Merge branch 'work.splice_read' of git://git.kernel.org/pub/scm/linux/kernel/git...
[cascardo/linux.git] / arch / powerpc / kernel / fpu.S
1 /*
2  *  FPU support code, moved here from head.S so that it can be used
3  *  by chips which use other head-whatever.S files.
4  *
5  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
6  *    Copyright (C) 1996 Cort Dougan <cort@cs.nmt.edu>
7  *    Copyright (C) 1996 Paul Mackerras.
8  *    Copyright (C) 1997 Dan Malek (dmalek@jlc.net).
9  *
10  *  This program is free software; you can redistribute it and/or
11  *  modify it under the terms of the GNU General Public License
12  *  as published by the Free Software Foundation; either version
13  *  2 of the License, or (at your option) any later version.
14  *
15  */
16
17 #include <asm/reg.h>
18 #include <asm/page.h>
19 #include <asm/mmu.h>
20 #include <asm/pgtable.h>
21 #include <asm/cputable.h>
22 #include <asm/cache.h>
23 #include <asm/thread_info.h>
24 #include <asm/ppc_asm.h>
25 #include <asm/asm-offsets.h>
26 #include <asm/ptrace.h>
27
28 #ifdef CONFIG_VSX
29 #define __REST_32FPVSRS(n,c,base)                                       \
30 BEGIN_FTR_SECTION                                                       \
31         b       2f;                                                     \
32 END_FTR_SECTION_IFSET(CPU_FTR_VSX);                                     \
33         REST_32FPRS(n,base);                                            \
34         b       3f;                                                     \
35 2:      REST_32VSRS(n,c,base);                                          \
36 3:
37
38 #define __SAVE_32FPVSRS(n,c,base)                                       \
39 BEGIN_FTR_SECTION                                                       \
40         b       2f;                                                     \
41 END_FTR_SECTION_IFSET(CPU_FTR_VSX);                                     \
42         SAVE_32FPRS(n,base);                                            \
43         b       3f;                                                     \
44 2:      SAVE_32VSRS(n,c,base);                                          \
45 3:
46 #else
47 #define __REST_32FPVSRS(n,b,base)       REST_32FPRS(n, base)
48 #define __SAVE_32FPVSRS(n,b,base)       SAVE_32FPRS(n, base)
49 #endif
50 #define REST_32FPVSRS(n,c,base) __REST_32FPVSRS(n,__REG_##c,__REG_##base)
51 #define SAVE_32FPVSRS(n,c,base) __SAVE_32FPVSRS(n,__REG_##c,__REG_##base)
52
53 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
54 /* void do_load_up_transact_fpu(struct thread_struct *thread)
55  *
56  * This is similar to load_up_fpu but for the transactional version of the FP
57  * register set.  It doesn't mess with the task MSR or valid flags.
58  * Furthermore, we don't do lazy FP with TM currently.
59  */
60 _GLOBAL(do_load_up_transact_fpu)
61         mfmsr   r6
62         ori     r5,r6,MSR_FP
63 #ifdef CONFIG_VSX
64 BEGIN_FTR_SECTION
65         oris    r5,r5,MSR_VSX@h
66 END_FTR_SECTION_IFSET(CPU_FTR_VSX)
67 #endif
68         SYNC
69         MTMSRD(r5)
70
71         addi    r7,r3,THREAD_TRANSACT_FPSTATE
72         lfd     fr0,FPSTATE_FPSCR(r7)
73         MTFSF_L(fr0)
74         REST_32FPVSRS(0, R4, R7)
75
76         blr
77 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
78
79 /*
80  * Load state from memory into FP registers including FPSCR.
81  * Assumes the caller has enabled FP in the MSR.
82  */
83 _GLOBAL(load_fp_state)
84         lfd     fr0,FPSTATE_FPSCR(r3)
85         MTFSF_L(fr0)
86         REST_32FPVSRS(0, R4, R3)
87         blr
88
89 /*
90  * Store FP state into memory, including FPSCR
91  * Assumes the caller has enabled FP in the MSR.
92  */
93 _GLOBAL(store_fp_state)
94         SAVE_32FPVSRS(0, R4, R3)
95         mffs    fr0
96         stfd    fr0,FPSTATE_FPSCR(r3)
97         blr
98
99 /*
100  * This task wants to use the FPU now.
101  * On UP, disable FP for the task which had the FPU previously,
102  * and save its floating-point registers in its thread_struct.
103  * Load up this task's FP registers from its thread_struct,
104  * enable the FPU for the current task and return to the task.
105  * Note that on 32-bit this can only use registers that will be
106  * restored by fast_exception_return, i.e. r3 - r6, r10 and r11.
107  */
108 _GLOBAL(load_up_fpu)
109         mfmsr   r5
110         ori     r5,r5,MSR_FP
111 #ifdef CONFIG_VSX
112 BEGIN_FTR_SECTION
113         oris    r5,r5,MSR_VSX@h
114 END_FTR_SECTION_IFSET(CPU_FTR_VSX)
115 #endif
116         SYNC
117         MTMSRD(r5)                      /* enable use of fpu now */
118         isync
119         /* enable use of FP after return */
120 #ifdef CONFIG_PPC32
121         mfspr   r5,SPRN_SPRG_THREAD     /* current task's THREAD (phys) */
122         lwz     r4,THREAD_FPEXC_MODE(r5)
123         ori     r9,r9,MSR_FP            /* enable FP for current */
124         or      r9,r9,r4
125 #else
126         ld      r4,PACACURRENT(r13)
127         addi    r5,r4,THREAD            /* Get THREAD */
128         lwz     r4,THREAD_FPEXC_MODE(r5)
129         ori     r12,r12,MSR_FP
130         or      r12,r12,r4
131         std     r12,_MSR(r1)
132 #endif
133         /* Don't care if r4 overflows, this is desired behaviour */
134         lbz     r4,THREAD_LOAD_FP(r5)
135         addi    r4,r4,1
136         stb     r4,THREAD_LOAD_FP(r5)
137         addi    r10,r5,THREAD_FPSTATE
138         lfd     fr0,FPSTATE_FPSCR(r10)
139         MTFSF_L(fr0)
140         REST_32FPVSRS(0, R4, R10)
141         /* restore registers and return */
142         /* we haven't used ctr or xer or lr */
143         blr
144
145 /*
146  * save_fpu(tsk)
147  * Save the floating-point registers in its thread_struct.
148  * Enables the FPU for use in the kernel on return.
149  */
150 _GLOBAL(save_fpu)
151         addi    r3,r3,THREAD            /* want THREAD of task */
152         PPC_LL  r6,THREAD_FPSAVEAREA(r3)
153         PPC_LL  r5,PT_REGS(r3)
154         PPC_LCMPI       0,r6,0
155         bne     2f
156         addi    r6,r3,THREAD_FPSTATE
157 2:      SAVE_32FPVSRS(0, R4, R6)
158         mffs    fr0
159         stfd    fr0,FPSTATE_FPSCR(r6)
160         blr
161
162 /*
163  * These are used in the alignment trap handler when emulating
164  * single-precision loads and stores.
165  */
166
167 _GLOBAL(cvt_fd)
168         lfs     0,0(r3)
169         stfd    0,0(r4)
170         blr
171
172 _GLOBAL(cvt_df)
173         lfd     0,0(r3)
174         stfs    0,0(r4)
175         blr