1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * In-kernel vector facility support functions
4 *
5 * Copyright IBM Corp. 2015
6 * Author(s): Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
7 */
8 #include <linux/kernel.h>
9 #include <linux/cpu.h>
10 #include <linux/sched.h>
11 #include <asm/fpu.h>
12
__kernel_fpu_begin(struct kernel_fpu * state,int flags)13 void __kernel_fpu_begin(struct kernel_fpu *state, int flags)
14 {
15 __vector128 *vxrs = state->vxrs;
16 int mask;
17
18 /*
19 * Limit the save to the FPU/vector registers already
20 * in use by the previous context.
21 */
22 flags &= state->hdr.mask;
23 if (flags & KERNEL_FPC)
24 fpu_stfpc(&state->hdr.fpc);
25 if (!cpu_has_vx()) {
26 if (flags & KERNEL_VXR_LOW)
27 save_fp_regs_vx(vxrs);
28 return;
29 }
30 mask = flags & KERNEL_VXR;
31 if (mask == KERNEL_VXR) {
32 vxrs += fpu_vstm(0, 15, vxrs);
33 vxrs += fpu_vstm(16, 31, vxrs);
34 return;
35 }
36 if (mask == KERNEL_VXR_MID) {
37 vxrs += fpu_vstm(8, 23, vxrs);
38 return;
39 }
40 mask = flags & KERNEL_VXR_LOW;
41 if (mask) {
42 if (mask == KERNEL_VXR_LOW)
43 vxrs += fpu_vstm(0, 15, vxrs);
44 else if (mask == KERNEL_VXR_V0V7)
45 vxrs += fpu_vstm(0, 7, vxrs);
46 else
47 vxrs += fpu_vstm(8, 15, vxrs);
48 }
49 mask = flags & KERNEL_VXR_HIGH;
50 if (mask) {
51 if (mask == KERNEL_VXR_HIGH)
52 vxrs += fpu_vstm(16, 31, vxrs);
53 else if (mask == KERNEL_VXR_V16V23)
54 vxrs += fpu_vstm(16, 23, vxrs);
55 else
56 vxrs += fpu_vstm(24, 31, vxrs);
57 }
58 }
59 EXPORT_SYMBOL(__kernel_fpu_begin);
60
__kernel_fpu_end(struct kernel_fpu * state,int flags)61 void __kernel_fpu_end(struct kernel_fpu *state, int flags)
62 {
63 __vector128 *vxrs = state->vxrs;
64 int mask;
65
66 /*
67 * Limit the restore to the FPU/vector registers of the
68 * previous context that have been overwritten by the
69 * current context.
70 */
71 flags &= state->hdr.mask;
72 if (flags & KERNEL_FPC)
73 fpu_lfpc(&state->hdr.fpc);
74 if (!cpu_has_vx()) {
75 if (flags & KERNEL_VXR_LOW)
76 load_fp_regs_vx(vxrs);
77 return;
78 }
79 mask = flags & KERNEL_VXR;
80 if (mask == KERNEL_VXR) {
81 vxrs += fpu_vlm(0, 15, vxrs);
82 vxrs += fpu_vlm(16, 31, vxrs);
83 return;
84 }
85 if (mask == KERNEL_VXR_MID) {
86 vxrs += fpu_vlm(8, 23, vxrs);
87 return;
88 }
89 mask = flags & KERNEL_VXR_LOW;
90 if (mask) {
91 if (mask == KERNEL_VXR_LOW)
92 vxrs += fpu_vlm(0, 15, vxrs);
93 else if (mask == KERNEL_VXR_V0V7)
94 vxrs += fpu_vlm(0, 7, vxrs);
95 else
96 vxrs += fpu_vlm(8, 15, vxrs);
97 }
98 mask = flags & KERNEL_VXR_HIGH;
99 if (mask) {
100 if (mask == KERNEL_VXR_HIGH)
101 vxrs += fpu_vlm(16, 31, vxrs);
102 else if (mask == KERNEL_VXR_V16V23)
103 vxrs += fpu_vlm(16, 23, vxrs);
104 else
105 vxrs += fpu_vlm(24, 31, vxrs);
106 }
107 }
108 EXPORT_SYMBOL(__kernel_fpu_end);
109
load_fpu_state(struct fpu * state,int flags)110 void load_fpu_state(struct fpu *state, int flags)
111 {
112 __vector128 *vxrs = &state->vxrs[0];
113 int mask;
114
115 if (flags & KERNEL_FPC)
116 fpu_lfpc_safe(&state->fpc);
117 if (!cpu_has_vx()) {
118 if (flags & KERNEL_VXR_V0V7)
119 load_fp_regs_vx(state->vxrs);
120 return;
121 }
122 mask = flags & KERNEL_VXR;
123 if (mask == KERNEL_VXR) {
124 fpu_vlm(0, 15, &vxrs[0]);
125 fpu_vlm(16, 31, &vxrs[16]);
126 return;
127 }
128 if (mask == KERNEL_VXR_MID) {
129 fpu_vlm(8, 23, &vxrs[8]);
130 return;
131 }
132 mask = flags & KERNEL_VXR_LOW;
133 if (mask) {
134 if (mask == KERNEL_VXR_LOW)
135 fpu_vlm(0, 15, &vxrs[0]);
136 else if (mask == KERNEL_VXR_V0V7)
137 fpu_vlm(0, 7, &vxrs[0]);
138 else
139 fpu_vlm(8, 15, &vxrs[8]);
140 }
141 mask = flags & KERNEL_VXR_HIGH;
142 if (mask) {
143 if (mask == KERNEL_VXR_HIGH)
144 fpu_vlm(16, 31, &vxrs[16]);
145 else if (mask == KERNEL_VXR_V16V23)
146 fpu_vlm(16, 23, &vxrs[16]);
147 else
148 fpu_vlm(24, 31, &vxrs[24]);
149 }
150 }
151
save_fpu_state(struct fpu * state,int flags)152 void save_fpu_state(struct fpu *state, int flags)
153 {
154 __vector128 *vxrs = &state->vxrs[0];
155 int mask;
156
157 if (flags & KERNEL_FPC)
158 fpu_stfpc(&state->fpc);
159 if (!cpu_has_vx()) {
160 if (flags & KERNEL_VXR_LOW)
161 save_fp_regs_vx(state->vxrs);
162 return;
163 }
164 mask = flags & KERNEL_VXR;
165 if (mask == KERNEL_VXR) {
166 fpu_vstm(0, 15, &vxrs[0]);
167 fpu_vstm(16, 31, &vxrs[16]);
168 return;
169 }
170 if (mask == KERNEL_VXR_MID) {
171 fpu_vstm(8, 23, &vxrs[8]);
172 return;
173 }
174 mask = flags & KERNEL_VXR_LOW;
175 if (mask) {
176 if (mask == KERNEL_VXR_LOW)
177 fpu_vstm(0, 15, &vxrs[0]);
178 else if (mask == KERNEL_VXR_V0V7)
179 fpu_vstm(0, 7, &vxrs[0]);
180 else
181 fpu_vstm(8, 15, &vxrs[8]);
182 }
183 mask = flags & KERNEL_VXR_HIGH;
184 if (mask) {
185 if (mask == KERNEL_VXR_HIGH)
186 fpu_vstm(16, 31, &vxrs[16]);
187 else if (mask == KERNEL_VXR_V16V23)
188 fpu_vstm(16, 23, &vxrs[16]);
189 else
190 fpu_vstm(24, 31, &vxrs[24]);
191 }
192 }
193 EXPORT_SYMBOL(save_fpu_state);
194