Merge branch 'work.splice_read' of git://git.kernel.org/pub/scm/linux/kernel/git...
[cascardo/linux.git] / arch / arm / mm / cache-l2x0.c
1 /*
2  * arch/arm/mm/cache-l2x0.c - L210/L220/L310 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/cpu.h>
20 #include <linux/err.h>
21 #include <linux/init.h>
22 #include <linux/smp.h>
23 #include <linux/spinlock.h>
24 #include <linux/log2.h>
25 #include <linux/io.h>
26 #include <linux/of.h>
27 #include <linux/of_address.h>
28
29 #include <asm/cacheflush.h>
30 #include <asm/cp15.h>
31 #include <asm/cputype.h>
32 #include <asm/hardware/cache-l2x0.h>
33 #include "cache-tauros3.h"
34 #include "cache-aurora-l2.h"
35
36 struct l2c_init_data {
37         const char *type;
38         unsigned way_size_0;
39         unsigned num_lock;
40         void (*of_parse)(const struct device_node *, u32 *, u32 *);
41         void (*enable)(void __iomem *, unsigned);
42         void (*fixup)(void __iomem *, u32, struct outer_cache_fns *);
43         void (*save)(void __iomem *);
44         void (*configure)(void __iomem *);
45         void (*unlock)(void __iomem *, unsigned);
46         struct outer_cache_fns outer_cache;
47 };
48
49 #define CACHE_LINE_SIZE         32
50
51 static void __iomem *l2x0_base;
52 static const struct l2c_init_data *l2x0_data;
53 static DEFINE_RAW_SPINLOCK(l2x0_lock);
54 static u32 l2x0_way_mask;       /* Bitmask of active ways */
55 static u32 l2x0_size;
56 static unsigned long sync_reg_offset = L2X0_CACHE_SYNC;
57
58 struct l2x0_regs l2x0_saved_regs;
59
60 /*
61  * Common code for all cache controllers.
62  */
63 static inline void l2c_wait_mask(void __iomem *reg, unsigned long mask)
64 {
65         /* wait for cache operation by line or way to complete */
66         while (readl_relaxed(reg) & mask)
67                 cpu_relax();
68 }
69
70 /*
71  * By default, we write directly to secure registers.  Platforms must
72  * override this if they are running non-secure.
73  */
74 static void l2c_write_sec(unsigned long val, void __iomem *base, unsigned reg)
75 {
76         if (val == readl_relaxed(base + reg))
77                 return;
78         if (outer_cache.write_sec)
79                 outer_cache.write_sec(val, reg);
80         else
81                 writel_relaxed(val, base + reg);
82 }
83
84 /*
85  * This should only be called when we have a requirement that the
86  * register be written due to a work-around, as platforms running
87  * in non-secure mode may not be able to access this register.
88  */
89 static inline void l2c_set_debug(void __iomem *base, unsigned long val)
90 {
91         l2c_write_sec(val, base, L2X0_DEBUG_CTRL);
92 }
93
94 static void __l2c_op_way(void __iomem *reg)
95 {
96         writel_relaxed(l2x0_way_mask, reg);
97         l2c_wait_mask(reg, l2x0_way_mask);
98 }
99
100 static inline void l2c_unlock(void __iomem *base, unsigned num)
101 {
102         unsigned i;
103
104         for (i = 0; i < num; i++) {
105                 writel_relaxed(0, base + L2X0_LOCKDOWN_WAY_D_BASE +
106                                i * L2X0_LOCKDOWN_STRIDE);
107                 writel_relaxed(0, base + L2X0_LOCKDOWN_WAY_I_BASE +
108                                i * L2X0_LOCKDOWN_STRIDE);
109         }
110 }
111
112 static void l2c_configure(void __iomem *base)
113 {
114         l2c_write_sec(l2x0_saved_regs.aux_ctrl, base, L2X0_AUX_CTRL);
115 }
116
117 /*
118  * Enable the L2 cache controller.  This function must only be
119  * called when the cache controller is known to be disabled.
120  */
121 static void l2c_enable(void __iomem *base, unsigned num_lock)
122 {
123         unsigned long flags;
124
125         if (outer_cache.configure)
126                 outer_cache.configure(&l2x0_saved_regs);
127         else
128                 l2x0_data->configure(base);
129
130         l2x0_data->unlock(base, num_lock);
131
132         local_irq_save(flags);
133         __l2c_op_way(base + L2X0_INV_WAY);
134         writel_relaxed(0, base + sync_reg_offset);
135         l2c_wait_mask(base + sync_reg_offset, 1);
136         local_irq_restore(flags);
137
138         l2c_write_sec(L2X0_CTRL_EN, base, L2X0_CTRL);
139 }
140
141 static void l2c_disable(void)
142 {
143         void __iomem *base = l2x0_base;
144
145         l2x0_pmu_suspend();
146
147         outer_cache.flush_all();
148         l2c_write_sec(0, base, L2X0_CTRL);
149         dsb(st);
150 }
151
152 static void l2c_save(void __iomem *base)
153 {
154         l2x0_saved_regs.aux_ctrl = readl_relaxed(l2x0_base + L2X0_AUX_CTRL);
155 }
156
157 static void l2c_resume(void)
158 {
159         void __iomem *base = l2x0_base;
160
161         /* Do not touch the controller if already enabled. */
162         if (!(readl_relaxed(base + L2X0_CTRL) & L2X0_CTRL_EN))
163                 l2c_enable(base, l2x0_data->num_lock);
164
165         l2x0_pmu_resume();
166 }
167
168 /*
169  * L2C-210 specific code.
170  *
171  * The L2C-2x0 PA, set/way and sync operations are atomic, but we must
172  * ensure that no background operation is running.  The way operations
173  * are all background tasks.
174  *
175  * While a background operation is in progress, any new operation is
176  * ignored (unspecified whether this causes an error.)  Thankfully, not
177  * used on SMP.
178  *
179  * Never has a different sync register other than L2X0_CACHE_SYNC, but
180  * we use sync_reg_offset here so we can share some of this with L2C-310.
181  */
182 static void __l2c210_cache_sync(void __iomem *base)
183 {
184         writel_relaxed(0, base + sync_reg_offset);
185 }
186
187 static void __l2c210_op_pa_range(void __iomem *reg, unsigned long start,
188         unsigned long end)
189 {
190         while (start < end) {
191                 writel_relaxed(start, reg);
192                 start += CACHE_LINE_SIZE;
193         }
194 }
195
196 static void l2c210_inv_range(unsigned long start, unsigned long end)
197 {
198         void __iomem *base = l2x0_base;
199
200         if (start & (CACHE_LINE_SIZE - 1)) {
201                 start &= ~(CACHE_LINE_SIZE - 1);
202                 writel_relaxed(start, base + L2X0_CLEAN_INV_LINE_PA);
203                 start += CACHE_LINE_SIZE;
204         }
205
206         if (end & (CACHE_LINE_SIZE - 1)) {
207                 end &= ~(CACHE_LINE_SIZE - 1);
208                 writel_relaxed(end, base + L2X0_CLEAN_INV_LINE_PA);
209         }
210
211         __l2c210_op_pa_range(base + L2X0_INV_LINE_PA, start, end);
212         __l2c210_cache_sync(base);
213 }
214
215 static void l2c210_clean_range(unsigned long start, unsigned long end)
216 {
217         void __iomem *base = l2x0_base;
218
219         start &= ~(CACHE_LINE_SIZE - 1);
220         __l2c210_op_pa_range(base + L2X0_CLEAN_LINE_PA, start, end);
221         __l2c210_cache_sync(base);
222 }
223
224 static void l2c210_flush_range(unsigned long start, unsigned long end)
225 {
226         void __iomem *base = l2x0_base;
227
228         start &= ~(CACHE_LINE_SIZE - 1);
229         __l2c210_op_pa_range(base + L2X0_CLEAN_INV_LINE_PA, start, end);
230         __l2c210_cache_sync(base);
231 }
232
233 static void l2c210_flush_all(void)
234 {
235         void __iomem *base = l2x0_base;
236
237         BUG_ON(!irqs_disabled());
238
239         __l2c_op_way(base + L2X0_CLEAN_INV_WAY);
240         __l2c210_cache_sync(base);
241 }
242
243 static void l2c210_sync(void)
244 {
245         __l2c210_cache_sync(l2x0_base);
246 }
247
248 static const struct l2c_init_data l2c210_data __initconst = {
249         .type = "L2C-210",
250         .way_size_0 = SZ_8K,
251         .num_lock = 1,
252         .enable = l2c_enable,
253         .save = l2c_save,
254         .configure = l2c_configure,
255         .unlock = l2c_unlock,
256         .outer_cache = {
257                 .inv_range = l2c210_inv_range,
258                 .clean_range = l2c210_clean_range,
259                 .flush_range = l2c210_flush_range,
260                 .flush_all = l2c210_flush_all,
261                 .disable = l2c_disable,
262                 .sync = l2c210_sync,
263                 .resume = l2c_resume,
264         },
265 };
266
267 /*
268  * L2C-220 specific code.
269  *
270  * All operations are background operations: they have to be waited for.
271  * Conflicting requests generate a slave error (which will cause an
272  * imprecise abort.)  Never uses sync_reg_offset, so we hard-code the
273  * sync register here.
274  *
275  * However, we can re-use the l2c210_resume call.
276  */
277 static inline void __l2c220_cache_sync(void __iomem *base)
278 {
279         writel_relaxed(0, base + L2X0_CACHE_SYNC);
280         l2c_wait_mask(base + L2X0_CACHE_SYNC, 1);
281 }
282
283 static void l2c220_op_way(void __iomem *base, unsigned reg)
284 {
285         unsigned long flags;
286
287         raw_spin_lock_irqsave(&l2x0_lock, flags);
288         __l2c_op_way(base + reg);
289         __l2c220_cache_sync(base);
290         raw_spin_unlock_irqrestore(&l2x0_lock, flags);
291 }
292
293 static unsigned long l2c220_op_pa_range(void __iomem *reg, unsigned long start,
294         unsigned long end, unsigned long flags)
295 {
296         raw_spinlock_t *lock = &l2x0_lock;
297
298         while (start < end) {
299                 unsigned long blk_end = start + min(end - start, 4096UL);
300
301                 while (start < blk_end) {
302                         l2c_wait_mask(reg, 1);
303                         writel_relaxed(start, reg);
304                         start += CACHE_LINE_SIZE;
305                 }
306
307                 if (blk_end < end) {
308                         raw_spin_unlock_irqrestore(lock, flags);
309                         raw_spin_lock_irqsave(lock, flags);
310                 }
311         }
312
313         return flags;
314 }
315
316 static void l2c220_inv_range(unsigned long start, unsigned long end)
317 {
318         void __iomem *base = l2x0_base;
319         unsigned long flags;
320
321         raw_spin_lock_irqsave(&l2x0_lock, flags);
322         if ((start | end) & (CACHE_LINE_SIZE - 1)) {
323                 if (start & (CACHE_LINE_SIZE - 1)) {
324                         start &= ~(CACHE_LINE_SIZE - 1);
325                         writel_relaxed(start, base + L2X0_CLEAN_INV_LINE_PA);
326                         start += CACHE_LINE_SIZE;
327                 }
328
329                 if (end & (CACHE_LINE_SIZE - 1)) {
330                         end &= ~(CACHE_LINE_SIZE - 1);
331                         l2c_wait_mask(base + L2X0_CLEAN_INV_LINE_PA, 1);
332                         writel_relaxed(end, base + L2X0_CLEAN_INV_LINE_PA);
333                 }
334         }
335
336         flags = l2c220_op_pa_range(base + L2X0_INV_LINE_PA,
337                                    start, end, flags);
338         l2c_wait_mask(base + L2X0_INV_LINE_PA, 1);
339         __l2c220_cache_sync(base);
340         raw_spin_unlock_irqrestore(&l2x0_lock, flags);
341 }
342
343 static void l2c220_clean_range(unsigned long start, unsigned long end)
344 {
345         void __iomem *base = l2x0_base;
346         unsigned long flags;
347
348         start &= ~(CACHE_LINE_SIZE - 1);
349         if ((end - start) >= l2x0_size) {
350                 l2c220_op_way(base, L2X0_CLEAN_WAY);
351                 return;
352         }
353
354         raw_spin_lock_irqsave(&l2x0_lock, flags);
355         flags = l2c220_op_pa_range(base + L2X0_CLEAN_LINE_PA,
356                                    start, end, flags);
357         l2c_wait_mask(base + L2X0_CLEAN_INV_LINE_PA, 1);
358         __l2c220_cache_sync(base);
359         raw_spin_unlock_irqrestore(&l2x0_lock, flags);
360 }
361
362 static void l2c220_flush_range(unsigned long start, unsigned long end)
363 {
364         void __iomem *base = l2x0_base;
365         unsigned long flags;
366
367         start &= ~(CACHE_LINE_SIZE - 1);
368         if ((end - start) >= l2x0_size) {
369                 l2c220_op_way(base, L2X0_CLEAN_INV_WAY);
370                 return;
371         }
372
373         raw_spin_lock_irqsave(&l2x0_lock, flags);
374         flags = l2c220_op_pa_range(base + L2X0_CLEAN_INV_LINE_PA,
375                                    start, end, flags);
376         l2c_wait_mask(base + L2X0_CLEAN_INV_LINE_PA, 1);
377         __l2c220_cache_sync(base);
378         raw_spin_unlock_irqrestore(&l2x0_lock, flags);
379 }
380
381 static void l2c220_flush_all(void)
382 {
383         l2c220_op_way(l2x0_base, L2X0_CLEAN_INV_WAY);
384 }
385
386 static void l2c220_sync(void)
387 {
388         unsigned long flags;
389
390         raw_spin_lock_irqsave(&l2x0_lock, flags);
391         __l2c220_cache_sync(l2x0_base);
392         raw_spin_unlock_irqrestore(&l2x0_lock, flags);
393 }
394
395 static void l2c220_enable(void __iomem *base, unsigned num_lock)
396 {
397         /*
398          * Always enable non-secure access to the lockdown registers -
399          * we write to them as part of the L2C enable sequence so they
400          * need to be accessible.
401          */
402         l2x0_saved_regs.aux_ctrl |= L220_AUX_CTRL_NS_LOCKDOWN;
403
404         l2c_enable(base, num_lock);
405 }
406
407 static void l2c220_unlock(void __iomem *base, unsigned num_lock)
408 {
409         if (readl_relaxed(base + L2X0_AUX_CTRL) & L220_AUX_CTRL_NS_LOCKDOWN)
410                 l2c_unlock(base, num_lock);
411 }
412
413 static const struct l2c_init_data l2c220_data = {
414         .type = "L2C-220",
415         .way_size_0 = SZ_8K,
416         .num_lock = 1,
417         .enable = l2c220_enable,
418         .save = l2c_save,
419         .configure = l2c_configure,
420         .unlock = l2c220_unlock,
421         .outer_cache = {
422                 .inv_range = l2c220_inv_range,
423                 .clean_range = l2c220_clean_range,
424                 .flush_range = l2c220_flush_range,
425                 .flush_all = l2c220_flush_all,
426                 .disable = l2c_disable,
427                 .sync = l2c220_sync,
428                 .resume = l2c_resume,
429         },
430 };
431
432 /*
433  * L2C-310 specific code.
434  *
435  * Very similar to L2C-210, the PA, set/way and sync operations are atomic,
436  * and the way operations are all background tasks.  However, issuing an
437  * operation while a background operation is in progress results in a
438  * SLVERR response.  We can reuse:
439  *
440  *  __l2c210_cache_sync (using sync_reg_offset)
441  *  l2c210_sync
442  *  l2c210_inv_range (if 588369 is not applicable)
443  *  l2c210_clean_range
444  *  l2c210_flush_range (if 588369 is not applicable)
445  *  l2c210_flush_all (if 727915 is not applicable)
446  *
447  * Errata:
448  * 588369: PL310 R0P0->R1P0, fixed R2P0.
449  *      Affects: all clean+invalidate operations
450  *      clean and invalidate skips the invalidate step, so we need to issue
451  *      separate operations.  We also require the above debug workaround
452  *      enclosing this code fragment on affected parts.  On unaffected parts,
453  *      we must not use this workaround without the debug register writes
454  *      to avoid exposing a problem similar to 727915.
455  *
456  * 727915: PL310 R2P0->R3P0, fixed R3P1.
457  *      Affects: clean+invalidate by way
458  *      clean and invalidate by way runs in the background, and a store can
459  *      hit the line between the clean operation and invalidate operation,
460  *      resulting in the store being lost.
461  *
462  * 752271: PL310 R3P0->R3P1-50REL0, fixed R3P2.
463  *      Affects: 8x64-bit (double fill) line fetches
464  *      double fill line fetches can fail to cause dirty data to be evicted
465  *      from the cache before the new data overwrites the second line.
466  *
467  * 753970: PL310 R3P0, fixed R3P1.
468  *      Affects: sync
469  *      prevents merging writes after the sync operation, until another L2C
470  *      operation is performed (or a number of other conditions.)
471  *
472  * 769419: PL310 R0P0->R3P1, fixed R3P2.
473  *      Affects: store buffer
474  *      store buffer is not automatically drained.
475  */
476 static void l2c310_inv_range_erratum(unsigned long start, unsigned long end)
477 {
478         void __iomem *base = l2x0_base;
479
480         if ((start | end) & (CACHE_LINE_SIZE - 1)) {
481                 unsigned long flags;
482
483                 /* Erratum 588369 for both clean+invalidate operations */
484                 raw_spin_lock_irqsave(&l2x0_lock, flags);
485                 l2c_set_debug(base, 0x03);
486
487                 if (start & (CACHE_LINE_SIZE - 1)) {
488                         start &= ~(CACHE_LINE_SIZE - 1);
489                         writel_relaxed(start, base + L2X0_CLEAN_LINE_PA);
490                         writel_relaxed(start, base + L2X0_INV_LINE_PA);
491                         start += CACHE_LINE_SIZE;
492                 }
493
494                 if (end & (CACHE_LINE_SIZE - 1)) {
495                         end &= ~(CACHE_LINE_SIZE - 1);
496                         writel_relaxed(end, base + L2X0_CLEAN_LINE_PA);
497                         writel_relaxed(end, base + L2X0_INV_LINE_PA);
498                 }
499
500                 l2c_set_debug(base, 0x00);
501                 raw_spin_unlock_irqrestore(&l2x0_lock, flags);
502         }
503
504         __l2c210_op_pa_range(base + L2X0_INV_LINE_PA, start, end);
505         __l2c210_cache_sync(base);
506 }
507
508 static void l2c310_flush_range_erratum(unsigned long start, unsigned long end)
509 {
510         raw_spinlock_t *lock = &l2x0_lock;
511         unsigned long flags;
512         void __iomem *base = l2x0_base;
513
514         raw_spin_lock_irqsave(lock, flags);
515         while (start < end) {
516                 unsigned long blk_end = start + min(end - start, 4096UL);
517
518                 l2c_set_debug(base, 0x03);
519                 while (start < blk_end) {
520                         writel_relaxed(start, base + L2X0_CLEAN_LINE_PA);
521                         writel_relaxed(start, base + L2X0_INV_LINE_PA);
522                         start += CACHE_LINE_SIZE;
523                 }
524                 l2c_set_debug(base, 0x00);
525
526                 if (blk_end < end) {
527                         raw_spin_unlock_irqrestore(lock, flags);
528                         raw_spin_lock_irqsave(lock, flags);
529                 }
530         }
531         raw_spin_unlock_irqrestore(lock, flags);
532         __l2c210_cache_sync(base);
533 }
534
535 static void l2c310_flush_all_erratum(void)
536 {
537         void __iomem *base = l2x0_base;
538         unsigned long flags;
539
540         raw_spin_lock_irqsave(&l2x0_lock, flags);
541         l2c_set_debug(base, 0x03);
542         __l2c_op_way(base + L2X0_CLEAN_INV_WAY);
543         l2c_set_debug(base, 0x00);
544         __l2c210_cache_sync(base);
545         raw_spin_unlock_irqrestore(&l2x0_lock, flags);
546 }
547
548 static void __init l2c310_save(void __iomem *base)
549 {
550         unsigned revision;
551
552         l2c_save(base);
553
554         l2x0_saved_regs.tag_latency = readl_relaxed(base +
555                 L310_TAG_LATENCY_CTRL);
556         l2x0_saved_regs.data_latency = readl_relaxed(base +
557                 L310_DATA_LATENCY_CTRL);
558         l2x0_saved_regs.filter_end = readl_relaxed(base +
559                 L310_ADDR_FILTER_END);
560         l2x0_saved_regs.filter_start = readl_relaxed(base +
561                 L310_ADDR_FILTER_START);
562
563         revision = readl_relaxed(base + L2X0_CACHE_ID) &
564                         L2X0_CACHE_ID_RTL_MASK;
565
566         /* From r2p0, there is Prefetch offset/control register */
567         if (revision >= L310_CACHE_ID_RTL_R2P0)
568                 l2x0_saved_regs.prefetch_ctrl = readl_relaxed(base +
569                                                         L310_PREFETCH_CTRL);
570
571         /* From r3p0, there is Power control register */
572         if (revision >= L310_CACHE_ID_RTL_R3P0)
573                 l2x0_saved_regs.pwr_ctrl = readl_relaxed(base +
574                                                         L310_POWER_CTRL);
575 }
576
577 static void l2c310_configure(void __iomem *base)
578 {
579         unsigned revision;
580
581         l2c_configure(base);
582
583         /* restore pl310 setup */
584         l2c_write_sec(l2x0_saved_regs.tag_latency, base,
585                       L310_TAG_LATENCY_CTRL);
586         l2c_write_sec(l2x0_saved_regs.data_latency, base,
587                       L310_DATA_LATENCY_CTRL);
588         l2c_write_sec(l2x0_saved_regs.filter_end, base,
589                       L310_ADDR_FILTER_END);
590         l2c_write_sec(l2x0_saved_regs.filter_start, base,
591                       L310_ADDR_FILTER_START);
592
593         revision = readl_relaxed(base + L2X0_CACHE_ID) &
594                                  L2X0_CACHE_ID_RTL_MASK;
595
596         if (revision >= L310_CACHE_ID_RTL_R2P0)
597                 l2c_write_sec(l2x0_saved_regs.prefetch_ctrl, base,
598                               L310_PREFETCH_CTRL);
599         if (revision >= L310_CACHE_ID_RTL_R3P0)
600                 l2c_write_sec(l2x0_saved_regs.pwr_ctrl, base,
601                               L310_POWER_CTRL);
602 }
603
604 static int l2c310_starting_cpu(unsigned int cpu)
605 {
606         set_auxcr(get_auxcr() | BIT(3) | BIT(2) | BIT(1));
607         return 0;
608 }
609
610 static int l2c310_dying_cpu(unsigned int cpu)
611 {
612         set_auxcr(get_auxcr() & ~(BIT(3) | BIT(2) | BIT(1)));
613         return 0;
614 }
615
616 static void __init l2c310_enable(void __iomem *base, unsigned num_lock)
617 {
618         unsigned rev = readl_relaxed(base + L2X0_CACHE_ID) & L2X0_CACHE_ID_RTL_MASK;
619         bool cortex_a9 = read_cpuid_part() == ARM_CPU_PART_CORTEX_A9;
620         u32 aux = l2x0_saved_regs.aux_ctrl;
621
622         if (rev >= L310_CACHE_ID_RTL_R2P0) {
623                 if (cortex_a9) {
624                         aux |= L310_AUX_CTRL_EARLY_BRESP;
625                         pr_info("L2C-310 enabling early BRESP for Cortex-A9\n");
626                 } else if (aux & L310_AUX_CTRL_EARLY_BRESP) {
627                         pr_warn("L2C-310 early BRESP only supported with Cortex-A9\n");
628                         aux &= ~L310_AUX_CTRL_EARLY_BRESP;
629                 }
630         }
631
632         if (cortex_a9) {
633                 u32 aux_cur = readl_relaxed(base + L2X0_AUX_CTRL);
634                 u32 acr = get_auxcr();
635
636                 pr_debug("Cortex-A9 ACR=0x%08x\n", acr);
637
638                 if (acr & BIT(3) && !(aux_cur & L310_AUX_CTRL_FULL_LINE_ZERO))
639                         pr_err("L2C-310: full line of zeros enabled in Cortex-A9 but not L2C-310 - invalid\n");
640
641                 if (aux & L310_AUX_CTRL_FULL_LINE_ZERO && !(acr & BIT(3)))
642                         pr_err("L2C-310: enabling full line of zeros but not enabled in Cortex-A9\n");
643
644                 if (!(aux & L310_AUX_CTRL_FULL_LINE_ZERO) && !outer_cache.write_sec) {
645                         aux |= L310_AUX_CTRL_FULL_LINE_ZERO;
646                         pr_info("L2C-310 full line of zeros enabled for Cortex-A9\n");
647                 }
648         } else if (aux & (L310_AUX_CTRL_FULL_LINE_ZERO | L310_AUX_CTRL_EARLY_BRESP)) {
649                 pr_err("L2C-310: disabling Cortex-A9 specific feature bits\n");
650                 aux &= ~(L310_AUX_CTRL_FULL_LINE_ZERO | L310_AUX_CTRL_EARLY_BRESP);
651         }
652
653         /*
654          * Always enable non-secure access to the lockdown registers -
655          * we write to them as part of the L2C enable sequence so they
656          * need to be accessible.
657          */
658         l2x0_saved_regs.aux_ctrl = aux | L310_AUX_CTRL_NS_LOCKDOWN;
659
660         l2c_enable(base, num_lock);
661
662         /* Read back resulting AUX_CTRL value as it could have been altered. */
663         aux = readl_relaxed(base + L2X0_AUX_CTRL);
664
665         if (aux & (L310_AUX_CTRL_DATA_PREFETCH | L310_AUX_CTRL_INSTR_PREFETCH)) {
666                 u32 prefetch = readl_relaxed(base + L310_PREFETCH_CTRL);
667
668                 pr_info("L2C-310 %s%s prefetch enabled, offset %u lines\n",
669                         aux & L310_AUX_CTRL_INSTR_PREFETCH ? "I" : "",
670                         aux & L310_AUX_CTRL_DATA_PREFETCH ? "D" : "",
671                         1 + (prefetch & L310_PREFETCH_CTRL_OFFSET_MASK));
672         }
673
674         /* r3p0 or later has power control register */
675         if (rev >= L310_CACHE_ID_RTL_R3P0) {
676                 u32 power_ctrl;
677
678                 power_ctrl = readl_relaxed(base + L310_POWER_CTRL);
679                 pr_info("L2C-310 dynamic clock gating %sabled, standby mode %sabled\n",
680                         power_ctrl & L310_DYNAMIC_CLK_GATING_EN ? "en" : "dis",
681                         power_ctrl & L310_STNDBY_MODE_EN ? "en" : "dis");
682         }
683
684         if (aux & L310_AUX_CTRL_FULL_LINE_ZERO)
685                 cpuhp_setup_state(CPUHP_AP_ARM_L2X0_STARTING,
686                                   "AP_ARM_L2X0_STARTING", l2c310_starting_cpu,
687                                   l2c310_dying_cpu);
688 }
689
690 static void __init l2c310_fixup(void __iomem *base, u32 cache_id,
691         struct outer_cache_fns *fns)
692 {
693         unsigned revision = cache_id & L2X0_CACHE_ID_RTL_MASK;
694         const char *errata[8];
695         unsigned n = 0;
696
697         if (IS_ENABLED(CONFIG_PL310_ERRATA_588369) &&
698             revision < L310_CACHE_ID_RTL_R2P0 &&
699             /* For bcm compatibility */
700             fns->inv_range == l2c210_inv_range) {
701                 fns->inv_range = l2c310_inv_range_erratum;
702                 fns->flush_range = l2c310_flush_range_erratum;
703                 errata[n++] = "588369";
704         }
705
706         if (IS_ENABLED(CONFIG_PL310_ERRATA_727915) &&
707             revision >= L310_CACHE_ID_RTL_R2P0 &&
708             revision < L310_CACHE_ID_RTL_R3P1) {
709                 fns->flush_all = l2c310_flush_all_erratum;
710                 errata[n++] = "727915";
711         }
712
713         if (revision >= L310_CACHE_ID_RTL_R3P0 &&
714             revision < L310_CACHE_ID_RTL_R3P2) {
715                 u32 val = l2x0_saved_regs.prefetch_ctrl;
716                 if (val & L310_PREFETCH_CTRL_DBL_LINEFILL) {
717                         val &= ~L310_PREFETCH_CTRL_DBL_LINEFILL;
718                         l2x0_saved_regs.prefetch_ctrl = val;
719                         errata[n++] = "752271";
720                 }
721         }
722
723         if (IS_ENABLED(CONFIG_PL310_ERRATA_753970) &&
724             revision == L310_CACHE_ID_RTL_R3P0) {
725                 sync_reg_offset = L2X0_DUMMY_REG;
726                 errata[n++] = "753970";
727         }
728
729         if (IS_ENABLED(CONFIG_PL310_ERRATA_769419))
730                 errata[n++] = "769419";
731
732         if (n) {
733                 unsigned i;
734
735                 pr_info("L2C-310 errat%s", n > 1 ? "a" : "um");
736                 for (i = 0; i < n; i++)
737                         pr_cont(" %s", errata[i]);
738                 pr_cont(" enabled\n");
739         }
740 }
741
742 static void l2c310_disable(void)
743 {
744         /*
745          * If full-line-of-zeros is enabled, we must first disable it in the
746          * Cortex-A9 auxiliary control register before disabling the L2 cache.
747          */
748         if (l2x0_saved_regs.aux_ctrl & L310_AUX_CTRL_FULL_LINE_ZERO)
749                 set_auxcr(get_auxcr() & ~(BIT(3) | BIT(2) | BIT(1)));
750
751         l2c_disable();
752 }
753
754 static void l2c310_resume(void)
755 {
756         l2c_resume();
757
758         /* Re-enable full-line-of-zeros for Cortex-A9 */
759         if (l2x0_saved_regs.aux_ctrl & L310_AUX_CTRL_FULL_LINE_ZERO)
760                 set_auxcr(get_auxcr() | BIT(3) | BIT(2) | BIT(1));
761 }
762
763 static void l2c310_unlock(void __iomem *base, unsigned num_lock)
764 {
765         if (readl_relaxed(base + L2X0_AUX_CTRL) & L310_AUX_CTRL_NS_LOCKDOWN)
766                 l2c_unlock(base, num_lock);
767 }
768
769 static const struct l2c_init_data l2c310_init_fns __initconst = {
770         .type = "L2C-310",
771         .way_size_0 = SZ_8K,
772         .num_lock = 8,
773         .enable = l2c310_enable,
774         .fixup = l2c310_fixup,
775         .save = l2c310_save,
776         .configure = l2c310_configure,
777         .unlock = l2c310_unlock,
778         .outer_cache = {
779                 .inv_range = l2c210_inv_range,
780                 .clean_range = l2c210_clean_range,
781                 .flush_range = l2c210_flush_range,
782                 .flush_all = l2c210_flush_all,
783                 .disable = l2c310_disable,
784                 .sync = l2c210_sync,
785                 .resume = l2c310_resume,
786         },
787 };
788
789 static int __init __l2c_init(const struct l2c_init_data *data,
790                              u32 aux_val, u32 aux_mask, u32 cache_id, bool nosync)
791 {
792         struct outer_cache_fns fns;
793         unsigned way_size_bits, ways;
794         u32 aux, old_aux;
795
796         /*
797          * Save the pointer globally so that callbacks which do not receive
798          * context from callers can access the structure.
799          */
800         l2x0_data = kmemdup(data, sizeof(*data), GFP_KERNEL);
801         if (!l2x0_data)
802                 return -ENOMEM;
803
804         /*
805          * Sanity check the aux values.  aux_mask is the bits we preserve
806          * from reading the hardware register, and aux_val is the bits we
807          * set.
808          */
809         if (aux_val & aux_mask)
810                 pr_alert("L2C: platform provided aux values permit register corruption.\n");
811
812         old_aux = aux = readl_relaxed(l2x0_base + L2X0_AUX_CTRL);
813         aux &= aux_mask;
814         aux |= aux_val;
815
816         if (old_aux != aux)
817                 pr_warn("L2C: DT/platform modifies aux control register: 0x%08x -> 0x%08x\n",
818                         old_aux, aux);
819
820         /* Determine the number of ways */
821         switch (cache_id & L2X0_CACHE_ID_PART_MASK) {
822         case L2X0_CACHE_ID_PART_L310:
823                 if ((aux_val | ~aux_mask) & (L2C_AUX_CTRL_WAY_SIZE_MASK | L310_AUX_CTRL_ASSOCIATIVITY_16))
824                         pr_warn("L2C: DT/platform tries to modify or specify cache size\n");
825                 if (aux & (1 << 16))
826                         ways = 16;
827                 else
828                         ways = 8;
829                 break;
830
831         case L2X0_CACHE_ID_PART_L210:
832         case L2X0_CACHE_ID_PART_L220:
833                 ways = (aux >> 13) & 0xf;
834                 break;
835
836         case AURORA_CACHE_ID:
837                 ways = (aux >> 13) & 0xf;
838                 ways = 2 << ((ways + 1) >> 2);
839                 break;
840
841         default:
842                 /* Assume unknown chips have 8 ways */
843                 ways = 8;
844                 break;
845         }
846
847         l2x0_way_mask = (1 << ways) - 1;
848
849         /*
850          * way_size_0 is the size that a way_size value of zero would be
851          * given the calculation: way_size = way_size_0 << way_size_bits.
852          * So, if way_size_bits=0 is reserved, but way_size_bits=1 is 16k,
853          * then way_size_0 would be 8k.
854          *
855          * L2 cache size = number of ways * way size.
856          */
857         way_size_bits = (aux & L2C_AUX_CTRL_WAY_SIZE_MASK) >>
858                         L2C_AUX_CTRL_WAY_SIZE_SHIFT;
859         l2x0_size = ways * (data->way_size_0 << way_size_bits);
860
861         fns = data->outer_cache;
862         fns.write_sec = outer_cache.write_sec;
863         fns.configure = outer_cache.configure;
864         if (data->fixup)
865                 data->fixup(l2x0_base, cache_id, &fns);
866         if (nosync) {
867                 pr_info("L2C: disabling outer sync\n");
868                 fns.sync = NULL;
869         }
870
871         /*
872          * Check if l2x0 controller is already enabled.  If we are booting
873          * in non-secure mode accessing the below registers will fault.
874          */
875         if (!(readl_relaxed(l2x0_base + L2X0_CTRL) & L2X0_CTRL_EN)) {
876                 l2x0_saved_regs.aux_ctrl = aux;
877
878                 data->enable(l2x0_base, data->num_lock);
879         }
880
881         outer_cache = fns;
882
883         /*
884          * It is strange to save the register state before initialisation,
885          * but hey, this is what the DT implementations decided to do.
886          */
887         if (data->save)
888                 data->save(l2x0_base);
889
890         /* Re-read it in case some bits are reserved. */
891         aux = readl_relaxed(l2x0_base + L2X0_AUX_CTRL);
892
893         pr_info("%s cache controller enabled, %d ways, %d kB\n",
894                 data->type, ways, l2x0_size >> 10);
895         pr_info("%s: CACHE_ID 0x%08x, AUX_CTRL 0x%08x\n",
896                 data->type, cache_id, aux);
897
898         l2x0_pmu_register(l2x0_base, cache_id);
899
900         return 0;
901 }
902
903 void __init l2x0_init(void __iomem *base, u32 aux_val, u32 aux_mask)
904 {
905         const struct l2c_init_data *data;
906         u32 cache_id;
907
908         l2x0_base = base;
909
910         cache_id = readl_relaxed(base + L2X0_CACHE_ID);
911
912         switch (cache_id & L2X0_CACHE_ID_PART_MASK) {
913         default:
914         case L2X0_CACHE_ID_PART_L210:
915                 data = &l2c210_data;
916                 break;
917
918         case L2X0_CACHE_ID_PART_L220:
919                 data = &l2c220_data;
920                 break;
921
922         case L2X0_CACHE_ID_PART_L310:
923                 data = &l2c310_init_fns;
924                 break;
925         }
926
927         /* Read back current (default) hardware configuration */
928         if (data->save)
929                 data->save(l2x0_base);
930
931         __l2c_init(data, aux_val, aux_mask, cache_id, false);
932 }
933
934 #ifdef CONFIG_OF
935 static int l2_wt_override;
936
937 /* Aurora don't have the cache ID register available, so we have to
938  * pass it though the device tree */
939 static u32 cache_id_part_number_from_dt;
940
941 /**
942  * l2x0_cache_size_of_parse() - read cache size parameters from DT
943  * @np: the device tree node for the l2 cache
944  * @aux_val: pointer to machine-supplied auxilary register value, to
945  * be augmented by the call (bits to be set to 1)
946  * @aux_mask: pointer to machine-supplied auxilary register mask, to
947  * be augmented by the call (bits to be set to 0)
948  * @associativity: variable to return the calculated associativity in
949  * @max_way_size: the maximum size in bytes for the cache ways
950  */
951 static int __init l2x0_cache_size_of_parse(const struct device_node *np,
952                                             u32 *aux_val, u32 *aux_mask,
953                                             u32 *associativity,
954                                             u32 max_way_size)
955 {
956         u32 mask = 0, val = 0;
957         u32 cache_size = 0, sets = 0;
958         u32 way_size_bits = 1;
959         u32 way_size = 0;
960         u32 block_size = 0;
961         u32 line_size = 0;
962
963         of_property_read_u32(np, "cache-size", &cache_size);
964         of_property_read_u32(np, "cache-sets", &sets);
965         of_property_read_u32(np, "cache-block-size", &block_size);
966         of_property_read_u32(np, "cache-line-size", &line_size);
967
968         if (!cache_size || !sets)
969                 return -ENODEV;
970
971         /* All these l2 caches have the same line = block size actually */
972         if (!line_size) {
973                 if (block_size) {
974                         /* If linesize is not given, it is equal to blocksize */
975                         line_size = block_size;
976                 } else {
977                         /* Fall back to known size */
978                         pr_warn("L2C OF: no cache block/line size given: "
979                                 "falling back to default size %d bytes\n",
980                                 CACHE_LINE_SIZE);
981                         line_size = CACHE_LINE_SIZE;
982                 }
983         }
984
985         if (line_size != CACHE_LINE_SIZE)
986                 pr_warn("L2C OF: DT supplied line size %d bytes does "
987                         "not match hardware line size of %d bytes\n",
988                         line_size,
989                         CACHE_LINE_SIZE);
990
991         /*
992          * Since:
993          * set size = cache size / sets
994          * ways = cache size / (sets * line size)
995          * way size = cache size / (cache size / (sets * line size))
996          * way size = sets * line size
997          * associativity = ways = cache size / way size
998          */
999         way_size = sets * line_size;
1000         *associativity = cache_size / way_size;
1001
1002         if (way_size > max_way_size) {
1003                 pr_err("L2C OF: set size %dKB is too large\n", way_size);
1004                 return -EINVAL;
1005         }
1006
1007         pr_info("L2C OF: override cache size: %d bytes (%dKB)\n",
1008                 cache_size, cache_size >> 10);
1009         pr_info("L2C OF: override line size: %d bytes\n", line_size);
1010         pr_info("L2C OF: override way size: %d bytes (%dKB)\n",
1011                 way_size, way_size >> 10);
1012         pr_info("L2C OF: override associativity: %d\n", *associativity);
1013
1014         /*
1015          * Calculates the bits 17:19 to set for way size:
1016          * 512KB -> 6, 256KB -> 5, ... 16KB -> 1
1017          */
1018         way_size_bits = ilog2(way_size >> 10) - 3;
1019         if (way_size_bits < 1 || way_size_bits > 6) {
1020                 pr_err("L2C OF: cache way size illegal: %dKB is not mapped\n",
1021                        way_size);
1022                 return -EINVAL;
1023         }
1024
1025         mask |= L2C_AUX_CTRL_WAY_SIZE_MASK;
1026         val |= (way_size_bits << L2C_AUX_CTRL_WAY_SIZE_SHIFT);
1027
1028         *aux_val &= ~mask;
1029         *aux_val |= val;
1030         *aux_mask &= ~mask;
1031
1032         return 0;
1033 }
1034
1035 static void __init l2x0_of_parse(const struct device_node *np,
1036                                  u32 *aux_val, u32 *aux_mask)
1037 {
1038         u32 data[2] = { 0, 0 };
1039         u32 tag = 0;
1040         u32 dirty = 0;
1041         u32 val = 0, mask = 0;
1042         u32 assoc;
1043         int ret;
1044
1045         of_property_read_u32(np, "arm,tag-latency", &tag);
1046         if (tag) {
1047                 mask |= L2X0_AUX_CTRL_TAG_LATENCY_MASK;
1048                 val |= (tag - 1) << L2X0_AUX_CTRL_TAG_LATENCY_SHIFT;
1049         }
1050
1051         of_property_read_u32_array(np, "arm,data-latency",
1052                                    data, ARRAY_SIZE(data));
1053         if (data[0] && data[1]) {
1054                 mask |= L2X0_AUX_CTRL_DATA_RD_LATENCY_MASK |
1055                         L2X0_AUX_CTRL_DATA_WR_LATENCY_MASK;
1056                 val |= ((data[0] - 1) << L2X0_AUX_CTRL_DATA_RD_LATENCY_SHIFT) |
1057                        ((data[1] - 1) << L2X0_AUX_CTRL_DATA_WR_LATENCY_SHIFT);
1058         }
1059
1060         of_property_read_u32(np, "arm,dirty-latency", &dirty);
1061         if (dirty) {
1062                 mask |= L2X0_AUX_CTRL_DIRTY_LATENCY_MASK;
1063                 val |= (dirty - 1) << L2X0_AUX_CTRL_DIRTY_LATENCY_SHIFT;
1064         }
1065
1066         if (of_property_read_bool(np, "arm,parity-enable")) {
1067                 mask &= ~L2C_AUX_CTRL_PARITY_ENABLE;
1068                 val |= L2C_AUX_CTRL_PARITY_ENABLE;
1069         } else if (of_property_read_bool(np, "arm,parity-disable")) {
1070                 mask &= ~L2C_AUX_CTRL_PARITY_ENABLE;
1071         }
1072
1073         if (of_property_read_bool(np, "arm,shared-override")) {
1074                 mask &= ~L2C_AUX_CTRL_SHARED_OVERRIDE;
1075                 val |= L2C_AUX_CTRL_SHARED_OVERRIDE;
1076         }
1077
1078         ret = l2x0_cache_size_of_parse(np, aux_val, aux_mask, &assoc, SZ_256K);
1079         if (ret)
1080                 return;
1081
1082         if (assoc > 8) {
1083                 pr_err("l2x0 of: cache setting yield too high associativity\n");
1084                 pr_err("l2x0 of: %d calculated, max 8\n", assoc);
1085         } else {
1086                 mask |= L2X0_AUX_CTRL_ASSOC_MASK;
1087                 val |= (assoc << L2X0_AUX_CTRL_ASSOC_SHIFT);
1088         }
1089
1090         *aux_val &= ~mask;
1091         *aux_val |= val;
1092         *aux_mask &= ~mask;
1093 }
1094
1095 static const struct l2c_init_data of_l2c210_data __initconst = {
1096         .type = "L2C-210",
1097         .way_size_0 = SZ_8K,
1098         .num_lock = 1,
1099         .of_parse = l2x0_of_parse,
1100         .enable = l2c_enable,
1101         .save = l2c_save,
1102         .configure = l2c_configure,
1103         .unlock = l2c_unlock,
1104         .outer_cache = {
1105                 .inv_range   = l2c210_inv_range,
1106                 .clean_range = l2c210_clean_range,
1107                 .flush_range = l2c210_flush_range,
1108                 .flush_all   = l2c210_flush_all,
1109                 .disable     = l2c_disable,
1110                 .sync        = l2c210_sync,
1111                 .resume      = l2c_resume,
1112         },
1113 };
1114
1115 static const struct l2c_init_data of_l2c220_data __initconst = {
1116         .type = "L2C-220",
1117         .way_size_0 = SZ_8K,
1118         .num_lock = 1,
1119         .of_parse = l2x0_of_parse,
1120         .enable = l2c220_enable,
1121         .save = l2c_save,
1122         .configure = l2c_configure,
1123         .unlock = l2c220_unlock,
1124         .outer_cache = {
1125                 .inv_range   = l2c220_inv_range,
1126                 .clean_range = l2c220_clean_range,
1127                 .flush_range = l2c220_flush_range,
1128                 .flush_all   = l2c220_flush_all,
1129                 .disable     = l2c_disable,
1130                 .sync        = l2c220_sync,
1131                 .resume      = l2c_resume,
1132         },
1133 };
1134
1135 static void __init l2c310_of_parse(const struct device_node *np,
1136         u32 *aux_val, u32 *aux_mask)
1137 {
1138         u32 data[3] = { 0, 0, 0 };
1139         u32 tag[3] = { 0, 0, 0 };
1140         u32 filter[2] = { 0, 0 };
1141         u32 assoc;
1142         u32 prefetch;
1143         u32 power;
1144         u32 val;
1145         int ret;
1146
1147         of_property_read_u32_array(np, "arm,tag-latency", tag, ARRAY_SIZE(tag));
1148         if (tag[0] && tag[1] && tag[2])
1149                 l2x0_saved_regs.tag_latency =
1150                         L310_LATENCY_CTRL_RD(tag[0] - 1) |
1151                         L310_LATENCY_CTRL_WR(tag[1] - 1) |
1152                         L310_LATENCY_CTRL_SETUP(tag[2] - 1);
1153
1154         of_property_read_u32_array(np, "arm,data-latency",
1155                                    data, ARRAY_SIZE(data));
1156         if (data[0] && data[1] && data[2])
1157                 l2x0_saved_regs.data_latency =
1158                         L310_LATENCY_CTRL_RD(data[0] - 1) |
1159                         L310_LATENCY_CTRL_WR(data[1] - 1) |
1160                         L310_LATENCY_CTRL_SETUP(data[2] - 1);
1161
1162         of_property_read_u32_array(np, "arm,filter-ranges",
1163                                    filter, ARRAY_SIZE(filter));
1164         if (filter[1]) {
1165                 l2x0_saved_regs.filter_end =
1166                                         ALIGN(filter[0] + filter[1], SZ_1M);
1167                 l2x0_saved_regs.filter_start = (filter[0] & ~(SZ_1M - 1))
1168                                         | L310_ADDR_FILTER_EN;
1169         }
1170
1171         ret = l2x0_cache_size_of_parse(np, aux_val, aux_mask, &assoc, SZ_512K);
1172         if (!ret) {
1173                 switch (assoc) {
1174                 case 16:
1175                         *aux_val &= ~L2X0_AUX_CTRL_ASSOC_MASK;
1176                         *aux_val |= L310_AUX_CTRL_ASSOCIATIVITY_16;
1177                         *aux_mask &= ~L2X0_AUX_CTRL_ASSOC_MASK;
1178                         break;
1179                 case 8:
1180                         *aux_val &= ~L2X0_AUX_CTRL_ASSOC_MASK;
1181                         *aux_mask &= ~L2X0_AUX_CTRL_ASSOC_MASK;
1182                         break;
1183                 default:
1184                         pr_err("L2C-310 OF cache associativity %d invalid, only 8 or 16 permitted\n",
1185                                assoc);
1186                         break;
1187                 }
1188         }
1189
1190         if (of_property_read_bool(np, "arm,shared-override")) {
1191                 *aux_val |= L2C_AUX_CTRL_SHARED_OVERRIDE;
1192                 *aux_mask &= ~L2C_AUX_CTRL_SHARED_OVERRIDE;
1193         }
1194
1195         if (of_property_read_bool(np, "arm,parity-enable")) {
1196                 *aux_val |= L2C_AUX_CTRL_PARITY_ENABLE;
1197                 *aux_mask &= ~L2C_AUX_CTRL_PARITY_ENABLE;
1198         } else if (of_property_read_bool(np, "arm,parity-disable")) {
1199                 *aux_val &= ~L2C_AUX_CTRL_PARITY_ENABLE;
1200                 *aux_mask &= ~L2C_AUX_CTRL_PARITY_ENABLE;
1201         }
1202
1203         prefetch = l2x0_saved_regs.prefetch_ctrl;
1204
1205         ret = of_property_read_u32(np, "arm,double-linefill", &val);
1206         if (ret == 0) {
1207                 if (val)
1208                         prefetch |= L310_PREFETCH_CTRL_DBL_LINEFILL;
1209                 else
1210                         prefetch &= ~L310_PREFETCH_CTRL_DBL_LINEFILL;
1211         } else if (ret != -EINVAL) {
1212                 pr_err("L2C-310 OF arm,double-linefill property value is missing\n");
1213         }
1214
1215         ret = of_property_read_u32(np, "arm,double-linefill-incr", &val);
1216         if (ret == 0) {
1217                 if (val)
1218                         prefetch |= L310_PREFETCH_CTRL_DBL_LINEFILL_INCR;
1219                 else
1220                         prefetch &= ~L310_PREFETCH_CTRL_DBL_LINEFILL_INCR;
1221         } else if (ret != -EINVAL) {
1222                 pr_err("L2C-310 OF arm,double-linefill-incr property value is missing\n");
1223         }
1224
1225         ret = of_property_read_u32(np, "arm,double-linefill-wrap", &val);
1226         if (ret == 0) {
1227                 if (!val)
1228                         prefetch |= L310_PREFETCH_CTRL_DBL_LINEFILL_WRAP;
1229                 else
1230                         prefetch &= ~L310_PREFETCH_CTRL_DBL_LINEFILL_WRAP;
1231         } else if (ret != -EINVAL) {
1232                 pr_err("L2C-310 OF arm,double-linefill-wrap property value is missing\n");
1233         }
1234
1235         ret = of_property_read_u32(np, "arm,prefetch-drop", &val);
1236         if (ret == 0) {
1237                 if (val)
1238                         prefetch |= L310_PREFETCH_CTRL_PREFETCH_DROP;
1239                 else
1240                         prefetch &= ~L310_PREFETCH_CTRL_PREFETCH_DROP;
1241         } else if (ret != -EINVAL) {
1242                 pr_err("L2C-310 OF arm,prefetch-drop property value is missing\n");
1243         }
1244
1245         ret = of_property_read_u32(np, "arm,prefetch-offset", &val);
1246         if (ret == 0) {
1247                 prefetch &= ~L310_PREFETCH_CTRL_OFFSET_MASK;
1248                 prefetch |= val & L310_PREFETCH_CTRL_OFFSET_MASK;
1249         } else if (ret != -EINVAL) {
1250                 pr_err("L2C-310 OF arm,prefetch-offset property value is missing\n");
1251         }
1252
1253         ret = of_property_read_u32(np, "prefetch-data", &val);
1254         if (ret == 0) {
1255                 if (val)
1256                         prefetch |= L310_PREFETCH_CTRL_DATA_PREFETCH;
1257                 else
1258                         prefetch &= ~L310_PREFETCH_CTRL_DATA_PREFETCH;
1259         } else if (ret != -EINVAL) {
1260                 pr_err("L2C-310 OF prefetch-data property value is missing\n");
1261         }
1262
1263         ret = of_property_read_u32(np, "prefetch-instr", &val);
1264         if (ret == 0) {
1265                 if (val)
1266                         prefetch |= L310_PREFETCH_CTRL_INSTR_PREFETCH;
1267                 else
1268                         prefetch &= ~L310_PREFETCH_CTRL_INSTR_PREFETCH;
1269         } else if (ret != -EINVAL) {
1270                 pr_err("L2C-310 OF prefetch-instr property value is missing\n");
1271         }
1272
1273         l2x0_saved_regs.prefetch_ctrl = prefetch;
1274
1275         power = l2x0_saved_regs.pwr_ctrl |
1276                 L310_DYNAMIC_CLK_GATING_EN | L310_STNDBY_MODE_EN;
1277
1278         ret = of_property_read_u32(np, "arm,dynamic-clock-gating", &val);
1279         if (!ret) {
1280                 if (!val)
1281                         power &= ~L310_DYNAMIC_CLK_GATING_EN;
1282         } else if (ret != -EINVAL) {
1283                 pr_err("L2C-310 OF dynamic-clock-gating property value is missing or invalid\n");
1284         }
1285         ret = of_property_read_u32(np, "arm,standby-mode", &val);
1286         if (!ret) {
1287                 if (!val)
1288                         power &= ~L310_STNDBY_MODE_EN;
1289         } else if (ret != -EINVAL) {
1290                 pr_err("L2C-310 OF standby-mode property value is missing or invalid\n");
1291         }
1292
1293         l2x0_saved_regs.pwr_ctrl = power;
1294 }
1295
1296 static const struct l2c_init_data of_l2c310_data __initconst = {
1297         .type = "L2C-310",
1298         .way_size_0 = SZ_8K,
1299         .num_lock = 8,
1300         .of_parse = l2c310_of_parse,
1301         .enable = l2c310_enable,
1302         .fixup = l2c310_fixup,
1303         .save  = l2c310_save,
1304         .configure = l2c310_configure,
1305         .unlock = l2c310_unlock,
1306         .outer_cache = {
1307                 .inv_range   = l2c210_inv_range,
1308                 .clean_range = l2c210_clean_range,
1309                 .flush_range = l2c210_flush_range,
1310                 .flush_all   = l2c210_flush_all,
1311                 .disable     = l2c310_disable,
1312                 .sync        = l2c210_sync,
1313                 .resume      = l2c310_resume,
1314         },
1315 };
1316
1317 /*
1318  * This is a variant of the of_l2c310_data with .sync set to
1319  * NULL. Outer sync operations are not needed when the system is I/O
1320  * coherent, and potentially harmful in certain situations (PCIe/PL310
1321  * deadlock on Armada 375/38x due to hardware I/O coherency). The
1322  * other operations are kept because they are infrequent (therefore do
1323  * not cause the deadlock in practice) and needed for secondary CPU
1324  * boot and other power management activities.
1325  */
1326 static const struct l2c_init_data of_l2c310_coherent_data __initconst = {
1327         .type = "L2C-310 Coherent",
1328         .way_size_0 = SZ_8K,
1329         .num_lock = 8,
1330         .of_parse = l2c310_of_parse,
1331         .enable = l2c310_enable,
1332         .fixup = l2c310_fixup,
1333         .save  = l2c310_save,
1334         .configure = l2c310_configure,
1335         .unlock = l2c310_unlock,
1336         .outer_cache = {
1337                 .inv_range   = l2c210_inv_range,
1338                 .clean_range = l2c210_clean_range,
1339                 .flush_range = l2c210_flush_range,
1340                 .flush_all   = l2c210_flush_all,
1341                 .disable     = l2c310_disable,
1342                 .resume      = l2c310_resume,
1343         },
1344 };
1345
1346 /*
1347  * Note that the end addresses passed to Linux primitives are
1348  * noninclusive, while the hardware cache range operations use
1349  * inclusive start and end addresses.
1350  */
1351 static unsigned long aurora_range_end(unsigned long start, unsigned long end)
1352 {
1353         /*
1354          * Limit the number of cache lines processed at once,
1355          * since cache range operations stall the CPU pipeline
1356          * until completion.
1357          */
1358         if (end > start + MAX_RANGE_SIZE)
1359                 end = start + MAX_RANGE_SIZE;
1360
1361         /*
1362          * Cache range operations can't straddle a page boundary.
1363          */
1364         if (end > PAGE_ALIGN(start+1))
1365                 end = PAGE_ALIGN(start+1);
1366
1367         return end;
1368 }
1369
1370 static void aurora_pa_range(unsigned long start, unsigned long end,
1371                             unsigned long offset)
1372 {
1373         void __iomem *base = l2x0_base;
1374         unsigned long range_end;
1375         unsigned long flags;
1376
1377         /*
1378          * round start and end adresses up to cache line size
1379          */
1380         start &= ~(CACHE_LINE_SIZE - 1);
1381         end = ALIGN(end, CACHE_LINE_SIZE);
1382
1383         /*
1384          * perform operation on all full cache lines between 'start' and 'end'
1385          */
1386         while (start < end) {
1387                 range_end = aurora_range_end(start, end);
1388
1389                 raw_spin_lock_irqsave(&l2x0_lock, flags);
1390                 writel_relaxed(start, base + AURORA_RANGE_BASE_ADDR_REG);
1391                 writel_relaxed(range_end - CACHE_LINE_SIZE, base + offset);
1392                 raw_spin_unlock_irqrestore(&l2x0_lock, flags);
1393
1394                 writel_relaxed(0, base + AURORA_SYNC_REG);
1395                 start = range_end;
1396         }
1397 }
1398 static void aurora_inv_range(unsigned long start, unsigned long end)
1399 {
1400         aurora_pa_range(start, end, AURORA_INVAL_RANGE_REG);
1401 }
1402
1403 static void aurora_clean_range(unsigned long start, unsigned long end)
1404 {
1405         /*
1406          * If L2 is forced to WT, the L2 will always be clean and we
1407          * don't need to do anything here.
1408          */
1409         if (!l2_wt_override)
1410                 aurora_pa_range(start, end, AURORA_CLEAN_RANGE_REG);
1411 }
1412
1413 static void aurora_flush_range(unsigned long start, unsigned long end)
1414 {
1415         if (l2_wt_override)
1416                 aurora_pa_range(start, end, AURORA_INVAL_RANGE_REG);
1417         else
1418                 aurora_pa_range(start, end, AURORA_FLUSH_RANGE_REG);
1419 }
1420
1421 static void aurora_flush_all(void)
1422 {
1423         void __iomem *base = l2x0_base;
1424         unsigned long flags;
1425
1426         /* clean all ways */
1427         raw_spin_lock_irqsave(&l2x0_lock, flags);
1428         __l2c_op_way(base + L2X0_CLEAN_INV_WAY);
1429         raw_spin_unlock_irqrestore(&l2x0_lock, flags);
1430
1431         writel_relaxed(0, base + AURORA_SYNC_REG);
1432 }
1433
1434 static void aurora_cache_sync(void)
1435 {
1436         writel_relaxed(0, l2x0_base + AURORA_SYNC_REG);
1437 }
1438
1439 static void aurora_disable(void)
1440 {
1441         void __iomem *base = l2x0_base;
1442         unsigned long flags;
1443
1444         raw_spin_lock_irqsave(&l2x0_lock, flags);
1445         __l2c_op_way(base + L2X0_CLEAN_INV_WAY);
1446         writel_relaxed(0, base + AURORA_SYNC_REG);
1447         l2c_write_sec(0, base, L2X0_CTRL);
1448         dsb(st);
1449         raw_spin_unlock_irqrestore(&l2x0_lock, flags);
1450 }
1451
1452 static void aurora_save(void __iomem *base)
1453 {
1454         l2x0_saved_regs.ctrl = readl_relaxed(base + L2X0_CTRL);
1455         l2x0_saved_regs.aux_ctrl = readl_relaxed(base + L2X0_AUX_CTRL);
1456 }
1457
1458 /*
1459  * For Aurora cache in no outer mode, enable via the CP15 coprocessor
1460  * broadcasting of cache commands to L2.
1461  */
1462 static void __init aurora_enable_no_outer(void __iomem *base,
1463         unsigned num_lock)
1464 {
1465         u32 u;
1466
1467         asm volatile("mrc p15, 1, %0, c15, c2, 0" : "=r" (u));
1468         u |= AURORA_CTRL_FW;            /* Set the FW bit */
1469         asm volatile("mcr p15, 1, %0, c15, c2, 0" : : "r" (u));
1470
1471         isb();
1472
1473         l2c_enable(base, num_lock);
1474 }
1475
1476 static void __init aurora_fixup(void __iomem *base, u32 cache_id,
1477         struct outer_cache_fns *fns)
1478 {
1479         sync_reg_offset = AURORA_SYNC_REG;
1480 }
1481
1482 static void __init aurora_of_parse(const struct device_node *np,
1483                                 u32 *aux_val, u32 *aux_mask)
1484 {
1485         u32 val = AURORA_ACR_REPLACEMENT_TYPE_SEMIPLRU;
1486         u32 mask =  AURORA_ACR_REPLACEMENT_MASK;
1487
1488         of_property_read_u32(np, "cache-id-part",
1489                         &cache_id_part_number_from_dt);
1490
1491         /* Determine and save the write policy */
1492         l2_wt_override = of_property_read_bool(np, "wt-override");
1493
1494         if (l2_wt_override) {
1495                 val |= AURORA_ACR_FORCE_WRITE_THRO_POLICY;
1496                 mask |= AURORA_ACR_FORCE_WRITE_POLICY_MASK;
1497         }
1498
1499         *aux_val &= ~mask;
1500         *aux_val |= val;
1501         *aux_mask &= ~mask;
1502 }
1503
1504 static const struct l2c_init_data of_aurora_with_outer_data __initconst = {
1505         .type = "Aurora",
1506         .way_size_0 = SZ_4K,
1507         .num_lock = 4,
1508         .of_parse = aurora_of_parse,
1509         .enable = l2c_enable,
1510         .fixup = aurora_fixup,
1511         .save  = aurora_save,
1512         .configure = l2c_configure,
1513         .unlock = l2c_unlock,
1514         .outer_cache = {
1515                 .inv_range   = aurora_inv_range,
1516                 .clean_range = aurora_clean_range,
1517                 .flush_range = aurora_flush_range,
1518                 .flush_all   = aurora_flush_all,
1519                 .disable     = aurora_disable,
1520                 .sync        = aurora_cache_sync,
1521                 .resume      = l2c_resume,
1522         },
1523 };
1524
1525 static const struct l2c_init_data of_aurora_no_outer_data __initconst = {
1526         .type = "Aurora",
1527         .way_size_0 = SZ_4K,
1528         .num_lock = 4,
1529         .of_parse = aurora_of_parse,
1530         .enable = aurora_enable_no_outer,
1531         .fixup = aurora_fixup,
1532         .save  = aurora_save,
1533         .configure = l2c_configure,
1534         .unlock = l2c_unlock,
1535         .outer_cache = {
1536                 .resume      = l2c_resume,
1537         },
1538 };
1539
1540 /*
1541  * For certain Broadcom SoCs, depending on the address range, different offsets
1542  * need to be added to the address before passing it to L2 for
1543  * invalidation/clean/flush
1544  *
1545  * Section Address Range              Offset        EMI
1546  *   1     0x00000000 - 0x3FFFFFFF    0x80000000    VC
1547  *   2     0x40000000 - 0xBFFFFFFF    0x40000000    SYS
1548  *   3     0xC0000000 - 0xFFFFFFFF    0x80000000    VC
1549  *
1550  * When the start and end addresses have crossed two different sections, we
1551  * need to break the L2 operation into two, each within its own section.
1552  * For example, if we need to invalidate addresses starts at 0xBFFF0000 and
1553  * ends at 0xC0001000, we need do invalidate 1) 0xBFFF0000 - 0xBFFFFFFF and 2)
1554  * 0xC0000000 - 0xC0001000
1555  *
1556  * Note 1:
1557  * By breaking a single L2 operation into two, we may potentially suffer some
1558  * performance hit, but keep in mind the cross section case is very rare
1559  *
1560  * Note 2:
1561  * We do not need to handle the case when the start address is in
1562  * Section 1 and the end address is in Section 3, since it is not a valid use
1563  * case
1564  *
1565  * Note 3:
1566  * Section 1 in practical terms can no longer be used on rev A2. Because of
1567  * that the code does not need to handle section 1 at all.
1568  *
1569  */
1570 #define BCM_SYS_EMI_START_ADDR        0x40000000UL
1571 #define BCM_VC_EMI_SEC3_START_ADDR    0xC0000000UL
1572
1573 #define BCM_SYS_EMI_OFFSET            0x40000000UL
1574 #define BCM_VC_EMI_OFFSET             0x80000000UL
1575
1576 static inline int bcm_addr_is_sys_emi(unsigned long addr)
1577 {
1578         return (addr >= BCM_SYS_EMI_START_ADDR) &&
1579                 (addr < BCM_VC_EMI_SEC3_START_ADDR);
1580 }
1581
1582 static inline unsigned long bcm_l2_phys_addr(unsigned long addr)
1583 {
1584         if (bcm_addr_is_sys_emi(addr))
1585                 return addr + BCM_SYS_EMI_OFFSET;
1586         else
1587                 return addr + BCM_VC_EMI_OFFSET;
1588 }
1589
1590 static void bcm_inv_range(unsigned long start, unsigned long end)
1591 {
1592         unsigned long new_start, new_end;
1593
1594         BUG_ON(start < BCM_SYS_EMI_START_ADDR);
1595
1596         if (unlikely(end <= start))
1597                 return;
1598
1599         new_start = bcm_l2_phys_addr(start);
1600         new_end = bcm_l2_phys_addr(end);
1601
1602         /* normal case, no cross section between start and end */
1603         if (likely(bcm_addr_is_sys_emi(end) || !bcm_addr_is_sys_emi(start))) {
1604                 l2c210_inv_range(new_start, new_end);
1605                 return;
1606         }
1607
1608         /* They cross sections, so it can only be a cross from section
1609          * 2 to section 3
1610          */
1611         l2c210_inv_range(new_start,
1612                 bcm_l2_phys_addr(BCM_VC_EMI_SEC3_START_ADDR-1));
1613         l2c210_inv_range(bcm_l2_phys_addr(BCM_VC_EMI_SEC3_START_ADDR),
1614                 new_end);
1615 }
1616
1617 static void bcm_clean_range(unsigned long start, unsigned long end)
1618 {
1619         unsigned long new_start, new_end;
1620
1621         BUG_ON(start < BCM_SYS_EMI_START_ADDR);
1622
1623         if (unlikely(end <= start))
1624                 return;
1625
1626         new_start = bcm_l2_phys_addr(start);
1627         new_end = bcm_l2_phys_addr(end);
1628
1629         /* normal case, no cross section between start and end */
1630         if (likely(bcm_addr_is_sys_emi(end) || !bcm_addr_is_sys_emi(start))) {
1631                 l2c210_clean_range(new_start, new_end);
1632                 return;
1633         }
1634
1635         /* They cross sections, so it can only be a cross from section
1636          * 2 to section 3
1637          */
1638         l2c210_clean_range(new_start,
1639                 bcm_l2_phys_addr(BCM_VC_EMI_SEC3_START_ADDR-1));
1640         l2c210_clean_range(bcm_l2_phys_addr(BCM_VC_EMI_SEC3_START_ADDR),
1641                 new_end);
1642 }
1643
1644 static void bcm_flush_range(unsigned long start, unsigned long end)
1645 {
1646         unsigned long new_start, new_end;
1647
1648         BUG_ON(start < BCM_SYS_EMI_START_ADDR);
1649
1650         if (unlikely(end <= start))
1651                 return;
1652
1653         if ((end - start) >= l2x0_size) {
1654                 outer_cache.flush_all();
1655                 return;
1656         }
1657
1658         new_start = bcm_l2_phys_addr(start);
1659         new_end = bcm_l2_phys_addr(end);
1660
1661         /* normal case, no cross section between start and end */
1662         if (likely(bcm_addr_is_sys_emi(end) || !bcm_addr_is_sys_emi(start))) {
1663                 l2c210_flush_range(new_start, new_end);
1664                 return;
1665         }
1666
1667         /* They cross sections, so it can only be a cross from section
1668          * 2 to section 3
1669          */
1670         l2c210_flush_range(new_start,
1671                 bcm_l2_phys_addr(BCM_VC_EMI_SEC3_START_ADDR-1));
1672         l2c210_flush_range(bcm_l2_phys_addr(BCM_VC_EMI_SEC3_START_ADDR),
1673                 new_end);
1674 }
1675
1676 /* Broadcom L2C-310 start from ARMs R3P2 or later, and require no fixups */
1677 static const struct l2c_init_data of_bcm_l2x0_data __initconst = {
1678         .type = "BCM-L2C-310",
1679         .way_size_0 = SZ_8K,
1680         .num_lock = 8,
1681         .of_parse = l2c310_of_parse,
1682         .enable = l2c310_enable,
1683         .save  = l2c310_save,
1684         .configure = l2c310_configure,
1685         .unlock = l2c310_unlock,
1686         .outer_cache = {
1687                 .inv_range   = bcm_inv_range,
1688                 .clean_range = bcm_clean_range,
1689                 .flush_range = bcm_flush_range,
1690                 .flush_all   = l2c210_flush_all,
1691                 .disable     = l2c310_disable,
1692                 .sync        = l2c210_sync,
1693                 .resume      = l2c310_resume,
1694         },
1695 };
1696
1697 static void __init tauros3_save(void __iomem *base)
1698 {
1699         l2c_save(base);
1700
1701         l2x0_saved_regs.aux2_ctrl =
1702                 readl_relaxed(base + TAUROS3_AUX2_CTRL);
1703         l2x0_saved_regs.prefetch_ctrl =
1704                 readl_relaxed(base + L310_PREFETCH_CTRL);
1705 }
1706
1707 static void tauros3_configure(void __iomem *base)
1708 {
1709         l2c_configure(base);
1710         writel_relaxed(l2x0_saved_regs.aux2_ctrl,
1711                        base + TAUROS3_AUX2_CTRL);
1712         writel_relaxed(l2x0_saved_regs.prefetch_ctrl,
1713                        base + L310_PREFETCH_CTRL);
1714 }
1715
1716 static const struct l2c_init_data of_tauros3_data __initconst = {
1717         .type = "Tauros3",
1718         .way_size_0 = SZ_8K,
1719         .num_lock = 8,
1720         .enable = l2c_enable,
1721         .save  = tauros3_save,
1722         .configure = tauros3_configure,
1723         .unlock = l2c_unlock,
1724         /* Tauros3 broadcasts L1 cache operations to L2 */
1725         .outer_cache = {
1726                 .resume      = l2c_resume,
1727         },
1728 };
1729
1730 #define L2C_ID(name, fns) { .compatible = name, .data = (void *)&fns }
1731 static const struct of_device_id l2x0_ids[] __initconst = {
1732         L2C_ID("arm,l210-cache", of_l2c210_data),
1733         L2C_ID("arm,l220-cache", of_l2c220_data),
1734         L2C_ID("arm,pl310-cache", of_l2c310_data),
1735         L2C_ID("brcm,bcm11351-a2-pl310-cache", of_bcm_l2x0_data),
1736         L2C_ID("marvell,aurora-outer-cache", of_aurora_with_outer_data),
1737         L2C_ID("marvell,aurora-system-cache", of_aurora_no_outer_data),
1738         L2C_ID("marvell,tauros3-cache", of_tauros3_data),
1739         /* Deprecated IDs */
1740         L2C_ID("bcm,bcm11351-a2-pl310-cache", of_bcm_l2x0_data),
1741         {}
1742 };
1743
1744 int __init l2x0_of_init(u32 aux_val, u32 aux_mask)
1745 {
1746         const struct l2c_init_data *data;
1747         struct device_node *np;
1748         struct resource res;
1749         u32 cache_id, old_aux;
1750         u32 cache_level = 2;
1751         bool nosync = false;
1752
1753         np = of_find_matching_node(NULL, l2x0_ids);
1754         if (!np)
1755                 return -ENODEV;
1756
1757         if (of_address_to_resource(np, 0, &res))
1758                 return -ENODEV;
1759
1760         l2x0_base = ioremap(res.start, resource_size(&res));
1761         if (!l2x0_base)
1762                 return -ENOMEM;
1763
1764         l2x0_saved_regs.phy_base = res.start;
1765
1766         data = of_match_node(l2x0_ids, np)->data;
1767
1768         if (of_device_is_compatible(np, "arm,pl310-cache") &&
1769             of_property_read_bool(np, "arm,io-coherent"))
1770                 data = &of_l2c310_coherent_data;
1771
1772         old_aux = readl_relaxed(l2x0_base + L2X0_AUX_CTRL);
1773         if (old_aux != ((old_aux & aux_mask) | aux_val)) {
1774                 pr_warn("L2C: platform modifies aux control register: 0x%08x -> 0x%08x\n",
1775                         old_aux, (old_aux & aux_mask) | aux_val);
1776         } else if (aux_mask != ~0U && aux_val != 0) {
1777                 pr_alert("L2C: platform provided aux values match the hardware, so have no effect.  Please remove them.\n");
1778         }
1779
1780         /* All L2 caches are unified, so this property should be specified */
1781         if (!of_property_read_bool(np, "cache-unified"))
1782                 pr_err("L2C: device tree omits to specify unified cache\n");
1783
1784         if (of_property_read_u32(np, "cache-level", &cache_level))
1785                 pr_err("L2C: device tree omits to specify cache-level\n");
1786
1787         if (cache_level != 2)
1788                 pr_err("L2C: device tree specifies invalid cache level\n");
1789
1790         nosync = of_property_read_bool(np, "arm,outer-sync-disable");
1791
1792         /* Read back current (default) hardware configuration */
1793         if (data->save)
1794                 data->save(l2x0_base);
1795
1796         /* L2 configuration can only be changed if the cache is disabled */
1797         if (!(readl_relaxed(l2x0_base + L2X0_CTRL) & L2X0_CTRL_EN))
1798                 if (data->of_parse)
1799                         data->of_parse(np, &aux_val, &aux_mask);
1800
1801         if (cache_id_part_number_from_dt)
1802                 cache_id = cache_id_part_number_from_dt;
1803         else
1804                 cache_id = readl_relaxed(l2x0_base + L2X0_CACHE_ID);
1805
1806         return __l2c_init(data, aux_val, aux_mask, cache_id, nosync);
1807 }
1808 #endif