irqchip: Prepare for local stub header removal
[deliverable/linux.git] / drivers / irqchip / irq-gic-v3-its.c
1 /*
2 * Copyright (C) 2013, 2014 ARM Limited, All Rights Reserved.
3 * Author: Marc Zyngier <marc.zyngier@arm.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License version 2 as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program. If not, see <http://www.gnu.org/licenses/>.
16 */
17
18 #include <linux/bitmap.h>
19 #include <linux/cpu.h>
20 #include <linux/delay.h>
21 #include <linux/interrupt.h>
22 #include <linux/log2.h>
23 #include <linux/mm.h>
24 #include <linux/msi.h>
25 #include <linux/of.h>
26 #include <linux/of_address.h>
27 #include <linux/of_irq.h>
28 #include <linux/of_pci.h>
29 #include <linux/of_platform.h>
30 #include <linux/percpu.h>
31 #include <linux/slab.h>
32
33 #include <linux/irqchip.h>
34 #include <linux/irqchip/arm-gic-v3.h>
35
36 #include <asm/cacheflush.h>
37 #include <asm/cputype.h>
38 #include <asm/exception.h>
39
40 #define ITS_FLAGS_CMDQ_NEEDS_FLUSHING (1 << 0)
41
42 #define RDIST_FLAGS_PROPBASE_NEEDS_FLUSHING (1 << 0)
43
44 /*
45 * Collection structure - just an ID, and a redistributor address to
46 * ping. We use one per CPU as a bag of interrupts assigned to this
47 * CPU.
48 */
49 struct its_collection {
50 u64 target_address;
51 u16 col_id;
52 };
53
54 /*
55 * The ITS structure - contains most of the infrastructure, with the
56 * msi_controller, the command queue, the collections, and the list of
57 * devices writing to it.
58 */
59 struct its_node {
60 raw_spinlock_t lock;
61 struct list_head entry;
62 struct msi_controller msi_chip;
63 struct irq_domain *domain;
64 void __iomem *base;
65 unsigned long phys_base;
66 struct its_cmd_block *cmd_base;
67 struct its_cmd_block *cmd_write;
68 void *tables[GITS_BASER_NR_REGS];
69 struct its_collection *collections;
70 struct list_head its_device_list;
71 u64 flags;
72 u32 ite_size;
73 };
74
75 #define ITS_ITT_ALIGN SZ_256
76
77 /*
78 * The ITS view of a device - belongs to an ITS, a collection, owns an
79 * interrupt translation table, and a list of interrupts.
80 */
81 struct its_device {
82 struct list_head entry;
83 struct its_node *its;
84 struct its_collection *collection;
85 void *itt;
86 unsigned long *lpi_map;
87 irq_hw_number_t lpi_base;
88 int nr_lpis;
89 u32 nr_ites;
90 u32 device_id;
91 };
92
93 static LIST_HEAD(its_nodes);
94 static DEFINE_SPINLOCK(its_lock);
95 static struct device_node *gic_root_node;
96 static struct rdists *gic_rdists;
97
98 #define gic_data_rdist() (raw_cpu_ptr(gic_rdists->rdist))
99 #define gic_data_rdist_rd_base() (gic_data_rdist()->rd_base)
100
101 /*
102 * ITS command descriptors - parameters to be encoded in a command
103 * block.
104 */
105 struct its_cmd_desc {
106 union {
107 struct {
108 struct its_device *dev;
109 u32 event_id;
110 } its_inv_cmd;
111
112 struct {
113 struct its_device *dev;
114 u32 event_id;
115 } its_int_cmd;
116
117 struct {
118 struct its_device *dev;
119 int valid;
120 } its_mapd_cmd;
121
122 struct {
123 struct its_collection *col;
124 int valid;
125 } its_mapc_cmd;
126
127 struct {
128 struct its_device *dev;
129 u32 phys_id;
130 u32 event_id;
131 } its_mapvi_cmd;
132
133 struct {
134 struct its_device *dev;
135 struct its_collection *col;
136 u32 id;
137 } its_movi_cmd;
138
139 struct {
140 struct its_device *dev;
141 u32 event_id;
142 } its_discard_cmd;
143
144 struct {
145 struct its_collection *col;
146 } its_invall_cmd;
147 };
148 };
149
150 /*
151 * The ITS command block, which is what the ITS actually parses.
152 */
153 struct its_cmd_block {
154 u64 raw_cmd[4];
155 };
156
157 #define ITS_CMD_QUEUE_SZ SZ_64K
158 #define ITS_CMD_QUEUE_NR_ENTRIES (ITS_CMD_QUEUE_SZ / sizeof(struct its_cmd_block))
159
160 typedef struct its_collection *(*its_cmd_builder_t)(struct its_cmd_block *,
161 struct its_cmd_desc *);
162
163 static void its_encode_cmd(struct its_cmd_block *cmd, u8 cmd_nr)
164 {
165 cmd->raw_cmd[0] &= ~0xffUL;
166 cmd->raw_cmd[0] |= cmd_nr;
167 }
168
169 static void its_encode_devid(struct its_cmd_block *cmd, u32 devid)
170 {
171 cmd->raw_cmd[0] &= BIT_ULL(32) - 1;
172 cmd->raw_cmd[0] |= ((u64)devid) << 32;
173 }
174
175 static void its_encode_event_id(struct its_cmd_block *cmd, u32 id)
176 {
177 cmd->raw_cmd[1] &= ~0xffffffffUL;
178 cmd->raw_cmd[1] |= id;
179 }
180
181 static void its_encode_phys_id(struct its_cmd_block *cmd, u32 phys_id)
182 {
183 cmd->raw_cmd[1] &= 0xffffffffUL;
184 cmd->raw_cmd[1] |= ((u64)phys_id) << 32;
185 }
186
187 static void its_encode_size(struct its_cmd_block *cmd, u8 size)
188 {
189 cmd->raw_cmd[1] &= ~0x1fUL;
190 cmd->raw_cmd[1] |= size & 0x1f;
191 }
192
193 static void its_encode_itt(struct its_cmd_block *cmd, u64 itt_addr)
194 {
195 cmd->raw_cmd[2] &= ~0xffffffffffffUL;
196 cmd->raw_cmd[2] |= itt_addr & 0xffffffffff00UL;
197 }
198
199 static void its_encode_valid(struct its_cmd_block *cmd, int valid)
200 {
201 cmd->raw_cmd[2] &= ~(1UL << 63);
202 cmd->raw_cmd[2] |= ((u64)!!valid) << 63;
203 }
204
205 static void its_encode_target(struct its_cmd_block *cmd, u64 target_addr)
206 {
207 cmd->raw_cmd[2] &= ~(0xffffffffUL << 16);
208 cmd->raw_cmd[2] |= (target_addr & (0xffffffffUL << 16));
209 }
210
211 static void its_encode_collection(struct its_cmd_block *cmd, u16 col)
212 {
213 cmd->raw_cmd[2] &= ~0xffffUL;
214 cmd->raw_cmd[2] |= col;
215 }
216
217 static inline void its_fixup_cmd(struct its_cmd_block *cmd)
218 {
219 /* Let's fixup BE commands */
220 cmd->raw_cmd[0] = cpu_to_le64(cmd->raw_cmd[0]);
221 cmd->raw_cmd[1] = cpu_to_le64(cmd->raw_cmd[1]);
222 cmd->raw_cmd[2] = cpu_to_le64(cmd->raw_cmd[2]);
223 cmd->raw_cmd[3] = cpu_to_le64(cmd->raw_cmd[3]);
224 }
225
226 static struct its_collection *its_build_mapd_cmd(struct its_cmd_block *cmd,
227 struct its_cmd_desc *desc)
228 {
229 unsigned long itt_addr;
230 u8 size = ilog2(desc->its_mapd_cmd.dev->nr_ites);
231
232 itt_addr = virt_to_phys(desc->its_mapd_cmd.dev->itt);
233 itt_addr = ALIGN(itt_addr, ITS_ITT_ALIGN);
234
235 its_encode_cmd(cmd, GITS_CMD_MAPD);
236 its_encode_devid(cmd, desc->its_mapd_cmd.dev->device_id);
237 its_encode_size(cmd, size - 1);
238 its_encode_itt(cmd, itt_addr);
239 its_encode_valid(cmd, desc->its_mapd_cmd.valid);
240
241 its_fixup_cmd(cmd);
242
243 return desc->its_mapd_cmd.dev->collection;
244 }
245
246 static struct its_collection *its_build_mapc_cmd(struct its_cmd_block *cmd,
247 struct its_cmd_desc *desc)
248 {
249 its_encode_cmd(cmd, GITS_CMD_MAPC);
250 its_encode_collection(cmd, desc->its_mapc_cmd.col->col_id);
251 its_encode_target(cmd, desc->its_mapc_cmd.col->target_address);
252 its_encode_valid(cmd, desc->its_mapc_cmd.valid);
253
254 its_fixup_cmd(cmd);
255
256 return desc->its_mapc_cmd.col;
257 }
258
259 static struct its_collection *its_build_mapvi_cmd(struct its_cmd_block *cmd,
260 struct its_cmd_desc *desc)
261 {
262 its_encode_cmd(cmd, GITS_CMD_MAPVI);
263 its_encode_devid(cmd, desc->its_mapvi_cmd.dev->device_id);
264 its_encode_event_id(cmd, desc->its_mapvi_cmd.event_id);
265 its_encode_phys_id(cmd, desc->its_mapvi_cmd.phys_id);
266 its_encode_collection(cmd, desc->its_mapvi_cmd.dev->collection->col_id);
267
268 its_fixup_cmd(cmd);
269
270 return desc->its_mapvi_cmd.dev->collection;
271 }
272
273 static struct its_collection *its_build_movi_cmd(struct its_cmd_block *cmd,
274 struct its_cmd_desc *desc)
275 {
276 its_encode_cmd(cmd, GITS_CMD_MOVI);
277 its_encode_devid(cmd, desc->its_movi_cmd.dev->device_id);
278 its_encode_event_id(cmd, desc->its_movi_cmd.id);
279 its_encode_collection(cmd, desc->its_movi_cmd.col->col_id);
280
281 its_fixup_cmd(cmd);
282
283 return desc->its_movi_cmd.dev->collection;
284 }
285
286 static struct its_collection *its_build_discard_cmd(struct its_cmd_block *cmd,
287 struct its_cmd_desc *desc)
288 {
289 its_encode_cmd(cmd, GITS_CMD_DISCARD);
290 its_encode_devid(cmd, desc->its_discard_cmd.dev->device_id);
291 its_encode_event_id(cmd, desc->its_discard_cmd.event_id);
292
293 its_fixup_cmd(cmd);
294
295 return desc->its_discard_cmd.dev->collection;
296 }
297
298 static struct its_collection *its_build_inv_cmd(struct its_cmd_block *cmd,
299 struct its_cmd_desc *desc)
300 {
301 its_encode_cmd(cmd, GITS_CMD_INV);
302 its_encode_devid(cmd, desc->its_inv_cmd.dev->device_id);
303 its_encode_event_id(cmd, desc->its_inv_cmd.event_id);
304
305 its_fixup_cmd(cmd);
306
307 return desc->its_inv_cmd.dev->collection;
308 }
309
310 static struct its_collection *its_build_invall_cmd(struct its_cmd_block *cmd,
311 struct its_cmd_desc *desc)
312 {
313 its_encode_cmd(cmd, GITS_CMD_INVALL);
314 its_encode_collection(cmd, desc->its_mapc_cmd.col->col_id);
315
316 its_fixup_cmd(cmd);
317
318 return NULL;
319 }
320
321 static u64 its_cmd_ptr_to_offset(struct its_node *its,
322 struct its_cmd_block *ptr)
323 {
324 return (ptr - its->cmd_base) * sizeof(*ptr);
325 }
326
327 static int its_queue_full(struct its_node *its)
328 {
329 int widx;
330 int ridx;
331
332 widx = its->cmd_write - its->cmd_base;
333 ridx = readl_relaxed(its->base + GITS_CREADR) / sizeof(struct its_cmd_block);
334
335 /* This is incredibly unlikely to happen, unless the ITS locks up. */
336 if (((widx + 1) % ITS_CMD_QUEUE_NR_ENTRIES) == ridx)
337 return 1;
338
339 return 0;
340 }
341
342 static struct its_cmd_block *its_allocate_entry(struct its_node *its)
343 {
344 struct its_cmd_block *cmd;
345 u32 count = 1000000; /* 1s! */
346
347 while (its_queue_full(its)) {
348 count--;
349 if (!count) {
350 pr_err_ratelimited("ITS queue not draining\n");
351 return NULL;
352 }
353 cpu_relax();
354 udelay(1);
355 }
356
357 cmd = its->cmd_write++;
358
359 /* Handle queue wrapping */
360 if (its->cmd_write == (its->cmd_base + ITS_CMD_QUEUE_NR_ENTRIES))
361 its->cmd_write = its->cmd_base;
362
363 return cmd;
364 }
365
366 static struct its_cmd_block *its_post_commands(struct its_node *its)
367 {
368 u64 wr = its_cmd_ptr_to_offset(its, its->cmd_write);
369
370 writel_relaxed(wr, its->base + GITS_CWRITER);
371
372 return its->cmd_write;
373 }
374
375 static void its_flush_cmd(struct its_node *its, struct its_cmd_block *cmd)
376 {
377 /*
378 * Make sure the commands written to memory are observable by
379 * the ITS.
380 */
381 if (its->flags & ITS_FLAGS_CMDQ_NEEDS_FLUSHING)
382 __flush_dcache_area(cmd, sizeof(*cmd));
383 else
384 dsb(ishst);
385 }
386
387 static void its_wait_for_range_completion(struct its_node *its,
388 struct its_cmd_block *from,
389 struct its_cmd_block *to)
390 {
391 u64 rd_idx, from_idx, to_idx;
392 u32 count = 1000000; /* 1s! */
393
394 from_idx = its_cmd_ptr_to_offset(its, from);
395 to_idx = its_cmd_ptr_to_offset(its, to);
396
397 while (1) {
398 rd_idx = readl_relaxed(its->base + GITS_CREADR);
399 if (rd_idx >= to_idx || rd_idx < from_idx)
400 break;
401
402 count--;
403 if (!count) {
404 pr_err_ratelimited("ITS queue timeout\n");
405 return;
406 }
407 cpu_relax();
408 udelay(1);
409 }
410 }
411
412 static void its_send_single_command(struct its_node *its,
413 its_cmd_builder_t builder,
414 struct its_cmd_desc *desc)
415 {
416 struct its_cmd_block *cmd, *sync_cmd, *next_cmd;
417 struct its_collection *sync_col;
418 unsigned long flags;
419
420 raw_spin_lock_irqsave(&its->lock, flags);
421
422 cmd = its_allocate_entry(its);
423 if (!cmd) { /* We're soooooo screewed... */
424 pr_err_ratelimited("ITS can't allocate, dropping command\n");
425 raw_spin_unlock_irqrestore(&its->lock, flags);
426 return;
427 }
428 sync_col = builder(cmd, desc);
429 its_flush_cmd(its, cmd);
430
431 if (sync_col) {
432 sync_cmd = its_allocate_entry(its);
433 if (!sync_cmd) {
434 pr_err_ratelimited("ITS can't SYNC, skipping\n");
435 goto post;
436 }
437 its_encode_cmd(sync_cmd, GITS_CMD_SYNC);
438 its_encode_target(sync_cmd, sync_col->target_address);
439 its_fixup_cmd(sync_cmd);
440 its_flush_cmd(its, sync_cmd);
441 }
442
443 post:
444 next_cmd = its_post_commands(its);
445 raw_spin_unlock_irqrestore(&its->lock, flags);
446
447 its_wait_for_range_completion(its, cmd, next_cmd);
448 }
449
450 static void its_send_inv(struct its_device *dev, u32 event_id)
451 {
452 struct its_cmd_desc desc;
453
454 desc.its_inv_cmd.dev = dev;
455 desc.its_inv_cmd.event_id = event_id;
456
457 its_send_single_command(dev->its, its_build_inv_cmd, &desc);
458 }
459
460 static void its_send_mapd(struct its_device *dev, int valid)
461 {
462 struct its_cmd_desc desc;
463
464 desc.its_mapd_cmd.dev = dev;
465 desc.its_mapd_cmd.valid = !!valid;
466
467 its_send_single_command(dev->its, its_build_mapd_cmd, &desc);
468 }
469
470 static void its_send_mapc(struct its_node *its, struct its_collection *col,
471 int valid)
472 {
473 struct its_cmd_desc desc;
474
475 desc.its_mapc_cmd.col = col;
476 desc.its_mapc_cmd.valid = !!valid;
477
478 its_send_single_command(its, its_build_mapc_cmd, &desc);
479 }
480
481 static void its_send_mapvi(struct its_device *dev, u32 irq_id, u32 id)
482 {
483 struct its_cmd_desc desc;
484
485 desc.its_mapvi_cmd.dev = dev;
486 desc.its_mapvi_cmd.phys_id = irq_id;
487 desc.its_mapvi_cmd.event_id = id;
488
489 its_send_single_command(dev->its, its_build_mapvi_cmd, &desc);
490 }
491
492 static void its_send_movi(struct its_device *dev,
493 struct its_collection *col, u32 id)
494 {
495 struct its_cmd_desc desc;
496
497 desc.its_movi_cmd.dev = dev;
498 desc.its_movi_cmd.col = col;
499 desc.its_movi_cmd.id = id;
500
501 its_send_single_command(dev->its, its_build_movi_cmd, &desc);
502 }
503
504 static void its_send_discard(struct its_device *dev, u32 id)
505 {
506 struct its_cmd_desc desc;
507
508 desc.its_discard_cmd.dev = dev;
509 desc.its_discard_cmd.event_id = id;
510
511 its_send_single_command(dev->its, its_build_discard_cmd, &desc);
512 }
513
514 static void its_send_invall(struct its_node *its, struct its_collection *col)
515 {
516 struct its_cmd_desc desc;
517
518 desc.its_invall_cmd.col = col;
519
520 its_send_single_command(its, its_build_invall_cmd, &desc);
521 }
522
523 /*
524 * irqchip functions - assumes MSI, mostly.
525 */
526
527 static inline u32 its_get_event_id(struct irq_data *d)
528 {
529 struct its_device *its_dev = irq_data_get_irq_chip_data(d);
530 return d->hwirq - its_dev->lpi_base;
531 }
532
533 static void lpi_set_config(struct irq_data *d, bool enable)
534 {
535 struct its_device *its_dev = irq_data_get_irq_chip_data(d);
536 irq_hw_number_t hwirq = d->hwirq;
537 u32 id = its_get_event_id(d);
538 u8 *cfg = page_address(gic_rdists->prop_page) + hwirq - 8192;
539
540 if (enable)
541 *cfg |= LPI_PROP_ENABLED;
542 else
543 *cfg &= ~LPI_PROP_ENABLED;
544
545 /*
546 * Make the above write visible to the redistributors.
547 * And yes, we're flushing exactly: One. Single. Byte.
548 * Humpf...
549 */
550 if (gic_rdists->flags & RDIST_FLAGS_PROPBASE_NEEDS_FLUSHING)
551 __flush_dcache_area(cfg, sizeof(*cfg));
552 else
553 dsb(ishst);
554 its_send_inv(its_dev, id);
555 }
556
557 static void its_mask_irq(struct irq_data *d)
558 {
559 lpi_set_config(d, false);
560 }
561
562 static void its_unmask_irq(struct irq_data *d)
563 {
564 lpi_set_config(d, true);
565 }
566
567 static void its_eoi_irq(struct irq_data *d)
568 {
569 gic_write_eoir(d->hwirq);
570 }
571
572 static int its_set_affinity(struct irq_data *d, const struct cpumask *mask_val,
573 bool force)
574 {
575 unsigned int cpu = cpumask_any_and(mask_val, cpu_online_mask);
576 struct its_device *its_dev = irq_data_get_irq_chip_data(d);
577 struct its_collection *target_col;
578 u32 id = its_get_event_id(d);
579
580 if (cpu >= nr_cpu_ids)
581 return -EINVAL;
582
583 target_col = &its_dev->its->collections[cpu];
584 its_send_movi(its_dev, target_col, id);
585 its_dev->collection = target_col;
586
587 return IRQ_SET_MASK_OK_DONE;
588 }
589
590 static void its_irq_compose_msi_msg(struct irq_data *d, struct msi_msg *msg)
591 {
592 struct its_device *its_dev = irq_data_get_irq_chip_data(d);
593 struct its_node *its;
594 u64 addr;
595
596 its = its_dev->its;
597 addr = its->phys_base + GITS_TRANSLATER;
598
599 msg->address_lo = addr & ((1UL << 32) - 1);
600 msg->address_hi = addr >> 32;
601 msg->data = its_get_event_id(d);
602 }
603
604 static struct irq_chip its_irq_chip = {
605 .name = "ITS",
606 .irq_mask = its_mask_irq,
607 .irq_unmask = its_unmask_irq,
608 .irq_eoi = its_eoi_irq,
609 .irq_set_affinity = its_set_affinity,
610 .irq_compose_msi_msg = its_irq_compose_msi_msg,
611 };
612
613 static void its_mask_msi_irq(struct irq_data *d)
614 {
615 pci_msi_mask_irq(d);
616 irq_chip_mask_parent(d);
617 }
618
619 static void its_unmask_msi_irq(struct irq_data *d)
620 {
621 pci_msi_unmask_irq(d);
622 irq_chip_unmask_parent(d);
623 }
624
625 static struct irq_chip its_msi_irq_chip = {
626 .name = "ITS-MSI",
627 .irq_unmask = its_unmask_msi_irq,
628 .irq_mask = its_mask_msi_irq,
629 .irq_eoi = irq_chip_eoi_parent,
630 .irq_write_msi_msg = pci_msi_domain_write_msg,
631 };
632
633 /*
634 * How we allocate LPIs:
635 *
636 * The GIC has id_bits bits for interrupt identifiers. From there, we
637 * must subtract 8192 which are reserved for SGIs/PPIs/SPIs. Then, as
638 * we allocate LPIs by chunks of 32, we can shift the whole thing by 5
639 * bits to the right.
640 *
641 * This gives us (((1UL << id_bits) - 8192) >> 5) possible allocations.
642 */
643 #define IRQS_PER_CHUNK_SHIFT 5
644 #define IRQS_PER_CHUNK (1 << IRQS_PER_CHUNK_SHIFT)
645
646 static unsigned long *lpi_bitmap;
647 static u32 lpi_chunks;
648 static DEFINE_SPINLOCK(lpi_lock);
649
650 static int its_lpi_to_chunk(int lpi)
651 {
652 return (lpi - 8192) >> IRQS_PER_CHUNK_SHIFT;
653 }
654
655 static int its_chunk_to_lpi(int chunk)
656 {
657 return (chunk << IRQS_PER_CHUNK_SHIFT) + 8192;
658 }
659
660 static int its_lpi_init(u32 id_bits)
661 {
662 lpi_chunks = its_lpi_to_chunk(1UL << id_bits);
663
664 lpi_bitmap = kzalloc(BITS_TO_LONGS(lpi_chunks) * sizeof(long),
665 GFP_KERNEL);
666 if (!lpi_bitmap) {
667 lpi_chunks = 0;
668 return -ENOMEM;
669 }
670
671 pr_info("ITS: Allocated %d chunks for LPIs\n", (int)lpi_chunks);
672 return 0;
673 }
674
675 static unsigned long *its_lpi_alloc_chunks(int nr_irqs, int *base, int *nr_ids)
676 {
677 unsigned long *bitmap = NULL;
678 int chunk_id;
679 int nr_chunks;
680 int i;
681
682 nr_chunks = DIV_ROUND_UP(nr_irqs, IRQS_PER_CHUNK);
683
684 spin_lock(&lpi_lock);
685
686 do {
687 chunk_id = bitmap_find_next_zero_area(lpi_bitmap, lpi_chunks,
688 0, nr_chunks, 0);
689 if (chunk_id < lpi_chunks)
690 break;
691
692 nr_chunks--;
693 } while (nr_chunks > 0);
694
695 if (!nr_chunks)
696 goto out;
697
698 bitmap = kzalloc(BITS_TO_LONGS(nr_chunks * IRQS_PER_CHUNK) * sizeof (long),
699 GFP_ATOMIC);
700 if (!bitmap)
701 goto out;
702
703 for (i = 0; i < nr_chunks; i++)
704 set_bit(chunk_id + i, lpi_bitmap);
705
706 *base = its_chunk_to_lpi(chunk_id);
707 *nr_ids = nr_chunks * IRQS_PER_CHUNK;
708
709 out:
710 spin_unlock(&lpi_lock);
711
712 return bitmap;
713 }
714
715 static void its_lpi_free(unsigned long *bitmap, int base, int nr_ids)
716 {
717 int lpi;
718
719 spin_lock(&lpi_lock);
720
721 for (lpi = base; lpi < (base + nr_ids); lpi += IRQS_PER_CHUNK) {
722 int chunk = its_lpi_to_chunk(lpi);
723 BUG_ON(chunk > lpi_chunks);
724 if (test_bit(chunk, lpi_bitmap)) {
725 clear_bit(chunk, lpi_bitmap);
726 } else {
727 pr_err("Bad LPI chunk %d\n", chunk);
728 }
729 }
730
731 spin_unlock(&lpi_lock);
732
733 kfree(bitmap);
734 }
735
736 /*
737 * We allocate 64kB for PROPBASE. That gives us at most 64K LPIs to
738 * deal with (one configuration byte per interrupt). PENDBASE has to
739 * be 64kB aligned (one bit per LPI, plus 8192 bits for SPI/PPI/SGI).
740 */
741 #define LPI_PROPBASE_SZ SZ_64K
742 #define LPI_PENDBASE_SZ (LPI_PROPBASE_SZ / 8 + SZ_1K)
743
744 /*
745 * This is how many bits of ID we need, including the useless ones.
746 */
747 #define LPI_NRBITS ilog2(LPI_PROPBASE_SZ + SZ_8K)
748
749 #define LPI_PROP_DEFAULT_PRIO 0xa0
750
751 static int __init its_alloc_lpi_tables(void)
752 {
753 phys_addr_t paddr;
754
755 gic_rdists->prop_page = alloc_pages(GFP_NOWAIT,
756 get_order(LPI_PROPBASE_SZ));
757 if (!gic_rdists->prop_page) {
758 pr_err("Failed to allocate PROPBASE\n");
759 return -ENOMEM;
760 }
761
762 paddr = page_to_phys(gic_rdists->prop_page);
763 pr_info("GIC: using LPI property table @%pa\n", &paddr);
764
765 /* Priority 0xa0, Group-1, disabled */
766 memset(page_address(gic_rdists->prop_page),
767 LPI_PROP_DEFAULT_PRIO | LPI_PROP_GROUP1,
768 LPI_PROPBASE_SZ);
769
770 /* Make sure the GIC will observe the written configuration */
771 __flush_dcache_area(page_address(gic_rdists->prop_page), LPI_PROPBASE_SZ);
772
773 return 0;
774 }
775
776 static const char *its_base_type_string[] = {
777 [GITS_BASER_TYPE_DEVICE] = "Devices",
778 [GITS_BASER_TYPE_VCPU] = "Virtual CPUs",
779 [GITS_BASER_TYPE_CPU] = "Physical CPUs",
780 [GITS_BASER_TYPE_COLLECTION] = "Interrupt Collections",
781 [GITS_BASER_TYPE_RESERVED5] = "Reserved (5)",
782 [GITS_BASER_TYPE_RESERVED6] = "Reserved (6)",
783 [GITS_BASER_TYPE_RESERVED7] = "Reserved (7)",
784 };
785
786 static void its_free_tables(struct its_node *its)
787 {
788 int i;
789
790 for (i = 0; i < GITS_BASER_NR_REGS; i++) {
791 if (its->tables[i]) {
792 free_page((unsigned long)its->tables[i]);
793 its->tables[i] = NULL;
794 }
795 }
796 }
797
798 static int its_alloc_tables(struct its_node *its)
799 {
800 int err;
801 int i;
802 int psz = SZ_64K;
803 u64 shr = GITS_BASER_InnerShareable;
804 u64 cache = GITS_BASER_WaWb;
805
806 for (i = 0; i < GITS_BASER_NR_REGS; i++) {
807 u64 val = readq_relaxed(its->base + GITS_BASER + i * 8);
808 u64 type = GITS_BASER_TYPE(val);
809 u64 entry_size = GITS_BASER_ENTRY_SIZE(val);
810 int order = get_order(psz);
811 int alloc_size;
812 u64 tmp;
813 void *base;
814
815 if (type == GITS_BASER_TYPE_NONE)
816 continue;
817
818 /*
819 * Allocate as many entries as required to fit the
820 * range of device IDs that the ITS can grok... The ID
821 * space being incredibly sparse, this results in a
822 * massive waste of memory.
823 *
824 * For other tables, only allocate a single page.
825 */
826 if (type == GITS_BASER_TYPE_DEVICE) {
827 u64 typer = readq_relaxed(its->base + GITS_TYPER);
828 u32 ids = GITS_TYPER_DEVBITS(typer);
829
830 /*
831 * 'order' was initialized earlier to the default page
832 * granule of the the ITS. We can't have an allocation
833 * smaller than that. If the requested allocation
834 * is smaller, round up to the default page granule.
835 */
836 order = max(get_order((1UL << ids) * entry_size),
837 order);
838 if (order >= MAX_ORDER) {
839 order = MAX_ORDER - 1;
840 pr_warn("%s: Device Table too large, reduce its page order to %u\n",
841 its->msi_chip.of_node->full_name, order);
842 }
843 }
844
845 alloc_size = (1 << order) * PAGE_SIZE;
846 base = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
847 if (!base) {
848 err = -ENOMEM;
849 goto out_free;
850 }
851
852 its->tables[i] = base;
853
854 retry_baser:
855 val = (virt_to_phys(base) |
856 (type << GITS_BASER_TYPE_SHIFT) |
857 ((entry_size - 1) << GITS_BASER_ENTRY_SIZE_SHIFT) |
858 cache |
859 shr |
860 GITS_BASER_VALID);
861
862 switch (psz) {
863 case SZ_4K:
864 val |= GITS_BASER_PAGE_SIZE_4K;
865 break;
866 case SZ_16K:
867 val |= GITS_BASER_PAGE_SIZE_16K;
868 break;
869 case SZ_64K:
870 val |= GITS_BASER_PAGE_SIZE_64K;
871 break;
872 }
873
874 val |= (alloc_size / psz) - 1;
875
876 writeq_relaxed(val, its->base + GITS_BASER + i * 8);
877 tmp = readq_relaxed(its->base + GITS_BASER + i * 8);
878
879 if ((val ^ tmp) & GITS_BASER_SHAREABILITY_MASK) {
880 /*
881 * Shareability didn't stick. Just use
882 * whatever the read reported, which is likely
883 * to be the only thing this redistributor
884 * supports. If that's zero, make it
885 * non-cacheable as well.
886 */
887 shr = tmp & GITS_BASER_SHAREABILITY_MASK;
888 if (!shr)
889 cache = GITS_BASER_nC;
890 goto retry_baser;
891 }
892
893 if ((val ^ tmp) & GITS_BASER_PAGE_SIZE_MASK) {
894 /*
895 * Page size didn't stick. Let's try a smaller
896 * size and retry. If we reach 4K, then
897 * something is horribly wrong...
898 */
899 switch (psz) {
900 case SZ_16K:
901 psz = SZ_4K;
902 goto retry_baser;
903 case SZ_64K:
904 psz = SZ_16K;
905 goto retry_baser;
906 }
907 }
908
909 if (val != tmp) {
910 pr_err("ITS: %s: GITS_BASER%d doesn't stick: %lx %lx\n",
911 its->msi_chip.of_node->full_name, i,
912 (unsigned long) val, (unsigned long) tmp);
913 err = -ENXIO;
914 goto out_free;
915 }
916
917 pr_info("ITS: allocated %d %s @%lx (psz %dK, shr %d)\n",
918 (int)(alloc_size / entry_size),
919 its_base_type_string[type],
920 (unsigned long)virt_to_phys(base),
921 psz / SZ_1K, (int)shr >> GITS_BASER_SHAREABILITY_SHIFT);
922 }
923
924 return 0;
925
926 out_free:
927 its_free_tables(its);
928
929 return err;
930 }
931
932 static int its_alloc_collections(struct its_node *its)
933 {
934 its->collections = kzalloc(nr_cpu_ids * sizeof(*its->collections),
935 GFP_KERNEL);
936 if (!its->collections)
937 return -ENOMEM;
938
939 return 0;
940 }
941
942 static void its_cpu_init_lpis(void)
943 {
944 void __iomem *rbase = gic_data_rdist_rd_base();
945 struct page *pend_page;
946 u64 val, tmp;
947
948 /* If we didn't allocate the pending table yet, do it now */
949 pend_page = gic_data_rdist()->pend_page;
950 if (!pend_page) {
951 phys_addr_t paddr;
952 /*
953 * The pending pages have to be at least 64kB aligned,
954 * hence the 'max(LPI_PENDBASE_SZ, SZ_64K)' below.
955 */
956 pend_page = alloc_pages(GFP_NOWAIT | __GFP_ZERO,
957 get_order(max(LPI_PENDBASE_SZ, SZ_64K)));
958 if (!pend_page) {
959 pr_err("Failed to allocate PENDBASE for CPU%d\n",
960 smp_processor_id());
961 return;
962 }
963
964 /* Make sure the GIC will observe the zero-ed page */
965 __flush_dcache_area(page_address(pend_page), LPI_PENDBASE_SZ);
966
967 paddr = page_to_phys(pend_page);
968 pr_info("CPU%d: using LPI pending table @%pa\n",
969 smp_processor_id(), &paddr);
970 gic_data_rdist()->pend_page = pend_page;
971 }
972
973 /* Disable LPIs */
974 val = readl_relaxed(rbase + GICR_CTLR);
975 val &= ~GICR_CTLR_ENABLE_LPIS;
976 writel_relaxed(val, rbase + GICR_CTLR);
977
978 /*
979 * Make sure any change to the table is observable by the GIC.
980 */
981 dsb(sy);
982
983 /* set PROPBASE */
984 val = (page_to_phys(gic_rdists->prop_page) |
985 GICR_PROPBASER_InnerShareable |
986 GICR_PROPBASER_WaWb |
987 ((LPI_NRBITS - 1) & GICR_PROPBASER_IDBITS_MASK));
988
989 writeq_relaxed(val, rbase + GICR_PROPBASER);
990 tmp = readq_relaxed(rbase + GICR_PROPBASER);
991
992 if ((tmp ^ val) & GICR_PROPBASER_SHAREABILITY_MASK) {
993 if (!(tmp & GICR_PROPBASER_SHAREABILITY_MASK)) {
994 /*
995 * The HW reports non-shareable, we must
996 * remove the cacheability attributes as
997 * well.
998 */
999 val &= ~(GICR_PROPBASER_SHAREABILITY_MASK |
1000 GICR_PROPBASER_CACHEABILITY_MASK);
1001 val |= GICR_PROPBASER_nC;
1002 writeq_relaxed(val, rbase + GICR_PROPBASER);
1003 }
1004 pr_info_once("GIC: using cache flushing for LPI property table\n");
1005 gic_rdists->flags |= RDIST_FLAGS_PROPBASE_NEEDS_FLUSHING;
1006 }
1007
1008 /* set PENDBASE */
1009 val = (page_to_phys(pend_page) |
1010 GICR_PENDBASER_InnerShareable |
1011 GICR_PENDBASER_WaWb);
1012
1013 writeq_relaxed(val, rbase + GICR_PENDBASER);
1014 tmp = readq_relaxed(rbase + GICR_PENDBASER);
1015
1016 if (!(tmp & GICR_PENDBASER_SHAREABILITY_MASK)) {
1017 /*
1018 * The HW reports non-shareable, we must remove the
1019 * cacheability attributes as well.
1020 */
1021 val &= ~(GICR_PENDBASER_SHAREABILITY_MASK |
1022 GICR_PENDBASER_CACHEABILITY_MASK);
1023 val |= GICR_PENDBASER_nC;
1024 writeq_relaxed(val, rbase + GICR_PENDBASER);
1025 }
1026
1027 /* Enable LPIs */
1028 val = readl_relaxed(rbase + GICR_CTLR);
1029 val |= GICR_CTLR_ENABLE_LPIS;
1030 writel_relaxed(val, rbase + GICR_CTLR);
1031
1032 /* Make sure the GIC has seen the above */
1033 dsb(sy);
1034 }
1035
1036 static void its_cpu_init_collection(void)
1037 {
1038 struct its_node *its;
1039 int cpu;
1040
1041 spin_lock(&its_lock);
1042 cpu = smp_processor_id();
1043
1044 list_for_each_entry(its, &its_nodes, entry) {
1045 u64 target;
1046
1047 /*
1048 * We now have to bind each collection to its target
1049 * redistributor.
1050 */
1051 if (readq_relaxed(its->base + GITS_TYPER) & GITS_TYPER_PTA) {
1052 /*
1053 * This ITS wants the physical address of the
1054 * redistributor.
1055 */
1056 target = gic_data_rdist()->phys_base;
1057 } else {
1058 /*
1059 * This ITS wants a linear CPU number.
1060 */
1061 target = readq_relaxed(gic_data_rdist_rd_base() + GICR_TYPER);
1062 target = GICR_TYPER_CPU_NUMBER(target) << 16;
1063 }
1064
1065 /* Perform collection mapping */
1066 its->collections[cpu].target_address = target;
1067 its->collections[cpu].col_id = cpu;
1068
1069 its_send_mapc(its, &its->collections[cpu], 1);
1070 its_send_invall(its, &its->collections[cpu]);
1071 }
1072
1073 spin_unlock(&its_lock);
1074 }
1075
1076 static struct its_device *its_find_device(struct its_node *its, u32 dev_id)
1077 {
1078 struct its_device *its_dev = NULL, *tmp;
1079 unsigned long flags;
1080
1081 raw_spin_lock_irqsave(&its->lock, flags);
1082
1083 list_for_each_entry(tmp, &its->its_device_list, entry) {
1084 if (tmp->device_id == dev_id) {
1085 its_dev = tmp;
1086 break;
1087 }
1088 }
1089
1090 raw_spin_unlock_irqrestore(&its->lock, flags);
1091
1092 return its_dev;
1093 }
1094
1095 static struct its_device *its_create_device(struct its_node *its, u32 dev_id,
1096 int nvecs)
1097 {
1098 struct its_device *dev;
1099 unsigned long *lpi_map;
1100 unsigned long flags;
1101 void *itt;
1102 int lpi_base;
1103 int nr_lpis;
1104 int nr_ites;
1105 int cpu;
1106 int sz;
1107
1108 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
1109 /*
1110 * At least one bit of EventID is being used, hence a minimum
1111 * of two entries. No, the architecture doesn't let you
1112 * express an ITT with a single entry.
1113 */
1114 nr_ites = max(2UL, roundup_pow_of_two(nvecs));
1115 sz = nr_ites * its->ite_size;
1116 sz = max(sz, ITS_ITT_ALIGN) + ITS_ITT_ALIGN - 1;
1117 itt = kzalloc(sz, GFP_KERNEL);
1118 lpi_map = its_lpi_alloc_chunks(nvecs, &lpi_base, &nr_lpis);
1119
1120 if (!dev || !itt || !lpi_map) {
1121 kfree(dev);
1122 kfree(itt);
1123 kfree(lpi_map);
1124 return NULL;
1125 }
1126
1127 dev->its = its;
1128 dev->itt = itt;
1129 dev->nr_ites = nr_ites;
1130 dev->lpi_map = lpi_map;
1131 dev->lpi_base = lpi_base;
1132 dev->nr_lpis = nr_lpis;
1133 dev->device_id = dev_id;
1134 INIT_LIST_HEAD(&dev->entry);
1135
1136 raw_spin_lock_irqsave(&its->lock, flags);
1137 list_add(&dev->entry, &its->its_device_list);
1138 raw_spin_unlock_irqrestore(&its->lock, flags);
1139
1140 /* Bind the device to the first possible CPU */
1141 cpu = cpumask_first(cpu_online_mask);
1142 dev->collection = &its->collections[cpu];
1143
1144 /* Map device to its ITT */
1145 its_send_mapd(dev, 1);
1146
1147 return dev;
1148 }
1149
1150 static void its_free_device(struct its_device *its_dev)
1151 {
1152 unsigned long flags;
1153
1154 raw_spin_lock_irqsave(&its_dev->its->lock, flags);
1155 list_del(&its_dev->entry);
1156 raw_spin_unlock_irqrestore(&its_dev->its->lock, flags);
1157 kfree(its_dev->itt);
1158 kfree(its_dev);
1159 }
1160
1161 static int its_alloc_device_irq(struct its_device *dev, irq_hw_number_t *hwirq)
1162 {
1163 int idx;
1164
1165 idx = find_first_zero_bit(dev->lpi_map, dev->nr_lpis);
1166 if (idx == dev->nr_lpis)
1167 return -ENOSPC;
1168
1169 *hwirq = dev->lpi_base + idx;
1170 set_bit(idx, dev->lpi_map);
1171
1172 return 0;
1173 }
1174
1175 struct its_pci_alias {
1176 struct pci_dev *pdev;
1177 u32 dev_id;
1178 u32 count;
1179 };
1180
1181 static int its_pci_msi_vec_count(struct pci_dev *pdev)
1182 {
1183 int msi, msix;
1184
1185 msi = max(pci_msi_vec_count(pdev), 0);
1186 msix = max(pci_msix_vec_count(pdev), 0);
1187
1188 return max(msi, msix);
1189 }
1190
1191 static int its_get_pci_alias(struct pci_dev *pdev, u16 alias, void *data)
1192 {
1193 struct its_pci_alias *dev_alias = data;
1194
1195 dev_alias->dev_id = alias;
1196 if (pdev != dev_alias->pdev)
1197 dev_alias->count += its_pci_msi_vec_count(dev_alias->pdev);
1198
1199 return 0;
1200 }
1201
1202 static int its_msi_prepare(struct irq_domain *domain, struct device *dev,
1203 int nvec, msi_alloc_info_t *info)
1204 {
1205 struct pci_dev *pdev;
1206 struct its_node *its;
1207 struct its_device *its_dev;
1208 struct its_pci_alias dev_alias;
1209
1210 if (!dev_is_pci(dev))
1211 return -EINVAL;
1212
1213 pdev = to_pci_dev(dev);
1214 dev_alias.pdev = pdev;
1215 dev_alias.count = nvec;
1216
1217 pci_for_each_dma_alias(pdev, its_get_pci_alias, &dev_alias);
1218 its = domain->parent->host_data;
1219
1220 its_dev = its_find_device(its, dev_alias.dev_id);
1221 if (its_dev) {
1222 /*
1223 * We already have seen this ID, probably through
1224 * another alias (PCI bridge of some sort). No need to
1225 * create the device.
1226 */
1227 dev_dbg(dev, "Reusing ITT for devID %x\n", dev_alias.dev_id);
1228 goto out;
1229 }
1230
1231 its_dev = its_create_device(its, dev_alias.dev_id, dev_alias.count);
1232 if (!its_dev)
1233 return -ENOMEM;
1234
1235 dev_dbg(&pdev->dev, "ITT %d entries, %d bits\n",
1236 dev_alias.count, ilog2(dev_alias.count));
1237 out:
1238 info->scratchpad[0].ptr = its_dev;
1239 info->scratchpad[1].ptr = dev;
1240 return 0;
1241 }
1242
1243 static struct msi_domain_ops its_pci_msi_ops = {
1244 .msi_prepare = its_msi_prepare,
1245 };
1246
1247 static struct msi_domain_info its_pci_msi_domain_info = {
1248 .flags = (MSI_FLAG_USE_DEF_DOM_OPS | MSI_FLAG_USE_DEF_CHIP_OPS |
1249 MSI_FLAG_MULTI_PCI_MSI | MSI_FLAG_PCI_MSIX),
1250 .ops = &its_pci_msi_ops,
1251 .chip = &its_msi_irq_chip,
1252 };
1253
1254 static int its_irq_gic_domain_alloc(struct irq_domain *domain,
1255 unsigned int virq,
1256 irq_hw_number_t hwirq)
1257 {
1258 struct of_phandle_args args;
1259
1260 args.np = domain->parent->of_node;
1261 args.args_count = 3;
1262 args.args[0] = GIC_IRQ_TYPE_LPI;
1263 args.args[1] = hwirq;
1264 args.args[2] = IRQ_TYPE_EDGE_RISING;
1265
1266 return irq_domain_alloc_irqs_parent(domain, virq, 1, &args);
1267 }
1268
1269 static int its_irq_domain_alloc(struct irq_domain *domain, unsigned int virq,
1270 unsigned int nr_irqs, void *args)
1271 {
1272 msi_alloc_info_t *info = args;
1273 struct its_device *its_dev = info->scratchpad[0].ptr;
1274 irq_hw_number_t hwirq;
1275 int err;
1276 int i;
1277
1278 for (i = 0; i < nr_irqs; i++) {
1279 err = its_alloc_device_irq(its_dev, &hwirq);
1280 if (err)
1281 return err;
1282
1283 err = its_irq_gic_domain_alloc(domain, virq + i, hwirq);
1284 if (err)
1285 return err;
1286
1287 irq_domain_set_hwirq_and_chip(domain, virq + i,
1288 hwirq, &its_irq_chip, its_dev);
1289 dev_dbg(info->scratchpad[1].ptr, "ID:%d pID:%d vID:%d\n",
1290 (int)(hwirq - its_dev->lpi_base), (int)hwirq, virq + i);
1291 }
1292
1293 return 0;
1294 }
1295
1296 static void its_irq_domain_activate(struct irq_domain *domain,
1297 struct irq_data *d)
1298 {
1299 struct its_device *its_dev = irq_data_get_irq_chip_data(d);
1300 u32 event = its_get_event_id(d);
1301
1302 /* Map the GIC IRQ and event to the device */
1303 its_send_mapvi(its_dev, d->hwirq, event);
1304 }
1305
1306 static void its_irq_domain_deactivate(struct irq_domain *domain,
1307 struct irq_data *d)
1308 {
1309 struct its_device *its_dev = irq_data_get_irq_chip_data(d);
1310 u32 event = its_get_event_id(d);
1311
1312 /* Stop the delivery of interrupts */
1313 its_send_discard(its_dev, event);
1314 }
1315
1316 static void its_irq_domain_free(struct irq_domain *domain, unsigned int virq,
1317 unsigned int nr_irqs)
1318 {
1319 struct irq_data *d = irq_domain_get_irq_data(domain, virq);
1320 struct its_device *its_dev = irq_data_get_irq_chip_data(d);
1321 int i;
1322
1323 for (i = 0; i < nr_irqs; i++) {
1324 struct irq_data *data = irq_domain_get_irq_data(domain,
1325 virq + i);
1326 u32 event = its_get_event_id(data);
1327
1328 /* Mark interrupt index as unused */
1329 clear_bit(event, its_dev->lpi_map);
1330
1331 /* Nuke the entry in the domain */
1332 irq_domain_reset_irq_data(data);
1333 }
1334
1335 /* If all interrupts have been freed, start mopping the floor */
1336 if (bitmap_empty(its_dev->lpi_map, its_dev->nr_lpis)) {
1337 its_lpi_free(its_dev->lpi_map,
1338 its_dev->lpi_base,
1339 its_dev->nr_lpis);
1340
1341 /* Unmap device/itt */
1342 its_send_mapd(its_dev, 0);
1343 its_free_device(its_dev);
1344 }
1345
1346 irq_domain_free_irqs_parent(domain, virq, nr_irqs);
1347 }
1348
1349 static const struct irq_domain_ops its_domain_ops = {
1350 .alloc = its_irq_domain_alloc,
1351 .free = its_irq_domain_free,
1352 .activate = its_irq_domain_activate,
1353 .deactivate = its_irq_domain_deactivate,
1354 };
1355
1356 static int its_force_quiescent(void __iomem *base)
1357 {
1358 u32 count = 1000000; /* 1s */
1359 u32 val;
1360
1361 val = readl_relaxed(base + GITS_CTLR);
1362 if (val & GITS_CTLR_QUIESCENT)
1363 return 0;
1364
1365 /* Disable the generation of all interrupts to this ITS */
1366 val &= ~GITS_CTLR_ENABLE;
1367 writel_relaxed(val, base + GITS_CTLR);
1368
1369 /* Poll GITS_CTLR and wait until ITS becomes quiescent */
1370 while (1) {
1371 val = readl_relaxed(base + GITS_CTLR);
1372 if (val & GITS_CTLR_QUIESCENT)
1373 return 0;
1374
1375 count--;
1376 if (!count)
1377 return -EBUSY;
1378
1379 cpu_relax();
1380 udelay(1);
1381 }
1382 }
1383
1384 static int its_probe(struct device_node *node, struct irq_domain *parent)
1385 {
1386 struct resource res;
1387 struct its_node *its;
1388 void __iomem *its_base;
1389 u32 val;
1390 u64 baser, tmp;
1391 int err;
1392
1393 err = of_address_to_resource(node, 0, &res);
1394 if (err) {
1395 pr_warn("%s: no regs?\n", node->full_name);
1396 return -ENXIO;
1397 }
1398
1399 its_base = ioremap(res.start, resource_size(&res));
1400 if (!its_base) {
1401 pr_warn("%s: unable to map registers\n", node->full_name);
1402 return -ENOMEM;
1403 }
1404
1405 val = readl_relaxed(its_base + GITS_PIDR2) & GIC_PIDR2_ARCH_MASK;
1406 if (val != 0x30 && val != 0x40) {
1407 pr_warn("%s: no ITS detected, giving up\n", node->full_name);
1408 err = -ENODEV;
1409 goto out_unmap;
1410 }
1411
1412 err = its_force_quiescent(its_base);
1413 if (err) {
1414 pr_warn("%s: failed to quiesce, giving up\n",
1415 node->full_name);
1416 goto out_unmap;
1417 }
1418
1419 pr_info("ITS: %s\n", node->full_name);
1420
1421 its = kzalloc(sizeof(*its), GFP_KERNEL);
1422 if (!its) {
1423 err = -ENOMEM;
1424 goto out_unmap;
1425 }
1426
1427 raw_spin_lock_init(&its->lock);
1428 INIT_LIST_HEAD(&its->entry);
1429 INIT_LIST_HEAD(&its->its_device_list);
1430 its->base = its_base;
1431 its->phys_base = res.start;
1432 its->msi_chip.of_node = node;
1433 its->ite_size = ((readl_relaxed(its_base + GITS_TYPER) >> 4) & 0xf) + 1;
1434
1435 its->cmd_base = kzalloc(ITS_CMD_QUEUE_SZ, GFP_KERNEL);
1436 if (!its->cmd_base) {
1437 err = -ENOMEM;
1438 goto out_free_its;
1439 }
1440 its->cmd_write = its->cmd_base;
1441
1442 err = its_alloc_tables(its);
1443 if (err)
1444 goto out_free_cmd;
1445
1446 err = its_alloc_collections(its);
1447 if (err)
1448 goto out_free_tables;
1449
1450 baser = (virt_to_phys(its->cmd_base) |
1451 GITS_CBASER_WaWb |
1452 GITS_CBASER_InnerShareable |
1453 (ITS_CMD_QUEUE_SZ / SZ_4K - 1) |
1454 GITS_CBASER_VALID);
1455
1456 writeq_relaxed(baser, its->base + GITS_CBASER);
1457 tmp = readq_relaxed(its->base + GITS_CBASER);
1458
1459 if ((tmp ^ baser) & GITS_CBASER_SHAREABILITY_MASK) {
1460 if (!(tmp & GITS_CBASER_SHAREABILITY_MASK)) {
1461 /*
1462 * The HW reports non-shareable, we must
1463 * remove the cacheability attributes as
1464 * well.
1465 */
1466 baser &= ~(GITS_CBASER_SHAREABILITY_MASK |
1467 GITS_CBASER_CACHEABILITY_MASK);
1468 baser |= GITS_CBASER_nC;
1469 writeq_relaxed(baser, its->base + GITS_CBASER);
1470 }
1471 pr_info("ITS: using cache flushing for cmd queue\n");
1472 its->flags |= ITS_FLAGS_CMDQ_NEEDS_FLUSHING;
1473 }
1474
1475 writeq_relaxed(0, its->base + GITS_CWRITER);
1476 writel_relaxed(GITS_CTLR_ENABLE, its->base + GITS_CTLR);
1477
1478 if (of_property_read_bool(its->msi_chip.of_node, "msi-controller")) {
1479 its->domain = irq_domain_add_tree(NULL, &its_domain_ops, its);
1480 if (!its->domain) {
1481 err = -ENOMEM;
1482 goto out_free_tables;
1483 }
1484
1485 its->domain->parent = parent;
1486
1487 its->msi_chip.domain = pci_msi_create_irq_domain(node,
1488 &its_pci_msi_domain_info,
1489 its->domain);
1490 if (!its->msi_chip.domain) {
1491 err = -ENOMEM;
1492 goto out_free_domains;
1493 }
1494
1495 err = of_pci_msi_chip_add(&its->msi_chip);
1496 if (err)
1497 goto out_free_domains;
1498 }
1499
1500 spin_lock(&its_lock);
1501 list_add(&its->entry, &its_nodes);
1502 spin_unlock(&its_lock);
1503
1504 return 0;
1505
1506 out_free_domains:
1507 if (its->msi_chip.domain)
1508 irq_domain_remove(its->msi_chip.domain);
1509 if (its->domain)
1510 irq_domain_remove(its->domain);
1511 out_free_tables:
1512 its_free_tables(its);
1513 out_free_cmd:
1514 kfree(its->cmd_base);
1515 out_free_its:
1516 kfree(its);
1517 out_unmap:
1518 iounmap(its_base);
1519 pr_err("ITS: failed probing %s (%d)\n", node->full_name, err);
1520 return err;
1521 }
1522
1523 static bool gic_rdists_supports_plpis(void)
1524 {
1525 return !!(readl_relaxed(gic_data_rdist_rd_base() + GICR_TYPER) & GICR_TYPER_PLPIS);
1526 }
1527
1528 int its_cpu_init(void)
1529 {
1530 if (!list_empty(&its_nodes)) {
1531 if (!gic_rdists_supports_plpis()) {
1532 pr_info("CPU%d: LPIs not supported\n", smp_processor_id());
1533 return -ENXIO;
1534 }
1535 its_cpu_init_lpis();
1536 its_cpu_init_collection();
1537 }
1538
1539 return 0;
1540 }
1541
1542 static struct of_device_id its_device_id[] = {
1543 { .compatible = "arm,gic-v3-its", },
1544 {},
1545 };
1546
1547 int its_init(struct device_node *node, struct rdists *rdists,
1548 struct irq_domain *parent_domain)
1549 {
1550 struct device_node *np;
1551
1552 for (np = of_find_matching_node(node, its_device_id); np;
1553 np = of_find_matching_node(np, its_device_id)) {
1554 its_probe(np, parent_domain);
1555 }
1556
1557 if (list_empty(&its_nodes)) {
1558 pr_warn("ITS: No ITS available, not enabling LPIs\n");
1559 return -ENXIO;
1560 }
1561
1562 gic_rdists = rdists;
1563 gic_root_node = node;
1564
1565 its_alloc_lpi_tables();
1566 its_lpi_init(rdists->id_bits);
1567
1568 return 0;
1569 }
This page took 0.065096 seconds and 5 git commands to generate.