xref: /openbmc/linux/arch/arm/mm/cache-l2x0.c (revision f777332b)
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 struct l2c_init_data {
32 	unsigned num_lock;
33 	void (*of_parse)(const struct device_node *, u32 *, u32 *);
34 	void (*enable)(void __iomem *, u32, unsigned);
35 	void (*fixup)(void __iomem *, u32, struct outer_cache_fns *);
36 	void (*save)(void __iomem *);
37 	struct outer_cache_fns outer_cache;
38 };
39 
40 #define CACHE_LINE_SIZE		32
41 
42 static void __iomem *l2x0_base;
43 static DEFINE_RAW_SPINLOCK(l2x0_lock);
44 static u32 l2x0_way_mask;	/* Bitmask of active ways */
45 static u32 l2x0_size;
46 static unsigned long sync_reg_offset = L2X0_CACHE_SYNC;
47 
48 struct l2x0_regs l2x0_saved_regs;
49 
50 /*
51  * Common code for all cache controllers.
52  */
53 static inline void l2c_wait_mask(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 /*
61  * This should only be called when we have a requirement that the
62  * register be written due to a work-around, as platforms running
63  * in non-secure mode may not be able to access this register.
64  */
65 static inline void l2c_set_debug(void __iomem *base, unsigned long val)
66 {
67 	outer_cache.set_debug(val);
68 }
69 
70 static void __l2c_op_way(void __iomem *reg)
71 {
72 	writel_relaxed(l2x0_way_mask, reg);
73 	l2c_wait_mask(reg, l2x0_way_mask);
74 }
75 
76 static inline void l2c_unlock(void __iomem *base, unsigned num)
77 {
78 	unsigned i;
79 
80 	for (i = 0; i < num; i++) {
81 		writel_relaxed(0, base + L2X0_LOCKDOWN_WAY_D_BASE +
82 			       i * L2X0_LOCKDOWN_STRIDE);
83 		writel_relaxed(0, base + L2X0_LOCKDOWN_WAY_I_BASE +
84 			       i * L2X0_LOCKDOWN_STRIDE);
85 	}
86 }
87 
88 /*
89  * Enable the L2 cache controller.  This function must only be
90  * called when the cache controller is known to be disabled.
91  */
92 static void l2c_enable(void __iomem *base, u32 aux, unsigned num_lock)
93 {
94 	unsigned long flags;
95 
96 	/* Only write the aux register if it needs changing */
97 	if (readl_relaxed(base + L2X0_AUX_CTRL) != aux)
98 		writel_relaxed(aux, base + L2X0_AUX_CTRL);
99 
100 	l2c_unlock(base, num_lock);
101 
102 	local_irq_save(flags);
103 	__l2c_op_way(base + L2X0_INV_WAY);
104 	writel_relaxed(0, base + sync_reg_offset);
105 	l2c_wait_mask(base + sync_reg_offset, 1);
106 	local_irq_restore(flags);
107 
108 	writel_relaxed(L2X0_CTRL_EN, base + L2X0_CTRL);
109 }
110 
111 static void l2c_disable(void)
112 {
113 	void __iomem *base = l2x0_base;
114 
115 	outer_cache.flush_all();
116 	writel_relaxed(0, base + L2X0_CTRL);
117 	dsb(st);
118 }
119 
120 #ifdef CONFIG_CACHE_PL310
121 static inline void cache_wait(void __iomem *reg, unsigned long mask)
122 {
123 	/* cache operations by line are atomic on PL310 */
124 }
125 #else
126 #define cache_wait	l2c_wait_mask
127 #endif
128 
129 static inline void cache_sync(void)
130 {
131 	void __iomem *base = l2x0_base;
132 
133 	writel_relaxed(0, base + sync_reg_offset);
134 	cache_wait(base + L2X0_CACHE_SYNC, 1);
135 }
136 
137 static inline void l2x0_clean_line(unsigned long addr)
138 {
139 	void __iomem *base = l2x0_base;
140 	cache_wait(base + L2X0_CLEAN_LINE_PA, 1);
141 	writel_relaxed(addr, base + L2X0_CLEAN_LINE_PA);
142 }
143 
144 static inline void l2x0_inv_line(unsigned long addr)
145 {
146 	void __iomem *base = l2x0_base;
147 	cache_wait(base + L2X0_INV_LINE_PA, 1);
148 	writel_relaxed(addr, base + L2X0_INV_LINE_PA);
149 }
150 
151 #if defined(CONFIG_PL310_ERRATA_588369) || defined(CONFIG_PL310_ERRATA_727915)
152 static inline void debug_writel(unsigned long val)
153 {
154 	if (outer_cache.set_debug)
155 		l2c_set_debug(l2x0_base, val);
156 }
157 #else
158 /* Optimised out for non-errata case */
159 static inline void debug_writel(unsigned long val)
160 {
161 }
162 #endif
163 
164 #ifdef CONFIG_PL310_ERRATA_588369
165 static inline void l2x0_flush_line(unsigned long addr)
166 {
167 	void __iomem *base = l2x0_base;
168 
169 	/* Clean by PA followed by Invalidate by PA */
170 	cache_wait(base + L2X0_CLEAN_LINE_PA, 1);
171 	writel_relaxed(addr, base + L2X0_CLEAN_LINE_PA);
172 	cache_wait(base + L2X0_INV_LINE_PA, 1);
173 	writel_relaxed(addr, base + L2X0_INV_LINE_PA);
174 }
175 #else
176 
177 static inline void l2x0_flush_line(unsigned long addr)
178 {
179 	void __iomem *base = l2x0_base;
180 	cache_wait(base + L2X0_CLEAN_INV_LINE_PA, 1);
181 	writel_relaxed(addr, base + L2X0_CLEAN_INV_LINE_PA);
182 }
183 #endif
184 
185 static void l2x0_cache_sync(void)
186 {
187 	unsigned long flags;
188 
189 	raw_spin_lock_irqsave(&l2x0_lock, flags);
190 	cache_sync();
191 	raw_spin_unlock_irqrestore(&l2x0_lock, flags);
192 }
193 
194 static void __l2x0_flush_all(void)
195 {
196 	debug_writel(0x03);
197 	__l2c_op_way(l2x0_base + L2X0_CLEAN_INV_WAY);
198 	cache_sync();
199 	debug_writel(0x00);
200 }
201 
202 static void l2x0_flush_all(void)
203 {
204 	unsigned long flags;
205 
206 	/* clean all ways */
207 	raw_spin_lock_irqsave(&l2x0_lock, flags);
208 	__l2x0_flush_all();
209 	raw_spin_unlock_irqrestore(&l2x0_lock, flags);
210 }
211 
212 static void l2x0_clean_all(void)
213 {
214 	unsigned long flags;
215 
216 	/* clean all ways */
217 	raw_spin_lock_irqsave(&l2x0_lock, flags);
218 	__l2c_op_way(l2x0_base + L2X0_CLEAN_WAY);
219 	cache_sync();
220 	raw_spin_unlock_irqrestore(&l2x0_lock, flags);
221 }
222 
223 static void l2x0_inv_all(void)
224 {
225 	unsigned long flags;
226 
227 	/* invalidate all ways */
228 	raw_spin_lock_irqsave(&l2x0_lock, flags);
229 	/* Invalidating when L2 is enabled is a nono */
230 	BUG_ON(readl(l2x0_base + L2X0_CTRL) & L2X0_CTRL_EN);
231 	__l2c_op_way(l2x0_base + L2X0_INV_WAY);
232 	cache_sync();
233 	raw_spin_unlock_irqrestore(&l2x0_lock, flags);
234 }
235 
236 static void l2x0_inv_range(unsigned long start, unsigned long end)
237 {
238 	void __iomem *base = l2x0_base;
239 	unsigned long flags;
240 
241 	raw_spin_lock_irqsave(&l2x0_lock, flags);
242 	if (start & (CACHE_LINE_SIZE - 1)) {
243 		start &= ~(CACHE_LINE_SIZE - 1);
244 		debug_writel(0x03);
245 		l2x0_flush_line(start);
246 		debug_writel(0x00);
247 		start += CACHE_LINE_SIZE;
248 	}
249 
250 	if (end & (CACHE_LINE_SIZE - 1)) {
251 		end &= ~(CACHE_LINE_SIZE - 1);
252 		debug_writel(0x03);
253 		l2x0_flush_line(end);
254 		debug_writel(0x00);
255 	}
256 
257 	while (start < end) {
258 		unsigned long blk_end = start + min(end - start, 4096UL);
259 
260 		while (start < blk_end) {
261 			l2x0_inv_line(start);
262 			start += CACHE_LINE_SIZE;
263 		}
264 
265 		if (blk_end < end) {
266 			raw_spin_unlock_irqrestore(&l2x0_lock, flags);
267 			raw_spin_lock_irqsave(&l2x0_lock, flags);
268 		}
269 	}
270 	cache_wait(base + L2X0_INV_LINE_PA, 1);
271 	cache_sync();
272 	raw_spin_unlock_irqrestore(&l2x0_lock, flags);
273 }
274 
275 static void l2x0_clean_range(unsigned long start, unsigned long end)
276 {
277 	void __iomem *base = l2x0_base;
278 	unsigned long flags;
279 
280 	if ((end - start) >= l2x0_size) {
281 		l2x0_clean_all();
282 		return;
283 	}
284 
285 	raw_spin_lock_irqsave(&l2x0_lock, flags);
286 	start &= ~(CACHE_LINE_SIZE - 1);
287 	while (start < end) {
288 		unsigned long blk_end = start + min(end - start, 4096UL);
289 
290 		while (start < blk_end) {
291 			l2x0_clean_line(start);
292 			start += CACHE_LINE_SIZE;
293 		}
294 
295 		if (blk_end < end) {
296 			raw_spin_unlock_irqrestore(&l2x0_lock, flags);
297 			raw_spin_lock_irqsave(&l2x0_lock, flags);
298 		}
299 	}
300 	cache_wait(base + L2X0_CLEAN_LINE_PA, 1);
301 	cache_sync();
302 	raw_spin_unlock_irqrestore(&l2x0_lock, flags);
303 }
304 
305 static void l2x0_flush_range(unsigned long start, unsigned long end)
306 {
307 	void __iomem *base = l2x0_base;
308 	unsigned long flags;
309 
310 	if ((end - start) >= l2x0_size) {
311 		l2x0_flush_all();
312 		return;
313 	}
314 
315 	raw_spin_lock_irqsave(&l2x0_lock, flags);
316 	start &= ~(CACHE_LINE_SIZE - 1);
317 	while (start < end) {
318 		unsigned long blk_end = start + min(end - start, 4096UL);
319 
320 		debug_writel(0x03);
321 		while (start < blk_end) {
322 			l2x0_flush_line(start);
323 			start += CACHE_LINE_SIZE;
324 		}
325 		debug_writel(0x00);
326 
327 		if (blk_end < end) {
328 			raw_spin_unlock_irqrestore(&l2x0_lock, flags);
329 			raw_spin_lock_irqsave(&l2x0_lock, flags);
330 		}
331 	}
332 	cache_wait(base + L2X0_CLEAN_INV_LINE_PA, 1);
333 	cache_sync();
334 	raw_spin_unlock_irqrestore(&l2x0_lock, flags);
335 }
336 
337 static void l2x0_disable(void)
338 {
339 	unsigned long flags;
340 
341 	raw_spin_lock_irqsave(&l2x0_lock, flags);
342 	__l2x0_flush_all();
343 	writel_relaxed(0, l2x0_base + L2X0_CTRL);
344 	dsb(st);
345 	raw_spin_unlock_irqrestore(&l2x0_lock, flags);
346 }
347 
348 static void l2x0_enable(void __iomem *base, u32 aux, unsigned num_lock)
349 {
350 	unsigned id;
351 
352 	id = readl_relaxed(base + L2X0_CACHE_ID) & L2X0_CACHE_ID_PART_MASK;
353 	if (id == L2X0_CACHE_ID_PART_L310)
354 		num_lock = 8;
355 	else
356 		num_lock = 1;
357 
358 	/* l2x0 controller is disabled */
359 	writel_relaxed(aux, base + L2X0_AUX_CTRL);
360 
361 	/* Make sure that I&D is not locked down when starting */
362 	l2c_unlock(base, num_lock);
363 
364 	l2x0_inv_all();
365 
366 	/* enable L2X0 */
367 	writel_relaxed(L2X0_CTRL_EN, base + L2X0_CTRL);
368 }
369 
370 static void l2x0_resume(void)
371 {
372 	void __iomem *base = l2x0_base;
373 
374 	if (!(readl_relaxed(base + L2X0_CTRL) & L2X0_CTRL_EN))
375 		l2x0_enable(base, l2x0_saved_regs.aux_ctrl, 0);
376 }
377 
378 static const struct l2c_init_data l2x0_init_fns __initconst = {
379 	.enable = l2x0_enable,
380 	.outer_cache = {
381 		.inv_range = l2x0_inv_range,
382 		.clean_range = l2x0_clean_range,
383 		.flush_range = l2x0_flush_range,
384 		.flush_all = l2x0_flush_all,
385 		.disable = l2x0_disable,
386 		.sync = l2x0_cache_sync,
387 		.resume = l2x0_resume,
388 	},
389 };
390 
391 /*
392  * L2C-210 specific code.
393  *
394  * The L2C-2x0 PA, set/way and sync operations are atomic, but we must
395  * ensure that no background operation is running.  The way operations
396  * are all background tasks.
397  *
398  * While a background operation is in progress, any new operation is
399  * ignored (unspecified whether this causes an error.)  Thankfully, not
400  * used on SMP.
401  *
402  * Never has a different sync register other than L2X0_CACHE_SYNC, but
403  * we use sync_reg_offset here so we can share some of this with L2C-310.
404  */
405 static void __l2c210_cache_sync(void __iomem *base)
406 {
407 	writel_relaxed(0, base + sync_reg_offset);
408 }
409 
410 static void __l2c210_op_pa_range(void __iomem *reg, unsigned long start,
411 	unsigned long end)
412 {
413 	while (start < end) {
414 		writel_relaxed(start, reg);
415 		start += CACHE_LINE_SIZE;
416 	}
417 }
418 
419 static void l2c210_inv_range(unsigned long start, unsigned long end)
420 {
421 	void __iomem *base = l2x0_base;
422 
423 	if (start & (CACHE_LINE_SIZE - 1)) {
424 		start &= ~(CACHE_LINE_SIZE - 1);
425 		writel_relaxed(start, base + L2X0_CLEAN_INV_LINE_PA);
426 		start += CACHE_LINE_SIZE;
427 	}
428 
429 	if (end & (CACHE_LINE_SIZE - 1)) {
430 		end &= ~(CACHE_LINE_SIZE - 1);
431 		writel_relaxed(end, base + L2X0_CLEAN_INV_LINE_PA);
432 	}
433 
434 	__l2c210_op_pa_range(base + L2X0_INV_LINE_PA, start, end);
435 	__l2c210_cache_sync(base);
436 }
437 
438 static void l2c210_clean_range(unsigned long start, unsigned long end)
439 {
440 	void __iomem *base = l2x0_base;
441 
442 	start &= ~(CACHE_LINE_SIZE - 1);
443 	__l2c210_op_pa_range(base + L2X0_CLEAN_LINE_PA, start, end);
444 	__l2c210_cache_sync(base);
445 }
446 
447 static void l2c210_flush_range(unsigned long start, unsigned long end)
448 {
449 	void __iomem *base = l2x0_base;
450 
451 	start &= ~(CACHE_LINE_SIZE - 1);
452 	__l2c210_op_pa_range(base + L2X0_CLEAN_INV_LINE_PA, start, end);
453 	__l2c210_cache_sync(base);
454 }
455 
456 static void l2c210_flush_all(void)
457 {
458 	void __iomem *base = l2x0_base;
459 
460 	BUG_ON(!irqs_disabled());
461 
462 	__l2c_op_way(base + L2X0_CLEAN_INV_WAY);
463 	__l2c210_cache_sync(base);
464 }
465 
466 static void l2c210_sync(void)
467 {
468 	__l2c210_cache_sync(l2x0_base);
469 }
470 
471 static void l2c210_resume(void)
472 {
473 	void __iomem *base = l2x0_base;
474 
475 	if (!(readl_relaxed(base + L2X0_CTRL) & L2X0_CTRL_EN))
476 		l2c_enable(base, l2x0_saved_regs.aux_ctrl, 1);
477 }
478 
479 static const struct l2c_init_data l2c210_data __initconst = {
480 	.num_lock = 1,
481 	.enable = l2c_enable,
482 	.outer_cache = {
483 		.inv_range = l2c210_inv_range,
484 		.clean_range = l2c210_clean_range,
485 		.flush_range = l2c210_flush_range,
486 		.flush_all = l2c210_flush_all,
487 		.disable = l2c_disable,
488 		.sync = l2c210_sync,
489 		.resume = l2c210_resume,
490 	},
491 };
492 
493 /*
494  * L2C-310 specific code.
495  *
496  * Very similar to L2C-210, the PA, set/way and sync operations are atomic,
497  * and the way operations are all background tasks.  However, issuing an
498  * operation while a background operation is in progress results in a
499  * SLVERR response.  We can reuse:
500  *
501  *  __l2c210_cache_sync (using sync_reg_offset)
502  *  l2c210_sync
503  *  l2c210_inv_range (if 588369 is not applicable)
504  *  l2c210_clean_range
505  *  l2c210_flush_range (if 588369 is not applicable)
506  *  l2c210_flush_all (if 727915 is not applicable)
507  *
508  * Errata:
509  * 588369: PL310 R0P0->R1P0, fixed R2P0.
510  *	Affects: all clean+invalidate operations
511  *	clean and invalidate skips the invalidate step, so we need to issue
512  *	separate operations.  We also require the above debug workaround
513  *	enclosing this code fragment on affected parts.  On unaffected parts,
514  *	we must not use this workaround without the debug register writes
515  *	to avoid exposing a problem similar to 727915.
516  *
517  * 727915: PL310 R2P0->R3P0, fixed R3P1.
518  *	Affects: clean+invalidate by way
519  *	clean and invalidate by way runs in the background, and a store can
520  *	hit the line between the clean operation and invalidate operation,
521  *	resulting in the store being lost.
522  *
523  * 753970: PL310 R3P0, fixed R3P1.
524  *	Affects: sync
525  *	prevents merging writes after the sync operation, until another L2C
526  *	operation is performed (or a number of other conditions.)
527  *
528  * 769419: PL310 R0P0->R3P1, fixed R3P2.
529  *	Affects: store buffer
530  *	store buffer is not automatically drained.
531  */
532 static void l2c310_set_debug(unsigned long val)
533 {
534 	writel_relaxed(val, l2x0_base + L2X0_DEBUG_CTRL);
535 }
536 
537 static void l2c310_inv_range_erratum(unsigned long start, unsigned long end)
538 {
539 	void __iomem *base = l2x0_base;
540 
541 	if ((start | end) & (CACHE_LINE_SIZE - 1)) {
542 		unsigned long flags;
543 
544 		/* Erratum 588369 for both clean+invalidate operations */
545 		raw_spin_lock_irqsave(&l2x0_lock, flags);
546 		l2c_set_debug(base, 0x03);
547 
548 		if (start & (CACHE_LINE_SIZE - 1)) {
549 			start &= ~(CACHE_LINE_SIZE - 1);
550 			writel_relaxed(start, base + L2X0_CLEAN_LINE_PA);
551 			writel_relaxed(start, base + L2X0_INV_LINE_PA);
552 			start += CACHE_LINE_SIZE;
553 		}
554 
555 		if (end & (CACHE_LINE_SIZE - 1)) {
556 			end &= ~(CACHE_LINE_SIZE - 1);
557 			writel_relaxed(end, base + L2X0_CLEAN_LINE_PA);
558 			writel_relaxed(end, base + L2X0_INV_LINE_PA);
559 		}
560 
561 		l2c_set_debug(base, 0x00);
562 		raw_spin_unlock_irqrestore(&l2x0_lock, flags);
563 	}
564 
565 	__l2c210_op_pa_range(base + L2X0_INV_LINE_PA, start, end);
566 	__l2c210_cache_sync(base);
567 }
568 
569 static void l2c310_flush_range_erratum(unsigned long start, unsigned long end)
570 {
571 	raw_spinlock_t *lock = &l2x0_lock;
572 	unsigned long flags;
573 	void __iomem *base = l2x0_base;
574 
575 	raw_spin_lock_irqsave(lock, flags);
576 	while (start < end) {
577 		unsigned long blk_end = start + min(end - start, 4096UL);
578 
579 		l2c_set_debug(base, 0x03);
580 		while (start < blk_end) {
581 			writel_relaxed(start, base + L2X0_CLEAN_LINE_PA);
582 			writel_relaxed(start, base + L2X0_INV_LINE_PA);
583 			start += CACHE_LINE_SIZE;
584 		}
585 		l2c_set_debug(base, 0x00);
586 
587 		if (blk_end < end) {
588 			raw_spin_unlock_irqrestore(lock, flags);
589 			raw_spin_lock_irqsave(lock, flags);
590 		}
591 	}
592 	raw_spin_unlock_irqrestore(lock, flags);
593 	__l2c210_cache_sync(base);
594 }
595 
596 static void l2c310_flush_all_erratum(void)
597 {
598 	void __iomem *base = l2x0_base;
599 	unsigned long flags;
600 
601 	raw_spin_lock_irqsave(&l2x0_lock, flags);
602 	l2c_set_debug(base, 0x03);
603 	__l2c_op_way(base + L2X0_CLEAN_INV_WAY);
604 	l2c_set_debug(base, 0x00);
605 	__l2c210_cache_sync(base);
606 	raw_spin_unlock_irqrestore(&l2x0_lock, flags);
607 }
608 
609 static void __init l2c310_save(void __iomem *base)
610 {
611 	unsigned revision;
612 
613 	l2x0_saved_regs.tag_latency = readl_relaxed(base +
614 		L2X0_TAG_LATENCY_CTRL);
615 	l2x0_saved_regs.data_latency = readl_relaxed(base +
616 		L2X0_DATA_LATENCY_CTRL);
617 	l2x0_saved_regs.filter_end = readl_relaxed(base +
618 		L2X0_ADDR_FILTER_END);
619 	l2x0_saved_regs.filter_start = readl_relaxed(base +
620 		L2X0_ADDR_FILTER_START);
621 
622 	revision = readl_relaxed(base + L2X0_CACHE_ID) &
623 			L2X0_CACHE_ID_RTL_MASK;
624 
625 	/* From r2p0, there is Prefetch offset/control register */
626 	if (revision >= L310_CACHE_ID_RTL_R2P0)
627 		l2x0_saved_regs.prefetch_ctrl = readl_relaxed(base +
628 							L2X0_PREFETCH_CTRL);
629 
630 	/* From r3p0, there is Power control register */
631 	if (revision >= L310_CACHE_ID_RTL_R3P0)
632 		l2x0_saved_regs.pwr_ctrl = readl_relaxed(base +
633 							L2X0_POWER_CTRL);
634 }
635 
636 static void l2c310_resume(void)
637 {
638 	void __iomem *base = l2x0_base;
639 
640 	if (!(readl_relaxed(base + L2X0_CTRL) & L2X0_CTRL_EN)) {
641 		unsigned revision;
642 
643 		/* restore pl310 setup */
644 		writel_relaxed(l2x0_saved_regs.tag_latency,
645 			       base + L2X0_TAG_LATENCY_CTRL);
646 		writel_relaxed(l2x0_saved_regs.data_latency,
647 			       base + L2X0_DATA_LATENCY_CTRL);
648 		writel_relaxed(l2x0_saved_regs.filter_end,
649 			       base + L2X0_ADDR_FILTER_END);
650 		writel_relaxed(l2x0_saved_regs.filter_start,
651 			       base + L2X0_ADDR_FILTER_START);
652 
653 		revision = readl_relaxed(base + L2X0_CACHE_ID) &
654 				L2X0_CACHE_ID_RTL_MASK;
655 
656 		if (revision >= L310_CACHE_ID_RTL_R2P0)
657 			writel_relaxed(l2x0_saved_regs.prefetch_ctrl,
658 				       base + L2X0_PREFETCH_CTRL);
659 		if (revision >= L310_CACHE_ID_RTL_R3P0)
660 			writel_relaxed(l2x0_saved_regs.pwr_ctrl,
661 				       base + L2X0_POWER_CTRL);
662 
663 		l2c_enable(base, l2x0_saved_regs.aux_ctrl, 8);
664 	}
665 }
666 
667 static void __init l2c310_fixup(void __iomem *base, u32 cache_id,
668 	struct outer_cache_fns *fns)
669 {
670 	unsigned revision = cache_id & L2X0_CACHE_ID_RTL_MASK;
671 	const char *errata[4];
672 	unsigned n = 0;
673 
674 	/* For compatibility */
675 	if (revision <= L310_CACHE_ID_RTL_R3P0)
676 		fns->set_debug = l2c310_set_debug;
677 
678 	if (IS_ENABLED(CONFIG_PL310_ERRATA_588369) &&
679 	    revision < L310_CACHE_ID_RTL_R2P0 &&
680 	    /* For bcm compatibility */
681 	    fns->inv_range == l2c210_inv_range) {
682 		fns->inv_range = l2c310_inv_range_erratum;
683 		fns->flush_range = l2c310_flush_range_erratum;
684 		errata[n++] = "588369";
685 	}
686 
687 	if (IS_ENABLED(CONFIG_PL310_ERRATA_727915) &&
688 	    revision >= L310_CACHE_ID_RTL_R2P0 &&
689 	    revision < L310_CACHE_ID_RTL_R3P1) {
690 		fns->flush_all = l2c310_flush_all_erratum;
691 		errata[n++] = "727915";
692 	}
693 
694 	if (IS_ENABLED(CONFIG_PL310_ERRATA_753970) &&
695 	    revision == L310_CACHE_ID_RTL_R3P0) {
696 		sync_reg_offset = L2X0_DUMMY_REG;
697 		errata[n++] = "753970";
698 	}
699 
700 	if (IS_ENABLED(CONFIG_PL310_ERRATA_769419))
701 		errata[n++] = "769419";
702 
703 	if (n) {
704 		unsigned i;
705 
706 		pr_info("L2C-310 errat%s", n > 1 ? "a" : "um");
707 		for (i = 0; i < n; i++)
708 			pr_cont(" %s", errata[i]);
709 		pr_cont(" enabled\n");
710 	}
711 }
712 
713 static const struct l2c_init_data l2c310_init_fns __initconst = {
714 	.num_lock = 8,
715 	.enable = l2c_enable,
716 	.fixup = l2c310_fixup,
717 	.save = l2c310_save,
718 	.outer_cache = {
719 		.inv_range = l2c210_inv_range,
720 		.clean_range = l2c210_clean_range,
721 		.flush_range = l2c210_flush_range,
722 		.flush_all = l2c210_flush_all,
723 		.disable = l2c_disable,
724 		.sync = l2c210_sync,
725 		.set_debug = l2c310_set_debug,
726 		.resume = l2c310_resume,
727 	},
728 };
729 
730 static void __init __l2c_init(const struct l2c_init_data *data,
731 	u32 aux_val, u32 aux_mask, u32 cache_id)
732 {
733 	struct outer_cache_fns fns;
734 	u32 aux;
735 	u32 way_size = 0;
736 	int ways;
737 	int way_size_shift = L2X0_WAY_SIZE_SHIFT;
738 	const char *type;
739 
740 	/*
741 	 * It is strange to save the register state before initialisation,
742 	 * but hey, this is what the DT implementations decided to do.
743 	 */
744 	if (data->save)
745 		data->save(l2x0_base);
746 
747 	aux = readl_relaxed(l2x0_base + L2X0_AUX_CTRL);
748 
749 	aux &= aux_mask;
750 	aux |= aux_val;
751 
752 	/* Determine the number of ways */
753 	switch (cache_id & L2X0_CACHE_ID_PART_MASK) {
754 	case L2X0_CACHE_ID_PART_L310:
755 		if (aux & (1 << 16))
756 			ways = 16;
757 		else
758 			ways = 8;
759 		type = "L310";
760 		break;
761 
762 	case L2X0_CACHE_ID_PART_L210:
763 		ways = (aux >> 13) & 0xf;
764 		type = "L210";
765 		break;
766 
767 	case AURORA_CACHE_ID:
768 		ways = (aux >> 13) & 0xf;
769 		ways = 2 << ((ways + 1) >> 2);
770 		way_size_shift = AURORA_WAY_SIZE_SHIFT;
771 		type = "Aurora";
772 		break;
773 
774 	default:
775 		/* Assume unknown chips have 8 ways */
776 		ways = 8;
777 		type = "L2x0 series";
778 		break;
779 	}
780 
781 	l2x0_way_mask = (1 << ways) - 1;
782 
783 	/*
784 	 * L2 cache Size =  Way size * Number of ways
785 	 */
786 	way_size = (aux & L2X0_AUX_CTRL_WAY_SIZE_MASK) >> 17;
787 	way_size = 1 << (way_size + way_size_shift);
788 
789 	l2x0_size = ways * way_size * SZ_1K;
790 
791 	fns = data->outer_cache;
792 	if (data->fixup)
793 		data->fixup(l2x0_base, cache_id, &fns);
794 
795 	/*
796 	 * Check if l2x0 controller is already enabled.  If we are booting
797 	 * in non-secure mode accessing the below registers will fault.
798 	 */
799 	if (!(readl_relaxed(l2x0_base + L2X0_CTRL) & L2X0_CTRL_EN))
800 		data->enable(l2x0_base, aux, data->num_lock);
801 
802 	/* Re-read it in case some bits are reserved. */
803 	aux = readl_relaxed(l2x0_base + L2X0_AUX_CTRL);
804 
805 	/* Save the value for resuming. */
806 	l2x0_saved_regs.aux_ctrl = aux;
807 
808 	outer_cache = fns;
809 
810 	pr_info("%s cache controller enabled, %d ways, %d kB\n",
811 		type, ways, l2x0_size >> 10);
812 	pr_info("%s: CACHE_ID 0x%08x, AUX_CTRL 0x%08x\n",
813 		type, cache_id, aux);
814 }
815 
816 void __init l2x0_init(void __iomem *base, u32 aux_val, u32 aux_mask)
817 {
818 	const struct l2c_init_data *data;
819 	u32 cache_id;
820 
821 	l2x0_base = base;
822 
823 	cache_id = readl_relaxed(base + L2X0_CACHE_ID);
824 
825 	switch (cache_id & L2X0_CACHE_ID_PART_MASK) {
826 	default:
827 		data = &l2x0_init_fns;
828 		break;
829 
830 	case L2X0_CACHE_ID_PART_L210:
831 		data = &l2c210_data;
832 		break;
833 
834 	case L2X0_CACHE_ID_PART_L310:
835 		data = &l2c310_init_fns;
836 		break;
837 	}
838 
839 	__l2c_init(data, aux_val, aux_mask, cache_id);
840 }
841 
842 #ifdef CONFIG_OF
843 static int l2_wt_override;
844 
845 /* Aurora don't have the cache ID register available, so we have to
846  * pass it though the device tree */
847 static u32 cache_id_part_number_from_dt;
848 
849 static void __init l2x0_of_parse(const struct device_node *np,
850 				 u32 *aux_val, u32 *aux_mask)
851 {
852 	u32 data[2] = { 0, 0 };
853 	u32 tag = 0;
854 	u32 dirty = 0;
855 	u32 val = 0, mask = 0;
856 
857 	of_property_read_u32(np, "arm,tag-latency", &tag);
858 	if (tag) {
859 		mask |= L2X0_AUX_CTRL_TAG_LATENCY_MASK;
860 		val |= (tag - 1) << L2X0_AUX_CTRL_TAG_LATENCY_SHIFT;
861 	}
862 
863 	of_property_read_u32_array(np, "arm,data-latency",
864 				   data, ARRAY_SIZE(data));
865 	if (data[0] && data[1]) {
866 		mask |= L2X0_AUX_CTRL_DATA_RD_LATENCY_MASK |
867 			L2X0_AUX_CTRL_DATA_WR_LATENCY_MASK;
868 		val |= ((data[0] - 1) << L2X0_AUX_CTRL_DATA_RD_LATENCY_SHIFT) |
869 		       ((data[1] - 1) << L2X0_AUX_CTRL_DATA_WR_LATENCY_SHIFT);
870 	}
871 
872 	of_property_read_u32(np, "arm,dirty-latency", &dirty);
873 	if (dirty) {
874 		mask |= L2X0_AUX_CTRL_DIRTY_LATENCY_MASK;
875 		val |= (dirty - 1) << L2X0_AUX_CTRL_DIRTY_LATENCY_SHIFT;
876 	}
877 
878 	*aux_val &= ~mask;
879 	*aux_val |= val;
880 	*aux_mask &= ~mask;
881 }
882 
883 static const struct l2c_init_data of_l2c210_data __initconst = {
884 	.num_lock = 1,
885 	.of_parse = l2x0_of_parse,
886 	.enable = l2c_enable,
887 	.outer_cache = {
888 		.inv_range   = l2c210_inv_range,
889 		.clean_range = l2c210_clean_range,
890 		.flush_range = l2c210_flush_range,
891 		.flush_all   = l2c210_flush_all,
892 		.disable     = l2c_disable,
893 		.sync        = l2c210_sync,
894 		.resume      = l2c210_resume,
895 	},
896 };
897 
898 static const struct l2c_init_data of_l2x0_data __initconst = {
899 	.of_parse = l2x0_of_parse,
900 	.enable = l2x0_enable,
901 	.outer_cache = {
902 		.inv_range   = l2x0_inv_range,
903 		.clean_range = l2x0_clean_range,
904 		.flush_range = l2x0_flush_range,
905 		.flush_all   = l2x0_flush_all,
906 		.disable     = l2x0_disable,
907 		.sync        = l2x0_cache_sync,
908 		.resume      = l2x0_resume,
909 	},
910 };
911 
912 static void __init l2c310_of_parse(const struct device_node *np,
913 	u32 *aux_val, u32 *aux_mask)
914 {
915 	u32 data[3] = { 0, 0, 0 };
916 	u32 tag[3] = { 0, 0, 0 };
917 	u32 filter[2] = { 0, 0 };
918 
919 	of_property_read_u32_array(np, "arm,tag-latency", tag, ARRAY_SIZE(tag));
920 	if (tag[0] && tag[1] && tag[2])
921 		writel_relaxed(
922 			((tag[0] - 1) << L2X0_LATENCY_CTRL_RD_SHIFT) |
923 			((tag[1] - 1) << L2X0_LATENCY_CTRL_WR_SHIFT) |
924 			((tag[2] - 1) << L2X0_LATENCY_CTRL_SETUP_SHIFT),
925 			l2x0_base + L2X0_TAG_LATENCY_CTRL);
926 
927 	of_property_read_u32_array(np, "arm,data-latency",
928 				   data, ARRAY_SIZE(data));
929 	if (data[0] && data[1] && data[2])
930 		writel_relaxed(
931 			((data[0] - 1) << L2X0_LATENCY_CTRL_RD_SHIFT) |
932 			((data[1] - 1) << L2X0_LATENCY_CTRL_WR_SHIFT) |
933 			((data[2] - 1) << L2X0_LATENCY_CTRL_SETUP_SHIFT),
934 			l2x0_base + L2X0_DATA_LATENCY_CTRL);
935 
936 	of_property_read_u32_array(np, "arm,filter-ranges",
937 				   filter, ARRAY_SIZE(filter));
938 	if (filter[1]) {
939 		writel_relaxed(ALIGN(filter[0] + filter[1], SZ_1M),
940 			       l2x0_base + L2X0_ADDR_FILTER_END);
941 		writel_relaxed((filter[0] & ~(SZ_1M - 1)) | L2X0_ADDR_FILTER_EN,
942 			       l2x0_base + L2X0_ADDR_FILTER_START);
943 	}
944 }
945 
946 static const struct l2c_init_data of_l2c310_data __initconst = {
947 	.num_lock = 8,
948 	.of_parse = l2c310_of_parse,
949 	.enable = l2c_enable,
950 	.fixup = l2c310_fixup,
951 	.save  = l2c310_save,
952 	.outer_cache = {
953 		.inv_range   = l2c210_inv_range,
954 		.clean_range = l2c210_clean_range,
955 		.flush_range = l2c210_flush_range,
956 		.flush_all   = l2c210_flush_all,
957 		.disable     = l2c_disable,
958 		.sync        = l2c210_sync,
959 		.set_debug   = l2c310_set_debug,
960 		.resume      = l2c310_resume,
961 	},
962 };
963 
964 /*
965  * Note that the end addresses passed to Linux primitives are
966  * noninclusive, while the hardware cache range operations use
967  * inclusive start and end addresses.
968  */
969 static unsigned long calc_range_end(unsigned long start, unsigned long end)
970 {
971 	/*
972 	 * Limit the number of cache lines processed at once,
973 	 * since cache range operations stall the CPU pipeline
974 	 * until completion.
975 	 */
976 	if (end > start + MAX_RANGE_SIZE)
977 		end = start + MAX_RANGE_SIZE;
978 
979 	/*
980 	 * Cache range operations can't straddle a page boundary.
981 	 */
982 	if (end > PAGE_ALIGN(start+1))
983 		end = PAGE_ALIGN(start+1);
984 
985 	return end;
986 }
987 
988 /*
989  * Make sure 'start' and 'end' reference the same page, as L2 is PIPT
990  * and range operations only do a TLB lookup on the start address.
991  */
992 static void aurora_pa_range(unsigned long start, unsigned long end,
993 			unsigned long offset)
994 {
995 	unsigned long flags;
996 
997 	raw_spin_lock_irqsave(&l2x0_lock, flags);
998 	writel_relaxed(start, l2x0_base + AURORA_RANGE_BASE_ADDR_REG);
999 	writel_relaxed(end, l2x0_base + offset);
1000 	raw_spin_unlock_irqrestore(&l2x0_lock, flags);
1001 
1002 	cache_sync();
1003 }
1004 
1005 static void aurora_inv_range(unsigned long start, unsigned long end)
1006 {
1007 	/*
1008 	 * round start and end adresses up to cache line size
1009 	 */
1010 	start &= ~(CACHE_LINE_SIZE - 1);
1011 	end = ALIGN(end, CACHE_LINE_SIZE);
1012 
1013 	/*
1014 	 * Invalidate all full cache lines between 'start' and 'end'.
1015 	 */
1016 	while (start < end) {
1017 		unsigned long range_end = calc_range_end(start, end);
1018 		aurora_pa_range(start, range_end - CACHE_LINE_SIZE,
1019 				AURORA_INVAL_RANGE_REG);
1020 		start = range_end;
1021 	}
1022 }
1023 
1024 static void aurora_clean_range(unsigned long start, unsigned long end)
1025 {
1026 	/*
1027 	 * If L2 is forced to WT, the L2 will always be clean and we
1028 	 * don't need to do anything here.
1029 	 */
1030 	if (!l2_wt_override) {
1031 		start &= ~(CACHE_LINE_SIZE - 1);
1032 		end = ALIGN(end, CACHE_LINE_SIZE);
1033 		while (start != end) {
1034 			unsigned long range_end = calc_range_end(start, end);
1035 			aurora_pa_range(start, range_end - CACHE_LINE_SIZE,
1036 					AURORA_CLEAN_RANGE_REG);
1037 			start = range_end;
1038 		}
1039 	}
1040 }
1041 
1042 static void aurora_flush_range(unsigned long start, unsigned long end)
1043 {
1044 	start &= ~(CACHE_LINE_SIZE - 1);
1045 	end = ALIGN(end, CACHE_LINE_SIZE);
1046 	while (start != end) {
1047 		unsigned long range_end = calc_range_end(start, end);
1048 		/*
1049 		 * If L2 is forced to WT, the L2 will always be clean and we
1050 		 * just need to invalidate.
1051 		 */
1052 		if (l2_wt_override)
1053 			aurora_pa_range(start, range_end - CACHE_LINE_SIZE,
1054 							AURORA_INVAL_RANGE_REG);
1055 		else
1056 			aurora_pa_range(start, range_end - CACHE_LINE_SIZE,
1057 							AURORA_FLUSH_RANGE_REG);
1058 		start = range_end;
1059 	}
1060 }
1061 
1062 static void aurora_save(void __iomem *base)
1063 {
1064 	l2x0_saved_regs.ctrl = readl_relaxed(base + L2X0_CTRL);
1065 	l2x0_saved_regs.aux_ctrl = readl_relaxed(base + L2X0_AUX_CTRL);
1066 }
1067 
1068 static void aurora_resume(void)
1069 {
1070 	void __iomem *base = l2x0_base;
1071 
1072 	if (!(readl(base + L2X0_CTRL) & L2X0_CTRL_EN)) {
1073 		writel_relaxed(l2x0_saved_regs.aux_ctrl, base + L2X0_AUX_CTRL);
1074 		writel_relaxed(l2x0_saved_regs.ctrl, base + L2X0_CTRL);
1075 	}
1076 }
1077 
1078 /*
1079  * For Aurora cache in no outer mode, enable via the CP15 coprocessor
1080  * broadcasting of cache commands to L2.
1081  */
1082 static void __init aurora_enable_no_outer(void __iomem *base, u32 aux,
1083 	unsigned num_lock)
1084 {
1085 	u32 u;
1086 
1087 	asm volatile("mrc p15, 1, %0, c15, c2, 0" : "=r" (u));
1088 	u |= AURORA_CTRL_FW;		/* Set the FW bit */
1089 	asm volatile("mcr p15, 1, %0, c15, c2, 0" : : "r" (u));
1090 
1091 	isb();
1092 
1093 	l2c_enable(base, aux, num_lock);
1094 }
1095 
1096 static void __init aurora_fixup(void __iomem *base, u32 cache_id,
1097 	struct outer_cache_fns *fns)
1098 {
1099 	sync_reg_offset = AURORA_SYNC_REG;
1100 }
1101 
1102 static void __init aurora_of_parse(const struct device_node *np,
1103 				u32 *aux_val, u32 *aux_mask)
1104 {
1105 	u32 val = AURORA_ACR_REPLACEMENT_TYPE_SEMIPLRU;
1106 	u32 mask =  AURORA_ACR_REPLACEMENT_MASK;
1107 
1108 	of_property_read_u32(np, "cache-id-part",
1109 			&cache_id_part_number_from_dt);
1110 
1111 	/* Determine and save the write policy */
1112 	l2_wt_override = of_property_read_bool(np, "wt-override");
1113 
1114 	if (l2_wt_override) {
1115 		val |= AURORA_ACR_FORCE_WRITE_THRO_POLICY;
1116 		mask |= AURORA_ACR_FORCE_WRITE_POLICY_MASK;
1117 	}
1118 
1119 	*aux_val &= ~mask;
1120 	*aux_val |= val;
1121 	*aux_mask &= ~mask;
1122 }
1123 
1124 static const struct l2c_init_data of_aurora_with_outer_data __initconst = {
1125 	.num_lock = 4,
1126 	.of_parse = aurora_of_parse,
1127 	.enable = l2c_enable,
1128 	.fixup = aurora_fixup,
1129 	.save  = aurora_save,
1130 	.outer_cache = {
1131 		.inv_range   = aurora_inv_range,
1132 		.clean_range = aurora_clean_range,
1133 		.flush_range = aurora_flush_range,
1134 		.flush_all   = l2x0_flush_all,
1135 		.disable     = l2x0_disable,
1136 		.sync        = l2x0_cache_sync,
1137 		.resume      = aurora_resume,
1138 	},
1139 };
1140 
1141 static const struct l2c_init_data of_aurora_no_outer_data __initconst = {
1142 	.num_lock = 4,
1143 	.of_parse = aurora_of_parse,
1144 	.enable = aurora_enable_no_outer,
1145 	.fixup = aurora_fixup,
1146 	.save  = aurora_save,
1147 	.outer_cache = {
1148 		.resume      = aurora_resume,
1149 	},
1150 };
1151 
1152 /*
1153  * For certain Broadcom SoCs, depending on the address range, different offsets
1154  * need to be added to the address before passing it to L2 for
1155  * invalidation/clean/flush
1156  *
1157  * Section Address Range              Offset        EMI
1158  *   1     0x00000000 - 0x3FFFFFFF    0x80000000    VC
1159  *   2     0x40000000 - 0xBFFFFFFF    0x40000000    SYS
1160  *   3     0xC0000000 - 0xFFFFFFFF    0x80000000    VC
1161  *
1162  * When the start and end addresses have crossed two different sections, we
1163  * need to break the L2 operation into two, each within its own section.
1164  * For example, if we need to invalidate addresses starts at 0xBFFF0000 and
1165  * ends at 0xC0001000, we need do invalidate 1) 0xBFFF0000 - 0xBFFFFFFF and 2)
1166  * 0xC0000000 - 0xC0001000
1167  *
1168  * Note 1:
1169  * By breaking a single L2 operation into two, we may potentially suffer some
1170  * performance hit, but keep in mind the cross section case is very rare
1171  *
1172  * Note 2:
1173  * We do not need to handle the case when the start address is in
1174  * Section 1 and the end address is in Section 3, since it is not a valid use
1175  * case
1176  *
1177  * Note 3:
1178  * Section 1 in practical terms can no longer be used on rev A2. Because of
1179  * that the code does not need to handle section 1 at all.
1180  *
1181  */
1182 #define BCM_SYS_EMI_START_ADDR        0x40000000UL
1183 #define BCM_VC_EMI_SEC3_START_ADDR    0xC0000000UL
1184 
1185 #define BCM_SYS_EMI_OFFSET            0x40000000UL
1186 #define BCM_VC_EMI_OFFSET             0x80000000UL
1187 
1188 static inline int bcm_addr_is_sys_emi(unsigned long addr)
1189 {
1190 	return (addr >= BCM_SYS_EMI_START_ADDR) &&
1191 		(addr < BCM_VC_EMI_SEC3_START_ADDR);
1192 }
1193 
1194 static inline unsigned long bcm_l2_phys_addr(unsigned long addr)
1195 {
1196 	if (bcm_addr_is_sys_emi(addr))
1197 		return addr + BCM_SYS_EMI_OFFSET;
1198 	else
1199 		return addr + BCM_VC_EMI_OFFSET;
1200 }
1201 
1202 static void bcm_inv_range(unsigned long start, unsigned long end)
1203 {
1204 	unsigned long new_start, new_end;
1205 
1206 	BUG_ON(start < BCM_SYS_EMI_START_ADDR);
1207 
1208 	if (unlikely(end <= start))
1209 		return;
1210 
1211 	new_start = bcm_l2_phys_addr(start);
1212 	new_end = bcm_l2_phys_addr(end);
1213 
1214 	/* normal case, no cross section between start and end */
1215 	if (likely(bcm_addr_is_sys_emi(end) || !bcm_addr_is_sys_emi(start))) {
1216 		l2x0_inv_range(new_start, new_end);
1217 		return;
1218 	}
1219 
1220 	/* They cross sections, so it can only be a cross from section
1221 	 * 2 to section 3
1222 	 */
1223 	l2x0_inv_range(new_start,
1224 		bcm_l2_phys_addr(BCM_VC_EMI_SEC3_START_ADDR-1));
1225 	l2x0_inv_range(bcm_l2_phys_addr(BCM_VC_EMI_SEC3_START_ADDR),
1226 		new_end);
1227 }
1228 
1229 static void bcm_clean_range(unsigned long start, unsigned long end)
1230 {
1231 	unsigned long new_start, new_end;
1232 
1233 	BUG_ON(start < BCM_SYS_EMI_START_ADDR);
1234 
1235 	if (unlikely(end <= start))
1236 		return;
1237 
1238 	if ((end - start) >= l2x0_size) {
1239 		l2x0_clean_all();
1240 		return;
1241 	}
1242 
1243 	new_start = bcm_l2_phys_addr(start);
1244 	new_end = bcm_l2_phys_addr(end);
1245 
1246 	/* normal case, no cross section between start and end */
1247 	if (likely(bcm_addr_is_sys_emi(end) || !bcm_addr_is_sys_emi(start))) {
1248 		l2x0_clean_range(new_start, new_end);
1249 		return;
1250 	}
1251 
1252 	/* They cross sections, so it can only be a cross from section
1253 	 * 2 to section 3
1254 	 */
1255 	l2x0_clean_range(new_start,
1256 		bcm_l2_phys_addr(BCM_VC_EMI_SEC3_START_ADDR-1));
1257 	l2x0_clean_range(bcm_l2_phys_addr(BCM_VC_EMI_SEC3_START_ADDR),
1258 		new_end);
1259 }
1260 
1261 static void bcm_flush_range(unsigned long start, unsigned long end)
1262 {
1263 	unsigned long new_start, new_end;
1264 
1265 	BUG_ON(start < BCM_SYS_EMI_START_ADDR);
1266 
1267 	if (unlikely(end <= start))
1268 		return;
1269 
1270 	if ((end - start) >= l2x0_size) {
1271 		l2x0_flush_all();
1272 		return;
1273 	}
1274 
1275 	new_start = bcm_l2_phys_addr(start);
1276 	new_end = bcm_l2_phys_addr(end);
1277 
1278 	/* normal case, no cross section between start and end */
1279 	if (likely(bcm_addr_is_sys_emi(end) || !bcm_addr_is_sys_emi(start))) {
1280 		l2x0_flush_range(new_start, new_end);
1281 		return;
1282 	}
1283 
1284 	/* They cross sections, so it can only be a cross from section
1285 	 * 2 to section 3
1286 	 */
1287 	l2x0_flush_range(new_start,
1288 		bcm_l2_phys_addr(BCM_VC_EMI_SEC3_START_ADDR-1));
1289 	l2x0_flush_range(bcm_l2_phys_addr(BCM_VC_EMI_SEC3_START_ADDR),
1290 		new_end);
1291 }
1292 
1293 static const struct l2c_init_data of_bcm_l2x0_data __initconst = {
1294 	.num_lock = 8,
1295 	.of_parse = l2c310_of_parse,
1296 	.enable = l2c_enable,
1297 	.fixup = l2c310_fixup,
1298 	.save  = l2c310_save,
1299 	.outer_cache = {
1300 		.inv_range   = bcm_inv_range,
1301 		.clean_range = bcm_clean_range,
1302 		.flush_range = bcm_flush_range,
1303 		.flush_all   = l2c210_flush_all,
1304 		.disable     = l2c_disable,
1305 		.sync        = l2c210_sync,
1306 		.resume      = l2c310_resume,
1307 	},
1308 };
1309 
1310 static void __init tauros3_save(void __iomem *base)
1311 {
1312 	l2x0_saved_regs.aux2_ctrl =
1313 		readl_relaxed(base + TAUROS3_AUX2_CTRL);
1314 	l2x0_saved_regs.prefetch_ctrl =
1315 		readl_relaxed(base + L2X0_PREFETCH_CTRL);
1316 }
1317 
1318 static void tauros3_resume(void)
1319 {
1320 	void __iomem *base = l2x0_base;
1321 
1322 	if (!(readl_relaxed(base + L2X0_CTRL) & L2X0_CTRL_EN)) {
1323 		writel_relaxed(l2x0_saved_regs.aux2_ctrl,
1324 			       base + TAUROS3_AUX2_CTRL);
1325 		writel_relaxed(l2x0_saved_regs.prefetch_ctrl,
1326 			       base + L2X0_PREFETCH_CTRL);
1327 
1328 		l2c_enable(base, l2x0_saved_regs.aux_ctrl, 8);
1329 	}
1330 }
1331 
1332 static const struct l2c_init_data of_tauros3_data __initconst = {
1333 	.num_lock = 8,
1334 	.enable = l2c_enable,
1335 	.save  = tauros3_save,
1336 	/* Tauros3 broadcasts L1 cache operations to L2 */
1337 	.outer_cache = {
1338 		.resume      = tauros3_resume,
1339 	},
1340 };
1341 
1342 #define L2C_ID(name, fns) { .compatible = name, .data = (void *)&fns }
1343 static const struct of_device_id l2x0_ids[] __initconst = {
1344 	L2C_ID("arm,l210-cache", of_l2c210_data),
1345 	L2C_ID("arm,l220-cache", of_l2x0_data),
1346 	L2C_ID("arm,pl310-cache", of_l2c310_data),
1347 	L2C_ID("brcm,bcm11351-a2-pl310-cache", of_bcm_l2x0_data),
1348 	L2C_ID("marvell,aurora-outer-cache", of_aurora_with_outer_data),
1349 	L2C_ID("marvell,aurora-system-cache", of_aurora_no_outer_data),
1350 	L2C_ID("marvell,tauros3-cache", of_tauros3_data),
1351 	/* Deprecated IDs */
1352 	L2C_ID("bcm,bcm11351-a2-pl310-cache", of_bcm_l2x0_data),
1353 	{}
1354 };
1355 
1356 int __init l2x0_of_init(u32 aux_val, u32 aux_mask)
1357 {
1358 	const struct l2c_init_data *data;
1359 	struct device_node *np;
1360 	struct resource res;
1361 	u32 cache_id;
1362 
1363 	np = of_find_matching_node(NULL, l2x0_ids);
1364 	if (!np)
1365 		return -ENODEV;
1366 
1367 	if (of_address_to_resource(np, 0, &res))
1368 		return -ENODEV;
1369 
1370 	l2x0_base = ioremap(res.start, resource_size(&res));
1371 	if (!l2x0_base)
1372 		return -ENOMEM;
1373 
1374 	l2x0_saved_regs.phy_base = res.start;
1375 
1376 	data = of_match_node(l2x0_ids, np)->data;
1377 
1378 	/* L2 configuration can only be changed if the cache is disabled */
1379 	if (!(readl_relaxed(l2x0_base + L2X0_CTRL) & L2X0_CTRL_EN))
1380 		if (data->of_parse)
1381 			data->of_parse(np, &aux_val, &aux_mask);
1382 
1383 	if (cache_id_part_number_from_dt)
1384 		cache_id = cache_id_part_number_from_dt;
1385 	else
1386 		cache_id = readl_relaxed(l2x0_base + L2X0_CACHE_ID);
1387 
1388 	__l2c_init(data, aux_val, aux_mask, cache_id);
1389 
1390 	return 0;
1391 }
1392 #endif
1393