]> git.ipfire.org Git - thirdparty/kernel/stable.git/blame - arch/riscv/kernel/cpufeature.c
Merge tag 'kvm-x86-mmu-6.7' of https://github.com/kvm-x86/linux into HEAD
[thirdparty/kernel/stable.git] / arch / riscv / kernel / cpufeature.c
CommitLineData
caab277b 1// SPDX-License-Identifier: GPL-2.0-only
e2c0cdfb
PD
2/*
3 * Copied from arch/arm64/kernel/cpufeature.c
4 *
5 * Copyright (C) 2015 ARM Ltd.
6 * Copyright (C) 2017 SiFive
e2c0cdfb
PD
7 */
8
396c0183 9#include <linux/acpi.h>
6bcff515 10#include <linux/bitmap.h>
2a31c54b 11#include <linux/ctype.h>
9daaca4a 12#include <linux/log2.h>
9493e6f3 13#include <linux/memory.h>
ff689fd2 14#include <linux/module.h>
e2c0cdfb 15#include <linux/of.h>
396c0183 16#include <asm/acpi.h>
ff689fd2 17#include <asm/alternative.h>
1631ba12 18#include <asm/cacheflush.h>
c2d3c844 19#include <asm/cpufeature.h>
e2c0cdfb 20#include <asm/hwcap.h>
584ea656 21#include <asm/hwprobe.h>
ff689fd2 22#include <asm/patch.h>
ff689fd2 23#include <asm/processor.h>
7017858e 24#include <asm/vector.h>
e2c0cdfb 25
584ea656
EG
26#include "copy-unaligned.h"
27
58004f26
TO
28#define NUM_ALPHA_EXTS ('z' - 'a' + 1)
29
584ea656
EG
30#define MISALIGNED_ACCESS_JIFFIES_LG2 1
31#define MISALIGNED_BUFFER_SIZE 0x4000
32#define MISALIGNED_COPY_SIZE ((MISALIGNED_BUFFER_SIZE / 2) - 0x80)
33
e2c0cdfb 34unsigned long elf_hwcap __read_mostly;
6bcff515
AP
35
36/* Host ISA bitmap */
37static DECLARE_BITMAP(riscv_isa, RISCV_ISA_EXT_MAX) __read_mostly;
38
82e9c66e
EG
39/* Per-cpu ISA extensions. */
40struct riscv_isainfo hart_isa[NR_CPUS];
41
62a31d6e
EG
42/* Performance information */
43DEFINE_PER_CPU(long, misaligned_access_speed);
44
6bcff515
AP
45/**
46 * riscv_isa_extension_base() - Get base extension word
47 *
48 * @isa_bitmap: ISA bitmap to use
49 * Return: base extension word as unsigned long value
50 *
51 * NOTE: If isa_bitmap is NULL then Host ISA bitmap will be used.
52 */
53unsigned long riscv_isa_extension_base(const unsigned long *isa_bitmap)
54{
55 if (!isa_bitmap)
56 return riscv_isa[0];
57 return isa_bitmap[0];
58}
59EXPORT_SYMBOL_GPL(riscv_isa_extension_base);
60
61/**
62 * __riscv_isa_extension_available() - Check whether given extension
63 * is available or not
64 *
65 * @isa_bitmap: ISA bitmap to use
66 * @bit: bit position of the desired extension
67 * Return: true or false
68 *
69 * NOTE: If isa_bitmap is NULL then Host ISA bitmap will be used.
70 */
71bool __riscv_isa_extension_available(const unsigned long *isa_bitmap, int bit)
72{
73 const unsigned long *bmap = (isa_bitmap) ? isa_bitmap : riscv_isa;
74
75 if (bit >= RISCV_ISA_EXT_MAX)
76 return false;
77
78 return test_bit(bit, bmap) ? true : false;
79}
80EXPORT_SYMBOL_GPL(__riscv_isa_extension_available);
81
fb0ff0a9
AJ
82static bool riscv_isa_extension_check(int id)
83{
9daaca4a
AJ
84 switch (id) {
85 case RISCV_ISA_EXT_ZICBOM:
86 if (!riscv_cbom_block_size) {
c818fea8 87 pr_err("Zicbom detected in ISA string, disabling as no cbom-block-size found\n");
9daaca4a
AJ
88 return false;
89 } else if (!is_power_of_2(riscv_cbom_block_size)) {
c818fea8 90 pr_err("Zicbom disabled as cbom-block-size present, but is not a power-of-2\n");
9daaca4a
AJ
91 return false;
92 }
93 return true;
7ea5a736
AJ
94 case RISCV_ISA_EXT_ZICBOZ:
95 if (!riscv_cboz_block_size) {
96 pr_err("Zicboz detected in ISA string, but no cboz-block-size found\n");
97 return false;
98 } else if (!is_power_of_2(riscv_cboz_block_size)) {
99 pr_err("cboz-block-size present, but is not a power-of-2\n");
100 return false;
101 }
102 return true;
9daaca4a
AJ
103 }
104
fb0ff0a9
AJ
105 return true;
106}
107
37f988dc
CD
108#define __RISCV_ISA_EXT_DATA(_name, _id) { \
109 .name = #_name, \
90700a4f 110 .property = #_name, \
37f988dc
CD
111 .id = _id, \
112}
8135ade3
CD
113
114/*
115 * The canonical order of ISA extension names in the ISA string is defined in
116 * chapter 27 of the unprivileged specification.
117 *
118 * Ordinarily, for in-kernel data structures, this order is unimportant but
119 * isa_ext_arr defines the order of the ISA string in /proc/cpuinfo.
120 *
121 * The specification uses vague wording, such as should, when it comes to
122 * ordering, so for our purposes the following rules apply:
123 *
124 * 1. All multi-letter extensions must be separated from other extensions by an
125 * underscore.
126 *
127 * 2. Additional standard extensions (starting with 'Z') must be sorted after
128 * single-letter extensions and before any higher-privileged extensions.
129 *
130 * 3. The first letter following the 'Z' conventionally indicates the most
131 * closely related alphabetical extension category, IMAFDQLCBKJTPVH.
132 * If multiple 'Z' extensions are named, they must be ordered first by
133 * category, then alphabetically within a category.
134 *
135 * 3. Standard supervisor-level extensions (starting with 'S') must be listed
136 * after standard unprivileged extensions. If multiple supervisor-level
137 * extensions are listed, they must be ordered alphabetically.
138 *
139 * 4. Standard machine-level extensions (starting with 'Zxm') must be listed
140 * after any lower-privileged, standard extensions. If multiple
141 * machine-level extensions are listed, they must be ordered
142 * alphabetically.
143 *
144 * 5. Non-standard extensions (starting with 'X') must be listed after all
145 * standard extensions. If multiple non-standard extensions are listed, they
146 * must be ordered alphabetically.
147 *
148 * An example string following the order is:
149 * rv64imadc_zifoo_zigoo_zafoo_sbar_scar_zxmbaz_xqux_xrux
150 *
151 * New entries to this struct should follow the ordering rules described above.
152 */
153const struct riscv_isa_ext_data riscv_isa_ext[] = {
effc122a
CD
154 __RISCV_ISA_EXT_DATA(i, RISCV_ISA_EXT_i),
155 __RISCV_ISA_EXT_DATA(m, RISCV_ISA_EXT_m),
156 __RISCV_ISA_EXT_DATA(a, RISCV_ISA_EXT_a),
157 __RISCV_ISA_EXT_DATA(f, RISCV_ISA_EXT_f),
158 __RISCV_ISA_EXT_DATA(d, RISCV_ISA_EXT_d),
159 __RISCV_ISA_EXT_DATA(q, RISCV_ISA_EXT_q),
160 __RISCV_ISA_EXT_DATA(c, RISCV_ISA_EXT_c),
161 __RISCV_ISA_EXT_DATA(b, RISCV_ISA_EXT_b),
162 __RISCV_ISA_EXT_DATA(k, RISCV_ISA_EXT_k),
163 __RISCV_ISA_EXT_DATA(j, RISCV_ISA_EXT_j),
164 __RISCV_ISA_EXT_DATA(p, RISCV_ISA_EXT_p),
165 __RISCV_ISA_EXT_DATA(v, RISCV_ISA_EXT_v),
166 __RISCV_ISA_EXT_DATA(h, RISCV_ISA_EXT_h),
8135ade3
CD
167 __RISCV_ISA_EXT_DATA(zicbom, RISCV_ISA_EXT_ZICBOM),
168 __RISCV_ISA_EXT_DATA(zicboz, RISCV_ISA_EXT_ZICBOZ),
169 __RISCV_ISA_EXT_DATA(zicntr, RISCV_ISA_EXT_ZICNTR),
662a601a 170 __RISCV_ISA_EXT_DATA(zicond, RISCV_ISA_EXT_ZICOND),
8135ade3
CD
171 __RISCV_ISA_EXT_DATA(zicsr, RISCV_ISA_EXT_ZICSR),
172 __RISCV_ISA_EXT_DATA(zifencei, RISCV_ISA_EXT_ZIFENCEI),
173 __RISCV_ISA_EXT_DATA(zihintpause, RISCV_ISA_EXT_ZIHINTPAUSE),
174 __RISCV_ISA_EXT_DATA(zihpm, RISCV_ISA_EXT_ZIHPM),
175 __RISCV_ISA_EXT_DATA(zba, RISCV_ISA_EXT_ZBA),
176 __RISCV_ISA_EXT_DATA(zbb, RISCV_ISA_EXT_ZBB),
177 __RISCV_ISA_EXT_DATA(zbs, RISCV_ISA_EXT_ZBS),
178 __RISCV_ISA_EXT_DATA(smaia, RISCV_ISA_EXT_SMAIA),
9dbaf381 179 __RISCV_ISA_EXT_DATA(smstateen, RISCV_ISA_EXT_SMSTATEEN),
8135ade3
CD
180 __RISCV_ISA_EXT_DATA(ssaia, RISCV_ISA_EXT_SSAIA),
181 __RISCV_ISA_EXT_DATA(sscofpmf, RISCV_ISA_EXT_SSCOFPMF),
182 __RISCV_ISA_EXT_DATA(sstc, RISCV_ISA_EXT_SSTC),
183 __RISCV_ISA_EXT_DATA(svinval, RISCV_ISA_EXT_SVINVAL),
184 __RISCV_ISA_EXT_DATA(svnapot, RISCV_ISA_EXT_SVNAPOT),
185 __RISCV_ISA_EXT_DATA(svpbmt, RISCV_ISA_EXT_SVPBMT),
186};
187
188const size_t riscv_isa_ext_count = ARRAY_SIZE(riscv_isa_ext);
189
4265b0ec
CD
190static void __init riscv_parse_isa_string(unsigned long *this_hwcap, struct riscv_isainfo *isainfo,
191 unsigned long *isa2hwcap, const char *isa)
192{
193 /*
194 * For all possible cpus, we have already validated in
195 * the boot process that they at least contain "rv" and
196 * whichever of "32"/"64" this kernel supports, and so this
197 * section can be skipped.
198 */
199 isa += 4;
200
201 while (*isa) {
202 const char *ext = isa++;
203 const char *ext_end = isa;
204 bool ext_long = false, ext_err = false;
205
206 switch (*ext) {
207 case 's':
208 /*
209 * Workaround for invalid single-letter 's' & 'u'(QEMU).
210 * No need to set the bit in riscv_isa as 's' & 'u' are
211 * not valid ISA extensions. It works until multi-letter
212 * extension starting with "Su" appears.
213 */
214 if (ext[-1] != '_' && ext[1] == 'u') {
215 ++isa;
216 ext_err = true;
217 break;
218 }
219 fallthrough;
220 case 'S':
221 case 'x':
222 case 'X':
223 case 'z':
224 case 'Z':
225 /*
226 * Before attempting to parse the extension itself, we find its end.
227 * As multi-letter extensions must be split from other multi-letter
228 * extensions with an "_", the end of a multi-letter extension will
229 * either be the null character or the "_" at the start of the next
230 * multi-letter extension.
231 *
232 * Next, as the extensions version is currently ignored, we
233 * eliminate that portion. This is done by parsing backwards from
234 * the end of the extension, removing any numbers. This may be a
235 * major or minor number however, so the process is repeated if a
236 * minor number was found.
237 *
238 * ext_end is intended to represent the first character *after* the
239 * name portion of an extension, but will be decremented to the last
240 * character itself while eliminating the extensions version number.
241 * A simple re-increment solves this problem.
242 */
243 ext_long = true;
244 for (; *isa && *isa != '_'; ++isa)
245 if (unlikely(!isalnum(*isa)))
246 ext_err = true;
247
248 ext_end = isa;
249 if (unlikely(ext_err))
250 break;
251
252 if (!isdigit(ext_end[-1]))
253 break;
254
255 while (isdigit(*--ext_end))
256 ;
257
258 if (tolower(ext_end[0]) != 'p' || !isdigit(ext_end[-1])) {
259 ++ext_end;
260 break;
261 }
262
263 while (isdigit(*--ext_end))
264 ;
265
266 ++ext_end;
267 break;
268 default:
269 /*
270 * Things are a little easier for single-letter extensions, as they
271 * are parsed forwards.
272 *
273 * After checking that our starting position is valid, we need to
274 * ensure that, when isa was incremented at the start of the loop,
275 * that it arrived at the start of the next extension.
276 *
277 * If we are already on a non-digit, there is nothing to do. Either
278 * we have a multi-letter extension's _, or the start of an
279 * extension.
280 *
281 * Otherwise we have found the current extension's major version
282 * number. Parse past it, and a subsequent p/minor version number
283 * if present. The `p` extension must not appear immediately after
284 * a number, so there is no fear of missing it.
285 *
286 */
287 if (unlikely(!isalpha(*ext))) {
288 ext_err = true;
289 break;
290 }
291
292 if (!isdigit(*isa))
293 break;
294
295 while (isdigit(*++isa))
296 ;
297
298 if (tolower(*isa) != 'p')
299 break;
300
301 if (!isdigit(*++isa)) {
302 --isa;
303 break;
304 }
305
306 while (isdigit(*++isa))
307 ;
308
309 break;
310 }
311
312 /*
313 * The parser expects that at the start of an iteration isa points to the
314 * first character of the next extension. As we stop parsing an extension
315 * on meeting a non-alphanumeric character, an extra increment is needed
316 * where the succeeding extension is a multi-letter prefixed with an "_".
317 */
318 if (*isa == '_')
319 ++isa;
320
321#define SET_ISA_EXT_MAP(name, bit) \
322 do { \
323 if ((ext_end - ext == strlen(name)) && \
324 !strncasecmp(ext, name, strlen(name)) && \
325 riscv_isa_extension_check(bit)) \
326 set_bit(bit, isainfo->isa); \
327 } while (false) \
328
329 if (unlikely(ext_err))
330 continue;
331 if (!ext_long) {
332 int nr = tolower(*ext) - 'a';
333
334 if (riscv_isa_extension_check(nr)) {
335 *this_hwcap |= isa2hwcap[nr];
336 set_bit(nr, isainfo->isa);
337 }
338 } else {
339 for (int i = 0; i < riscv_isa_ext_count; i++)
340 SET_ISA_EXT_MAP(riscv_isa_ext[i].name,
341 riscv_isa_ext[i].id);
342 }
343#undef SET_ISA_EXT_MAP
344 }
345}
346
347static void __init riscv_fill_hwcap_from_isa_string(unsigned long *isa2hwcap)
e2c0cdfb 348{
dd81c8ab 349 struct device_node *node;
e2c0cdfb 350 const char *isa;
4265b0ec 351 int rc;
396c0183
S
352 struct acpi_table_header *rhct;
353 acpi_status status;
914d6f44 354 unsigned int cpu;
e2c0cdfb 355
396c0183
S
356 if (!acpi_disabled) {
357 status = acpi_get_table(ACPI_SIG_RHCT, 0, &rhct);
358 if (ACPI_FAILURE(status))
359 return;
360 }
361
914d6f44 362 for_each_possible_cpu(cpu) {
16252e01 363 struct riscv_isainfo *isainfo = &hart_isa[cpu];
fbdc6193 364 unsigned long this_hwcap = 0;
e2c0cdfb 365
396c0183
S
366 if (acpi_disabled) {
367 node = of_cpu_device_node_get(cpu);
368 if (!node) {
369 pr_warn("Unable to find cpu node\n");
370 continue;
371 }
e2c0cdfb 372
396c0183
S
373 rc = of_property_read_string(node, "riscv,isa", &isa);
374 of_node_put(node);
375 if (rc) {
376 pr_warn("Unable to find \"riscv,isa\" devicetree entry\n");
377 continue;
378 }
379 } else {
380 rc = acpi_get_riscv_isa(rhct, cpu, &isa);
381 if (rc < 0) {
382 pr_warn("Unable to get ISA for the hart - %d\n", cpu);
383 continue;
384 }
fbdc6193
AP
385 }
386
4265b0ec 387 riscv_parse_isa_string(&this_hwcap, isainfo, isa2hwcap, isa);
fbdc6193 388
07edc327
CD
389 /*
390 * These ones were as they were part of the base ISA when the
391 * port & dt-bindings were upstreamed, and so can be set
392 * unconditionally where `i` is in riscv,isa on DT systems.
393 */
394 if (acpi_disabled) {
ab2dbc7a
PD
395 set_bit(RISCV_ISA_EXT_ZICSR, isainfo->isa);
396 set_bit(RISCV_ISA_EXT_ZIFENCEI, isainfo->isa);
42b89447
PD
397 set_bit(RISCV_ISA_EXT_ZICNTR, isainfo->isa);
398 set_bit(RISCV_ISA_EXT_ZIHPM, isainfo->isa);
07edc327
CD
399 }
400
fbdc6193
AP
401 /*
402 * All "okay" hart should have same isa. Set HWCAP based on
403 * common capabilities of every "okay" hart, in case they don't
404 * have.
405 */
406 if (elf_hwcap)
407 elf_hwcap &= this_hwcap;
408 else
409 elf_hwcap = this_hwcap;
6bcff515 410
8f51558e 411 if (bitmap_empty(riscv_isa, RISCV_ISA_EXT_MAX))
82e9c66e 412 bitmap_copy(riscv_isa, isainfo->isa, RISCV_ISA_EXT_MAX);
8f51558e 413 else
82e9c66e 414 bitmap_and(riscv_isa, riscv_isa, isainfo->isa, RISCV_ISA_EXT_MAX);
fbdc6193 415 }
e2c0cdfb 416
396c0183
S
417 if (!acpi_disabled && rhct)
418 acpi_put_table((struct acpi_table_header *)rhct);
4265b0ec
CD
419}
420
90700a4f
CD
421static int __init riscv_fill_hwcap_from_ext_list(unsigned long *isa2hwcap)
422{
423 unsigned int cpu;
424
425 for_each_possible_cpu(cpu) {
426 unsigned long this_hwcap = 0;
427 struct device_node *cpu_node;
428 struct riscv_isainfo *isainfo = &hart_isa[cpu];
429
430 cpu_node = of_cpu_device_node_get(cpu);
431 if (!cpu_node) {
432 pr_warn("Unable to find cpu node\n");
433 continue;
434 }
435
436 if (!of_property_present(cpu_node, "riscv,isa-extensions")) {
437 of_node_put(cpu_node);
438 continue;
439 }
440
441 for (int i = 0; i < riscv_isa_ext_count; i++) {
442 if (of_property_match_string(cpu_node, "riscv,isa-extensions",
443 riscv_isa_ext[i].property) < 0)
444 continue;
445
446 if (!riscv_isa_extension_check(riscv_isa_ext[i].id))
447 continue;
448
449 /* Only single letter extensions get set in hwcap */
450 if (strnlen(riscv_isa_ext[i].name, 2) == 1)
451 this_hwcap |= isa2hwcap[riscv_isa_ext[i].id];
452
453 set_bit(riscv_isa_ext[i].id, isainfo->isa);
454 }
455
456 of_node_put(cpu_node);
457
458 /*
459 * All "okay" harts should have same isa. Set HWCAP based on
460 * common capabilities of every "okay" hart, in case they don't.
461 */
462 if (elf_hwcap)
463 elf_hwcap &= this_hwcap;
464 else
465 elf_hwcap = this_hwcap;
466
467 if (bitmap_empty(riscv_isa, RISCV_ISA_EXT_MAX))
468 bitmap_copy(riscv_isa, isainfo->isa, RISCV_ISA_EXT_MAX);
469 else
470 bitmap_and(riscv_isa, riscv_isa, isainfo->isa, RISCV_ISA_EXT_MAX);
471 }
472
473 if (bitmap_empty(riscv_isa, RISCV_ISA_EXT_MAX))
474 return -ENOENT;
475
476 return 0;
477}
478
496ea826
CD
479#ifdef CONFIG_RISCV_ISA_FALLBACK
480bool __initdata riscv_isa_fallback = true;
481#else
482bool __initdata riscv_isa_fallback;
483static int __init riscv_isa_fallback_setup(char *__unused)
484{
485 riscv_isa_fallback = true;
486 return 1;
487}
488early_param("riscv_isa_fallback", riscv_isa_fallback_setup);
489#endif
490
4265b0ec
CD
491void __init riscv_fill_hwcap(void)
492{
493 char print_str[NUM_ALPHA_EXTS + 1];
4265b0ec 494 unsigned long isa2hwcap[26] = {0};
90700a4f 495 int i, j;
4265b0ec
CD
496
497 isa2hwcap['i' - 'a'] = COMPAT_HWCAP_ISA_I;
498 isa2hwcap['m' - 'a'] = COMPAT_HWCAP_ISA_M;
499 isa2hwcap['a' - 'a'] = COMPAT_HWCAP_ISA_A;
500 isa2hwcap['f' - 'a'] = COMPAT_HWCAP_ISA_F;
501 isa2hwcap['d' - 'a'] = COMPAT_HWCAP_ISA_D;
502 isa2hwcap['c' - 'a'] = COMPAT_HWCAP_ISA_C;
503 isa2hwcap['v' - 'a'] = COMPAT_HWCAP_ISA_V;
504
90700a4f
CD
505 if (!acpi_disabled) {
506 riscv_fill_hwcap_from_isa_string(isa2hwcap);
507 } else {
508 int ret = riscv_fill_hwcap_from_ext_list(isa2hwcap);
396c0183 509
496ea826 510 if (ret && riscv_isa_fallback) {
90700a4f
CD
511 pr_info("Falling back to deprecated \"riscv,isa\"\n");
512 riscv_fill_hwcap_from_isa_string(isa2hwcap);
513 }
514 }
396c0183 515
90700a4f
CD
516 /*
517 * We don't support systems with F but without D, so mask those out
518 * here.
519 */
86e581e3 520 if ((elf_hwcap & COMPAT_HWCAP_ISA_F) && !(elf_hwcap & COMPAT_HWCAP_ISA_D)) {
7265d103 521 pr_info("This kernel does not support systems with F but not D\n");
86e581e3
PD
522 elf_hwcap &= ~COMPAT_HWCAP_ISA_F;
523 }
524
dc6667a4 525 if (elf_hwcap & COMPAT_HWCAP_ISA_V) {
7017858e 526 riscv_v_setup_vsize();
dc6667a4
GR
527 /*
528 * ISA string in device tree might have 'v' flag, but
529 * CONFIG_RISCV_ISA_V is disabled in kernel.
530 * Clear V flag in elf_hwcap if CONFIG_RISCV_ISA_V is disabled.
531 */
532 if (!IS_ENABLED(CONFIG_RISCV_ISA_V))
533 elf_hwcap &= ~COMPAT_HWCAP_ISA_V;
534 }
535
6bcff515 536 memset(print_str, 0, sizeof(print_str));
58004f26 537 for (i = 0, j = 0; i < NUM_ALPHA_EXTS; i++)
6bcff515
AP
538 if (riscv_isa[0] & BIT_MASK(i))
539 print_str[j++] = (char)('a' + i);
02d52fbd 540 pr_info("riscv: base ISA extensions %s\n", print_str);
6bcff515
AP
541
542 memset(print_str, 0, sizeof(print_str));
58004f26 543 for (i = 0, j = 0; i < NUM_ALPHA_EXTS; i++)
6bcff515
AP
544 if (elf_hwcap & BIT_MASK(i))
545 print_str[j++] = (char)('a' + i);
546 pr_info("riscv: ELF capabilities %s\n", print_str);
e2c0cdfb 547}
ff689fd2 548
50724efc
AC
549unsigned long riscv_get_elf_hwcap(void)
550{
1fd96a3e
AC
551 unsigned long hwcap;
552
553 hwcap = (elf_hwcap & ((1UL << RISCV_ISA_EXT_BASE) - 1));
554
555 if (!riscv_v_vstate_ctrl_user_allowed())
556 hwcap &= ~COMPAT_HWCAP_ISA_V;
557
558 return hwcap;
50724efc
AC
559}
560
584ea656
EG
561void check_unaligned_access(int cpu)
562{
563 u64 start_cycles, end_cycles;
564 u64 word_cycles;
565 u64 byte_cycles;
566 int ratio;
567 unsigned long start_jiffies, now;
568 struct page *page;
569 void *dst;
570 void *src;
571 long speed = RISCV_HWPROBE_MISALIGNED_SLOW;
572
573 page = alloc_pages(GFP_NOWAIT, get_order(MISALIGNED_BUFFER_SIZE));
574 if (!page) {
575 pr_warn("Can't alloc pages to measure memcpy performance");
576 return;
577 }
578
579 /* Make an unaligned destination buffer. */
580 dst = (void *)((unsigned long)page_address(page) | 0x1);
581 /* Unalign src as well, but differently (off by 1 + 2 = 3). */
582 src = dst + (MISALIGNED_BUFFER_SIZE / 2);
583 src += 2;
584 word_cycles = -1ULL;
585 /* Do a warmup. */
586 __riscv_copy_words_unaligned(dst, src, MISALIGNED_COPY_SIZE);
587 preempt_disable();
588 start_jiffies = jiffies;
589 while ((now = jiffies) == start_jiffies)
590 cpu_relax();
591
592 /*
593 * For a fixed amount of time, repeatedly try the function, and take
594 * the best time in cycles as the measurement.
595 */
596 while (time_before(jiffies, now + (1 << MISALIGNED_ACCESS_JIFFIES_LG2))) {
597 start_cycles = get_cycles64();
598 /* Ensure the CSR read can't reorder WRT to the copy. */
599 mb();
600 __riscv_copy_words_unaligned(dst, src, MISALIGNED_COPY_SIZE);
601 /* Ensure the copy ends before the end time is snapped. */
602 mb();
603 end_cycles = get_cycles64();
604 if ((end_cycles - start_cycles) < word_cycles)
605 word_cycles = end_cycles - start_cycles;
606 }
607
608 byte_cycles = -1ULL;
609 __riscv_copy_bytes_unaligned(dst, src, MISALIGNED_COPY_SIZE);
610 start_jiffies = jiffies;
611 while ((now = jiffies) == start_jiffies)
612 cpu_relax();
613
614 while (time_before(jiffies, now + (1 << MISALIGNED_ACCESS_JIFFIES_LG2))) {
615 start_cycles = get_cycles64();
616 mb();
617 __riscv_copy_bytes_unaligned(dst, src, MISALIGNED_COPY_SIZE);
618 mb();
619 end_cycles = get_cycles64();
620 if ((end_cycles - start_cycles) < byte_cycles)
621 byte_cycles = end_cycles - start_cycles;
622 }
623
624 preempt_enable();
625
626 /* Don't divide by zero. */
627 if (!word_cycles || !byte_cycles) {
628 pr_warn("cpu%d: rdtime lacks granularity needed to measure unaligned access speed\n",
629 cpu);
630
631 goto out;
632 }
633
634 if (word_cycles < byte_cycles)
635 speed = RISCV_HWPROBE_MISALIGNED_FAST;
636
637 ratio = div_u64((byte_cycles * 100), word_cycles);
638 pr_info("cpu%d: Ratio of byte access time to unaligned word access is %d.%02d, unaligned accesses are %s\n",
639 cpu,
640 ratio / 100,
641 ratio % 100,
642 (speed == RISCV_HWPROBE_MISALIGNED_FAST) ? "fast" : "slow");
643
644 per_cpu(misaligned_access_speed, cpu) = speed;
645
646out:
647 __free_pages(page, get_order(MISALIGNED_BUFFER_SIZE));
648}
649
650static int check_unaligned_access_boot_cpu(void)
651{
652 check_unaligned_access(0);
653 return 0;
654}
655
656arch_initcall(check_unaligned_access_boot_cpu);
657
ff689fd2 658#ifdef CONFIG_RISCV_ALTERNATIVE
d25f2563
AJ
659/*
660 * Alternative patch sites consider 48 bits when determining when to patch
661 * the old instruction sequence with the new. These bits are broken into a
662 * 16-bit vendor ID and a 32-bit patch ID. A non-zero vendor ID means the
663 * patch site is for an erratum, identified by the 32-bit patch ID. When
664 * the vendor ID is zero, the patch site is for a cpufeature. cpufeatures
665 * further break down patch ID into two 16-bit numbers. The lower 16 bits
666 * are the cpufeature ID and the upper 16 bits are used for a value specific
667 * to the cpufeature and patch site. If the upper 16 bits are zero, then it
668 * implies no specific value is specified. cpufeatures that want to control
669 * patching on a per-site basis will provide non-zero values and implement
670 * checks here. The checks return true when patching should be done, and
671 * false otherwise.
672 */
673static bool riscv_cpufeature_patch_check(u16 id, u16 value)
674{
675 if (!value)
676 return true;
677
ab0f7746
AJ
678 switch (id) {
679 case RISCV_ISA_EXT_ZICBOZ:
680 /*
681 * Zicboz alternative applications provide the maximum
682 * supported block size order, or zero when it doesn't
683 * matter. If the current block size exceeds the maximum,
684 * then the alternative cannot be applied.
685 */
686 return riscv_cboz_block_size <= (1U << value);
687 }
688
d25f2563
AJ
689 return false;
690}
691
ff689fd2
HS
692void __init_or_module riscv_cpufeature_patch_func(struct alt_entry *begin,
693 struct alt_entry *end,
694 unsigned int stage)
695{
ff689fd2 696 struct alt_entry *alt;
8d23e94a 697 void *oldptr, *altptr;
d25f2563 698 u16 id, value;
ff689fd2 699
191b27c7
JZ
700 if (stage == RISCV_ALTERNATIVES_EARLY_BOOT)
701 return;
ff689fd2
HS
702
703 for (alt = begin; alt < end; alt++) {
704 if (alt->vendor_id != 0)
705 continue;
d25f2563
AJ
706
707 id = PATCH_ID_CPUFEATURE_ID(alt->patch_id);
708
709 if (id >= RISCV_ISA_EXT_MAX) {
710 WARN(1, "This extension id:%d is not in ISA extension list", id);
ff689fd2
HS
711 continue;
712 }
713
d25f2563
AJ
714 if (!__riscv_isa_extension_available(NULL, id))
715 continue;
716
717 value = PATCH_ID_CPUFEATURE_VALUE(alt->patch_id);
718 if (!riscv_cpufeature_patch_check(id, value))
4bf88607
JZ
719 continue;
720
8d23e94a
JZ
721 oldptr = ALT_OLD_PTR(alt);
722 altptr = ALT_ALT_PTR(alt);
9493e6f3
CD
723
724 mutex_lock(&text_mutex);
8d23e94a
JZ
725 patch_text_nosync(oldptr, altptr, alt->alt_len);
726 riscv_alternative_fix_offsets(oldptr, alt->alt_len, oldptr - altptr);
9493e6f3 727 mutex_unlock(&text_mutex);
ff689fd2
HS
728 }
729}
730#endif