1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
|
// SPDX-License-Identifier: GPL-2.0
/*
* In-kernel vector facility support functions
*
* Copyright IBM Corp. 2015
* Author(s): Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
*/
#include <linux/kernel.h>
#include <linux/cpu.h>
#include <linux/sched.h>
#include <asm/fpu.h>
void __kernel_fpu_begin(struct kernel_fpu *state, u32 flags)
{
/*
* Limit the save to the FPU/vector registers already
* in use by the previous context.
*/
flags &= state->mask;
if (flags & KERNEL_FPC)
fpu_stfpc(&state->fpc);
if (!cpu_has_vx()) {
if (flags & KERNEL_VXR_LOW)
save_fp_regs(state->fprs);
return;
}
/* Test and save vector registers */
asm volatile (
/*
* Test if any vector register must be saved and, if so,
* test if all register can be saved.
*/
" la 1,%[vxrs]\n" /* load save area */
" tmll %[m],30\n" /* KERNEL_VXR */
" jz 7f\n" /* no work -> done */
" jo 5f\n" /* -> save V0..V31 */
/*
* Test for special case KERNEL_FPU_MID only. In this
* case a vstm V8..V23 is the best instruction
*/
" chi %[m],12\n" /* KERNEL_VXR_MID */
" jne 0f\n" /* -> save V8..V23 */
" VSTM 8,23,128,1\n" /* vstm %v8,%v23,128(%r1) */
" j 7f\n"
/* Test and save the first half of 16 vector registers */
"0: tmll %[m],6\n" /* KERNEL_VXR_LOW */
" jz 3f\n" /* -> KERNEL_VXR_HIGH */
" jo 2f\n" /* 11 -> save V0..V15 */
" brc 2,1f\n" /* 10 -> save V8..V15 */
" VSTM 0,7,0,1\n" /* vstm %v0,%v7,0(%r1) */
" j 3f\n"
"1: VSTM 8,15,128,1\n" /* vstm %v8,%v15,128(%r1) */
" j 3f\n"
"2: VSTM 0,15,0,1\n" /* vstm %v0,%v15,0(%r1) */
/* Test and save the second half of 16 vector registers */
"3: tmll %[m],24\n" /* KERNEL_VXR_HIGH */
" jz 7f\n"
" jo 6f\n" /* 11 -> save V16..V31 */
" brc 2,4f\n" /* 10 -> save V24..V31 */
" VSTM 16,23,256,1\n" /* vstm %v16,%v23,256(%r1) */
" j 7f\n"
"4: VSTM 24,31,384,1\n" /* vstm %v24,%v31,384(%r1) */
" j 7f\n"
"5: VSTM 0,15,0,1\n" /* vstm %v0,%v15,0(%r1) */
"6: VSTM 16,31,256,1\n" /* vstm %v16,%v31,256(%r1) */
"7:"
: [vxrs] "=Q" (*(struct vx_array *) &state->vxrs)
: [m] "d" (flags)
: "1", "cc");
}
EXPORT_SYMBOL(__kernel_fpu_begin);
void __kernel_fpu_end(struct kernel_fpu *state, u32 flags)
{
/*
* Limit the restore to the FPU/vector registers of the
* previous context that have been overwritten by the
* current context.
*/
flags &= state->mask;
if (flags & KERNEL_FPC)
fpu_lfpc(&state->fpc);
if (!cpu_has_vx()) {
if (flags & KERNEL_VXR_LOW)
load_fp_regs(state->fprs);
return;
}
/* Test and restore (load) vector registers */
asm volatile (
/*
* Test if any vector register must be loaded and, if so,
* test if all registers can be loaded at once.
*/
" la 1,%[vxrs]\n" /* load restore area */
" tmll %[m],30\n" /* KERNEL_VXR */
" jz 7f\n" /* no work -> done */
" jo 5f\n" /* -> restore V0..V31 */
/*
* Test for special case KERNEL_FPU_MID only. In this
* case a vlm V8..V23 is the best instruction
*/
" chi %[m],12\n" /* KERNEL_VXR_MID */
" jne 0f\n" /* -> restore V8..V23 */
" VLM 8,23,128,1\n" /* vlm %v8,%v23,128(%r1) */
" j 7f\n"
/* Test and restore the first half of 16 vector registers */
"0: tmll %[m],6\n" /* KERNEL_VXR_LOW */
" jz 3f\n" /* -> KERNEL_VXR_HIGH */
" jo 2f\n" /* 11 -> restore V0..V15 */
" brc 2,1f\n" /* 10 -> restore V8..V15 */
" VLM 0,7,0,1\n" /* vlm %v0,%v7,0(%r1) */
" j 3f\n"
"1: VLM 8,15,128,1\n" /* vlm %v8,%v15,128(%r1) */
" j 3f\n"
"2: VLM 0,15,0,1\n" /* vlm %v0,%v15,0(%r1) */
/* Test and restore the second half of 16 vector registers */
"3: tmll %[m],24\n" /* KERNEL_VXR_HIGH */
" jz 7f\n"
" jo 6f\n" /* 11 -> restore V16..V31 */
" brc 2,4f\n" /* 10 -> restore V24..V31 */
" VLM 16,23,256,1\n" /* vlm %v16,%v23,256(%r1) */
" j 7f\n"
"4: VLM 24,31,384,1\n" /* vlm %v24,%v31,384(%r1) */
" j 7f\n"
"5: VLM 0,15,0,1\n" /* vlm %v0,%v15,0(%r1) */
"6: VLM 16,31,256,1\n" /* vlm %v16,%v31,256(%r1) */
"7:"
: [vxrs] "=Q" (*(struct vx_array *) &state->vxrs)
: [m] "d" (flags)
: "1", "cc");
}
EXPORT_SYMBOL(__kernel_fpu_end);
void __load_fpu_regs(void)
{
struct fpu *state = ¤t->thread.fpu;
void *regs = current->thread.fpu.regs;
fpu_lfpc_safe(&state->fpc);
if (likely(cpu_has_vx())) {
asm volatile("lgr 1,%0\n"
"VLM 0,15,0,1\n"
"VLM 16,31,256,1\n"
:
: "d" (regs)
: "1", "cc", "memory");
} else {
load_fp_regs(regs);
}
clear_cpu_flag(CIF_FPU);
}
void load_fpu_regs(void)
{
raw_local_irq_disable();
__load_fpu_regs();
raw_local_irq_enable();
}
EXPORT_SYMBOL(load_fpu_regs);
void save_fpu_regs(void)
{
unsigned long flags;
struct fpu *state;
void *regs;
local_irq_save(flags);
if (test_cpu_flag(CIF_FPU))
goto out;
state = ¤t->thread.fpu;
regs = current->thread.fpu.regs;
fpu_stfpc(&state->fpc);
if (likely(cpu_has_vx())) {
asm volatile("lgr 1,%0\n"
"VSTM 0,15,0,1\n"
"VSTM 16,31,256,1\n"
:
: "d" (regs)
: "1", "cc", "memory");
} else {
save_fp_regs(regs);
}
set_cpu_flag(CIF_FPU);
out:
local_irq_restore(flags);
}
EXPORT_SYMBOL(save_fpu_regs);
|