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
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
|
/* SPDX-License-Identifier: GPL-2.0 */
/*
* In-kernel FPU support functions
*
*
* Consider these guidelines before using in-kernel FPU functions:
*
* 1. Use kernel_fpu_begin() and kernel_fpu_end() to enclose all in-kernel
* use of floating-point or vector registers and instructions.
*
* 2. For kernel_fpu_begin(), specify the vector register range you want to
* use with the KERNEL_VXR_* constants. Consider these usage guidelines:
*
* a) If your function typically runs in process-context, use the lower
* half of the vector registers, for example, specify KERNEL_VXR_LOW.
* b) If your function typically runs in soft-irq or hard-irq context,
* prefer using the upper half of the vector registers, for example,
* specify KERNEL_VXR_HIGH.
*
* If you adhere to these guidelines, an interrupted process context
* does not require to save and restore vector registers because of
* disjoint register ranges.
*
* Also note that the __kernel_fpu_begin()/__kernel_fpu_end() functions
* includes logic to save and restore up to 16 vector registers at once.
*
* 3. You can nest kernel_fpu_begin()/kernel_fpu_end() by using different
* struct kernel_fpu states. Vector registers that are in use by outer
* levels are saved and restored. You can minimize the save and restore
* effort by choosing disjoint vector register ranges.
*
* 5. To use vector floating-point instructions, specify the KERNEL_FPC
* flag to save and restore floating-point controls in addition to any
* vector register range.
*
* 6. To use floating-point registers and instructions only, specify the
* KERNEL_FPR flag. This flag triggers a save and restore of vector
* registers V0 to V15 and floating-point controls.
*
* Copyright IBM Corp. 2015
* Author(s): Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
*/
#ifndef _ASM_S390_FPU_H
#define _ASM_S390_FPU_H
#include <linux/processor.h>
#include <linux/preempt.h>
#include <linux/string.h>
#include <linux/sched.h>
#include <asm/sigcontext.h>
#include <asm/fpu-types.h>
#include <asm/fpu-insn.h>
#include <asm/facility.h>
static inline bool cpu_has_vx(void)
{
return likely(test_facility(129));
}
void save_user_fpu_regs(void);
void load_user_fpu_regs(void);
void __load_user_fpu_regs(void);
enum {
KERNEL_FPC_BIT = 0,
KERNEL_VXR_V0V7_BIT,
KERNEL_VXR_V8V15_BIT,
KERNEL_VXR_V16V23_BIT,
KERNEL_VXR_V24V31_BIT,
};
#define KERNEL_FPC BIT(KERNEL_FPC_BIT)
#define KERNEL_VXR_V0V7 BIT(KERNEL_VXR_V0V7_BIT)
#define KERNEL_VXR_V8V15 BIT(KERNEL_VXR_V8V15_BIT)
#define KERNEL_VXR_V16V23 BIT(KERNEL_VXR_V16V23_BIT)
#define KERNEL_VXR_V24V31 BIT(KERNEL_VXR_V24V31_BIT)
#define KERNEL_VXR_LOW (KERNEL_VXR_V0V7 | KERNEL_VXR_V8V15)
#define KERNEL_VXR_MID (KERNEL_VXR_V8V15 | KERNEL_VXR_V16V23)
#define KERNEL_VXR_HIGH (KERNEL_VXR_V16V23 | KERNEL_VXR_V24V31)
#define KERNEL_VXR (KERNEL_VXR_LOW | KERNEL_VXR_HIGH)
#define KERNEL_FPR (KERNEL_FPC | KERNEL_VXR_LOW)
void __kernel_fpu_begin(struct kernel_fpu *state, int flags);
void __kernel_fpu_end(struct kernel_fpu *state, int flags);
static __always_inline void save_vx_regs(__vector128 *vxrs)
{
fpu_vstm(0, 15, &vxrs[0]);
fpu_vstm(16, 31, &vxrs[16]);
}
static __always_inline void load_vx_regs(__vector128 *vxrs)
{
fpu_vlm(0, 15, &vxrs[0]);
fpu_vlm(16, 31, &vxrs[16]);
}
static __always_inline void __save_fp_regs(freg_t *fprs, unsigned int offset)
{
fpu_std(0, &fprs[0 * offset]);
fpu_std(1, &fprs[1 * offset]);
fpu_std(2, &fprs[2 * offset]);
fpu_std(3, &fprs[3 * offset]);
fpu_std(4, &fprs[4 * offset]);
fpu_std(5, &fprs[5 * offset]);
fpu_std(6, &fprs[6 * offset]);
fpu_std(7, &fprs[7 * offset]);
fpu_std(8, &fprs[8 * offset]);
fpu_std(9, &fprs[9 * offset]);
fpu_std(10, &fprs[10 * offset]);
fpu_std(11, &fprs[11 * offset]);
fpu_std(12, &fprs[12 * offset]);
fpu_std(13, &fprs[13 * offset]);
fpu_std(14, &fprs[14 * offset]);
fpu_std(15, &fprs[15 * offset]);
}
static __always_inline void __load_fp_regs(freg_t *fprs, unsigned int offset)
{
fpu_ld(0, &fprs[0 * offset]);
fpu_ld(1, &fprs[1 * offset]);
fpu_ld(2, &fprs[2 * offset]);
fpu_ld(3, &fprs[3 * offset]);
fpu_ld(4, &fprs[4 * offset]);
fpu_ld(5, &fprs[5 * offset]);
fpu_ld(6, &fprs[6 * offset]);
fpu_ld(7, &fprs[7 * offset]);
fpu_ld(8, &fprs[8 * offset]);
fpu_ld(9, &fprs[9 * offset]);
fpu_ld(10, &fprs[10 * offset]);
fpu_ld(11, &fprs[11 * offset]);
fpu_ld(12, &fprs[12 * offset]);
fpu_ld(13, &fprs[13 * offset]);
fpu_ld(14, &fprs[14 * offset]);
fpu_ld(15, &fprs[15 * offset]);
}
static __always_inline void save_fp_regs(freg_t *fprs)
{
__save_fp_regs(fprs, sizeof(freg_t) / sizeof(freg_t));
}
static __always_inline void load_fp_regs(freg_t *fprs)
{
__load_fp_regs(fprs, sizeof(freg_t) / sizeof(freg_t));
}
static __always_inline void save_fp_regs_vx(__vector128 *vxrs)
{
freg_t *fprs = (freg_t *)&vxrs[0].high;
__save_fp_regs(fprs, sizeof(__vector128) / sizeof(freg_t));
}
static __always_inline void load_fp_regs_vx(__vector128 *vxrs)
{
freg_t *fprs = (freg_t *)&vxrs[0].high;
__load_fp_regs(fprs, sizeof(__vector128) / sizeof(freg_t));
}
static inline void kernel_fpu_begin(struct kernel_fpu *state, int flags)
{
state->mask = READ_ONCE(current->thread.kfpu_flags);
if (!test_thread_flag(TIF_FPU)) {
/* Save user space FPU state and register contents */
save_user_fpu_regs();
} else if (state->mask & flags) {
/* Save FPU/vector register in-use by the kernel */
__kernel_fpu_begin(state, flags);
}
__atomic_or(flags, ¤t->thread.kfpu_flags);
}
static inline void kernel_fpu_end(struct kernel_fpu *state, int flags)
{
WRITE_ONCE(current->thread.kfpu_flags, state->mask);
if (state->mask & flags) {
/* Restore FPU/vector register in-use by the kernel */
__kernel_fpu_end(state, flags);
}
}
static inline void save_kernel_fpu_regs(struct thread_struct *thread)
{
struct fpu *state = &thread->kfpu;
if (!thread->kfpu_flags)
return;
fpu_stfpc(&state->fpc);
if (likely(cpu_has_vx()))
save_vx_regs(state->vxrs);
else
save_fp_regs_vx(state->vxrs);
}
static inline void restore_kernel_fpu_regs(struct thread_struct *thread)
{
struct fpu *state = &thread->kfpu;
if (!thread->kfpu_flags)
return;
fpu_lfpc(&state->fpc);
if (likely(cpu_has_vx()))
load_vx_regs(state->vxrs);
else
load_fp_regs_vx(state->vxrs);
}
static inline void convert_vx_to_fp(freg_t *fprs, __vector128 *vxrs)
{
int i;
for (i = 0; i < __NUM_FPRS; i++)
fprs[i].ui = vxrs[i].high;
}
static inline void convert_fp_to_vx(__vector128 *vxrs, freg_t *fprs)
{
int i;
for (i = 0; i < __NUM_FPRS; i++)
vxrs[i].high = fprs[i].ui;
}
static inline void fpregs_store(_s390_fp_regs *fpregs, struct fpu *fpu)
{
fpregs->pad = 0;
fpregs->fpc = fpu->fpc;
convert_vx_to_fp((freg_t *)&fpregs->fprs, fpu->vxrs);
}
static inline void fpregs_load(_s390_fp_regs *fpregs, struct fpu *fpu)
{
fpu->fpc = fpregs->fpc;
convert_fp_to_vx(fpu->vxrs, (freg_t *)&fpregs->fprs);
}
#endif /* _ASM_S390_FPU_H */
|