zero PCU area when loading used area.diff -r1.4 -r1.5 src/sys/arch/powerpc/booke/spe.c
(matt)
--- src/sys/arch/powerpc/booke/spe.c 2011/05/02 06:43:16 1.4
+++ src/sys/arch/powerpc/booke/spe.c 2011/06/07 01:01:42 1.5
@@ -1,237 +1,242 @@ | @@ -1,237 +1,242 @@ | |||
1 | /* $NetBSD: spe.c,v 1.4 2011/05/02 06:43:16 matt Exp $ */ | 1 | /* $NetBSD: spe.c,v 1.5 2011/06/07 01:01:42 matt Exp $ */ | |
2 | 2 | |||
3 | /*- | 3 | /*- | |
4 | * Copyright (c) 2011 The NetBSD Foundation, Inc. | 4 | * Copyright (c) 2011 The NetBSD Foundation, Inc. | |
5 | * All rights reserved. | 5 | * All rights reserved. | |
6 | * | 6 | * | |
7 | * This code is derived from software contributed to The NetBSD Foundation | 7 | * This code is derived from software contributed to The NetBSD Foundation | |
8 | * by Matt Thomas of 3am Software Foundry. | 8 | * by Matt Thomas of 3am Software Foundry. | |
9 | * | 9 | * | |
10 | * Redistribution and use in source and binary forms, with or without | 10 | * Redistribution and use in source and binary forms, with or without | |
11 | * modification, are permitted provided that the following conditions | 11 | * modification, are permitted provided that the following conditions | |
12 | * are met: | 12 | * are met: | |
13 | * 1. Redistributions of source code must retain the above copyright | 13 | * 1. Redistributions of source code must retain the above copyright | |
14 | * notice, this list of conditions and the following disclaimer. | 14 | * notice, this list of conditions and the following disclaimer. | |
15 | * 2. Redistributions in binary form must reproduce the above copyright | 15 | * 2. Redistributions in binary form must reproduce the above copyright | |
16 | * notice, this list of conditions and the following disclaimer in the | 16 | * notice, this list of conditions and the following disclaimer in the | |
17 | * documentation and/or other materials provided with the distribution. | 17 | * documentation and/or other materials provided with the distribution. | |
18 | * | 18 | * | |
19 | * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS | 19 | * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS | |
20 | * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED | 20 | * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED | |
21 | * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR | 21 | * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR | |
22 | * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS | 22 | * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS | |
23 | * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR | 23 | * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR | |
24 | * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF | 24 | * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF | |
25 | * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS | 25 | * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS | |
26 | * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN | 26 | * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN | |
27 | * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) | 27 | * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) | |
28 | * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE | 28 | * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE | |
29 | * POSSIBILITY OF SUCH DAMAGE. | 29 | * POSSIBILITY OF SUCH DAMAGE. | |
30 | */ | 30 | */ | |
31 | 31 | |||
32 | #include <sys/cdefs.h> | 32 | #include <sys/cdefs.h> | |
33 | __KERNEL_RCSID(0, "$NetBSD: spe.c,v 1.4 2011/05/02 06:43:16 matt Exp $"); | 33 | __KERNEL_RCSID(0, "$NetBSD: spe.c,v 1.5 2011/06/07 01:01:42 matt Exp $"); | |
34 | 34 | |||
35 | #include "opt_altivec.h" | 35 | #include "opt_altivec.h" | |
36 | 36 | |||
37 | #ifdef PPC_HAVE_SPE | 37 | #ifdef PPC_HAVE_SPE | |
38 | 38 | |||
39 | #include <sys/param.h> | 39 | #include <sys/param.h> | |
40 | #include <sys/proc.h> | 40 | #include <sys/proc.h> | |
41 | #include <sys/systm.h> | 41 | #include <sys/systm.h> | |
42 | #include <sys/atomic.h> | 42 | #include <sys/atomic.h> | |
43 | #include <sys/siginfo.h> | 43 | #include <sys/siginfo.h> | |
44 | #include <sys/pcu.h> | 44 | #include <sys/pcu.h> | |
45 | 45 | |||
46 | #include <powerpc/altivec.h> | 46 | #include <powerpc/altivec.h> | |
47 | #include <powerpc/spr.h> | 47 | #include <powerpc/spr.h> | |
48 | #include <powerpc/booke/spr.h> | 48 | #include <powerpc/booke/spr.h> | |
49 | #include <powerpc/psl.h> | 49 | #include <powerpc/psl.h> | |
50 | #include <powerpc/pcb.h> | 50 | #include <powerpc/pcb.h> | |
51 | 51 | |||
52 | static void vec_state_load(lwp_t *, bool); | 52 | static void vec_state_load(lwp_t *, bool); | |
53 | static void vec_state_save(lwp_t *); | 53 | static void vec_state_save(lwp_t *); | |
54 | static void vec_state_release(lwp_t *); | 54 | static void vec_state_release(lwp_t *); | |
55 | 55 | |||
56 | const pcu_ops_t vec_ops = { | 56 | const pcu_ops_t vec_ops = { | |
57 | .pcu_id = PCU_VEC, | 57 | .pcu_id = PCU_VEC, | |
58 | .pcu_state_load = vec_state_load, | 58 | .pcu_state_load = vec_state_load, | |
59 | .pcu_state_save = vec_state_save, | 59 | .pcu_state_save = vec_state_save, | |
60 | .pcu_state_release = vec_state_release, | 60 | .pcu_state_release = vec_state_release, | |
61 | }; | 61 | }; | |
62 | 62 | |||
63 | bool | 63 | bool | |
64 | vec_used_p(lwp_t *l) | 64 | vec_used_p(lwp_t *l) | |
65 | { | 65 | { | |
66 | return (l->l_md.md_flags & MDLWP_USEDVEC) != 0; | 66 | return (l->l_md.md_flags & MDLWP_USEDVEC) != 0; | |
67 | } | 67 | } | |
68 | 68 | |||
69 | void | 69 | void | |
70 | vec_mark_used(lwp_t *l) | 70 | vec_mark_used(lwp_t *l) | |
71 | { | 71 | { | |
72 | l->l_md.md_flags |= MDLWP_USEDVEC; | 72 | l->l_md.md_flags |= MDLWP_USEDVEC; | |
73 | } | 73 | } | |
74 | 74 | |||
75 | void | 75 | void | |
76 | vec_state_load(lwp_t *l, bool used) | 76 | vec_state_load(lwp_t *l, bool used) | |
77 | { | 77 | { | |
78 | struct pcb * const pcb = lwp_getpcb(l); | 78 | struct pcb * const pcb = lwp_getpcb(l); | |
79 | 79 | |||
80 | if (__predict_false(!vec_used_p(l))) { | |||
81 | memset(&pcb->pcb_vr, 0, sizeof(pcb->pcb_vr)); | |||
82 | vec_mark_used(l); | |||
83 | } | |||
84 | ||||
80 | /* | 85 | /* | |
81 | * Enable SPE temporarily (and disable interrupts). | 86 | * Enable SPE temporarily (and disable interrupts). | |
82 | */ | 87 | */ | |
83 | const register_t msr = mfmsr(); | 88 | const register_t msr = mfmsr(); | |
84 | mtmsr((msr & ~PSL_EE) | PSL_SPV); | 89 | mtmsr((msr & ~PSL_EE) | PSL_SPV); | |
85 | __asm volatile ("isync"); | 90 | __asm volatile ("isync"); | |
86 | 91 | |||
87 | /* | 92 | /* | |
88 | * Call an assembly routine to do load everything. | 93 | * Call an assembly routine to do load everything. | |
89 | */ | 94 | */ | |
90 | vec_load_from_vreg(&pcb->pcb_vr); | 95 | vec_load_from_vreg(&pcb->pcb_vr); | |
91 | __asm volatile ("sync"); | 96 | __asm volatile ("sync"); | |
92 | 97 | |||
93 | 98 | |||
94 | /* | 99 | /* | |
95 | * Restore MSR (turn off SPE) | 100 | * Restore MSR (turn off SPE) | |
96 | */ | 101 | */ | |
97 | mtmsr(msr); | 102 | mtmsr(msr); | |
98 | __asm volatile ("isync"); | 103 | __asm volatile ("isync"); | |
99 | 104 | |||
100 | /* | 105 | /* | |
101 | * Note that vector has now been used. | 106 | * Note that vector has now been used. | |
102 | */ | 107 | */ | |
103 | l->l_md.md_flags |= MDLWP_USEDVEC; | 108 | l->l_md.md_flags |= MDLWP_USEDVEC; | |
104 | l->l_md.md_utf->tf_srr1 |= PSL_SPV; | 109 | l->l_md.md_utf->tf_srr1 |= PSL_SPV; | |
105 | } | 110 | } | |
106 | 111 | |||
107 | void | 112 | void | |
108 | vec_state_save(lwp_t *l) | 113 | vec_state_save(lwp_t *l) | |
109 | { | 114 | { | |
110 | struct pcb * const pcb = lwp_getpcb(l); | 115 | struct pcb * const pcb = lwp_getpcb(l); | |
111 | 116 | |||
112 | /* | 117 | /* | |
113 | * Turn on SPE, turn off interrupts. | 118 | * Turn on SPE, turn off interrupts. | |
114 | */ | 119 | */ | |
115 | const register_t msr = mfmsr(); | 120 | const register_t msr = mfmsr(); | |
116 | mtmsr((msr & ~PSL_EE) | PSL_SPV); | 121 | mtmsr((msr & ~PSL_EE) | PSL_SPV); | |
117 | __asm volatile ("isync"); | 122 | __asm volatile ("isync"); | |
118 | 123 | |||
119 | /* | 124 | /* | |
120 | * Save the vector state which is best done in assembly. | 125 | * Save the vector state which is best done in assembly. | |
121 | */ | 126 | */ | |
122 | vec_unload_to_vreg(&pcb->pcb_vr); | 127 | vec_unload_to_vreg(&pcb->pcb_vr); | |
123 | __asm volatile ("sync"); | 128 | __asm volatile ("sync"); | |
124 | 129 | |||
125 | /* | 130 | /* | |
126 | * Restore MSR (turn off SPE) | 131 | * Restore MSR (turn off SPE) | |
127 | */ | 132 | */ | |
128 | mtmsr(msr); | 133 | mtmsr(msr); | |
129 | __asm volatile ("isync"); | 134 | __asm volatile ("isync"); | |
130 | } | 135 | } | |
131 | 136 | |||
132 | void | 137 | void | |
133 | vec_state_release(lwp_t *l) | 138 | vec_state_release(lwp_t *l) | |
134 | { | 139 | { | |
135 | /* | 140 | /* | |
136 | * Turn off SPV so the next SPE instruction will cause a | 141 | * Turn off SPV so the next SPE instruction will cause a | |
137 | * SPE unavailable exception | 142 | * SPE unavailable exception | |
138 | */ | 143 | */ | |
139 | l->l_md.md_utf->tf_srr1 &= ~PSL_SPV; | 144 | l->l_md.md_utf->tf_srr1 &= ~PSL_SPV; | |
140 | } | 145 | } | |
141 | 146 | |||
142 | void | 147 | void | |
143 | vec_restore_from_mcontext(lwp_t *l, const mcontext_t *mcp) | 148 | vec_restore_from_mcontext(lwp_t *l, const mcontext_t *mcp) | |
144 | { | 149 | { | |
145 | struct pcb * const pcb = lwp_getpcb(l); | 150 | struct pcb * const pcb = lwp_getpcb(l); | |
146 | const union __vr *vr = mcp->__vrf.__vrs; | 151 | const union __vr *vr = mcp->__vrf.__vrs; | |
147 | 152 | |||
148 | KASSERT(l == curlwp); | 153 | KASSERT(l == curlwp); | |
149 | 154 | |||
150 | vec_save(); | 155 | vec_save(); | |
151 | 156 | |||
152 | /* grab the accumulator */ | 157 | /* grab the accumulator */ | |
153 | pcb->pcb_vr.vreg[8][0] = vr->__vr32[2]; | 158 | pcb->pcb_vr.vreg[8][0] = vr->__vr32[2]; | |
154 | pcb->pcb_vr.vreg[8][1] = vr->__vr32[3]; | 159 | pcb->pcb_vr.vreg[8][1] = vr->__vr32[3]; | |
155 | 160 | |||
156 | /* | 161 | /* | |
157 | * We store the high parts of each register in the first 8 vectors. | 162 | * We store the high parts of each register in the first 8 vectors. | |
158 | */ | 163 | */ | |
159 | for (u_int i = 0; i < 8; i++, vr += 4) { | 164 | for (u_int i = 0; i < 8; i++, vr += 4) { | |
160 | pcb->pcb_vr.vreg[i][0] = vr[0].__vr32[0]; | 165 | pcb->pcb_vr.vreg[i][0] = vr[0].__vr32[0]; | |
161 | pcb->pcb_vr.vreg[i][1] = vr[1].__vr32[0]; | 166 | pcb->pcb_vr.vreg[i][1] = vr[1].__vr32[0]; | |
162 | pcb->pcb_vr.vreg[i][2] = vr[2].__vr32[0]; | 167 | pcb->pcb_vr.vreg[i][2] = vr[2].__vr32[0]; | |
163 | pcb->pcb_vr.vreg[i][3] = vr[3].__vr32[0]; | 168 | pcb->pcb_vr.vreg[i][3] = vr[3].__vr32[0]; | |
164 | } | 169 | } | |
165 | l->l_md.md_utf->tf_spefscr = pcb->pcb_vr.vscr = mcp->__vrf.__vscr; | 170 | l->l_md.md_utf->tf_spefscr = pcb->pcb_vr.vscr = mcp->__vrf.__vscr; | |
166 | pcb->pcb_vr.vrsave = mcp->__vrf.__vrsave; | 171 | pcb->pcb_vr.vrsave = mcp->__vrf.__vrsave; | |
167 | } | 172 | } | |
168 | 173 | |||
169 | bool | 174 | bool | |
170 | vec_save_to_mcontext(lwp_t *l, mcontext_t *mcp, unsigned int *flagp) | 175 | vec_save_to_mcontext(lwp_t *l, mcontext_t *mcp, unsigned int *flagp) | |
171 | { | 176 | { | |
172 | struct pcb * const pcb = lwp_getpcb(l); | 177 | struct pcb * const pcb = lwp_getpcb(l); | |
173 | 178 | |||
174 | KASSERT(l == curlwp); | 179 | KASSERT(l == curlwp); | |
175 | 180 | |||
176 | if (!vec_used_p(l)) | 181 | if (!vec_used_p(l)) | |
177 | return false; | 182 | return false; | |
178 | 183 | |||
179 | vec_save(); | 184 | vec_save(); | |
180 | 185 | |||
181 | mcp->__gregs[_REG_MSR] |= PSL_SPV; | 186 | mcp->__gregs[_REG_MSR] |= PSL_SPV; | |
182 | 187 | |||
183 | union __vr *vr = mcp->__vrf.__vrs; | 188 | union __vr *vr = mcp->__vrf.__vrs; | |
184 | const register_t *fixreg = l->l_md.md_utf->tf_fixreg; | 189 | const register_t *fixreg = l->l_md.md_utf->tf_fixreg; | |
185 | for (u_int i = 0; i < 32; i++, vr += 4, fixreg += 4) { | 190 | for (u_int i = 0; i < 32; i++, vr += 4, fixreg += 4) { | |
186 | vr[0].__vr32[0] = pcb->pcb_vr.vreg[i][0]; | 191 | vr[0].__vr32[0] = pcb->pcb_vr.vreg[i][0]; | |
187 | vr[0].__vr32[1] = fixreg[0]; | 192 | vr[0].__vr32[1] = fixreg[0]; | |
188 | vr[0].__vr32[2] = 0; | 193 | vr[0].__vr32[2] = 0; | |
189 | vr[0].__vr32[3] = 0; | 194 | vr[0].__vr32[3] = 0; | |
190 | vr[1].__vr32[0] = pcb->pcb_vr.vreg[i][1]; | 195 | vr[1].__vr32[0] = pcb->pcb_vr.vreg[i][1]; | |
191 | vr[1].__vr32[1] = fixreg[1]; | 196 | vr[1].__vr32[1] = fixreg[1]; | |
192 | vr[1].__vr32[2] = 0; | 197 | vr[1].__vr32[2] = 0; | |
193 | vr[1].__vr32[3] = 0; | 198 | vr[1].__vr32[3] = 0; | |
194 | vr[2].__vr32[0] = pcb->pcb_vr.vreg[i][2]; | 199 | vr[2].__vr32[0] = pcb->pcb_vr.vreg[i][2]; | |
195 | vr[2].__vr32[1] = fixreg[2]; | 200 | vr[2].__vr32[1] = fixreg[2]; | |
196 | vr[2].__vr32[2] = 0; | 201 | vr[2].__vr32[2] = 0; | |
197 | vr[2].__vr32[3] = 0; | 202 | vr[2].__vr32[3] = 0; | |
198 | vr[3].__vr32[0] = pcb->pcb_vr.vreg[i][3]; | 203 | vr[3].__vr32[0] = pcb->pcb_vr.vreg[i][3]; | |
199 | vr[3].__vr32[1] = fixreg[3]; | 204 | vr[3].__vr32[1] = fixreg[3]; | |
200 | vr[3].__vr32[2] = 0; | 205 | vr[3].__vr32[2] = 0; | |
201 | vr[3].__vr32[3] = 0; | 206 | vr[3].__vr32[3] = 0; | |
202 | } | 207 | } | |
203 | 208 | |||
204 | mcp->__vrf.__vrs[0].__vr32[2] = pcb->pcb_vr.vreg[8][0]; | 209 | mcp->__vrf.__vrs[0].__vr32[2] = pcb->pcb_vr.vreg[8][0]; | |
205 | mcp->__vrf.__vrs[0].__vr32[3] = pcb->pcb_vr.vreg[8][1]; | 210 | mcp->__vrf.__vrs[0].__vr32[3] = pcb->pcb_vr.vreg[8][1]; | |
206 | 211 | |||
207 | mcp->__vrf.__vrsave = pcb->pcb_vr.vrsave; | 212 | mcp->__vrf.__vrsave = pcb->pcb_vr.vrsave; | |
208 | mcp->__vrf.__vscr = l->l_md.md_utf->tf_spefscr; | 213 | mcp->__vrf.__vscr = l->l_md.md_utf->tf_spefscr; | |
209 | 214 | |||
210 | *flagp |= _UC_POWERPC_SPE; | 215 | *flagp |= _UC_POWERPC_SPE; | |
211 | 216 | |||
212 | return true; | 217 | return true; | |
213 | } | 218 | } | |
214 | 219 | |||
215 | static const struct { | 220 | static const struct { | |
216 | uint32_t mask; | 221 | uint32_t mask; | |
217 | int code; | 222 | int code; | |
218 | } spefscr_siginfo_map[] = { | 223 | } spefscr_siginfo_map[] = { | |
219 | { SPEFSCR_FINV|SPEFSCR_FINVH, FPE_FLTINV }, | 224 | { SPEFSCR_FINV|SPEFSCR_FINVH, FPE_FLTINV }, | |
220 | { SPEFSCR_FOVF|SPEFSCR_FOVFH, FPE_FLTOVF }, | 225 | { SPEFSCR_FOVF|SPEFSCR_FOVFH, FPE_FLTOVF }, | |
221 | { SPEFSCR_FUNF|SPEFSCR_FUNFH, FPE_FLTUND }, | 226 | { SPEFSCR_FUNF|SPEFSCR_FUNFH, FPE_FLTUND }, | |
222 | { SPEFSCR_FX |SPEFSCR_FXH, FPE_FLTRES }, | 227 | { SPEFSCR_FX |SPEFSCR_FXH, FPE_FLTRES }, | |
223 | { SPEFSCR_FDBZ|SPEFSCR_FDBZH, FPE_FLTDIV }, | 228 | { SPEFSCR_FDBZ|SPEFSCR_FDBZH, FPE_FLTDIV }, | |
224 | { SPEFSCR_OV |SPEFSCR_OVH, FPE_INTOVF }, | 229 | { SPEFSCR_OV |SPEFSCR_OVH, FPE_INTOVF }, | |
225 | }; | 230 | }; | |
226 | 231 | |||
227 | int | 232 | int | |
228 | vec_siginfo_code(const struct trapframe *tf) | 233 | vec_siginfo_code(const struct trapframe *tf) | |
229 | { | 234 | { | |
230 | for (u_int i = 0; i < __arraycount(spefscr_siginfo_map); i++) { | 235 | for (u_int i = 0; i < __arraycount(spefscr_siginfo_map); i++) { | |
231 | if (tf->tf_spefscr & spefscr_siginfo_map[i].mask) | 236 | if (tf->tf_spefscr & spefscr_siginfo_map[i].mask) | |
232 | return spefscr_siginfo_map[i].code; | 237 | return spefscr_siginfo_map[i].code; | |
233 | } | 238 | } | |
234 | return 0; | 239 | return 0; | |
235 | } | 240 | } | |
236 | 241 | |||
237 | #endif /* PPC_HAVE_SPE */ | 242 | #endif /* PPC_HAVE_SPE */ |
--- src/sys/arch/powerpc/oea/altivec.c 2011/05/25 05:42:37 1.24
+++ src/sys/arch/powerpc/oea/altivec.c 2011/06/07 01:01:43 1.25
@@ -1,314 +1,319 @@ | @@ -1,314 +1,319 @@ | |||
1 | /* $NetBSD: altivec.c,v 1.24 2011/05/25 05:42:37 matt Exp $ */ | 1 | /* $NetBSD: altivec.c,v 1.25 2011/06/07 01:01:43 matt Exp $ */ | |
2 | 2 | |||
3 | /* | 3 | /* | |
4 | * Copyright (C) 1996 Wolfgang Solfrank. | 4 | * Copyright (C) 1996 Wolfgang Solfrank. | |
5 | * Copyright (C) 1996 TooLs GmbH. | 5 | * Copyright (C) 1996 TooLs GmbH. | |
6 | * All rights reserved. | 6 | * All rights reserved. | |
7 | * | 7 | * | |
8 | * Redistribution and use in source and binary forms, with or without | 8 | * Redistribution and use in source and binary forms, with or without | |
9 | * modification, are permitted provided that the following conditions | 9 | * modification, are permitted provided that the following conditions | |
10 | * are met: | 10 | * are met: | |
11 | * 1. Redistributions of source code must retain the above copyright | 11 | * 1. Redistributions of source code must retain the above copyright | |
12 | * notice, this list of conditions and the following disclaimer. | 12 | * notice, this list of conditions and the following disclaimer. | |
13 | * 2. Redistributions in binary form must reproduce the above copyright | 13 | * 2. Redistributions in binary form must reproduce the above copyright | |
14 | * notice, this list of conditions and the following disclaimer in the | 14 | * notice, this list of conditions and the following disclaimer in the | |
15 | * documentation and/or other materials provided with the distribution. | 15 | * documentation and/or other materials provided with the distribution. | |
16 | * 3. All advertising materials mentioning features or use of this software | 16 | * 3. All advertising materials mentioning features or use of this software | |
17 | * must display the following acknowledgement: | 17 | * must display the following acknowledgement: | |
18 | * This product includes software developed by TooLs GmbH. | 18 | * This product includes software developed by TooLs GmbH. | |
19 | * 4. The name of TooLs GmbH may not be used to endorse or promote products | 19 | * 4. The name of TooLs GmbH may not be used to endorse or promote products | |
20 | * derived from this software without specific prior written permission. | 20 | * derived from this software without specific prior written permission. | |
21 | * | 21 | * | |
22 | * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR | 22 | * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR | |
23 | * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES | 23 | * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES | |
24 | * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. | 24 | * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. | |
25 | * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, | 25 | * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, | |
26 | * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, | 26 | * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, | |
27 | * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; | 27 | * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; | |
28 | * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, | 28 | * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, | |
29 | * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR | 29 | * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR | |
30 | * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF | 30 | * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF | |
31 | * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. | 31 | * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. | |
32 | */ | 32 | */ | |
33 | 33 | |||
34 | #include <sys/cdefs.h> | 34 | #include <sys/cdefs.h> | |
35 | __KERNEL_RCSID(0, "$NetBSD: altivec.c,v 1.24 2011/05/25 05:42:37 matt Exp $"); | 35 | __KERNEL_RCSID(0, "$NetBSD: altivec.c,v 1.25 2011/06/07 01:01:43 matt Exp $"); | |
36 | 36 | |||
37 | #include "opt_multiprocessor.h" | 37 | #include "opt_multiprocessor.h" | |
38 | 38 | |||
39 | #include <sys/param.h> | 39 | #include <sys/param.h> | |
40 | #include <sys/proc.h> | 40 | #include <sys/proc.h> | |
41 | #include <sys/systm.h> | 41 | #include <sys/systm.h> | |
42 | #include <sys/atomic.h> | 42 | #include <sys/atomic.h> | |
43 | 43 | |||
44 | #include <uvm/uvm_extern.h> /* for vcopypage/vzeropage */ | 44 | #include <uvm/uvm_extern.h> /* for vcopypage/vzeropage */ | |
45 | 45 | |||
46 | #include <powerpc/pcb.h> | 46 | #include <powerpc/pcb.h> | |
47 | #include <powerpc/altivec.h> | 47 | #include <powerpc/altivec.h> | |
48 | #include <powerpc/spr.h> | 48 | #include <powerpc/spr.h> | |
49 | #include <powerpc/oea/spr.h> | 49 | #include <powerpc/oea/spr.h> | |
50 | #include <powerpc/psl.h> | 50 | #include <powerpc/psl.h> | |
51 | 51 | |||
52 | static void vec_state_load(lwp_t *, bool); | 52 | static void vec_state_load(lwp_t *, bool); | |
53 | static void vec_state_save(lwp_t *); | 53 | static void vec_state_save(lwp_t *); | |
54 | static void vec_state_release(lwp_t *); | 54 | static void vec_state_release(lwp_t *); | |
55 | 55 | |||
56 | const pcu_ops_t vec_ops = { | 56 | const pcu_ops_t vec_ops = { | |
57 | .pcu_id = PCU_VEC, | 57 | .pcu_id = PCU_VEC, | |
58 | .pcu_state_load = vec_state_load, | 58 | .pcu_state_load = vec_state_load, | |
59 | .pcu_state_save = vec_state_save, | 59 | .pcu_state_save = vec_state_save, | |
60 | .pcu_state_release = vec_state_release, | 60 | .pcu_state_release = vec_state_release, | |
61 | }; | 61 | }; | |
62 | 62 | |||
63 | bool | 63 | bool | |
64 | vec_used_p(lwp_t *l) | 64 | vec_used_p(lwp_t *l) | |
65 | { | 65 | { | |
66 | return (l->l_md.md_flags & MDLWP_USEDVEC) != 0; | 66 | return (l->l_md.md_flags & MDLWP_USEDVEC) != 0; | |
67 | } | 67 | } | |
68 | 68 | |||
69 | void | 69 | void | |
70 | vec_mark_used(lwp_t *l) | 70 | vec_mark_used(lwp_t *l) | |
71 | { | 71 | { | |
72 | l->l_md.md_flags |= MDLWP_USEDVEC; | 72 | l->l_md.md_flags |= MDLWP_USEDVEC; | |
73 | } | 73 | } | |
74 | 74 | |||
75 | void | 75 | void | |
76 | vec_state_load(lwp_t *l, bool used) | 76 | vec_state_load(lwp_t *l, bool used) | |
77 | { | 77 | { | |
78 | struct pcb * const pcb = lwp_getpcb(l); | 78 | struct pcb * const pcb = lwp_getpcb(l); | |
79 | 79 | |||
80 | if (__predict_false(!vec_used_p(l))) { | |||
81 | memset(&pcb->pcb_vr, 0, sizeof(pcb->pcb_vr)); | |||
82 | vec_mark_used(l); | |||
83 | } | |||
84 | ||||
80 | /* | 85 | /* | |
81 | * Enable AltiVec temporarily (and disable interrupts). | 86 | * Enable AltiVec temporarily (and disable interrupts). | |
82 | */ | 87 | */ | |
83 | const register_t msr = mfmsr(); | 88 | const register_t msr = mfmsr(); | |
84 | mtmsr((msr & ~PSL_EE) | PSL_VEC); | 89 | mtmsr((msr & ~PSL_EE) | PSL_VEC); | |
85 | __asm volatile ("isync"); | 90 | __asm volatile ("isync"); | |
86 | 91 | |||
87 | /* | 92 | /* | |
88 | * Load the vector unit from vreg which is best done in | 93 | * Load the vector unit from vreg which is best done in | |
89 | * assembly. | 94 | * assembly. | |
90 | */ | 95 | */ | |
91 | vec_load_from_vreg(&pcb->pcb_vr); | 96 | vec_load_from_vreg(&pcb->pcb_vr); | |
92 | 97 | |||
93 | /* | 98 | /* | |
94 | * VRSAVE will be restored when trap frame returns | 99 | * VRSAVE will be restored when trap frame returns | |
95 | */ | 100 | */ | |
96 | l->l_md.md_utf->tf_vrsave = pcb->pcb_vr.vrsave; | 101 | l->l_md.md_utf->tf_vrsave = pcb->pcb_vr.vrsave; | |
97 | 102 | |||
98 | /* | 103 | /* | |
99 | * Restore MSR (turn off AltiVec) | 104 | * Restore MSR (turn off AltiVec) | |
100 | */ | 105 | */ | |
101 | mtmsr(msr); | 106 | mtmsr(msr); | |
102 | __asm volatile ("isync"); | 107 | __asm volatile ("isync"); | |
103 | 108 | |||
104 | /* | 109 | /* | |
105 | * Mark vector registers as modified. | 110 | * Mark vector registers as modified. | |
106 | */ | 111 | */ | |
107 | l->l_md.md_flags |= MDLWP_USEDVEC|PSL_VEC; | 112 | l->l_md.md_flags |= MDLWP_USEDVEC|PSL_VEC; | |
108 | l->l_md.md_utf->tf_srr1 |= PSL_VEC; | 113 | l->l_md.md_utf->tf_srr1 |= PSL_VEC; | |
109 | } | 114 | } | |
110 | 115 | |||
111 | void | 116 | void | |
112 | vec_state_save(lwp_t *l) | 117 | vec_state_save(lwp_t *l) | |
113 | { | 118 | { | |
114 | struct pcb * const pcb = lwp_getpcb(l); | 119 | struct pcb * const pcb = lwp_getpcb(l); | |
115 | 120 | |||
116 | /* | 121 | /* | |
117 | * Turn on AltiVEC, turn off interrupts. | 122 | * Turn on AltiVEC, turn off interrupts. | |
118 | */ | 123 | */ | |
119 | const register_t msr = mfmsr(); | 124 | const register_t msr = mfmsr(); | |
120 | mtmsr((msr & ~PSL_EE) | PSL_VEC); | 125 | mtmsr((msr & ~PSL_EE) | PSL_VEC); | |
121 | __asm volatile ("isync"); | 126 | __asm volatile ("isync"); | |
122 | 127 | |||
123 | /* | 128 | /* | |
124 | * Grab contents of vector unit. | 129 | * Grab contents of vector unit. | |
125 | */ | 130 | */ | |
126 | vec_unload_to_vreg(&pcb->pcb_vr); | 131 | vec_unload_to_vreg(&pcb->pcb_vr); | |
127 | 132 | |||
128 | /* | 133 | /* | |
129 | * Save VRSAVE | 134 | * Save VRSAVE | |
130 | */ | 135 | */ | |
131 | pcb->pcb_vr.vrsave = l->l_md.md_utf->tf_vrsave; | 136 | pcb->pcb_vr.vrsave = l->l_md.md_utf->tf_vrsave; | |
132 | 137 | |||
133 | /* | 138 | /* | |
134 | * Note that we aren't using any CPU resources and stop any | 139 | * Note that we aren't using any CPU resources and stop any | |
135 | * data streams. | 140 | * data streams. | |
136 | */ | 141 | */ | |
137 | __asm volatile ("dssall; sync"); | 142 | __asm volatile ("dssall; sync"); | |
138 | 143 | |||
139 | /* | 144 | /* | |
140 | * Restore MSR (turn off AltiVec) | 145 | * Restore MSR (turn off AltiVec) | |
141 | */ | 146 | */ | |
142 | mtmsr(msr); | 147 | mtmsr(msr); | |
143 | __asm volatile ("isync"); | 148 | __asm volatile ("isync"); | |
144 | } | 149 | } | |
145 | 150 | |||
146 | void | 151 | void | |
147 | vec_state_release(lwp_t *l) | 152 | vec_state_release(lwp_t *l) | |
148 | { | 153 | { | |
149 | __asm volatile("dssall;sync"); | 154 | __asm volatile("dssall;sync"); | |
150 | l->l_md.md_utf->tf_srr1 &= ~PSL_VEC; | 155 | l->l_md.md_utf->tf_srr1 &= ~PSL_VEC; | |
151 | l->l_md.md_flags &= ~PSL_VEC; | 156 | l->l_md.md_flags &= ~PSL_VEC; | |
152 | } | 157 | } | |
153 | 158 | |||
154 | void | 159 | void | |
155 | vec_restore_from_mcontext(struct lwp *l, const mcontext_t *mcp) | 160 | vec_restore_from_mcontext(struct lwp *l, const mcontext_t *mcp) | |
156 | { | 161 | { | |
157 | struct pcb * const pcb = lwp_getpcb(l); | 162 | struct pcb * const pcb = lwp_getpcb(l); | |
158 | 163 | |||
159 | KASSERT(l == curlwp); | 164 | KASSERT(l == curlwp); | |
160 | 165 | |||
161 | /* we don't need to save the state, just drop it */ | 166 | /* we don't need to save the state, just drop it */ | |
162 | pcu_discard(&vec_ops); | 167 | pcu_discard(&vec_ops); | |
163 | memcpy(pcb->pcb_vr.vreg, &mcp->__vrf.__vrs, sizeof (pcb->pcb_vr.vreg)); | 168 | memcpy(pcb->pcb_vr.vreg, &mcp->__vrf.__vrs, sizeof (pcb->pcb_vr.vreg)); | |
164 | pcb->pcb_vr.vscr = mcp->__vrf.__vscr; | 169 | pcb->pcb_vr.vscr = mcp->__vrf.__vscr; | |
165 | pcb->pcb_vr.vrsave = mcp->__vrf.__vrsave; | 170 | pcb->pcb_vr.vrsave = mcp->__vrf.__vrsave; | |
166 | l->l_md.md_utf->tf_vrsave = pcb->pcb_vr.vrsave; | 171 | l->l_md.md_utf->tf_vrsave = pcb->pcb_vr.vrsave; | |
167 | } | 172 | } | |
168 | 173 | |||
169 | bool | 174 | bool | |
170 | vec_save_to_mcontext(struct lwp *l, mcontext_t *mcp, unsigned int *flagp) | 175 | vec_save_to_mcontext(struct lwp *l, mcontext_t *mcp, unsigned int *flagp) | |
171 | { | 176 | { | |
172 | struct pcb * const pcb = lwp_getpcb(l); | 177 | struct pcb * const pcb = lwp_getpcb(l); | |
173 | 178 | |||
174 | KASSERT(l == curlwp); | 179 | KASSERT(l == curlwp); | |
175 | 180 | |||
176 | /* Save AltiVec context, if any. */ | 181 | /* Save AltiVec context, if any. */ | |
177 | if (!vec_used_p(l)) | 182 | if (!vec_used_p(l)) | |
178 | return false; | 183 | return false; | |
179 | 184 | |||
180 | /* | 185 | /* | |
181 | * If we're the AltiVec owner, dump its context to the PCB first. | 186 | * If we're the AltiVec owner, dump its context to the PCB first. | |
182 | */ | 187 | */ | |
183 | pcu_save(&vec_ops); | 188 | pcu_save(&vec_ops); | |
184 | 189 | |||
185 | mcp->__gregs[_REG_MSR] |= PSL_VEC; | 190 | mcp->__gregs[_REG_MSR] |= PSL_VEC; | |
186 | mcp->__vrf.__vscr = pcb->pcb_vr.vscr; | 191 | mcp->__vrf.__vscr = pcb->pcb_vr.vscr; | |
187 | mcp->__vrf.__vrsave = l->l_md.md_utf->tf_vrsave; | 192 | mcp->__vrf.__vrsave = l->l_md.md_utf->tf_vrsave; | |
188 | memcpy(mcp->__vrf.__vrs, pcb->pcb_vr.vreg, sizeof (mcp->__vrf.__vrs)); | 193 | memcpy(mcp->__vrf.__vrs, pcb->pcb_vr.vreg, sizeof (mcp->__vrf.__vrs)); | |
189 | *flagp |= _UC_POWERPC_VEC; | 194 | *flagp |= _UC_POWERPC_VEC; | |
190 | return true; | 195 | return true; | |
191 | } | 196 | } | |
192 | 197 | |||
193 | #define ZERO_VEC 19 | 198 | #define ZERO_VEC 19 | |
194 | 199 | |||
195 | void | 200 | void | |
196 | vzeropage(paddr_t pa) | 201 | vzeropage(paddr_t pa) | |
197 | { | 202 | { | |
198 | const paddr_t ea = pa + PAGE_SIZE; | 203 | const paddr_t ea = pa + PAGE_SIZE; | |
199 | uint32_t vec[7], *vp = (void *) roundup((uintptr_t) vec, 16); | 204 | uint32_t vec[7], *vp = (void *) roundup((uintptr_t) vec, 16); | |
200 | register_t omsr, msr; | 205 | register_t omsr, msr; | |
201 | 206 | |||
202 | __asm volatile("mfmsr %0" : "=r"(omsr) :); | 207 | __asm volatile("mfmsr %0" : "=r"(omsr) :); | |
203 | 208 | |||
204 | /* | 209 | /* | |
205 | * Turn on AltiVec, turn off interrupts. | 210 | * Turn on AltiVec, turn off interrupts. | |
206 | */ | 211 | */ | |
207 | msr = (omsr & ~PSL_EE) | PSL_VEC; | 212 | msr = (omsr & ~PSL_EE) | PSL_VEC; | |
208 | __asm volatile("sync; mtmsr %0; isync" :: "r"(msr)); | 213 | __asm volatile("sync; mtmsr %0; isync" :: "r"(msr)); | |
209 | 214 | |||
210 | /* | 215 | /* | |
211 | * Save the VEC register we are going to use before we disable | 216 | * Save the VEC register we are going to use before we disable | |
212 | * relocation. | 217 | * relocation. | |
213 | */ | 218 | */ | |
214 | __asm("stvx %1,0,%0" :: "r"(vp), "n"(ZERO_VEC)); | 219 | __asm("stvx %1,0,%0" :: "r"(vp), "n"(ZERO_VEC)); | |
215 | __asm("vxor %0,%0,%0" :: "n"(ZERO_VEC)); | 220 | __asm("vxor %0,%0,%0" :: "n"(ZERO_VEC)); | |
216 | 221 | |||
217 | /* | 222 | /* | |
218 | * Zero the page using a single cache line. | 223 | * Zero the page using a single cache line. | |
219 | */ | 224 | */ | |
220 | __asm volatile( | 225 | __asm volatile( | |
221 | " sync ;" | 226 | " sync ;" | |
222 | " mfmsr %[msr];" | 227 | " mfmsr %[msr];" | |
223 | " rlwinm %[msr],%[msr],0,28,26;" /* Clear PSL_DR */ | 228 | " rlwinm %[msr],%[msr],0,28,26;" /* Clear PSL_DR */ | |
224 | " mtmsr %[msr];" /* Turn off DMMU */ | 229 | " mtmsr %[msr];" /* Turn off DMMU */ | |
225 | " isync;" | 230 | " isync;" | |
226 | "1: stvx %[zv], %[pa], %[off0];" | 231 | "1: stvx %[zv], %[pa], %[off0];" | |
227 | " stvxl %[zv], %[pa], %[off16];" | 232 | " stvxl %[zv], %[pa], %[off16];" | |
228 | " stvx %[zv], %[pa], %[off32];" | 233 | " stvx %[zv], %[pa], %[off32];" | |
229 | " stvxl %[zv], %[pa], %[off48];" | 234 | " stvxl %[zv], %[pa], %[off48];" | |
230 | " addi %[pa], %[pa], 64;" | 235 | " addi %[pa], %[pa], 64;" | |
231 | " cmplw %[pa], %[ea];" | 236 | " cmplw %[pa], %[ea];" | |
232 | " blt+ 1b;" | 237 | " blt+ 1b;" | |
233 | " ori %[msr], %[msr], 0x10;" /* Set PSL_DR */ | 238 | " ori %[msr], %[msr], 0x10;" /* Set PSL_DR */ | |
234 | " sync;" | 239 | " sync;" | |
235 | " mtmsr %[msr];" /* Turn on DMMU */ | 240 | " mtmsr %[msr];" /* Turn on DMMU */ | |
236 | " isync;" | 241 | " isync;" | |
237 | :: [msr] "r"(msr), [pa] "b"(pa), [ea] "b"(ea), | 242 | :: [msr] "r"(msr), [pa] "b"(pa), [ea] "b"(ea), | |
238 | [off0] "r"(0), [off16] "r"(16), [off32] "r"(32), [off48] "r"(48), | 243 | [off0] "r"(0), [off16] "r"(16), [off32] "r"(32), [off48] "r"(48), | |
239 | [zv] "n"(ZERO_VEC)); | 244 | [zv] "n"(ZERO_VEC)); | |
240 | 245 | |||
241 | /* | 246 | /* | |
242 | * Restore VEC register (now that we can access the stack again). | 247 | * Restore VEC register (now that we can access the stack again). | |
243 | */ | 248 | */ | |
244 | __asm("lvx %1,0,%0" :: "r"(vp), "n"(ZERO_VEC)); | 249 | __asm("lvx %1,0,%0" :: "r"(vp), "n"(ZERO_VEC)); | |
245 | 250 | |||
246 | /* | 251 | /* | |
247 | * Restore old MSR (AltiVec OFF). | 252 | * Restore old MSR (AltiVec OFF). | |
248 | */ | 253 | */ | |
249 | __asm volatile("sync; mtmsr %0; isync" :: "r"(omsr)); | 254 | __asm volatile("sync; mtmsr %0; isync" :: "r"(omsr)); | |
250 | } | 255 | } | |
251 | 256 | |||
252 | #define LO_VEC 16 | 257 | #define LO_VEC 16 | |
253 | #define HI_VEC 17 | 258 | #define HI_VEC 17 | |
254 | 259 | |||
255 | void | 260 | void | |
256 | vcopypage(paddr_t dst, paddr_t src) | 261 | vcopypage(paddr_t dst, paddr_t src) | |
257 | { | 262 | { | |
258 | const paddr_t edst = dst + PAGE_SIZE; | 263 | const paddr_t edst = dst + PAGE_SIZE; | |
259 | uint32_t vec[11], *vp = (void *) roundup((uintptr_t) vec, 16); | 264 | uint32_t vec[11], *vp = (void *) roundup((uintptr_t) vec, 16); | |
260 | register_t omsr, msr; | 265 | register_t omsr, msr; | |
261 | 266 | |||
262 | __asm volatile("mfmsr %0" : "=r"(omsr) :); | 267 | __asm volatile("mfmsr %0" : "=r"(omsr) :); | |
263 | 268 | |||
264 | /* | 269 | /* | |
265 | * Turn on AltiVec, turn off interrupts. | 270 | * Turn on AltiVec, turn off interrupts. | |
266 | */ | 271 | */ | |
267 | msr = (omsr & ~PSL_EE) | PSL_VEC; | 272 | msr = (omsr & ~PSL_EE) | PSL_VEC; | |
268 | __asm volatile("sync; mtmsr %0; isync" :: "r"(msr)); | 273 | __asm volatile("sync; mtmsr %0; isync" :: "r"(msr)); | |
269 | 274 | |||
270 | /* | 275 | /* | |
271 | * Save the VEC registers we will be using before we disable | 276 | * Save the VEC registers we will be using before we disable | |
272 | * relocation. | 277 | * relocation. | |
273 | */ | 278 | */ | |
274 | __asm("stvx %2,%1,%0" :: "b"(vp), "r"( 0), "n"(LO_VEC)); | 279 | __asm("stvx %2,%1,%0" :: "b"(vp), "r"( 0), "n"(LO_VEC)); | |
275 | __asm("stvx %2,%1,%0" :: "b"(vp), "r"(16), "n"(HI_VEC)); | 280 | __asm("stvx %2,%1,%0" :: "b"(vp), "r"(16), "n"(HI_VEC)); | |
276 | 281 | |||
277 | /* | 282 | /* | |
278 | * Copy the page using a single cache line, with DMMU | 283 | * Copy the page using a single cache line, with DMMU | |
279 | * disabled. On most PPCs, two vector registers occupy one | 284 | * disabled. On most PPCs, two vector registers occupy one | |
280 | * cache line. | 285 | * cache line. | |
281 | */ | 286 | */ | |
282 | __asm volatile( | 287 | __asm volatile( | |
283 | " sync ;" | 288 | " sync ;" | |
284 | " mfmsr %[msr];" | 289 | " mfmsr %[msr];" | |
285 | " rlwinm %[msr],%[msr],0,28,26;" /* Clear PSL_DR */ | 290 | " rlwinm %[msr],%[msr],0,28,26;" /* Clear PSL_DR */ | |
286 | " mtmsr %[msr];" /* Turn off DMMU */ | 291 | " mtmsr %[msr];" /* Turn off DMMU */ | |
287 | " isync;" | 292 | " isync;" | |
288 | "1: lvx %[lv], %[src], %[off0];" | 293 | "1: lvx %[lv], %[src], %[off0];" | |
289 | " stvx %[lv], %[dst], %[off0];" | 294 | " stvx %[lv], %[dst], %[off0];" | |
290 | " lvxl %[hv], %[src], %[off16];" | 295 | " lvxl %[hv], %[src], %[off16];" | |
291 | " stvxl %[hv], %[dst], %[off16];" | 296 | " stvxl %[hv], %[dst], %[off16];" | |
292 | " addi %[src], %[src], 32;" | 297 | " addi %[src], %[src], 32;" | |
293 | " addi %[dst], %[dst], 32;" | 298 | " addi %[dst], %[dst], 32;" | |
294 | " cmplw %[dst], %[edst];" | 299 | " cmplw %[dst], %[edst];" | |
295 | " blt+ 1b;" | 300 | " blt+ 1b;" | |
296 | " ori %[msr], %[msr], 0x10;" /* Set PSL_DR */ | 301 | " ori %[msr], %[msr], 0x10;" /* Set PSL_DR */ | |
297 | " sync;" | 302 | " sync;" | |
298 | " mtmsr %[msr];" /* Turn on DMMU */ | 303 | " mtmsr %[msr];" /* Turn on DMMU */ | |
299 | " isync;" | 304 | " isync;" | |
300 | :: [msr] "r"(msr), [src] "b"(src), [dst] "b"(dst), | 305 | :: [msr] "r"(msr), [src] "b"(src), [dst] "b"(dst), | |
301 | [edst] "b"(edst), [off0] "r"(0), [off16] "r"(16), | 306 | [edst] "b"(edst), [off0] "r"(0), [off16] "r"(16), | |
302 | [lv] "n"(LO_VEC), [hv] "n"(HI_VEC)); | 307 | [lv] "n"(LO_VEC), [hv] "n"(HI_VEC)); | |
303 | 308 | |||
304 | /* | 309 | /* | |
305 | * Restore VEC registers (now that we can access the stack again). | 310 | * Restore VEC registers (now that we can access the stack again). | |
306 | */ | 311 | */ | |
307 | __asm("lvx %2,%1,%0" :: "b"(vp), "r"( 0), "n"(LO_VEC)); | 312 | __asm("lvx %2,%1,%0" :: "b"(vp), "r"( 0), "n"(LO_VEC)); | |
308 | __asm("lvx %2,%1,%0" :: "b"(vp), "r"(16), "n"(HI_VEC)); | 313 | __asm("lvx %2,%1,%0" :: "b"(vp), "r"(16), "n"(HI_VEC)); | |
309 | 314 | |||
310 | /* | 315 | /* | |
311 | * Restore old MSR (AltiVec OFF). | 316 | * Restore old MSR (AltiVec OFF). | |
312 | */ | 317 | */ | |
313 | __asm volatile("sync; mtmsr %0; isync" :: "r"(omsr)); | 318 | __asm volatile("sync; mtmsr %0; isync" :: "r"(omsr)); | |
314 | } | 319 | } |
--- src/sys/arch/powerpc/powerpc/fpu.c 2011/06/07 00:48:32 1.30
+++ src/sys/arch/powerpc/powerpc/fpu.c 2011/06/07 01:01:43 1.31
@@ -1,249 +1,250 @@ | @@ -1,249 +1,250 @@ | |||
1 | /* $NetBSD: fpu.c,v 1.30 2011/06/07 00:48:32 matt Exp $ */ | 1 | /* $NetBSD: fpu.c,v 1.31 2011/06/07 01:01:43 matt Exp $ */ | |
2 | 2 | |||
3 | /* | 3 | /* | |
4 | * Copyright (C) 1996 Wolfgang Solfrank. | 4 | * Copyright (C) 1996 Wolfgang Solfrank. | |
5 | * Copyright (C) 1996 TooLs GmbH. | 5 | * Copyright (C) 1996 TooLs GmbH. | |
6 | * All rights reserved. | 6 | * All rights reserved. | |
7 | * | 7 | * | |
8 | * Redistribution and use in source and binary forms, with or without | 8 | * Redistribution and use in source and binary forms, with or without | |
9 | * modification, are permitted provided that the following conditions | 9 | * modification, are permitted provided that the following conditions | |
10 | * are met: | 10 | * are met: | |
11 | * 1. Redistributions of source code must retain the above copyright | 11 | * 1. Redistributions of source code must retain the above copyright | |
12 | * notice, this list of conditions and the following disclaimer. | 12 | * notice, this list of conditions and the following disclaimer. | |
13 | * 2. Redistributions in binary form must reproduce the above copyright | 13 | * 2. Redistributions in binary form must reproduce the above copyright | |
14 | * notice, this list of conditions and the following disclaimer in the | 14 | * notice, this list of conditions and the following disclaimer in the | |
15 | * documentation and/or other materials provided with the distribution. | 15 | * documentation and/or other materials provided with the distribution. | |
16 | * 3. All advertising materials mentioning features or use of this software | 16 | * 3. All advertising materials mentioning features or use of this software | |
17 | * must display the following acknowledgement: | 17 | * must display the following acknowledgement: | |
18 | * This product includes software developed by TooLs GmbH. | 18 | * This product includes software developed by TooLs GmbH. | |
19 | * 4. The name of TooLs GmbH may not be used to endorse or promote products | 19 | * 4. The name of TooLs GmbH may not be used to endorse or promote products | |
20 | * derived from this software without specific prior written permission. | 20 | * derived from this software without specific prior written permission. | |
21 | * | 21 | * | |
22 | * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR | 22 | * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR | |
23 | * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES | 23 | * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES | |
24 | * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. | 24 | * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. | |
25 | * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, | 25 | * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, | |
26 | * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, | 26 | * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, | |
27 | * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; | 27 | * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; | |
28 | * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, | 28 | * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, | |
29 | * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR | 29 | * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR | |
30 | * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF | 30 | * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF | |
31 | * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. | 31 | * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. | |
32 | */ | 32 | */ | |
33 | 33 | |||
34 | #include <sys/cdefs.h> | 34 | #include <sys/cdefs.h> | |
35 | __KERNEL_RCSID(0, "$NetBSD: fpu.c,v 1.30 2011/06/07 00:48:32 matt Exp $"); | 35 | __KERNEL_RCSID(0, "$NetBSD: fpu.c,v 1.31 2011/06/07 01:01:43 matt Exp $"); | |
36 | 36 | |||
37 | #include "opt_multiprocessor.h" | 37 | #include "opt_multiprocessor.h" | |
38 | 38 | |||
39 | #include <sys/param.h> | 39 | #include <sys/param.h> | |
40 | #include <sys/proc.h> | 40 | #include <sys/proc.h> | |
41 | #include <sys/systm.h> | 41 | #include <sys/systm.h> | |
42 | #include <sys/atomic.h> | 42 | #include <sys/atomic.h> | |
43 | #include <sys/siginfo.h> | 43 | #include <sys/siginfo.h> | |
44 | #include <sys/pcu.h> | 44 | #include <sys/pcu.h> | |
45 | 45 | |||
46 | #include <machine/pcb.h> | 46 | #include <machine/pcb.h> | |
47 | #include <machine/fpu.h> | 47 | #include <machine/fpu.h> | |
48 | #include <machine/psl.h> | 48 | #include <machine/psl.h> | |
49 | 49 | |||
50 | #ifdef PPC_HAVE_FPU | 50 | #ifdef PPC_HAVE_FPU | |
51 | static void fpu_state_load(lwp_t *, bool); | 51 | static void fpu_state_load(lwp_t *, bool); | |
52 | static void fpu_state_save(lwp_t *); | 52 | static void fpu_state_save(lwp_t *); | |
53 | static void fpu_state_release(lwp_t *); | 53 | static void fpu_state_release(lwp_t *); | |
54 | #endif | 54 | #endif | |
55 | 55 | |||
56 | const pcu_ops_t fpu_ops = { | 56 | const pcu_ops_t fpu_ops = { | |
57 | .pcu_id = PCU_FPU, | 57 | .pcu_id = PCU_FPU, | |
58 | #ifdef PPC_HAVE_FPU | 58 | #ifdef PPC_HAVE_FPU | |
59 | .pcu_state_load = fpu_state_load, | 59 | .pcu_state_load = fpu_state_load, | |
60 | .pcu_state_save = fpu_state_save, | 60 | .pcu_state_save = fpu_state_save, | |
61 | .pcu_state_release = fpu_state_release, | 61 | .pcu_state_release = fpu_state_release, | |
62 | #endif | 62 | #endif | |
63 | }; | 63 | }; | |
64 | 64 | |||
65 | bool | 65 | bool | |
66 | fpu_used_p(lwp_t *l) | 66 | fpu_used_p(lwp_t *l) | |
67 | { | 67 | { | |
68 | return (l->l_md.md_flags & MDLWP_USEDFPU) != 0; | 68 | return (l->l_md.md_flags & MDLWP_USEDFPU) != 0; | |
69 | } | 69 | } | |
70 | 70 | |||
71 | void | 71 | void | |
72 | fpu_mark_used(lwp_t *l) | 72 | fpu_mark_used(lwp_t *l) | |
73 | { | 73 | { | |
74 | l->l_md.md_flags |= MDLWP_USEDFPU; | 74 | l->l_md.md_flags |= MDLWP_USEDFPU; | |
75 | } | 75 | } | |
76 | 76 | |||
77 | #ifdef PPC_HAVE_FPU | 77 | #ifdef PPC_HAVE_FPU | |
78 | void | 78 | void | |
79 | fpu_state_load(lwp_t *l, bool used) | 79 | fpu_state_load(lwp_t *l, bool used) | |
80 | { | 80 | { | |
81 | struct pcb * const pcb = lwp_getpcb(l); | 81 | struct pcb * const pcb = lwp_getpcb(l); | |
82 | 82 | |||
83 | if (__predict_false(!used)) { | 83 | if (__predict_false(!fpu_used_p(l))) { | |
84 | memset(&pcb->pcb_fpu, 0, sizeof(pcb->pcb_fpu)); | 84 | memset(&pcb->pcb_fpu, 0, sizeof(pcb->pcb_fpu)); | |
85 | fpu_mark_used(l); | |||
85 | } | 86 | } | |
86 | 87 | |||
87 | const register_t msr = mfmsr(); | 88 | const register_t msr = mfmsr(); | |
88 | mtmsr((msr & ~PSL_EE) | PSL_FP); | 89 | mtmsr((msr & ~PSL_EE) | PSL_FP); | |
89 | __asm volatile ("isync"); | 90 | __asm volatile ("isync"); | |
90 | 91 | |||
91 | fpu_load_from_fpreg(&pcb->pcb_fpu); | 92 | fpu_load_from_fpreg(&pcb->pcb_fpu); | |
92 | __asm volatile ("sync"); | 93 | __asm volatile ("sync"); | |
93 | 94 | |||
94 | mtmsr(msr); | 95 | mtmsr(msr); | |
95 | __asm volatile ("isync"); | 96 | __asm volatile ("isync"); | |
96 | 97 | |||
97 | curcpu()->ci_ev_fpusw.ev_count++; | 98 | curcpu()->ci_ev_fpusw.ev_count++; | |
98 | l->l_md.md_utf->tf_srr1 |= PSL_FP|(pcb->pcb_flags & (PCB_FE0|PCB_FE1)); | 99 | l->l_md.md_utf->tf_srr1 |= PSL_FP|(pcb->pcb_flags & (PCB_FE0|PCB_FE1)); | |
99 | l->l_md.md_flags |= MDLWP_USEDFPU; | 100 | l->l_md.md_flags |= MDLWP_USEDFPU; | |
100 | } | 101 | } | |
101 | 102 | |||
102 | /* | 103 | /* | |
103 | * Save the contents of the current CPU's FPU to its PCB. | 104 | * Save the contents of the current CPU's FPU to its PCB. | |
104 | */ | 105 | */ | |
105 | void | 106 | void | |
106 | fpu_state_save(lwp_t *l) | 107 | fpu_state_save(lwp_t *l) | |
107 | { | 108 | { | |
108 | struct pcb * const pcb = lwp_getpcb(l); | 109 | struct pcb * const pcb = lwp_getpcb(l); | |
109 | 110 | |||
110 | const register_t msr = mfmsr(); | 111 | const register_t msr = mfmsr(); | |
111 | mtmsr((msr & ~PSL_EE) | PSL_FP); | 112 | mtmsr((msr & ~PSL_EE) | PSL_FP); | |
112 | __asm volatile ("isync"); | 113 | __asm volatile ("isync"); | |
113 | 114 | |||
114 | fpu_unload_to_fpreg(&pcb->pcb_fpu); | 115 | fpu_unload_to_fpreg(&pcb->pcb_fpu); | |
115 | __asm volatile ("sync"); | 116 | __asm volatile ("sync"); | |
116 | 117 | |||
117 | mtmsr(msr); | 118 | mtmsr(msr); | |
118 | __asm volatile ("isync"); | 119 | __asm volatile ("isync"); | |
119 | } | 120 | } | |
120 | 121 | |||
121 | void | 122 | void | |
122 | fpu_state_release(lwp_t *l) | 123 | fpu_state_release(lwp_t *l) | |
123 | { | 124 | { | |
124 | l->l_md.md_utf->tf_srr1 &= ~PSL_FP; | 125 | l->l_md.md_utf->tf_srr1 &= ~PSL_FP; | |
125 | } | 126 | } | |
126 | 127 | |||
127 | #define STICKYBITS (FPSCR_VX|FPSCR_OX|FPSCR_UX|FPSCR_ZX|FPSCR_XX) | 128 | #define STICKYBITS (FPSCR_VX|FPSCR_OX|FPSCR_UX|FPSCR_ZX|FPSCR_XX) | |
128 | #define STICKYSHIFT 25 | 129 | #define STICKYSHIFT 25 | |
129 | #define MASKBITS (FPSCR_VE|FPSCR_OE|FPSCR_UE|FPSCR_ZE|FPSCR_XE) | 130 | #define MASKBITS (FPSCR_VE|FPSCR_OE|FPSCR_UE|FPSCR_ZE|FPSCR_XE) | |
130 | #define MASKSHIFT 3 | 131 | #define MASKSHIFT 3 | |
131 | 132 | |||
132 | int | 133 | int | |
133 | fpu_get_fault_code(void) | 134 | fpu_get_fault_code(void) | |
134 | { | 135 | { | |
135 | lwp_t * const l = curlwp; | 136 | lwp_t * const l = curlwp; | |
136 | struct pcb * const pcb = lwp_getpcb(l); | 137 | struct pcb * const pcb = lwp_getpcb(l); | |
137 | uint64_t fpscr64; | 138 | uint64_t fpscr64; | |
138 | uint32_t fpscr, ofpscr; | 139 | uint32_t fpscr, ofpscr; | |
139 | int code; | 140 | int code; | |
140 | 141 | |||
141 | int s = splsoftclock(); /* disable preemption */ | 142 | int s = splsoftclock(); /* disable preemption */ | |
142 | 143 | |||
143 | struct cpu_info * const ci = curcpu(); | 144 | struct cpu_info * const ci = curcpu(); | |
144 | /* | 145 | /* | |
145 | * If we got preempted, we may be running on a different CPU. So we | 146 | * If we got preempted, we may be running on a different CPU. So we | |
146 | * need to check for that. | 147 | * need to check for that. | |
147 | */ | 148 | */ | |
148 | KASSERT(fpu_used_p(l)); | 149 | KASSERT(fpu_used_p(l)); | |
149 | if (__predict_true(l->l_pcu_cpu[PCU_FPU] == ci)) { | 150 | if (__predict_true(l->l_pcu_cpu[PCU_FPU] == ci)) { | |
150 | uint64_t tmp; | 151 | uint64_t tmp; | |
151 | const register_t msr = mfmsr(); | 152 | const register_t msr = mfmsr(); | |
152 | mtmsr((msr & ~PSL_EE) | PSL_FP); | 153 | mtmsr((msr & ~PSL_EE) | PSL_FP); | |
153 | __asm volatile ("isync"); | 154 | __asm volatile ("isync"); | |
154 | __asm volatile ( | 155 | __asm volatile ( | |
155 | "stfd 0,0(%[tmp])\n" /* save f0 */ | 156 | "stfd 0,0(%[tmp])\n" /* save f0 */ | |
156 | "mffs 0\n" /* get FPSCR */ | 157 | "mffs 0\n" /* get FPSCR */ | |
157 | "stfd 0,0(%[fpscr64])\n" /* store a temp copy */ | 158 | "stfd 0,0(%[fpscr64])\n" /* store a temp copy */ | |
158 | "mtfsb0 0\n" /* clear FPSCR_FX */ | 159 | "mtfsb0 0\n" /* clear FPSCR_FX */ | |
159 | "mtfsb0 24\n" /* clear FPSCR_VE */ | 160 | "mtfsb0 24\n" /* clear FPSCR_VE */ | |
160 | "mtfsb0 25\n" /* clear FPSCR_OE */ | 161 | "mtfsb0 25\n" /* clear FPSCR_OE */ | |
161 | "mtfsb0 26\n" /* clear FPSCR_UE */ | 162 | "mtfsb0 26\n" /* clear FPSCR_UE */ | |
162 | "mtfsb0 27\n" /* clear FPSCR_ZE */ | 163 | "mtfsb0 27\n" /* clear FPSCR_ZE */ | |
163 | "mtfsb0 28\n" /* clear FPSCR_XE */ | 164 | "mtfsb0 28\n" /* clear FPSCR_XE */ | |
164 | "mffs 0\n" /* get FPSCR */ | 165 | "mffs 0\n" /* get FPSCR */ | |
165 | "stfd 0,0(%[fpscr])\n" /* store it */ | 166 | "stfd 0,0(%[fpscr])\n" /* store it */ | |
166 | "lfd 0,0(%[tmp])\n" /* restore f0 */ | 167 | "lfd 0,0(%[tmp])\n" /* restore f0 */ | |
167 | :: [tmp] "b"(&tmp), | 168 | :: [tmp] "b"(&tmp), | |
168 | [fpscr] "b"(&pcb->pcb_fpu.fpscr), | 169 | [fpscr] "b"(&pcb->pcb_fpu.fpscr), | |
169 | [fpscr64] "b"(&fpscr64)); | 170 | [fpscr64] "b"(&fpscr64)); | |
170 | mtmsr(msr); | 171 | mtmsr(msr); | |
171 | __asm volatile ("isync"); | 172 | __asm volatile ("isync"); | |
172 | } else { | 173 | } else { | |
173 | /* | 174 | /* | |
174 | * We got preempted to a different CPU so we need to save | 175 | * We got preempted to a different CPU so we need to save | |
175 | * our FPU state. | 176 | * our FPU state. | |
176 | */ | 177 | */ | |
177 | fpu_save(); | 178 | fpu_save(); | |
178 | fpscr64 = *(uint64_t *)&pcb->pcb_fpu.fpscr; | 179 | fpscr64 = *(uint64_t *)&pcb->pcb_fpu.fpscr; | |
179 | ((uint32_t *)&pcb->pcb_fpu.fpscr)[_QUAD_LOWWORD] &= ~MASKBITS; | 180 | ((uint32_t *)&pcb->pcb_fpu.fpscr)[_QUAD_LOWWORD] &= ~MASKBITS; | |
180 | } | 181 | } | |
181 | 182 | |||
182 | splx(s); /* allow preemption */ | 183 | splx(s); /* allow preemption */ | |
183 | 184 | |||
184 | /* | 185 | /* | |
185 | * Now determine the fault type. First we test to see if any of sticky | 186 | * Now determine the fault type. First we test to see if any of sticky | |
186 | * bits correspond to the enabled exceptions. If so, we only test | 187 | * bits correspond to the enabled exceptions. If so, we only test | |
187 | * those bits. If not, we look at all the bits. (In reality, we only | 188 | * those bits. If not, we look at all the bits. (In reality, we only | |
188 | * could get an exception if FPSCR_FEX changed state. So we should | 189 | * could get an exception if FPSCR_FEX changed state. So we should | |
189 | * have at least one bit that corresponds). | 190 | * have at least one bit that corresponds). | |
190 | */ | 191 | */ | |
191 | ofpscr = (uint32_t)fpscr64; | 192 | ofpscr = (uint32_t)fpscr64; | |
192 | ofpscr &= ofpscr << (STICKYSHIFT - MASKSHIFT); | 193 | ofpscr &= ofpscr << (STICKYSHIFT - MASKSHIFT); | |
193 | fpscr = ((uint32_t *)&pcb->pcb_fpu.fpscr)[_QUAD_LOWWORD]; | 194 | fpscr = ((uint32_t *)&pcb->pcb_fpu.fpscr)[_QUAD_LOWWORD]; | |
194 | if (fpscr & ofpscr & STICKYBITS) | 195 | if (fpscr & ofpscr & STICKYBITS) | |
195 | fpscr &= ofpscr; | 196 | fpscr &= ofpscr; | |
196 | 197 | |||
197 | /* | 198 | /* | |
198 | * Let's determine what the appropriate code is. | 199 | * Let's determine what the appropriate code is. | |
199 | */ | 200 | */ | |
200 | if (fpscr & FPSCR_VX) code = FPE_FLTINV; | 201 | if (fpscr & FPSCR_VX) code = FPE_FLTINV; | |
201 | else if (fpscr & FPSCR_OX) code = FPE_FLTOVF; | 202 | else if (fpscr & FPSCR_OX) code = FPE_FLTOVF; | |
202 | else if (fpscr & FPSCR_UX) code = FPE_FLTUND; | 203 | else if (fpscr & FPSCR_UX) code = FPE_FLTUND; | |
203 | else if (fpscr & FPSCR_ZX) code = FPE_FLTDIV; | 204 | else if (fpscr & FPSCR_ZX) code = FPE_FLTDIV; | |
204 | else if (fpscr & FPSCR_XX) code = FPE_FLTRES; | 205 | else if (fpscr & FPSCR_XX) code = FPE_FLTRES; | |
205 | else code = 0; | 206 | else code = 0; | |
206 | return code; | 207 | return code; | |
207 | } | 208 | } | |
208 | #endif /* PPC_HAVE_FPU */ | 209 | #endif /* PPC_HAVE_FPU */ | |
209 | 210 | |||
210 | bool | 211 | bool | |
211 | fpu_save_to_mcontext(lwp_t *l, mcontext_t *mcp, unsigned int *flagp) | 212 | fpu_save_to_mcontext(lwp_t *l, mcontext_t *mcp, unsigned int *flagp) | |
212 | { | 213 | { | |
213 | KASSERT(l == curlwp); | 214 | KASSERT(l == curlwp); | |
214 | 215 | |||
215 | if (!pcu_used_p(&fpu_ops)) | 216 | if (!pcu_used_p(&fpu_ops)) | |
216 | return false; | 217 | return false; | |
217 | 218 | |||
218 | struct pcb * const pcb = lwp_getpcb(l); | 219 | struct pcb * const pcb = lwp_getpcb(l); | |
219 | 220 | |||
220 | #ifdef PPC_HAVE_FPU | 221 | #ifdef PPC_HAVE_FPU | |
221 | /* If we're the FPU owner, dump its context to the PCB first. */ | 222 | /* If we're the FPU owner, dump its context to the PCB first. */ | |
222 | pcu_save(&fpu_ops); | 223 | pcu_save(&fpu_ops); | |
223 | #endif | 224 | #endif | |
224 | (void)memcpy(mcp->__fpregs.__fpu_regs, pcb->pcb_fpu.fpreg, | 225 | (void)memcpy(mcp->__fpregs.__fpu_regs, pcb->pcb_fpu.fpreg, | |
225 | sizeof (mcp->__fpregs.__fpu_regs)); | 226 | sizeof (mcp->__fpregs.__fpu_regs)); | |
226 | mcp->__fpregs.__fpu_fpscr = | 227 | mcp->__fpregs.__fpu_fpscr = | |
227 | ((int *)&pcb->pcb_fpu.fpscr)[_QUAD_LOWWORD]; | 228 | ((int *)&pcb->pcb_fpu.fpscr)[_QUAD_LOWWORD]; | |
228 | mcp->__fpregs.__fpu_valid = 1; | 229 | mcp->__fpregs.__fpu_valid = 1; | |
229 | *flagp |= _UC_FPU; | 230 | *flagp |= _UC_FPU; | |
230 | return true; | 231 | return true; | |
231 | } | 232 | } | |
232 | 233 | |||
233 | void | 234 | void | |
234 | fpu_restore_from_mcontext(lwp_t *l, const mcontext_t *mcp) | 235 | fpu_restore_from_mcontext(lwp_t *l, const mcontext_t *mcp) | |
235 | { | 236 | { | |
236 | if (!mcp->__fpregs.__fpu_valid) | 237 | if (!mcp->__fpregs.__fpu_valid) | |
237 | return; | 238 | return; | |
238 | 239 | |||
239 | struct pcb * const pcb = lwp_getpcb(l); | 240 | struct pcb * const pcb = lwp_getpcb(l); | |
240 | 241 | |||
241 | #ifdef PPC_HAVE_FPU | 242 | #ifdef PPC_HAVE_FPU | |
242 | /* we don't need to save the state, just drop it */ | 243 | /* we don't need to save the state, just drop it */ | |
243 | if (l == curlwp) | 244 | if (l == curlwp) | |
244 | pcu_discard(&fpu_ops); | 245 | pcu_discard(&fpu_ops); | |
245 | #endif | 246 | #endif | |
246 | (void)memcpy(&pcb->pcb_fpu.fpreg, &mcp->__fpregs.__fpu_regs, | 247 | (void)memcpy(&pcb->pcb_fpu.fpreg, &mcp->__fpregs.__fpu_regs, | |
247 | sizeof (pcb->pcb_fpu.fpreg)); | 248 | sizeof (pcb->pcb_fpu.fpreg)); | |
248 | ((int *)&pcb->pcb_fpu.fpscr)[_QUAD_LOWWORD] = mcp->__fpregs.__fpu_fpscr; | 249 | ((int *)&pcb->pcb_fpu.fpscr)[_QUAD_LOWWORD] = mcp->__fpregs.__fpu_fpscr; | |
249 | } | 250 | } |