3 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
4 * Rewritten by Cort Dougan (cort@cs.nmt.edu) for PReP
5 * Copyright (C) 1996 Cort Dougan <cort@cs.nmt.edu>
6 * Adapted for Power Macintosh by Paul Mackerras.
7 * Low-level exception handlers and MMU support
8 * rewritten by Paul Mackerras.
9 * Copyright (C) 1996 Paul Mackerras.
10 * MPC8xx modifications Copyright (C) 1997 Dan Malek (dmalek@jlc.net).
12 * This file contains the system call entry code, context switch
13 * code, and exception/interrupt return code for PowerPC.
15 * This program is free software; you can redistribute it and/or
16 * modify it under the terms of the GNU General Public License
17 * as published by the Free Software Foundation; either version
18 * 2 of the License, or (at your option) any later version.
21 #include <linux/errno.h>
22 #include <asm/unistd.h>
23 #include <asm/processor.h>
26 #include <asm/thread_info.h>
27 #include <asm/ppc_asm.h>
28 #include <asm/asm-offsets.h>
29 #include <asm/cputable.h>
30 #include <asm/firmware.h>
32 #include <asm/ptrace.h>
33 #include <asm/irqflags.h>
34 #include <asm/ftrace.h>
41 .tc .sys_call_table[TC],.sys_call_table
43 /* This value is used to mark exception frames on the stack. */
45 .tc ID_EXC_MARKER[TC],STACK_FRAME_REGS_MARKER
52 .globl system_call_common
56 addi r1,r1,-INT_FRAME_SIZE
64 ACCOUNT_CPU_USER_ENTRY(r10, r11)
66 * This "crclr so" clears CR0.SO, which is the error indication on
67 * return from this system call. There must be no cmp instruction
68 * between it and the "mfcr r9" below, otherwise if XER.SO is set,
69 * CR0.SO will get set, causing all system calls to appear to fail.
97 addi r9,r1,STACK_FRAME_OVERHEAD
98 ld r11,exception_marker@toc(r2)
99 std r11,-16(r9) /* "regshere" marker */
100 #if defined(CONFIG_VIRT_CPU_ACCOUNTING) && defined(CONFIG_PPC_SPLPAR)
103 /* if from user, see if there are any DTL entries to process */
104 ld r10,PACALPPACAPTR(r13) /* get ptr to VPA */
105 ld r11,PACA_DTL_RIDX(r13) /* get log read index */
106 ld r10,LPPACA_DTLIDX(r10) /* get log write index */
109 bl .accumulate_stolen_time
113 addi r9,r1,STACK_FRAME_OVERHEAD
115 END_FW_FTR_SECTION_IFSET(FW_FEATURE_SPLPAR)
116 #endif /* CONFIG_VIRT_CPU_ACCOUNTING && CONFIG_PPC_SPLPAR */
118 #ifdef CONFIG_TRACE_IRQFLAGS
119 bl .trace_hardirqs_on
123 addi r9,r1,STACK_FRAME_OVERHEAD
125 #endif /* CONFIG_TRACE_IRQFLAGS */
127 stb r10,PACASOFTIRQEN(r13)
128 stb r10,PACAHARDIRQEN(r13)
130 #ifdef CONFIG_PPC_ISERIES
132 /* Hack for handling interrupts when soft-enabling on iSeries */
133 cmpdi cr1,r0,0x5555 /* syscall 0x5555 */
134 andi. r10,r12,MSR_PR /* from kernel */
135 crand 4*cr0+eq,4*cr1+eq,4*cr0+eq
137 b hardware_interrupt_entry
139 END_FW_FTR_SECTION_IFSET(FW_FEATURE_ISERIES)
140 #endif /* CONFIG_PPC_ISERIES */
142 /* Hard enable interrupts */
143 #ifdef CONFIG_PPC_BOOK3E
149 #endif /* CONFIG_PPC_BOOK3E */
156 addi r9,r1,STACK_FRAME_OVERHEAD
158 clrrdi r11,r1,THREAD_SHIFT
160 andi. r11,r10,_TIF_SYSCALL_T_OR_A
162 syscall_dotrace_cont:
163 cmpldi 0,r0,NR_syscalls
166 system_call: /* label this so stack traces look sane */
168 * Need to vector to 32 Bit or default sys_call_table here,
169 * based on caller's run-mode / personality.
171 ld r11,.SYS_CALL_TABLE@toc(2)
172 andi. r10,r10,_TIF_32BIT
174 addi r11,r11,8 /* use 32-bit syscall entries */
183 ldx r10,r11,r0 /* Fetch system call handler [ptr] */
185 bctrl /* Call handler */
190 bl .do_show_syscall_exit
193 clrrdi r12,r1,THREAD_SHIFT
196 #ifdef CONFIG_PPC_BOOK3S
197 /* No MSR:RI on BookE */
202 /* Disable interrupts so current_thread_info()->flags can't change,
203 * and so that we don't get interrupted after loading SRR0/1.
205 #ifdef CONFIG_PPC_BOOK3E
212 #endif /* CONFIG_PPC_BOOK3E */
216 andi. r0,r9,(_TIF_SYSCALL_T_OR_A|_TIF_SINGLESTEP|_TIF_USER_WORK_MASK|_TIF_PERSYSCALL_MASK)
217 bne- syscall_exit_work
224 stdcx. r0,0,r1 /* to clear the reservation */
225 END_FTR_SECTION_IFCLR(CPU_FTR_STCX_CHECKS_ADDRESS)
229 * Clear RI before restoring r13. If we are returning to
230 * userspace and we take an exception after restoring r13,
231 * we end up corrupting the userspace r13 value.
233 #ifdef CONFIG_PPC_BOOK3S
234 /* No MSR:RI on BookE */
237 mtmsrd r11,1 /* clear MSR.RI */
238 #endif /* CONFIG_PPC_BOOK3S */
241 ACCOUNT_CPU_USER_EXIT(r11, r12)
242 ld r13,GPR13(r1) /* only restore r13 if returning to usermode */
250 b . /* prevent speculative execution */
253 oris r5,r5,0x1000 /* Set SO bit in CR */
258 /* Traced system call support */
261 addi r3,r1,STACK_FRAME_OVERHEAD
262 bl .do_syscall_trace_enter
264 * Restore argument registers possibly just changed.
265 * We use the return value of do_syscall_trace_enter
266 * for the call number to look up in the table (r0).
275 addi r9,r1,STACK_FRAME_OVERHEAD
276 clrrdi r10,r1,THREAD_SHIFT
278 b syscall_dotrace_cont
285 /* If TIF_RESTOREALL is set, don't scribble on either r3 or ccr.
286 If TIF_NOERROR is set, just save r3 as it is. */
288 andi. r0,r9,_TIF_RESTOREALL
292 0: cmpld r3,r11 /* r10 is -LAST_ERRNO */
294 andi. r0,r9,_TIF_NOERROR
298 oris r5,r5,0x1000 /* Set SO bit in CR */
301 2: andi. r0,r9,(_TIF_PERSYSCALL_MASK)
304 /* Clear per-syscall TIF flags if any are set. */
306 li r11,_TIF_PERSYSCALL_MASK
307 addi r12,r12,TI_FLAGS
312 subi r12,r12,TI_FLAGS
314 4: /* Anything else left to do? */
315 andi. r0,r9,(_TIF_SYSCALL_T_OR_A|_TIF_SINGLESTEP)
316 beq .ret_from_except_lite
318 /* Re-enable interrupts */
319 #ifdef CONFIG_PPC_BOOK3E
325 #endif /* CONFIG_PPC_BOOK3E */
328 addi r3,r1,STACK_FRAME_OVERHEAD
329 bl .do_syscall_trace_leave
332 /* Save non-volatile GPRs, if not already saved. */
344 * The sigsuspend and rt_sigsuspend system calls can call do_signal
345 * and thus put the process into the stopped state where we might
346 * want to examine its user state with ptrace. Therefore we need
347 * to save all the nonvolatile registers (r14 - r31) before calling
348 * the C code. Similarly, fork, vfork and clone need the full
349 * register state on the stack so that it can be copied to the child.
367 _GLOBAL(ppc32_swapcontext)
369 bl .compat_sys_swapcontext
372 _GLOBAL(ppc64_swapcontext)
377 _GLOBAL(ret_from_fork)
384 * This routine switches between two different tasks. The process
385 * state of one is saved on its kernel stack. Then the state
386 * of the other is restored from its kernel stack. The memory
387 * management hardware is updated to the second process's state.
388 * Finally, we can return to the second process, via ret_from_except.
389 * On entry, r3 points to the THREAD for the current task, r4
390 * points to the THREAD for the new task.
392 * Note: there are two ways to get to the "going out" portion
393 * of this code; either by coming in via the entry (_switch)
394 * or via "fork" which must set up an environment equivalent
395 * to the "_switch" path. If you change this you'll have to change
396 * the fork code also.
398 * The code which creates the new task context is in 'copy_thread'
399 * in arch/powerpc/kernel/process.c
405 stdu r1,-SWITCH_FRAME_SIZE(r1)
406 /* r3-r13 are caller saved -- Cort */
409 mflr r20 /* Return to switch caller */
414 oris r0,r0,MSR_VSX@h /* Disable VSX */
415 END_FTR_SECTION_IFSET(CPU_FTR_VSX)
416 #endif /* CONFIG_VSX */
417 #ifdef CONFIG_ALTIVEC
419 oris r0,r0,MSR_VEC@h /* Disable altivec */
420 mfspr r24,SPRN_VRSAVE /* save vrsave register value */
421 std r24,THREAD_VRSAVE(r3)
422 END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
423 #endif /* CONFIG_ALTIVEC */
432 std r1,KSP(r3) /* Set old stack pointer */
435 /* We need a sync somewhere here to make sure that if the
436 * previous task gets rescheduled on another CPU, it sees all
437 * stores it has performed on this one.
440 #endif /* CONFIG_SMP */
443 * If we optimise away the clear of the reservation in system
444 * calls because we know the CPU tracks the address of the
445 * reservation, then we need to clear it here to cover the
446 * case that the kernel context switch path has no larx
451 END_FTR_SECTION_IFSET(CPU_FTR_STCX_CHECKS_ADDRESS)
453 addi r6,r4,-THREAD /* Convert THREAD to 'current' */
454 std r6,PACACURRENT(r13) /* Set new 'current' */
456 ld r8,KSP(r4) /* new stack pointer */
457 #ifdef CONFIG_PPC_BOOK3S
459 BEGIN_FTR_SECTION_NESTED(95)
460 clrrdi r6,r8,28 /* get its ESID */
461 clrrdi r9,r1,28 /* get current sp ESID */
462 FTR_SECTION_ELSE_NESTED(95)
463 clrrdi r6,r8,40 /* get its 1T ESID */
464 clrrdi r9,r1,40 /* get current sp 1T ESID */
465 ALT_FTR_SECTION_END_NESTED_IFCLR(CPU_FTR_1T_SEGMENT, 95)
468 ALT_FTR_SECTION_END_IFSET(CPU_FTR_SLB)
469 clrldi. r0,r6,2 /* is new ESID c00000000? */
470 cmpd cr1,r6,r9 /* or is new ESID the same as current ESID? */
472 beq 2f /* if yes, don't slbie it */
474 /* Bolt in the new stack SLB entry */
475 ld r7,KSP_VSID(r4) /* Get new stack's VSID */
476 oris r0,r6,(SLB_ESID_V)@h
477 ori r0,r0,(SLB_NUM_BOLTED-1)@l
479 li r9,MMU_SEGSIZE_1T /* insert B field */
480 oris r6,r6,(MMU_SEGSIZE_1T << SLBIE_SSIZE_SHIFT)@h
481 rldimi r7,r9,SLB_VSID_SSIZE_SHIFT,0
482 END_FTR_SECTION_IFSET(CPU_FTR_1T_SEGMENT)
484 /* Update the last bolted SLB. No write barriers are needed
485 * here, provided we only update the current CPU's SLB shadow
488 ld r9,PACA_SLBSHADOWPTR(r13)
490 std r12,SLBSHADOW_STACKESID(r9) /* Clear ESID */
491 std r7,SLBSHADOW_STACKVSID(r9) /* Save VSID */
492 std r0,SLBSHADOW_STACKESID(r9) /* Save ESID */
494 /* No need to check for CPU_FTR_NO_SLBIE_B here, since when
495 * we have 1TB segments, the only CPUs known to have the errata
496 * only support less than 1TB of system memory and we'll never
497 * actually hit this code path.
501 slbie r6 /* Workaround POWER5 < DD2.1 issue */
505 #endif /* !CONFIG_PPC_BOOK3S */
507 clrrdi r7,r8,THREAD_SHIFT /* base of new stack */
508 /* Note: this uses SWITCH_FRAME_SIZE rather than INT_FRAME_SIZE
509 because we don't need to leave the 288-byte ABI gap at the
510 top of the kernel stack. */
511 addi r7,r7,THREAD_SIZE-SWITCH_FRAME_SIZE
513 mr r1,r8 /* start using new stack pointer */
514 std r7,PACAKSAVE(r13)
519 #ifdef CONFIG_ALTIVEC
521 ld r0,THREAD_VRSAVE(r4)
522 mtspr SPRN_VRSAVE,r0 /* if G4, restore VRSAVE reg */
523 END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
524 #endif /* CONFIG_ALTIVEC */
526 /* r3-r13 are destroyed -- Cort */
530 /* convert old thread to its task_struct for return value */
532 ld r7,_NIP(r1) /* Return to _switch caller in new task */
534 addi r1,r1,SWITCH_FRAME_SIZE
538 _GLOBAL(ret_from_except)
541 bne .ret_from_except_lite
544 _GLOBAL(ret_from_except_lite)
546 * Disable interrupts so that current_thread_info()->flags
547 * can't change between when we test it and when we return
548 * from the interrupt.
550 #ifdef CONFIG_PPC_BOOK3E
553 mfmsr r10 /* Get current interrupt state */
554 rldicl r9,r10,48,1 /* clear MSR_EE */
556 mtmsrd r9,1 /* Update machine state */
557 #endif /* CONFIG_PPC_BOOK3E */
559 #ifdef CONFIG_PREEMPT
560 clrrdi r9,r1,THREAD_SHIFT /* current_thread_info() */
561 li r0,_TIF_NEED_RESCHED /* bits to check */
564 /* Move MSR_PR bit in r3 to _TIF_SIGPENDING position in r0 */
565 rlwimi r0,r3,32+TIF_SIGPENDING-MSR_PR_LG,_TIF_SIGPENDING
566 and. r0,r4,r0 /* check NEED_RESCHED and maybe SIGPENDING */
569 #else /* !CONFIG_PREEMPT */
570 ld r3,_MSR(r1) /* Returning to user mode? */
572 beq restore /* if not, just restore regs and return */
574 /* Check current_thread_info()->flags */
575 clrrdi r9,r1,THREAD_SHIFT
577 andi. r0,r4,_TIF_USER_WORK_MASK
585 b .Liseries_check_pending_irqs
586 ALT_FW_FTR_SECTION_END_IFCLR(FW_FEATURE_ISERIES)
588 TRACE_AND_RESTORE_IRQ(r5);
590 /* extract EE bit and use it to restore paca->hard_enabled */
592 rldicl r4,r3,49,63 /* r0 = (r3 >> 15) & 1 */
593 stb r4,PACAHARDIRQEN(r13)
595 #ifdef CONFIG_PPC_BOOK3E
596 b .exception_return_book3e
611 * Clear the reservation. If we know the CPU tracks the address of
612 * the reservation then we can potentially save some cycles and use
613 * a larx. On POWER6 and POWER7 this is significantly faster.
616 stdcx. r0,0,r1 /* to clear the reservation */
619 ALT_FTR_SECTION_END_IFCLR(CPU_FTR_STCX_CHECKS_ADDRESS)
622 * Clear RI before restoring r13. If we are returning to
623 * userspace and we take an exception after restoring r13,
624 * we end up corrupting the userspace r13 value.
627 andc r4,r4,r0 /* r0 contains MSR_RI here */
631 * r13 is our per cpu area, only restore it if we are returning to
636 ACCOUNT_CPU_USER_EXIT(r2, r4)
653 b . /* prevent speculative execution */
655 #endif /* CONFIG_PPC_BOOK3E */
657 .Liseries_check_pending_irqs:
658 #ifdef CONFIG_PPC_ISERIES
662 /* Check for pending interrupts (iSeries) */
663 ld r3,PACALPPACAPTR(r13)
664 ld r3,LPPACAANYINT(r3)
666 beq+ 2b /* skip do_IRQ if no interrupts */
669 stb r3,PACASOFTIRQEN(r13) /* ensure we are soft-disabled */
670 #ifdef CONFIG_TRACE_IRQFLAGS
671 bl .trace_hardirqs_off
675 mtmsrd r10 /* hard-enable again */
676 addi r3,r1,STACK_FRAME_OVERHEAD
678 b .ret_from_except_lite /* loop back and handle more */
682 #ifdef CONFIG_PREEMPT
683 andi. r0,r3,MSR_PR /* Returning to user mode? */
685 /* Check that preempt_count() == 0 and interrupts are enabled */
686 lwz r8,TI_PREEMPT(r9)
690 crandc eq,cr1*4+eq,eq
693 /* Here we are preempting the current task.
695 * Ensure interrupts are soft-disabled. We also properly mark
696 * the PACA to reflect the fact that they are hard-disabled
697 * and trace the change
700 stb r0,PACASOFTIRQEN(r13)
701 stb r0,PACAHARDIRQEN(r13)
704 /* Call the scheduler with soft IRQs off */
705 1: bl .preempt_schedule_irq
707 /* Hard-disable interrupts again (and update PACA) */
708 #ifdef CONFIG_PPC_BOOK3E
715 #endif /* CONFIG_PPC_BOOK3E */
717 stb r0,PACAHARDIRQEN(r13)
719 /* Re-test flags and eventually loop */
720 clrrdi r9,r1,THREAD_SHIFT
722 andi. r0,r4,_TIF_NEED_RESCHED
727 #endif /* CONFIG_PREEMPT */
729 /* Enable interrupts */
730 #ifdef CONFIG_PPC_BOOK3E
735 #endif /* CONFIG_PPC_BOOK3E */
737 andi. r0,r4,_TIF_NEED_RESCHED
740 b .ret_from_except_lite
743 addi r3,r1,STACK_FRAME_OVERHEAD
748 addi r3,r1,STACK_FRAME_OVERHEAD
749 bl .unrecoverable_exception
752 #ifdef CONFIG_PPC_RTAS
754 * On CHRP, the Run-Time Abstraction Services (RTAS) have to be
755 * called with the MMU off.
757 * In addition, we need to be in 32b mode, at least for now.
759 * Note: r3 is an input parameter to rtas, so don't trash it...
764 stdu r1,-RTAS_FRAME_SIZE(r1) /* Save SP and create stack space. */
766 /* Because RTAS is running in 32b mode, it clobbers the high order half
767 * of all registers that it saves. We therefore save those registers
768 * RTAS might touch to the stack. (r0, r3-r13 are caller saved)
770 SAVE_GPR(2, r1) /* Save the TOC */
771 SAVE_GPR(13, r1) /* Save paca */
772 SAVE_8GPRS(14, r1) /* Save the non-volatiles */
773 SAVE_10GPRS(22, r1) /* ditto */
786 /* Temporary workaround to clear CR until RTAS can be modified to
793 /* There is no way it is acceptable to get here with interrupts enabled,
794 * check it with the asm equivalent of WARN_ON
796 lbz r0,PACASOFTIRQEN(r13)
798 EMIT_BUG_ENTRY 1b,__FILE__,__LINE__,BUGFLAG_WARNING
801 /* Hard-disable interrupts */
807 /* Unfortunately, the stack pointer and the MSR are also clobbered,
808 * so they are saved in the PACA which allows us to restore
809 * our original state after RTAS returns.
812 std r6,PACASAVEDMSR(r13)
814 /* Setup our real return addr */
815 LOAD_REG_ADDR(r4,.rtas_return_loc)
816 clrldi r4,r4,2 /* convert to realmode address */
820 ori r0,r0,MSR_EE|MSR_SE|MSR_BE|MSR_RI
824 rldicr r9,r9,MSR_SF_LG,(63-MSR_SF_LG)
825 ori r9,r9,MSR_IR|MSR_DR|MSR_FE0|MSR_FE1|MSR_FP|MSR_RI
827 sync /* disable interrupts so SRR0/1 */
828 mtmsrd r0 /* don't get trashed */
830 LOAD_REG_ADDR(r4, rtas)
831 ld r5,RTASENTRY(r4) /* get the rtas->entry value */
832 ld r4,RTASBASE(r4) /* get the rtas->base value */
837 b . /* prevent speculative execution */
839 _STATIC(rtas_return_loc)
840 /* relocation is off at this point */
841 mfspr r4,SPRN_SPRG_PACA /* Get PACA */
842 clrldi r4,r4,2 /* convert to realmode address */
846 ld r3,(1f-0b)(r3) /* get &.rtas_restore_regs */
854 ld r1,PACAR1(r4) /* Restore our SP */
855 ld r4,PACASAVEDMSR(r4) /* Restore our MSR */
860 b . /* prevent speculative execution */
863 1: .llong .rtas_restore_regs
865 _STATIC(rtas_restore_regs)
866 /* relocation is on at this point */
867 REST_GPR(2, r1) /* Restore the TOC */
868 REST_GPR(13, r1) /* Restore paca */
869 REST_8GPRS(14, r1) /* Restore the non-volatiles */
870 REST_10GPRS(22, r1) /* ditto */
872 mfspr r13,SPRN_SPRG_PACA
885 addi r1,r1,RTAS_FRAME_SIZE /* Unstack our frame */
886 ld r0,16(r1) /* get return address */
889 blr /* return to caller */
891 #endif /* CONFIG_PPC_RTAS */
896 stdu r1,-PROM_FRAME_SIZE(r1) /* Save SP and create stack space */
898 /* Because PROM is running in 32b mode, it clobbers the high order half
899 * of all registers that it saves. We therefore save those registers
900 * PROM might touch to the stack. (r0, r3-r13 are caller saved)
911 /* Get the PROM entrypoint */
914 /* Switch MSR to 32 bits mode
916 #ifdef CONFIG_PPC_BOOK3E
917 rlwinm r11,r11,0,1,31
919 #else /* CONFIG_PPC_BOOK3E */
922 rldicr r12,r12,MSR_SF_LG,(63-MSR_SF_LG)
925 rldicr r12,r12,MSR_ISF_LG,(63-MSR_ISF_LG)
928 #endif /* CONFIG_PPC_BOOK3E */
931 /* Enter PROM here... */
934 /* Just make sure that r1 top 32 bits didn't get
939 /* Restore the MSR (back to 64 bits) */
944 /* Restore other registers */
952 addi r1,r1,PROM_FRAME_SIZE
957 #ifdef CONFIG_FUNCTION_TRACER
958 #ifdef CONFIG_DYNAMIC_FTRACE
963 _GLOBAL(ftrace_caller)
964 /* Taken from output of objdump from lib64/glibc */
970 subi r3, r3, MCOUNT_INSN_SIZE
975 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
976 .globl ftrace_graph_call
979 _GLOBAL(ftrace_graph_stub)
991 /* Taken from output of objdump from lib64/glibc */
998 subi r3, r3, MCOUNT_INSN_SIZE
999 LOAD_REG_ADDR(r5,ftrace_trace_function)
1007 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1008 b ftrace_graph_caller
1013 _GLOBAL(ftrace_stub)
1016 #endif /* CONFIG_DYNAMIC_FTRACE */
1018 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1019 _GLOBAL(ftrace_graph_caller)
1020 /* load r4 with local address */
1022 subi r4, r4, MCOUNT_INSN_SIZE
1024 /* get the parent address */
1028 bl .prepare_ftrace_return
1036 _GLOBAL(return_to_handler)
1037 /* need to save return values */
1044 bl .ftrace_return_to_handler
1047 /* return value has real return address */
1055 /* Jump back to real return address */
1058 _GLOBAL(mod_return_to_handler)
1059 /* need to save return values */
1069 * We are in a module using the module's TOC.
1070 * Switch to our TOC to run inside the core kernel.
1074 bl .ftrace_return_to_handler
1077 /* return value has real return address */
1086 /* Jump back to real return address */
1088 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
1089 #endif /* CONFIG_FUNCTION_TRACER */