xref: /openbmc/linux/arch/arm/mm/cache-l2x0.c (revision e68f31f4)
1 /*
2  * arch/arm/mm/cache-l2x0.c - L210/L220 cache controller support
3  *
4  * Copyright (C) 2007 ARM Limited
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 as
8  * published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License
16  * along with this program; if not, write to the Free Software
17  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18  */
19 #include <linux/err.h>
20 #include <linux/init.h>
21 #include <linux/spinlock.h>
22 #include <linux/io.h>
23 #include <linux/of.h>
24 #include <linux/of_address.h>
25 
26 #include <asm/cacheflush.h>
27 #include <asm/hardware/cache-l2x0.h>
28 #include "cache-tauros3.h"
29 #include "cache-aurora-l2.h"
30 
31 #define CACHE_LINE_SIZE		32
32 
33 static void __iomem *l2x0_base;
34 static DEFINE_RAW_SPINLOCK(l2x0_lock);
35 static u32 l2x0_way_mask;	/* Bitmask of active ways */
36 static u32 l2x0_size;
37 static unsigned long sync_reg_offset = L2X0_CACHE_SYNC;
38 
39 /* Aurora don't have the cache ID register available, so we have to
40  * pass it though the device tree */
41 static u32  cache_id_part_number_from_dt;
42 
43 struct l2x0_regs l2x0_saved_regs;
44 
45 struct l2x0_of_data {
46 	void (*setup)(const struct device_node *, u32 *, u32 *);
47 	void (*save)(void);
48 	struct outer_cache_fns outer_cache;
49 };
50 
51 static bool of_init = false;
52 
53 static inline void cache_wait_way(void __iomem *reg, unsigned long mask)
54 {
55 	/* wait for cache operation by line or way to complete */
56 	while (readl_relaxed(reg) & mask)
57 		cpu_relax();
58 }
59 
60 #ifdef CONFIG_CACHE_PL310
61 static inline void cache_wait(void __iomem *reg, unsigned long mask)
62 {
63 	/* cache operations by line are atomic on PL310 */
64 }
65 #else
66 #define cache_wait	cache_wait_way
67 #endif
68 
69 static inline void cache_sync(void)
70 {
71 	void __iomem *base = l2x0_base;
72 
73 	writel_relaxed(0, base + sync_reg_offset);
74 	cache_wait(base + L2X0_CACHE_SYNC, 1);
75 }
76 
77 static inline void l2x0_clean_line(unsigned long addr)
78 {
79 	void __iomem *base = l2x0_base;
80 	cache_wait(base + L2X0_CLEAN_LINE_PA, 1);
81 	writel_relaxed(addr, base + L2X0_CLEAN_LINE_PA);
82 }
83 
84 static inline void l2x0_inv_line(unsigned long addr)
85 {
86 	void __iomem *base = l2x0_base;
87 	cache_wait(base + L2X0_INV_LINE_PA, 1);
88 	writel_relaxed(addr, base + L2X0_INV_LINE_PA);
89 }
90 
91 #if defined(CONFIG_PL310_ERRATA_588369) || defined(CONFIG_PL310_ERRATA_727915)
92 static inline void debug_writel(unsigned long val)
93 {
94 	if (outer_cache.set_debug)
95 		outer_cache.set_debug(val);
96 }
97 
98 static void pl310_set_debug(unsigned long val)
99 {
100 	writel_relaxed(val, l2x0_base + L2X0_DEBUG_CTRL);
101 }
102 #else
103 /* Optimised out for non-errata case */
104 static inline void debug_writel(unsigned long val)
105 {
106 }
107 
108 #define pl310_set_debug	NULL
109 #endif
110 
111 #ifdef CONFIG_PL310_ERRATA_588369
112 static inline void l2x0_flush_line(unsigned long addr)
113 {
114 	void __iomem *base = l2x0_base;
115 
116 	/* Clean by PA followed by Invalidate by PA */
117 	cache_wait(base + L2X0_CLEAN_LINE_PA, 1);
118 	writel_relaxed(addr, base + L2X0_CLEAN_LINE_PA);
119 	cache_wait(base + L2X0_INV_LINE_PA, 1);
120 	writel_relaxed(addr, base + L2X0_INV_LINE_PA);
121 }
122 #else
123 
124 static inline void l2x0_flush_line(unsigned long addr)
125 {
126 	void __iomem *base = l2x0_base;
127 	cache_wait(base + L2X0_CLEAN_INV_LINE_PA, 1);
128 	writel_relaxed(addr, base + L2X0_CLEAN_INV_LINE_PA);
129 }
130 #endif
131 
132 static void l2x0_cache_sync(void)
133 {
134 	unsigned long flags;
135 
136 	raw_spin_lock_irqsave(&l2x0_lock, flags);
137 	cache_sync();
138 	raw_spin_unlock_irqrestore(&l2x0_lock, flags);
139 }
140 
141 static void __l2x0_flush_all(void)
142 {
143 	debug_writel(0x03);
144 	writel_relaxed(l2x0_way_mask, l2x0_base + L2X0_CLEAN_INV_WAY);
145 	cache_wait_way(l2x0_base + L2X0_CLEAN_INV_WAY, l2x0_way_mask);
146 	cache_sync();
147 	debug_writel(0x00);
148 }
149 
150 static void l2x0_flush_all(void)
151 {
152 	unsigned long flags;
153 
154 	/* clean all ways */
155 	raw_spin_lock_irqsave(&l2x0_lock, flags);
156 	__l2x0_flush_all();
157 	raw_spin_unlock_irqrestore(&l2x0_lock, flags);
158 }
159 
160 static void l2x0_clean_all(void)
161 {
162 	unsigned long flags;
163 
164 	/* clean all ways */
165 	raw_spin_lock_irqsave(&l2x0_lock, flags);
166 	writel_relaxed(l2x0_way_mask, l2x0_base + L2X0_CLEAN_WAY);
167 	cache_wait_way(l2x0_base + L2X0_CLEAN_WAY, l2x0_way_mask);
168 	cache_sync();
169 	raw_spin_unlock_irqrestore(&l2x0_lock, flags);
170 }
171 
172 static void l2x0_inv_all(void)
173 {
174 	unsigned long flags;
175 
176 	/* invalidate all ways */
177 	raw_spin_lock_irqsave(&l2x0_lock, flags);
178 	/* Invalidating when L2 is enabled is a nono */
179 	BUG_ON(readl(l2x0_base + L2X0_CTRL) & L2X0_CTRL_EN);
180 	writel_relaxed(l2x0_way_mask, l2x0_base + L2X0_INV_WAY);
181 	cache_wait_way(l2x0_base + L2X0_INV_WAY, l2x0_way_mask);
182 	cache_sync();
183 	raw_spin_unlock_irqrestore(&l2x0_lock, flags);
184 }
185 
186 static void l2x0_inv_range(unsigned long start, unsigned long end)
187 {
188 	void __iomem *base = l2x0_base;
189 	unsigned long flags;
190 
191 	raw_spin_lock_irqsave(&l2x0_lock, flags);
192 	if (start & (CACHE_LINE_SIZE - 1)) {
193 		start &= ~(CACHE_LINE_SIZE - 1);
194 		debug_writel(0x03);
195 		l2x0_flush_line(start);
196 		debug_writel(0x00);
197 		start += CACHE_LINE_SIZE;
198 	}
199 
200 	if (end & (CACHE_LINE_SIZE - 1)) {
201 		end &= ~(CACHE_LINE_SIZE - 1);
202 		debug_writel(0x03);
203 		l2x0_flush_line(end);
204 		debug_writel(0x00);
205 	}
206 
207 	while (start < end) {
208 		unsigned long blk_end = start + min(end - start, 4096UL);
209 
210 		while (start < blk_end) {
211 			l2x0_inv_line(start);
212 			start += CACHE_LINE_SIZE;
213 		}
214 
215 		if (blk_end < end) {
216 			raw_spin_unlock_irqrestore(&l2x0_lock, flags);
217 			raw_spin_lock_irqsave(&l2x0_lock, flags);
218 		}
219 	}
220 	cache_wait(base + L2X0_INV_LINE_PA, 1);
221 	cache_sync();
222 	raw_spin_unlock_irqrestore(&l2x0_lock, flags);
223 }
224 
225 static void l2x0_clean_range(unsigned long start, unsigned long end)
226 {
227 	void __iomem *base = l2x0_base;
228 	unsigned long flags;
229 
230 	if ((end - start) >= l2x0_size) {
231 		l2x0_clean_all();
232 		return;
233 	}
234 
235 	raw_spin_lock_irqsave(&l2x0_lock, flags);
236 	start &= ~(CACHE_LINE_SIZE - 1);
237 	while (start < end) {
238 		unsigned long blk_end = start + min(end - start, 4096UL);
239 
240 		while (start < blk_end) {
241 			l2x0_clean_line(start);
242 			start += CACHE_LINE_SIZE;
243 		}
244 
245 		if (blk_end < end) {
246 			raw_spin_unlock_irqrestore(&l2x0_lock, flags);
247 			raw_spin_lock_irqsave(&l2x0_lock, flags);
248 		}
249 	}
250 	cache_wait(base + L2X0_CLEAN_LINE_PA, 1);
251 	cache_sync();
252 	raw_spin_unlock_irqrestore(&l2x0_lock, flags);
253 }
254 
255 static void l2x0_flush_range(unsigned long start, unsigned long end)
256 {
257 	void __iomem *base = l2x0_base;
258 	unsigned long flags;
259 
260 	if ((end - start) >= l2x0_size) {
261 		l2x0_flush_all();
262 		return;
263 	}
264 
265 	raw_spin_lock_irqsave(&l2x0_lock, flags);
266 	start &= ~(CACHE_LINE_SIZE - 1);
267 	while (start < end) {
268 		unsigned long blk_end = start + min(end - start, 4096UL);
269 
270 		debug_writel(0x03);
271 		while (start < blk_end) {
272 			l2x0_flush_line(start);
273 			start += CACHE_LINE_SIZE;
274 		}
275 		debug_writel(0x00);
276 
277 		if (blk_end < end) {
278 			raw_spin_unlock_irqrestore(&l2x0_lock, flags);
279 			raw_spin_lock_irqsave(&l2x0_lock, flags);
280 		}
281 	}
282 	cache_wait(base + L2X0_CLEAN_INV_LINE_PA, 1);
283 	cache_sync();
284 	raw_spin_unlock_irqrestore(&l2x0_lock, flags);
285 }
286 
287 static void l2x0_disable(void)
288 {
289 	unsigned long flags;
290 
291 	raw_spin_lock_irqsave(&l2x0_lock, flags);
292 	__l2x0_flush_all();
293 	writel_relaxed(0, l2x0_base + L2X0_CTRL);
294 	dsb(st);
295 	raw_spin_unlock_irqrestore(&l2x0_lock, flags);
296 }
297 
298 static void l2x0_unlock(u32 cache_id)
299 {
300 	int lockregs;
301 	int i;
302 
303 	switch (cache_id & L2X0_CACHE_ID_PART_MASK) {
304 	case L2X0_CACHE_ID_PART_L310:
305 		lockregs = 8;
306 		break;
307 	case AURORA_CACHE_ID:
308 		lockregs = 4;
309 		break;
310 	default:
311 		/* L210 and unknown types */
312 		lockregs = 1;
313 		break;
314 	}
315 
316 	for (i = 0; i < lockregs; i++) {
317 		writel_relaxed(0x0, l2x0_base + L2X0_LOCKDOWN_WAY_D_BASE +
318 			       i * L2X0_LOCKDOWN_STRIDE);
319 		writel_relaxed(0x0, l2x0_base + L2X0_LOCKDOWN_WAY_I_BASE +
320 			       i * L2X0_LOCKDOWN_STRIDE);
321 	}
322 }
323 
324 void __init l2x0_init(void __iomem *base, u32 aux_val, u32 aux_mask)
325 {
326 	u32 aux;
327 	u32 cache_id;
328 	u32 way_size = 0;
329 	int ways;
330 	int way_size_shift = L2X0_WAY_SIZE_SHIFT;
331 	const char *type;
332 
333 	l2x0_base = base;
334 	if (cache_id_part_number_from_dt)
335 		cache_id = cache_id_part_number_from_dt;
336 	else
337 		cache_id = readl_relaxed(l2x0_base + L2X0_CACHE_ID);
338 	aux = readl_relaxed(l2x0_base + L2X0_AUX_CTRL);
339 
340 	aux &= aux_mask;
341 	aux |= aux_val;
342 
343 	/* Determine the number of ways */
344 	switch (cache_id & L2X0_CACHE_ID_PART_MASK) {
345 	case L2X0_CACHE_ID_PART_L310:
346 		if (aux & (1 << 16))
347 			ways = 16;
348 		else
349 			ways = 8;
350 		type = "L310";
351 #ifdef CONFIG_PL310_ERRATA_753970
352 		/* Unmapped register. */
353 		sync_reg_offset = L2X0_DUMMY_REG;
354 #endif
355 		if ((cache_id & L2X0_CACHE_ID_RTL_MASK) <= L2X0_CACHE_ID_RTL_R3P0)
356 			outer_cache.set_debug = pl310_set_debug;
357 		break;
358 	case L2X0_CACHE_ID_PART_L210:
359 		ways = (aux >> 13) & 0xf;
360 		type = "L210";
361 		break;
362 
363 	case AURORA_CACHE_ID:
364 		sync_reg_offset = AURORA_SYNC_REG;
365 		ways = (aux >> 13) & 0xf;
366 		ways = 2 << ((ways + 1) >> 2);
367 		way_size_shift = AURORA_WAY_SIZE_SHIFT;
368 		type = "Aurora";
369 		break;
370 	default:
371 		/* Assume unknown chips have 8 ways */
372 		ways = 8;
373 		type = "L2x0 series";
374 		break;
375 	}
376 
377 	l2x0_way_mask = (1 << ways) - 1;
378 
379 	/*
380 	 * L2 cache Size =  Way size * Number of ways
381 	 */
382 	way_size = (aux & L2X0_AUX_CTRL_WAY_SIZE_MASK) >> 17;
383 	way_size = 1 << (way_size + way_size_shift);
384 
385 	l2x0_size = ways * way_size * SZ_1K;
386 
387 	/*
388 	 * Check if l2x0 controller is already enabled.
389 	 * If you are booting from non-secure mode
390 	 * accessing the below registers will fault.
391 	 */
392 	if (!(readl_relaxed(l2x0_base + L2X0_CTRL) & L2X0_CTRL_EN)) {
393 		/* Make sure that I&D is not locked down when starting */
394 		l2x0_unlock(cache_id);
395 
396 		/* l2x0 controller is disabled */
397 		writel_relaxed(aux, l2x0_base + L2X0_AUX_CTRL);
398 
399 		l2x0_inv_all();
400 
401 		/* enable L2X0 */
402 		writel_relaxed(L2X0_CTRL_EN, l2x0_base + L2X0_CTRL);
403 	}
404 
405 	/* Re-read it in case some bits are reserved. */
406 	aux = readl_relaxed(l2x0_base + L2X0_AUX_CTRL);
407 
408 	/* Save the value for resuming. */
409 	l2x0_saved_regs.aux_ctrl = aux;
410 
411 	if (!of_init) {
412 		outer_cache.inv_range = l2x0_inv_range;
413 		outer_cache.clean_range = l2x0_clean_range;
414 		outer_cache.flush_range = l2x0_flush_range;
415 		outer_cache.sync = l2x0_cache_sync;
416 		outer_cache.flush_all = l2x0_flush_all;
417 		outer_cache.inv_all = l2x0_inv_all;
418 		outer_cache.disable = l2x0_disable;
419 	}
420 
421 	pr_info("%s cache controller enabled\n", type);
422 	pr_info("l2x0: %d ways, CACHE_ID 0x%08x, AUX_CTRL 0x%08x, Cache size: %d kB\n",
423 		ways, cache_id, aux, l2x0_size >> 10);
424 }
425 
426 #ifdef CONFIG_OF
427 static int l2_wt_override;
428 
429 /*
430  * Note that the end addresses passed to Linux primitives are
431  * noninclusive, while the hardware cache range operations use
432  * inclusive start and end addresses.
433  */
434 static unsigned long calc_range_end(unsigned long start, unsigned long end)
435 {
436 	/*
437 	 * Limit the number of cache lines processed at once,
438 	 * since cache range operations stall the CPU pipeline
439 	 * until completion.
440 	 */
441 	if (end > start + MAX_RANGE_SIZE)
442 		end = start + MAX_RANGE_SIZE;
443 
444 	/*
445 	 * Cache range operations can't straddle a page boundary.
446 	 */
447 	if (end > PAGE_ALIGN(start+1))
448 		end = PAGE_ALIGN(start+1);
449 
450 	return end;
451 }
452 
453 /*
454  * Make sure 'start' and 'end' reference the same page, as L2 is PIPT
455  * and range operations only do a TLB lookup on the start address.
456  */
457 static void aurora_pa_range(unsigned long start, unsigned long end,
458 			unsigned long offset)
459 {
460 	unsigned long flags;
461 
462 	raw_spin_lock_irqsave(&l2x0_lock, flags);
463 	writel_relaxed(start, l2x0_base + AURORA_RANGE_BASE_ADDR_REG);
464 	writel_relaxed(end, l2x0_base + offset);
465 	raw_spin_unlock_irqrestore(&l2x0_lock, flags);
466 
467 	cache_sync();
468 }
469 
470 static void aurora_inv_range(unsigned long start, unsigned long end)
471 {
472 	/*
473 	 * round start and end adresses up to cache line size
474 	 */
475 	start &= ~(CACHE_LINE_SIZE - 1);
476 	end = ALIGN(end, CACHE_LINE_SIZE);
477 
478 	/*
479 	 * Invalidate all full cache lines between 'start' and 'end'.
480 	 */
481 	while (start < end) {
482 		unsigned long range_end = calc_range_end(start, end);
483 		aurora_pa_range(start, range_end - CACHE_LINE_SIZE,
484 				AURORA_INVAL_RANGE_REG);
485 		start = range_end;
486 	}
487 }
488 
489 static void aurora_clean_range(unsigned long start, unsigned long end)
490 {
491 	/*
492 	 * If L2 is forced to WT, the L2 will always be clean and we
493 	 * don't need to do anything here.
494 	 */
495 	if (!l2_wt_override) {
496 		start &= ~(CACHE_LINE_SIZE - 1);
497 		end = ALIGN(end, CACHE_LINE_SIZE);
498 		while (start != end) {
499 			unsigned long range_end = calc_range_end(start, end);
500 			aurora_pa_range(start, range_end - CACHE_LINE_SIZE,
501 					AURORA_CLEAN_RANGE_REG);
502 			start = range_end;
503 		}
504 	}
505 }
506 
507 static void aurora_flush_range(unsigned long start, unsigned long end)
508 {
509 	start &= ~(CACHE_LINE_SIZE - 1);
510 	end = ALIGN(end, CACHE_LINE_SIZE);
511 	while (start != end) {
512 		unsigned long range_end = calc_range_end(start, end);
513 		/*
514 		 * If L2 is forced to WT, the L2 will always be clean and we
515 		 * just need to invalidate.
516 		 */
517 		if (l2_wt_override)
518 			aurora_pa_range(start, range_end - CACHE_LINE_SIZE,
519 							AURORA_INVAL_RANGE_REG);
520 		else
521 			aurora_pa_range(start, range_end - CACHE_LINE_SIZE,
522 							AURORA_FLUSH_RANGE_REG);
523 		start = range_end;
524 	}
525 }
526 
527 /*
528  * For certain Broadcom SoCs, depending on the address range, different offsets
529  * need to be added to the address before passing it to L2 for
530  * invalidation/clean/flush
531  *
532  * Section Address Range              Offset        EMI
533  *   1     0x00000000 - 0x3FFFFFFF    0x80000000    VC
534  *   2     0x40000000 - 0xBFFFFFFF    0x40000000    SYS
535  *   3     0xC0000000 - 0xFFFFFFFF    0x80000000    VC
536  *
537  * When the start and end addresses have crossed two different sections, we
538  * need to break the L2 operation into two, each within its own section.
539  * For example, if we need to invalidate addresses starts at 0xBFFF0000 and
540  * ends at 0xC0001000, we need do invalidate 1) 0xBFFF0000 - 0xBFFFFFFF and 2)
541  * 0xC0000000 - 0xC0001000
542  *
543  * Note 1:
544  * By breaking a single L2 operation into two, we may potentially suffer some
545  * performance hit, but keep in mind the cross section case is very rare
546  *
547  * Note 2:
548  * We do not need to handle the case when the start address is in
549  * Section 1 and the end address is in Section 3, since it is not a valid use
550  * case
551  *
552  * Note 3:
553  * Section 1 in practical terms can no longer be used on rev A2. Because of
554  * that the code does not need to handle section 1 at all.
555  *
556  */
557 #define BCM_SYS_EMI_START_ADDR        0x40000000UL
558 #define BCM_VC_EMI_SEC3_START_ADDR    0xC0000000UL
559 
560 #define BCM_SYS_EMI_OFFSET            0x40000000UL
561 #define BCM_VC_EMI_OFFSET             0x80000000UL
562 
563 static inline int bcm_addr_is_sys_emi(unsigned long addr)
564 {
565 	return (addr >= BCM_SYS_EMI_START_ADDR) &&
566 		(addr < BCM_VC_EMI_SEC3_START_ADDR);
567 }
568 
569 static inline unsigned long bcm_l2_phys_addr(unsigned long addr)
570 {
571 	if (bcm_addr_is_sys_emi(addr))
572 		return addr + BCM_SYS_EMI_OFFSET;
573 	else
574 		return addr + BCM_VC_EMI_OFFSET;
575 }
576 
577 static void bcm_inv_range(unsigned long start, unsigned long end)
578 {
579 	unsigned long new_start, new_end;
580 
581 	BUG_ON(start < BCM_SYS_EMI_START_ADDR);
582 
583 	if (unlikely(end <= start))
584 		return;
585 
586 	new_start = bcm_l2_phys_addr(start);
587 	new_end = bcm_l2_phys_addr(end);
588 
589 	/* normal case, no cross section between start and end */
590 	if (likely(bcm_addr_is_sys_emi(end) || !bcm_addr_is_sys_emi(start))) {
591 		l2x0_inv_range(new_start, new_end);
592 		return;
593 	}
594 
595 	/* They cross sections, so it can only be a cross from section
596 	 * 2 to section 3
597 	 */
598 	l2x0_inv_range(new_start,
599 		bcm_l2_phys_addr(BCM_VC_EMI_SEC3_START_ADDR-1));
600 	l2x0_inv_range(bcm_l2_phys_addr(BCM_VC_EMI_SEC3_START_ADDR),
601 		new_end);
602 }
603 
604 static void bcm_clean_range(unsigned long start, unsigned long end)
605 {
606 	unsigned long new_start, new_end;
607 
608 	BUG_ON(start < BCM_SYS_EMI_START_ADDR);
609 
610 	if (unlikely(end <= start))
611 		return;
612 
613 	if ((end - start) >= l2x0_size) {
614 		l2x0_clean_all();
615 		return;
616 	}
617 
618 	new_start = bcm_l2_phys_addr(start);
619 	new_end = bcm_l2_phys_addr(end);
620 
621 	/* normal case, no cross section between start and end */
622 	if (likely(bcm_addr_is_sys_emi(end) || !bcm_addr_is_sys_emi(start))) {
623 		l2x0_clean_range(new_start, new_end);
624 		return;
625 	}
626 
627 	/* They cross sections, so it can only be a cross from section
628 	 * 2 to section 3
629 	 */
630 	l2x0_clean_range(new_start,
631 		bcm_l2_phys_addr(BCM_VC_EMI_SEC3_START_ADDR-1));
632 	l2x0_clean_range(bcm_l2_phys_addr(BCM_VC_EMI_SEC3_START_ADDR),
633 		new_end);
634 }
635 
636 static void bcm_flush_range(unsigned long start, unsigned long end)
637 {
638 	unsigned long new_start, new_end;
639 
640 	BUG_ON(start < BCM_SYS_EMI_START_ADDR);
641 
642 	if (unlikely(end <= start))
643 		return;
644 
645 	if ((end - start) >= l2x0_size) {
646 		l2x0_flush_all();
647 		return;
648 	}
649 
650 	new_start = bcm_l2_phys_addr(start);
651 	new_end = bcm_l2_phys_addr(end);
652 
653 	/* normal case, no cross section between start and end */
654 	if (likely(bcm_addr_is_sys_emi(end) || !bcm_addr_is_sys_emi(start))) {
655 		l2x0_flush_range(new_start, new_end);
656 		return;
657 	}
658 
659 	/* They cross sections, so it can only be a cross from section
660 	 * 2 to section 3
661 	 */
662 	l2x0_flush_range(new_start,
663 		bcm_l2_phys_addr(BCM_VC_EMI_SEC3_START_ADDR-1));
664 	l2x0_flush_range(bcm_l2_phys_addr(BCM_VC_EMI_SEC3_START_ADDR),
665 		new_end);
666 }
667 
668 static void __init l2x0_of_setup(const struct device_node *np,
669 				 u32 *aux_val, u32 *aux_mask)
670 {
671 	u32 data[2] = { 0, 0 };
672 	u32 tag = 0;
673 	u32 dirty = 0;
674 	u32 val = 0, mask = 0;
675 
676 	of_property_read_u32(np, "arm,tag-latency", &tag);
677 	if (tag) {
678 		mask |= L2X0_AUX_CTRL_TAG_LATENCY_MASK;
679 		val |= (tag - 1) << L2X0_AUX_CTRL_TAG_LATENCY_SHIFT;
680 	}
681 
682 	of_property_read_u32_array(np, "arm,data-latency",
683 				   data, ARRAY_SIZE(data));
684 	if (data[0] && data[1]) {
685 		mask |= L2X0_AUX_CTRL_DATA_RD_LATENCY_MASK |
686 			L2X0_AUX_CTRL_DATA_WR_LATENCY_MASK;
687 		val |= ((data[0] - 1) << L2X0_AUX_CTRL_DATA_RD_LATENCY_SHIFT) |
688 		       ((data[1] - 1) << L2X0_AUX_CTRL_DATA_WR_LATENCY_SHIFT);
689 	}
690 
691 	of_property_read_u32(np, "arm,dirty-latency", &dirty);
692 	if (dirty) {
693 		mask |= L2X0_AUX_CTRL_DIRTY_LATENCY_MASK;
694 		val |= (dirty - 1) << L2X0_AUX_CTRL_DIRTY_LATENCY_SHIFT;
695 	}
696 
697 	*aux_val &= ~mask;
698 	*aux_val |= val;
699 	*aux_mask &= ~mask;
700 }
701 
702 static void __init pl310_of_setup(const struct device_node *np,
703 				  u32 *aux_val, u32 *aux_mask)
704 {
705 	u32 data[3] = { 0, 0, 0 };
706 	u32 tag[3] = { 0, 0, 0 };
707 	u32 filter[2] = { 0, 0 };
708 
709 	of_property_read_u32_array(np, "arm,tag-latency", tag, ARRAY_SIZE(tag));
710 	if (tag[0] && tag[1] && tag[2])
711 		writel_relaxed(
712 			((tag[0] - 1) << L2X0_LATENCY_CTRL_RD_SHIFT) |
713 			((tag[1] - 1) << L2X0_LATENCY_CTRL_WR_SHIFT) |
714 			((tag[2] - 1) << L2X0_LATENCY_CTRL_SETUP_SHIFT),
715 			l2x0_base + L2X0_TAG_LATENCY_CTRL);
716 
717 	of_property_read_u32_array(np, "arm,data-latency",
718 				   data, ARRAY_SIZE(data));
719 	if (data[0] && data[1] && data[2])
720 		writel_relaxed(
721 			((data[0] - 1) << L2X0_LATENCY_CTRL_RD_SHIFT) |
722 			((data[1] - 1) << L2X0_LATENCY_CTRL_WR_SHIFT) |
723 			((data[2] - 1) << L2X0_LATENCY_CTRL_SETUP_SHIFT),
724 			l2x0_base + L2X0_DATA_LATENCY_CTRL);
725 
726 	of_property_read_u32_array(np, "arm,filter-ranges",
727 				   filter, ARRAY_SIZE(filter));
728 	if (filter[1]) {
729 		writel_relaxed(ALIGN(filter[0] + filter[1], SZ_1M),
730 			       l2x0_base + L2X0_ADDR_FILTER_END);
731 		writel_relaxed((filter[0] & ~(SZ_1M - 1)) | L2X0_ADDR_FILTER_EN,
732 			       l2x0_base + L2X0_ADDR_FILTER_START);
733 	}
734 }
735 
736 static void __init pl310_save(void)
737 {
738 	u32 l2x0_revision = readl_relaxed(l2x0_base + L2X0_CACHE_ID) &
739 		L2X0_CACHE_ID_RTL_MASK;
740 
741 	l2x0_saved_regs.tag_latency = readl_relaxed(l2x0_base +
742 		L2X0_TAG_LATENCY_CTRL);
743 	l2x0_saved_regs.data_latency = readl_relaxed(l2x0_base +
744 		L2X0_DATA_LATENCY_CTRL);
745 	l2x0_saved_regs.filter_end = readl_relaxed(l2x0_base +
746 		L2X0_ADDR_FILTER_END);
747 	l2x0_saved_regs.filter_start = readl_relaxed(l2x0_base +
748 		L2X0_ADDR_FILTER_START);
749 
750 	if (l2x0_revision >= L2X0_CACHE_ID_RTL_R2P0) {
751 		/*
752 		 * From r2p0, there is Prefetch offset/control register
753 		 */
754 		l2x0_saved_regs.prefetch_ctrl = readl_relaxed(l2x0_base +
755 			L2X0_PREFETCH_CTRL);
756 		/*
757 		 * From r3p0, there is Power control register
758 		 */
759 		if (l2x0_revision >= L2X0_CACHE_ID_RTL_R3P0)
760 			l2x0_saved_regs.pwr_ctrl = readl_relaxed(l2x0_base +
761 				L2X0_POWER_CTRL);
762 	}
763 }
764 
765 static void aurora_save(void)
766 {
767 	l2x0_saved_regs.ctrl = readl_relaxed(l2x0_base + L2X0_CTRL);
768 	l2x0_saved_regs.aux_ctrl = readl_relaxed(l2x0_base + L2X0_AUX_CTRL);
769 }
770 
771 static void __init tauros3_save(void)
772 {
773 	l2x0_saved_regs.aux2_ctrl =
774 		readl_relaxed(l2x0_base + TAUROS3_AUX2_CTRL);
775 	l2x0_saved_regs.prefetch_ctrl =
776 		readl_relaxed(l2x0_base + L2X0_PREFETCH_CTRL);
777 }
778 
779 static void l2x0_resume(void)
780 {
781 	if (!(readl_relaxed(l2x0_base + L2X0_CTRL) & L2X0_CTRL_EN)) {
782 		/* restore aux ctrl and enable l2 */
783 		l2x0_unlock(readl_relaxed(l2x0_base + L2X0_CACHE_ID));
784 
785 		writel_relaxed(l2x0_saved_regs.aux_ctrl, l2x0_base +
786 			L2X0_AUX_CTRL);
787 
788 		l2x0_inv_all();
789 
790 		writel_relaxed(L2X0_CTRL_EN, l2x0_base + L2X0_CTRL);
791 	}
792 }
793 
794 static void pl310_resume(void)
795 {
796 	u32 l2x0_revision;
797 
798 	if (!(readl_relaxed(l2x0_base + L2X0_CTRL) & L2X0_CTRL_EN)) {
799 		/* restore pl310 setup */
800 		writel_relaxed(l2x0_saved_regs.tag_latency,
801 			l2x0_base + L2X0_TAG_LATENCY_CTRL);
802 		writel_relaxed(l2x0_saved_regs.data_latency,
803 			l2x0_base + L2X0_DATA_LATENCY_CTRL);
804 		writel_relaxed(l2x0_saved_regs.filter_end,
805 			l2x0_base + L2X0_ADDR_FILTER_END);
806 		writel_relaxed(l2x0_saved_regs.filter_start,
807 			l2x0_base + L2X0_ADDR_FILTER_START);
808 
809 		l2x0_revision = readl_relaxed(l2x0_base + L2X0_CACHE_ID) &
810 			L2X0_CACHE_ID_RTL_MASK;
811 
812 		if (l2x0_revision >= L2X0_CACHE_ID_RTL_R2P0) {
813 			writel_relaxed(l2x0_saved_regs.prefetch_ctrl,
814 				l2x0_base + L2X0_PREFETCH_CTRL);
815 			if (l2x0_revision >= L2X0_CACHE_ID_RTL_R3P0)
816 				writel_relaxed(l2x0_saved_regs.pwr_ctrl,
817 					l2x0_base + L2X0_POWER_CTRL);
818 		}
819 	}
820 
821 	l2x0_resume();
822 }
823 
824 static void aurora_resume(void)
825 {
826 	if (!(readl(l2x0_base + L2X0_CTRL) & L2X0_CTRL_EN)) {
827 		writel_relaxed(l2x0_saved_regs.aux_ctrl,
828 				l2x0_base + L2X0_AUX_CTRL);
829 		writel_relaxed(l2x0_saved_regs.ctrl, l2x0_base + L2X0_CTRL);
830 	}
831 }
832 
833 static void tauros3_resume(void)
834 {
835 	if (!(readl_relaxed(l2x0_base + L2X0_CTRL) & L2X0_CTRL_EN)) {
836 		writel_relaxed(l2x0_saved_regs.aux2_ctrl,
837 			       l2x0_base + TAUROS3_AUX2_CTRL);
838 		writel_relaxed(l2x0_saved_regs.prefetch_ctrl,
839 			       l2x0_base + L2X0_PREFETCH_CTRL);
840 	}
841 
842 	l2x0_resume();
843 }
844 
845 static void __init aurora_broadcast_l2_commands(void)
846 {
847 	__u32 u;
848 	/* Enable Broadcasting of cache commands to L2*/
849 	__asm__ __volatile__("mrc p15, 1, %0, c15, c2, 0" : "=r"(u));
850 	u |= AURORA_CTRL_FW;		/* Set the FW bit */
851 	__asm__ __volatile__("mcr p15, 1, %0, c15, c2, 0\n" : : "r"(u));
852 	isb();
853 }
854 
855 static void __init aurora_of_setup(const struct device_node *np,
856 				u32 *aux_val, u32 *aux_mask)
857 {
858 	u32 val = AURORA_ACR_REPLACEMENT_TYPE_SEMIPLRU;
859 	u32 mask =  AURORA_ACR_REPLACEMENT_MASK;
860 
861 	of_property_read_u32(np, "cache-id-part",
862 			&cache_id_part_number_from_dt);
863 
864 	/* Determine and save the write policy */
865 	l2_wt_override = of_property_read_bool(np, "wt-override");
866 
867 	if (l2_wt_override) {
868 		val |= AURORA_ACR_FORCE_WRITE_THRO_POLICY;
869 		mask |= AURORA_ACR_FORCE_WRITE_POLICY_MASK;
870 	}
871 
872 	*aux_val &= ~mask;
873 	*aux_val |= val;
874 	*aux_mask &= ~mask;
875 }
876 
877 static const struct l2x0_of_data pl310_data = {
878 	.setup = pl310_of_setup,
879 	.save  = pl310_save,
880 	.outer_cache = {
881 		.resume      = pl310_resume,
882 		.inv_range   = l2x0_inv_range,
883 		.clean_range = l2x0_clean_range,
884 		.flush_range = l2x0_flush_range,
885 		.sync        = l2x0_cache_sync,
886 		.flush_all   = l2x0_flush_all,
887 		.inv_all     = l2x0_inv_all,
888 		.disable     = l2x0_disable,
889 	},
890 };
891 
892 static const struct l2x0_of_data l2x0_data = {
893 	.setup = l2x0_of_setup,
894 	.save  = NULL,
895 	.outer_cache = {
896 		.resume      = l2x0_resume,
897 		.inv_range   = l2x0_inv_range,
898 		.clean_range = l2x0_clean_range,
899 		.flush_range = l2x0_flush_range,
900 		.sync        = l2x0_cache_sync,
901 		.flush_all   = l2x0_flush_all,
902 		.inv_all     = l2x0_inv_all,
903 		.disable     = l2x0_disable,
904 	},
905 };
906 
907 static const struct l2x0_of_data aurora_with_outer_data = {
908 	.setup = aurora_of_setup,
909 	.save  = aurora_save,
910 	.outer_cache = {
911 		.resume      = aurora_resume,
912 		.inv_range   = aurora_inv_range,
913 		.clean_range = aurora_clean_range,
914 		.flush_range = aurora_flush_range,
915 		.sync        = l2x0_cache_sync,
916 		.flush_all   = l2x0_flush_all,
917 		.inv_all     = l2x0_inv_all,
918 		.disable     = l2x0_disable,
919 	},
920 };
921 
922 static const struct l2x0_of_data aurora_no_outer_data = {
923 	.setup = aurora_of_setup,
924 	.save  = aurora_save,
925 	.outer_cache = {
926 		.resume      = aurora_resume,
927 	},
928 };
929 
930 static const struct l2x0_of_data tauros3_data = {
931 	.setup = NULL,
932 	.save  = tauros3_save,
933 	/* Tauros3 broadcasts L1 cache operations to L2 */
934 	.outer_cache = {
935 		.resume      = tauros3_resume,
936 	},
937 };
938 
939 static const struct l2x0_of_data bcm_l2x0_data = {
940 	.setup = pl310_of_setup,
941 	.save  = pl310_save,
942 	.outer_cache = {
943 		.resume      = pl310_resume,
944 		.inv_range   = bcm_inv_range,
945 		.clean_range = bcm_clean_range,
946 		.flush_range = bcm_flush_range,
947 		.sync        = l2x0_cache_sync,
948 		.flush_all   = l2x0_flush_all,
949 		.inv_all     = l2x0_inv_all,
950 		.disable     = l2x0_disable,
951 	},
952 };
953 
954 static const struct of_device_id l2x0_ids[] __initconst = {
955 	{ .compatible = "arm,l210-cache", .data = (void *)&l2x0_data },
956 	{ .compatible = "arm,l220-cache", .data = (void *)&l2x0_data },
957 	{ .compatible = "arm,pl310-cache", .data = (void *)&pl310_data },
958 	{ .compatible = "bcm,bcm11351-a2-pl310-cache", /* deprecated name */
959 	  .data = (void *)&bcm_l2x0_data},
960 	{ .compatible = "brcm,bcm11351-a2-pl310-cache",
961 	  .data = (void *)&bcm_l2x0_data},
962 	{ .compatible = "marvell,aurora-outer-cache",
963 	  .data = (void *)&aurora_with_outer_data},
964 	{ .compatible = "marvell,aurora-system-cache",
965 	  .data = (void *)&aurora_no_outer_data},
966 	{ .compatible = "marvell,tauros3-cache",
967 	  .data = (void *)&tauros3_data },
968 	{}
969 };
970 
971 int __init l2x0_of_init(u32 aux_val, u32 aux_mask)
972 {
973 	struct device_node *np;
974 	const struct l2x0_of_data *data;
975 	struct resource res;
976 
977 	np = of_find_matching_node(NULL, l2x0_ids);
978 	if (!np)
979 		return -ENODEV;
980 
981 	if (of_address_to_resource(np, 0, &res))
982 		return -ENODEV;
983 
984 	l2x0_base = ioremap(res.start, resource_size(&res));
985 	if (!l2x0_base)
986 		return -ENOMEM;
987 
988 	l2x0_saved_regs.phy_base = res.start;
989 
990 	data = of_match_node(l2x0_ids, np)->data;
991 
992 	/* L2 configuration can only be changed if the cache is disabled */
993 	if (!(readl_relaxed(l2x0_base + L2X0_CTRL) & L2X0_CTRL_EN)) {
994 		if (data->setup)
995 			data->setup(np, &aux_val, &aux_mask);
996 
997 		/* For aurora cache in no outer mode select the
998 		 * correct mode using the coprocessor*/
999 		if (data == &aurora_no_outer_data)
1000 			aurora_broadcast_l2_commands();
1001 	}
1002 
1003 	if (data->save)
1004 		data->save();
1005 
1006 	of_init = true;
1007 	memcpy(&outer_cache, &data->outer_cache, sizeof(outer_cache));
1008 	l2x0_init(l2x0_base, aux_val, aux_mask);
1009 
1010 	return 0;
1011 }
1012 #endif
1013