drm/panthor: Don't add write fences to the shared BOs
[drm/drm-misc.git] / arch / riscv / kernel / traps_misaligned.c
blobd4fd8af7aaf5a937de9c3f50f0d52cd51f845645
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Copyright (C) 2020 Western Digital Corporation or its affiliates.
4 */
5 #include <linux/kernel.h>
6 #include <linux/init.h>
7 #include <linux/mm.h>
8 #include <linux/module.h>
9 #include <linux/perf_event.h>
10 #include <linux/irq.h>
11 #include <linux/stringify.h>
13 #include <asm/processor.h>
14 #include <asm/ptrace.h>
15 #include <asm/csr.h>
16 #include <asm/entry-common.h>
17 #include <asm/hwprobe.h>
18 #include <asm/cpufeature.h>
20 #define INSN_MATCH_LB 0x3
21 #define INSN_MASK_LB 0x707f
22 #define INSN_MATCH_LH 0x1003
23 #define INSN_MASK_LH 0x707f
24 #define INSN_MATCH_LW 0x2003
25 #define INSN_MASK_LW 0x707f
26 #define INSN_MATCH_LD 0x3003
27 #define INSN_MASK_LD 0x707f
28 #define INSN_MATCH_LBU 0x4003
29 #define INSN_MASK_LBU 0x707f
30 #define INSN_MATCH_LHU 0x5003
31 #define INSN_MASK_LHU 0x707f
32 #define INSN_MATCH_LWU 0x6003
33 #define INSN_MASK_LWU 0x707f
34 #define INSN_MATCH_SB 0x23
35 #define INSN_MASK_SB 0x707f
36 #define INSN_MATCH_SH 0x1023
37 #define INSN_MASK_SH 0x707f
38 #define INSN_MATCH_SW 0x2023
39 #define INSN_MASK_SW 0x707f
40 #define INSN_MATCH_SD 0x3023
41 #define INSN_MASK_SD 0x707f
43 #define INSN_MATCH_FLW 0x2007
44 #define INSN_MASK_FLW 0x707f
45 #define INSN_MATCH_FLD 0x3007
46 #define INSN_MASK_FLD 0x707f
47 #define INSN_MATCH_FLQ 0x4007
48 #define INSN_MASK_FLQ 0x707f
49 #define INSN_MATCH_FSW 0x2027
50 #define INSN_MASK_FSW 0x707f
51 #define INSN_MATCH_FSD 0x3027
52 #define INSN_MASK_FSD 0x707f
53 #define INSN_MATCH_FSQ 0x4027
54 #define INSN_MASK_FSQ 0x707f
56 #define INSN_MATCH_C_LD 0x6000
57 #define INSN_MASK_C_LD 0xe003
58 #define INSN_MATCH_C_SD 0xe000
59 #define INSN_MASK_C_SD 0xe003
60 #define INSN_MATCH_C_LW 0x4000
61 #define INSN_MASK_C_LW 0xe003
62 #define INSN_MATCH_C_SW 0xc000
63 #define INSN_MASK_C_SW 0xe003
64 #define INSN_MATCH_C_LDSP 0x6002
65 #define INSN_MASK_C_LDSP 0xe003
66 #define INSN_MATCH_C_SDSP 0xe002
67 #define INSN_MASK_C_SDSP 0xe003
68 #define INSN_MATCH_C_LWSP 0x4002
69 #define INSN_MASK_C_LWSP 0xe003
70 #define INSN_MATCH_C_SWSP 0xc002
71 #define INSN_MASK_C_SWSP 0xe003
73 #define INSN_MATCH_C_FLD 0x2000
74 #define INSN_MASK_C_FLD 0xe003
75 #define INSN_MATCH_C_FLW 0x6000
76 #define INSN_MASK_C_FLW 0xe003
77 #define INSN_MATCH_C_FSD 0xa000
78 #define INSN_MASK_C_FSD 0xe003
79 #define INSN_MATCH_C_FSW 0xe000
80 #define INSN_MASK_C_FSW 0xe003
81 #define INSN_MATCH_C_FLDSP 0x2002
82 #define INSN_MASK_C_FLDSP 0xe003
83 #define INSN_MATCH_C_FSDSP 0xa002
84 #define INSN_MASK_C_FSDSP 0xe003
85 #define INSN_MATCH_C_FLWSP 0x6002
86 #define INSN_MASK_C_FLWSP 0xe003
87 #define INSN_MATCH_C_FSWSP 0xe002
88 #define INSN_MASK_C_FSWSP 0xe003
90 #define INSN_LEN(insn) ((((insn) & 0x3) < 0x3) ? 2 : 4)
92 #if defined(CONFIG_64BIT)
93 #define LOG_REGBYTES 3
94 #define XLEN 64
95 #else
96 #define LOG_REGBYTES 2
97 #define XLEN 32
98 #endif
99 #define REGBYTES (1 << LOG_REGBYTES)
100 #define XLEN_MINUS_16 ((XLEN) - 16)
102 #define SH_RD 7
103 #define SH_RS1 15
104 #define SH_RS2 20
105 #define SH_RS2C 2
107 #define RV_X(x, s, n) (((x) >> (s)) & ((1 << (n)) - 1))
108 #define RVC_LW_IMM(x) ((RV_X(x, 6, 1) << 2) | \
109 (RV_X(x, 10, 3) << 3) | \
110 (RV_X(x, 5, 1) << 6))
111 #define RVC_LD_IMM(x) ((RV_X(x, 10, 3) << 3) | \
112 (RV_X(x, 5, 2) << 6))
113 #define RVC_LWSP_IMM(x) ((RV_X(x, 4, 3) << 2) | \
114 (RV_X(x, 12, 1) << 5) | \
115 (RV_X(x, 2, 2) << 6))
116 #define RVC_LDSP_IMM(x) ((RV_X(x, 5, 2) << 3) | \
117 (RV_X(x, 12, 1) << 5) | \
118 (RV_X(x, 2, 3) << 6))
119 #define RVC_SWSP_IMM(x) ((RV_X(x, 9, 4) << 2) | \
120 (RV_X(x, 7, 2) << 6))
121 #define RVC_SDSP_IMM(x) ((RV_X(x, 10, 3) << 3) | \
122 (RV_X(x, 7, 3) << 6))
123 #define RVC_RS1S(insn) (8 + RV_X(insn, SH_RD, 3))
124 #define RVC_RS2S(insn) (8 + RV_X(insn, SH_RS2C, 3))
125 #define RVC_RS2(insn) RV_X(insn, SH_RS2C, 5)
127 #define SHIFT_RIGHT(x, y) \
128 ((y) < 0 ? ((x) << -(y)) : ((x) >> (y)))
130 #define REG_MASK \
131 ((1 << (5 + LOG_REGBYTES)) - (1 << LOG_REGBYTES))
133 #define REG_OFFSET(insn, pos) \
134 (SHIFT_RIGHT((insn), (pos) - LOG_REGBYTES) & REG_MASK)
136 #define REG_PTR(insn, pos, regs) \
137 (ulong *)((ulong)(regs) + REG_OFFSET(insn, pos))
139 #define GET_RM(insn) (((insn) >> 12) & 7)
141 #define GET_RS1(insn, regs) (*REG_PTR(insn, SH_RS1, regs))
142 #define GET_RS2(insn, regs) (*REG_PTR(insn, SH_RS2, regs))
143 #define GET_RS1S(insn, regs) (*REG_PTR(RVC_RS1S(insn), 0, regs))
144 #define GET_RS2S(insn, regs) (*REG_PTR(RVC_RS2S(insn), 0, regs))
145 #define GET_RS2C(insn, regs) (*REG_PTR(insn, SH_RS2C, regs))
146 #define GET_SP(regs) (*REG_PTR(2, 0, regs))
147 #define SET_RD(insn, regs, val) (*REG_PTR(insn, SH_RD, regs) = (val))
148 #define IMM_I(insn) ((s32)(insn) >> 20)
149 #define IMM_S(insn) (((s32)(insn) >> 25 << 5) | \
150 (s32)(((insn) >> 7) & 0x1f))
151 #define MASK_FUNCT3 0x7000
153 #define GET_PRECISION(insn) (((insn) >> 25) & 3)
154 #define GET_RM(insn) (((insn) >> 12) & 7)
155 #define PRECISION_S 0
156 #define PRECISION_D 1
158 #ifdef CONFIG_FPU
160 #define FP_GET_RD(insn) (insn >> 7 & 0x1F)
162 extern void put_f32_reg(unsigned long fp_reg, unsigned long value);
164 static int set_f32_rd(unsigned long insn, struct pt_regs *regs,
165 unsigned long val)
167 unsigned long fp_reg = FP_GET_RD(insn);
169 put_f32_reg(fp_reg, val);
170 regs->status |= SR_FS_DIRTY;
172 return 0;
175 extern void put_f64_reg(unsigned long fp_reg, unsigned long value);
177 static int set_f64_rd(unsigned long insn, struct pt_regs *regs, u64 val)
179 unsigned long fp_reg = FP_GET_RD(insn);
180 unsigned long value;
182 #if __riscv_xlen == 32
183 value = (unsigned long) &val;
184 #else
185 value = val;
186 #endif
187 put_f64_reg(fp_reg, value);
188 regs->status |= SR_FS_DIRTY;
190 return 0;
193 #if __riscv_xlen == 32
194 extern void get_f64_reg(unsigned long fp_reg, u64 *value);
196 static u64 get_f64_rs(unsigned long insn, u8 fp_reg_offset,
197 struct pt_regs *regs)
199 unsigned long fp_reg = (insn >> fp_reg_offset) & 0x1F;
200 u64 val;
202 get_f64_reg(fp_reg, &val);
203 regs->status |= SR_FS_DIRTY;
205 return val;
207 #else
209 extern unsigned long get_f64_reg(unsigned long fp_reg);
211 static unsigned long get_f64_rs(unsigned long insn, u8 fp_reg_offset,
212 struct pt_regs *regs)
214 unsigned long fp_reg = (insn >> fp_reg_offset) & 0x1F;
215 unsigned long val;
217 val = get_f64_reg(fp_reg);
218 regs->status |= SR_FS_DIRTY;
220 return val;
223 #endif
225 extern unsigned long get_f32_reg(unsigned long fp_reg);
227 static unsigned long get_f32_rs(unsigned long insn, u8 fp_reg_offset,
228 struct pt_regs *regs)
230 unsigned long fp_reg = (insn >> fp_reg_offset) & 0x1F;
231 unsigned long val;
233 val = get_f32_reg(fp_reg);
234 regs->status |= SR_FS_DIRTY;
236 return val;
239 #else /* CONFIG_FPU */
240 static void set_f32_rd(unsigned long insn, struct pt_regs *regs,
241 unsigned long val) {}
243 static void set_f64_rd(unsigned long insn, struct pt_regs *regs, u64 val) {}
245 static unsigned long get_f64_rs(unsigned long insn, u8 fp_reg_offset,
246 struct pt_regs *regs)
248 return 0;
251 static unsigned long get_f32_rs(unsigned long insn, u8 fp_reg_offset,
252 struct pt_regs *regs)
254 return 0;
257 #endif
259 #define GET_F64_RS2(insn, regs) (get_f64_rs(insn, 20, regs))
260 #define GET_F64_RS2C(insn, regs) (get_f64_rs(insn, 2, regs))
261 #define GET_F64_RS2S(insn, regs) (get_f64_rs(RVC_RS2S(insn), 0, regs))
263 #define GET_F32_RS2(insn, regs) (get_f32_rs(insn, 20, regs))
264 #define GET_F32_RS2C(insn, regs) (get_f32_rs(insn, 2, regs))
265 #define GET_F32_RS2S(insn, regs) (get_f32_rs(RVC_RS2S(insn), 0, regs))
267 #define __read_insn(regs, insn, insn_addr, type) \
268 ({ \
269 int __ret; \
271 if (user_mode(regs)) { \
272 __ret = __get_user(insn, (type __user *) insn_addr); \
273 } else { \
274 insn = *(type *)insn_addr; \
275 __ret = 0; \
278 __ret; \
281 static inline int get_insn(struct pt_regs *regs, ulong epc, ulong *r_insn)
283 ulong insn = 0;
285 if (epc & 0x2) {
286 ulong tmp = 0;
288 if (__read_insn(regs, insn, epc, u16))
289 return -EFAULT;
290 /* __get_user() uses regular "lw" which sign extend the loaded
291 * value make sure to clear higher order bits in case we "or" it
292 * below with the upper 16 bits half.
294 insn &= GENMASK(15, 0);
295 if ((insn & __INSN_LENGTH_MASK) != __INSN_LENGTH_32) {
296 *r_insn = insn;
297 return 0;
299 epc += sizeof(u16);
300 if (__read_insn(regs, tmp, epc, u16))
301 return -EFAULT;
302 *r_insn = (tmp << 16) | insn;
304 return 0;
305 } else {
306 if (__read_insn(regs, insn, epc, u32))
307 return -EFAULT;
308 if ((insn & __INSN_LENGTH_MASK) == __INSN_LENGTH_32) {
309 *r_insn = insn;
310 return 0;
312 insn &= GENMASK(15, 0);
313 *r_insn = insn;
315 return 0;
319 union reg_data {
320 u8 data_bytes[8];
321 ulong data_ulong;
322 u64 data_u64;
325 static bool unaligned_ctl __read_mostly;
327 /* sysctl hooks */
328 int unaligned_enabled __read_mostly = 1; /* Enabled by default */
330 int handle_misaligned_load(struct pt_regs *regs)
332 union reg_data val;
333 unsigned long epc = regs->epc;
334 unsigned long insn;
335 unsigned long addr = regs->badaddr;
336 int fp = 0, shift = 0, len = 0;
338 perf_sw_event(PERF_COUNT_SW_ALIGNMENT_FAULTS, 1, regs, addr);
340 #ifdef CONFIG_RISCV_PROBE_UNALIGNED_ACCESS
341 *this_cpu_ptr(&misaligned_access_speed) = RISCV_HWPROBE_MISALIGNED_SCALAR_EMULATED;
342 #endif
344 if (!unaligned_enabled)
345 return -1;
347 if (user_mode(regs) && (current->thread.align_ctl & PR_UNALIGN_SIGBUS))
348 return -1;
350 if (get_insn(regs, epc, &insn))
351 return -1;
353 regs->epc = 0;
355 if ((insn & INSN_MASK_LW) == INSN_MATCH_LW) {
356 len = 4;
357 shift = 8 * (sizeof(unsigned long) - len);
358 #if defined(CONFIG_64BIT)
359 } else if ((insn & INSN_MASK_LD) == INSN_MATCH_LD) {
360 len = 8;
361 shift = 8 * (sizeof(unsigned long) - len);
362 } else if ((insn & INSN_MASK_LWU) == INSN_MATCH_LWU) {
363 len = 4;
364 #endif
365 } else if ((insn & INSN_MASK_FLD) == INSN_MATCH_FLD) {
366 fp = 1;
367 len = 8;
368 } else if ((insn & INSN_MASK_FLW) == INSN_MATCH_FLW) {
369 fp = 1;
370 len = 4;
371 } else if ((insn & INSN_MASK_LH) == INSN_MATCH_LH) {
372 len = 2;
373 shift = 8 * (sizeof(unsigned long) - len);
374 } else if ((insn & INSN_MASK_LHU) == INSN_MATCH_LHU) {
375 len = 2;
376 #if defined(CONFIG_64BIT)
377 } else if ((insn & INSN_MASK_C_LD) == INSN_MATCH_C_LD) {
378 len = 8;
379 shift = 8 * (sizeof(unsigned long) - len);
380 insn = RVC_RS2S(insn) << SH_RD;
381 } else if ((insn & INSN_MASK_C_LDSP) == INSN_MATCH_C_LDSP &&
382 ((insn >> SH_RD) & 0x1f)) {
383 len = 8;
384 shift = 8 * (sizeof(unsigned long) - len);
385 #endif
386 } else if ((insn & INSN_MASK_C_LW) == INSN_MATCH_C_LW) {
387 len = 4;
388 shift = 8 * (sizeof(unsigned long) - len);
389 insn = RVC_RS2S(insn) << SH_RD;
390 } else if ((insn & INSN_MASK_C_LWSP) == INSN_MATCH_C_LWSP &&
391 ((insn >> SH_RD) & 0x1f)) {
392 len = 4;
393 shift = 8 * (sizeof(unsigned long) - len);
394 } else if ((insn & INSN_MASK_C_FLD) == INSN_MATCH_C_FLD) {
395 fp = 1;
396 len = 8;
397 insn = RVC_RS2S(insn) << SH_RD;
398 } else if ((insn & INSN_MASK_C_FLDSP) == INSN_MATCH_C_FLDSP) {
399 fp = 1;
400 len = 8;
401 #if defined(CONFIG_32BIT)
402 } else if ((insn & INSN_MASK_C_FLW) == INSN_MATCH_C_FLW) {
403 fp = 1;
404 len = 4;
405 insn = RVC_RS2S(insn) << SH_RD;
406 } else if ((insn & INSN_MASK_C_FLWSP) == INSN_MATCH_C_FLWSP) {
407 fp = 1;
408 len = 4;
409 #endif
410 } else {
411 regs->epc = epc;
412 return -1;
415 if (!IS_ENABLED(CONFIG_FPU) && fp)
416 return -EOPNOTSUPP;
418 val.data_u64 = 0;
419 if (user_mode(regs)) {
420 if (copy_from_user(&val, (u8 __user *)addr, len))
421 return -1;
422 } else {
423 memcpy(&val, (u8 *)addr, len);
426 if (!fp)
427 SET_RD(insn, regs, val.data_ulong << shift >> shift);
428 else if (len == 8)
429 set_f64_rd(insn, regs, val.data_u64);
430 else
431 set_f32_rd(insn, regs, val.data_ulong);
433 regs->epc = epc + INSN_LEN(insn);
435 return 0;
438 int handle_misaligned_store(struct pt_regs *regs)
440 union reg_data val;
441 unsigned long epc = regs->epc;
442 unsigned long insn;
443 unsigned long addr = regs->badaddr;
444 int len = 0, fp = 0;
446 perf_sw_event(PERF_COUNT_SW_ALIGNMENT_FAULTS, 1, regs, addr);
448 if (!unaligned_enabled)
449 return -1;
451 if (user_mode(regs) && (current->thread.align_ctl & PR_UNALIGN_SIGBUS))
452 return -1;
454 if (get_insn(regs, epc, &insn))
455 return -1;
457 regs->epc = 0;
459 val.data_ulong = GET_RS2(insn, regs);
461 if ((insn & INSN_MASK_SW) == INSN_MATCH_SW) {
462 len = 4;
463 #if defined(CONFIG_64BIT)
464 } else if ((insn & INSN_MASK_SD) == INSN_MATCH_SD) {
465 len = 8;
466 #endif
467 } else if ((insn & INSN_MASK_FSD) == INSN_MATCH_FSD) {
468 fp = 1;
469 len = 8;
470 val.data_u64 = GET_F64_RS2(insn, regs);
471 } else if ((insn & INSN_MASK_FSW) == INSN_MATCH_FSW) {
472 fp = 1;
473 len = 4;
474 val.data_ulong = GET_F32_RS2(insn, regs);
475 } else if ((insn & INSN_MASK_SH) == INSN_MATCH_SH) {
476 len = 2;
477 #if defined(CONFIG_64BIT)
478 } else if ((insn & INSN_MASK_C_SD) == INSN_MATCH_C_SD) {
479 len = 8;
480 val.data_ulong = GET_RS2S(insn, regs);
481 } else if ((insn & INSN_MASK_C_SDSP) == INSN_MATCH_C_SDSP) {
482 len = 8;
483 val.data_ulong = GET_RS2C(insn, regs);
484 #endif
485 } else if ((insn & INSN_MASK_C_SW) == INSN_MATCH_C_SW) {
486 len = 4;
487 val.data_ulong = GET_RS2S(insn, regs);
488 } else if ((insn & INSN_MASK_C_SWSP) == INSN_MATCH_C_SWSP) {
489 len = 4;
490 val.data_ulong = GET_RS2C(insn, regs);
491 } else if ((insn & INSN_MASK_C_FSD) == INSN_MATCH_C_FSD) {
492 fp = 1;
493 len = 8;
494 val.data_u64 = GET_F64_RS2S(insn, regs);
495 } else if ((insn & INSN_MASK_C_FSDSP) == INSN_MATCH_C_FSDSP) {
496 fp = 1;
497 len = 8;
498 val.data_u64 = GET_F64_RS2C(insn, regs);
499 #if !defined(CONFIG_64BIT)
500 } else if ((insn & INSN_MASK_C_FSW) == INSN_MATCH_C_FSW) {
501 fp = 1;
502 len = 4;
503 val.data_ulong = GET_F32_RS2S(insn, regs);
504 } else if ((insn & INSN_MASK_C_FSWSP) == INSN_MATCH_C_FSWSP) {
505 fp = 1;
506 len = 4;
507 val.data_ulong = GET_F32_RS2C(insn, regs);
508 #endif
509 } else {
510 regs->epc = epc;
511 return -1;
514 if (!IS_ENABLED(CONFIG_FPU) && fp)
515 return -EOPNOTSUPP;
517 if (user_mode(regs)) {
518 if (copy_to_user((u8 __user *)addr, &val, len))
519 return -1;
520 } else {
521 memcpy((u8 *)addr, &val, len);
524 regs->epc = epc + INSN_LEN(insn);
526 return 0;
529 static bool check_unaligned_access_emulated(int cpu)
531 long *mas_ptr = per_cpu_ptr(&misaligned_access_speed, cpu);
532 unsigned long tmp_var, tmp_val;
533 bool misaligned_emu_detected;
535 *mas_ptr = RISCV_HWPROBE_MISALIGNED_SCALAR_UNKNOWN;
537 __asm__ __volatile__ (
538 " "REG_L" %[tmp], 1(%[ptr])\n"
539 : [tmp] "=r" (tmp_val) : [ptr] "r" (&tmp_var) : "memory");
541 misaligned_emu_detected = (*mas_ptr == RISCV_HWPROBE_MISALIGNED_SCALAR_EMULATED);
543 * If unaligned_ctl is already set, this means that we detected that all
544 * CPUS uses emulated misaligned access at boot time. If that changed
545 * when hotplugging the new cpu, this is something we don't handle.
547 if (unlikely(unaligned_ctl && !misaligned_emu_detected)) {
548 pr_crit("CPU misaligned accesses non homogeneous (expected all emulated)\n");
549 while (true)
550 cpu_relax();
553 return misaligned_emu_detected;
556 bool check_unaligned_access_emulated_all_cpus(void)
558 int cpu;
561 * We can only support PR_UNALIGN controls if all CPUs have misaligned
562 * accesses emulated since tasks requesting such control can run on any
563 * CPU.
565 for_each_online_cpu(cpu)
566 if (!check_unaligned_access_emulated(cpu))
567 return false;
569 unaligned_ctl = true;
570 return true;
573 bool unaligned_ctl_available(void)
575 return unaligned_ctl;