cachepc-linux

Fork of AMDESE/linux with modifications for CachePC side-channel attack
git clone https://git.sinitax.com/sinitax/cachepc-linux
Log | Files | Refs | README | LICENSE | sfeed.txt

entry_64.S (8705B)


      1/* SPDX-License-Identifier: GPL-2.0-or-later */
      2/*
      3 *  PowerPC version 
      4 *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
      5 *  Rewritten by Cort Dougan (cort@cs.nmt.edu) for PReP
      6 *    Copyright (C) 1996 Cort Dougan <cort@cs.nmt.edu>
      7 *  Adapted for Power Macintosh by Paul Mackerras.
      8 *  Low-level exception handlers and MMU support
      9 *  rewritten by Paul Mackerras.
     10 *    Copyright (C) 1996 Paul Mackerras.
     11 *  MPC8xx modifications Copyright (C) 1997 Dan Malek (dmalek@jlc.net).
     12 *
     13 *  This file contains the system call entry code, context switch
     14 *  code, and exception/interrupt return code for PowerPC.
     15 */
     16
     17#include <linux/errno.h>
     18#include <linux/err.h>
     19#include <asm/cache.h>
     20#include <asm/unistd.h>
     21#include <asm/processor.h>
     22#include <asm/page.h>
     23#include <asm/mmu.h>
     24#include <asm/thread_info.h>
     25#include <asm/code-patching-asm.h>
     26#include <asm/ppc_asm.h>
     27#include <asm/asm-offsets.h>
     28#include <asm/cputable.h>
     29#include <asm/firmware.h>
     30#include <asm/bug.h>
     31#include <asm/ptrace.h>
     32#include <asm/irqflags.h>
     33#include <asm/hw_irq.h>
     34#include <asm/context_tracking.h>
     35#include <asm/ppc-opcode.h>
     36#include <asm/barrier.h>
     37#include <asm/export.h>
     38#include <asm/asm-compat.h>
     39#ifdef CONFIG_PPC_BOOK3S
     40#include <asm/exception-64s.h>
     41#else
     42#include <asm/exception-64e.h>
     43#endif
     44#include <asm/feature-fixups.h>
     45#include <asm/kup.h>
     46
     47/*
     48 * System calls.
     49 */
     50	.section	".text"
     51
     52#ifdef CONFIG_PPC_BOOK3S_64
     53
     54#define FLUSH_COUNT_CACHE	\
     551:	nop;			\
     56	patch_site 1b, patch__call_flush_branch_caches1; \
     571:	nop;			\
     58	patch_site 1b, patch__call_flush_branch_caches2; \
     591:	nop;			\
     60	patch_site 1b, patch__call_flush_branch_caches3
     61
     62.macro nops number
     63	.rept \number
     64	nop
     65	.endr
     66.endm
     67
     68.balign 32
     69.global flush_branch_caches
     70flush_branch_caches:
     71	/* Save LR into r9 */
     72	mflr	r9
     73
     74	// Flush the link stack
     75	.rept 64
     76	bl	.+4
     77	.endr
     78	b	1f
     79	nops	6
     80
     81	.balign 32
     82	/* Restore LR */
     831:	mtlr	r9
     84
     85	// If we're just flushing the link stack, return here
     863:	nop
     87	patch_site 3b patch__flush_link_stack_return
     88
     89	li	r9,0x7fff
     90	mtctr	r9
     91
     92	PPC_BCCTR_FLUSH
     93
     942:	nop
     95	patch_site 2b patch__flush_count_cache_return
     96
     97	nops	3
     98
     99	.rept 278
    100	.balign 32
    101	PPC_BCCTR_FLUSH
    102	nops	7
    103	.endr
    104
    105	blr
    106#else
    107#define FLUSH_COUNT_CACHE
    108#endif /* CONFIG_PPC_BOOK3S_64 */
    109
    110/*
    111 * This routine switches between two different tasks.  The process
    112 * state of one is saved on its kernel stack.  Then the state
    113 * of the other is restored from its kernel stack.  The memory
    114 * management hardware is updated to the second process's state.
    115 * Finally, we can return to the second process, via interrupt_return.
    116 * On entry, r3 points to the THREAD for the current task, r4
    117 * points to the THREAD for the new task.
    118 *
    119 * Note: there are two ways to get to the "going out" portion
    120 * of this code; either by coming in via the entry (_switch)
    121 * or via "fork" which must set up an environment equivalent
    122 * to the "_switch" path.  If you change this you'll have to change
    123 * the fork code also.
    124 *
    125 * The code which creates the new task context is in 'copy_thread'
    126 * in arch/powerpc/kernel/process.c 
    127 */
    128	.align	7
    129_GLOBAL(_switch)
    130	mflr	r0
    131	std	r0,16(r1)
    132	stdu	r1,-SWITCH_FRAME_SIZE(r1)
    133	/* r3-r13 are caller saved -- Cort */
    134	SAVE_NVGPRS(r1)
    135	std	r0,_NIP(r1)	/* Return to switch caller */
    136	mfcr	r23
    137	std	r23,_CCR(r1)
    138	std	r1,KSP(r3)	/* Set old stack pointer */
    139
    140	kuap_check_amr r9, r10
    141
    142	FLUSH_COUNT_CACHE	/* Clobbers r9, ctr */
    143
    144	/*
    145	 * On SMP kernels, care must be taken because a task may be
    146	 * scheduled off CPUx and on to CPUy. Memory ordering must be
    147	 * considered.
    148	 *
    149	 * Cacheable stores on CPUx will be visible when the task is
    150	 * scheduled on CPUy by virtue of the core scheduler barriers
    151	 * (see "Notes on Program-Order guarantees on SMP systems." in
    152	 * kernel/sched/core.c).
    153	 *
    154	 * Uncacheable stores in the case of involuntary preemption must
    155	 * be taken care of. The smp_mb__after_spinlock() in __schedule()
    156	 * is implemented as hwsync on powerpc, which orders MMIO too. So
    157	 * long as there is an hwsync in the context switch path, it will
    158	 * be executed on the source CPU after the task has performed
    159	 * all MMIO ops on that CPU, and on the destination CPU before the
    160	 * task performs any MMIO ops there.
    161	 */
    162
    163	/*
    164	 * The kernel context switch path must contain a spin_lock,
    165	 * which contains larx/stcx, which will clear any reservation
    166	 * of the task being switched.
    167	 */
    168#ifdef CONFIG_PPC_BOOK3S
    169/* Cancel all explict user streams as they will have no use after context
    170 * switch and will stop the HW from creating streams itself
    171 */
    172	DCBT_BOOK3S_STOP_ALL_STREAM_IDS(r6)
    173#endif
    174
    175	addi	r6,r4,-THREAD	/* Convert THREAD to 'current' */
    176	std	r6,PACACURRENT(r13)	/* Set new 'current' */
    177#if defined(CONFIG_STACKPROTECTOR)
    178	ld	r6, TASK_CANARY(r6)
    179	std	r6, PACA_CANARY(r13)
    180#endif
    181
    182	ld	r8,KSP(r4)	/* new stack pointer */
    183#ifdef CONFIG_PPC_64S_HASH_MMU
    184BEGIN_MMU_FTR_SECTION
    185	b	2f
    186END_MMU_FTR_SECTION_IFSET(MMU_FTR_TYPE_RADIX)
    187BEGIN_FTR_SECTION
    188	clrrdi	r6,r8,28	/* get its ESID */
    189	clrrdi	r9,r1,28	/* get current sp ESID */
    190FTR_SECTION_ELSE
    191	clrrdi	r6,r8,40	/* get its 1T ESID */
    192	clrrdi	r9,r1,40	/* get current sp 1T ESID */
    193ALT_MMU_FTR_SECTION_END_IFCLR(MMU_FTR_1T_SEGMENT)
    194	clrldi.	r0,r6,2		/* is new ESID c00000000? */
    195	cmpd	cr1,r6,r9	/* or is new ESID the same as current ESID? */
    196	cror	eq,4*cr1+eq,eq
    197	beq	2f		/* if yes, don't slbie it */
    198
    199	/* Bolt in the new stack SLB entry */
    200	ld	r7,KSP_VSID(r4)	/* Get new stack's VSID */
    201	oris	r0,r6,(SLB_ESID_V)@h
    202	ori	r0,r0,(SLB_NUM_BOLTED-1)@l
    203BEGIN_FTR_SECTION
    204	li	r9,MMU_SEGSIZE_1T	/* insert B field */
    205	oris	r6,r6,(MMU_SEGSIZE_1T << SLBIE_SSIZE_SHIFT)@h
    206	rldimi	r7,r9,SLB_VSID_SSIZE_SHIFT,0
    207END_MMU_FTR_SECTION_IFSET(MMU_FTR_1T_SEGMENT)
    208
    209	/* Update the last bolted SLB.  No write barriers are needed
    210	 * here, provided we only update the current CPU's SLB shadow
    211	 * buffer.
    212	 */
    213	ld	r9,PACA_SLBSHADOWPTR(r13)
    214	li	r12,0
    215	std	r12,SLBSHADOW_STACKESID(r9)	/* Clear ESID */
    216	li	r12,SLBSHADOW_STACKVSID
    217	STDX_BE	r7,r12,r9			/* Save VSID */
    218	li	r12,SLBSHADOW_STACKESID
    219	STDX_BE	r0,r12,r9			/* Save ESID */
    220
    221	/* No need to check for MMU_FTR_NO_SLBIE_B here, since when
    222	 * we have 1TB segments, the only CPUs known to have the errata
    223	 * only support less than 1TB of system memory and we'll never
    224	 * actually hit this code path.
    225	 */
    226
    227	isync
    228	slbie	r6
    229BEGIN_FTR_SECTION
    230	slbie	r6		/* Workaround POWER5 < DD2.1 issue */
    231END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
    232	slbmte	r7,r0
    233	isync
    2342:
    235#endif /* CONFIG_PPC_64S_HASH_MMU */
    236
    237	clrrdi	r7, r8, THREAD_SHIFT	/* base of new stack */
    238	/* Note: this uses SWITCH_FRAME_SIZE rather than INT_FRAME_SIZE
    239	   because we don't need to leave the 288-byte ABI gap at the
    240	   top of the kernel stack. */
    241	addi	r7,r7,THREAD_SIZE-SWITCH_FRAME_SIZE
    242
    243	/*
    244	 * PMU interrupts in radix may come in here. They will use r1, not
    245	 * PACAKSAVE, so this stack switch will not cause a problem. They
    246	 * will store to the process stack, which may then be migrated to
    247	 * another CPU. However the rq lock release on this CPU paired with
    248	 * the rq lock acquire on the new CPU before the stack becomes
    249	 * active on the new CPU, will order those stores.
    250	 */
    251	mr	r1,r8		/* start using new stack pointer */
    252	std	r7,PACAKSAVE(r13)
    253
    254	ld	r6,_CCR(r1)
    255	mtcrf	0xFF,r6
    256
    257	/* r3-r13 are destroyed -- Cort */
    258	REST_NVGPRS(r1)
    259
    260	/* convert old thread to its task_struct for return value */
    261	addi	r3,r3,-THREAD
    262	ld	r7,_NIP(r1)	/* Return to _switch caller in new task */
    263	mtlr	r7
    264	addi	r1,r1,SWITCH_FRAME_SIZE
    265	blr
    266
    267_GLOBAL(enter_prom)
    268	mflr	r0
    269	std	r0,16(r1)
    270        stdu	r1,-SWITCH_FRAME_SIZE(r1) /* Save SP and create stack space */
    271
    272	/* Because PROM is running in 32b mode, it clobbers the high order half
    273	 * of all registers that it saves.  We therefore save those registers
    274	 * PROM might touch to the stack.  (r0, r3-r13 are caller saved)
    275   	 */
    276	SAVE_GPR(2, r1)
    277	SAVE_GPR(13, r1)
    278	SAVE_NVGPRS(r1)
    279	mfcr	r10
    280	mfmsr	r11
    281	std	r10,_CCR(r1)
    282	std	r11,_MSR(r1)
    283
    284	/* Put PROM address in SRR0 */
    285	mtsrr0	r4
    286
    287	/* Setup our trampoline return addr in LR */
    288	bcl	20,31,$+4
    2890:	mflr	r4
    290	addi	r4,r4,(1f - 0b)
    291       	mtlr	r4
    292
    293	/* Prepare a 32-bit mode big endian MSR
    294	 */
    295#ifdef CONFIG_PPC_BOOK3E
    296	rlwinm	r11,r11,0,1,31
    297	mtsrr1	r11
    298	rfi
    299#else /* CONFIG_PPC_BOOK3E */
    300	LOAD_REG_IMMEDIATE(r12, MSR_SF | MSR_LE)
    301	andc	r11,r11,r12
    302	mtsrr1	r11
    303	RFI_TO_KERNEL
    304#endif /* CONFIG_PPC_BOOK3E */
    305
    3061:	/* Return from OF */
    307	FIXUP_ENDIAN
    308
    309	/* Just make sure that r1 top 32 bits didn't get
    310	 * corrupt by OF
    311	 */
    312	rldicl	r1,r1,0,32
    313
    314	/* Restore the MSR (back to 64 bits) */
    315	ld	r0,_MSR(r1)
    316	MTMSRD(r0)
    317        isync
    318
    319	/* Restore other registers */
    320	REST_GPR(2, r1)
    321	REST_GPR(13, r1)
    322	REST_NVGPRS(r1)
    323	ld	r4,_CCR(r1)
    324	mtcr	r4
    325
    326        addi	r1,r1,SWITCH_FRAME_SIZE
    327	ld	r0,16(r1)
    328	mtlr    r0
    329        blr