);
 970
 971out_and_saveregs:
 972	/* Always save the regs here, even if a transaction's not active.
 973	 * This context-switches a thread's TM info SPRs.  We do it here to
 974	 * be consistent with the restore path (in recheckpoint) which
 975	 * cannot happen later in _switch().
 976	 */
 977	tm_save_sprs(thr);
 978}
 979
 980extern void __tm_recheckpoint(struct thread_struct *thread);
 981
 982void tm_recheckpoint(struct thread_struct *thread)
 983{
 984	unsigned long flags;
 985
 986	if (!(thread->regs->msr & MSR_TM))
 987		return;
 988
 989	/* We really can't be interrupted here as the TEXASR registers can't
 990	 * change and later in the trecheckpoint code, we have a userspace R1.
 991	 * So let's hard disable over this region.
 992	 */
 993	local_irq_save(flags);
 994	hard_irq_disable();
 995
 996	/* The TM SPRs are restored here, so that TEXASR.FS can be set
 997	 * before the trecheckpoint and no explosion occurs.
 998	 */
 999	tm_restore_sprs(thread);
1000
1001	__tm_recheckpoint(thread);
1002
1003	local_irq_restore(flags);
1004}
1005
1006static inline void tm_recheckpoint_new_task(struct task_struct *new)
1007{
1008	if (!cpu_has_feature(CPU_FTR_TM))
1009		return;
1010
1011	/* Recheckpoint the registers of the thread we're about to switch to.
1012	 *
1013	 * If the task was using FP, we non-lazily reload both the original and
1014	 * the speculative FP register states.  This is because the kernel
1015	 * doesn't see if/when a TM rollback occurs, so if we take an FP
1016	 * unavailable later, we are unable to determine which set of FP regs
1017	 * need to be restored.
1018	 */
1019	if (!tm_enabled(new))
1020		return;
1021
1022	if (!MSR_TM_ACTIVE(new->thread.regs->msr)){
1023		tm_restore_sprs(&new->thread);
1024		return;
1025	}
1026	/* Recheckpoint to restore original checkpointed register state. */
1027	TM_DEBUG("*** tm_recheckpoint of pid %d (new->msr 0x%lx)\n",
1028		 new->pid, new->thread.regs->msr);
1029
1030	tm_recheckpoint(&new->thread);
1031
1032	/*
1033	 * The checkpointed state has been restored but the live state has
1034	 * not, ensure all the math functionality is turned off to trigger
1035	 * restore_math() to reload.
1036	 */
1037	new->thread.regs->msr &= ~(MSR_FP | MSR_VEC | MSR_VSX);
1038
1039	TM_DEBUG("*** tm_recheckpoint of pid %d complete "
1040		 "(kernel msr 0x%lx)\n",
1041		 new->pid, mfmsr());
1042}
1043
1044static inline void __switch_to_tm(struct task_struct *prev,
1045		struct task_struct *new)
1046{
1047	if (cpu_has_feature(CPU_FTR_TM)) {
1048		if (tm_enabled(prev) || tm_enabled(new))
1049			tm_enable();
1050
1051		if (tm_enabled(prev)) {
1052			prev->thread.load_tm++;
1053			tm_reclaim_task(prev);
1054			if (!MSR_TM_ACTIVE(prev->thread.regs->msr) && prev->thread.load_tm == 0)
1055				prev->thread.regs->msr &= ~MSR_TM;
1056		}
1057
1058		tm_recheckpoint_new_task(new);
1059	}
1060}
1061
1062/*
1063 * This is called if we are on the way out to userspace and the
1064 * TIF_RESTORE_TM flag is set.  It checks if we need to reload
1065 * FP and/or vector state and does so if necessary.
1066 * If userspace is inside a transaction (whether active or
1067 * suspended) and FP/VMX/VSX instructions have ever been enabled
1068 * inside that transaction, then we have to keep them enabled
1069 * and keep the FP/VMX/VSX state loaded while ever the transaction
1070 * continues.  The reason is that if we didn't, and subsequently
1071 * got a FP/VMX/VSX unavailable interrupt inside a transaction,
1072 * we don't know whether it's the same transaction, and thus we
1073 * don't know which of the checkpointed state and the transactional
1074 * state to use.
1075 */
1076void restore_tm_state(struct pt_regs *regs)
1077{
1078	unsigned long msr_diff;
1079
1080	/*
1081	 * This is the only moment we should clear TIF_RESTORE_TM as
1082	 * it is here that ckpt_regs.msr and pt_regs.msr become the same
1083	 * again, anything else could lead to an incorrect ckpt_msr being
1084	 * saved and therefore incorrect signal contexts.
1085	 */
1086	clear_thread_flag(TIF_RESTORE_TM);
1087	if (!MSR_TM_ACTIVE(regs->msr))
1088		return;
1089
1090	msr_diff = current->thread.ckpt_regs.msr & ~regs->msr;
1091	msr_diff &= MSR_FP | MSR_VEC | MSR_VSX;
1092
1093	/* Ensure that restore_math() will restore */
1094	if (msr_diff & MSR_FP)
1095		current->thread.load_fp = 1;
1096#ifdef CONFIG_ALTIVEC
1097	if (cpu_has_feature(CPU_FTR_ALTIVEC) && msr_diff & MSR_VEC)
1098		current->thread.load_vec = 1;
1099#endif
1100	restore_math(regs);
1101
1102	regs->msr |= msr_diff;
1103}
1104
1105#else
1106#define tm_recheckpoint_new_task(new)
1107#define __switch_to_tm(prev, new)
1108#endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1109
1110static inline void save_sprs(struct thread_struct *t)
1111{
1112#ifdef CONFIG_ALTIVEC
1113	if (cpu_has_feature(CPU_FTR_ALTIVEC))
1114		t->vrsave = mfspr(SPRN_VRSAVE);
1115#endif
1116#ifdef CONFIG_PPC_BOOK3S_64
1117	if (cpu_has_feature(CPU_FTR_DSCR))
1118		t->dscr = mfspr(SPRN_DSCR);
1119
1120	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
1121		t->bescr = mfspr(SPRN_BESCR);
1122		t->ebbhr = mfspr(SPRN_EBBHR);
1123		t->ebbrr = mfspr(SPRN_EBBRR);
1124
1125		t->fscr = mfspr(SPRN_FSCR);
1126
1127		/*
1128		 * Note that the TAR is not available for use in the kernel.
1129		 * (To provide this, the TAR should be backed up/restored on
1130		 * exception entry/exit instead, and be in pt_regs.  FIXME,
1131		 * this should be in pt_regs anyway (for debug).)
1132		 */
1133		t->tar = mfspr(SPRN_TAR);
1134	}
1135#endif
1136
1137	thread_pkey_regs_save(t);
1138}
1139
1140static inline void restore_sprs(struct thread_struct *old_thread,
1141				struct thread_struct *new_thread)
1142{
1143#ifdef CONFIG_ALTIVEC
1144	if (cpu_has_feature(CPU_FTR_ALTIVEC) &&
1145	    old_thread->vrsave != new_thread->vrsave)
1146		mtspr(SPRN_VRSAVE, new_thread->vrsave);
1147#endif
1148#ifdef CONFIG_PPC_BOOK3S_64
1149	if (cpu_has_feature(CPU_FTR_DSCR)) {
1150		u64 dscr = get_paca()->dscr_default;
1151		if (new_thread->dscr_inherit)
1152			dscr = new_thread->dscr;
1153
1154		if (old_thread->dscr != dscr)
1155			mtspr(SPRN_DSCR, dscr);
1156	}
1157
1158	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
1159		if (old_thread->bescr != new_thread->bescr)
1160			mtspr(SPRN_BESCR, new_thread->bescr);
1161		if (old_thread->ebbhr != new_thread->ebbhr)
1162			mtspr(SPRN_EBBHR, new_thread->ebbhr);
1163		if (old_thread->ebbrr != new_thread->ebbrr)
1164			mtspr(SPRN_EBBRR, new_thread->ebbrr);
1165
1166		if (old_thread->fscr != new_thread->fscr)
1167			mtspr(SPRN_FSCR, new_thread->fscr);
1168
1169		if (old_thread->tar != new_thread->tar)
1170			mtspr(SPRN_TAR, new_thread->tar);
1171	}
1172
1173	if (cpu_has_feature(CPU_FTR_P9_TIDR) &&
1174	    old_thread->tidr != new_thread->tidr)
1175		mtspr(SPRN_TIDR, new_thread->tidr);
1176#endif
1177
1178	thread_pkey_regs_restore(new_thread, old_thread);
1179}
1180
1181struct task_struct *__switch_to(struct task_struct *prev,
1182	struct task_struct *new)
1183{
1184	struct thread_struct *new_thread, *old_thread;
1185	struct task_struct *last;
1186#ifdef CONFIG_PPC_BOOK3S_64
1187	struct ppc64_tlb_batch *batch;
1188#endif
1189
1190	new_thread = &new->thread;
1191	old_thread = &current->thread;
1192
1193	WARN_ON(!irqs_disabled());
1194
1195#ifdef CONFIG_PPC_BOOK3S_64
1196	batch = this_cpu_ptr(&ppc64_tlb_batch);
1197	if (batch->active) {
1198		current_thread_info()->local_flags |= _TLF_LAZY_MMU;
1199		if (batch->index)
1200			__flush_tlb_pending(batch);
1201		batch->active = 0;
1202	}
1203#endif /* CONFIG_PPC_BOOK3S_64 */
1204
1205#ifdef CONFIG_PPC_ADV_DEBUG_REGS
1206	switch_booke_debug_regs(&new->thread.debug);
1207#else
1208/*
1209 * For PPC_BOOK3S_64, we use the hw-breakpoint interfaces that would
1210 * schedule DABR
1211 */
1212#ifndef CONFIG_HAVE_HW_BREAKPOINT
1213	switch_hw_breakpoint(new);
 
1214#endif /* CONFIG_HAVE_HW_BREAKPOINT */
1215#endif
1216
1217	/*
1218	 * We need to save SPRs before treclaim/trecheckpoint as these will
1219	 * change a number of them.
1220	 */
1221	save_sprs(&prev->thread);
1222
1223	/* Save FPU, Altivec, VSX and SPE state */
1224	giveup_all(prev);
1225
1226	__switch_to_tm(prev, new);
1227
1228	if (!radix_enabled()) {
1229		/*
1230		 * We can't take a PMU exception inside _switch() since there
1231		 * is a window where the kernel stack SLB and the kernel stack
1232		 * are out of sync. Hard disable here.
1233		 */
1234		hard_irq_disable();
1235	}
1236
1237	/*
1238	 * Call restore_sprs() before calling _switch(). If we move it after
1239	 * _switch() then we miss out on calling it for new tasks. The reason
1240	 * for this is we manually create a stack frame for new tasks that
1241	 * directly returns through ret_from_fork() or
1242	 * ret_from_kernel_thread(). See copy_thread() for details.
1243	 */
1244	restore_sprs(old_thread, new_thread);
1245
1246	last = _switch(old_thread, new_thread);
1247
1248#ifdef CONFIG_PPC_BOOK3S_64
1249	if (current_thread_info()->local_flags & _TLF_LAZY_MMU) {
1250		current_thread_info()->local_flags &= ~_TLF_LAZY_MMU;
1251		batch = this_cpu_ptr(&ppc64_tlb_batch);
1252		batch->active = 1;
1253	}
1254
1255	if (current->thread.regs) {
1256		restore_math(current->thread.regs);
1257
1258		/*
1259		 * The copy-paste buffer can only store into foreign real
1260		 * addresses, so unprivileged processes can not see the
1261		 * data or use it in any way unless they have foreign real
1262		 * mappings. If the new process has the foreign real address
1263		 * mappings, we must issue a cp_abort to clear any state and
1264		 * prevent snooping, corruption or a covert channel.
1265		 */
1266		if (current->mm &&
1267			atomic_read(&current->mm->context.vas_windows))
1268			asm volatile(PPC_CP_ABORT);
1269	}
1270#endif /* CONFIG_PPC_BOOK3S_64 */
1271
1272	return last;
1273}
1274
1275#define NR_INSN_TO_PRINT	16
1276
1277static void show_instructions(struct pt_regs *regs)
1278{
1279	int i;
1280	unsigned long nip = regs->nip;
1281	unsigned long pc = regs->nip - (NR_INSN_TO_PRINT * 3 / 4 * sizeof(int));
1282
1283	printk("Instruction dump:");
1284
1285	/*
1286	 * If we were executing with the MMU off for instructions, adjust pc
1287	 * rather than printing XXXXXXXX.
1288	 */
1289	if (!IS_ENABLED(CONFIG_BOOKE) && !(regs->msr & MSR_IR)) {
1290		pc = (unsigned long)phys_to_virt(pc);
1291		nip = (unsigned long)phys_to_virt(regs->nip);
1292	}
1293
1294	for (i = 0; i < NR_INSN_TO_PRINT; i++) {
1295		int instr;
1296
1297		if (!(i % 8))
1298			pr_cont("\n");
1299
 
 
 
 
 
 
 
 
1300		if (!__kernel_text_address(pc) ||
1301		    get_kernel_nofault(instr, (const void *)pc)) {
1302			pr_cont("XXXXXXXX ");
1303		} else {
1304			if (nip == pc)
1305				pr_cont("<%08x> ", instr);
1306			else
1307				pr_cont("%08x ", instr);
1308		}
1309
1310		pc += sizeof(int);
1311	}
1312
1313	pr_cont("\n");
1314}
1315
1316void show_user_instructions(struct pt_regs *regs)
1317{
1318	unsigned long pc;
1319	int n = NR_INSN_TO_PRINT;
1320	struct seq_buf s;
1321	char buf[96]; /* enough for 8 times 9 + 2 chars */
1322
1323	pc = regs->nip - (NR_INSN_TO_PRINT * 3 / 4 * sizeof(int));
1324
 
 
 
 
 
 
 
 
 
 
1325	seq_buf_init(&s, buf, sizeof(buf));
1326
1327	while (n) {
1328		int i;
1329
1330		seq_buf_clear(&s);
1331
1332		for (i = 0; i < 8 && n; i++, n--, pc += sizeof(int)) {
1333			int instr;
1334
1335			if (copy_from_user_nofault(&instr, (void __user *)pc,
1336					sizeof(instr))) {
1337				seq_buf_printf(&s, "XXXXXXXX ");
1338				continue;
1339			}
1340			seq_buf_printf(&s, regs->nip == pc ? "<%08x> " : "%08x ", instr);
1341		}
1342
1343		if (!seq_buf_has_overflowed(&s))
1344			pr_info("%s[%d]: code: %s\n", current->comm,
1345				current->pid, s.buffer);
1346	}
1347}
1348
1349struct regbit {
1350	unsigned long bit;
1351	const char *name;
1352};
1353
1354static struct regbit msr_bits[] = {
1355#if defined(CONFIG_PPC64) && !defined(CONFIG_BOOKE)
1356	{MSR_SF,	"SF"},
1357	{MSR_HV,	"HV"},
1358#endif
1359	{MSR_VEC,	"VEC"},
1360	{MSR_VSX,	"VSX"},
1361#ifdef CONFIG_BOOKE
1362	{MSR_CE,	"CE"},
1363#endif
1364	{MSR_EE,	"EE"},
1365	{MSR_PR,	"PR"},
1366	{MSR_FP,	"FP"},
1367	{MSR_ME,	"ME"},
1368#ifdef CONFIG_BOOKE
1369	{MSR_DE,	"DE"},
1370#else
1371	{MSR_SE,	"SE"},
1372	{MSR_BE,	"BE"},
1373#endif
1374	{MSR_IR,	"IR"},
1375	{MSR_DR,	"DR"},
1376	{MSR_PMM,	"PMM"},
1377#ifndef CONFIG_BOOKE
1378	{MSR_RI,	"RI"},
1379	{MSR_LE,	"LE"},
1380#endif
1381	{0,		NULL}
1382};
1383
1384static void print_bits(unsigned long val, struct regbit *bits, const char *sep)
1385{
1386	const char *s = "";
1387
1388	for (; bits->bit; ++bits)
1389		if (val & bits->bit) {
1390			pr_cont("%s%s", s, bits->name);
1391			s = sep;
1392		}
1393}
1394
1395#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1396static struct regbit msr_tm_bits[] = {
1397	{MSR_TS_T,	"T"},
1398	{MSR_TS_S,	"S"},
1399	{MSR_TM,	"E"},
1400	{0,		NULL}
1401};
1402
1403static void print_tm_bits(unsigned long val)
1404{
1405/*
1406 * This only prints something if at least one of the TM bit is set.
1407 * Inside the TM[], the output means:
1408 *   E: Enabled		(bit 32)
1409 *   S: Suspended	(bit 33)
1410 *   T: Transactional	(bit 34)
1411 */
1412	if (val & (MSR_TM | MSR_TS_S | MSR_TS_T)) {
1413		pr_cont(",TM[");
1414		print_bits(val, msr_tm_bits, "");
1415		pr_cont("]");
1416	}
1417}
1418#else
1419static void print_tm_bits(unsigned long val) {}
1420#endif
1421
1422static void print_msr_bits(unsigned long val)
1423{
1424	pr_cont("<");
1425	print_bits(val, msr_bits, ",");
1426	print_tm_bits(val);
1427	pr_cont(">");
1428}
1429
1430#ifdef CONFIG_PPC64
1431#define REG		"%016lx"
1432#define REGS_PER_LINE	4
1433#define LAST_VOLATILE	13
1434#else
1435#define REG		"%08lx"
1436#define REGS_PER_LINE	8
1437#define LAST_VOLATILE	12
1438#endif
1439
1440void show_regs(struct pt_regs * regs)
1441{
1442	int i, trap;
1443
1444	show_regs_print_info(KERN_DEFAULT);
1445
1446	printk("NIP:  "REG" LR: "REG" CTR: "REG"\n",
1447	       regs->nip, regs->link, regs->ctr);
1448	printk("REGS: %px TRAP: %04lx   %s  (%s)\n",
1449	       regs, regs->trap, print_tainted(), init_utsname()->release);
1450	printk("MSR:  "REG" ", regs->msr);
1451	print_msr_bits(regs->msr);
1452	pr_cont("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
1453	trap = TRAP(regs);
1454	if (!trap_is_syscall(regs) && cpu_has_feature(CPU_FTR_CFAR))
1455		pr_cont("CFAR: "REG" ", regs->orig_gpr3);
1456	if (trap == 0x200 || trap == 0x300 || trap == 0x600)
1457#if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
1458		pr_cont("DEAR: "REG" ESR: "REG" ", regs->dar, regs->dsisr);
1459#else
1460		pr_cont("DAR: "REG" DSISR: %08lx ", regs->dar, regs->dsisr);
1461#endif
1462#ifdef CONFIG_PPC64
1463	pr_cont("IRQMASK: %lx ", regs->softe);
1464#endif
1465#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1466	if (MSR_TM_ACTIVE(regs->msr))
1467		pr_cont("\nPACATMSCRATCH: %016llx ", get_paca()->tm_scratch);
1468#endif
1469
1470	for (i = 0;  i < 32;  i++) {
1471		if ((i % REGS_PER_LINE) == 0)
1472			pr_cont("\nGPR%02d: ", i);
1473		pr_cont(REG " ", regs->gpr[i]);
1474		if (i == LAST_VOLATILE && !FULL_REGS(regs))
1475			break;
1476	}
1477	pr_cont("\n");
1478#ifdef CONFIG_KALLSYMS
1479	/*
1480	 * Lookup NIP late so we have the best change of getting the
1481	 * above info out without failing
1482	 */
1483	printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
1484	printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
1485#endif
1486	show_stack(current, (unsigned long *) regs->gpr[1], KERN_DEFAULT);
1487	if (!user_mode(regs))
1488		show_instructions(regs);
1489}
1490
1491void flush_thread(void)
1492{
1493#ifdef CONFIG_HAVE_HW_BREAKPOINT
1494	flush_ptrace_hw_breakpoint(current);
1495#else /* CONFIG_HAVE_HW_BREAKPOINT */
1496	set_debug_reg_defaults(&current->thread);
1497#endif /* CONFIG_HAVE_HW_BREAKPOINT */
1498}
1499
1500#ifdef CONFIG_PPC_BOOK3S_64
1501void arch_setup_new_exec(void)
1502{
1503	if (radix_enabled())
1504		return;
1505	hash__setup_new_exec();
1506}
1507#endif
1508
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1509#ifdef CONFIG_PPC64
1510/**
1511 * Assign a TIDR (thread ID) for task @t and set it in the thread
1512 * structure. For now, we only support setting TIDR for 'current' task.
1513 *
1514 * Since the TID value is a truncated form of it PID, it is possible
1515 * (but unlikely) for 2 threads to have the same TID. In the unlikely event
1516 * that 2 threads share the same TID and are waiting, one of the following
1517 * cases will happen:
1518 *
1519 * 1. The correct thread is running, the wrong thread is not
1520 * In this situation, the correct thread is woken and proceeds to pass it's
1521 * condition check.
1522 *
1523 * 2. Neither threads are running
1524 * In this situation, neither thread will be woken. When scheduled, the waiting
1525 * threads will execute either a wait, which will return immediately, followed
1526 * by a condition check, which will pass for the correct thread and fail
1527 * for the wrong thread, or they will execute the condition check immediately.
1528 *
1529 * 3. The wrong thread is running, the correct thread is not
1530 * The wrong thread will be woken, but will fail it's condition check and
1531 * re-execute wait. The correct thread, when scheduled, will execute either
1532 * it's condition check (which will pass), or wait, which returns immediately
1533 * when called the first time after the thread is scheduled, followed by it's
1534 * condition check (which will pass).
1535 *
1536 * 4. Both threads are running
1537 * Both threads will be woken. The wrong thread will fail it's condition check
1538 * and execute another wait, while the correct thread will pass it's condition
1539 * check.
1540 *
1541 * @t: the task to set the thread ID for
1542 */
1543int set_thread_tidr(struct task_struct *t)
1544{
1545	if (!cpu_has_feature(CPU_FTR_P9_TIDR))
1546		return -EINVAL;
1547
1548	if (t != current)
1549		return -EINVAL;
1550
1551	if (t->thread.tidr)
1552		return 0;
1553
1554	t->thread.tidr = (u16)task_pid_nr(t);
1555	mtspr(SPRN_TIDR, t->thread.tidr);
1556
1557	return 0;
1558}
1559EXPORT_SYMBOL_GPL(set_thread_tidr);
1560
1561#endif /* CONFIG_PPC64 */
1562
1563void
1564release_thread(struct task_struct *t)
1565{
1566}
1567
1568/*
1569 * this gets called so that we can store coprocessor state into memory and
1570 * copy the current task into the new thread.
1571 */
1572int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
1573{
1574	flush_all_to_thread(src);
1575	/*
1576	 * Flush TM state out so we can copy it.  __switch_to_tm() does this
1577	 * flush but it removes the checkpointed state from the current CPU and
1578	 * transitions the CPU out of TM mode.  Hence we need to call
1579	 * tm_recheckpoint_new_task() (on the same task) to restore the
1580	 * checkpointed state back and the TM mode.
1581	 *
1582	 * Can't pass dst because it isn't ready. Doesn't matter, passing
1583	 * dst is only important for __switch_to()
1584	 */
1585	__switch_to_tm(src, src);
1586
1587	*dst = *src;
1588
1589	clear_task_ebb(dst);
1590
1591	return 0;
1592}
1593
1594static void setup_ksp_vsid(struct task_struct *p, unsigned long sp)
1595{
1596#ifdef CONFIG_PPC_BOOK3S_64
1597	unsigned long sp_vsid;
1598	unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
1599
1600	if (radix_enabled())
1601		return;
1602
1603	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1604		sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
1605			<< SLB_VSID_SHIFT_1T;
1606	else
1607		sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
1608			<< SLB_VSID_SHIFT;
1609	sp_vsid |= SLB_VSID_KERNEL | llp;
1610	p->thread.ksp_vsid = sp_vsid;
1611#endif
1612}
1613
1614/*
1615 * Copy a thread..
1616 */
1617
1618/*
1619 * Copy architecture-specific thread state
1620 */
1621int copy_thread(unsigned long clone_flags, unsigned long usp,
1622		unsigned long kthread_arg, struct task_struct *p,
1623		unsigned long tls)
1624{
1625	struct pt_regs *childregs, *kregs;
1626	extern void ret_from_fork(void);
1627	extern void ret_from_fork_scv(void);
1628	extern void ret_from_kernel_thread(void);
1629	void (*f)(void);
1630	unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
1631	struct thread_info *ti = task_thread_info(p);
1632#ifdef CONFIG_HAVE_HW_BREAKPOINT
1633	int i;
1634#endif
1635
1636	klp_init_thread_info(p);
1637
1638	/* Copy registers */
1639	sp -= sizeof(struct pt_regs);
1640	childregs = (struct pt_regs *) sp;
1641	if (unlikely(p->flags & PF_KTHREAD)) {
1642		/* kernel thread */
1643		memset(childregs, 0, sizeof(struct pt_regs));
1644		childregs->gpr[1] = sp + sizeof(struct pt_regs);
1645		/* function */
1646		if (usp)
1647			childregs->gpr[14] = ppc_function_entry((void *)usp);
1648#ifdef CONFIG_PPC64
1649		clear_tsk_thread_flag(p, TIF_32BIT);
1650		childregs->softe = IRQS_ENABLED;
1651#endif
1652		childregs->gpr[15] = kthread_arg;
1653		p->thread.regs = NULL;	/* no user register state */
1654		ti->flags |= _TIF_RESTOREALL;
1655		f = ret_from_kernel_thread;
1656	} else {
1657		/* user thread */
1658		struct pt_regs *regs = current_pt_regs();
1659		CHECK_FULL_REGS(regs);
1660		*childregs = *regs;
1661		if (usp)
1662			childregs->gpr[1] = usp;
1663		p->thread.regs = childregs;
1664		/* 64s sets this in ret_from_fork */
1665		if (!IS_ENABLED(CONFIG_PPC_BOOK3S_64))
1666			childregs->gpr[3] = 0;  /* Result from fork() */
1667		if (clone_flags & CLONE_SETTLS) {
 
1668			if (!is_32bit_task())
1669				childregs->gpr[13] = tls;
1670			else
 
1671				childregs->gpr[2] = tls;
1672		}
1673
1674		if (trap_is_scv(regs))
1675			f = ret_from_fork_scv;
1676		else
1677			f = ret_from_fork;
1678	}
1679	childregs->msr &= ~(MSR_FP|MSR_VEC|MSR_VSX);
1680	sp -= STACK_FRAME_OVERHEAD;
1681
1682	/*
1683	 * The way this works is that at some point in the future
1684	 * some task will call _switch to switch to the new task.
1685	 * That will pop off the stack frame created below and start
1686	 * the new task running at ret_from_fork.  The new task will
1687	 * do some house keeping and then return from the fork or clone
1688	 * system call, using the stack frame created above.
1689	 */
1690	((unsigned long *)sp)[0] = 0;
1691	sp -= sizeof(struct pt_regs);
1692	kregs = (struct pt_regs *) sp;
1693	sp -= STACK_FRAME_OVERHEAD;
1694	p->thread.ksp = sp;
1695#ifdef CONFIG_PPC32
1696	p->thread.ksp_limit = (unsigned long)end_of_stack(p);
1697#endif
1698#ifdef CONFIG_HAVE_HW_BREAKPOINT
1699	for (i = 0; i < nr_wp_slots(); i++)
1700		p->thread.ptrace_bps[i] = NULL;
1701#endif
1702
1703	p->thread.fp_save_area = NULL;
1704#ifdef CONFIG_ALTIVEC
1705	p->thread.vr_save_area = NULL;
1706#endif
1707
1708	setup_ksp_vsid(p, sp);
1709
1710#ifdef CONFIG_PPC64 
1711	if (cpu_has_feature(CPU_FTR_DSCR)) {
1712		p->thread.dscr_inherit = current->thread.dscr_inherit;
1713		p->thread.dscr = mfspr(SPRN_DSCR);
1714	}
1715	if (cpu_has_feature(CPU_FTR_HAS_PPR))
1716		childregs->ppr = DEFAULT_PPR;
1717
1718	p->thread.tidr = 0;
1719#endif
1720	kregs->nip = ppc_function_entry(f);
1721	return 0;
1722}
1723
1724void preload_new_slb_context(unsigned long start, unsigned long sp);
1725
1726/*
1727 * Set up a thread for executing a new program
1728 */
1729void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
1730{
1731#ifdef CONFIG_PPC64
1732	unsigned long load_addr = regs->gpr[2];	/* saved by ELF_PLAT_INIT */
1733
1734#ifdef CONFIG_PPC_BOOK3S_64
1735	if (!radix_enabled())
1736		preload_new_slb_context(start, sp);
1737#endif
1738#endif
1739
1740	/*
1741	 * If we exec out of a kernel thread then thread.regs will not be
1742	 * set.  Do it now.
1743	 */
1744	if (!current->thread.regs) {
1745		struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
1746		current->thread.regs = regs - 1;
1747	}
1748
1749#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1750	/*
1751	 * Clear any transactional state, we're exec()ing. The cause is
1752	 * not important as there will never be a recheckpoint so it's not
1753	 * user visible.
1754	 */
1755	if (MSR_TM_SUSPENDED(mfmsr()))
1756		tm_reclaim_current(0);
1757#endif
1758
1759	memset(regs->gpr, 0, sizeof(regs->gpr));
1760	regs->ctr = 0;
1761	regs->link = 0;
1762	regs->xer = 0;
1763	regs->ccr = 0;
1764	regs->gpr[1] = sp;
1765
1766	/*
1767	 * We have just cleared all the nonvolatile GPRs, so make
1768	 * FULL_REGS(regs) return true.  This is necessary to allow
1769	 * ptrace to examine the thread immediately after exec.
1770	 */
1771	SET_FULL_REGS(regs);
1772
1773#ifdef CONFIG_PPC32
1774	regs->mq = 0;
1775	regs->nip = start;
1776	regs->msr = MSR_USER;
1777#else
1778	if (!is_32bit_task()) {
1779		unsigned long entry;
1780
1781		if (is_elf2_task()) {
1782			/* Look ma, no function descriptors! */
1783			entry = start;
1784
1785			/*
1786			 * Ulrich says:
1787			 *   The latest iteration of the ABI requires that when
1788			 *   calling a function (at its global entry point),
1789			 *   the caller must ensure r12 holds the entry point
1790			 *   address (so that the function can quickly
1791			 *   establish addressability).
1792			 */
1793			regs->gpr[12] = start;
1794			/* Make sure that's restored on entry to userspace. */
1795			set_thread_flag(TIF_RESTOREALL);
1796		} else {
1797			unsigned long toc;
1798
1799			/* start is a relocated pointer to the function
1800			 * descriptor for the elf _start routine.  The first
1801			 * entry in the function descriptor is the entry
1802			 * address of _start and the second entry is the TOC
1803			 * value we need to use.
1804			 */
1805			__get_user(entry, (unsigned long __user *)start);
1806			__get_user(toc, (unsigned long __user *)start+1);
1807
1808			/* Check whether the e_entry function descriptor entries
1809			 * need to be relocated before we can use them.
1810			 */
1811			if (load_addr != 0) {
1812				entry += load_addr;
1813				toc   += load_addr;
1814			}
1815			regs->gpr[2] = toc;
1816		}
1817		regs->nip = entry;
1818		regs->msr = MSR_USER64;
1819	} else {
1820		regs->nip = start;
1821		regs->gpr[2] = 0;
1822		regs->msr = MSR_USER32;
1823	}
1824#endif
1825#ifdef CONFIG_VSX
1826	current->thread.used_vsr = 0;
1827#endif
1828	current->thread.load_slb = 0;
1829	current->thread.load_fp = 0;
1830	memset(&current->thread.fp_state, 0, sizeof(current->thread.fp_state));
1831	current->thread.fp_save_area = NULL;
1832#ifdef CONFIG_ALTIVEC
1833	memset(&current->thread.vr_state, 0, sizeof(current->thread.vr_state));
1834	current->thread.vr_state.vscr.u[3] = 0x00010000; /* Java mode disabled */
1835	current->thread.vr_save_area = NULL;
1836	current->thread.vrsave = 0;
1837	current->thread.used_vr = 0;
1838	current->thread.load_vec = 0;
1839#endif /* CONFIG_ALTIVEC */
1840#ifdef CONFIG_SPE
1841	memset(current->thread.evr, 0, sizeof(current->thread.evr));
1842	current->thread.acc = 0;
1843	current->thread.spefscr = 0;
1844	current->thread.used_spe = 0;
1845#endif /* CONFIG_SPE */
1846#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1847	current->thread.tm_tfhar = 0;
1848	current->thread.tm_texasr = 0;
1849	current->thread.tm_tfiar = 0;
1850	current->thread.load_tm = 0;
1851#endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1852
1853	thread_pkey_regs_init(&current->thread);
1854}
1855EXPORT_SYMBOL(start_thread);
1856
1857#define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
1858		| PR_FP_EXC_RES | PR_FP_EXC_INV)
1859
1860int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
1861{
1862	struct pt_regs *regs = tsk->thread.regs;
1863
1864	/* This is a bit hairy.  If we are an SPE enabled  processor
1865	 * (have embedded fp) we store the IEEE exception enable flags in
1866	 * fpexc_mode.  fpexc_mode is also used for setting FP exception
1867	 * mode (asyn, precise, disabled) for 'Classic' FP. */
1868	if (val & PR_FP_EXC_SW_ENABLE) {
1869#ifdef CONFIG_SPE
1870		if (cpu_has_feature(CPU_FTR_SPE)) {
1871			/*
1872			 * When the sticky exception bits are set
1873			 * directly by userspace, it must call prctl
1874			 * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
1875			 * in the existing prctl settings) or
1876			 * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
1877			 * the bits being set).  <fenv.h> functions
1878			 * saving and restoring the whole
1879			 * floating-point environment need to do so
1880			 * anyway to restore the prctl settings from
1881			 * the saved environment.
1882			 */
1883			tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
1884			tsk->thread.fpexc_mode = val &
1885				(PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
1886			return 0;
1887		} else {
1888			return -EINVAL;
1889		}
1890#else
1891		return -EINVAL;
1892#endif
1893	}
1894
1895	/* on a CONFIG_SPE this does not hurt us.  The bits that
1896	 * __pack_fe01 use do not overlap with bits used for
1897	 * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
1898	 * on CONFIG_SPE implementations are reserved so writing to
1899	 * them does not change anything */
1900	if (val > PR_FP_EXC_PRECISE)
1901		return -EINVAL;
1902	tsk->thread.fpexc_mode = __pack_fe01(val);
1903	if (regs != NULL && (regs->msr & MSR_FP) != 0)
1904		regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
1905			| tsk->thread.fpexc_mode;
1906	return 0;
1907}
1908
1909int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
1910{
1911	unsigned int val;
1912
1913	if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
1914#ifdef CONFIG_SPE
1915		if (cpu_has_feature(CPU_FTR_SPE)) {
1916			/*
1917			 * When the sticky exception bits are set
1918			 * directly by userspace, it must call prctl
1919			 * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
1920			 * in the existing prctl settings) or
1921			 * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
1922			 * the bits being set).  <fenv.h> functions
1923			 * saving and restoring the whole
1924			 * floating-point environment need to do so
1925			 * anyway to restore the prctl settings from
1926			 * the saved environment.
1927			 */
1928			tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
1929			val = tsk->thread.fpexc_mode;
1930		} else
1931			return -EINVAL;
1932#else
1933		return -EINVAL;
1934#endif
1935	else
1936		val = __unpack_fe01(tsk->thread.fpexc_mode);
1937	return put_user(val, (unsigned int __user *) adr);
1938}
1939
1940int set_endian(struct task_struct *tsk, unsigned int val)
1941{
1942	struct pt_regs *regs = tsk->thread.regs;
1943
1944	if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
1945	    (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
1946		return -EINVAL;
1947
1948	if (regs == NULL)
1949		return -EINVAL;
1950
1951	if (val == PR_ENDIAN_BIG)
1952		regs->msr &= ~MSR_LE;
1953	else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
1954		regs->msr |= MSR_LE;
1955	else
1956		return -EINVAL;
1957
1958	return 0;
1959}
1960
1961int get_endian(struct task_struct *tsk, unsigned long adr)
1962{
1963	struct pt_regs *regs = tsk->thread.regs;
1964	unsigned int val;
1965
1966	if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
1967	    !cpu_has_feature(CPU_FTR_REAL_LE))
1968		return -EINVAL;
1969
1970	if (regs == NULL)
1971		return -EINVAL;
1972
1973	if (regs->msr & MSR_LE) {
1974		if (cpu_has_feature(CPU_FTR_REAL_LE))
1975			val = PR_ENDIAN_LITTLE;
1976		else
1977			val = PR_ENDIAN_PPC_LITTLE;
1978	} else
1979		val = PR_ENDIAN_BIG;
1980
1981	return put_user(val, (unsigned int __user *)adr);
1982}
1983
1984int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
1985{
1986	tsk->thread.align_ctl = val;
1987	return 0;
1988}
1989
1990int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
1991{
1992	return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
1993}
1994
1995static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
1996				  unsigned long nbytes)
1997{
1998	unsigned long stack_page;
1999	unsigned long cpu = task_cpu(p);
2000
2001	stack_page = (unsigned long)hardirq_ctx[cpu];
2002	if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2003		return 1;
2004
2005	stack_page = (unsigned long)softirq_ctx[cpu];
2006	if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2007		return 1;
2008
2009	return 0;
2010}
2011
2012static inline int valid_emergency_stack(unsigned long sp, struct task_struct *p,
2013					unsigned long nbytes)
2014{
2015#ifdef CONFIG_PPC64
2016	unsigned long stack_page;
2017	unsigned long cpu = task_cpu(p);
2018
2019	stack_page = (unsigned long)paca_ptrs[cpu]->emergency_sp - THREAD_SIZE;
2020	if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2021		return 1;
2022
2023# ifdef CONFIG_PPC_BOOK3S_64
2024	stack_page = (unsigned long)paca_ptrs[cpu]->nmi_emergency_sp - THREAD_SIZE;
2025	if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2026		return 1;
2027
2028	stack_page = (unsigned long)paca_ptrs[cpu]->mc_emergency_sp - THREAD_SIZE;
2029	if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2030		return 1;
2031# endif
2032#endif
2033
2034	return 0;
2035}
2036
2037
2038int validate_sp(unsigned long sp, struct task_struct *p,
2039		       unsigned long nbytes)
2040{
2041	unsigned long stack_page = (unsigned long)task_stack_page(p);
2042
2043	if (sp < THREAD_SIZE)
2044		return 0;
2045
2046	if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
2047		return 1;
2048
2049	if (valid_irq_stack(sp, p, nbytes))
2050		return 1;
2051
2052	return valid_emergency_stack(sp, p, nbytes);
2053}
2054
2055EXPORT_SYMBOL(validate_sp);
2056
2057static unsigned long __get_wchan(struct task_struct *p)
2058{
2059	unsigned long ip, sp;
2060	int count = 0;
2061
2062	if (!p || p == current || p->state == TASK_RUNNING)
2063		return 0;
2064
2065	sp = p->thread.ksp;
2066	if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
2067		return 0;
2068
2069	do {
2070		sp = *(unsigned long *)sp;
2071		if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD) ||
2072		    p->state == TASK_RUNNING)
2073			return 0;
2074		if (count > 0) {
2075			ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
2076			if (!in_sched_functions(ip))
2077				return ip;
2078		}
2079	} while (count++ < 16);
2080	return 0;
2081}
2082
2083unsigned long get_wchan(struct task_struct *p)
2084{
2085	unsigned long ret;
2086
2087	if (!try_get_task_stack(p))
2088		return 0;
2089
2090	ret = __get_wchan(p);
2091
2092	put_task_stack(p);
2093
2094	return ret;
2095}
2096
2097static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
2098
2099void show_stack(struct task_struct *tsk, unsigned long *stack,
2100		const char *loglvl)
2101{
2102	unsigned long sp, ip, lr, newsp;
2103	int count = 0;
2104	int firstframe = 1;
2105#ifdef CONFIG_FUNCTION_GRAPH_TRACER
2106	unsigned long ret_addr;
2107	int ftrace_idx = 0;
2108#endif
2109
2110	if (tsk == NULL)
2111		tsk = current;
2112
2113	if (!try_get_task_stack(tsk))
2114		return;
2115
2116	sp = (unsigned long) stack;
2117	if (sp == 0) {
2118		if (tsk == current)
2119			sp = current_stack_frame();
2120		else
2121			sp = tsk->thread.ksp;
2122	}
2123
2124	lr = 0;
2125	printk("%sCall Trace:\n", loglvl);
2126	do {
2127		if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
2128			break;
2129
2130		stack = (unsigned long *) sp;
2131		newsp = stack[0];
2132		ip = stack[STACK_FRAME_LR_SAVE];
2133		if (!firstframe || ip != lr) {
2134			printk("%s["REG"] ["REG"] %pS",
2135				loglvl, sp, ip, (void *)ip);
2136#ifdef CONFIG_FUNCTION_GRAPH_TRACER
2137			ret_addr = ftrace_graph_ret_addr(current,
2138						&ftrace_idx, ip, stack);
2139			if (ret_addr != ip)
2140				pr_cont(" (%pS)", (void *)ret_addr);
2141#endif
2142			if (firstframe)
2143				pr_cont(" (unreliable)");
2144			pr_cont("\n");
2145		}
2146		firstframe = 0;
2147
2148		/*
2149		 * See if this is an exception frame.
2150		 * We look for the "regshere" marker in the current frame.
2151		 */
2152		if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
2153		    && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
2154			struct pt_regs *regs = (struct pt_regs *)
2155				(sp + STACK_FRAME_OVERHEAD);
2156			lr = regs->link;
2157			printk("%s--- interrupt: %lx at %pS\n    LR = %pS\n",
2158			       loglvl, regs->trap,
2159			       (void *)regs->nip, (void *)lr);
2160			firstframe = 1;
2161		}
2162
2163		sp = newsp;
2164	} while (count++ < kstack_depth_to_print);
2165
2166	put_task_stack(tsk);
2167}
2168
2169#ifdef CONFIG_PPC64
2170/* Called with hard IRQs off */
2171void notrace __ppc64_runlatch_on(void)
2172{
2173	struct thread_info *ti = current_thread_info();
2174
2175	if (cpu_has_feature(CPU_FTR_ARCH_206)) {
2176		/*
2177		 * Least significant bit (RUN) is the only writable bit of
2178		 * the CTRL register, so we can avoid mfspr. 2.06 is not the
2179		 * earliest ISA where this is the case, but it's convenient.
2180		 */
2181		mtspr(SPRN_CTRLT, CTRL_RUNLATCH);
2182	} else {
2183		unsigned long ctrl;
2184
2185		/*
2186		 * Some architectures (e.g., Cell) have writable fields other
2187		 * than RUN, so do the read-modify-write.
2188		 */
2189		ctrl = mfspr(SPRN_CTRLF);
2190		ctrl |= CTRL_RUNLATCH;
2191		mtspr(SPRN_CTRLT, ctrl);
2192	}
2193
2194	ti->local_flags |= _TLF_RUNLATCH;
2195}
2196
2197/* Called with hard IRQs off */
2198void notrace __ppc64_runlatch_off(void)
2199{
2200	struct thread_info *ti = current_thread_info();
2201
2202	ti->local_flags &= ~_TLF_RUNLATCH;
2203
2204	if (cpu_has_feature(CPU_FTR_ARCH_206)) {
2205		mtspr(SPRN_CTRLT, 0);
2206	} else {
2207		unsigned long ctrl;
2208
2209		ctrl = mfspr(SPRN_CTRLF);
2210		ctrl &= ~CTRL_RUNLATCH;
2211		mtspr(SPRN_CTRLT, ctrl);
2212	}
2213}
2214#endif /* CONFIG_PPC64 */
2215
2216unsigned long arch_align_stack(unsigned long sp)
2217{
2218	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
2219		sp -= get_random_int() & ~PAGE_MASK;
2220	return sp & ~0xf;
2221}
2222
2223static inline unsigned long brk_rnd(void)
2224{
2225        unsigned long rnd = 0;
2226
2227	/* 8MB for 32bit, 1GB for 64bit */
2228	if (is_32bit_task())
2229		rnd = (get_random_long() % (1UL<<(23-PAGE_SHIFT)));
2230	else
2231		rnd = (get_random_long() % (1UL<<(30-PAGE_SHIFT)));
2232
2233	return rnd << PAGE_SHIFT;
2234}
2235
2236unsigned long arch_randomize_brk(struct mm_struct *mm)
2237{
2238	unsigned long base = mm->brk;
2239	unsigned long ret;
2240
2241#ifdef CONFIG_PPC_BOOK3S_64
2242	/*
2243	 * If we are using 1TB segments and we are allowed to randomise
2244	 * the heap, we can put it above 1TB so it is backed by a 1TB
2245	 * segment. Otherwise the heap will be in the bottom 1TB
2246	 * which always uses 256MB segments and this may result in a
2247	 * performance penalty. We don't need to worry about radix. For
2248	 * radix, mmu_highuser_ssize remains unchanged from 256MB.
2249	 */
2250	if (!is_32bit_task() && (mmu_highuser_ssize == MMU_SEGSIZE_1T))
2251		base = max_t(unsigned long, mm->brk, 1UL << SID_SHIFT_1T);
2252#endif
2253
2254	ret = PAGE_ALIGN(base + brk_rnd());
2255
2256	if (ret < mm->brk)
2257		return mm->brk;
2258
2259	return ret;
2260}
2261
v5.4
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 *  Derived from "arch/i386/kernel/process.c"
   4 *    Copyright (C) 1995  Linus Torvalds
   5 *
   6 *  Updated and modified by Cort Dougan (cort@cs.nmt.edu) and
   7 *  Paul Mackerras (paulus@cs.anu.edu.au)
   8 *
   9 *  PowerPC version
  10 *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
  11 */
  12
  13#include <linux/errno.h>
  14#include <linux/sched.h>
  15#include <linux/sched/debug.h>
  16#include <linux/sched/task.h>
  17#include <linux/sched/task_stack.h>
  18#include <linux/kernel.h>
  19#include <linux/mm.h>
  20#include <linux/smp.h>
  21#include <linux/stddef.h>
  22#include <linux/unistd.h>
  23#include <linux/ptrace.h>
  24#include <linux/slab.h>
  25#include <linux/user.h>
  26#include <linux/elf.h>
  27#include <linux/prctl.h>
  28#include <linux/init_task.h>
  29#include <linux/export.h>
  30#include <linux/kallsyms.h>
  31#include <linux/mqueue.h>
  32#include <linux/hardirq.h>
  33#include <linux/utsname.h>
  34#include <linux/ftrace.h>
  35#include <linux/kernel_stat.h>
  36#include <linux/personality.h>
  37#include <linux/random.h>
  38#include <linux/hw_breakpoint.h>
  39#include <linux/uaccess.h>
  40#include <linux/elf-randomize.h>
  41#include <linux/pkeys.h>
  42#include <linux/seq_buf.h>
  43
  44#include <asm/pgtable.h>
  45#include <asm/io.h>
  46#include <asm/processor.h>
  47#include <asm/mmu.h>
  48#include <asm/prom.h>
  49#include <asm/machdep.h>
  50#include <asm/time.h>
  51#include <asm/runlatch.h>
  52#include <asm/syscalls.h>
  53#include <asm/switch_to.h>
  54#include <asm/tm.h>
  55#include <asm/debug.h>
  56#ifdef CONFIG_PPC64
  57#include <asm/firmware.h>
  58#include <asm/hw_irq.h>
  59#endif
  60#include <asm/code-patching.h>
  61#include <asm/exec.h>
  62#include <asm/livepatch.h>
  63#include <asm/cpu_has_feature.h>
  64#include <asm/asm-prototypes.h>
  65#include <asm/stacktrace.h>
  66#include <asm/hw_breakpoint.h>
  67
  68#include <linux/kprobes.h>
  69#include <linux/kdebug.h>
  70
  71/* Transactional Memory debug */
  72#ifdef TM_DEBUG_SW
  73#define TM_DEBUG(x...) printk(KERN_INFO x)
  74#else
  75#define TM_DEBUG(x...) do { } while(0)
  76#endif
  77
  78extern unsigned long _get_SP(void);
  79
  80#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
  81/*
  82 * Are we running in "Suspend disabled" mode? If so we have to block any
  83 * sigreturn that would get us into suspended state, and we also warn in some
  84 * other paths that we should never reach with suspend disabled.
  85 */
  86bool tm_suspend_disabled __ro_after_init = false;
  87
  88static void check_if_tm_restore_required(struct task_struct *tsk)
  89{
  90	/*
  91	 * If we are saving the current thread's registers, and the
  92	 * thread is in a transactional state, set the TIF_RESTORE_TM
  93	 * bit so that we know to restore the registers before
  94	 * returning to userspace.
  95	 */
  96	if (tsk == current && tsk->thread.regs &&
  97	    MSR_TM_ACTIVE(tsk->thread.regs->msr) &&
  98	    !test_thread_flag(TIF_RESTORE_TM)) {
  99		tsk->thread.ckpt_regs.msr = tsk->thread.regs->msr;
 100		set_thread_flag(TIF_RESTORE_TM);
 101	}
 102}
 103
 104#else
 105static inline void check_if_tm_restore_required(struct task_struct *tsk) { }
 106#endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
 107
 108bool strict_msr_control;
 109EXPORT_SYMBOL(strict_msr_control);
 110
 111static int __init enable_strict_msr_control(char *str)
 112{
 113	strict_msr_control = true;
 114	pr_info("Enabling strict facility control\n");
 115
 116	return 0;
 117}
 118early_param("ppc_strict_facility_enable", enable_strict_msr_control);
 119
 120/* notrace because it's called by restore_math */
 121unsigned long notrace msr_check_and_set(unsigned long bits)
 122{
 123	unsigned long oldmsr = mfmsr();
 124	unsigned long newmsr;
 125
 126	newmsr = oldmsr | bits;
 127
 128#ifdef CONFIG_VSX
 129	if (cpu_has_feature(CPU_FTR_VSX) && (bits & MSR_FP))
 130		newmsr |= MSR_VSX;
 131#endif
 132
 133	if (oldmsr != newmsr)
 134		mtmsr_isync(newmsr);
 135
 136	return newmsr;
 137}
 138EXPORT_SYMBOL_GPL(msr_check_and_set);
 139
 140/* notrace because it's called by restore_math */
 141void notrace __msr_check_and_clear(unsigned long bits)
 142{
 143	unsigned long oldmsr = mfmsr();
 144	unsigned long newmsr;
 145
 146	newmsr = oldmsr & ~bits;
 147
 148#ifdef CONFIG_VSX
 149	if (cpu_has_feature(CPU_FTR_VSX) && (bits & MSR_FP))
 150		newmsr &= ~MSR_VSX;
 151#endif
 152
 153	if (oldmsr != newmsr)
 154		mtmsr_isync(newmsr);
 155}
 156EXPORT_SYMBOL(__msr_check_and_clear);
 157
 158#ifdef CONFIG_PPC_FPU
 159static void __giveup_fpu(struct task_struct *tsk)
 160{
 161	unsigned long msr;
 162
 163	save_fpu(tsk);
 164	msr = tsk->thread.regs->msr;
 165	msr &= ~(MSR_FP|MSR_FE0|MSR_FE1);
 166#ifdef CONFIG_VSX
 167	if (cpu_has_feature(CPU_FTR_VSX))
 168		msr &= ~MSR_VSX;
 169#endif
 170	tsk->thread.regs->msr = msr;
 171}
 172
 173void giveup_fpu(struct task_struct *tsk)
 174{
 175	check_if_tm_restore_required(tsk);
 176
 177	msr_check_and_set(MSR_FP);
 178	__giveup_fpu(tsk);
 179	msr_check_and_clear(MSR_FP);
 180}
 181EXPORT_SYMBOL(giveup_fpu);
 182
 183/*
 184 * Make sure the floating-point register state in the
 185 * the thread_struct is up to date for task tsk.
 186 */
 187void flush_fp_to_thread(struct task_struct *tsk)
 188{
 189	if (tsk->thread.regs) {
 190		/*
 191		 * We need to disable preemption here because if we didn't,
 192		 * another process could get scheduled after the regs->msr
 193		 * test but before we have finished saving the FP registers
 194		 * to the thread_struct.  That process could take over the
 195		 * FPU, and then when we get scheduled again we would store
 196		 * bogus values for the remaining FP registers.
 197		 */
 198		preempt_disable();
 199		if (tsk->thread.regs->msr & MSR_FP) {
 200			/*
 201			 * This should only ever be called for current or
 202			 * for a stopped child process.  Since we save away
 203			 * the FP register state on context switch,
 204			 * there is something wrong if a stopped child appears
 205			 * to still have its FP state in the CPU registers.
 206			 */
 207			BUG_ON(tsk != current);
 208			giveup_fpu(tsk);
 209		}
 210		preempt_enable();
 211	}
 212}
 213EXPORT_SYMBOL_GPL(flush_fp_to_thread);
 214
 215void enable_kernel_fp(void)
 216{
 217	unsigned long cpumsr;
 218
 219	WARN_ON(preemptible());
 220
 221	cpumsr = msr_check_and_set(MSR_FP);
 222
 223	if (current->thread.regs && (current->thread.regs->msr & MSR_FP)) {
 224		check_if_tm_restore_required(current);
 225		/*
 226		 * If a thread has already been reclaimed then the
 227		 * checkpointed registers are on the CPU but have definitely
 228		 * been saved by the reclaim code. Don't need to and *cannot*
 229		 * giveup as this would save  to the 'live' structure not the
 230		 * checkpointed structure.
 231		 */
 232		if (!MSR_TM_ACTIVE(cpumsr) &&
 233		     MSR_TM_ACTIVE(current->thread.regs->msr))
 234			return;
 235		__giveup_fpu(current);
 236	}
 237}
 238EXPORT_SYMBOL(enable_kernel_fp);
 239
 240static int restore_fp(struct task_struct *tsk)
 241{
 242	if (tsk->thread.load_fp) {
 243		load_fp_state(&current->thread.fp_state);
 244		current->thread.load_fp++;
 245		return 1;
 246	}
 247	return 0;
 248}
 249#else
 250static int restore_fp(struct task_struct *tsk) { return 0; }
 251#endif /* CONFIG_PPC_FPU */
 252
 253#ifdef CONFIG_ALTIVEC
 254#define loadvec(thr) ((thr).load_vec)
 255
 256static void __giveup_altivec(struct task_struct *tsk)
 257{
 258	unsigned long msr;
 259
 260	save_altivec(tsk);
 261	msr = tsk->thread.regs->msr;
 262	msr &= ~MSR_VEC;
 263#ifdef CONFIG_VSX
 264	if (cpu_has_feature(CPU_FTR_VSX))
 265		msr &= ~MSR_VSX;
 266#endif
 267	tsk->thread.regs->msr = msr;
 268}
 269
 270void giveup_altivec(struct task_struct *tsk)
 271{
 272	check_if_tm_restore_required(tsk);
 273
 274	msr_check_and_set(MSR_VEC);
 275	__giveup_altivec(tsk);
 276	msr_check_and_clear(MSR_VEC);
 277}
 278EXPORT_SYMBOL(giveup_altivec);
 279
 280void enable_kernel_altivec(void)
 281{
 282	unsigned long cpumsr;
 283
 284	WARN_ON(preemptible());
 285
 286	cpumsr = msr_check_and_set(MSR_VEC);
 287
 288	if (current->thread.regs && (current->thread.regs->msr & MSR_VEC)) {
 289		check_if_tm_restore_required(current);
 290		/*
 291		 * If a thread has already been reclaimed then the
 292		 * checkpointed registers are on the CPU but have definitely
 293		 * been saved by the reclaim code. Don't need to and *cannot*
 294		 * giveup as this would save  to the 'live' structure not the
 295		 * checkpointed structure.
 296		 */
 297		if (!MSR_TM_ACTIVE(cpumsr) &&
 298		     MSR_TM_ACTIVE(current->thread.regs->msr))
 299			return;
 300		__giveup_altivec(current);
 301	}
 302}
 303EXPORT_SYMBOL(enable_kernel_altivec);
 304
 305/*
 306 * Make sure the VMX/Altivec register state in the
 307 * the thread_struct is up to date for task tsk.
 308 */
 309void flush_altivec_to_thread(struct task_struct *tsk)
 310{
 311	if (tsk->thread.regs) {
 312		preempt_disable();
 313		if (tsk->thread.regs->msr & MSR_VEC) {
 314			BUG_ON(tsk != current);
 315			giveup_altivec(tsk);
 316		}
 317		preempt_enable();
 318	}
 319}
 320EXPORT_SYMBOL_GPL(flush_altivec_to_thread);
 321
 322static int restore_altivec(struct task_struct *tsk)
 323{
 324	if (cpu_has_feature(CPU_FTR_ALTIVEC) && (tsk->thread.load_vec)) {
 325		load_vr_state(&tsk->thread.vr_state);
 326		tsk->thread.used_vr = 1;
 327		tsk->thread.load_vec++;
 328
 329		return 1;
 330	}
 331	return 0;
 332}
 333#else
 334#define loadvec(thr) 0
 335static inline int restore_altivec(struct task_struct *tsk) { return 0; }
 336#endif /* CONFIG_ALTIVEC */
 337
 338#ifdef CONFIG_VSX
 339static void __giveup_vsx(struct task_struct *tsk)
 340{
 341	unsigned long msr = tsk->thread.regs->msr;
 342
 343	/*
 344	 * We should never be ssetting MSR_VSX without also setting
 345	 * MSR_FP and MSR_VEC
 346	 */
 347	WARN_ON((msr & MSR_VSX) && !((msr & MSR_FP) && (msr & MSR_VEC)));
 348
 349	/* __giveup_fpu will clear MSR_VSX */
 350	if (msr & MSR_FP)
 351		__giveup_fpu(tsk);
 352	if (msr & MSR_VEC)
 353		__giveup_altivec(tsk);
 354}
 355
 356static void giveup_vsx(struct task_struct *tsk)
 357{
 358	check_if_tm_restore_required(tsk);
 359
 360	msr_check_and_set(MSR_FP|MSR_VEC|MSR_VSX);
 361	__giveup_vsx(tsk);
 362	msr_check_and_clear(MSR_FP|MSR_VEC|MSR_VSX);
 363}
 364
 365void enable_kernel_vsx(void)
 366{
 367	unsigned long cpumsr;
 368
 369	WARN_ON(preemptible());
 370
 371	cpumsr = msr_check_and_set(MSR_FP|MSR_VEC|MSR_VSX);
 372
 373	if (current->thread.regs &&
 374	    (current->thread.regs->msr & (MSR_VSX|MSR_VEC|MSR_FP))) {
 375		check_if_tm_restore_required(current);
 376		/*
 377		 * If a thread has already been reclaimed then the
 378		 * checkpointed registers are on the CPU but have definitely
 379		 * been saved by the reclaim code. Don't need to and *cannot*
 380		 * giveup as this would save  to the 'live' structure not the
 381		 * checkpointed structure.
 382		 */
 383		if (!MSR_TM_ACTIVE(cpumsr) &&
 384		     MSR_TM_ACTIVE(current->thread.regs->msr))
 385			return;
 386		__giveup_vsx(current);
 387	}
 388}
 389EXPORT_SYMBOL(enable_kernel_vsx);
 390
 391void flush_vsx_to_thread(struct task_struct *tsk)
 392{
 393	if (tsk->thread.regs) {
 394		preempt_disable();
 395		if (tsk->thread.regs->msr & (MSR_VSX|MSR_VEC|MSR_FP)) {
 396			BUG_ON(tsk != current);
 397			giveup_vsx(tsk);
 398		}
 399		preempt_enable();
 400	}
 401}
 402EXPORT_SYMBOL_GPL(flush_vsx_to_thread);
 403
 404static int restore_vsx(struct task_struct *tsk)
 405{
 406	if (cpu_has_feature(CPU_FTR_VSX)) {
 407		tsk->thread.used_vsr = 1;
 408		return 1;
 409	}
 410
 411	return 0;
 412}
 413#else
 414static inline int restore_vsx(struct task_struct *tsk) { return 0; }
 415#endif /* CONFIG_VSX */
 416
 417#ifdef CONFIG_SPE
 418void giveup_spe(struct task_struct *tsk)
 419{
 420	check_if_tm_restore_required(tsk);
 421
 422	msr_check_and_set(MSR_SPE);
 423	__giveup_spe(tsk);
 424	msr_check_and_clear(MSR_SPE);
 425}
 426EXPORT_SYMBOL(giveup_spe);
 427
 428void enable_kernel_spe(void)
 429{
 430	WARN_ON(preemptible());
 431
 432	msr_check_and_set(MSR_SPE);
 433
 434	if (current->thread.regs && (current->thread.regs->msr & MSR_SPE)) {
 435		check_if_tm_restore_required(current);
 436		__giveup_spe(current);
 437	}
 438}
 439EXPORT_SYMBOL(enable_kernel_spe);
 440
 441void flush_spe_to_thread(struct task_struct *tsk)
 442{
 443	if (tsk->thread.regs) {
 444		preempt_disable();
 445		if (tsk->thread.regs->msr & MSR_SPE) {
 446			BUG_ON(tsk != current);
 447			tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
 448			giveup_spe(tsk);
 449		}
 450		preempt_enable();
 451	}
 452}
 453#endif /* CONFIG_SPE */
 454
 455static unsigned long msr_all_available;
 456
 457static int __init init_msr_all_available(void)
 458{
 459#ifdef CONFIG_PPC_FPU
 460	msr_all_available |= MSR_FP;
 461#endif
 462#ifdef CONFIG_ALTIVEC
 463	if (cpu_has_feature(CPU_FTR_ALTIVEC))
 464		msr_all_available |= MSR_VEC;
 465#endif
 466#ifdef CONFIG_VSX
 467	if (cpu_has_feature(CPU_FTR_VSX))
 468		msr_all_available |= MSR_VSX;
 469#endif
 470#ifdef CONFIG_SPE
 471	if (cpu_has_feature(CPU_FTR_SPE))
 472		msr_all_available |= MSR_SPE;
 473#endif
 474
 475	return 0;
 476}
 477early_initcall(init_msr_all_available);
 478
 479void giveup_all(struct task_struct *tsk)
 480{
 481	unsigned long usermsr;
 482
 483	if (!tsk->thread.regs)
 484		return;
 485
 486	check_if_tm_restore_required(tsk);
 487
 488	usermsr = tsk->thread.regs->msr;
 489
 490	if ((usermsr & msr_all_available) == 0)
 491		return;
 492
 493	msr_check_and_set(msr_all_available);
 494
 495	WARN_ON((usermsr & MSR_VSX) && !((usermsr & MSR_FP) && (usermsr & MSR_VEC)));
 496
 497#ifdef CONFIG_PPC_FPU
 498	if (usermsr & MSR_FP)
 499		__giveup_fpu(tsk);
 500#endif
 501#ifdef CONFIG_ALTIVEC
 502	if (usermsr & MSR_VEC)
 503		__giveup_altivec(tsk);
 504#endif
 505#ifdef CONFIG_SPE
 506	if (usermsr & MSR_SPE)
 507		__giveup_spe(tsk);
 508#endif
 509
 510	msr_check_and_clear(msr_all_available);
 511}
 512EXPORT_SYMBOL(giveup_all);
 513
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 514/*
 515 * The exception exit path calls restore_math() with interrupts hard disabled
 516 * but the soft irq state not "reconciled". ftrace code that calls
 517 * local_irq_save/restore causes warnings.
 518 *
 519 * Rather than complicate the exit path, just don't trace restore_math. This
 520 * could be done by having ftrace entry code check for this un-reconciled
 521 * condition where MSR[EE]=0 and PACA_IRQ_HARD_DIS is not set, and
 522 * temporarily fix it up for the duration of the ftrace call.
 523 */
 524void notrace restore_math(struct pt_regs *regs)
 525{
 526	unsigned long msr;
 527
 528	if (!MSR_TM_ACTIVE(regs->msr) &&
 529		!current->thread.load_fp && !loadvec(current->thread))
 530		return;
 531
 532	msr = regs->msr;
 533	msr_check_and_set(msr_all_available);
 534
 535	/*
 536	 * Only reload if the bit is not set in the user MSR, the bit BEING set
 537	 * indicates that the registers are hot
 
 538	 */
 539	if ((!(msr & MSR_FP)) && restore_fp(current))
 540		msr |= MSR_FP | current->thread.fpexc_mode;
 541
 542	if ((!(msr & MSR_VEC)) && restore_altivec(current))
 543		msr |= MSR_VEC;
 544
 545	if ((msr & (MSR_FP | MSR_VEC)) == (MSR_FP | MSR_VEC) &&
 546			restore_vsx(current)) {
 547		msr |= MSR_VSX;
 548	}
 549
 550	msr_check_and_clear(msr_all_available);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 551
 552	regs->msr = msr;
 
 553}
 
 554
 555static void save_all(struct task_struct *tsk)
 556{
 557	unsigned long usermsr;
 558
 559	if (!tsk->thread.regs)
 560		return;
 561
 562	usermsr = tsk->thread.regs->msr;
 563
 564	if ((usermsr & msr_all_available) == 0)
 565		return;
 566
 567	msr_check_and_set(msr_all_available);
 568
 569	WARN_ON((usermsr & MSR_VSX) && !((usermsr & MSR_FP) && (usermsr & MSR_VEC)));
 570
 571	if (usermsr & MSR_FP)
 572		save_fpu(tsk);
 573
 574	if (usermsr & MSR_VEC)
 575		save_altivec(tsk);
 576
 577	if (usermsr & MSR_SPE)
 578		__giveup_spe(tsk);
 579
 580	msr_check_and_clear(msr_all_available);
 581	thread_pkey_regs_save(&tsk->thread);
 582}
 583
 584void flush_all_to_thread(struct task_struct *tsk)
 585{
 586	if (tsk->thread.regs) {
 587		preempt_disable();
 588		BUG_ON(tsk != current);
 589#ifdef CONFIG_SPE
 590		if (tsk->thread.regs->msr & MSR_SPE)
 591			tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
 592#endif
 593		save_all(tsk);
 594
 595		preempt_enable();
 596	}
 597}
 598EXPORT_SYMBOL(flush_all_to_thread);
 599
 600#ifdef CONFIG_PPC_ADV_DEBUG_REGS
 601void do_send_trap(struct pt_regs *regs, unsigned long address,
 602		  unsigned long error_code, int breakpt)
 603{
 604	current->thread.trap_nr = TRAP_HWBKPT;
 605	if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
 606			11, SIGSEGV) == NOTIFY_STOP)
 607		return;
 608
 609	/* Deliver the signal to userspace */
 610	force_sig_ptrace_errno_trap(breakpt, /* breakpoint or watchpoint id */
 611				    (void __user *)address);
 612}
 613#else	/* !CONFIG_PPC_ADV_DEBUG_REGS */
 614void do_break (struct pt_regs *regs, unsigned long address,
 615		    unsigned long error_code)
 616{
 617	current->thread.trap_nr = TRAP_HWBKPT;
 618	if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
 619			11, SIGSEGV) == NOTIFY_STOP)
 620		return;
 621
 622	if (debugger_break_match(regs))
 623		return;
 624
 625	/* Clear the breakpoint */
 626	hw_breakpoint_disable();
 627
 628	/* Deliver the signal to userspace */
 629	force_sig_fault(SIGTRAP, TRAP_HWBKPT, (void __user *)address);
 630}
 631#endif	/* CONFIG_PPC_ADV_DEBUG_REGS */
 632
 633static DEFINE_PER_CPU(struct arch_hw_breakpoint, current_brk);
 634
 635#ifdef CONFIG_PPC_ADV_DEBUG_REGS
 636/*
 637 * Set the debug registers back to their default "safe" values.
 638 */
 639static void set_debug_reg_defaults(struct thread_struct *thread)
 640{
 641	thread->debug.iac1 = thread->debug.iac2 = 0;
 642#if CONFIG_PPC_ADV_DEBUG_IACS > 2
 643	thread->debug.iac3 = thread->debug.iac4 = 0;
 644#endif
 645	thread->debug.dac1 = thread->debug.dac2 = 0;
 646#if CONFIG_PPC_ADV_DEBUG_DVCS > 0
 647	thread->debug.dvc1 = thread->debug.dvc2 = 0;
 648#endif
 649	thread->debug.dbcr0 = 0;
 650#ifdef CONFIG_BOOKE
 651	/*
 652	 * Force User/Supervisor bits to b11 (user-only MSR[PR]=1)
 653	 */
 654	thread->debug.dbcr1 = DBCR1_IAC1US | DBCR1_IAC2US |
 655			DBCR1_IAC3US | DBCR1_IAC4US;
 656	/*
 657	 * Force Data Address Compare User/Supervisor bits to be User-only
 658	 * (0b11 MSR[PR]=1) and set all other bits in DBCR2 register to be 0.
 659	 */
 660	thread->debug.dbcr2 = DBCR2_DAC1US | DBCR2_DAC2US;
 661#else
 662	thread->debug.dbcr1 = 0;
 663#endif
 664}
 665
 666static void prime_debug_regs(struct debug_reg *debug)
 667{
 668	/*
 669	 * We could have inherited MSR_DE from userspace, since
 670	 * it doesn't get cleared on exception entry.  Make sure
 671	 * MSR_DE is clear before we enable any debug events.
 672	 */
 673	mtmsr(mfmsr() & ~MSR_DE);
 674
 675	mtspr(SPRN_IAC1, debug->iac1);
 676	mtspr(SPRN_IAC2, debug->iac2);
 677#if CONFIG_PPC_ADV_DEBUG_IACS > 2
 678	mtspr(SPRN_IAC3, debug->iac3);
 679	mtspr(SPRN_IAC4, debug->iac4);
 680#endif
 681	mtspr(SPRN_DAC1, debug->dac1);
 682	mtspr(SPRN_DAC2, debug->dac2);
 683#if CONFIG_PPC_ADV_DEBUG_DVCS > 0
 684	mtspr(SPRN_DVC1, debug->dvc1);
 685	mtspr(SPRN_DVC2, debug->dvc2);
 686#endif
 687	mtspr(SPRN_DBCR0, debug->dbcr0);
 688	mtspr(SPRN_DBCR1, debug->dbcr1);
 689#ifdef CONFIG_BOOKE
 690	mtspr(SPRN_DBCR2, debug->dbcr2);
 691#endif
 692}
 693/*
 694 * Unless neither the old or new thread are making use of the
 695 * debug registers, set the debug registers from the values
 696 * stored in the new thread.
 697 */
 698void switch_booke_debug_regs(struct debug_reg *new_debug)
 699{
 700	if ((current->thread.debug.dbcr0 & DBCR0_IDM)
 701		|| (new_debug->dbcr0 & DBCR0_IDM))
 702			prime_debug_regs(new_debug);
 703}
 704EXPORT_SYMBOL_GPL(switch_booke_debug_regs);
 705#else	/* !CONFIG_PPC_ADV_DEBUG_REGS */
 706#ifndef CONFIG_HAVE_HW_BREAKPOINT
 707static void set_breakpoint(struct arch_hw_breakpoint *brk)
 708{
 709	preempt_disable();
 710	__set_breakpoint(brk);
 711	preempt_enable();
 712}
 713
 714static void set_debug_reg_defaults(struct thread_struct *thread)
 715{
 716	thread->hw_brk.address = 0;
 717	thread->hw_brk.type = 0;
 718	if (ppc_breakpoint_available())
 719		set_breakpoint(&thread->hw_brk);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 720}
 721#endif /* !CONFIG_HAVE_HW_BREAKPOINT */
 722#endif	/* CONFIG_PPC_ADV_DEBUG_REGS */
 723
 724#ifdef CONFIG_PPC_ADV_DEBUG_REGS
 725static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
 726{
 727	mtspr(SPRN_DAC1, dabr);
 728#ifdef CONFIG_PPC_47x
 729	isync();
 730#endif
 731	return 0;
 732}
 733#elif defined(CONFIG_PPC_BOOK3S)
 734static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
 735{
 736	mtspr(SPRN_DABR, dabr);
 737	if (cpu_has_feature(CPU_FTR_DABRX))
 738		mtspr(SPRN_DABRX, dabrx);
 739	return 0;
 740}
 741#elif defined(CONFIG_PPC_8xx)
 742static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
 743{
 744	unsigned long addr = dabr & ~HW_BRK_TYPE_DABR;
 745	unsigned long lctrl1 = 0x90000000; /* compare type: equal on E & F */
 746	unsigned long lctrl2 = 0x8e000002; /* watchpoint 1 on cmp E | F */
 747
 748	if ((dabr & HW_BRK_TYPE_RDWR) == HW_BRK_TYPE_READ)
 749		lctrl1 |= 0xa0000;
 750	else if ((dabr & HW_BRK_TYPE_RDWR) == HW_BRK_TYPE_WRITE)
 751		lctrl1 |= 0xf0000;
 752	else if ((dabr & HW_BRK_TYPE_RDWR) == 0)
 753		lctrl2 = 0;
 754
 755	mtspr(SPRN_LCTRL2, 0);
 756	mtspr(SPRN_CMPE, addr);
 757	mtspr(SPRN_CMPF, addr + 4);
 758	mtspr(SPRN_LCTRL1, lctrl1);
 759	mtspr(SPRN_LCTRL2, lctrl2);
 760
 761	return 0;
 762}
 763#else
 764static inline int __set_dabr(unsigned long dabr, unsigned long dabrx)
 765{
 766	return -EINVAL;
 767}
 768#endif
 769
 770static inline int set_dabr(struct arch_hw_breakpoint *brk)
 771{
 772	unsigned long dabr, dabrx;
 773
 774	dabr = brk->address | (brk->type & HW_BRK_TYPE_DABR);
 775	dabrx = ((brk->type >> 3) & 0x7);
 776
 777	if (ppc_md.set_dabr)
 778		return ppc_md.set_dabr(dabr, dabrx);
 779
 780	return __set_dabr(dabr, dabrx);
 781}
 782
 783void __set_breakpoint(struct arch_hw_breakpoint *brk)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 784{
 785	memcpy(this_cpu_ptr(&current_brk), brk, sizeof(*brk));
 786
 787	if (dawr_enabled())
 788		// Power8 or later
 789		set_dawr(brk);
 
 
 790	else if (!cpu_has_feature(CPU_FTR_ARCH_207S))
 791		// Power7 or earlier
 792		set_dabr(brk);
 793	else
 794		// Shouldn't happen due to higher level checks
 795		WARN_ON_ONCE(1);
 796}
 797
 798/* Check if we have DAWR or DABR hardware */
 799bool ppc_breakpoint_available(void)
 800{
 801	if (dawr_enabled())
 802		return true; /* POWER8 DAWR or POWER9 forced DAWR */
 803	if (cpu_has_feature(CPU_FTR_ARCH_207S))
 804		return false; /* POWER9 with DAWR disabled */
 805	/* DABR: Everything but POWER8 and POWER9 */
 806	return true;
 807}
 808EXPORT_SYMBOL_GPL(ppc_breakpoint_available);
 809
 810static inline bool hw_brk_match(struct arch_hw_breakpoint *a,
 811			      struct arch_hw_breakpoint *b)
 812{
 813	if (a->address != b->address)
 814		return false;
 815	if (a->type != b->type)
 816		return false;
 817	if (a->len != b->len)
 818		return false;
 819	return true;
 820}
 821
 822#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
 823
 824static inline bool tm_enabled(struct task_struct *tsk)
 825{
 826	return tsk && tsk->thread.regs && (tsk->thread.regs->msr & MSR_TM);
 827}
 828
 829static void tm_reclaim_thread(struct thread_struct *thr, uint8_t cause)
 830{
 831	/*
 832	 * Use the current MSR TM suspended bit to track if we have
 833	 * checkpointed state outstanding.
 834	 * On signal delivery, we'd normally reclaim the checkpointed
 835	 * state to obtain stack pointer (see:get_tm_stackpointer()).
 836	 * This will then directly return to userspace without going
 837	 * through __switch_to(). However, if the stack frame is bad,
 838	 * we need to exit this thread which calls __switch_to() which
 839	 * will again attempt to reclaim the already saved tm state.
 840	 * Hence we need to check that we've not already reclaimed
 841	 * this state.
 842	 * We do this using the current MSR, rather tracking it in
 843	 * some specific thread_struct bit, as it has the additional
 844	 * benefit of checking for a potential TM bad thing exception.
 845	 */
 846	if (!MSR_TM_SUSPENDED(mfmsr()))
 847		return;
 848
 849	giveup_all(container_of(thr, struct task_struct, thread));
 850
 851	tm_reclaim(thr, cause);
 852
 853	/*
 854	 * If we are in a transaction and FP is off then we can't have
 855	 * used FP inside that transaction. Hence the checkpointed
 856	 * state is the same as the live state. We need to copy the
 857	 * live state to the checkpointed state so that when the
 858	 * transaction is restored, the checkpointed state is correct
 859	 * and the aborted transaction sees the correct state. We use
 860	 * ckpt_regs.msr here as that's what tm_reclaim will use to
 861	 * determine if it's going to write the checkpointed state or
 862	 * not. So either this will write the checkpointed registers,
 863	 * or reclaim will. Similarly for VMX.
 864	 */
 865	if ((thr->ckpt_regs.msr & MSR_FP) == 0)
 866		memcpy(&thr->ckfp_state, &thr->fp_state,
 867		       sizeof(struct thread_fp_state));
 868	if ((thr->ckpt_regs.msr & MSR_VEC) == 0)
 869		memcpy(&thr->ckvr_state, &thr->vr_state,
 870		       sizeof(struct thread_vr_state));
 871}
 872
 873void tm_reclaim_current(uint8_t cause)
 874{
 875	tm_enable();
 876	tm_reclaim_thread(&current->thread, cause);
 877}
 878
 879static inline void tm_reclaim_task(struct task_struct *tsk)
 880{
 881	/* We have to work out if we're switching from/to a task that's in the
 882	 * middle of a transaction.
 883	 *
 884	 * In switching we need to maintain a 2nd register state as
 885	 * oldtask->thread.ckpt_regs.  We tm_reclaim(oldproc); this saves the
 886	 * checkpointed (tbegin) state in ckpt_regs, ckfp_state and
 887	 * ckvr_state
 888	 *
 889	 * We also context switch (save) TFHAR/TEXASR/TFIAR in here.
 890	 */
 891	struct thread_struct *thr = &tsk->thread;
 892
 893	if (!thr->regs)
 894		return;
 895
 896	if (!MSR_TM_ACTIVE(thr->regs->msr))
 897		goto out_and_saveregs;
 898
 899	WARN_ON(tm_suspend_disabled);
 900
 901	TM_DEBUG("--- tm_reclaim on pid %d (NIP=%lx, "
 902		 "ccr=%lx, msr=%lx, trap=%lx)\n",
 903		 tsk->pid, thr->regs->nip,
 904		 thr->regs->ccr, thr->regs->msr,
 905		 thr->regs->trap);
 906
 907	tm_reclaim_thread(thr, TM_CAUSE_RESCHED);
 908
 909	TM_DEBUG("--- tm_reclaim on pid %d complete\n",
 910		 tsk->pid);
 911
 912out_and_saveregs:
 913	/* Always save the regs here, even if a transaction's not active.
 914	 * This context-switches a thread's TM info SPRs.  We do it here to
 915	 * be consistent with the restore path (in recheckpoint) which
 916	 * cannot happen later in _switch().
 917	 */
 918	tm_save_sprs(thr);
 919}
 920
 921extern void __tm_recheckpoint(struct thread_struct *thread);
 922
 923void tm_recheckpoint(struct thread_struct *thread)
 924{
 925	unsigned long flags;
 926
 927	if (!(thread->regs->msr & MSR_TM))
 928		return;
 929
 930	/* We really can't be interrupted here as the TEXASR registers can't
 931	 * change and later in the trecheckpoint code, we have a userspace R1.
 932	 * So let's hard disable over this region.
 933	 */
 934	local_irq_save(flags);
 935	hard_irq_disable();
 936
 937	/* The TM SPRs are restored here, so that TEXASR.FS can be set
 938	 * before the trecheckpoint and no explosion occurs.
 939	 */
 940	tm_restore_sprs(thread);
 941
 942	__tm_recheckpoint(thread);
 943
 944	local_irq_restore(flags);
 945}
 946
 947static inline void tm_recheckpoint_new_task(struct task_struct *new)
 948{
 949	if (!cpu_has_feature(CPU_FTR_TM))
 950		return;
 951
 952	/* Recheckpoint the registers of the thread we're about to switch to.
 953	 *
 954	 * If the task was using FP, we non-lazily reload both the original and
 955	 * the speculative FP register states.  This is because the kernel
 956	 * doesn't see if/when a TM rollback occurs, so if we take an FP
 957	 * unavailable later, we are unable to determine which set of FP regs
 958	 * need to be restored.
 959	 */
 960	if (!tm_enabled(new))
 961		return;
 962
 963	if (!MSR_TM_ACTIVE(new->thread.regs->msr)){
 964		tm_restore_sprs(&new->thread);
 965		return;
 966	}
 967	/* Recheckpoint to restore original checkpointed register state. */
 968	TM_DEBUG("*** tm_recheckpoint of pid %d (new->msr 0x%lx)\n",
 969		 new->pid, new->thread.regs->msr);
 970
 971	tm_recheckpoint(&new->thread);
 972
 973	/*
 974	 * The checkpointed state has been restored but the live state has
 975	 * not, ensure all the math functionality is turned off to trigger
 976	 * restore_math() to reload.
 977	 */
 978	new->thread.regs->msr &= ~(MSR_FP | MSR_VEC | MSR_VSX);
 979
 980	TM_DEBUG("*** tm_recheckpoint of pid %d complete "
 981		 "(kernel msr 0x%lx)\n",
 982		 new->pid, mfmsr());
 983}
 984
 985static inline void __switch_to_tm(struct task_struct *prev,
 986		struct task_struct *new)
 987{
 988	if (cpu_has_feature(CPU_FTR_TM)) {
 989		if (tm_enabled(prev) || tm_enabled(new))
 990			tm_enable();
 991
 992		if (tm_enabled(prev)) {
 993			prev->thread.load_tm++;
 994			tm_reclaim_task(prev);
 995			if (!MSR_TM_ACTIVE(prev->thread.regs->msr) && prev->thread.load_tm == 0)
 996				prev->thread.regs->msr &= ~MSR_TM;
 997		}
 998
 999		tm_recheckpoint_new_task(new);
1000	}
1001}
1002
1003/*
1004 * This is called if we are on the way out to userspace and the
1005 * TIF_RESTORE_TM flag is set.  It checks if we need to reload
1006 * FP and/or vector state and does so if necessary.
1007 * If userspace is inside a transaction (whether active or
1008 * suspended) and FP/VMX/VSX instructions have ever been enabled
1009 * inside that transaction, then we have to keep them enabled
1010 * and keep the FP/VMX/VSX state loaded while ever the transaction
1011 * continues.  The reason is that if we didn't, and subsequently
1012 * got a FP/VMX/VSX unavailable interrupt inside a transaction,
1013 * we don't know whether it's the same transaction, and thus we
1014 * don't know which of the checkpointed state and the transactional
1015 * state to use.
1016 */
1017void restore_tm_state(struct pt_regs *regs)
1018{
1019	unsigned long msr_diff;
1020
1021	/*
1022	 * This is the only moment we should clear TIF_RESTORE_TM as
1023	 * it is here that ckpt_regs.msr and pt_regs.msr become the same
1024	 * again, anything else could lead to an incorrect ckpt_msr being
1025	 * saved and therefore incorrect signal contexts.
1026	 */
1027	clear_thread_flag(TIF_RESTORE_TM);
1028	if (!MSR_TM_ACTIVE(regs->msr))
1029		return;
1030
1031	msr_diff = current->thread.ckpt_regs.msr & ~regs->msr;
1032	msr_diff &= MSR_FP | MSR_VEC | MSR_VSX;
1033
1034	/* Ensure that restore_math() will restore */
1035	if (msr_diff & MSR_FP)
1036		current->thread.load_fp = 1;
1037#ifdef CONFIG_ALTIVEC
1038	if (cpu_has_feature(CPU_FTR_ALTIVEC) && msr_diff & MSR_VEC)
1039		current->thread.load_vec = 1;
1040#endif
1041	restore_math(regs);
1042
1043	regs->msr |= msr_diff;
1044}
1045
1046#else
1047#define tm_recheckpoint_new_task(new)
1048#define __switch_to_tm(prev, new)
1049#endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1050
1051static inline void save_sprs(struct thread_struct *t)
1052{
1053#ifdef CONFIG_ALTIVEC
1054	if (cpu_has_feature(CPU_FTR_ALTIVEC))
1055		t->vrsave = mfspr(SPRN_VRSAVE);
1056#endif
1057#ifdef CONFIG_PPC_BOOK3S_64
1058	if (cpu_has_feature(CPU_FTR_DSCR))
1059		t->dscr = mfspr(SPRN_DSCR);
1060
1061	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
1062		t->bescr = mfspr(SPRN_BESCR);
1063		t->ebbhr = mfspr(SPRN_EBBHR);
1064		t->ebbrr = mfspr(SPRN_EBBRR);
1065
1066		t->fscr = mfspr(SPRN_FSCR);
1067
1068		/*
1069		 * Note that the TAR is not available for use in the kernel.
1070		 * (To provide this, the TAR should be backed up/restored on
1071		 * exception entry/exit instead, and be in pt_regs.  FIXME,
1072		 * this should be in pt_regs anyway (for debug).)
1073		 */
1074		t->tar = mfspr(SPRN_TAR);
1075	}
1076#endif
1077
1078	thread_pkey_regs_save(t);
1079}
1080
1081static inline void restore_sprs(struct thread_struct *old_thread,
1082				struct thread_struct *new_thread)
1083{
1084#ifdef CONFIG_ALTIVEC
1085	if (cpu_has_feature(CPU_FTR_ALTIVEC) &&
1086	    old_thread->vrsave != new_thread->vrsave)
1087		mtspr(SPRN_VRSAVE, new_thread->vrsave);
1088#endif
1089#ifdef CONFIG_PPC_BOOK3S_64
1090	if (cpu_has_feature(CPU_FTR_DSCR)) {
1091		u64 dscr = get_paca()->dscr_default;
1092		if (new_thread->dscr_inherit)
1093			dscr = new_thread->dscr;
1094
1095		if (old_thread->dscr != dscr)
1096			mtspr(SPRN_DSCR, dscr);
1097	}
1098
1099	if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
1100		if (old_thread->bescr != new_thread->bescr)
1101			mtspr(SPRN_BESCR, new_thread->bescr);
1102		if (old_thread->ebbhr != new_thread->ebbhr)
1103			mtspr(SPRN_EBBHR, new_thread->ebbhr);
1104		if (old_thread->ebbrr != new_thread->ebbrr)
1105			mtspr(SPRN_EBBRR, new_thread->ebbrr);
1106
1107		if (old_thread->fscr != new_thread->fscr)
1108			mtspr(SPRN_FSCR, new_thread->fscr);
1109
1110		if (old_thread->tar != new_thread->tar)
1111			mtspr(SPRN_TAR, new_thread->tar);
1112	}
1113
1114	if (cpu_has_feature(CPU_FTR_P9_TIDR) &&
1115	    old_thread->tidr != new_thread->tidr)
1116		mtspr(SPRN_TIDR, new_thread->tidr);
1117#endif
1118
1119	thread_pkey_regs_restore(new_thread, old_thread);
1120}
1121
1122struct task_struct *__switch_to(struct task_struct *prev,
1123	struct task_struct *new)
1124{
1125	struct thread_struct *new_thread, *old_thread;
1126	struct task_struct *last;
1127#ifdef CONFIG_PPC_BOOK3S_64
1128	struct ppc64_tlb_batch *batch;
1129#endif
1130
1131	new_thread = &new->thread;
1132	old_thread = &current->thread;
1133
1134	WARN_ON(!irqs_disabled());
1135
1136#ifdef CONFIG_PPC_BOOK3S_64
1137	batch = this_cpu_ptr(&ppc64_tlb_batch);
1138	if (batch->active) {
1139		current_thread_info()->local_flags |= _TLF_LAZY_MMU;
1140		if (batch->index)
1141			__flush_tlb_pending(batch);
1142		batch->active = 0;
1143	}
1144#endif /* CONFIG_PPC_BOOK3S_64 */
1145
1146#ifdef CONFIG_PPC_ADV_DEBUG_REGS
1147	switch_booke_debug_regs(&new->thread.debug);
1148#else
1149/*
1150 * For PPC_BOOK3S_64, we use the hw-breakpoint interfaces that would
1151 * schedule DABR
1152 */
1153#ifndef CONFIG_HAVE_HW_BREAKPOINT
1154	if (unlikely(!hw_brk_match(this_cpu_ptr(&current_brk), &new->thread.hw_brk)))
1155		__set_breakpoint(&new->thread.hw_brk);
1156#endif /* CONFIG_HAVE_HW_BREAKPOINT */
1157#endif
1158
1159	/*
1160	 * We need to save SPRs before treclaim/trecheckpoint as these will
1161	 * change a number of them.
1162	 */
1163	save_sprs(&prev->thread);
1164
1165	/* Save FPU, Altivec, VSX and SPE state */
1166	giveup_all(prev);
1167
1168	__switch_to_tm(prev, new);
1169
1170	if (!radix_enabled()) {
1171		/*
1172		 * We can't take a PMU exception inside _switch() since there
1173		 * is a window where the kernel stack SLB and the kernel stack
1174		 * are out of sync. Hard disable here.
1175		 */
1176		hard_irq_disable();
1177	}
1178
1179	/*
1180	 * Call restore_sprs() before calling _switch(). If we move it after
1181	 * _switch() then we miss out on calling it for new tasks. The reason
1182	 * for this is we manually create a stack frame for new tasks that
1183	 * directly returns through ret_from_fork() or
1184	 * ret_from_kernel_thread(). See copy_thread() for details.
1185	 */
1186	restore_sprs(old_thread, new_thread);
1187
1188	last = _switch(old_thread, new_thread);
1189
1190#ifdef CONFIG_PPC_BOOK3S_64
1191	if (current_thread_info()->local_flags & _TLF_LAZY_MMU) {
1192		current_thread_info()->local_flags &= ~_TLF_LAZY_MMU;
1193		batch = this_cpu_ptr(&ppc64_tlb_batch);
1194		batch->active = 1;
1195	}
1196
1197	if (current->thread.regs) {
1198		restore_math(current->thread.regs);
1199
1200		/*
1201		 * The copy-paste buffer can only store into foreign real
1202		 * addresses, so unprivileged processes can not see the
1203		 * data or use it in any way unless they have foreign real
1204		 * mappings. If the new process has the foreign real address
1205		 * mappings, we must issue a cp_abort to clear any state and
1206		 * prevent snooping, corruption or a covert channel.
1207		 */
1208		if (current->thread.used_vas)
 
1209			asm volatile(PPC_CP_ABORT);
1210	}
1211#endif /* CONFIG_PPC_BOOK3S_64 */
1212
1213	return last;
1214}
1215
1216#define NR_INSN_TO_PRINT	16
1217
1218static void show_instructions(struct pt_regs *regs)
1219{
1220	int i;
 
1221	unsigned long pc = regs->nip - (NR_INSN_TO_PRINT * 3 / 4 * sizeof(int));
1222
1223	printk("Instruction dump:");
1224
 
 
 
 
 
 
 
 
 
1225	for (i = 0; i < NR_INSN_TO_PRINT; i++) {
1226		int instr;
1227
1228		if (!(i % 8))
1229			pr_cont("\n");
1230
1231#if !defined(CONFIG_BOOKE)
1232		/* If executing with the IMMU off, adjust pc rather
1233		 * than print XXXXXXXX.
1234		 */
1235		if (!(regs->msr & MSR_IR))
1236			pc = (unsigned long)phys_to_virt(pc);
1237#endif
1238
1239		if (!__kernel_text_address(pc) ||
1240		    probe_kernel_address((const void *)pc, instr)) {
1241			pr_cont("XXXXXXXX ");
1242		} else {
1243			if (regs->nip == pc)
1244				pr_cont("<%08x> ", instr);
1245			else
1246				pr_cont("%08x ", instr);
1247		}
1248
1249		pc += sizeof(int);
1250	}
1251
1252	pr_cont("\n");
1253}
1254
1255void show_user_instructions(struct pt_regs *regs)
1256{
1257	unsigned long pc;
1258	int n = NR_INSN_TO_PRINT;
1259	struct seq_buf s;
1260	char buf[96]; /* enough for 8 times 9 + 2 chars */
1261
1262	pc = regs->nip - (NR_INSN_TO_PRINT * 3 / 4 * sizeof(int));
1263
1264	/*
1265	 * Make sure the NIP points at userspace, not kernel text/data or
1266	 * elsewhere.
1267	 */
1268	if (!__access_ok(pc, NR_INSN_TO_PRINT * sizeof(int), USER_DS)) {
1269		pr_info("%s[%d]: Bad NIP, not dumping instructions.\n",
1270			current->comm, current->pid);
1271		return;
1272	}
1273
1274	seq_buf_init(&s, buf, sizeof(buf));
1275
1276	while (n) {
1277		int i;
1278
1279		seq_buf_clear(&s);
1280
1281		for (i = 0; i < 8 && n; i++, n--, pc += sizeof(int)) {
1282			int instr;
1283
1284			if (probe_kernel_address((const void *)pc, instr)) {
 
1285				seq_buf_printf(&s, "XXXXXXXX ");
1286				continue;
1287			}
1288			seq_buf_printf(&s, regs->nip == pc ? "<%08x> " : "%08x ", instr);
1289		}
1290
1291		if (!seq_buf_has_overflowed(&s))
1292			pr_info("%s[%d]: code: %s\n", current->comm,
1293				current->pid, s.buffer);
1294	}
1295}
1296
1297struct regbit {
1298	unsigned long bit;
1299	const char *name;
1300};
1301
1302static struct regbit msr_bits[] = {
1303#if defined(CONFIG_PPC64) && !defined(CONFIG_BOOKE)
1304	{MSR_SF,	"SF"},
1305	{MSR_HV,	"HV"},
1306#endif
1307	{MSR_VEC,	"VEC"},
1308	{MSR_VSX,	"VSX"},
1309#ifdef CONFIG_BOOKE
1310	{MSR_CE,	"CE"},
1311#endif
1312	{MSR_EE,	"EE"},
1313	{MSR_PR,	"PR"},
1314	{MSR_FP,	"FP"},
1315	{MSR_ME,	"ME"},
1316#ifdef CONFIG_BOOKE
1317	{MSR_DE,	"DE"},
1318#else
1319	{MSR_SE,	"SE"},
1320	{MSR_BE,	"BE"},
1321#endif
1322	{MSR_IR,	"IR"},
1323	{MSR_DR,	"DR"},
1324	{MSR_PMM,	"PMM"},
1325#ifndef CONFIG_BOOKE
1326	{MSR_RI,	"RI"},
1327	{MSR_LE,	"LE"},
1328#endif
1329	{0,		NULL}
1330};
1331
1332static void print_bits(unsigned long val, struct regbit *bits, const char *sep)
1333{
1334	const char *s = "";
1335
1336	for (; bits->bit; ++bits)
1337		if (val & bits->bit) {
1338			pr_cont("%s%s", s, bits->name);
1339			s = sep;
1340		}
1341}
1342
1343#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1344static struct regbit msr_tm_bits[] = {
1345	{MSR_TS_T,	"T"},
1346	{MSR_TS_S,	"S"},
1347	{MSR_TM,	"E"},
1348	{0,		NULL}
1349};
1350
1351static void print_tm_bits(unsigned long val)
1352{
1353/*
1354 * This only prints something if at least one of the TM bit is set.
1355 * Inside the TM[], the output means:
1356 *   E: Enabled		(bit 32)
1357 *   S: Suspended	(bit 33)
1358 *   T: Transactional	(bit 34)
1359 */
1360	if (val & (MSR_TM | MSR_TS_S | MSR_TS_T)) {
1361		pr_cont(",TM[");
1362		print_bits(val, msr_tm_bits, "");
1363		pr_cont("]");
1364	}
1365}
1366#else
1367static void print_tm_bits(unsigned long val) {}
1368#endif
1369
1370static void print_msr_bits(unsigned long val)
1371{
1372	pr_cont("<");
1373	print_bits(val, msr_bits, ",");
1374	print_tm_bits(val);
1375	pr_cont(">");
1376}
1377
1378#ifdef CONFIG_PPC64
1379#define REG		"%016lx"
1380#define REGS_PER_LINE	4
1381#define LAST_VOLATILE	13
1382#else
1383#define REG		"%08lx"
1384#define REGS_PER_LINE	8
1385#define LAST_VOLATILE	12
1386#endif
1387
1388void show_regs(struct pt_regs * regs)
1389{
1390	int i, trap;
1391
1392	show_regs_print_info(KERN_DEFAULT);
1393
1394	printk("NIP:  "REG" LR: "REG" CTR: "REG"\n",
1395	       regs->nip, regs->link, regs->ctr);
1396	printk("REGS: %px TRAP: %04lx   %s  (%s)\n",
1397	       regs, regs->trap, print_tainted(), init_utsname()->release);
1398	printk("MSR:  "REG" ", regs->msr);
1399	print_msr_bits(regs->msr);
1400	pr_cont("  CR: %08lx  XER: %08lx\n", regs->ccr, regs->xer);
1401	trap = TRAP(regs);
1402	if ((TRAP(regs) != 0xc00) && cpu_has_feature(CPU_FTR_CFAR))
1403		pr_cont("CFAR: "REG" ", regs->orig_gpr3);
1404	if (trap == 0x200 || trap == 0x300 || trap == 0x600)
1405#if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
1406		pr_cont("DEAR: "REG" ESR: "REG" ", regs->dar, regs->dsisr);
1407#else
1408		pr_cont("DAR: "REG" DSISR: %08lx ", regs->dar, regs->dsisr);
1409#endif
1410#ifdef CONFIG_PPC64
1411	pr_cont("IRQMASK: %lx ", regs->softe);
1412#endif
1413#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1414	if (MSR_TM_ACTIVE(regs->msr))
1415		pr_cont("\nPACATMSCRATCH: %016llx ", get_paca()->tm_scratch);
1416#endif
1417
1418	for (i = 0;  i < 32;  i++) {
1419		if ((i % REGS_PER_LINE) == 0)
1420			pr_cont("\nGPR%02d: ", i);
1421		pr_cont(REG " ", regs->gpr[i]);
1422		if (i == LAST_VOLATILE && !FULL_REGS(regs))
1423			break;
1424	}
1425	pr_cont("\n");
1426#ifdef CONFIG_KALLSYMS
1427	/*
1428	 * Lookup NIP late so we have the best change of getting the
1429	 * above info out without failing
1430	 */
1431	printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
1432	printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
1433#endif
1434	show_stack(current, (unsigned long *) regs->gpr[1]);
1435	if (!user_mode(regs))
1436		show_instructions(regs);
1437}
1438
1439void flush_thread(void)
1440{
1441#ifdef CONFIG_HAVE_HW_BREAKPOINT
1442	flush_ptrace_hw_breakpoint(current);
1443#else /* CONFIG_HAVE_HW_BREAKPOINT */
1444	set_debug_reg_defaults(&current->thread);
1445#endif /* CONFIG_HAVE_HW_BREAKPOINT */
1446}
1447
1448#ifdef CONFIG_PPC_BOOK3S_64
1449void arch_setup_new_exec(void)
1450{
1451	if (radix_enabled())
1452		return;
1453	hash__setup_new_exec();
1454}
1455#endif
1456
1457int set_thread_uses_vas(void)
1458{
1459#ifdef CONFIG_PPC_BOOK3S_64
1460	if (!cpu_has_feature(CPU_FTR_ARCH_300))
1461		return -EINVAL;
1462
1463	current->thread.used_vas = 1;
1464
1465	/*
1466	 * Even a process that has no foreign real address mapping can use
1467	 * an unpaired COPY instruction (to no real effect). Issue CP_ABORT
1468	 * to clear any pending COPY and prevent a covert channel.
1469	 *
1470	 * __switch_to() will issue CP_ABORT on future context switches.
1471	 */
1472	asm volatile(PPC_CP_ABORT);
1473
1474#endif /* CONFIG_PPC_BOOK3S_64 */
1475	return 0;
1476}
1477
1478#ifdef CONFIG_PPC64
1479/**
1480 * Assign a TIDR (thread ID) for task @t and set it in the thread
1481 * structure. For now, we only support setting TIDR for 'current' task.
1482 *
1483 * Since the TID value is a truncated form of it PID, it is possible
1484 * (but unlikely) for 2 threads to have the same TID. In the unlikely event
1485 * that 2 threads share the same TID and are waiting, one of the following
1486 * cases will happen:
1487 *
1488 * 1. The correct thread is running, the wrong thread is not
1489 * In this situation, the correct thread is woken and proceeds to pass it's
1490 * condition check.
1491 *
1492 * 2. Neither threads are running
1493 * In this situation, neither thread will be woken. When scheduled, the waiting
1494 * threads will execute either a wait, which will return immediately, followed
1495 * by a condition check, which will pass for the correct thread and fail
1496 * for the wrong thread, or they will execute the condition check immediately.
1497 *
1498 * 3. The wrong thread is running, the correct thread is not
1499 * The wrong thread will be woken, but will fail it's condition check and
1500 * re-execute wait. The correct thread, when scheduled, will execute either
1501 * it's condition check (which will pass), or wait, which returns immediately
1502 * when called the first time after the thread is scheduled, followed by it's
1503 * condition check (which will pass).
1504 *
1505 * 4. Both threads are running
1506 * Both threads will be woken. The wrong thread will fail it's condition check
1507 * and execute another wait, while the correct thread will pass it's condition
1508 * check.
1509 *
1510 * @t: the task to set the thread ID for
1511 */
1512int set_thread_tidr(struct task_struct *t)
1513{
1514	if (!cpu_has_feature(CPU_FTR_P9_TIDR))
1515		return -EINVAL;
1516
1517	if (t != current)
1518		return -EINVAL;
1519
1520	if (t->thread.tidr)
1521		return 0;
1522
1523	t->thread.tidr = (u16)task_pid_nr(t);
1524	mtspr(SPRN_TIDR, t->thread.tidr);
1525
1526	return 0;
1527}
1528EXPORT_SYMBOL_GPL(set_thread_tidr);
1529
1530#endif /* CONFIG_PPC64 */
1531
1532void
1533release_thread(struct task_struct *t)
1534{
1535}
1536
1537/*
1538 * this gets called so that we can store coprocessor state into memory and
1539 * copy the current task into the new thread.
1540 */
1541int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
1542{
1543	flush_all_to_thread(src);
1544	/*
1545	 * Flush TM state out so we can copy it.  __switch_to_tm() does this
1546	 * flush but it removes the checkpointed state from the current CPU and
1547	 * transitions the CPU out of TM mode.  Hence we need to call
1548	 * tm_recheckpoint_new_task() (on the same task) to restore the
1549	 * checkpointed state back and the TM mode.
1550	 *
1551	 * Can't pass dst because it isn't ready. Doesn't matter, passing
1552	 * dst is only important for __switch_to()
1553	 */
1554	__switch_to_tm(src, src);
1555
1556	*dst = *src;
1557
1558	clear_task_ebb(dst);
1559
1560	return 0;
1561}
1562
1563static void setup_ksp_vsid(struct task_struct *p, unsigned long sp)
1564{
1565#ifdef CONFIG_PPC_BOOK3S_64
1566	unsigned long sp_vsid;
1567	unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
1568
1569	if (radix_enabled())
1570		return;
1571
1572	if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
1573		sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
1574			<< SLB_VSID_SHIFT_1T;
1575	else
1576		sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
1577			<< SLB_VSID_SHIFT;
1578	sp_vsid |= SLB_VSID_KERNEL | llp;
1579	p->thread.ksp_vsid = sp_vsid;
1580#endif
1581}
1582
1583/*
1584 * Copy a thread..
1585 */
1586
1587/*
1588 * Copy architecture-specific thread state
1589 */
1590int copy_thread_tls(unsigned long clone_flags, unsigned long usp,
1591		unsigned long kthread_arg, struct task_struct *p,
1592		unsigned long tls)
1593{
1594	struct pt_regs *childregs, *kregs;
1595	extern void ret_from_fork(void);
 
1596	extern void ret_from_kernel_thread(void);
1597	void (*f)(void);
1598	unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
1599	struct thread_info *ti = task_thread_info(p);
 
 
 
1600
1601	klp_init_thread_info(p);
1602
1603	/* Copy registers */
1604	sp -= sizeof(struct pt_regs);
1605	childregs = (struct pt_regs *) sp;
1606	if (unlikely(p->flags & PF_KTHREAD)) {
1607		/* kernel thread */
1608		memset(childregs, 0, sizeof(struct pt_regs));
1609		childregs->gpr[1] = sp + sizeof(struct pt_regs);
1610		/* function */
1611		if (usp)
1612			childregs->gpr[14] = ppc_function_entry((void *)usp);
1613#ifdef CONFIG_PPC64
1614		clear_tsk_thread_flag(p, TIF_32BIT);
1615		childregs->softe = IRQS_ENABLED;
1616#endif
1617		childregs->gpr[15] = kthread_arg;
1618		p->thread.regs = NULL;	/* no user register state */
1619		ti->flags |= _TIF_RESTOREALL;
1620		f = ret_from_kernel_thread;
1621	} else {
1622		/* user thread */
1623		struct pt_regs *regs = current_pt_regs();
1624		CHECK_FULL_REGS(regs);
1625		*childregs = *regs;
1626		if (usp)
1627			childregs->gpr[1] = usp;
1628		p->thread.regs = childregs;
1629		childregs->gpr[3] = 0;  /* Result from fork() */
 
 
1630		if (clone_flags & CLONE_SETTLS) {
1631#ifdef CONFIG_PPC64
1632			if (!is_32bit_task())
1633				childregs->gpr[13] = tls;
1634			else
1635#endif
1636				childregs->gpr[2] = tls;
1637		}
1638
1639		f = ret_from_fork;
 
 
 
1640	}
1641	childregs->msr &= ~(MSR_FP|MSR_VEC|MSR_VSX);
1642	sp -= STACK_FRAME_OVERHEAD;
1643
1644	/*
1645	 * The way this works is that at some point in the future
1646	 * some task will call _switch to switch to the new task.
1647	 * That will pop off the stack frame created below and start
1648	 * the new task running at ret_from_fork.  The new task will
1649	 * do some house keeping and then return from the fork or clone
1650	 * system call, using the stack frame created above.
1651	 */
1652	((unsigned long *)sp)[0] = 0;
1653	sp -= sizeof(struct pt_regs);
1654	kregs = (struct pt_regs *) sp;
1655	sp -= STACK_FRAME_OVERHEAD;
1656	p->thread.ksp = sp;
1657#ifdef CONFIG_PPC32
1658	p->thread.ksp_limit = (unsigned long)end_of_stack(p);
1659#endif
1660#ifdef CONFIG_HAVE_HW_BREAKPOINT
1661	p->thread.ptrace_bps[0] = NULL;
 
1662#endif
1663
1664	p->thread.fp_save_area = NULL;
1665#ifdef CONFIG_ALTIVEC
1666	p->thread.vr_save_area = NULL;
1667#endif
1668
1669	setup_ksp_vsid(p, sp);
1670
1671#ifdef CONFIG_PPC64 
1672	if (cpu_has_feature(CPU_FTR_DSCR)) {
1673		p->thread.dscr_inherit = current->thread.dscr_inherit;
1674		p->thread.dscr = mfspr(SPRN_DSCR);
1675	}
1676	if (cpu_has_feature(CPU_FTR_HAS_PPR))
1677		childregs->ppr = DEFAULT_PPR;
1678
1679	p->thread.tidr = 0;
1680#endif
1681	kregs->nip = ppc_function_entry(f);
1682	return 0;
1683}
1684
1685void preload_new_slb_context(unsigned long start, unsigned long sp);
1686
1687/*
1688 * Set up a thread for executing a new program
1689 */
1690void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
1691{
1692#ifdef CONFIG_PPC64
1693	unsigned long load_addr = regs->gpr[2];	/* saved by ELF_PLAT_INIT */
1694
1695#ifdef CONFIG_PPC_BOOK3S_64
1696	if (!radix_enabled())
1697		preload_new_slb_context(start, sp);
1698#endif
1699#endif
1700
1701	/*
1702	 * If we exec out of a kernel thread then thread.regs will not be
1703	 * set.  Do it now.
1704	 */
1705	if (!current->thread.regs) {
1706		struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
1707		current->thread.regs = regs - 1;
1708	}
1709
1710#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1711	/*
1712	 * Clear any transactional state, we're exec()ing. The cause is
1713	 * not important as there will never be a recheckpoint so it's not
1714	 * user visible.
1715	 */
1716	if (MSR_TM_SUSPENDED(mfmsr()))
1717		tm_reclaim_current(0);
1718#endif
1719
1720	memset(regs->gpr, 0, sizeof(regs->gpr));
1721	regs->ctr = 0;
1722	regs->link = 0;
1723	regs->xer = 0;
1724	regs->ccr = 0;
1725	regs->gpr[1] = sp;
1726
1727	/*
1728	 * We have just cleared all the nonvolatile GPRs, so make
1729	 * FULL_REGS(regs) return true.  This is necessary to allow
1730	 * ptrace to examine the thread immediately after exec.
1731	 */
1732	regs->trap &= ~1UL;
1733
1734#ifdef CONFIG_PPC32
1735	regs->mq = 0;
1736	regs->nip = start;
1737	regs->msr = MSR_USER;
1738#else
1739	if (!is_32bit_task()) {
1740		unsigned long entry;
1741
1742		if (is_elf2_task()) {
1743			/* Look ma, no function descriptors! */
1744			entry = start;
1745
1746			/*
1747			 * Ulrich says:
1748			 *   The latest iteration of the ABI requires that when
1749			 *   calling a function (at its global entry point),
1750			 *   the caller must ensure r12 holds the entry point
1751			 *   address (so that the function can quickly
1752			 *   establish addressability).
1753			 */
1754			regs->gpr[12] = start;
1755			/* Make sure that's restored on entry to userspace. */
1756			set_thread_flag(TIF_RESTOREALL);
1757		} else {
1758			unsigned long toc;
1759
1760			/* start is a relocated pointer to the function
1761			 * descriptor for the elf _start routine.  The first
1762			 * entry in the function descriptor is the entry
1763			 * address of _start and the second entry is the TOC
1764			 * value we need to use.
1765			 */
1766			__get_user(entry, (unsigned long __user *)start);
1767			__get_user(toc, (unsigned long __user *)start+1);
1768
1769			/* Check whether the e_entry function descriptor entries
1770			 * need to be relocated before we can use them.
1771			 */
1772			if (load_addr != 0) {
1773				entry += load_addr;
1774				toc   += load_addr;
1775			}
1776			regs->gpr[2] = toc;
1777		}
1778		regs->nip = entry;
1779		regs->msr = MSR_USER64;
1780	} else {
1781		regs->nip = start;
1782		regs->gpr[2] = 0;
1783		regs->msr = MSR_USER32;
1784	}
1785#endif
1786#ifdef CONFIG_VSX
1787	current->thread.used_vsr = 0;
1788#endif
1789	current->thread.load_slb = 0;
1790	current->thread.load_fp = 0;
1791	memset(&current->thread.fp_state, 0, sizeof(current->thread.fp_state));
1792	current->thread.fp_save_area = NULL;
1793#ifdef CONFIG_ALTIVEC
1794	memset(&current->thread.vr_state, 0, sizeof(current->thread.vr_state));
1795	current->thread.vr_state.vscr.u[3] = 0x00010000; /* Java mode disabled */
1796	current->thread.vr_save_area = NULL;
1797	current->thread.vrsave = 0;
1798	current->thread.used_vr = 0;
1799	current->thread.load_vec = 0;
1800#endif /* CONFIG_ALTIVEC */
1801#ifdef CONFIG_SPE
1802	memset(current->thread.evr, 0, sizeof(current->thread.evr));
1803	current->thread.acc = 0;
1804	current->thread.spefscr = 0;
1805	current->thread.used_spe = 0;
1806#endif /* CONFIG_SPE */
1807#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1808	current->thread.tm_tfhar = 0;
1809	current->thread.tm_texasr = 0;
1810	current->thread.tm_tfiar = 0;
1811	current->thread.load_tm = 0;
1812#endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
1813
1814	thread_pkey_regs_init(&current->thread);
1815}
1816EXPORT_SYMBOL(start_thread);
1817
1818#define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
1819		| PR_FP_EXC_RES | PR_FP_EXC_INV)
1820
1821int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
1822{
1823	struct pt_regs *regs = tsk->thread.regs;
1824
1825	/* This is a bit hairy.  If we are an SPE enabled  processor
1826	 * (have embedded fp) we store the IEEE exception enable flags in
1827	 * fpexc_mode.  fpexc_mode is also used for setting FP exception
1828	 * mode (asyn, precise, disabled) for 'Classic' FP. */
1829	if (val & PR_FP_EXC_SW_ENABLE) {
1830#ifdef CONFIG_SPE
1831		if (cpu_has_feature(CPU_FTR_SPE)) {
1832			/*
1833			 * When the sticky exception bits are set
1834			 * directly by userspace, it must call prctl
1835			 * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
1836			 * in the existing prctl settings) or
1837			 * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
1838			 * the bits being set).  <fenv.h> functions
1839			 * saving and restoring the whole
1840			 * floating-point environment need to do so
1841			 * anyway to restore the prctl settings from
1842			 * the saved environment.
1843			 */
1844			tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
1845			tsk->thread.fpexc_mode = val &
1846				(PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
1847			return 0;
1848		} else {
1849			return -EINVAL;
1850		}
1851#else
1852		return -EINVAL;
1853#endif
1854	}
1855
1856	/* on a CONFIG_SPE this does not hurt us.  The bits that
1857	 * __pack_fe01 use do not overlap with bits used for
1858	 * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
1859	 * on CONFIG_SPE implementations are reserved so writing to
1860	 * them does not change anything */
1861	if (val > PR_FP_EXC_PRECISE)
1862		return -EINVAL;
1863	tsk->thread.fpexc_mode = __pack_fe01(val);
1864	if (regs != NULL && (regs->msr & MSR_FP) != 0)
1865		regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
1866			| tsk->thread.fpexc_mode;
1867	return 0;
1868}
1869
1870int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
1871{
1872	unsigned int val;
1873
1874	if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
1875#ifdef CONFIG_SPE
1876		if (cpu_has_feature(CPU_FTR_SPE)) {
1877			/*
1878			 * When the sticky exception bits are set
1879			 * directly by userspace, it must call prctl
1880			 * with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
1881			 * in the existing prctl settings) or
1882			 * PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
1883			 * the bits being set).  <fenv.h> functions
1884			 * saving and restoring the whole
1885			 * floating-point environment need to do so
1886			 * anyway to restore the prctl settings from
1887			 * the saved environment.
1888			 */
1889			tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
1890			val = tsk->thread.fpexc_mode;
1891		} else
1892			return -EINVAL;
1893#else
1894		return -EINVAL;
1895#endif
1896	else
1897		val = __unpack_fe01(tsk->thread.fpexc_mode);
1898	return put_user(val, (unsigned int __user *) adr);
1899}
1900
1901int set_endian(struct task_struct *tsk, unsigned int val)
1902{
1903	struct pt_regs *regs = tsk->thread.regs;
1904
1905	if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
1906	    (val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
1907		return -EINVAL;
1908
1909	if (regs == NULL)
1910		return -EINVAL;
1911
1912	if (val == PR_ENDIAN_BIG)
1913		regs->msr &= ~MSR_LE;
1914	else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
1915		regs->msr |= MSR_LE;
1916	else
1917		return -EINVAL;
1918
1919	return 0;
1920}
1921
1922int get_endian(struct task_struct *tsk, unsigned long adr)
1923{
1924	struct pt_regs *regs = tsk->thread.regs;
1925	unsigned int val;
1926
1927	if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
1928	    !cpu_has_feature(CPU_FTR_REAL_LE))
1929		return -EINVAL;
1930
1931	if (regs == NULL)
1932		return -EINVAL;
1933
1934	if (regs->msr & MSR_LE) {
1935		if (cpu_has_feature(CPU_FTR_REAL_LE))
1936			val = PR_ENDIAN_LITTLE;
1937		else
1938			val = PR_ENDIAN_PPC_LITTLE;
1939	} else
1940		val = PR_ENDIAN_BIG;
1941
1942	return put_user(val, (unsigned int __user *)adr);
1943}
1944
1945int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
1946{
1947	tsk->thread.align_ctl = val;
1948	return 0;
1949}
1950
1951int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
1952{
1953	return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
1954}
1955
1956static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
1957				  unsigned long nbytes)
1958{
1959	unsigned long stack_page;
1960	unsigned long cpu = task_cpu(p);
1961
1962	stack_page = (unsigned long)hardirq_ctx[cpu];
1963	if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
1964		return 1;
1965
1966	stack_page = (unsigned long)softirq_ctx[cpu];
1967	if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
1968		return 1;
1969
1970	return 0;
1971}
1972
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1973int validate_sp(unsigned long sp, struct task_struct *p,
1974		       unsigned long nbytes)
1975{
1976	unsigned long stack_page = (unsigned long)task_stack_page(p);
1977
1978	if (sp < THREAD_SIZE)
1979		return 0;
1980
1981	if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
1982		return 1;
1983
1984	return valid_irq_stack(sp, p, nbytes);
 
 
 
1985}
1986
1987EXPORT_SYMBOL(validate_sp);
1988
1989static unsigned long __get_wchan(struct task_struct *p)
1990{
1991	unsigned long ip, sp;
1992	int count = 0;
1993
1994	if (!p || p == current || p->state == TASK_RUNNING)
1995		return 0;
1996
1997	sp = p->thread.ksp;
1998	if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
1999		return 0;
2000
2001	do {
2002		sp = *(unsigned long *)sp;
2003		if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD) ||
2004		    p->state == TASK_RUNNING)
2005			return 0;
2006		if (count > 0) {
2007			ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
2008			if (!in_sched_functions(ip))
2009				return ip;
2010		}
2011	} while (count++ < 16);
2012	return 0;
2013}
2014
2015unsigned long get_wchan(struct task_struct *p)
2016{
2017	unsigned long ret;
2018
2019	if (!try_get_task_stack(p))
2020		return 0;
2021
2022	ret = __get_wchan(p);
2023
2024	put_task_stack(p);
2025
2026	return ret;
2027}
2028
2029static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
2030
2031void show_stack(struct task_struct *tsk, unsigned long *stack)
 
2032{
2033	unsigned long sp, ip, lr, newsp;
2034	int count = 0;
2035	int firstframe = 1;
2036#ifdef CONFIG_FUNCTION_GRAPH_TRACER
2037	unsigned long ret_addr;
2038	int ftrace_idx = 0;
2039#endif
2040
2041	if (tsk == NULL)
2042		tsk = current;
2043
2044	if (!try_get_task_stack(tsk))
2045		return;
2046
2047	sp = (unsigned long) stack;
2048	if (sp == 0) {
2049		if (tsk == current)
2050			sp = current_stack_pointer();
2051		else
2052			sp = tsk->thread.ksp;
2053	}
2054
2055	lr = 0;
2056	printk("Call Trace:\n");
2057	do {
2058		if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
2059			break;
2060
2061		stack = (unsigned long *) sp;
2062		newsp = stack[0];
2063		ip = stack[STACK_FRAME_LR_SAVE];
2064		if (!firstframe || ip != lr) {
2065			printk("["REG"] ["REG"] %pS", sp, ip, (void *)ip);
 
2066#ifdef CONFIG_FUNCTION_GRAPH_TRACER
2067			ret_addr = ftrace_graph_ret_addr(current,
2068						&ftrace_idx, ip, stack);
2069			if (ret_addr != ip)
2070				pr_cont(" (%pS)", (void *)ret_addr);
2071#endif
2072			if (firstframe)
2073				pr_cont(" (unreliable)");
2074			pr_cont("\n");
2075		}
2076		firstframe = 0;
2077
2078		/*
2079		 * See if this is an exception frame.
2080		 * We look for the "regshere" marker in the current frame.
2081		 */
2082		if (validate_sp(sp, tsk, STACK_INT_FRAME_SIZE)
2083		    && stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
2084			struct pt_regs *regs = (struct pt_regs *)
2085				(sp + STACK_FRAME_OVERHEAD);
2086			lr = regs->link;
2087			printk("--- interrupt: %lx at %pS\n    LR = %pS\n",
2088			       regs->trap, (void *)regs->nip, (void *)lr);
 
2089			firstframe = 1;
2090		}
2091
2092		sp = newsp;
2093	} while (count++ < kstack_depth_to_print);
2094
2095	put_task_stack(tsk);
2096}
2097
2098#ifdef CONFIG_PPC64
2099/* Called with hard IRQs off */
2100void notrace __ppc64_runlatch_on(void)
2101{
2102	struct thread_info *ti = current_thread_info();
2103
2104	if (cpu_has_feature(CPU_FTR_ARCH_206)) {
2105		/*
2106		 * Least significant bit (RUN) is the only writable bit of
2107		 * the CTRL register, so we can avoid mfspr. 2.06 is not the
2108		 * earliest ISA where this is the case, but it's convenient.
2109		 */
2110		mtspr(SPRN_CTRLT, CTRL_RUNLATCH);
2111	} else {
2112		unsigned long ctrl;
2113
2114		/*
2115		 * Some architectures (e.g., Cell) have writable fields other
2116		 * than RUN, so do the read-modify-write.
2117		 */
2118		ctrl = mfspr(SPRN_CTRLF);
2119		ctrl |= CTRL_RUNLATCH;
2120		mtspr(SPRN_CTRLT, ctrl);
2121	}
2122
2123	ti->local_flags |= _TLF_RUNLATCH;
2124}
2125
2126/* Called with hard IRQs off */
2127void notrace __ppc64_runlatch_off(void)
2128{
2129	struct thread_info *ti = current_thread_info();
2130
2131	ti->local_flags &= ~_TLF_RUNLATCH;
2132
2133	if (cpu_has_feature(CPU_FTR_ARCH_206)) {
2134		mtspr(SPRN_CTRLT, 0);
2135	} else {
2136		unsigned long ctrl;
2137
2138		ctrl = mfspr(SPRN_CTRLF);
2139		ctrl &= ~CTRL_RUNLATCH;
2140		mtspr(SPRN_CTRLT, ctrl);
2141	}
2142}
2143#endif /* CONFIG_PPC64 */
2144
2145unsigned long arch_align_stack(unsigned long sp)
2146{
2147	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
2148		sp -= get_random_int() & ~PAGE_MASK;
2149	return sp & ~0xf;
2150}
2151
2152static inline unsigned long brk_rnd(void)
2153{
2154        unsigned long rnd = 0;
2155
2156	/* 8MB for 32bit, 1GB for 64bit */
2157	if (is_32bit_task())
2158		rnd = (get_random_long() % (1UL<<(23-PAGE_SHIFT)));
2159	else
2160		rnd = (get_random_long() % (1UL<<(30-PAGE_SHIFT)));
2161
2162	return rnd << PAGE_SHIFT;
2163}
2164
2165unsigned long arch_randomize_brk(struct mm_struct *mm)
2166{
2167	unsigned long base = mm->brk;
2168	unsigned long ret;
2169
2170#ifdef CONFIG_PPC_BOOK3S_64
2171	/*
2172	 * If we are using 1TB segments and we are allowed to randomise
2173	 * the heap, we can put it above 1TB so it is backed by a 1TB
2174	 * segment. Otherwise the heap will be in the bottom 1TB
2175	 * which always uses 256MB segments and this may result in a
2176	 * performance penalty. We don't need to worry about radix. For
2177	 * radix, mmu_highuser_ssize remains unchanged from 256MB.
2178	 */
2179	if (!is_32bit_task() && (mmu_highuser_ssize == MMU_SEGSIZE_1T))
2180		base = max_t(unsigned long, mm->brk, 1UL << SID_SHIFT_1T);
2181#endif
2182
2183	ret = PAGE_ALIGN(base + brk_rnd());
2184
2185	if (ret < mm->brk)
2186		return mm->brk;
2187
2188	return ret;
2189}
2190