xref: /openbmc/linux/arch/arm/mm/pmsa-v7.c (revision d2ba09c1)
1 /*
2  * Based on linux/arch/arm/mm/nommu.c
3  *
4  * ARM PMSAv7 supporting functions.
5  */
6 
7 #include <linux/bitops.h>
8 #include <linux/memblock.h>
9 #include <linux/string.h>
10 
11 #include <asm/cacheflush.h>
12 #include <asm/cp15.h>
13 #include <asm/cputype.h>
14 #include <asm/mpu.h>
15 #include <asm/sections.h>
16 
17 #include "mm.h"
18 
19 struct region {
20 	phys_addr_t base;
21 	phys_addr_t size;
22 	unsigned long subreg;
23 };
24 
25 static struct region __initdata mem[MPU_MAX_REGIONS];
26 #ifdef CONFIG_XIP_KERNEL
27 static struct region __initdata xip[MPU_MAX_REGIONS];
28 #endif
29 
30 static unsigned int __initdata mpu_min_region_order;
31 static unsigned int __initdata mpu_max_regions;
32 
33 static int __init __mpu_min_region_order(void);
34 static int __init __mpu_max_regions(void);
35 
36 #ifndef CONFIG_CPU_V7M
37 
38 #define DRBAR	__ACCESS_CP15(c6, 0, c1, 0)
39 #define IRBAR	__ACCESS_CP15(c6, 0, c1, 1)
40 #define DRSR	__ACCESS_CP15(c6, 0, c1, 2)
41 #define IRSR	__ACCESS_CP15(c6, 0, c1, 3)
42 #define DRACR	__ACCESS_CP15(c6, 0, c1, 4)
43 #define IRACR	__ACCESS_CP15(c6, 0, c1, 5)
44 #define RNGNR	__ACCESS_CP15(c6, 0, c2, 0)
45 
46 /* Region number */
47 static inline void rgnr_write(u32 v)
48 {
49 	write_sysreg(v, RNGNR);
50 }
51 
52 /* Data-side / unified region attributes */
53 
54 /* Region access control register */
55 static inline void dracr_write(u32 v)
56 {
57 	write_sysreg(v, DRACR);
58 }
59 
60 /* Region size register */
61 static inline void drsr_write(u32 v)
62 {
63 	write_sysreg(v, DRSR);
64 }
65 
66 /* Region base address register */
67 static inline void drbar_write(u32 v)
68 {
69 	write_sysreg(v, DRBAR);
70 }
71 
72 static inline u32 drbar_read(void)
73 {
74 	return read_sysreg(DRBAR);
75 }
76 /* Optional instruction-side region attributes */
77 
78 /* I-side Region access control register */
79 static inline void iracr_write(u32 v)
80 {
81 	write_sysreg(v, IRACR);
82 }
83 
84 /* I-side Region size register */
85 static inline void irsr_write(u32 v)
86 {
87 	write_sysreg(v, IRSR);
88 }
89 
90 /* I-side Region base address register */
91 static inline void irbar_write(u32 v)
92 {
93 	write_sysreg(v, IRBAR);
94 }
95 
96 static inline u32 irbar_read(void)
97 {
98 	return read_sysreg(IRBAR);
99 }
100 
101 #else
102 
103 static inline void rgnr_write(u32 v)
104 {
105 	writel_relaxed(v, BASEADDR_V7M_SCB + MPU_RNR);
106 }
107 
108 /* Data-side / unified region attributes */
109 
110 /* Region access control register */
111 static inline void dracr_write(u32 v)
112 {
113 	u32 rsr = readl_relaxed(BASEADDR_V7M_SCB + MPU_RASR) & GENMASK(15, 0);
114 
115 	writel_relaxed((v << 16) | rsr, BASEADDR_V7M_SCB + MPU_RASR);
116 }
117 
118 /* Region size register */
119 static inline void drsr_write(u32 v)
120 {
121 	u32 racr = readl_relaxed(BASEADDR_V7M_SCB + MPU_RASR) & GENMASK(31, 16);
122 
123 	writel_relaxed(v | racr, BASEADDR_V7M_SCB + MPU_RASR);
124 }
125 
126 /* Region base address register */
127 static inline void drbar_write(u32 v)
128 {
129 	writel_relaxed(v, BASEADDR_V7M_SCB + MPU_RBAR);
130 }
131 
132 static inline u32 drbar_read(void)
133 {
134 	return readl_relaxed(BASEADDR_V7M_SCB + MPU_RBAR);
135 }
136 
137 /* ARMv7-M only supports a unified MPU, so I-side operations are nop */
138 
139 static inline void iracr_write(u32 v) {}
140 static inline void irsr_write(u32 v) {}
141 static inline void irbar_write(u32 v) {}
142 static inline unsigned long irbar_read(void) {return 0;}
143 
144 #endif
145 
146 static int __init mpu_present(void)
147 {
148 	return ((read_cpuid_ext(CPUID_EXT_MMFR0) & MMFR0_PMSA) == MMFR0_PMSAv7);
149 }
150 
151 static bool __init try_split_region(phys_addr_t base, phys_addr_t size, struct region *region)
152 {
153 	unsigned long  subreg, bslots, sslots;
154 	phys_addr_t abase = base & ~(size - 1);
155 	phys_addr_t asize = base + size - abase;
156 	phys_addr_t p2size = 1 << __fls(asize);
157 	phys_addr_t bdiff, sdiff;
158 
159 	if (p2size != asize)
160 		p2size *= 2;
161 
162 	bdiff = base - abase;
163 	sdiff = p2size - asize;
164 	subreg = p2size / MPU_NR_SUBREGS;
165 
166 	if ((bdiff % subreg) || (sdiff % subreg))
167 		return false;
168 
169 	bslots = bdiff / subreg;
170 	sslots = sdiff / subreg;
171 
172 	if (bslots || sslots) {
173 		int i;
174 
175 		if (subreg < MPU_MIN_SUBREG_SIZE)
176 			return false;
177 
178 		if (bslots + sslots > MPU_NR_SUBREGS)
179 			return false;
180 
181 		for (i = 0; i < bslots; i++)
182 			_set_bit(i, &region->subreg);
183 
184 		for (i = 1; i <= sslots; i++)
185 			_set_bit(MPU_NR_SUBREGS - i, &region->subreg);
186 	}
187 
188 	region->base = abase;
189 	region->size = p2size;
190 
191 	return true;
192 }
193 
194 static int __init allocate_region(phys_addr_t base, phys_addr_t size,
195 				  unsigned int limit, struct region *regions)
196 {
197 	int count = 0;
198 	phys_addr_t diff = size;
199 	int attempts = MPU_MAX_REGIONS;
200 
201 	while (diff) {
202 		/* Try cover region as is (maybe with help of subregions) */
203 		if (try_split_region(base, size, &regions[count])) {
204 			count++;
205 			base += size;
206 			diff -= size;
207 			size = diff;
208 		} else {
209 			/*
210 			 * Maximum aligned region might overflow phys_addr_t
211 			 * if "base" is 0. Hence we keep everything below 4G
212 			 * until we take the smaller of the aligned region
213 			 * size ("asize") and rounded region size ("p2size"),
214 			 * one of which is guaranteed to be smaller than the
215 			 * maximum physical address.
216 			 */
217 			phys_addr_t asize = (base - 1) ^ base;
218 			phys_addr_t p2size = (1 <<  __fls(diff)) - 1;
219 
220 			size = asize < p2size ? asize + 1 : p2size + 1;
221 		}
222 
223 		if (count > limit)
224 			break;
225 
226 		if (!attempts)
227 			break;
228 
229 		attempts--;
230 	}
231 
232 	return count;
233 }
234 
235 /* MPU initialisation functions */
236 void __init adjust_lowmem_bounds_mpu(void)
237 {
238 	phys_addr_t  specified_mem_size = 0, total_mem_size = 0;
239 	struct memblock_region *reg;
240 	bool first = true;
241 	phys_addr_t mem_start;
242 	phys_addr_t mem_end;
243 	unsigned int mem_max_regions;
244 	int num, i;
245 
246 	if (!mpu_present())
247 		return;
248 
249 	/* Free-up MPU_PROBE_REGION */
250 	mpu_min_region_order = __mpu_min_region_order();
251 
252 	/* How many regions are supported */
253 	mpu_max_regions = __mpu_max_regions();
254 
255 	mem_max_regions = min((unsigned int)MPU_MAX_REGIONS, mpu_max_regions);
256 
257 	/* We need to keep one slot for background region */
258 	mem_max_regions--;
259 
260 #ifndef CONFIG_CPU_V7M
261 	/* ... and one for vectors */
262 	mem_max_regions--;
263 #endif
264 
265 #ifdef CONFIG_XIP_KERNEL
266 	/* plus some regions to cover XIP ROM */
267 	num = allocate_region(CONFIG_XIP_PHYS_ADDR, __pa(_exiprom) - CONFIG_XIP_PHYS_ADDR,
268 			      mem_max_regions, xip);
269 
270 	mem_max_regions -= num;
271 #endif
272 
273 	for_each_memblock(memory, reg) {
274 		if (first) {
275 			phys_addr_t phys_offset = PHYS_OFFSET;
276 
277 			/*
278 			 * Initially only use memory continuous from
279 			 * PHYS_OFFSET */
280 			if (reg->base != phys_offset)
281 				panic("First memory bank must be contiguous from PHYS_OFFSET");
282 
283 			mem_start = reg->base;
284 			mem_end = reg->base + reg->size;
285 			specified_mem_size = reg->size;
286 			first = false;
287 		} else {
288 			/*
289 			 * memblock auto merges contiguous blocks, remove
290 			 * all blocks afterwards in one go (we can't remove
291 			 * blocks separately while iterating)
292 			 */
293 			pr_notice("Ignoring RAM after %pa, memory at %pa ignored\n",
294 				  &mem_end, &reg->base);
295 			memblock_remove(reg->base, 0 - reg->base);
296 			break;
297 		}
298 	}
299 
300 	memset(mem, 0, sizeof(mem));
301 	num = allocate_region(mem_start, specified_mem_size, mem_max_regions, mem);
302 
303 	for (i = 0; i < num; i++) {
304 		unsigned long  subreg = mem[i].size / MPU_NR_SUBREGS;
305 
306 		total_mem_size += mem[i].size - subreg * hweight_long(mem[i].subreg);
307 
308 		pr_debug("MPU: base %pa size %pa disable subregions: %*pbl\n",
309 			 &mem[i].base, &mem[i].size, MPU_NR_SUBREGS, &mem[i].subreg);
310 	}
311 
312 	if (total_mem_size != specified_mem_size) {
313 		pr_warn("Truncating memory from %pa to %pa (MPU region constraints)",
314 				&specified_mem_size, &total_mem_size);
315 		memblock_remove(mem_start + total_mem_size,
316 				specified_mem_size - total_mem_size);
317 	}
318 }
319 
320 static int __init __mpu_max_regions(void)
321 {
322 	/*
323 	 * We don't support a different number of I/D side regions so if we
324 	 * have separate instruction and data memory maps then return
325 	 * whichever side has a smaller number of supported regions.
326 	 */
327 	u32 dregions, iregions, mpuir;
328 
329 	mpuir = read_cpuid_mputype();
330 
331 	dregions = iregions = (mpuir & MPUIR_DREGION_SZMASK) >> MPUIR_DREGION;
332 
333 	/* Check for separate d-side and i-side memory maps */
334 	if (mpuir & MPUIR_nU)
335 		iregions = (mpuir & MPUIR_IREGION_SZMASK) >> MPUIR_IREGION;
336 
337 	/* Use the smallest of the two maxima */
338 	return min(dregions, iregions);
339 }
340 
341 static int __init mpu_iside_independent(void)
342 {
343 	/* MPUIR.nU specifies whether there is *not* a unified memory map */
344 	return read_cpuid_mputype() & MPUIR_nU;
345 }
346 
347 static int __init __mpu_min_region_order(void)
348 {
349 	u32 drbar_result, irbar_result;
350 
351 	/* We've kept a region free for this probing */
352 	rgnr_write(MPU_PROBE_REGION);
353 	isb();
354 	/*
355 	 * As per ARM ARM, write 0xFFFFFFFC to DRBAR to find the minimum
356 	 * region order
357 	*/
358 	drbar_write(0xFFFFFFFC);
359 	drbar_result = irbar_result = drbar_read();
360 	drbar_write(0x0);
361 	/* If the MPU is non-unified, we use the larger of the two minima*/
362 	if (mpu_iside_independent()) {
363 		irbar_write(0xFFFFFFFC);
364 		irbar_result = irbar_read();
365 		irbar_write(0x0);
366 	}
367 	isb(); /* Ensure that MPU region operations have completed */
368 	/* Return whichever result is larger */
369 
370 	return __ffs(max(drbar_result, irbar_result));
371 }
372 
373 static int __init mpu_setup_region(unsigned int number, phys_addr_t start,
374 				   unsigned int size_order, unsigned int properties,
375 				   unsigned int subregions, bool need_flush)
376 {
377 	u32 size_data;
378 
379 	/* We kept a region free for probing resolution of MPU regions*/
380 	if (number > mpu_max_regions
381 	    || number >= MPU_MAX_REGIONS)
382 		return -ENOENT;
383 
384 	if (size_order > 32)
385 		return -ENOMEM;
386 
387 	if (size_order < mpu_min_region_order)
388 		return -ENOMEM;
389 
390 	/* Writing N to bits 5:1 (RSR_SZ)  specifies region size 2^N+1 */
391 	size_data = ((size_order - 1) << MPU_RSR_SZ) | 1 << MPU_RSR_EN;
392 	size_data |= subregions << MPU_RSR_SD;
393 
394 	if (need_flush)
395 		flush_cache_all();
396 
397 	dsb(); /* Ensure all previous data accesses occur with old mappings */
398 	rgnr_write(number);
399 	isb();
400 	drbar_write(start);
401 	dracr_write(properties);
402 	isb(); /* Propagate properties before enabling region */
403 	drsr_write(size_data);
404 
405 	/* Check for independent I-side registers */
406 	if (mpu_iside_independent()) {
407 		irbar_write(start);
408 		iracr_write(properties);
409 		isb();
410 		irsr_write(size_data);
411 	}
412 	isb();
413 
414 	/* Store region info (we treat i/d side the same, so only store d) */
415 	mpu_rgn_info.rgns[number].dracr = properties;
416 	mpu_rgn_info.rgns[number].drbar = start;
417 	mpu_rgn_info.rgns[number].drsr = size_data;
418 
419 	mpu_rgn_info.used++;
420 
421 	return 0;
422 }
423 
424 /*
425 * Set up default MPU regions, doing nothing if there is no MPU
426 */
427 void __init mpu_setup(void)
428 {
429 	int i, region = 0, err = 0;
430 
431 	if (!mpu_present())
432 		return;
433 
434 	/* Setup MPU (order is important) */
435 
436 	/* Background */
437 	err |= mpu_setup_region(region++, 0, 32,
438 				MPU_ACR_XN | MPU_RGN_STRONGLY_ORDERED | MPU_AP_PL1RW_PL0RW,
439 				0, false);
440 
441 #ifdef CONFIG_XIP_KERNEL
442 	/* ROM */
443 	for (i = 0; i < ARRAY_SIZE(xip); i++) {
444 		/*
445                  * In case we overwrite RAM region we set earlier in
446                  * head-nommu.S (which is cachable) all subsequent
447                  * data access till we setup RAM bellow would be done
448                  * with BG region (which is uncachable), thus we need
449                  * to clean and invalidate cache.
450 		 */
451 		bool need_flush = region == MPU_RAM_REGION;
452 
453 		if (!xip[i].size)
454 			continue;
455 
456 		err |= mpu_setup_region(region++, xip[i].base, ilog2(xip[i].size),
457 					MPU_AP_PL1RO_PL0NA | MPU_RGN_NORMAL,
458 					xip[i].subreg, need_flush);
459 	}
460 #endif
461 
462 	/* RAM */
463 	for (i = 0; i < ARRAY_SIZE(mem); i++) {
464 		if (!mem[i].size)
465 			continue;
466 
467 		err |= mpu_setup_region(region++, mem[i].base, ilog2(mem[i].size),
468 					MPU_AP_PL1RW_PL0RW | MPU_RGN_NORMAL,
469 					mem[i].subreg, false);
470 	}
471 
472 	/* Vectors */
473 #ifndef CONFIG_CPU_V7M
474 	err |= mpu_setup_region(region++, vectors_base, ilog2(2 * PAGE_SIZE),
475 				MPU_AP_PL1RW_PL0NA | MPU_RGN_NORMAL,
476 				0, false);
477 #endif
478 	if (err) {
479 		panic("MPU region initialization failure! %d", err);
480 	} else {
481 		pr_info("Using ARMv7 PMSA Compliant MPU. "
482 			 "Region independence: %s, Used %d of %d regions\n",
483 			mpu_iside_independent() ? "Yes" : "No",
484 			mpu_rgn_info.used, mpu_max_regions);
485 	}
486 }
487