diff options
Diffstat (limited to 'arch/mips/kernel/octeon_switch.S')
| -rw-r--r-- | arch/mips/kernel/octeon_switch.S | 171 |
1 files changed, 94 insertions, 77 deletions
diff --git a/arch/mips/kernel/octeon_switch.S b/arch/mips/kernel/octeon_switch.S index dd18b26a358..f6547680c81 100644 --- a/arch/mips/kernel/octeon_switch.S +++ b/arch/mips/kernel/octeon_switch.S @@ -4,34 +4,21 @@ * for more details. * * Copyright (C) 1994, 1995, 1996, 1998, 1999, 2002, 2003 Ralf Baechle - * Copyright (C) 1996 David S. Miller (dm@engr.sgi.com) + * Copyright (C) 1996 David S. Miller (davem@davemloft.net) * Copyright (C) 1994, 1995, 1996, by Andreas Busse * Copyright (C) 1999 Silicon Graphics, Inc. * Copyright (C) 2000 MIPS Technologies, Inc. * written by Carsten Langgaard, carstenl@mips.com */ -#include <asm/asm.h> -#include <asm/cachectl.h> -#include <asm/fpregdef.h> -#include <asm/mipsregs.h> -#include <asm/asm-offsets.h> -#include <asm/page.h> -#include <asm/pgtable-bits.h> -#include <asm/regdef.h> -#include <asm/stackframe.h> -#include <asm/thread_info.h> - -#include <asm/asmmacro.h> - -/* - * Offset to the current process status flags, the first 32 bytes of the - * stack are not used. - */ -#define ST_OFF (_THREAD_SIZE - 32 - PT_SIZE + PT_STATUS) +#define USE_ALTERNATE_RESUME_IMPL 1 + .set push + .set arch=mips64r2 +#include "r4k_switch.S" + .set pop /* * task_struct *resume(task_struct *prev, task_struct *next, - * struct thread_info *next_ti) + * struct thread_info *next_ti, int usedfpu) */ .align 7 LEAF(resume) @@ -41,6 +28,34 @@ cpu_save_nonscratch a0 LONG_S ra, THREAD_REG31(a0) + /* + * check if we need to save FPU registers + */ + PTR_L t3, TASK_THREAD_INFO(a0) + LONG_L t0, TI_FLAGS(t3) + li t1, _TIF_USEDFPU + and t2, t0, t1 + beqz t2, 1f + nor t1, zero, t1 + + and t0, t0, t1 + LONG_S t0, TI_FLAGS(t3) + + /* + * clear saved user stack CU1 bit + */ + LONG_L t0, ST_OFF(t3) + li t1, ~ST0_CU1 + and t0, t0, t1 + LONG_S t0, ST_OFF(t3) + + .set push + .set arch=mips64r2 + fpu_save_double a0 t0 t1 # c0_status passed in t0 + # clobbers t1 + .set pop +1: + /* check if we need to save COP2 registers */ PTR_L t2, TASK_THREAD_INFO(a0) LONG_L t0, ST_OFF(t2) @@ -70,7 +85,7 @@ 1: #if CONFIG_CAVIUM_OCTEON_CVMSEG_SIZE > 0 /* Check if we need to store CVMSEG state */ - mfc0 t0, $11,7 /* CvmMemCtl */ + mfc0 t0, $11,7 /* CvmMemCtl */ bbit0 t0, 6, 3f /* Is user access enabled? */ /* Store the CVMSEG state */ @@ -78,8 +93,8 @@ andi t0, 0x3f /* Multiply * (cache line size/sizeof(long)/2) */ sll t0, 7-LONGLOG-1 - li t1, -32768 /* Base address of CVMSEG */ - LONG_ADDI t2, a0, THREAD_CVMSEG /* Where to store CVMSEG to */ + li t1, -32768 /* Base address of CVMSEG */ + LONG_ADDI t2, a0, THREAD_CVMSEG /* Where to store CVMSEG to */ synciobdma 2: .set noreorder @@ -90,15 +105,22 @@ LONG_S t8, 0(t2) /* Store CVMSEG to thread storage */ LONG_ADDU t2, LONGSIZE*2 /* Increment loc in thread storage */ bnez t0, 2b /* Loop until we've copied it all */ - LONG_S t9, -LONGSIZE(t2)/* Store CVMSEG to thread storage */ + LONG_S t9, -LONGSIZE(t2)/* Store CVMSEG to thread storage */ .set reorder /* Disable access to CVMSEG */ - mfc0 t0, $11,7 /* CvmMemCtl */ + mfc0 t0, $11,7 /* CvmMemCtl */ xori t0, t0, 0x40 /* Bit 6 is CVMSEG user enable */ - mtc0 t0, $11,7 /* CvmMemCtl */ + mtc0 t0, $11,7 /* CvmMemCtl */ #endif 3: + +#if defined(CONFIG_CC_STACKPROTECTOR) && !defined(CONFIG_SMP) + PTR_LA t8, __stack_chk_guard + LONG_L t9, TASK_STACK_CANARY(a1) + LONG_S t9, 0(t8) +#endif + /* * The order of restoring the registers takes care of the race * updating $28, $29 and kernelsp without disabling ints. @@ -106,12 +128,7 @@ move $28, a2 cpu_restore_nonscratch a1 -#if (_THREAD_SIZE - 32) < 0x8000 - PTR_ADDIU t0, $28, _THREAD_SIZE - 32 -#else - PTR_LI t0, _THREAD_SIZE - 32 - PTR_ADDU t0, $28 -#endif + PTR_ADDU t0, $28, _THREAD_SIZE - 32 set_saved_sp t0, t1, t2 mfc0 t1, CP0_STATUS /* Do we really need this? */ @@ -134,7 +151,7 @@ dmfc0 t9, $9,7 /* CvmCtl register. */ - /* Save the COP2 CRC state */ + /* Save the COP2 CRC state */ dmfc2 t0, 0x0201 dmfc2 t1, 0x0202 dmfc2 t2, 0x0200 @@ -150,30 +167,30 @@ sd t0, OCTEON_CP2_LLM_DAT(a0) sd t1, OCTEON_CP2_LLM_DAT+8(a0) -1: bbit1 t9, 26, 3f /* done if CvmCtl[NOCRYPTO] set */ +1: bbit1 t9, 26, 3f /* done if CvmCtl[NOCRYPTO] set */ /* Save the COP2 crypto state */ - /* this part is mostly common to both pass 1 and later revisions */ - dmfc2 t0, 0x0084 - dmfc2 t1, 0x0080 - dmfc2 t2, 0x0081 - dmfc2 t3, 0x0082 + /* this part is mostly common to both pass 1 and later revisions */ + dmfc2 t0, 0x0084 + dmfc2 t1, 0x0080 + dmfc2 t2, 0x0081 + dmfc2 t3, 0x0082 sd t0, OCTEON_CP2_3DES_IV(a0) - dmfc2 t0, 0x0088 + dmfc2 t0, 0x0088 sd t1, OCTEON_CP2_3DES_KEY(a0) - dmfc2 t1, 0x0111 /* only necessary for pass 1 */ + dmfc2 t1, 0x0111 /* only necessary for pass 1 */ sd t2, OCTEON_CP2_3DES_KEY+8(a0) - dmfc2 t2, 0x0102 + dmfc2 t2, 0x0102 sd t3, OCTEON_CP2_3DES_KEY+16(a0) - dmfc2 t3, 0x0103 + dmfc2 t3, 0x0103 sd t0, OCTEON_CP2_3DES_RESULT(a0) - dmfc2 t0, 0x0104 - sd t1, OCTEON_CP2_AES_INP0(a0) /* only necessary for pass 1 */ - dmfc2 t1, 0x0105 + dmfc2 t0, 0x0104 + sd t1, OCTEON_CP2_AES_INP0(a0) /* only necessary for pass 1 */ + dmfc2 t1, 0x0105 sd t2, OCTEON_CP2_AES_IV(a0) dmfc2 t2, 0x0106 sd t3, OCTEON_CP2_AES_IV+8(a0) - dmfc2 t3, 0x0107 + dmfc2 t3, 0x0107 sd t0, OCTEON_CP2_AES_KEY(a0) dmfc2 t0, 0x0110 sd t1, OCTEON_CP2_AES_KEY+8(a0) @@ -181,7 +198,7 @@ sd t2, OCTEON_CP2_AES_KEY+16(a0) dmfc2 t2, 0x0101 sd t3, OCTEON_CP2_AES_KEY+24(a0) - mfc0 t3, $15,0 /* Get the processor ID register */ + mfc0 t3, $15,0 /* Get the processor ID register */ sd t0, OCTEON_CP2_AES_KEYLEN(a0) li t0, 0x000d0000 /* This is the processor ID of Octeon Pass1 */ sd t1, OCTEON_CP2_AES_RESULT(a0) @@ -189,7 +206,7 @@ /* Skip to the Pass1 version of the remainder of the COP2 state */ beq t3, t0, 2f - /* the non-pass1 state when !CvmCtl[NOCRYPTO] */ + /* the non-pass1 state when !CvmCtl[NOCRYPTO] */ dmfc2 t1, 0x0240 dmfc2 t2, 0x0241 dmfc2 t3, 0x0242 @@ -215,7 +232,7 @@ sd t2, OCTEON_CP2_HSH_DATW+72(a0) dmfc2 t2, 0x024D sd t3, OCTEON_CP2_HSH_DATW+80(a0) - dmfc2 t3, 0x024E + dmfc2 t3, 0x024E sd t0, OCTEON_CP2_HSH_DATW+88(a0) dmfc2 t0, 0x0250 sd t1, OCTEON_CP2_HSH_DATW+96(a0) @@ -233,9 +250,9 @@ sd t3, OCTEON_CP2_HSH_IVW+24(a0) dmfc2 t3, 0x0257 sd t0, OCTEON_CP2_HSH_IVW+32(a0) - dmfc2 t0, 0x0258 + dmfc2 t0, 0x0258 sd t1, OCTEON_CP2_HSH_IVW+40(a0) - dmfc2 t1, 0x0259 + dmfc2 t1, 0x0259 sd t2, OCTEON_CP2_HSH_IVW+48(a0) dmfc2 t2, 0x025E sd t3, OCTEON_CP2_HSH_IVW+56(a0) @@ -248,7 +265,7 @@ sd t0, OCTEON_CP2_GFM_RESULT+8(a0) jr ra -2: /* pass 1 special stuff when !CvmCtl[NOCRYPTO] */ +2: /* pass 1 special stuff when !CvmCtl[NOCRYPTO] */ dmfc2 t3, 0x0040 dmfc2 t0, 0x0041 dmfc2 t1, 0x0042 @@ -270,7 +287,7 @@ sd t3, OCTEON_CP2_HSH_IVW+8(a0) sd t0, OCTEON_CP2_HSH_IVW+16(a0) -3: /* pass 1 or CvmCtl[NOCRYPTO] set */ +3: /* pass 1 or CvmCtl[NOCRYPTO] set */ jr ra END(octeon_cop2_save) @@ -281,19 +298,19 @@ .set push .set noreorder LEAF(octeon_cop2_restore) - /* First cache line was prefetched before the call */ - pref 4, 128(a0) + /* First cache line was prefetched before the call */ + pref 4, 128(a0) dmfc0 t9, $9,7 /* CvmCtl register. */ - pref 4, 256(a0) + pref 4, 256(a0) ld t0, OCTEON_CP2_CRC_IV(a0) - pref 4, 384(a0) + pref 4, 384(a0) ld t1, OCTEON_CP2_CRC_LENGTH(a0) ld t2, OCTEON_CP2_CRC_POLY(a0) /* Restore the COP2 CRC state */ dmtc2 t0, 0x0201 - dmtc2 t1, 0x1202 + dmtc2 t1, 0x1202 bbit1 t9, 28, 2f /* Skip LLM if CvmCtl[NODFA_CP2] is set */ dmtc2 t2, 0x4200 @@ -311,19 +328,19 @@ ld t0, OCTEON_CP2_3DES_IV(a0) ld t1, OCTEON_CP2_3DES_KEY(a0) ld t2, OCTEON_CP2_3DES_KEY+8(a0) - dmtc2 t0, 0x0084 + dmtc2 t0, 0x0084 ld t0, OCTEON_CP2_3DES_KEY+16(a0) - dmtc2 t1, 0x0080 + dmtc2 t1, 0x0080 ld t1, OCTEON_CP2_3DES_RESULT(a0) - dmtc2 t2, 0x0081 + dmtc2 t2, 0x0081 ld t2, OCTEON_CP2_AES_INP0(a0) /* only really needed for pass 1 */ dmtc2 t0, 0x0082 ld t0, OCTEON_CP2_AES_IV(a0) - dmtc2 t1, 0x0098 + dmtc2 t1, 0x0098 ld t1, OCTEON_CP2_AES_IV+8(a0) - dmtc2 t2, 0x010A /* only really needed for pass 1 */ + dmtc2 t2, 0x010A /* only really needed for pass 1 */ ld t2, OCTEON_CP2_AES_KEY(a0) - dmtc2 t0, 0x0102 + dmtc2 t0, 0x0102 ld t0, OCTEON_CP2_AES_KEY+8(a0) dmtc2 t1, 0x0103 ld t1, OCTEON_CP2_AES_KEY+16(a0) @@ -335,14 +352,14 @@ ld t1, OCTEON_CP2_AES_RESULT(a0) dmtc2 t2, 0x0107 ld t2, OCTEON_CP2_AES_RESULT+8(a0) - mfc0 t3, $15,0 /* Get the processor ID register */ + mfc0 t3, $15,0 /* Get the processor ID register */ dmtc2 t0, 0x0110 li t0, 0x000d0000 /* This is the processor ID of Octeon Pass1 */ dmtc2 t1, 0x0100 bne t0, t3, 3f /* Skip the next stuff for non-pass1 */ dmtc2 t2, 0x0101 - /* this code is specific for pass 1 */ + /* this code is specific for pass 1 */ ld t0, OCTEON_CP2_HSH_DATW(a0) ld t1, OCTEON_CP2_HSH_DATW+8(a0) ld t2, OCTEON_CP2_HSH_DATW+16(a0) @@ -362,10 +379,10 @@ ld t0, OCTEON_CP2_HSH_IVW+16(a0) dmtc2 t1, 0x0048 dmtc2 t2, 0x0049 - b done_restore /* unconditional branch */ + b done_restore /* unconditional branch */ dmtc2 t0, 0x004A -3: /* this is post-pass1 code */ +3: /* this is post-pass1 code */ ld t2, OCTEON_CP2_HSH_DATW(a0) ld t0, OCTEON_CP2_HSH_DATW+8(a0) ld t1, OCTEON_CP2_HSH_DATW+16(a0) @@ -434,7 +451,7 @@ done_restore: * sp is assumed to point to a struct pt_regs * * NOTE: This is called in SAVE_SOME in stackframe.h. It can only - * safely modify k0 and k1. + * safely modify k0 and k1. */ .align 7 .set push @@ -447,14 +464,14 @@ done_restore: /* Save the multiplier state */ v3mulu k0, $0, $0 v3mulu k1, $0, $0 - sd k0, PT_MTP(sp) /* PT_MTP has P0 */ + sd k0, PT_MTP(sp) /* PT_MTP has P0 */ v3mulu k0, $0, $0 sd k1, PT_MTP+8(sp) /* PT_MTP+8 has P1 */ ori k1, $0, 1 v3mulu k1, k1, $0 sd k0, PT_MTP+16(sp) /* PT_MTP+16 has P2 */ v3mulu k0, $0, $0 - sd k1, PT_MPL(sp) /* PT_MPL has MPL0 */ + sd k1, PT_MPL(sp) /* PT_MPL has MPL0 */ v3mulu k1, $0, $0 sd k0, PT_MPL+8(sp) /* PT_MPL+8 has MPL1 */ jr ra @@ -476,19 +493,19 @@ done_restore: .set noreorder LEAF(octeon_mult_restore) dmfc0 k1, $9,7 /* CvmCtl register. */ - ld v0, PT_MPL(sp) /* MPL0 */ - ld v1, PT_MPL+8(sp) /* MPL1 */ - ld k0, PT_MPL+16(sp) /* MPL2 */ + ld v0, PT_MPL(sp) /* MPL0 */ + ld v1, PT_MPL+8(sp) /* MPL1 */ + ld k0, PT_MPL+16(sp) /* MPL2 */ bbit1 k1, 27, 1f /* Skip CvmCtl[NOMUL] */ /* Normally falls through, so no time wasted here */ nop /* Restore the multiplier state */ - ld k1, PT_MTP+16(sp) /* P2 */ + ld k1, PT_MTP+16(sp) /* P2 */ MTM0 v0 /* MPL0 */ ld v0, PT_MTP+8(sp) /* P1 */ MTM1 v1 /* MPL1 */ - ld v1, PT_MTP(sp) /* P0 */ + ld v1, PT_MTP(sp) /* P0 */ MTM2 k0 /* MPL2 */ MTP2 k1 /* P2 */ MTP1 v0 /* P1 */ |