Commit | Line | Data |
---|---|---|
caab277b | 1 | /* SPDX-License-Identifier: GPL-2.0-only */ |
53631b54 CM |
2 | /* |
3 | * Copyright (C) 2012 ARM Ltd. | |
53631b54 CM |
4 | */ |
5 | #ifndef __ASM_FP_H | |
6 | #define __ASM_FP_H | |
7 | ||
2d2123bc | 8 | #include <asm/errno.h> |
f9209e26 | 9 | #include <asm/ptrace.h> |
9a6e5948 DM |
10 | #include <asm/processor.h> |
11 | #include <asm/sigcontext.h> | |
f9209e26 | 12 | #include <asm/sysreg.h> |
53631b54 CM |
13 | |
14 | #ifndef __ASSEMBLY__ | |
15 | ||
ead9e430 | 16 | #include <linux/bitmap.h> |
f9209e26 | 17 | #include <linux/build_bug.h> |
ead9e430 | 18 | #include <linux/bug.h> |
7582e220 | 19 | #include <linux/cache.h> |
b4f9b390 | 20 | #include <linux/init.h> |
bc0ee476 | 21 | #include <linux/stddef.h> |
ead9e430 | 22 | #include <linux/types.h> |
bc0ee476 | 23 | |
b907b80d | 24 | #ifdef CONFIG_COMPAT |
53631b54 CM |
25 | /* Masks for extracting the FPSR and FPCR from the FPSCR */ |
26 | #define VFP_FPSCR_STAT_MASK 0xf800009f | |
27 | #define VFP_FPSCR_CTRL_MASK 0x07f79f00 | |
28 | /* | |
29 | * The VFP state has 32x64-bit registers and a single 32-bit | |
30 | * control/status register. | |
31 | */ | |
32 | #define VFP_STATE_SIZE ((32 * 8) + 4) | |
33 | #endif | |
34 | ||
d158a060 MB |
35 | /* |
36 | * When we defined the maximum SVE vector length we defined the ABI so | |
37 | * that the maximum vector length included all the reserved for future | |
38 | * expansion bits in ZCR rather than those just currently defined by | |
39 | * the architecture. While SME follows a similar pattern the fact that | |
40 | * it includes a square matrix means that any allocations that attempt | |
41 | * to cover the maximum potential vector length (such as happen with | |
42 | * the regset used for ptrace) end up being extremely large. Define | |
43 | * the much lower actual limit for use in such situations. | |
44 | */ | |
45 | #define SME_VQ_MAX 16 | |
46 | ||
53631b54 CM |
47 | struct task_struct; |
48 | ||
20b85472 DM |
49 | extern void fpsimd_save_state(struct user_fpsimd_state *state); |
50 | extern void fpsimd_load_state(struct user_fpsimd_state *state); | |
53631b54 CM |
51 | |
52 | extern void fpsimd_thread_switch(struct task_struct *next); | |
53 | extern void fpsimd_flush_thread(void); | |
54 | ||
8cd969d2 | 55 | extern void fpsimd_signal_preserve_current_state(void); |
c51f9269 | 56 | extern void fpsimd_preserve_current_state(void); |
005f78cd | 57 | extern void fpsimd_restore_current_state(void); |
0abdeff5 | 58 | extern void fpsimd_update_current_state(struct user_fpsimd_state const *state); |
93ae6b01 | 59 | extern void fpsimd_kvm_prepare(void); |
c51f9269 | 60 | |
1192b93b MB |
61 | struct cpu_fp_state { |
62 | struct user_fpsimd_state *st; | |
63 | void *sve_state; | |
ce514000 | 64 | void *sme_state; |
1192b93b MB |
65 | u64 *svcr; |
66 | unsigned int sve_vl; | |
67 | unsigned int sme_vl; | |
68 | enum fp_type *fp_type; | |
69 | enum fp_type to_save; | |
70 | }; | |
71 | ||
72 | extern void fpsimd_bind_state_to_cpu(struct cpu_fp_state *fp_state); | |
e6b673b7 | 73 | |
005f78cd | 74 | extern void fpsimd_flush_task_state(struct task_struct *target); |
54b8c7cb | 75 | extern void fpsimd_save_and_flush_cpu_state(void); |
005f78cd | 76 | |
af7167d6 MB |
77 | static inline bool thread_sm_enabled(struct thread_struct *thread) |
78 | { | |
ec0067a6 | 79 | return system_supports_sme() && (thread->svcr & SVCR_SM_MASK); |
af7167d6 MB |
80 | } |
81 | ||
82 | static inline bool thread_za_enabled(struct thread_struct *thread) | |
83 | { | |
ec0067a6 | 84 | return system_supports_sme() && (thread->svcr & SVCR_ZA_MASK); |
af7167d6 MB |
85 | } |
86 | ||
30c43e73 MB |
87 | /* Maximum VL that SVE/SME VL-agnostic software can transparently support */ |
88 | #define VL_ARCH_MAX 0x100 | |
7582e220 | 89 | |
9a6e5948 DM |
90 | /* Offset of FFR in the SVE register dump */ |
91 | static inline size_t sve_ffr_offset(int vl) | |
92 | { | |
93 | return SVE_SIG_FFR_OFFSET(sve_vq_from_vl(vl)) - SVE_SIG_REGS_OFFSET; | |
94 | } | |
95 | ||
96 | static inline void *sve_pffr(struct thread_struct *thread) | |
97 | { | |
af7167d6 MB |
98 | unsigned int vl; |
99 | ||
100 | if (system_supports_sme() && thread_sm_enabled(thread)) | |
101 | vl = thread_get_sme_vl(thread); | |
102 | else | |
103 | vl = thread_get_sve_vl(thread); | |
104 | ||
105 | return (char *)thread->sve_state + sve_ffr_offset(vl); | |
9a6e5948 DM |
106 | } |
107 | ||
d6138b4a MB |
108 | static inline void *thread_zt_state(struct thread_struct *thread) |
109 | { | |
110 | /* The ZT register state is stored immediately after the ZA state */ | |
111 | unsigned int sme_vq = sve_vq_from_vl(thread_get_sme_vl(thread)); | |
112 | return thread->sme_state + ZA_SIG_REGS_SIZE(sme_vq); | |
113 | } | |
114 | ||
9f584866 | 115 | extern void sve_save_state(void *state, u32 *pfpsr, int save_ffr); |
1fc5dce7 | 116 | extern void sve_load_state(void const *state, u32 const *pfpsr, |
ddc806b5 | 117 | int restore_ffr); |
9f584866 | 118 | extern void sve_flush_live(bool flush_ffr, unsigned long vq_minus_1); |
1fc5dce7 | 119 | extern unsigned int sve_get_vl(void); |
cccb78ce | 120 | extern void sve_set_vq(unsigned long vq_minus_1); |
af7167d6 | 121 | extern void sme_set_vq(unsigned long vq_minus_1); |
95fcec71 MB |
122 | extern void sme_save_state(void *state, int zt); |
123 | extern void sme_load_state(void const *state, int zt); | |
c0cda3b8 DM |
124 | |
125 | struct arm64_cpu_capabilities; | |
126 | extern void sve_kernel_enable(const struct arm64_cpu_capabilities *__unused); | |
5e64b862 | 127 | extern void sme_kernel_enable(const struct arm64_cpu_capabilities *__unused); |
d4913eee | 128 | extern void sme2_kernel_enable(const struct arm64_cpu_capabilities *__unused); |
5e64b862 | 129 | extern void fa64_kernel_enable(const struct arm64_cpu_capabilities *__unused); |
1fc5dce7 | 130 | |
31dc52b3 | 131 | extern u64 read_zcr_features(void); |
b42990d3 | 132 | extern u64 read_smcr_features(void); |
31dc52b3 | 133 | |
ead9e430 DM |
134 | /* |
135 | * Helpers to translate bit indices in sve_vq_map to VQ values (and | |
136 | * vice versa). This allows find_next_bit() to be used to find the | |
137 | * _maximum_ VQ not exceeding a certain value. | |
138 | */ | |
139 | static inline unsigned int __vq_to_bit(unsigned int vq) | |
140 | { | |
141 | return SVE_VQ_MAX - vq; | |
142 | } | |
143 | ||
144 | static inline unsigned int __bit_to_vq(unsigned int bit) | |
145 | { | |
ead9e430 DM |
146 | return SVE_VQ_MAX - bit; |
147 | } | |
148 | ||
b5bc00ff MB |
149 | |
150 | struct vl_info { | |
151 | enum vec_type type; | |
152 | const char *name; /* For display purposes */ | |
153 | ||
154 | /* Minimum supported vector length across all CPUs */ | |
155 | int min_vl; | |
156 | ||
157 | /* Maximum supported vector length across all CPUs */ | |
158 | int max_vl; | |
159 | int max_virtualisable_vl; | |
160 | ||
161 | /* | |
162 | * Set of available vector lengths, | |
163 | * where length vq encoded as bit __vq_to_bit(vq): | |
164 | */ | |
165 | DECLARE_BITMAP(vq_map, SVE_VQ_MAX); | |
166 | ||
167 | /* Set of vector lengths present on at least one cpu: */ | |
168 | DECLARE_BITMAP(vq_partial_map, SVE_VQ_MAX); | |
169 | }; | |
7582e220 | 170 | |
bc0ee476 DM |
171 | #ifdef CONFIG_ARM64_SVE |
172 | ||
826a4fdd | 173 | extern void sve_alloc(struct task_struct *task, bool flush); |
bc0ee476 | 174 | extern void fpsimd_release_task(struct task_struct *task); |
43d4da2c | 175 | extern void fpsimd_sync_to_sve(struct task_struct *task); |
e12310a0 | 176 | extern void fpsimd_force_sync_to_sve(struct task_struct *task); |
43d4da2c DM |
177 | extern void sve_sync_to_fpsimd(struct task_struct *task); |
178 | extern void sve_sync_from_fpsimd_zeropad(struct task_struct *task); | |
179 | ||
30c43e73 | 180 | extern int vec_set_vector_length(struct task_struct *task, enum vec_type type, |
7582e220 | 181 | unsigned long vl, unsigned long flags); |
bc0ee476 | 182 | |
2d2123bc DM |
183 | extern int sve_set_current_vl(unsigned long arg); |
184 | extern int sve_get_current_vl(void); | |
185 | ||
f9209e26 MR |
186 | static inline void sve_user_disable(void) |
187 | { | |
188 | sysreg_clear_set(cpacr_el1, CPACR_EL1_ZEN_EL0EN, 0); | |
189 | } | |
190 | ||
191 | static inline void sve_user_enable(void) | |
192 | { | |
193 | sysreg_clear_set(cpacr_el1, 0, CPACR_EL1_ZEN_EL0EN); | |
194 | } | |
195 | ||
71ce1ae5 MZ |
196 | #define sve_cond_update_zcr_vq(val, reg) \ |
197 | do { \ | |
198 | u64 __zcr = read_sysreg_s((reg)); \ | |
199 | u64 __new = __zcr & ~ZCR_ELx_LEN_MASK; \ | |
200 | __new |= (val) & ZCR_ELx_LEN_MASK; \ | |
201 | if (__zcr != __new) \ | |
202 | write_sysreg_s(__new, (reg)); \ | |
203 | } while (0) | |
204 | ||
2e0f2478 DM |
205 | /* |
206 | * Probing and setup functions. | |
207 | * Calls to these functions must be serialised with one another. | |
208 | */ | |
b5bc00ff MB |
209 | enum vec_type; |
210 | ||
211 | extern void __init vec_init_vq_map(enum vec_type type); | |
212 | extern void vec_update_vq_map(enum vec_type type); | |
213 | extern int vec_verify_vq_map(enum vec_type type); | |
2e0f2478 DM |
214 | extern void __init sve_setup(void); |
215 | ||
b5bc00ff MB |
216 | extern __ro_after_init struct vl_info vl_info[ARM64_VEC_MAX]; |
217 | ||
218 | static inline void write_vl(enum vec_type type, u64 val) | |
219 | { | |
220 | u64 tmp; | |
221 | ||
222 | switch (type) { | |
223 | #ifdef CONFIG_ARM64_SVE | |
224 | case ARM64_VEC_SVE: | |
225 | tmp = read_sysreg_s(SYS_ZCR_EL1) & ~ZCR_ELx_LEN_MASK; | |
226 | write_sysreg_s(tmp | val, SYS_ZCR_EL1); | |
227 | break; | |
b42990d3 MB |
228 | #endif |
229 | #ifdef CONFIG_ARM64_SME | |
230 | case ARM64_VEC_SME: | |
231 | tmp = read_sysreg_s(SYS_SMCR_EL1) & ~SMCR_ELx_LEN_MASK; | |
232 | write_sysreg_s(tmp | val, SYS_SMCR_EL1); | |
233 | break; | |
b5bc00ff MB |
234 | #endif |
235 | default: | |
236 | WARN_ON_ONCE(1); | |
237 | break; | |
238 | } | |
239 | } | |
240 | ||
241 | static inline int vec_max_vl(enum vec_type type) | |
242 | { | |
243 | return vl_info[type].max_vl; | |
244 | } | |
245 | ||
246 | static inline int vec_max_virtualisable_vl(enum vec_type type) | |
247 | { | |
248 | return vl_info[type].max_virtualisable_vl; | |
249 | } | |
250 | ||
251 | static inline int sve_max_vl(void) | |
252 | { | |
253 | return vec_max_vl(ARM64_VEC_SVE); | |
254 | } | |
255 | ||
256 | static inline int sve_max_virtualisable_vl(void) | |
257 | { | |
258 | return vec_max_virtualisable_vl(ARM64_VEC_SVE); | |
259 | } | |
260 | ||
261 | /* Ensure vq >= SVE_VQ_MIN && vq <= SVE_VQ_MAX before calling this function */ | |
262 | static inline bool vq_available(enum vec_type type, unsigned int vq) | |
263 | { | |
264 | return test_bit(__vq_to_bit(vq), vl_info[type].vq_map); | |
265 | } | |
266 | ||
267 | static inline bool sve_vq_available(unsigned int vq) | |
268 | { | |
269 | return vq_available(ARM64_VEC_SVE, vq); | |
270 | } | |
271 | ||
8bd7f91c MB |
272 | size_t sve_state_size(struct task_struct const *task); |
273 | ||
bc0ee476 DM |
274 | #else /* ! CONFIG_ARM64_SVE */ |
275 | ||
826a4fdd | 276 | static inline void sve_alloc(struct task_struct *task, bool flush) { } |
bc0ee476 | 277 | static inline void fpsimd_release_task(struct task_struct *task) { } |
43d4da2c DM |
278 | static inline void sve_sync_to_fpsimd(struct task_struct *task) { } |
279 | static inline void sve_sync_from_fpsimd_zeropad(struct task_struct *task) { } | |
280 | ||
49ed9204 MB |
281 | static inline int sve_max_virtualisable_vl(void) |
282 | { | |
283 | return 0; | |
284 | } | |
285 | ||
2d2123bc DM |
286 | static inline int sve_set_current_vl(unsigned long arg) |
287 | { | |
288 | return -EINVAL; | |
289 | } | |
290 | ||
291 | static inline int sve_get_current_vl(void) | |
292 | { | |
293 | return -EINVAL; | |
294 | } | |
295 | ||
b5bc00ff MB |
296 | static inline int sve_max_vl(void) |
297 | { | |
298 | return -EINVAL; | |
299 | } | |
300 | ||
301 | static inline bool sve_vq_available(unsigned int vq) { return false; } | |
302 | ||
f9209e26 MR |
303 | static inline void sve_user_disable(void) { BUILD_BUG(); } |
304 | static inline void sve_user_enable(void) { BUILD_BUG(); } | |
305 | ||
a9f8696d XT |
306 | #define sve_cond_update_zcr_vq(val, reg) do { } while (0) |
307 | ||
b5bc00ff MB |
308 | static inline void vec_init_vq_map(enum vec_type t) { } |
309 | static inline void vec_update_vq_map(enum vec_type t) { } | |
310 | static inline int vec_verify_vq_map(enum vec_type t) { return 0; } | |
2e0f2478 | 311 | static inline void sve_setup(void) { } |
bc0ee476 | 312 | |
8bd7f91c MB |
313 | static inline size_t sve_state_size(struct task_struct const *task) |
314 | { | |
315 | return 0; | |
316 | } | |
317 | ||
bc0ee476 DM |
318 | #endif /* ! CONFIG_ARM64_SVE */ |
319 | ||
ca8a4ebc MB |
320 | #ifdef CONFIG_ARM64_SME |
321 | ||
8bd7f91c MB |
322 | static inline void sme_user_disable(void) |
323 | { | |
324 | sysreg_clear_set(cpacr_el1, CPACR_EL1_SMEN_EL0EN, 0); | |
325 | } | |
326 | ||
327 | static inline void sme_user_enable(void) | |
328 | { | |
329 | sysreg_clear_set(cpacr_el1, 0, CPACR_EL1_SMEN_EL0EN); | |
330 | } | |
331 | ||
ca8a4ebc MB |
332 | static inline void sme_smstart_sm(void) |
333 | { | |
334 | asm volatile(__msr_s(SYS_SVCR_SMSTART_SM_EL0, "xzr")); | |
335 | } | |
336 | ||
337 | static inline void sme_smstop_sm(void) | |
338 | { | |
339 | asm volatile(__msr_s(SYS_SVCR_SMSTOP_SM_EL0, "xzr")); | |
340 | } | |
341 | ||
342 | static inline void sme_smstop(void) | |
343 | { | |
344 | asm volatile(__msr_s(SYS_SVCR_SMSTOP_SMZA_EL0, "xzr")); | |
345 | } | |
346 | ||
b42990d3 MB |
347 | extern void __init sme_setup(void); |
348 | ||
349 | static inline int sme_max_vl(void) | |
350 | { | |
351 | return vec_max_vl(ARM64_VEC_SME); | |
352 | } | |
353 | ||
354 | static inline int sme_max_virtualisable_vl(void) | |
355 | { | |
356 | return vec_max_virtualisable_vl(ARM64_VEC_SME); | |
357 | } | |
358 | ||
5d0a8d2f | 359 | extern void sme_alloc(struct task_struct *task, bool flush); |
b42990d3 | 360 | extern unsigned int sme_get_vl(void); |
9e4ab6c8 MB |
361 | extern int sme_set_current_vl(unsigned long arg); |
362 | extern int sme_get_current_vl(void); | |
b42990d3 | 363 | |
8bd7f91c MB |
364 | /* |
365 | * Return how many bytes of memory are required to store the full SME | |
ce514000 MB |
366 | * specific state for task, given task's currently configured vector |
367 | * length. | |
8bd7f91c | 368 | */ |
ce514000 | 369 | static inline size_t sme_state_size(struct task_struct const *task) |
8bd7f91c MB |
370 | { |
371 | unsigned int vl = task_get_sme_vl(task); | |
ce514000 | 372 | size_t size; |
8bd7f91c | 373 | |
ce514000 MB |
374 | size = ZA_SIG_REGS_SIZE(sve_vq_from_vl(vl)); |
375 | ||
d6138b4a MB |
376 | if (system_supports_sme2()) |
377 | size += ZT_SIG_REG_SIZE; | |
378 | ||
ce514000 | 379 | return size; |
8bd7f91c MB |
380 | } |
381 | ||
ca8a4ebc MB |
382 | #else |
383 | ||
8bd7f91c MB |
384 | static inline void sme_user_disable(void) { BUILD_BUG(); } |
385 | static inline void sme_user_enable(void) { BUILD_BUG(); } | |
386 | ||
ca8a4ebc MB |
387 | static inline void sme_smstart_sm(void) { } |
388 | static inline void sme_smstop_sm(void) { } | |
389 | static inline void sme_smstop(void) { } | |
390 | ||
5d0a8d2f | 391 | static inline void sme_alloc(struct task_struct *task, bool flush) { } |
b42990d3 MB |
392 | static inline void sme_setup(void) { } |
393 | static inline unsigned int sme_get_vl(void) { return 0; } | |
394 | static inline int sme_max_vl(void) { return 0; } | |
395 | static inline int sme_max_virtualisable_vl(void) { return 0; } | |
9e4ab6c8 MB |
396 | static inline int sme_set_current_vl(unsigned long arg) { return -EINVAL; } |
397 | static inline int sme_get_current_vl(void) { return -EINVAL; } | |
b42990d3 | 398 | |
ce514000 | 399 | static inline size_t sme_state_size(struct task_struct const *task) |
8bd7f91c MB |
400 | { |
401 | return 0; | |
402 | } | |
403 | ||
ca8a4ebc MB |
404 | #endif /* ! CONFIG_ARM64_SME */ |
405 | ||
4328825d DM |
406 | /* For use by EFI runtime services calls only */ |
407 | extern void __efi_fpsimd_begin(void); | |
408 | extern void __efi_fpsimd_end(void); | |
409 | ||
53631b54 CM |
410 | #endif |
411 | ||
412 | #endif |