xref: /freebsd/sys/arm/include/cpu.h (revision f126890ac5386406dadf7c4cfa9566cbb56537c5)
1 /*-
2  * Copyright 2014 Svatopluk Kraus <onwahe@gmail.com>
3  * Copyright 2014 Michal Meloun <meloun@miracle.cz>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 /* $NetBSD: cpu.h,v 1.2 2001/02/23 21:23:52 reinoud Exp $ */
28 
29 #ifndef MACHINE_CPU_H
30 #define MACHINE_CPU_H
31 
32 #include <machine/armreg.h>
33 #include <machine/frame.h>
34 
35 void	cpu_halt(void);
36 
37 #ifdef _KERNEL
38 #include <machine/atomic.h>
39 #include <machine/cpufunc.h>
40 #include <machine/cpuinfo.h>
41 #include <machine/sysreg.h>
42 
43 /*
44  * Some kernel modules (dtrace all for example) are compiled
45  * unconditionally with -DSMP. Although it looks like a bug,
46  * handle this case here and in #elif condition in ARM_SMP_UP macro.
47  */
48 #if __ARM_ARCH <= 6 && defined(SMP) && !defined(KLD_MODULE)
49 #error SMP option is not supported on ARMv6
50 #endif
51 
52 #if __ARM_ARCH <= 6 && defined(SMP_ON_UP)
53 #error SMP_ON_UP option is only supported on ARMv7+ CPUs
54 #endif
55 
56 #if !defined(SMP) && defined(SMP_ON_UP)
57 #error SMP option must be defined for SMP_ON_UP option
58 #endif
59 
60 #define CPU_ASID_KERNEL 0
61 
62 #if defined(SMP_ON_UP)
63 #define ARM_SMP_UP(smp_code, up_code)				\
64 do {								\
65 	if (cpuinfo.mp_ext != 0) {				\
66 		smp_code;					\
67 	} else {						\
68 		up_code;					\
69 	}							\
70 } while (0)
71 #elif defined(SMP) && __ARM_ARCH > 6
72 #define ARM_SMP_UP(smp_code, up_code)				\
73 do {								\
74 	smp_code;						\
75 } while (0)
76 #else
77 #define ARM_SMP_UP(smp_code, up_code)				\
78 do {								\
79 	up_code;						\
80 } while (0)
81 #endif
82 
83 void dcache_wbinv_poc_all(void); /* !!! NOT SMP coherent function !!! */
84 vm_offset_t dcache_wb_pou_checked(vm_offset_t, vm_size_t);
85 vm_offset_t icache_inv_pou_checked(vm_offset_t, vm_size_t);
86 
87 #ifdef DEV_PMU
88 #include <sys/pcpu.h>
89 #define	PMU_OVSR_C		0x80000000	/* Cycle Counter */
90 extern uint32_t	ccnt_hi[MAXCPU];
91 extern int pmu_attched;
92 #endif /* DEV_PMU */
93 
94 #define sev()  __asm __volatile("sev" : : : "memory")
95 #define wfe()  __asm __volatile("wfe" : : : "memory")
96 
97 /*
98  * Macros to generate CP15 (system control processor) read/write functions.
99  */
100 #define _FX(s...) #s
101 
102 #define _RF0(fname, aname...)						\
103 static __inline uint32_t						\
104 fname(void)								\
105 {									\
106 	uint32_t reg;							\
107 	__asm __volatile("mrc\t" _FX(aname): "=r" (reg));		\
108 	return(reg);							\
109 }
110 
111 #define _R64F0(fname, aname)						\
112 static __inline uint64_t						\
113 fname(void)								\
114 {									\
115 	uint64_t reg;							\
116 	__asm __volatile("mrrc\t" _FX(aname): "=r" (reg));		\
117 	return(reg);							\
118 }
119 
120 #define _WF0(fname, aname...)						\
121 static __inline void							\
122 fname(void)								\
123 {									\
124 	__asm __volatile("mcr\t" _FX(aname));				\
125 }
126 
127 #define _WF1(fname, aname...)						\
128 static __inline void							\
129 fname(uint32_t reg)							\
130 {									\
131 	__asm __volatile("mcr\t" _FX(aname):: "r" (reg));		\
132 }
133 
134 #define _W64F1(fname, aname...)						\
135 static __inline void							\
136 fname(uint64_t reg)							\
137 {									\
138 	__asm __volatile("mcrr\t" _FX(aname):: "r" (reg));		\
139 }
140 
141 /*
142  * Raw CP15  maintenance operations
143  * !!! not for external use !!!
144  */
145 
146 /* TLB */
147 
148 _WF0(_CP15_TLBIALL, CP15_TLBIALL)		/* Invalidate entire unified TLB */
149 #if __ARM_ARCH >= 7 && defined(SMP)
150 _WF0(_CP15_TLBIALLIS, CP15_TLBIALLIS)		/* Invalidate entire unified TLB IS */
151 #endif
152 _WF1(_CP15_TLBIASID, CP15_TLBIASID(%0))		/* Invalidate unified TLB by ASID */
153 #if __ARM_ARCH >= 7 && defined(SMP)
154 _WF1(_CP15_TLBIASIDIS, CP15_TLBIASIDIS(%0))	/* Invalidate unified TLB by ASID IS */
155 #endif
156 _WF1(_CP15_TLBIMVAA, CP15_TLBIMVAA(%0))		/* Invalidate unified TLB by MVA, all ASID */
157 #if __ARM_ARCH >= 7 && defined(SMP)
158 _WF1(_CP15_TLBIMVAAIS, CP15_TLBIMVAAIS(%0))	/* Invalidate unified TLB by MVA, all ASID IS */
159 #endif
160 _WF1(_CP15_TLBIMVA, CP15_TLBIMVA(%0))		/* Invalidate unified TLB by MVA */
161 
162 _WF1(_CP15_TTB_SET, CP15_TTBR0(%0))
163 
164 /* Cache and Branch predictor */
165 
166 _WF0(_CP15_BPIALL, CP15_BPIALL)			/* Branch predictor invalidate all */
167 #if __ARM_ARCH >= 7 && defined(SMP)
168 _WF0(_CP15_BPIALLIS, CP15_BPIALLIS)		/* Branch predictor invalidate all IS */
169 #endif
170 _WF1(_CP15_BPIMVA, CP15_BPIMVA(%0))		/* Branch predictor invalidate by MVA */
171 _WF1(_CP15_DCCIMVAC, CP15_DCCIMVAC(%0))		/* Data cache clean and invalidate by MVA PoC */
172 _WF1(_CP15_DCCISW, CP15_DCCISW(%0))		/* Data cache clean and invalidate by set/way */
173 _WF1(_CP15_DCCMVAC, CP15_DCCMVAC(%0))		/* Data cache clean by MVA PoC */
174 #if __ARM_ARCH >= 7
175 _WF1(_CP15_DCCMVAU, CP15_DCCMVAU(%0))		/* Data cache clean by MVA PoU */
176 #endif
177 _WF1(_CP15_DCCSW, CP15_DCCSW(%0))		/* Data cache clean by set/way */
178 _WF1(_CP15_DCIMVAC, CP15_DCIMVAC(%0))		/* Data cache invalidate by MVA PoC */
179 _WF1(_CP15_DCISW, CP15_DCISW(%0))		/* Data cache invalidate by set/way */
180 _WF0(_CP15_ICIALLU, CP15_ICIALLU)		/* Instruction cache invalidate all PoU */
181 #if __ARM_ARCH >= 7 && defined(SMP)
182 _WF0(_CP15_ICIALLUIS, CP15_ICIALLUIS)		/* Instruction cache invalidate all PoU IS */
183 #endif
184 _WF1(_CP15_ICIMVAU, CP15_ICIMVAU(%0))		/* Instruction cache invalidate */
185 
186 /*
187  * Publicly accessible functions
188  */
189 
190 /* CP14 Debug Registers */
191 _RF0(cp14_dbgdidr_get, CP14_DBGDIDR(%0))
192 _RF0(cp14_dbgprsr_get, CP14_DBGPRSR(%0))
193 _RF0(cp14_dbgoslsr_get, CP14_DBGOSLSR(%0))
194 _RF0(cp14_dbgosdlr_get, CP14_DBGOSDLR(%0))
195 _RF0(cp14_dbgdscrint_get, CP14_DBGDSCRint(%0))
196 
197 _WF1(cp14_dbgdscr_v6_set, CP14_DBGDSCRext_V6(%0))
198 _WF1(cp14_dbgdscr_v7_set, CP14_DBGDSCRext_V7(%0))
199 _WF1(cp14_dbgvcr_set, CP14_DBGVCR(%0))
200 _WF1(cp14_dbgoslar_set, CP14_DBGOSLAR(%0))
201 
202 /* Various control registers */
203 
204 _RF0(cp15_cpacr_get, CP15_CPACR(%0))
205 _WF1(cp15_cpacr_set, CP15_CPACR(%0))
206 _RF0(cp15_dfsr_get, CP15_DFSR(%0))
207 _RF0(cp15_ifsr_get, CP15_IFSR(%0))
208 _WF1(cp15_prrr_set, CP15_PRRR(%0))
209 _WF1(cp15_nmrr_set, CP15_NMRR(%0))
210 _RF0(cp15_ttbr_get, CP15_TTBR0(%0))
211 _RF0(cp15_dfar_get, CP15_DFAR(%0))
212 #if __ARM_ARCH >= 7
213 _RF0(cp15_ifar_get, CP15_IFAR(%0))
214 _RF0(cp15_l2ctlr_get, CP15_L2CTLR(%0))
215 #endif
216 _RF0(cp15_actlr_get, CP15_ACTLR(%0))
217 _WF1(cp15_actlr_set, CP15_ACTLR(%0))
218 _WF1(cp15_ats1cpr_set, CP15_ATS1CPR(%0))
219 _WF1(cp15_ats1cpw_set, CP15_ATS1CPW(%0))
220 _WF1(cp15_ats1cur_set, CP15_ATS1CUR(%0))
221 _WF1(cp15_ats1cuw_set, CP15_ATS1CUW(%0))
222 _RF0(cp15_par_get, CP15_PAR(%0))
223 _RF0(cp15_sctlr_get, CP15_SCTLR(%0))
224 
225 /*CPU id registers */
226 _RF0(cp15_midr_get, CP15_MIDR(%0))
227 _RF0(cp15_ctr_get, CP15_CTR(%0))
228 _RF0(cp15_tcmtr_get, CP15_TCMTR(%0))
229 _RF0(cp15_tlbtr_get, CP15_TLBTR(%0))
230 _RF0(cp15_mpidr_get, CP15_MPIDR(%0))
231 _RF0(cp15_revidr_get, CP15_REVIDR(%0))
232 _RF0(cp15_ccsidr_get, CP15_CCSIDR(%0))
233 _RF0(cp15_clidr_get, CP15_CLIDR(%0))
234 _RF0(cp15_aidr_get, CP15_AIDR(%0))
235 _WF1(cp15_csselr_set, CP15_CSSELR(%0))
236 _RF0(cp15_id_pfr0_get, CP15_ID_PFR0(%0))
237 _RF0(cp15_id_pfr1_get, CP15_ID_PFR1(%0))
238 _RF0(cp15_id_dfr0_get, CP15_ID_DFR0(%0))
239 _RF0(cp15_id_afr0_get, CP15_ID_AFR0(%0))
240 _RF0(cp15_id_mmfr0_get, CP15_ID_MMFR0(%0))
241 _RF0(cp15_id_mmfr1_get, CP15_ID_MMFR1(%0))
242 _RF0(cp15_id_mmfr2_get, CP15_ID_MMFR2(%0))
243 _RF0(cp15_id_mmfr3_get, CP15_ID_MMFR3(%0))
244 _RF0(cp15_id_isar0_get, CP15_ID_ISAR0(%0))
245 _RF0(cp15_id_isar1_get, CP15_ID_ISAR1(%0))
246 _RF0(cp15_id_isar2_get, CP15_ID_ISAR2(%0))
247 _RF0(cp15_id_isar3_get, CP15_ID_ISAR3(%0))
248 _RF0(cp15_id_isar4_get, CP15_ID_ISAR4(%0))
249 _RF0(cp15_id_isar5_get, CP15_ID_ISAR5(%0))
250 _RF0(cp15_cbar_get, CP15_CBAR(%0))
251 
252 /* Performance Monitor registers */
253 
254 #if __ARM_ARCH == 6 && defined(CPU_ARM1176)
255 _RF0(cp15_pmuserenr_get, CP15_PMUSERENR(%0))
256 _WF1(cp15_pmuserenr_set, CP15_PMUSERENR(%0))
257 _RF0(cp15_pmcr_get, CP15_PMCR(%0))
258 _WF1(cp15_pmcr_set, CP15_PMCR(%0))
259 _RF0(cp15_pmccntr_get, CP15_PMCCNTR(%0))
260 _WF1(cp15_pmccntr_set, CP15_PMCCNTR(%0))
261 #elif __ARM_ARCH > 6
262 _RF0(cp15_pmcr_get, CP15_PMCR(%0))
263 _WF1(cp15_pmcr_set, CP15_PMCR(%0))
264 _RF0(cp15_pmcnten_get, CP15_PMCNTENSET(%0))
265 _WF1(cp15_pmcnten_set, CP15_PMCNTENSET(%0))
266 _WF1(cp15_pmcnten_clr, CP15_PMCNTENCLR(%0))
267 _RF0(cp15_pmovsr_get, CP15_PMOVSR(%0))
268 _WF1(cp15_pmovsr_set, CP15_PMOVSR(%0))
269 _WF1(cp15_pmswinc_set, CP15_PMSWINC(%0))
270 _RF0(cp15_pmselr_get, CP15_PMSELR(%0))
271 _WF1(cp15_pmselr_set, CP15_PMSELR(%0))
272 _RF0(cp15_pmccntr_get, CP15_PMCCNTR(%0))
273 _WF1(cp15_pmccntr_set, CP15_PMCCNTR(%0))
274 _RF0(cp15_pmxevtyper_get, CP15_PMXEVTYPER(%0))
275 _WF1(cp15_pmxevtyper_set, CP15_PMXEVTYPER(%0))
276 _RF0(cp15_pmxevcntr_get, CP15_PMXEVCNTRR(%0))
277 _WF1(cp15_pmxevcntr_set, CP15_PMXEVCNTRR(%0))
278 _RF0(cp15_pmuserenr_get, CP15_PMUSERENR(%0))
279 _WF1(cp15_pmuserenr_set, CP15_PMUSERENR(%0))
280 _RF0(cp15_pminten_get, CP15_PMINTENSET(%0))
281 _WF1(cp15_pminten_set, CP15_PMINTENSET(%0))
282 _WF1(cp15_pminten_clr, CP15_PMINTENCLR(%0))
283 #endif
284 
285 _RF0(cp15_tpidrurw_get, CP15_TPIDRURW(%0))
286 _WF1(cp15_tpidrurw_set, CP15_TPIDRURW(%0))
287 _RF0(cp15_tpidruro_get, CP15_TPIDRURO(%0))
288 _WF1(cp15_tpidruro_set, CP15_TPIDRURO(%0))
289 _RF0(cp15_tpidrpwr_get, CP15_TPIDRPRW(%0))
290 _WF1(cp15_tpidrpwr_set, CP15_TPIDRPRW(%0))
291 
292 /* Generic Timer registers - only use when you know the hardware is available */
293 _RF0(cp15_cntfrq_get, CP15_CNTFRQ(%0))
294 _WF1(cp15_cntfrq_set, CP15_CNTFRQ(%0))
295 _RF0(cp15_cntkctl_get, CP15_CNTKCTL(%0))
296 _WF1(cp15_cntkctl_set, CP15_CNTKCTL(%0))
297 _RF0(cp15_cntp_tval_get, CP15_CNTP_TVAL(%0))
298 _WF1(cp15_cntp_tval_set, CP15_CNTP_TVAL(%0))
299 _RF0(cp15_cntp_ctl_get, CP15_CNTP_CTL(%0))
300 _WF1(cp15_cntp_ctl_set, CP15_CNTP_CTL(%0))
301 _RF0(cp15_cntv_tval_get, CP15_CNTV_TVAL(%0))
302 _WF1(cp15_cntv_tval_set, CP15_CNTV_TVAL(%0))
303 _RF0(cp15_cntv_ctl_get, CP15_CNTV_CTL(%0))
304 _WF1(cp15_cntv_ctl_set, CP15_CNTV_CTL(%0))
305 _RF0(cp15_cnthctl_get, CP15_CNTHCTL(%0))
306 _WF1(cp15_cnthctl_set, CP15_CNTHCTL(%0))
307 _RF0(cp15_cnthp_tval_get, CP15_CNTHP_TVAL(%0))
308 _WF1(cp15_cnthp_tval_set, CP15_CNTHP_TVAL(%0))
309 _RF0(cp15_cnthp_ctl_get, CP15_CNTHP_CTL(%0))
310 _WF1(cp15_cnthp_ctl_set, CP15_CNTHP_CTL(%0))
311 
312 _R64F0(cp15_cntpct_get, CP15_CNTPCT(%Q0, %R0))
313 _R64F0(cp15_cntvct_get, CP15_CNTVCT(%Q0, %R0))
314 _R64F0(cp15_cntp_cval_get, CP15_CNTP_CVAL(%Q0, %R0))
315 _W64F1(cp15_cntp_cval_set, CP15_CNTP_CVAL(%Q0, %R0))
316 _R64F0(cp15_cntv_cval_get, CP15_CNTV_CVAL(%Q0, %R0))
317 _W64F1(cp15_cntv_cval_set, CP15_CNTV_CVAL(%Q0, %R0))
318 _R64F0(cp15_cntvoff_get, CP15_CNTVOFF(%Q0, %R0))
319 _W64F1(cp15_cntvoff_set, CP15_CNTVOFF(%Q0, %R0))
320 _R64F0(cp15_cnthp_cval_get, CP15_CNTHP_CVAL(%Q0, %R0))
321 _W64F1(cp15_cnthp_cval_set, CP15_CNTHP_CVAL(%Q0, %R0))
322 
323 #undef	_FX
324 #undef	_RF0
325 #undef	_WF0
326 #undef	_WF1
327 
328 /*
329  * TLB maintenance operations.
330  */
331 
332 /* Local (i.e. not broadcasting ) operations.  */
333 
334 /* Flush all TLB entries (even global). */
335 static __inline void
336 tlb_flush_all_local(void)
337 {
338 
339 	dsb();
340 	_CP15_TLBIALL();
341 	dsb();
342 }
343 
344 /* Flush all not global TLB entries. */
345 static __inline void
346 tlb_flush_all_ng_local(void)
347 {
348 
349 	dsb();
350 	_CP15_TLBIASID(CPU_ASID_KERNEL);
351 	dsb();
352 }
353 
354 /* Flush single TLB entry (even global). */
355 static __inline void
356 tlb_flush_local(vm_offset_t va)
357 {
358 
359 	KASSERT((va & PAGE_MASK) == 0, ("%s: va %#x not aligned", __func__, va));
360 
361 	dsb();
362 	_CP15_TLBIMVA(va | CPU_ASID_KERNEL);
363 	dsb();
364 }
365 
366 /* Flush range of TLB entries (even global). */
367 static __inline void
368 tlb_flush_range_local(vm_offset_t va, vm_size_t size)
369 {
370 	vm_offset_t eva = va + size;
371 
372 	KASSERT((va & PAGE_MASK) == 0, ("%s: va %#x not aligned", __func__, va));
373 	KASSERT((size & PAGE_MASK) == 0, ("%s: size %#x not aligned", __func__,
374 	    size));
375 
376 	dsb();
377 	for (; va < eva; va += PAGE_SIZE)
378 		_CP15_TLBIMVA(va | CPU_ASID_KERNEL);
379 	dsb();
380 }
381 
382 /* Broadcasting operations. */
383 #if __ARM_ARCH >= 7 && defined(SMP)
384 
385 static __inline void
386 tlb_flush_all(void)
387 {
388 
389 	dsb();
390 	ARM_SMP_UP(
391 	    _CP15_TLBIALLIS(),
392 	    _CP15_TLBIALL()
393 	);
394 	dsb();
395 }
396 
397 static __inline void
398 tlb_flush_all_ng(void)
399 {
400 
401 	dsb();
402 	ARM_SMP_UP(
403 	    _CP15_TLBIASIDIS(CPU_ASID_KERNEL),
404 	    _CP15_TLBIASID(CPU_ASID_KERNEL)
405 	);
406 	dsb();
407 }
408 
409 static __inline void
410 tlb_flush(vm_offset_t va)
411 {
412 
413 	KASSERT((va & PAGE_MASK) == 0, ("%s: va %#x not aligned", __func__, va));
414 
415 	dsb();
416 	ARM_SMP_UP(
417 	    _CP15_TLBIMVAAIS(va),
418 	    _CP15_TLBIMVA(va | CPU_ASID_KERNEL)
419 	);
420 	dsb();
421 }
422 
423 static __inline void
424 tlb_flush_range(vm_offset_t va,  vm_size_t size)
425 {
426 	vm_offset_t eva = va + size;
427 
428 	KASSERT((va & PAGE_MASK) == 0, ("%s: va %#x not aligned", __func__, va));
429 	KASSERT((size & PAGE_MASK) == 0, ("%s: size %#x not aligned", __func__,
430 	    size));
431 
432 	dsb();
433 	ARM_SMP_UP(
434 		{
435 			for (; va < eva; va += PAGE_SIZE)
436 				_CP15_TLBIMVAAIS(va);
437 		},
438 		{
439 			for (; va < eva; va += PAGE_SIZE)
440 				_CP15_TLBIMVA(va | CPU_ASID_KERNEL);
441 		}
442 	);
443 	dsb();
444 }
445 #else /* __ARM_ARCH < 7 */
446 
447 #define tlb_flush_all() 		tlb_flush_all_local()
448 #define tlb_flush_all_ng() 		tlb_flush_all_ng_local()
449 #define tlb_flush(va) 			tlb_flush_local(va)
450 #define tlb_flush_range(va, size) 	tlb_flush_range_local(va, size)
451 
452 #endif /* __ARM_ARCH < 7 */
453 
454 /*
455  * Cache maintenance operations.
456  */
457 
458 /*  Sync I and D caches to PoU */
459 static __inline void
460 icache_sync(vm_offset_t va, vm_size_t size)
461 {
462 	vm_offset_t eva = va + size;
463 
464 	dsb();
465 	va &= ~cpuinfo.dcache_line_mask;
466 
467 	for ( ; va < eva; va += cpuinfo.dcache_line_size) {
468 #if __ARM_ARCH >= 7
469 		_CP15_DCCMVAU(va);
470 #else
471 		_CP15_DCCMVAC(va);
472 #endif
473 	}
474 	dsb();
475 	ARM_SMP_UP(
476 			_CP15_ICIALLUIS(),
477 			_CP15_ICIALLU()
478 	);
479 	dsb();
480 	isb();
481 }
482 
483 /*  Invalidate I cache */
484 static __inline void
485 icache_inv_all(void)
486 {
487 
488 	ARM_SMP_UP(
489 		_CP15_ICIALLUIS(),
490 		_CP15_ICIALLU()
491 	);
492 	dsb();
493 	isb();
494 }
495 
496 /* Invalidate branch predictor buffer */
497 static __inline void
498 bpb_inv_all(void)
499 {
500 
501 	ARM_SMP_UP(
502 		_CP15_BPIALLIS(),
503 		_CP15_BPIALL()
504 	);
505 	dsb();
506 	isb();
507 }
508 
509 /* Write back D-cache to PoU */
510 static __inline void
511 dcache_wb_pou(vm_offset_t va, vm_size_t size)
512 {
513 	vm_offset_t eva = va + size;
514 
515 	dsb();
516 	va &= ~cpuinfo.dcache_line_mask;
517 	for ( ; va < eva; va += cpuinfo.dcache_line_size) {
518 #if __ARM_ARCH >= 7
519 		_CP15_DCCMVAU(va);
520 #else
521 		_CP15_DCCMVAC(va);
522 #endif
523 	}
524 	dsb();
525 }
526 
527 /*
528  * Invalidate D-cache to PoC
529  *
530  * Caches are invalidated from outermost to innermost as fresh cachelines
531  * flow in this direction. In given range, if there was no dirty cacheline
532  * in any cache before, no stale cacheline should remain in them after this
533  * operation finishes.
534  */
535 static __inline void
536 dcache_inv_poc(vm_offset_t va, vm_paddr_t pa, vm_size_t size)
537 {
538 	vm_offset_t eva = va + size;
539 
540 	dsb();
541 	/* invalidate L2 first */
542 	cpu_l2cache_inv_range(pa, size);
543 
544 	/* then L1 */
545 	va &= ~cpuinfo.dcache_line_mask;
546 	for ( ; va < eva; va += cpuinfo.dcache_line_size) {
547 		_CP15_DCIMVAC(va);
548 	}
549 	dsb();
550 }
551 
552 /*
553  * Discard D-cache lines to PoC, prior to overwrite by DMA engine.
554  *
555  * Normal invalidation does L2 then L1 to ensure that stale data from L2 doesn't
556  * flow into L1 while invalidating.  This routine is intended to be used only
557  * when invalidating a buffer before a DMA operation loads new data into memory.
558  * The concern in this case is that dirty lines are not evicted to main memory,
559  * overwriting the DMA data.  For that reason, the L1 is done first to ensure
560  * that an evicted L1 line doesn't flow to L2 after the L2 has been cleaned.
561  */
562 static __inline void
563 dcache_inv_poc_dma(vm_offset_t va, vm_paddr_t pa, vm_size_t size)
564 {
565 	vm_offset_t eva = va + size;
566 
567 	/* invalidate L1 first */
568 	dsb();
569 	va &= ~cpuinfo.dcache_line_mask;
570 	for ( ; va < eva; va += cpuinfo.dcache_line_size) {
571 		_CP15_DCIMVAC(va);
572 	}
573 	dsb();
574 
575 	/* then L2 */
576 	cpu_l2cache_inv_range(pa, size);
577 }
578 
579 /*
580  * Write back D-cache to PoC
581  *
582  * Caches are written back from innermost to outermost as dirty cachelines
583  * flow in this direction. In given range, no dirty cacheline should remain
584  * in any cache after this operation finishes.
585  */
586 static __inline void
587 dcache_wb_poc(vm_offset_t va, vm_paddr_t pa, vm_size_t size)
588 {
589 	vm_offset_t eva = va + size;
590 
591 	dsb();
592 	va &= ~cpuinfo.dcache_line_mask;
593 	for ( ; va < eva; va += cpuinfo.dcache_line_size) {
594 		_CP15_DCCMVAC(va);
595 	}
596 	dsb();
597 
598 	cpu_l2cache_wb_range(pa, size);
599 }
600 
601 /* Write back and invalidate D-cache to PoC */
602 static __inline void
603 dcache_wbinv_poc(vm_offset_t sva, vm_paddr_t pa, vm_size_t size)
604 {
605 	vm_offset_t va;
606 	vm_offset_t eva = sva + size;
607 
608 	dsb();
609 	/* write back L1 first */
610 	va = sva & ~cpuinfo.dcache_line_mask;
611 	for ( ; va < eva; va += cpuinfo.dcache_line_size) {
612 		_CP15_DCCMVAC(va);
613 	}
614 	dsb();
615 
616 	/* then write back and invalidate L2 */
617 	cpu_l2cache_wbinv_range(pa, size);
618 
619 	/* then invalidate L1 */
620 	va = sva & ~cpuinfo.dcache_line_mask;
621 	for ( ; va < eva; va += cpuinfo.dcache_line_size) {
622 		_CP15_DCIMVAC(va);
623 	}
624 	dsb();
625 }
626 
627 /* Set TTB0 register */
628 static __inline void
629 cp15_ttbr_set(uint32_t reg)
630 {
631 	dsb();
632 	_CP15_TTB_SET(reg);
633 	dsb();
634 	_CP15_BPIALL();
635 	dsb();
636 	isb();
637 	tlb_flush_all_ng_local();
638 }
639 
640 /*
641  * Functions for address checking:
642  *
643  *  cp15_ats1cpr_check() ... check stage 1 privileged (PL1) read access
644  *  cp15_ats1cpw_check() ... check stage 1 privileged (PL1) write access
645  *  cp15_ats1cur_check() ... check stage 1 unprivileged (PL0) read access
646  *  cp15_ats1cuw_check() ... check stage 1 unprivileged (PL0) write access
647  *
648  * They must be called while interrupts are disabled to get consistent result.
649  */
650 static __inline int
651 cp15_ats1cpr_check(vm_offset_t addr)
652 {
653 
654 	cp15_ats1cpr_set(addr);
655 	isb();
656 	return (cp15_par_get() & 0x01 ? EFAULT : 0);
657 }
658 
659 static __inline int
660 cp15_ats1cpw_check(vm_offset_t addr)
661 {
662 
663 	cp15_ats1cpw_set(addr);
664 	isb();
665 	return (cp15_par_get() & 0x01 ? EFAULT : 0);
666 }
667 
668 static __inline int
669 cp15_ats1cur_check(vm_offset_t addr)
670 {
671 
672 	cp15_ats1cur_set(addr);
673 	isb();
674 	return (cp15_par_get() & 0x01 ? EFAULT : 0);
675 }
676 
677 static __inline int
678 cp15_ats1cuw_check(vm_offset_t addr)
679 {
680 
681 	cp15_ats1cuw_set(addr);
682 	isb();
683 	return (cp15_par_get() & 0x01 ? EFAULT : 0);
684 }
685 
686 static __inline uint64_t
687 get_cyclecount(void)
688 {
689 #if __ARM_ARCH > 6 || (__ARM_ARCH == 6 && defined(CPU_ARM1176))
690 #if (__ARM_ARCH > 6) && defined(DEV_PMU)
691 	if (pmu_attched) {
692 		u_int cpu;
693 		uint64_t h, h2;
694 		uint32_t l, r;
695 
696 		cpu = PCPU_GET(cpuid);
697 		h = (uint64_t)atomic_load_acq_32(&ccnt_hi[cpu]);
698 		l = cp15_pmccntr_get();
699 		/* In case interrupts are disabled we need to check for overflow. */
700 		r = cp15_pmovsr_get();
701 		if (r & PMU_OVSR_C) {
702 			atomic_add_32(&ccnt_hi[cpu], 1);
703 			/* Clear the event. */
704 			cp15_pmovsr_set(PMU_OVSR_C);
705 		}
706 		/* Make sure there was no wrap-around while we read the lo half. */
707 		h2 = (uint64_t)atomic_load_acq_32(&ccnt_hi[cpu]);
708 		if (h != h2)
709 			l = cp15_pmccntr_get();
710 		return (h2 << 32 | l);
711 	} else
712 #endif
713 		return cp15_pmccntr_get();
714 #else /* No performance counters, so use nanotime(9). */
715 	struct timespec tv;
716 
717 	nanotime(&tv);
718 	return (tv.tv_sec * (uint64_t)1000000000ull + tv.tv_nsec);
719 #endif
720 }
721 #endif
722 
723 #define TRAPF_USERMODE(frame)	((frame->tf_spsr & PSR_MODE) == PSR_USR32_MODE)
724 
725 #define TRAPF_PC(tfp)		((tfp)->tf_pc)
726 
727 #define cpu_getstack(td)	((td)->td_frame->tf_usr_sp)
728 #define cpu_setstack(td, sp)	((td)->td_frame->tf_usr_sp = (sp))
729 #define cpu_spinwait()		/* nothing */
730 #define	cpu_lock_delay()	DELAY(1)
731 
732 #define ARM_NVEC		8
733 #define ARM_VEC_ALL		0xffffffff
734 
735 extern vm_offset_t vector_page;
736 
737 /*
738  * Params passed into initarm. If you change the size of this you will
739  * need to update locore.S to allocate more memory on the stack before
740  * it calls initarm.
741  */
742 struct arm_boot_params {
743 	register_t	abp_size;	/* Size of this structure */
744 	register_t	abp_r0;		/* r0 from the boot loader */
745 	register_t	abp_r1;		/* r1 from the boot loader */
746 	register_t	abp_r2;		/* r2 from the boot loader */
747 	register_t	abp_r3;		/* r3 from the boot loader */
748 	vm_offset_t	abp_physaddr;	/* The kernel physical address */
749 	vm_offset_t	abp_pagetable;	/* The early page table */
750 };
751 
752 void	arm_vector_init(vm_offset_t, int);
753 void	fork_trampoline(void);
754 void	identify_arm_cpu(void);
755 void	*initarm(struct arm_boot_params *);
756 
757 extern char btext[];
758 extern char etext[];
759 int badaddr_read(void *, size_t, void *);
760 #endif /* !MACHINE_CPU_H */
761