xref: /openbmc/linux/arch/arm/mm/cache-l2x0.c (revision 505caa66)
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-aurora-l2.h"
29 
30 #define CACHE_LINE_SIZE		32
31 
32 static void __iomem *l2x0_base;
33 static DEFINE_RAW_SPINLOCK(l2x0_lock);
34 static u32 l2x0_way_mask;	/* Bitmask of active ways */
35 static u32 l2x0_size;
36 static unsigned long sync_reg_offset = L2X0_CACHE_SYNC;
37 
38 /* Aurora don't have the cache ID register available, so we have to
39  * pass it though the device tree */
40 static u32  cache_id_part_number_from_dt;
41 
42 struct l2x0_regs l2x0_saved_regs;
43 
44 struct l2x0_of_data {
45 	void (*setup)(const struct device_node *, u32 *, u32 *);
46 	void (*save)(void);
47 	struct outer_cache_fns outer_cache;
48 };
49 
50 static bool of_init = false;
51 
52 static inline void cache_wait_way(void __iomem *reg, unsigned long mask)
53 {
54 	/* wait for cache operation by line or way to complete */
55 	while (readl_relaxed(reg) & mask)
56 		cpu_relax();
57 }
58 
59 #ifdef CONFIG_CACHE_PL310
60 static inline void cache_wait(void __iomem *reg, unsigned long mask)
61 {
62 	/* cache operations by line are atomic on PL310 */
63 }
64 #else
65 #define cache_wait	cache_wait_way
66 #endif
67 
68 static inline void cache_sync(void)
69 {
70 	void __iomem *base = l2x0_base;
71 
72 	writel_relaxed(0, base + sync_reg_offset);
73 	cache_wait(base + L2X0_CACHE_SYNC, 1);
74 }
75 
76 static inline void l2x0_clean_line(unsigned long addr)
77 {
78 	void __iomem *base = l2x0_base;
79 	cache_wait(base + L2X0_CLEAN_LINE_PA, 1);
80 	writel_relaxed(addr, base + L2X0_CLEAN_LINE_PA);
81 }
82 
83 static inline void l2x0_inv_line(unsigned long addr)
84 {
85 	void __iomem *base = l2x0_base;
86 	cache_wait(base + L2X0_INV_LINE_PA, 1);
87 	writel_relaxed(addr, base + L2X0_INV_LINE_PA);
88 }
89 
90 #if defined(CONFIG_PL310_ERRATA_588369) || defined(CONFIG_PL310_ERRATA_727915)
91 static inline void debug_writel(unsigned long val)
92 {
93 	if (outer_cache.set_debug)
94 		outer_cache.set_debug(val);
95 }
96 
97 static void pl310_set_debug(unsigned long val)
98 {
99 	writel_relaxed(val, l2x0_base + L2X0_DEBUG_CTRL);
100 }
101 #else
102 /* Optimised out for non-errata case */
103 static inline void debug_writel(unsigned long val)
104 {
105 }
106 
107 #define pl310_set_debug	NULL
108 #endif
109 
110 #ifdef CONFIG_PL310_ERRATA_588369
111 static inline void l2x0_flush_line(unsigned long addr)
112 {
113 	void __iomem *base = l2x0_base;
114 
115 	/* Clean by PA followed by Invalidate by PA */
116 	cache_wait(base + L2X0_CLEAN_LINE_PA, 1);
117 	writel_relaxed(addr, base + L2X0_CLEAN_LINE_PA);
118 	cache_wait(base + L2X0_INV_LINE_PA, 1);
119 	writel_relaxed(addr, base + L2X0_INV_LINE_PA);
120 }
121 #else
122 
123 static inline void l2x0_flush_line(unsigned long addr)
124 {
125 	void __iomem *base = l2x0_base;
126 	cache_wait(base + L2X0_CLEAN_INV_LINE_PA, 1);
127 	writel_relaxed(addr, base + L2X0_CLEAN_INV_LINE_PA);
128 }
129 #endif
130 
131 static void l2x0_cache_sync(void)
132 {
133 	unsigned long flags;
134 
135 	raw_spin_lock_irqsave(&l2x0_lock, flags);
136 	cache_sync();
137 	raw_spin_unlock_irqrestore(&l2x0_lock, flags);
138 }
139 
140 static void __l2x0_flush_all(void)
141 {
142 	debug_writel(0x03);
143 	writel_relaxed(l2x0_way_mask, l2x0_base + L2X0_CLEAN_INV_WAY);
144 	cache_wait_way(l2x0_base + L2X0_CLEAN_INV_WAY, l2x0_way_mask);
145 	cache_sync();
146 	debug_writel(0x00);
147 }
148 
149 static void l2x0_flush_all(void)
150 {
151 	unsigned long flags;
152 
153 	/* clean all ways */
154 	raw_spin_lock_irqsave(&l2x0_lock, flags);
155 	__l2x0_flush_all();
156 	raw_spin_unlock_irqrestore(&l2x0_lock, flags);
157 }
158 
159 static void l2x0_clean_all(void)
160 {
161 	unsigned long flags;
162 
163 	/* clean all ways */
164 	raw_spin_lock_irqsave(&l2x0_lock, flags);
165 	writel_relaxed(l2x0_way_mask, l2x0_base + L2X0_CLEAN_WAY);
166 	cache_wait_way(l2x0_base + L2X0_CLEAN_WAY, l2x0_way_mask);
167 	cache_sync();
168 	raw_spin_unlock_irqrestore(&l2x0_lock, flags);
169 }
170 
171 static void l2x0_inv_all(void)
172 {
173 	unsigned long flags;
174 
175 	/* invalidate all ways */
176 	raw_spin_lock_irqsave(&l2x0_lock, flags);
177 	/* Invalidating when L2 is enabled is a nono */
178 	BUG_ON(readl(l2x0_base + L2X0_CTRL) & L2X0_CTRL_EN);
179 	writel_relaxed(l2x0_way_mask, l2x0_base + L2X0_INV_WAY);
180 	cache_wait_way(l2x0_base + L2X0_INV_WAY, l2x0_way_mask);
181 	cache_sync();
182 	raw_spin_unlock_irqrestore(&l2x0_lock, flags);
183 }
184 
185 static void l2x0_inv_range(unsigned long start, unsigned long end)
186 {
187 	void __iomem *base = l2x0_base;
188 	unsigned long flags;
189 
190 	raw_spin_lock_irqsave(&l2x0_lock, flags);
191 	if (start & (CACHE_LINE_SIZE - 1)) {
192 		start &= ~(CACHE_LINE_SIZE - 1);
193 		debug_writel(0x03);
194 		l2x0_flush_line(start);
195 		debug_writel(0x00);
196 		start += CACHE_LINE_SIZE;
197 	}
198 
199 	if (end & (CACHE_LINE_SIZE - 1)) {
200 		end &= ~(CACHE_LINE_SIZE - 1);
201 		debug_writel(0x03);
202 		l2x0_flush_line(end);
203 		debug_writel(0x00);
204 	}
205 
206 	while (start < end) {
207 		unsigned long blk_end = start + min(end - start, 4096UL);
208 
209 		while (start < blk_end) {
210 			l2x0_inv_line(start);
211 			start += CACHE_LINE_SIZE;
212 		}
213 
214 		if (blk_end < end) {
215 			raw_spin_unlock_irqrestore(&l2x0_lock, flags);
216 			raw_spin_lock_irqsave(&l2x0_lock, flags);
217 		}
218 	}
219 	cache_wait(base + L2X0_INV_LINE_PA, 1);
220 	cache_sync();
221 	raw_spin_unlock_irqrestore(&l2x0_lock, flags);
222 }
223 
224 static void l2x0_clean_range(unsigned long start, unsigned long end)
225 {
226 	void __iomem *base = l2x0_base;
227 	unsigned long flags;
228 
229 	if ((end - start) >= l2x0_size) {
230 		l2x0_clean_all();
231 		return;
232 	}
233 
234 	raw_spin_lock_irqsave(&l2x0_lock, flags);
235 	start &= ~(CACHE_LINE_SIZE - 1);
236 	while (start < end) {
237 		unsigned long blk_end = start + min(end - start, 4096UL);
238 
239 		while (start < blk_end) {
240 			l2x0_clean_line(start);
241 			start += CACHE_LINE_SIZE;
242 		}
243 
244 		if (blk_end < end) {
245 			raw_spin_unlock_irqrestore(&l2x0_lock, flags);
246 			raw_spin_lock_irqsave(&l2x0_lock, flags);
247 		}
248 	}
249 	cache_wait(base + L2X0_CLEAN_LINE_PA, 1);
250 	cache_sync();
251 	raw_spin_unlock_irqrestore(&l2x0_lock, flags);
252 }
253 
254 static void l2x0_flush_range(unsigned long start, unsigned long end)
255 {
256 	void __iomem *base = l2x0_base;
257 	unsigned long flags;
258 
259 	if ((end - start) >= l2x0_size) {
260 		l2x0_flush_all();
261 		return;
262 	}
263 
264 	raw_spin_lock_irqsave(&l2x0_lock, flags);
265 	start &= ~(CACHE_LINE_SIZE - 1);
266 	while (start < end) {
267 		unsigned long blk_end = start + min(end - start, 4096UL);
268 
269 		debug_writel(0x03);
270 		while (start < blk_end) {
271 			l2x0_flush_line(start);
272 			start += CACHE_LINE_SIZE;
273 		}
274 		debug_writel(0x00);
275 
276 		if (blk_end < end) {
277 			raw_spin_unlock_irqrestore(&l2x0_lock, flags);
278 			raw_spin_lock_irqsave(&l2x0_lock, flags);
279 		}
280 	}
281 	cache_wait(base + L2X0_CLEAN_INV_LINE_PA, 1);
282 	cache_sync();
283 	raw_spin_unlock_irqrestore(&l2x0_lock, flags);
284 }
285 
286 static void l2x0_disable(void)
287 {
288 	unsigned long flags;
289 
290 	raw_spin_lock_irqsave(&l2x0_lock, flags);
291 	__l2x0_flush_all();
292 	writel_relaxed(0, l2x0_base + L2X0_CTRL);
293 	dsb();
294 	raw_spin_unlock_irqrestore(&l2x0_lock, flags);
295 }
296 
297 static void l2x0_unlock(u32 cache_id)
298 {
299 	int lockregs;
300 	int i;
301 
302 	switch (cache_id & L2X0_CACHE_ID_PART_MASK) {
303 	case L2X0_CACHE_ID_PART_L310:
304 		lockregs = 8;
305 		break;
306 	case AURORA_CACHE_ID:
307 		lockregs = 4;
308 		break;
309 	default:
310 		/* L210 and unknown types */
311 		lockregs = 1;
312 		break;
313 	}
314 
315 	for (i = 0; i < lockregs; i++) {
316 		writel_relaxed(0x0, l2x0_base + L2X0_LOCKDOWN_WAY_D_BASE +
317 			       i * L2X0_LOCKDOWN_STRIDE);
318 		writel_relaxed(0x0, l2x0_base + L2X0_LOCKDOWN_WAY_I_BASE +
319 			       i * L2X0_LOCKDOWN_STRIDE);
320 	}
321 }
322 
323 void __init l2x0_init(void __iomem *base, u32 aux_val, u32 aux_mask)
324 {
325 	u32 aux;
326 	u32 cache_id;
327 	u32 way_size = 0;
328 	int ways;
329 	int way_size_shift = L2X0_WAY_SIZE_SHIFT;
330 	const char *type;
331 
332 	l2x0_base = base;
333 	if (cache_id_part_number_from_dt)
334 		cache_id = cache_id_part_number_from_dt;
335 	else
336 		cache_id = readl_relaxed(l2x0_base + L2X0_CACHE_ID);
337 	aux = readl_relaxed(l2x0_base + L2X0_AUX_CTRL);
338 
339 	aux &= aux_mask;
340 	aux |= aux_val;
341 
342 	/* Determine the number of ways */
343 	switch (cache_id & L2X0_CACHE_ID_PART_MASK) {
344 	case L2X0_CACHE_ID_PART_L310:
345 		if (aux & (1 << 16))
346 			ways = 16;
347 		else
348 			ways = 8;
349 		type = "L310";
350 #ifdef CONFIG_PL310_ERRATA_753970
351 		/* Unmapped register. */
352 		sync_reg_offset = L2X0_DUMMY_REG;
353 #endif
354 		if ((cache_id & L2X0_CACHE_ID_RTL_MASK) <= L2X0_CACHE_ID_RTL_R3P0)
355 			outer_cache.set_debug = pl310_set_debug;
356 		break;
357 	case L2X0_CACHE_ID_PART_L210:
358 		ways = (aux >> 13) & 0xf;
359 		type = "L210";
360 		break;
361 
362 	case AURORA_CACHE_ID:
363 		sync_reg_offset = AURORA_SYNC_REG;
364 		ways = (aux >> 13) & 0xf;
365 		ways = 2 << ((ways + 1) >> 2);
366 		way_size_shift = AURORA_WAY_SIZE_SHIFT;
367 		type = "Aurora";
368 		break;
369 	default:
370 		/* Assume unknown chips have 8 ways */
371 		ways = 8;
372 		type = "L2x0 series";
373 		break;
374 	}
375 
376 	l2x0_way_mask = (1 << ways) - 1;
377 
378 	/*
379 	 * L2 cache Size =  Way size * Number of ways
380 	 */
381 	way_size = (aux & L2X0_AUX_CTRL_WAY_SIZE_MASK) >> 17;
382 	way_size = 1 << (way_size + way_size_shift);
383 
384 	l2x0_size = ways * way_size * SZ_1K;
385 
386 	/*
387 	 * Check if l2x0 controller is already enabled.
388 	 * If you are booting from non-secure mode
389 	 * accessing the below registers will fault.
390 	 */
391 	if (!(readl_relaxed(l2x0_base + L2X0_CTRL) & L2X0_CTRL_EN)) {
392 		/* Make sure that I&D is not locked down when starting */
393 		l2x0_unlock(cache_id);
394 
395 		/* l2x0 controller is disabled */
396 		writel_relaxed(aux, l2x0_base + L2X0_AUX_CTRL);
397 
398 		l2x0_inv_all();
399 
400 		/* enable L2X0 */
401 		writel_relaxed(L2X0_CTRL_EN, l2x0_base + L2X0_CTRL);
402 	}
403 
404 	/* Re-read it in case some bits are reserved. */
405 	aux = readl_relaxed(l2x0_base + L2X0_AUX_CTRL);
406 
407 	/* Save the value for resuming. */
408 	l2x0_saved_regs.aux_ctrl = aux;
409 
410 	if (!of_init) {
411 		outer_cache.inv_range = l2x0_inv_range;
412 		outer_cache.clean_range = l2x0_clean_range;
413 		outer_cache.flush_range = l2x0_flush_range;
414 		outer_cache.sync = l2x0_cache_sync;
415 		outer_cache.flush_all = l2x0_flush_all;
416 		outer_cache.inv_all = l2x0_inv_all;
417 		outer_cache.disable = l2x0_disable;
418 	}
419 
420 	pr_info("%s cache controller enabled\n", type);
421 	pr_info("l2x0: %d ways, CACHE_ID 0x%08x, AUX_CTRL 0x%08x, Cache size: %d kB\n",
422 		ways, cache_id, aux, l2x0_size >> 10);
423 }
424 
425 #ifdef CONFIG_OF
426 static int l2_wt_override;
427 
428 /*
429  * Note that the end addresses passed to Linux primitives are
430  * noninclusive, while the hardware cache range operations use
431  * inclusive start and end addresses.
432  */
433 static unsigned long calc_range_end(unsigned long start, unsigned long end)
434 {
435 	/*
436 	 * Limit the number of cache lines processed at once,
437 	 * since cache range operations stall the CPU pipeline
438 	 * until completion.
439 	 */
440 	if (end > start + MAX_RANGE_SIZE)
441 		end = start + MAX_RANGE_SIZE;
442 
443 	/*
444 	 * Cache range operations can't straddle a page boundary.
445 	 */
446 	if (end > PAGE_ALIGN(start+1))
447 		end = PAGE_ALIGN(start+1);
448 
449 	return end;
450 }
451 
452 /*
453  * Make sure 'start' and 'end' reference the same page, as L2 is PIPT
454  * and range operations only do a TLB lookup on the start address.
455  */
456 static void aurora_pa_range(unsigned long start, unsigned long end,
457 			unsigned long offset)
458 {
459 	unsigned long flags;
460 
461 	raw_spin_lock_irqsave(&l2x0_lock, flags);
462 	writel_relaxed(start, l2x0_base + AURORA_RANGE_BASE_ADDR_REG);
463 	writel_relaxed(end, l2x0_base + offset);
464 	raw_spin_unlock_irqrestore(&l2x0_lock, flags);
465 
466 	cache_sync();
467 }
468 
469 static void aurora_inv_range(unsigned long start, unsigned long end)
470 {
471 	/*
472 	 * round start and end adresses up to cache line size
473 	 */
474 	start &= ~(CACHE_LINE_SIZE - 1);
475 	end = ALIGN(end, CACHE_LINE_SIZE);
476 
477 	/*
478 	 * Invalidate all full cache lines between 'start' and 'end'.
479 	 */
480 	while (start < end) {
481 		unsigned long range_end = calc_range_end(start, end);
482 		aurora_pa_range(start, range_end - CACHE_LINE_SIZE,
483 				AURORA_INVAL_RANGE_REG);
484 		start = range_end;
485 	}
486 }
487 
488 static void aurora_clean_range(unsigned long start, unsigned long end)
489 {
490 	/*
491 	 * If L2 is forced to WT, the L2 will always be clean and we
492 	 * don't need to do anything here.
493 	 */
494 	if (!l2_wt_override) {
495 		start &= ~(CACHE_LINE_SIZE - 1);
496 		end = ALIGN(end, CACHE_LINE_SIZE);
497 		while (start != end) {
498 			unsigned long range_end = calc_range_end(start, end);
499 			aurora_pa_range(start, range_end - CACHE_LINE_SIZE,
500 					AURORA_CLEAN_RANGE_REG);
501 			start = range_end;
502 		}
503 	}
504 }
505 
506 static void aurora_flush_range(unsigned long start, unsigned long end)
507 {
508 	start &= ~(CACHE_LINE_SIZE - 1);
509 	end = ALIGN(end, CACHE_LINE_SIZE);
510 	while (start != end) {
511 		unsigned long range_end = calc_range_end(start, end);
512 		/*
513 		 * If L2 is forced to WT, the L2 will always be clean and we
514 		 * just need to invalidate.
515 		 */
516 		if (l2_wt_override)
517 			aurora_pa_range(start, range_end - CACHE_LINE_SIZE,
518 							AURORA_INVAL_RANGE_REG);
519 		else
520 			aurora_pa_range(start, range_end - CACHE_LINE_SIZE,
521 							AURORA_FLUSH_RANGE_REG);
522 		start = range_end;
523 	}
524 }
525 
526 /*
527  * For certain Broadcom SoCs, depending on the address range, different offsets
528  * need to be added to the address before passing it to L2 for
529  * invalidation/clean/flush
530  *
531  * Section Address Range              Offset        EMI
532  *   1     0x00000000 - 0x3FFFFFFF    0x80000000    VC
533  *   2     0x40000000 - 0xBFFFFFFF    0x40000000    SYS
534  *   3     0xC0000000 - 0xFFFFFFFF    0x80000000    VC
535  *
536  * When the start and end addresses have crossed two different sections, we
537  * need to break the L2 operation into two, each within its own section.
538  * For example, if we need to invalidate addresses starts at 0xBFFF0000 and
539  * ends at 0xC0001000, we need do invalidate 1) 0xBFFF0000 - 0xBFFFFFFF and 2)
540  * 0xC0000000 - 0xC0001000
541  *
542  * Note 1:
543  * By breaking a single L2 operation into two, we may potentially suffer some
544  * performance hit, but keep in mind the cross section case is very rare
545  *
546  * Note 2:
547  * We do not need to handle the case when the start address is in
548  * Section 1 and the end address is in Section 3, since it is not a valid use
549  * case
550  *
551  * Note 3:
552  * Section 1 in practical terms can no longer be used on rev A2. Because of
553  * that the code does not need to handle section 1 at all.
554  *
555  */
556 #define BCM_SYS_EMI_START_ADDR        0x40000000UL
557 #define BCM_VC_EMI_SEC3_START_ADDR    0xC0000000UL
558 
559 #define BCM_SYS_EMI_OFFSET            0x40000000UL
560 #define BCM_VC_EMI_OFFSET             0x80000000UL
561 
562 static inline int bcm_addr_is_sys_emi(unsigned long addr)
563 {
564 	return (addr >= BCM_SYS_EMI_START_ADDR) &&
565 		(addr < BCM_VC_EMI_SEC3_START_ADDR);
566 }
567 
568 static inline unsigned long bcm_l2_phys_addr(unsigned long addr)
569 {
570 	if (bcm_addr_is_sys_emi(addr))
571 		return addr + BCM_SYS_EMI_OFFSET;
572 	else
573 		return addr + BCM_VC_EMI_OFFSET;
574 }
575 
576 static void bcm_inv_range(unsigned long start, unsigned long end)
577 {
578 	unsigned long new_start, new_end;
579 
580 	BUG_ON(start < BCM_SYS_EMI_START_ADDR);
581 
582 	if (unlikely(end <= start))
583 		return;
584 
585 	new_start = bcm_l2_phys_addr(start);
586 	new_end = bcm_l2_phys_addr(end);
587 
588 	/* normal case, no cross section between start and end */
589 	if (likely(bcm_addr_is_sys_emi(end) || !bcm_addr_is_sys_emi(start))) {
590 		l2x0_inv_range(new_start, new_end);
591 		return;
592 	}
593 
594 	/* They cross sections, so it can only be a cross from section
595 	 * 2 to section 3
596 	 */
597 	l2x0_inv_range(new_start,
598 		bcm_l2_phys_addr(BCM_VC_EMI_SEC3_START_ADDR-1));
599 	l2x0_inv_range(bcm_l2_phys_addr(BCM_VC_EMI_SEC3_START_ADDR),
600 		new_end);
601 }
602 
603 static void bcm_clean_range(unsigned long start, unsigned long end)
604 {
605 	unsigned long new_start, new_end;
606 
607 	BUG_ON(start < BCM_SYS_EMI_START_ADDR);
608 
609 	if (unlikely(end <= start))
610 		return;
611 
612 	if ((end - start) >= l2x0_size) {
613 		l2x0_clean_all();
614 		return;
615 	}
616 
617 	new_start = bcm_l2_phys_addr(start);
618 	new_end = bcm_l2_phys_addr(end);
619 
620 	/* normal case, no cross section between start and end */
621 	if (likely(bcm_addr_is_sys_emi(end) || !bcm_addr_is_sys_emi(start))) {
622 		l2x0_clean_range(new_start, new_end);
623 		return;
624 	}
625 
626 	/* They cross sections, so it can only be a cross from section
627 	 * 2 to section 3
628 	 */
629 	l2x0_clean_range(new_start,
630 		bcm_l2_phys_addr(BCM_VC_EMI_SEC3_START_ADDR-1));
631 	l2x0_clean_range(bcm_l2_phys_addr(BCM_VC_EMI_SEC3_START_ADDR),
632 		new_end);
633 }
634 
635 static void bcm_flush_range(unsigned long start, unsigned long end)
636 {
637 	unsigned long new_start, new_end;
638 
639 	BUG_ON(start < BCM_SYS_EMI_START_ADDR);
640 
641 	if (unlikely(end <= start))
642 		return;
643 
644 	if ((end - start) >= l2x0_size) {
645 		l2x0_flush_all();
646 		return;
647 	}
648 
649 	new_start = bcm_l2_phys_addr(start);
650 	new_end = bcm_l2_phys_addr(end);
651 
652 	/* normal case, no cross section between start and end */
653 	if (likely(bcm_addr_is_sys_emi(end) || !bcm_addr_is_sys_emi(start))) {
654 		l2x0_flush_range(new_start, new_end);
655 		return;
656 	}
657 
658 	/* They cross sections, so it can only be a cross from section
659 	 * 2 to section 3
660 	 */
661 	l2x0_flush_range(new_start,
662 		bcm_l2_phys_addr(BCM_VC_EMI_SEC3_START_ADDR-1));
663 	l2x0_flush_range(bcm_l2_phys_addr(BCM_VC_EMI_SEC3_START_ADDR),
664 		new_end);
665 }
666 
667 static void __init l2x0_of_setup(const struct device_node *np,
668 				 u32 *aux_val, u32 *aux_mask)
669 {
670 	u32 data[2] = { 0, 0 };
671 	u32 tag = 0;
672 	u32 dirty = 0;
673 	u32 val = 0, mask = 0;
674 
675 	of_property_read_u32(np, "arm,tag-latency", &tag);
676 	if (tag) {
677 		mask |= L2X0_AUX_CTRL_TAG_LATENCY_MASK;
678 		val |= (tag - 1) << L2X0_AUX_CTRL_TAG_LATENCY_SHIFT;
679 	}
680 
681 	of_property_read_u32_array(np, "arm,data-latency",
682 				   data, ARRAY_SIZE(data));
683 	if (data[0] && data[1]) {
684 		mask |= L2X0_AUX_CTRL_DATA_RD_LATENCY_MASK |
685 			L2X0_AUX_CTRL_DATA_WR_LATENCY_MASK;
686 		val |= ((data[0] - 1) << L2X0_AUX_CTRL_DATA_RD_LATENCY_SHIFT) |
687 		       ((data[1] - 1) << L2X0_AUX_CTRL_DATA_WR_LATENCY_SHIFT);
688 	}
689 
690 	of_property_read_u32(np, "arm,dirty-latency", &dirty);
691 	if (dirty) {
692 		mask |= L2X0_AUX_CTRL_DIRTY_LATENCY_MASK;
693 		val |= (dirty - 1) << L2X0_AUX_CTRL_DIRTY_LATENCY_SHIFT;
694 	}
695 
696 	*aux_val &= ~mask;
697 	*aux_val |= val;
698 	*aux_mask &= ~mask;
699 }
700 
701 static void __init pl310_of_setup(const struct device_node *np,
702 				  u32 *aux_val, u32 *aux_mask)
703 {
704 	u32 data[3] = { 0, 0, 0 };
705 	u32 tag[3] = { 0, 0, 0 };
706 	u32 filter[2] = { 0, 0 };
707 
708 	of_property_read_u32_array(np, "arm,tag-latency", tag, ARRAY_SIZE(tag));
709 	if (tag[0] && tag[1] && tag[2])
710 		writel_relaxed(
711 			((tag[0] - 1) << L2X0_LATENCY_CTRL_RD_SHIFT) |
712 			((tag[1] - 1) << L2X0_LATENCY_CTRL_WR_SHIFT) |
713 			((tag[2] - 1) << L2X0_LATENCY_CTRL_SETUP_SHIFT),
714 			l2x0_base + L2X0_TAG_LATENCY_CTRL);
715 
716 	of_property_read_u32_array(np, "arm,data-latency",
717 				   data, ARRAY_SIZE(data));
718 	if (data[0] && data[1] && data[2])
719 		writel_relaxed(
720 			((data[0] - 1) << L2X0_LATENCY_CTRL_RD_SHIFT) |
721 			((data[1] - 1) << L2X0_LATENCY_CTRL_WR_SHIFT) |
722 			((data[2] - 1) << L2X0_LATENCY_CTRL_SETUP_SHIFT),
723 			l2x0_base + L2X0_DATA_LATENCY_CTRL);
724 
725 	of_property_read_u32_array(np, "arm,filter-ranges",
726 				   filter, ARRAY_SIZE(filter));
727 	if (filter[1]) {
728 		writel_relaxed(ALIGN(filter[0] + filter[1], SZ_1M),
729 			       l2x0_base + L2X0_ADDR_FILTER_END);
730 		writel_relaxed((filter[0] & ~(SZ_1M - 1)) | L2X0_ADDR_FILTER_EN,
731 			       l2x0_base + L2X0_ADDR_FILTER_START);
732 	}
733 }
734 
735 static void __init pl310_save(void)
736 {
737 	u32 l2x0_revision = readl_relaxed(l2x0_base + L2X0_CACHE_ID) &
738 		L2X0_CACHE_ID_RTL_MASK;
739 
740 	l2x0_saved_regs.tag_latency = readl_relaxed(l2x0_base +
741 		L2X0_TAG_LATENCY_CTRL);
742 	l2x0_saved_regs.data_latency = readl_relaxed(l2x0_base +
743 		L2X0_DATA_LATENCY_CTRL);
744 	l2x0_saved_regs.filter_end = readl_relaxed(l2x0_base +
745 		L2X0_ADDR_FILTER_END);
746 	l2x0_saved_regs.filter_start = readl_relaxed(l2x0_base +
747 		L2X0_ADDR_FILTER_START);
748 
749 	if (l2x0_revision >= L2X0_CACHE_ID_RTL_R2P0) {
750 		/*
751 		 * From r2p0, there is Prefetch offset/control register
752 		 */
753 		l2x0_saved_regs.prefetch_ctrl = readl_relaxed(l2x0_base +
754 			L2X0_PREFETCH_CTRL);
755 		/*
756 		 * From r3p0, there is Power control register
757 		 */
758 		if (l2x0_revision >= L2X0_CACHE_ID_RTL_R3P0)
759 			l2x0_saved_regs.pwr_ctrl = readl_relaxed(l2x0_base +
760 				L2X0_POWER_CTRL);
761 	}
762 }
763 
764 static void aurora_save(void)
765 {
766 	l2x0_saved_regs.ctrl = readl_relaxed(l2x0_base + L2X0_CTRL);
767 	l2x0_saved_regs.aux_ctrl = readl_relaxed(l2x0_base + L2X0_AUX_CTRL);
768 }
769 
770 static void l2x0_resume(void)
771 {
772 	if (!(readl_relaxed(l2x0_base + L2X0_CTRL) & L2X0_CTRL_EN)) {
773 		/* restore aux ctrl and enable l2 */
774 		l2x0_unlock(readl_relaxed(l2x0_base + L2X0_CACHE_ID));
775 
776 		writel_relaxed(l2x0_saved_regs.aux_ctrl, l2x0_base +
777 			L2X0_AUX_CTRL);
778 
779 		l2x0_inv_all();
780 
781 		writel_relaxed(L2X0_CTRL_EN, l2x0_base + L2X0_CTRL);
782 	}
783 }
784 
785 static void pl310_resume(void)
786 {
787 	u32 l2x0_revision;
788 
789 	if (!(readl_relaxed(l2x0_base + L2X0_CTRL) & L2X0_CTRL_EN)) {
790 		/* restore pl310 setup */
791 		writel_relaxed(l2x0_saved_regs.tag_latency,
792 			l2x0_base + L2X0_TAG_LATENCY_CTRL);
793 		writel_relaxed(l2x0_saved_regs.data_latency,
794 			l2x0_base + L2X0_DATA_LATENCY_CTRL);
795 		writel_relaxed(l2x0_saved_regs.filter_end,
796 			l2x0_base + L2X0_ADDR_FILTER_END);
797 		writel_relaxed(l2x0_saved_regs.filter_start,
798 			l2x0_base + L2X0_ADDR_FILTER_START);
799 
800 		l2x0_revision = readl_relaxed(l2x0_base + L2X0_CACHE_ID) &
801 			L2X0_CACHE_ID_RTL_MASK;
802 
803 		if (l2x0_revision >= L2X0_CACHE_ID_RTL_R2P0) {
804 			writel_relaxed(l2x0_saved_regs.prefetch_ctrl,
805 				l2x0_base + L2X0_PREFETCH_CTRL);
806 			if (l2x0_revision >= L2X0_CACHE_ID_RTL_R3P0)
807 				writel_relaxed(l2x0_saved_regs.pwr_ctrl,
808 					l2x0_base + L2X0_POWER_CTRL);
809 		}
810 	}
811 
812 	l2x0_resume();
813 }
814 
815 static void aurora_resume(void)
816 {
817 	if (!(readl(l2x0_base + L2X0_CTRL) & L2X0_CTRL_EN)) {
818 		writel_relaxed(l2x0_saved_regs.aux_ctrl,
819 				l2x0_base + L2X0_AUX_CTRL);
820 		writel_relaxed(l2x0_saved_regs.ctrl, l2x0_base + L2X0_CTRL);
821 	}
822 }
823 
824 static void __init aurora_broadcast_l2_commands(void)
825 {
826 	__u32 u;
827 	/* Enable Broadcasting of cache commands to L2*/
828 	__asm__ __volatile__("mrc p15, 1, %0, c15, c2, 0" : "=r"(u));
829 	u |= AURORA_CTRL_FW;		/* Set the FW bit */
830 	__asm__ __volatile__("mcr p15, 1, %0, c15, c2, 0\n" : : "r"(u));
831 	isb();
832 }
833 
834 static void __init aurora_of_setup(const struct device_node *np,
835 				u32 *aux_val, u32 *aux_mask)
836 {
837 	u32 val = AURORA_ACR_REPLACEMENT_TYPE_SEMIPLRU;
838 	u32 mask =  AURORA_ACR_REPLACEMENT_MASK;
839 
840 	of_property_read_u32(np, "cache-id-part",
841 			&cache_id_part_number_from_dt);
842 
843 	/* Determine and save the write policy */
844 	l2_wt_override = of_property_read_bool(np, "wt-override");
845 
846 	if (l2_wt_override) {
847 		val |= AURORA_ACR_FORCE_WRITE_THRO_POLICY;
848 		mask |= AURORA_ACR_FORCE_WRITE_POLICY_MASK;
849 	}
850 
851 	*aux_val &= ~mask;
852 	*aux_val |= val;
853 	*aux_mask &= ~mask;
854 }
855 
856 static const struct l2x0_of_data pl310_data = {
857 	.setup = pl310_of_setup,
858 	.save  = pl310_save,
859 	.outer_cache = {
860 		.resume      = pl310_resume,
861 		.inv_range   = l2x0_inv_range,
862 		.clean_range = l2x0_clean_range,
863 		.flush_range = l2x0_flush_range,
864 		.sync        = l2x0_cache_sync,
865 		.flush_all   = l2x0_flush_all,
866 		.inv_all     = l2x0_inv_all,
867 		.disable     = l2x0_disable,
868 	},
869 };
870 
871 static const struct l2x0_of_data l2x0_data = {
872 	.setup = l2x0_of_setup,
873 	.save  = NULL,
874 	.outer_cache = {
875 		.resume      = l2x0_resume,
876 		.inv_range   = l2x0_inv_range,
877 		.clean_range = l2x0_clean_range,
878 		.flush_range = l2x0_flush_range,
879 		.sync        = l2x0_cache_sync,
880 		.flush_all   = l2x0_flush_all,
881 		.inv_all     = l2x0_inv_all,
882 		.disable     = l2x0_disable,
883 	},
884 };
885 
886 static const struct l2x0_of_data aurora_with_outer_data = {
887 	.setup = aurora_of_setup,
888 	.save  = aurora_save,
889 	.outer_cache = {
890 		.resume      = aurora_resume,
891 		.inv_range   = aurora_inv_range,
892 		.clean_range = aurora_clean_range,
893 		.flush_range = aurora_flush_range,
894 		.sync        = l2x0_cache_sync,
895 		.flush_all   = l2x0_flush_all,
896 		.inv_all     = l2x0_inv_all,
897 		.disable     = l2x0_disable,
898 	},
899 };
900 
901 static const struct l2x0_of_data aurora_no_outer_data = {
902 	.setup = aurora_of_setup,
903 	.save  = aurora_save,
904 	.outer_cache = {
905 		.resume      = aurora_resume,
906 	},
907 };
908 
909 static const struct l2x0_of_data bcm_l2x0_data = {
910 	.setup = pl310_of_setup,
911 	.save  = pl310_save,
912 	.outer_cache = {
913 		.resume      = pl310_resume,
914 		.inv_range   = bcm_inv_range,
915 		.clean_range = bcm_clean_range,
916 		.flush_range = bcm_flush_range,
917 		.sync        = l2x0_cache_sync,
918 		.flush_all   = l2x0_flush_all,
919 		.inv_all     = l2x0_inv_all,
920 		.disable     = l2x0_disable,
921 	},
922 };
923 
924 static const struct of_device_id l2x0_ids[] __initconst = {
925 	{ .compatible = "arm,pl310-cache", .data = (void *)&pl310_data },
926 	{ .compatible = "arm,l220-cache", .data = (void *)&l2x0_data },
927 	{ .compatible = "arm,l210-cache", .data = (void *)&l2x0_data },
928 	{ .compatible = "marvell,aurora-system-cache",
929 	  .data = (void *)&aurora_no_outer_data},
930 	{ .compatible = "marvell,aurora-outer-cache",
931 	  .data = (void *)&aurora_with_outer_data},
932 	{ .compatible = "brcm,bcm11351-a2-pl310-cache",
933 	  .data = (void *)&bcm_l2x0_data},
934 	{ .compatible = "bcm,bcm11351-a2-pl310-cache", /* deprecated name */
935 	  .data = (void *)&bcm_l2x0_data},
936 	{}
937 };
938 
939 int __init l2x0_of_init(u32 aux_val, u32 aux_mask)
940 {
941 	struct device_node *np;
942 	const struct l2x0_of_data *data;
943 	struct resource res;
944 
945 	np = of_find_matching_node(NULL, l2x0_ids);
946 	if (!np)
947 		return -ENODEV;
948 
949 	if (of_address_to_resource(np, 0, &res))
950 		return -ENODEV;
951 
952 	l2x0_base = ioremap(res.start, resource_size(&res));
953 	if (!l2x0_base)
954 		return -ENOMEM;
955 
956 	l2x0_saved_regs.phy_base = res.start;
957 
958 	data = of_match_node(l2x0_ids, np)->data;
959 
960 	/* L2 configuration can only be changed if the cache is disabled */
961 	if (!(readl_relaxed(l2x0_base + L2X0_CTRL) & L2X0_CTRL_EN)) {
962 		if (data->setup)
963 			data->setup(np, &aux_val, &aux_mask);
964 
965 		/* For aurora cache in no outer mode select the
966 		 * correct mode using the coprocessor*/
967 		if (data == &aurora_no_outer_data)
968 			aurora_broadcast_l2_commands();
969 	}
970 
971 	if (data->save)
972 		data->save();
973 
974 	of_init = true;
975 	memcpy(&outer_cache, &data->outer_cache, sizeof(outer_cache));
976 	l2x0_init(l2x0_base, aux_val, aux_mask);
977 
978 	return 0;
979 }
980 #endif
981