xref: /openbmc/linux/arch/arm/mm/cache-uniphier.c (revision c4c3c32d)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright (C) 2015-2016 Socionext Inc.
4  *   Author: Masahiro Yamada <yamada.masahiro@socionext.com>
5  */
6 
7 #define pr_fmt(fmt)		"uniphier: " fmt
8 
9 #include <linux/bitops.h>
10 #include <linux/init.h>
11 #include <linux/io.h>
12 #include <linux/log2.h>
13 #include <linux/of_address.h>
14 #include <linux/slab.h>
15 #include <asm/hardware/cache-uniphier.h>
16 #include <asm/outercache.h>
17 
18 /* control registers */
19 #define UNIPHIER_SSCC		0x0	/* Control Register */
20 #define    UNIPHIER_SSCC_BST			BIT(20)	/* UCWG burst read */
21 #define    UNIPHIER_SSCC_ACT			BIT(19)	/* Inst-Data separate */
22 #define    UNIPHIER_SSCC_WTG			BIT(18)	/* WT gathering on */
23 #define    UNIPHIER_SSCC_PRD			BIT(17)	/* enable pre-fetch */
24 #define    UNIPHIER_SSCC_ON			BIT(0)	/* enable cache */
25 #define UNIPHIER_SSCLPDAWCR	0x30	/* Unified/Data Active Way Control */
26 #define UNIPHIER_SSCLPIAWCR	0x34	/* Instruction Active Way Control */
27 
28 /* revision registers */
29 #define UNIPHIER_SSCID		0x0	/* ID Register */
30 
31 /* operation registers */
32 #define UNIPHIER_SSCOPE		0x244	/* Cache Operation Primitive Entry */
33 #define    UNIPHIER_SSCOPE_CM_INV		0x0	/* invalidate */
34 #define    UNIPHIER_SSCOPE_CM_CLEAN		0x1	/* clean */
35 #define    UNIPHIER_SSCOPE_CM_FLUSH		0x2	/* flush */
36 #define    UNIPHIER_SSCOPE_CM_SYNC		0x8	/* sync (drain bufs) */
37 #define    UNIPHIER_SSCOPE_CM_FLUSH_PREFETCH	0x9	/* flush p-fetch buf */
38 #define UNIPHIER_SSCOQM		0x248	/* Cache Operation Queue Mode */
39 #define    UNIPHIER_SSCOQM_S_MASK		(0x3 << 17)
40 #define    UNIPHIER_SSCOQM_S_RANGE		(0x0 << 17)
41 #define    UNIPHIER_SSCOQM_S_ALL		(0x1 << 17)
42 #define    UNIPHIER_SSCOQM_CE			BIT(15)	/* notify completion */
43 #define    UNIPHIER_SSCOQM_CM_INV		0x0	/* invalidate */
44 #define    UNIPHIER_SSCOQM_CM_CLEAN		0x1	/* clean */
45 #define    UNIPHIER_SSCOQM_CM_FLUSH		0x2	/* flush */
46 #define UNIPHIER_SSCOQAD	0x24c	/* Cache Operation Queue Address */
47 #define UNIPHIER_SSCOQSZ	0x250	/* Cache Operation Queue Size */
48 #define UNIPHIER_SSCOPPQSEF	0x25c	/* Cache Operation Queue Set Complete*/
49 #define    UNIPHIER_SSCOPPQSEF_FE		BIT(1)
50 #define    UNIPHIER_SSCOPPQSEF_OE		BIT(0)
51 #define UNIPHIER_SSCOLPQS	0x260	/* Cache Operation Queue Status */
52 #define    UNIPHIER_SSCOLPQS_EF			BIT(2)
53 #define    UNIPHIER_SSCOLPQS_EST		BIT(1)
54 #define    UNIPHIER_SSCOLPQS_QST		BIT(0)
55 
56 /* Is the operation region specified by address range? */
57 #define UNIPHIER_SSCOQM_S_IS_RANGE(op) \
58 		((op & UNIPHIER_SSCOQM_S_MASK) == UNIPHIER_SSCOQM_S_RANGE)
59 
60 /**
61  * uniphier_cache_data - UniPhier outer cache specific data
62  *
63  * @ctrl_base: virtual base address of control registers
64  * @rev_base: virtual base address of revision registers
65  * @op_base: virtual base address of operation registers
66  * @way_mask: each bit specifies if the way is present
67  * @nsets: number of associativity sets
68  * @line_size: line size in bytes
69  * @range_op_max_size: max size that can be handled by a single range operation
70  * @list: list node to include this level in the whole cache hierarchy
71  */
72 struct uniphier_cache_data {
73 	void __iomem *ctrl_base;
74 	void __iomem *rev_base;
75 	void __iomem *op_base;
76 	void __iomem *way_ctrl_base;
77 	u32 way_mask;
78 	u32 nsets;
79 	u32 line_size;
80 	u32 range_op_max_size;
81 	struct list_head list;
82 };
83 
84 /*
85  * List of the whole outer cache hierarchy.  This list is only modified during
86  * the early boot stage, so no mutex is taken for the access to the list.
87  */
88 static LIST_HEAD(uniphier_cache_list);
89 
90 /**
91  * __uniphier_cache_sync - perform a sync point for a particular cache level
92  *
93  * @data: cache controller specific data
94  */
95 static void __uniphier_cache_sync(struct uniphier_cache_data *data)
96 {
97 	/* This sequence need not be atomic.  Do not disable IRQ. */
98 	writel_relaxed(UNIPHIER_SSCOPE_CM_SYNC,
99 		       data->op_base + UNIPHIER_SSCOPE);
100 	/* need a read back to confirm */
101 	readl_relaxed(data->op_base + UNIPHIER_SSCOPE);
102 }
103 
104 /**
105  * __uniphier_cache_maint_common - run a queue operation for a particular level
106  *
107  * @data: cache controller specific data
108  * @start: start address of range operation (don't care for "all" operation)
109  * @size: data size of range operation (don't care for "all" operation)
110  * @operation: flags to specify the desired cache operation
111  */
112 static void __uniphier_cache_maint_common(struct uniphier_cache_data *data,
113 					  unsigned long start,
114 					  unsigned long size,
115 					  u32 operation)
116 {
117 	unsigned long flags;
118 
119 	/*
120 	 * No spin lock is necessary here because:
121 	 *
122 	 * [1] This outer cache controller is able to accept maintenance
123 	 * operations from multiple CPUs at a time in an SMP system; if a
124 	 * maintenance operation is under way and another operation is issued,
125 	 * the new one is stored in the queue.  The controller performs one
126 	 * operation after another.  If the queue is full, the status register,
127 	 * UNIPHIER_SSCOPPQSEF, indicates that the queue registration has
128 	 * failed.  The status registers, UNIPHIER_{SSCOPPQSEF, SSCOLPQS}, have
129 	 * different instances for each CPU, i.e. each CPU can track the status
130 	 * of the maintenance operations triggered by itself.
131 	 *
132 	 * [2] The cache command registers, UNIPHIER_{SSCOQM, SSCOQAD, SSCOQSZ,
133 	 * SSCOQWN}, are shared between multiple CPUs, but the hardware still
134 	 * guarantees the registration sequence is atomic; the write access to
135 	 * them are arbitrated by the hardware.  The first accessor to the
136 	 * register, UNIPHIER_SSCOQM, holds the access right and it is released
137 	 * by reading the status register, UNIPHIER_SSCOPPQSEF.  While one CPU
138 	 * is holding the access right, other CPUs fail to register operations.
139 	 * One CPU should not hold the access right for a long time, so local
140 	 * IRQs should be disabled while the following sequence.
141 	 */
142 	local_irq_save(flags);
143 
144 	/* clear the complete notification flag */
145 	writel_relaxed(UNIPHIER_SSCOLPQS_EF, data->op_base + UNIPHIER_SSCOLPQS);
146 
147 	do {
148 		/* set cache operation */
149 		writel_relaxed(UNIPHIER_SSCOQM_CE | operation,
150 			       data->op_base + UNIPHIER_SSCOQM);
151 
152 		/* set address range if needed */
153 		if (likely(UNIPHIER_SSCOQM_S_IS_RANGE(operation))) {
154 			writel_relaxed(start, data->op_base + UNIPHIER_SSCOQAD);
155 			writel_relaxed(size, data->op_base + UNIPHIER_SSCOQSZ);
156 		}
157 	} while (unlikely(readl_relaxed(data->op_base + UNIPHIER_SSCOPPQSEF) &
158 			  (UNIPHIER_SSCOPPQSEF_FE | UNIPHIER_SSCOPPQSEF_OE)));
159 
160 	/* wait until the operation is completed */
161 	while (likely(readl_relaxed(data->op_base + UNIPHIER_SSCOLPQS) !=
162 		      UNIPHIER_SSCOLPQS_EF))
163 		cpu_relax();
164 
165 	local_irq_restore(flags);
166 }
167 
168 static void __uniphier_cache_maint_all(struct uniphier_cache_data *data,
169 				       u32 operation)
170 {
171 	__uniphier_cache_maint_common(data, 0, 0,
172 				      UNIPHIER_SSCOQM_S_ALL | operation);
173 
174 	__uniphier_cache_sync(data);
175 }
176 
177 static void __uniphier_cache_maint_range(struct uniphier_cache_data *data,
178 					 unsigned long start, unsigned long end,
179 					 u32 operation)
180 {
181 	unsigned long size;
182 
183 	/*
184 	 * If the start address is not aligned,
185 	 * perform a cache operation for the first cache-line
186 	 */
187 	start = start & ~(data->line_size - 1);
188 
189 	size = end - start;
190 
191 	if (unlikely(size >= (unsigned long)(-data->line_size))) {
192 		/* this means cache operation for all range */
193 		__uniphier_cache_maint_all(data, operation);
194 		return;
195 	}
196 
197 	/*
198 	 * If the end address is not aligned,
199 	 * perform a cache operation for the last cache-line
200 	 */
201 	size = ALIGN(size, data->line_size);
202 
203 	while (size) {
204 		unsigned long chunk_size = min_t(unsigned long, size,
205 						 data->range_op_max_size);
206 
207 		__uniphier_cache_maint_common(data, start, chunk_size,
208 					UNIPHIER_SSCOQM_S_RANGE | operation);
209 
210 		start += chunk_size;
211 		size -= chunk_size;
212 	}
213 
214 	__uniphier_cache_sync(data);
215 }
216 
217 static void __uniphier_cache_enable(struct uniphier_cache_data *data, bool on)
218 {
219 	u32 val = 0;
220 
221 	if (on)
222 		val = UNIPHIER_SSCC_WTG | UNIPHIER_SSCC_PRD | UNIPHIER_SSCC_ON;
223 
224 	writel_relaxed(val, data->ctrl_base + UNIPHIER_SSCC);
225 }
226 
227 static void __init __uniphier_cache_set_active_ways(
228 					struct uniphier_cache_data *data)
229 {
230 	unsigned int cpu;
231 
232 	for_each_possible_cpu(cpu)
233 		writel_relaxed(data->way_mask, data->way_ctrl_base + 4 * cpu);
234 }
235 
236 static void uniphier_cache_maint_range(unsigned long start, unsigned long end,
237 				       u32 operation)
238 {
239 	struct uniphier_cache_data *data;
240 
241 	list_for_each_entry(data, &uniphier_cache_list, list)
242 		__uniphier_cache_maint_range(data, start, end, operation);
243 }
244 
245 static void uniphier_cache_maint_all(u32 operation)
246 {
247 	struct uniphier_cache_data *data;
248 
249 	list_for_each_entry(data, &uniphier_cache_list, list)
250 		__uniphier_cache_maint_all(data, operation);
251 }
252 
253 static void uniphier_cache_inv_range(unsigned long start, unsigned long end)
254 {
255 	uniphier_cache_maint_range(start, end, UNIPHIER_SSCOQM_CM_INV);
256 }
257 
258 static void uniphier_cache_clean_range(unsigned long start, unsigned long end)
259 {
260 	uniphier_cache_maint_range(start, end, UNIPHIER_SSCOQM_CM_CLEAN);
261 }
262 
263 static void uniphier_cache_flush_range(unsigned long start, unsigned long end)
264 {
265 	uniphier_cache_maint_range(start, end, UNIPHIER_SSCOQM_CM_FLUSH);
266 }
267 
268 static void __init uniphier_cache_inv_all(void)
269 {
270 	uniphier_cache_maint_all(UNIPHIER_SSCOQM_CM_INV);
271 }
272 
273 static void uniphier_cache_flush_all(void)
274 {
275 	uniphier_cache_maint_all(UNIPHIER_SSCOQM_CM_FLUSH);
276 }
277 
278 static void uniphier_cache_disable(void)
279 {
280 	struct uniphier_cache_data *data;
281 
282 	list_for_each_entry_reverse(data, &uniphier_cache_list, list)
283 		__uniphier_cache_enable(data, false);
284 
285 	uniphier_cache_flush_all();
286 }
287 
288 static void __init uniphier_cache_enable(void)
289 {
290 	struct uniphier_cache_data *data;
291 
292 	uniphier_cache_inv_all();
293 
294 	list_for_each_entry(data, &uniphier_cache_list, list) {
295 		__uniphier_cache_enable(data, true);
296 		__uniphier_cache_set_active_ways(data);
297 	}
298 }
299 
300 static void uniphier_cache_sync(void)
301 {
302 	struct uniphier_cache_data *data;
303 
304 	list_for_each_entry(data, &uniphier_cache_list, list)
305 		__uniphier_cache_sync(data);
306 }
307 
308 static const struct of_device_id uniphier_cache_match[] __initconst = {
309 	{ .compatible = "socionext,uniphier-system-cache" },
310 	{ /* sentinel */ }
311 };
312 
313 static int __init __uniphier_cache_init(struct device_node *np,
314 					unsigned int *cache_level)
315 {
316 	struct uniphier_cache_data *data;
317 	u32 level, cache_size;
318 	struct device_node *next_np;
319 	int ret = 0;
320 
321 	if (!of_match_node(uniphier_cache_match, np)) {
322 		pr_err("L%d: not compatible with uniphier cache\n",
323 		       *cache_level);
324 		return -EINVAL;
325 	}
326 
327 	if (of_property_read_u32(np, "cache-level", &level)) {
328 		pr_err("L%d: cache-level is not specified\n", *cache_level);
329 		return -EINVAL;
330 	}
331 
332 	if (level != *cache_level) {
333 		pr_err("L%d: cache-level is unexpected value %d\n",
334 		       *cache_level, level);
335 		return -EINVAL;
336 	}
337 
338 	if (!of_property_read_bool(np, "cache-unified")) {
339 		pr_err("L%d: cache-unified is not specified\n", *cache_level);
340 		return -EINVAL;
341 	}
342 
343 	data = kzalloc(sizeof(*data), GFP_KERNEL);
344 	if (!data)
345 		return -ENOMEM;
346 
347 	if (of_property_read_u32(np, "cache-line-size", &data->line_size) ||
348 	    !is_power_of_2(data->line_size)) {
349 		pr_err("L%d: cache-line-size is unspecified or invalid\n",
350 		       *cache_level);
351 		ret = -EINVAL;
352 		goto err;
353 	}
354 
355 	if (of_property_read_u32(np, "cache-sets", &data->nsets) ||
356 	    !is_power_of_2(data->nsets)) {
357 		pr_err("L%d: cache-sets is unspecified or invalid\n",
358 		       *cache_level);
359 		ret = -EINVAL;
360 		goto err;
361 	}
362 
363 	if (of_property_read_u32(np, "cache-size", &cache_size) ||
364 	    cache_size == 0 || cache_size % (data->nsets * data->line_size)) {
365 		pr_err("L%d: cache-size is unspecified or invalid\n",
366 		       *cache_level);
367 		ret = -EINVAL;
368 		goto err;
369 	}
370 
371 	data->way_mask = GENMASK(cache_size / data->nsets / data->line_size - 1,
372 				 0);
373 
374 	data->ctrl_base = of_iomap(np, 0);
375 	if (!data->ctrl_base) {
376 		pr_err("L%d: failed to map control register\n", *cache_level);
377 		ret = -ENOMEM;
378 		goto err;
379 	}
380 
381 	data->rev_base = of_iomap(np, 1);
382 	if (!data->rev_base) {
383 		pr_err("L%d: failed to map revision register\n", *cache_level);
384 		ret = -ENOMEM;
385 		goto err;
386 	}
387 
388 	data->op_base = of_iomap(np, 2);
389 	if (!data->op_base) {
390 		pr_err("L%d: failed to map operation register\n", *cache_level);
391 		ret = -ENOMEM;
392 		goto err;
393 	}
394 
395 	data->way_ctrl_base = data->ctrl_base + 0xc00;
396 
397 	if (*cache_level == 2) {
398 		u32 revision = readl(data->rev_base + UNIPHIER_SSCID);
399 		/*
400 		 * The size of range operation is limited to (1 << 22) or less
401 		 * for PH-sLD8 or older SoCs.
402 		 */
403 		if (revision <= 0x16)
404 			data->range_op_max_size = (u32)1 << 22;
405 
406 		/*
407 		 * Unfortunatly, the offset address of active way control base
408 		 * varies from SoC to SoC.
409 		 */
410 		switch (revision) {
411 		case 0x11:	/* sLD3 */
412 			data->way_ctrl_base = data->ctrl_base + 0x870;
413 			break;
414 		case 0x12:	/* LD4 */
415 		case 0x16:	/* sld8 */
416 			data->way_ctrl_base = data->ctrl_base + 0x840;
417 			break;
418 		default:
419 			break;
420 		}
421 	}
422 
423 	data->range_op_max_size -= data->line_size;
424 
425 	INIT_LIST_HEAD(&data->list);
426 	list_add_tail(&data->list, &uniphier_cache_list); /* no mutex */
427 
428 	/*
429 	 * OK, this level has been successfully initialized.  Look for the next
430 	 * level cache.  Do not roll back even if the initialization of the
431 	 * next level cache fails because we want to continue with available
432 	 * cache levels.
433 	 */
434 	next_np = of_find_next_cache_node(np);
435 	if (next_np) {
436 		(*cache_level)++;
437 		ret = __uniphier_cache_init(next_np, cache_level);
438 	}
439 	of_node_put(next_np);
440 
441 	return ret;
442 err:
443 	iounmap(data->op_base);
444 	iounmap(data->rev_base);
445 	iounmap(data->ctrl_base);
446 	kfree(data);
447 
448 	return ret;
449 }
450 
451 int __init uniphier_cache_init(void)
452 {
453 	struct device_node *np = NULL;
454 	unsigned int cache_level;
455 	int ret = 0;
456 
457 	/* look for level 2 cache */
458 	while ((np = of_find_matching_node(np, uniphier_cache_match)))
459 		if (!of_property_read_u32(np, "cache-level", &cache_level) &&
460 		    cache_level == 2)
461 			break;
462 
463 	if (!np)
464 		return -ENODEV;
465 
466 	ret = __uniphier_cache_init(np, &cache_level);
467 	of_node_put(np);
468 
469 	if (ret) {
470 		/*
471 		 * Error out iif L2 initialization fails.  Continue with any
472 		 * error on L3 or outer because they are optional.
473 		 */
474 		if (cache_level == 2) {
475 			pr_err("failed to initialize L2 cache\n");
476 			return ret;
477 		}
478 
479 		cache_level--;
480 		ret = 0;
481 	}
482 
483 	outer_cache.inv_range = uniphier_cache_inv_range;
484 	outer_cache.clean_range = uniphier_cache_clean_range;
485 	outer_cache.flush_range = uniphier_cache_flush_range;
486 	outer_cache.flush_all = uniphier_cache_flush_all;
487 	outer_cache.disable = uniphier_cache_disable;
488 	outer_cache.sync = uniphier_cache_sync;
489 
490 	uniphier_cache_enable();
491 
492 	pr_info("enabled outer cache (cache level: %d)\n", cache_level);
493 
494 	return ret;
495 }
496