2 * Copyright (C) 2013, 2014 ARM Limited, All Rights Reserved.
3 * Author: Marc Zyngier <marc.zyngier@arm.com>
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.
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.
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/>.
18 #include <linux/acpi.h>
19 #include <linux/bitmap.h>
20 #include <linux/cpu.h>
21 #include <linux/delay.h>
22 #include <linux/interrupt.h>
23 #include <linux/irqdomain.h>
24 #include <linux/acpi_iort.h>
25 #include <linux/log2.h>
27 #include <linux/msi.h>
29 #include <linux/of_address.h>
30 #include <linux/of_irq.h>
31 #include <linux/of_pci.h>
32 #include <linux/of_platform.h>
33 #include <linux/percpu.h>
34 #include <linux/slab.h>
36 #include <linux/irqchip.h>
37 #include <linux/irqchip/arm-gic-v3.h>
39 #include <asm/cacheflush.h>
40 #include <asm/cputype.h>
41 #include <asm/exception.h>
43 #include "irq-gic-common.h"
45 #define ITS_FLAGS_CMDQ_NEEDS_FLUSHING (1ULL << 0)
46 #define ITS_FLAGS_WORKAROUND_CAVIUM_22375 (1ULL << 1)
47 #define ITS_FLAGS_WORKAROUND_CAVIUM_23144 (1ULL << 2)
49 #define RDIST_FLAGS_PROPBASE_NEEDS_FLUSHING (1 << 0)
52 * Collection structure - just an ID, and a redistributor address to
53 * ping. We use one per CPU as a bag of interrupts assigned to this
56 struct its_collection {
62 * The ITS_BASER structure - contains memory information, cached
63 * value of BASER register configuration and ITS page size.
73 * The ITS structure - contains most of the infrastructure, with the
74 * top-level MSI domain, the command queue, the collections, and the
75 * list of devices writing to it.
79 struct list_head entry;
81 phys_addr_t phys_base;
82 struct its_cmd_block *cmd_base;
83 struct its_cmd_block *cmd_write;
84 struct its_baser tables[GITS_BASER_NR_REGS];
85 struct its_collection *collections;
86 struct list_head its_device_list;
93 #define ITS_ITT_ALIGN SZ_256
95 /* Convert page order to size in bytes */
96 #define PAGE_ORDER_TO_SIZE(o) (PAGE_SIZE << (o))
98 struct event_lpi_map {
99 unsigned long *lpi_map;
101 irq_hw_number_t lpi_base;
106 * The ITS view of a device - belongs to an ITS, a collection, owns an
107 * interrupt translation table, and a list of interrupts.
110 struct list_head entry;
111 struct its_node *its;
112 struct event_lpi_map event_map;
118 static LIST_HEAD(its_nodes);
119 static DEFINE_SPINLOCK(its_lock);
120 static struct rdists *gic_rdists;
121 static struct irq_domain *its_parent;
123 #define gic_data_rdist() (raw_cpu_ptr(gic_rdists->rdist))
124 #define gic_data_rdist_rd_base() (gic_data_rdist()->rd_base)
126 static struct its_collection *dev_event_to_col(struct its_device *its_dev,
129 struct its_node *its = its_dev->its;
131 return its->collections + its_dev->event_map.col_map[event];
135 * ITS command descriptors - parameters to be encoded in a command
138 struct its_cmd_desc {
141 struct its_device *dev;
146 struct its_device *dev;
151 struct its_device *dev;
156 struct its_collection *col;
161 struct its_device *dev;
167 struct its_device *dev;
168 struct its_collection *col;
173 struct its_device *dev;
178 struct its_collection *col;
184 * The ITS command block, which is what the ITS actually parses.
186 struct its_cmd_block {
190 #define ITS_CMD_QUEUE_SZ SZ_64K
191 #define ITS_CMD_QUEUE_NR_ENTRIES (ITS_CMD_QUEUE_SZ / sizeof(struct its_cmd_block))
193 typedef struct its_collection *(*its_cmd_builder_t)(struct its_cmd_block *,
194 struct its_cmd_desc *);
196 static void its_encode_cmd(struct its_cmd_block *cmd, u8 cmd_nr)
198 cmd->raw_cmd[0] &= ~0xffUL;
199 cmd->raw_cmd[0] |= cmd_nr;
202 static void its_encode_devid(struct its_cmd_block *cmd, u32 devid)
204 cmd->raw_cmd[0] &= BIT_ULL(32) - 1;
205 cmd->raw_cmd[0] |= ((u64)devid) << 32;
208 static void its_encode_event_id(struct its_cmd_block *cmd, u32 id)
210 cmd->raw_cmd[1] &= ~0xffffffffUL;
211 cmd->raw_cmd[1] |= id;
214 static void its_encode_phys_id(struct its_cmd_block *cmd, u32 phys_id)
216 cmd->raw_cmd[1] &= 0xffffffffUL;
217 cmd->raw_cmd[1] |= ((u64)phys_id) << 32;
220 static void its_encode_size(struct its_cmd_block *cmd, u8 size)
222 cmd->raw_cmd[1] &= ~0x1fUL;
223 cmd->raw_cmd[1] |= size & 0x1f;
226 static void its_encode_itt(struct its_cmd_block *cmd, u64 itt_addr)
228 cmd->raw_cmd[2] &= ~0xffffffffffffUL;
229 cmd->raw_cmd[2] |= itt_addr & 0xffffffffff00UL;
232 static void its_encode_valid(struct its_cmd_block *cmd, int valid)
234 cmd->raw_cmd[2] &= ~(1UL << 63);
235 cmd->raw_cmd[2] |= ((u64)!!valid) << 63;
238 static void its_encode_target(struct its_cmd_block *cmd, u64 target_addr)
240 cmd->raw_cmd[2] &= ~(0xffffffffUL << 16);
241 cmd->raw_cmd[2] |= (target_addr & (0xffffffffUL << 16));
244 static void its_encode_collection(struct its_cmd_block *cmd, u16 col)
246 cmd->raw_cmd[2] &= ~0xffffUL;
247 cmd->raw_cmd[2] |= col;
250 static inline void its_fixup_cmd(struct its_cmd_block *cmd)
252 /* Let's fixup BE commands */
253 cmd->raw_cmd[0] = cpu_to_le64(cmd->raw_cmd[0]);
254 cmd->raw_cmd[1] = cpu_to_le64(cmd->raw_cmd[1]);
255 cmd->raw_cmd[2] = cpu_to_le64(cmd->raw_cmd[2]);
256 cmd->raw_cmd[3] = cpu_to_le64(cmd->raw_cmd[3]);
259 static struct its_collection *its_build_mapd_cmd(struct its_cmd_block *cmd,
260 struct its_cmd_desc *desc)
262 unsigned long itt_addr;
263 u8 size = ilog2(desc->its_mapd_cmd.dev->nr_ites);
265 itt_addr = virt_to_phys(desc->its_mapd_cmd.dev->itt);
266 itt_addr = ALIGN(itt_addr, ITS_ITT_ALIGN);
268 its_encode_cmd(cmd, GITS_CMD_MAPD);
269 its_encode_devid(cmd, desc->its_mapd_cmd.dev->device_id);
270 its_encode_size(cmd, size - 1);
271 its_encode_itt(cmd, itt_addr);
272 its_encode_valid(cmd, desc->its_mapd_cmd.valid);
279 static struct its_collection *its_build_mapc_cmd(struct its_cmd_block *cmd,
280 struct its_cmd_desc *desc)
282 its_encode_cmd(cmd, GITS_CMD_MAPC);
283 its_encode_collection(cmd, desc->its_mapc_cmd.col->col_id);
284 its_encode_target(cmd, desc->its_mapc_cmd.col->target_address);
285 its_encode_valid(cmd, desc->its_mapc_cmd.valid);
289 return desc->its_mapc_cmd.col;
292 static struct its_collection *its_build_mapvi_cmd(struct its_cmd_block *cmd,
293 struct its_cmd_desc *desc)
295 struct its_collection *col;
297 col = dev_event_to_col(desc->its_mapvi_cmd.dev,
298 desc->its_mapvi_cmd.event_id);
300 its_encode_cmd(cmd, GITS_CMD_MAPVI);
301 its_encode_devid(cmd, desc->its_mapvi_cmd.dev->device_id);
302 its_encode_event_id(cmd, desc->its_mapvi_cmd.event_id);
303 its_encode_phys_id(cmd, desc->its_mapvi_cmd.phys_id);
304 its_encode_collection(cmd, col->col_id);
311 static struct its_collection *its_build_movi_cmd(struct its_cmd_block *cmd,
312 struct its_cmd_desc *desc)
314 struct its_collection *col;
316 col = dev_event_to_col(desc->its_movi_cmd.dev,
317 desc->its_movi_cmd.event_id);
319 its_encode_cmd(cmd, GITS_CMD_MOVI);
320 its_encode_devid(cmd, desc->its_movi_cmd.dev->device_id);
321 its_encode_event_id(cmd, desc->its_movi_cmd.event_id);
322 its_encode_collection(cmd, desc->its_movi_cmd.col->col_id);
329 static struct its_collection *its_build_discard_cmd(struct its_cmd_block *cmd,
330 struct its_cmd_desc *desc)
332 struct its_collection *col;
334 col = dev_event_to_col(desc->its_discard_cmd.dev,
335 desc->its_discard_cmd.event_id);
337 its_encode_cmd(cmd, GITS_CMD_DISCARD);
338 its_encode_devid(cmd, desc->its_discard_cmd.dev->device_id);
339 its_encode_event_id(cmd, desc->its_discard_cmd.event_id);
346 static struct its_collection *its_build_inv_cmd(struct its_cmd_block *cmd,
347 struct its_cmd_desc *desc)
349 struct its_collection *col;
351 col = dev_event_to_col(desc->its_inv_cmd.dev,
352 desc->its_inv_cmd.event_id);
354 its_encode_cmd(cmd, GITS_CMD_INV);
355 its_encode_devid(cmd, desc->its_inv_cmd.dev->device_id);
356 its_encode_event_id(cmd, desc->its_inv_cmd.event_id);
363 static struct its_collection *its_build_invall_cmd(struct its_cmd_block *cmd,
364 struct its_cmd_desc *desc)
366 its_encode_cmd(cmd, GITS_CMD_INVALL);
367 its_encode_collection(cmd, desc->its_mapc_cmd.col->col_id);
374 static u64 its_cmd_ptr_to_offset(struct its_node *its,
375 struct its_cmd_block *ptr)
377 return (ptr - its->cmd_base) * sizeof(*ptr);
380 static int its_queue_full(struct its_node *its)
385 widx = its->cmd_write - its->cmd_base;
386 ridx = readl_relaxed(its->base + GITS_CREADR) / sizeof(struct its_cmd_block);
388 /* This is incredibly unlikely to happen, unless the ITS locks up. */
389 if (((widx + 1) % ITS_CMD_QUEUE_NR_ENTRIES) == ridx)
395 static struct its_cmd_block *its_allocate_entry(struct its_node *its)
397 struct its_cmd_block *cmd;
398 u32 count = 1000000; /* 1s! */
400 while (its_queue_full(its)) {
403 pr_err_ratelimited("ITS queue not draining\n");
410 cmd = its->cmd_write++;
412 /* Handle queue wrapping */
413 if (its->cmd_write == (its->cmd_base + ITS_CMD_QUEUE_NR_ENTRIES))
414 its->cmd_write = its->cmd_base;
419 static struct its_cmd_block *its_post_commands(struct its_node *its)
421 u64 wr = its_cmd_ptr_to_offset(its, its->cmd_write);
423 writel_relaxed(wr, its->base + GITS_CWRITER);
425 return its->cmd_write;
428 static void its_flush_cmd(struct its_node *its, struct its_cmd_block *cmd)
431 * Make sure the commands written to memory are observable by
434 if (its->flags & ITS_FLAGS_CMDQ_NEEDS_FLUSHING)
435 __flush_dcache_area(cmd, sizeof(*cmd));
440 static void its_wait_for_range_completion(struct its_node *its,
441 struct its_cmd_block *from,
442 struct its_cmd_block *to)
444 u64 rd_idx, from_idx, to_idx;
445 u32 count = 1000000; /* 1s! */
447 from_idx = its_cmd_ptr_to_offset(its, from);
448 to_idx = its_cmd_ptr_to_offset(its, to);
451 rd_idx = readl_relaxed(its->base + GITS_CREADR);
452 if (rd_idx >= to_idx || rd_idx < from_idx)
457 pr_err_ratelimited("ITS queue timeout\n");
465 static void its_send_single_command(struct its_node *its,
466 its_cmd_builder_t builder,
467 struct its_cmd_desc *desc)
469 struct its_cmd_block *cmd, *sync_cmd, *next_cmd;
470 struct its_collection *sync_col;
473 raw_spin_lock_irqsave(&its->lock, flags);
475 cmd = its_allocate_entry(its);
476 if (!cmd) { /* We're soooooo screewed... */
477 pr_err_ratelimited("ITS can't allocate, dropping command\n");
478 raw_spin_unlock_irqrestore(&its->lock, flags);
481 sync_col = builder(cmd, desc);
482 its_flush_cmd(its, cmd);
485 sync_cmd = its_allocate_entry(its);
487 pr_err_ratelimited("ITS can't SYNC, skipping\n");
490 its_encode_cmd(sync_cmd, GITS_CMD_SYNC);
491 its_encode_target(sync_cmd, sync_col->target_address);
492 its_fixup_cmd(sync_cmd);
493 its_flush_cmd(its, sync_cmd);
497 next_cmd = its_post_commands(its);
498 raw_spin_unlock_irqrestore(&its->lock, flags);
500 its_wait_for_range_completion(its, cmd, next_cmd);
503 static void its_send_inv(struct its_device *dev, u32 event_id)
505 struct its_cmd_desc desc;
507 desc.its_inv_cmd.dev = dev;
508 desc.its_inv_cmd.event_id = event_id;
510 its_send_single_command(dev->its, its_build_inv_cmd, &desc);
513 static void its_send_mapd(struct its_device *dev, int valid)
515 struct its_cmd_desc desc;
517 desc.its_mapd_cmd.dev = dev;
518 desc.its_mapd_cmd.valid = !!valid;
520 its_send_single_command(dev->its, its_build_mapd_cmd, &desc);
523 static void its_send_mapc(struct its_node *its, struct its_collection *col,
526 struct its_cmd_desc desc;
528 desc.its_mapc_cmd.col = col;
529 desc.its_mapc_cmd.valid = !!valid;
531 its_send_single_command(its, its_build_mapc_cmd, &desc);
534 static void its_send_mapvi(struct its_device *dev, u32 irq_id, u32 id)
536 struct its_cmd_desc desc;
538 desc.its_mapvi_cmd.dev = dev;
539 desc.its_mapvi_cmd.phys_id = irq_id;
540 desc.its_mapvi_cmd.event_id = id;
542 its_send_single_command(dev->its, its_build_mapvi_cmd, &desc);
545 static void its_send_movi(struct its_device *dev,
546 struct its_collection *col, u32 id)
548 struct its_cmd_desc desc;
550 desc.its_movi_cmd.dev = dev;
551 desc.its_movi_cmd.col = col;
552 desc.its_movi_cmd.event_id = id;
554 its_send_single_command(dev->its, its_build_movi_cmd, &desc);
557 static void its_send_discard(struct its_device *dev, u32 id)
559 struct its_cmd_desc desc;
561 desc.its_discard_cmd.dev = dev;
562 desc.its_discard_cmd.event_id = id;
564 its_send_single_command(dev->its, its_build_discard_cmd, &desc);
567 static void its_send_invall(struct its_node *its, struct its_collection *col)
569 struct its_cmd_desc desc;
571 desc.its_invall_cmd.col = col;
573 its_send_single_command(its, its_build_invall_cmd, &desc);
577 * irqchip functions - assumes MSI, mostly.
580 static inline u32 its_get_event_id(struct irq_data *d)
582 struct its_device *its_dev = irq_data_get_irq_chip_data(d);
583 return d->hwirq - its_dev->event_map.lpi_base;
586 static void lpi_set_config(struct irq_data *d, bool enable)
588 struct its_device *its_dev = irq_data_get_irq_chip_data(d);
589 irq_hw_number_t hwirq = d->hwirq;
590 u32 id = its_get_event_id(d);
591 u8 *cfg = page_address(gic_rdists->prop_page) + hwirq - 8192;
594 *cfg |= LPI_PROP_ENABLED;
596 *cfg &= ~LPI_PROP_ENABLED;
599 * Make the above write visible to the redistributors.
600 * And yes, we're flushing exactly: One. Single. Byte.
603 if (gic_rdists->flags & RDIST_FLAGS_PROPBASE_NEEDS_FLUSHING)
604 __flush_dcache_area(cfg, sizeof(*cfg));
607 its_send_inv(its_dev, id);
610 static void its_mask_irq(struct irq_data *d)
612 lpi_set_config(d, false);
615 static void its_unmask_irq(struct irq_data *d)
617 lpi_set_config(d, true);
620 static int its_set_affinity(struct irq_data *d, const struct cpumask *mask_val,
624 const struct cpumask *cpu_mask = cpu_online_mask;
625 struct its_device *its_dev = irq_data_get_irq_chip_data(d);
626 struct its_collection *target_col;
627 u32 id = its_get_event_id(d);
629 /* lpi cannot be routed to a redistributor that is on a foreign node */
630 if (its_dev->its->flags & ITS_FLAGS_WORKAROUND_CAVIUM_23144) {
631 if (its_dev->its->numa_node >= 0) {
632 cpu_mask = cpumask_of_node(its_dev->its->numa_node);
633 if (!cpumask_intersects(mask_val, cpu_mask))
638 cpu = cpumask_any_and(mask_val, cpu_mask);
640 if (cpu >= nr_cpu_ids)
643 target_col = &its_dev->its->collections[cpu];
644 its_send_movi(its_dev, target_col, id);
645 its_dev->event_map.col_map[id] = cpu;
647 return IRQ_SET_MASK_OK_DONE;
650 static void its_irq_compose_msi_msg(struct irq_data *d, struct msi_msg *msg)
652 struct its_device *its_dev = irq_data_get_irq_chip_data(d);
653 struct its_node *its;
657 addr = its->phys_base + GITS_TRANSLATER;
659 msg->address_lo = addr & ((1UL << 32) - 1);
660 msg->address_hi = addr >> 32;
661 msg->data = its_get_event_id(d);
664 static struct irq_chip its_irq_chip = {
666 .irq_mask = its_mask_irq,
667 .irq_unmask = its_unmask_irq,
668 .irq_eoi = irq_chip_eoi_parent,
669 .irq_set_affinity = its_set_affinity,
670 .irq_compose_msi_msg = its_irq_compose_msi_msg,
674 * How we allocate LPIs:
676 * The GIC has id_bits bits for interrupt identifiers. From there, we
677 * must subtract 8192 which are reserved for SGIs/PPIs/SPIs. Then, as
678 * we allocate LPIs by chunks of 32, we can shift the whole thing by 5
681 * This gives us (((1UL << id_bits) - 8192) >> 5) possible allocations.
683 #define IRQS_PER_CHUNK_SHIFT 5
684 #define IRQS_PER_CHUNK (1 << IRQS_PER_CHUNK_SHIFT)
686 static unsigned long *lpi_bitmap;
687 static u32 lpi_chunks;
688 static DEFINE_SPINLOCK(lpi_lock);
690 static int its_lpi_to_chunk(int lpi)
692 return (lpi - 8192) >> IRQS_PER_CHUNK_SHIFT;
695 static int its_chunk_to_lpi(int chunk)
697 return (chunk << IRQS_PER_CHUNK_SHIFT) + 8192;
700 static int __init its_lpi_init(u32 id_bits)
702 lpi_chunks = its_lpi_to_chunk(1UL << id_bits);
704 lpi_bitmap = kzalloc(BITS_TO_LONGS(lpi_chunks) * sizeof(long),
711 pr_info("ITS: Allocated %d chunks for LPIs\n", (int)lpi_chunks);
715 static unsigned long *its_lpi_alloc_chunks(int nr_irqs, int *base, int *nr_ids)
717 unsigned long *bitmap = NULL;
722 nr_chunks = DIV_ROUND_UP(nr_irqs, IRQS_PER_CHUNK);
724 spin_lock(&lpi_lock);
727 chunk_id = bitmap_find_next_zero_area(lpi_bitmap, lpi_chunks,
729 if (chunk_id < lpi_chunks)
733 } while (nr_chunks > 0);
738 bitmap = kzalloc(BITS_TO_LONGS(nr_chunks * IRQS_PER_CHUNK) * sizeof (long),
743 for (i = 0; i < nr_chunks; i++)
744 set_bit(chunk_id + i, lpi_bitmap);
746 *base = its_chunk_to_lpi(chunk_id);
747 *nr_ids = nr_chunks * IRQS_PER_CHUNK;
750 spin_unlock(&lpi_lock);
758 static void its_lpi_free(struct event_lpi_map *map)
760 int base = map->lpi_base;
761 int nr_ids = map->nr_lpis;
764 spin_lock(&lpi_lock);
766 for (lpi = base; lpi < (base + nr_ids); lpi += IRQS_PER_CHUNK) {
767 int chunk = its_lpi_to_chunk(lpi);
768 BUG_ON(chunk > lpi_chunks);
769 if (test_bit(chunk, lpi_bitmap)) {
770 clear_bit(chunk, lpi_bitmap);
772 pr_err("Bad LPI chunk %d\n", chunk);
776 spin_unlock(&lpi_lock);
783 * We allocate 64kB for PROPBASE. That gives us at most 64K LPIs to
784 * deal with (one configuration byte per interrupt). PENDBASE has to
785 * be 64kB aligned (one bit per LPI, plus 8192 bits for SPI/PPI/SGI).
787 #define LPI_PROPBASE_SZ SZ_64K
788 #define LPI_PENDBASE_SZ (LPI_PROPBASE_SZ / 8 + SZ_1K)
791 * This is how many bits of ID we need, including the useless ones.
793 #define LPI_NRBITS ilog2(LPI_PROPBASE_SZ + SZ_8K)
795 #define LPI_PROP_DEFAULT_PRIO 0xa0
797 static int __init its_alloc_lpi_tables(void)
801 gic_rdists->prop_page = alloc_pages(GFP_NOWAIT,
802 get_order(LPI_PROPBASE_SZ));
803 if (!gic_rdists->prop_page) {
804 pr_err("Failed to allocate PROPBASE\n");
808 paddr = page_to_phys(gic_rdists->prop_page);
809 pr_info("GIC: using LPI property table @%pa\n", &paddr);
811 /* Priority 0xa0, Group-1, disabled */
812 memset(page_address(gic_rdists->prop_page),
813 LPI_PROP_DEFAULT_PRIO | LPI_PROP_GROUP1,
816 /* Make sure the GIC will observe the written configuration */
817 __flush_dcache_area(page_address(gic_rdists->prop_page), LPI_PROPBASE_SZ);
822 static const char *its_base_type_string[] = {
823 [GITS_BASER_TYPE_DEVICE] = "Devices",
824 [GITS_BASER_TYPE_VCPU] = "Virtual CPUs",
825 [GITS_BASER_TYPE_CPU] = "Physical CPUs",
826 [GITS_BASER_TYPE_COLLECTION] = "Interrupt Collections",
827 [GITS_BASER_TYPE_RESERVED5] = "Reserved (5)",
828 [GITS_BASER_TYPE_RESERVED6] = "Reserved (6)",
829 [GITS_BASER_TYPE_RESERVED7] = "Reserved (7)",
832 static u64 its_read_baser(struct its_node *its, struct its_baser *baser)
834 u32 idx = baser - its->tables;
836 return readq_relaxed(its->base + GITS_BASER + (idx << 3));
839 static void its_write_baser(struct its_node *its, struct its_baser *baser,
842 u32 idx = baser - its->tables;
844 writeq_relaxed(val, its->base + GITS_BASER + (idx << 3));
845 baser->val = its_read_baser(its, baser);
848 static int its_setup_baser(struct its_node *its, struct its_baser *baser,
849 u64 cache, u64 shr, u32 psz, u32 order,
852 u64 val = its_read_baser(its, baser);
853 u64 esz = GITS_BASER_ENTRY_SIZE(val);
854 u64 type = GITS_BASER_TYPE(val);
860 alloc_pages = (PAGE_ORDER_TO_SIZE(order) / psz);
861 if (alloc_pages > GITS_BASER_PAGES_MAX) {
862 pr_warn("ITS@%pa: %s too large, reduce ITS pages %u->%u\n",
863 &its->phys_base, its_base_type_string[type],
864 alloc_pages, GITS_BASER_PAGES_MAX);
865 alloc_pages = GITS_BASER_PAGES_MAX;
866 order = get_order(GITS_BASER_PAGES_MAX * psz);
869 base = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
874 val = (virt_to_phys(base) |
875 (type << GITS_BASER_TYPE_SHIFT) |
876 ((esz - 1) << GITS_BASER_ENTRY_SIZE_SHIFT) |
877 ((alloc_pages - 1) << GITS_BASER_PAGES_SHIFT) |
882 val |= indirect ? GITS_BASER_INDIRECT : 0x0;
886 val |= GITS_BASER_PAGE_SIZE_4K;
889 val |= GITS_BASER_PAGE_SIZE_16K;
892 val |= GITS_BASER_PAGE_SIZE_64K;
896 its_write_baser(its, baser, val);
899 if ((val ^ tmp) & GITS_BASER_SHAREABILITY_MASK) {
901 * Shareability didn't stick. Just use
902 * whatever the read reported, which is likely
903 * to be the only thing this redistributor
904 * supports. If that's zero, make it
905 * non-cacheable as well.
907 shr = tmp & GITS_BASER_SHAREABILITY_MASK;
909 cache = GITS_BASER_nC;
910 __flush_dcache_area(base, PAGE_ORDER_TO_SIZE(order));
915 if ((val ^ tmp) & GITS_BASER_PAGE_SIZE_MASK) {
917 * Page size didn't stick. Let's try a smaller
918 * size and retry. If we reach 4K, then
919 * something is horribly wrong...
921 free_pages((unsigned long)base, order);
927 goto retry_alloc_baser;
930 goto retry_alloc_baser;
935 pr_err("ITS@%pa: %s doesn't stick: %lx %lx\n",
936 &its->phys_base, its_base_type_string[type],
937 (unsigned long) val, (unsigned long) tmp);
938 free_pages((unsigned long)base, order);
942 baser->order = order;
945 tmp = indirect ? GITS_LVL1_ENTRY_SIZE : esz;
947 pr_info("ITS@%pa: allocated %d %s @%lx (%s, esz %d, psz %dK, shr %d)\n",
948 &its->phys_base, (int)(PAGE_ORDER_TO_SIZE(order) / tmp),
949 its_base_type_string[type],
950 (unsigned long)virt_to_phys(base),
951 indirect ? "indirect" : "flat", (int)esz,
952 psz / SZ_1K, (int)shr >> GITS_BASER_SHAREABILITY_SHIFT);
957 static bool its_parse_baser_device(struct its_node *its, struct its_baser *baser,
960 u64 esz = GITS_BASER_ENTRY_SIZE(its_read_baser(its, baser));
961 u64 val = GITS_BASER_InnerShareable | GITS_BASER_WaWb;
962 u32 ids = its->device_ids;
963 u32 new_order = *order;
964 bool indirect = false;
966 /* No need to enable Indirection if memory requirement < (psz*2)bytes */
967 if ((esz << ids) > (psz * 2)) {
969 * Find out whether hw supports a single or two-level table by
970 * table by reading bit at offset '62' after writing '1' to it.
972 its_write_baser(its, baser, val | GITS_BASER_INDIRECT);
973 indirect = !!(baser->val & GITS_BASER_INDIRECT);
977 * The size of the lvl2 table is equal to ITS page size
978 * which is 'psz'. For computing lvl1 table size,
979 * subtract ID bits that sparse lvl2 table from 'ids'
980 * which is reported by ITS hardware times lvl1 table
983 ids -= ilog2(psz / esz);
984 esz = GITS_LVL1_ENTRY_SIZE;
989 * Allocate as many entries as required to fit the
990 * range of device IDs that the ITS can grok... The ID
991 * space being incredibly sparse, this results in a
992 * massive waste of memory if two-level device table
993 * feature is not supported by hardware.
995 new_order = max_t(u32, get_order(esz << ids), new_order);
996 if (new_order >= MAX_ORDER) {
997 new_order = MAX_ORDER - 1;
998 ids = ilog2(PAGE_ORDER_TO_SIZE(new_order) / esz);
999 pr_warn("ITS@%pa: Device Table too large, reduce ids %u->%u\n",
1000 &its->phys_base, its->device_ids, ids);
1008 static void its_free_tables(struct its_node *its)
1012 for (i = 0; i < GITS_BASER_NR_REGS; i++) {
1013 if (its->tables[i].base) {
1014 free_pages((unsigned long)its->tables[i].base,
1015 its->tables[i].order);
1016 its->tables[i].base = NULL;
1021 static int its_alloc_tables(struct its_node *its)
1023 u64 typer = readq_relaxed(its->base + GITS_TYPER);
1024 u32 ids = GITS_TYPER_DEVBITS(typer);
1025 u64 shr = GITS_BASER_InnerShareable;
1026 u64 cache = GITS_BASER_WaWb;
1030 if (its->flags & ITS_FLAGS_WORKAROUND_CAVIUM_22375) {
1032 * erratum 22375: only alloc 8MB table size
1033 * erratum 24313: ignore memory access type
1035 cache = GITS_BASER_nCnB;
1036 ids = 0x14; /* 20 bits, 8MB */
1039 its->device_ids = ids;
1041 for (i = 0; i < GITS_BASER_NR_REGS; i++) {
1042 struct its_baser *baser = its->tables + i;
1043 u64 val = its_read_baser(its, baser);
1044 u64 type = GITS_BASER_TYPE(val);
1045 u32 order = get_order(psz);
1046 bool indirect = false;
1048 if (type == GITS_BASER_TYPE_NONE)
1051 if (type == GITS_BASER_TYPE_DEVICE)
1052 indirect = its_parse_baser_device(its, baser, psz, &order);
1054 err = its_setup_baser(its, baser, cache, shr, psz, order, indirect);
1056 its_free_tables(its);
1060 /* Update settings which will be used for next BASERn */
1062 cache = baser->val & GITS_BASER_CACHEABILITY_MASK;
1063 shr = baser->val & GITS_BASER_SHAREABILITY_MASK;
1069 static int its_alloc_collections(struct its_node *its)
1071 its->collections = kzalloc(nr_cpu_ids * sizeof(*its->collections),
1073 if (!its->collections)
1079 static void its_cpu_init_lpis(void)
1081 void __iomem *rbase = gic_data_rdist_rd_base();
1082 struct page *pend_page;
1085 /* If we didn't allocate the pending table yet, do it now */
1086 pend_page = gic_data_rdist()->pend_page;
1090 * The pending pages have to be at least 64kB aligned,
1091 * hence the 'max(LPI_PENDBASE_SZ, SZ_64K)' below.
1093 pend_page = alloc_pages(GFP_NOWAIT | __GFP_ZERO,
1094 get_order(max(LPI_PENDBASE_SZ, SZ_64K)));
1096 pr_err("Failed to allocate PENDBASE for CPU%d\n",
1097 smp_processor_id());
1101 /* Make sure the GIC will observe the zero-ed page */
1102 __flush_dcache_area(page_address(pend_page), LPI_PENDBASE_SZ);
1104 paddr = page_to_phys(pend_page);
1105 pr_info("CPU%d: using LPI pending table @%pa\n",
1106 smp_processor_id(), &paddr);
1107 gic_data_rdist()->pend_page = pend_page;
1111 val = readl_relaxed(rbase + GICR_CTLR);
1112 val &= ~GICR_CTLR_ENABLE_LPIS;
1113 writel_relaxed(val, rbase + GICR_CTLR);
1116 * Make sure any change to the table is observable by the GIC.
1121 val = (page_to_phys(gic_rdists->prop_page) |
1122 GICR_PROPBASER_InnerShareable |
1123 GICR_PROPBASER_WaWb |
1124 ((LPI_NRBITS - 1) & GICR_PROPBASER_IDBITS_MASK));
1126 writeq_relaxed(val, rbase + GICR_PROPBASER);
1127 tmp = readq_relaxed(rbase + GICR_PROPBASER);
1129 if ((tmp ^ val) & GICR_PROPBASER_SHAREABILITY_MASK) {
1130 if (!(tmp & GICR_PROPBASER_SHAREABILITY_MASK)) {
1132 * The HW reports non-shareable, we must
1133 * remove the cacheability attributes as
1136 val &= ~(GICR_PROPBASER_SHAREABILITY_MASK |
1137 GICR_PROPBASER_CACHEABILITY_MASK);
1138 val |= GICR_PROPBASER_nC;
1139 writeq_relaxed(val, rbase + GICR_PROPBASER);
1141 pr_info_once("GIC: using cache flushing for LPI property table\n");
1142 gic_rdists->flags |= RDIST_FLAGS_PROPBASE_NEEDS_FLUSHING;
1146 val = (page_to_phys(pend_page) |
1147 GICR_PENDBASER_InnerShareable |
1148 GICR_PENDBASER_WaWb);
1150 writeq_relaxed(val, rbase + GICR_PENDBASER);
1151 tmp = readq_relaxed(rbase + GICR_PENDBASER);
1153 if (!(tmp & GICR_PENDBASER_SHAREABILITY_MASK)) {
1155 * The HW reports non-shareable, we must remove the
1156 * cacheability attributes as well.
1158 val &= ~(GICR_PENDBASER_SHAREABILITY_MASK |
1159 GICR_PENDBASER_CACHEABILITY_MASK);
1160 val |= GICR_PENDBASER_nC;
1161 writeq_relaxed(val, rbase + GICR_PENDBASER);
1165 val = readl_relaxed(rbase + GICR_CTLR);
1166 val |= GICR_CTLR_ENABLE_LPIS;
1167 writel_relaxed(val, rbase + GICR_CTLR);
1169 /* Make sure the GIC has seen the above */
1173 static void its_cpu_init_collection(void)
1175 struct its_node *its;
1178 spin_lock(&its_lock);
1179 cpu = smp_processor_id();
1181 list_for_each_entry(its, &its_nodes, entry) {
1184 /* avoid cross node collections and its mapping */
1185 if (its->flags & ITS_FLAGS_WORKAROUND_CAVIUM_23144) {
1186 struct device_node *cpu_node;
1188 cpu_node = of_get_cpu_node(cpu, NULL);
1189 if (its->numa_node != NUMA_NO_NODE &&
1190 its->numa_node != of_node_to_nid(cpu_node))
1195 * We now have to bind each collection to its target
1198 if (readq_relaxed(its->base + GITS_TYPER) & GITS_TYPER_PTA) {
1200 * This ITS wants the physical address of the
1203 target = gic_data_rdist()->phys_base;
1206 * This ITS wants a linear CPU number.
1208 target = readq_relaxed(gic_data_rdist_rd_base() + GICR_TYPER);
1209 target = GICR_TYPER_CPU_NUMBER(target) << 16;
1212 /* Perform collection mapping */
1213 its->collections[cpu].target_address = target;
1214 its->collections[cpu].col_id = cpu;
1216 its_send_mapc(its, &its->collections[cpu], 1);
1217 its_send_invall(its, &its->collections[cpu]);
1220 spin_unlock(&its_lock);
1223 static struct its_device *its_find_device(struct its_node *its, u32 dev_id)
1225 struct its_device *its_dev = NULL, *tmp;
1226 unsigned long flags;
1228 raw_spin_lock_irqsave(&its->lock, flags);
1230 list_for_each_entry(tmp, &its->its_device_list, entry) {
1231 if (tmp->device_id == dev_id) {
1237 raw_spin_unlock_irqrestore(&its->lock, flags);
1242 static struct its_baser *its_get_baser(struct its_node *its, u32 type)
1246 for (i = 0; i < GITS_BASER_NR_REGS; i++) {
1247 if (GITS_BASER_TYPE(its->tables[i].val) == type)
1248 return &its->tables[i];
1254 static bool its_alloc_device_table(struct its_node *its, u32 dev_id)
1256 struct its_baser *baser;
1261 baser = its_get_baser(its, GITS_BASER_TYPE_DEVICE);
1263 /* Don't allow device id that exceeds ITS hardware limit */
1265 return (ilog2(dev_id) < its->device_ids);
1267 /* Don't allow device id that exceeds single, flat table limit */
1268 esz = GITS_BASER_ENTRY_SIZE(baser->val);
1269 if (!(baser->val & GITS_BASER_INDIRECT))
1270 return (dev_id < (PAGE_ORDER_TO_SIZE(baser->order) / esz));
1272 /* Compute 1st level table index & check if that exceeds table limit */
1273 idx = dev_id >> ilog2(baser->psz / esz);
1274 if (idx >= (PAGE_ORDER_TO_SIZE(baser->order) / GITS_LVL1_ENTRY_SIZE))
1277 table = baser->base;
1279 /* Allocate memory for 2nd level table */
1281 page = alloc_pages(GFP_KERNEL | __GFP_ZERO, get_order(baser->psz));
1285 /* Flush Lvl2 table to PoC if hw doesn't support coherency */
1286 if (!(baser->val & GITS_BASER_SHAREABILITY_MASK))
1287 __flush_dcache_area(page_address(page), baser->psz);
1289 table[idx] = cpu_to_le64(page_to_phys(page) | GITS_BASER_VALID);
1291 /* Flush Lvl1 entry to PoC if hw doesn't support coherency */
1292 if (!(baser->val & GITS_BASER_SHAREABILITY_MASK))
1293 __flush_dcache_area(table + idx, GITS_LVL1_ENTRY_SIZE);
1295 /* Ensure updated table contents are visible to ITS hardware */
1302 static struct its_device *its_create_device(struct its_node *its, u32 dev_id,
1305 struct its_device *dev;
1306 unsigned long *lpi_map;
1307 unsigned long flags;
1308 u16 *col_map = NULL;
1315 if (!its_alloc_device_table(its, dev_id))
1318 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
1320 * At least one bit of EventID is being used, hence a minimum
1321 * of two entries. No, the architecture doesn't let you
1322 * express an ITT with a single entry.
1324 nr_ites = max(2UL, roundup_pow_of_two(nvecs));
1325 sz = nr_ites * its->ite_size;
1326 sz = max(sz, ITS_ITT_ALIGN) + ITS_ITT_ALIGN - 1;
1327 itt = kzalloc(sz, GFP_KERNEL);
1328 lpi_map = its_lpi_alloc_chunks(nvecs, &lpi_base, &nr_lpis);
1330 col_map = kzalloc(sizeof(*col_map) * nr_lpis, GFP_KERNEL);
1332 if (!dev || !itt || !lpi_map || !col_map) {
1340 __flush_dcache_area(itt, sz);
1344 dev->nr_ites = nr_ites;
1345 dev->event_map.lpi_map = lpi_map;
1346 dev->event_map.col_map = col_map;
1347 dev->event_map.lpi_base = lpi_base;
1348 dev->event_map.nr_lpis = nr_lpis;
1349 dev->device_id = dev_id;
1350 INIT_LIST_HEAD(&dev->entry);
1352 raw_spin_lock_irqsave(&its->lock, flags);
1353 list_add(&dev->entry, &its->its_device_list);
1354 raw_spin_unlock_irqrestore(&its->lock, flags);
1356 /* Map device to its ITT */
1357 its_send_mapd(dev, 1);
1362 static void its_free_device(struct its_device *its_dev)
1364 unsigned long flags;
1366 raw_spin_lock_irqsave(&its_dev->its->lock, flags);
1367 list_del(&its_dev->entry);
1368 raw_spin_unlock_irqrestore(&its_dev->its->lock, flags);
1369 kfree(its_dev->itt);
1373 static int its_alloc_device_irq(struct its_device *dev, irq_hw_number_t *hwirq)
1377 idx = find_first_zero_bit(dev->event_map.lpi_map,
1378 dev->event_map.nr_lpis);
1379 if (idx == dev->event_map.nr_lpis)
1382 *hwirq = dev->event_map.lpi_base + idx;
1383 set_bit(idx, dev->event_map.lpi_map);
1388 static int its_msi_prepare(struct irq_domain *domain, struct device *dev,
1389 int nvec, msi_alloc_info_t *info)
1391 struct its_node *its;
1392 struct its_device *its_dev;
1393 struct msi_domain_info *msi_info;
1397 * We ignore "dev" entierely, and rely on the dev_id that has
1398 * been passed via the scratchpad. This limits this domain's
1399 * usefulness to upper layers that definitely know that they
1400 * are built on top of the ITS.
1402 dev_id = info->scratchpad[0].ul;
1404 msi_info = msi_get_domain_info(domain);
1405 its = msi_info->data;
1407 its_dev = its_find_device(its, dev_id);
1410 * We already have seen this ID, probably through
1411 * another alias (PCI bridge of some sort). No need to
1412 * create the device.
1414 pr_debug("Reusing ITT for devID %x\n", dev_id);
1418 its_dev = its_create_device(its, dev_id, nvec);
1422 pr_debug("ITT %d entries, %d bits\n", nvec, ilog2(nvec));
1424 info->scratchpad[0].ptr = its_dev;
1428 static struct msi_domain_ops its_msi_domain_ops = {
1429 .msi_prepare = its_msi_prepare,
1432 static int its_irq_gic_domain_alloc(struct irq_domain *domain,
1434 irq_hw_number_t hwirq)
1436 struct irq_fwspec fwspec;
1438 if (irq_domain_get_of_node(domain->parent)) {
1439 fwspec.fwnode = domain->parent->fwnode;
1440 fwspec.param_count = 3;
1441 fwspec.param[0] = GIC_IRQ_TYPE_LPI;
1442 fwspec.param[1] = hwirq;
1443 fwspec.param[2] = IRQ_TYPE_EDGE_RISING;
1444 } else if (is_fwnode_irqchip(domain->parent->fwnode)) {
1445 fwspec.fwnode = domain->parent->fwnode;
1446 fwspec.param_count = 2;
1447 fwspec.param[0] = hwirq;
1448 fwspec.param[1] = IRQ_TYPE_EDGE_RISING;
1453 return irq_domain_alloc_irqs_parent(domain, virq, 1, &fwspec);
1456 static int its_irq_domain_alloc(struct irq_domain *domain, unsigned int virq,
1457 unsigned int nr_irqs, void *args)
1459 msi_alloc_info_t *info = args;
1460 struct its_device *its_dev = info->scratchpad[0].ptr;
1461 irq_hw_number_t hwirq;
1465 for (i = 0; i < nr_irqs; i++) {
1466 err = its_alloc_device_irq(its_dev, &hwirq);
1470 err = its_irq_gic_domain_alloc(domain, virq + i, hwirq);
1474 irq_domain_set_hwirq_and_chip(domain, virq + i,
1475 hwirq, &its_irq_chip, its_dev);
1476 pr_debug("ID:%d pID:%d vID:%d\n",
1477 (int)(hwirq - its_dev->event_map.lpi_base),
1478 (int) hwirq, virq + i);
1484 static void its_irq_domain_activate(struct irq_domain *domain,
1487 struct its_device *its_dev = irq_data_get_irq_chip_data(d);
1488 u32 event = its_get_event_id(d);
1489 const struct cpumask *cpu_mask = cpu_online_mask;
1491 /* get the cpu_mask of local node */
1492 if (its_dev->its->numa_node >= 0)
1493 cpu_mask = cpumask_of_node(its_dev->its->numa_node);
1495 /* Bind the LPI to the first possible CPU */
1496 its_dev->event_map.col_map[event] = cpumask_first(cpu_mask);
1498 /* Map the GIC IRQ and event to the device */
1499 its_send_mapvi(its_dev, d->hwirq, event);
1502 static void its_irq_domain_deactivate(struct irq_domain *domain,
1505 struct its_device *its_dev = irq_data_get_irq_chip_data(d);
1506 u32 event = its_get_event_id(d);
1508 /* Stop the delivery of interrupts */
1509 its_send_discard(its_dev, event);
1512 static void its_irq_domain_free(struct irq_domain *domain, unsigned int virq,
1513 unsigned int nr_irqs)
1515 struct irq_data *d = irq_domain_get_irq_data(domain, virq);
1516 struct its_device *its_dev = irq_data_get_irq_chip_data(d);
1519 for (i = 0; i < nr_irqs; i++) {
1520 struct irq_data *data = irq_domain_get_irq_data(domain,
1522 u32 event = its_get_event_id(data);
1524 /* Mark interrupt index as unused */
1525 clear_bit(event, its_dev->event_map.lpi_map);
1527 /* Nuke the entry in the domain */
1528 irq_domain_reset_irq_data(data);
1531 /* If all interrupts have been freed, start mopping the floor */
1532 if (bitmap_empty(its_dev->event_map.lpi_map,
1533 its_dev->event_map.nr_lpis)) {
1534 its_lpi_free(&its_dev->event_map);
1536 /* Unmap device/itt */
1537 its_send_mapd(its_dev, 0);
1538 its_free_device(its_dev);
1541 irq_domain_free_irqs_parent(domain, virq, nr_irqs);
1544 static const struct irq_domain_ops its_domain_ops = {
1545 .alloc = its_irq_domain_alloc,
1546 .free = its_irq_domain_free,
1547 .activate = its_irq_domain_activate,
1548 .deactivate = its_irq_domain_deactivate,
1551 static int its_force_quiescent(void __iomem *base)
1553 u32 count = 1000000; /* 1s */
1556 val = readl_relaxed(base + GITS_CTLR);
1558 * GIC architecture specification requires the ITS to be both
1559 * disabled and quiescent for writes to GITS_BASER<n> or
1560 * GITS_CBASER to not have UNPREDICTABLE results.
1562 if ((val & GITS_CTLR_QUIESCENT) && !(val & GITS_CTLR_ENABLE))
1565 /* Disable the generation of all interrupts to this ITS */
1566 val &= ~GITS_CTLR_ENABLE;
1567 writel_relaxed(val, base + GITS_CTLR);
1569 /* Poll GITS_CTLR and wait until ITS becomes quiescent */
1571 val = readl_relaxed(base + GITS_CTLR);
1572 if (val & GITS_CTLR_QUIESCENT)
1584 static void __maybe_unused its_enable_quirk_cavium_22375(void *data)
1586 struct its_node *its = data;
1588 its->flags |= ITS_FLAGS_WORKAROUND_CAVIUM_22375;
1591 static void __maybe_unused its_enable_quirk_cavium_23144(void *data)
1593 struct its_node *its = data;
1595 its->flags |= ITS_FLAGS_WORKAROUND_CAVIUM_23144;
1598 static const struct gic_quirk its_quirks[] = {
1599 #ifdef CONFIG_CAVIUM_ERRATUM_22375
1601 .desc = "ITS: Cavium errata 22375, 24313",
1602 .iidr = 0xa100034c, /* ThunderX pass 1.x */
1604 .init = its_enable_quirk_cavium_22375,
1607 #ifdef CONFIG_CAVIUM_ERRATUM_23144
1609 .desc = "ITS: Cavium erratum 23144",
1610 .iidr = 0xa100034c, /* ThunderX pass 1.x */
1612 .init = its_enable_quirk_cavium_23144,
1619 static void its_enable_quirks(struct its_node *its)
1621 u32 iidr = readl_relaxed(its->base + GITS_IIDR);
1623 gic_enable_quirks(iidr, its_quirks, its);
1626 static int its_init_domain(struct fwnode_handle *handle, struct its_node *its)
1628 struct irq_domain *inner_domain;
1629 struct msi_domain_info *info;
1631 info = kzalloc(sizeof(*info), GFP_KERNEL);
1635 inner_domain = irq_domain_create_tree(handle, &its_domain_ops, its);
1636 if (!inner_domain) {
1641 inner_domain->parent = its_parent;
1642 inner_domain->bus_token = DOMAIN_BUS_NEXUS;
1643 info->ops = &its_msi_domain_ops;
1645 inner_domain->host_data = info;
1650 static int __init its_probe_one(struct resource *res,
1651 struct fwnode_handle *handle, int numa_node)
1653 struct its_node *its;
1654 void __iomem *its_base;
1659 its_base = ioremap(res->start, resource_size(res));
1661 pr_warn("ITS@%pa: Unable to map ITS registers\n", &res->start);
1665 val = readl_relaxed(its_base + GITS_PIDR2) & GIC_PIDR2_ARCH_MASK;
1666 if (val != 0x30 && val != 0x40) {
1667 pr_warn("ITS@%pa: No ITS detected, giving up\n", &res->start);
1672 err = its_force_quiescent(its_base);
1674 pr_warn("ITS@%pa: Failed to quiesce, giving up\n", &res->start);
1678 pr_info("ITS %pR\n", res);
1680 its = kzalloc(sizeof(*its), GFP_KERNEL);
1686 raw_spin_lock_init(&its->lock);
1687 INIT_LIST_HEAD(&its->entry);
1688 INIT_LIST_HEAD(&its->its_device_list);
1689 its->base = its_base;
1690 its->phys_base = res->start;
1691 its->ite_size = ((readl_relaxed(its_base + GITS_TYPER) >> 4) & 0xf) + 1;
1692 its->numa_node = numa_node;
1694 its->cmd_base = kzalloc(ITS_CMD_QUEUE_SZ, GFP_KERNEL);
1695 if (!its->cmd_base) {
1699 its->cmd_write = its->cmd_base;
1701 its_enable_quirks(its);
1703 err = its_alloc_tables(its);
1707 err = its_alloc_collections(its);
1709 goto out_free_tables;
1711 baser = (virt_to_phys(its->cmd_base) |
1713 GITS_CBASER_InnerShareable |
1714 (ITS_CMD_QUEUE_SZ / SZ_4K - 1) |
1717 writeq_relaxed(baser, its->base + GITS_CBASER);
1718 tmp = readq_relaxed(its->base + GITS_CBASER);
1720 if ((tmp ^ baser) & GITS_CBASER_SHAREABILITY_MASK) {
1721 if (!(tmp & GITS_CBASER_SHAREABILITY_MASK)) {
1723 * The HW reports non-shareable, we must
1724 * remove the cacheability attributes as
1727 baser &= ~(GITS_CBASER_SHAREABILITY_MASK |
1728 GITS_CBASER_CACHEABILITY_MASK);
1729 baser |= GITS_CBASER_nC;
1730 writeq_relaxed(baser, its->base + GITS_CBASER);
1732 pr_info("ITS: using cache flushing for cmd queue\n");
1733 its->flags |= ITS_FLAGS_CMDQ_NEEDS_FLUSHING;
1736 writeq_relaxed(0, its->base + GITS_CWRITER);
1737 writel_relaxed(GITS_CTLR_ENABLE, its->base + GITS_CTLR);
1739 err = its_init_domain(handle, its);
1741 goto out_free_tables;
1743 spin_lock(&its_lock);
1744 list_add(&its->entry, &its_nodes);
1745 spin_unlock(&its_lock);
1750 its_free_tables(its);
1752 kfree(its->cmd_base);
1757 pr_err("ITS@%pa: failed probing (%d)\n", &res->start, err);
1761 static bool gic_rdists_supports_plpis(void)
1763 return !!(readl_relaxed(gic_data_rdist_rd_base() + GICR_TYPER) & GICR_TYPER_PLPIS);
1766 int its_cpu_init(void)
1768 if (!list_empty(&its_nodes)) {
1769 if (!gic_rdists_supports_plpis()) {
1770 pr_info("CPU%d: LPIs not supported\n", smp_processor_id());
1773 its_cpu_init_lpis();
1774 its_cpu_init_collection();
1780 static struct of_device_id its_device_id[] = {
1781 { .compatible = "arm,gic-v3-its", },
1785 static int __init its_of_probe(struct device_node *node)
1787 struct device_node *np;
1788 struct resource res;
1790 for (np = of_find_matching_node(node, its_device_id); np;
1791 np = of_find_matching_node(np, its_device_id)) {
1792 if (!of_property_read_bool(np, "msi-controller")) {
1793 pr_warn("%s: no msi-controller property, ITS ignored\n",
1798 if (of_address_to_resource(np, 0, &res)) {
1799 pr_warn("%s: no regs?\n", np->full_name);
1803 its_probe_one(&res, &np->fwnode, of_node_to_nid(np));
1810 #define ACPI_GICV3_ITS_MEM_SIZE (SZ_128K)
1812 static int __init gic_acpi_parse_madt_its(struct acpi_subtable_header *header,
1813 const unsigned long end)
1815 struct acpi_madt_generic_translator *its_entry;
1816 struct fwnode_handle *dom_handle;
1817 struct resource res;
1820 its_entry = (struct acpi_madt_generic_translator *)header;
1821 memset(&res, 0, sizeof(res));
1822 res.start = its_entry->base_address;
1823 res.end = its_entry->base_address + ACPI_GICV3_ITS_MEM_SIZE - 1;
1824 res.flags = IORESOURCE_MEM;
1826 dom_handle = irq_domain_alloc_fwnode((void *)its_entry->base_address);
1828 pr_err("ITS@%pa: Unable to allocate GICv3 ITS domain token\n",
1833 err = iort_register_domain_token(its_entry->translation_id, dom_handle);
1835 pr_err("ITS@%pa: Unable to register GICv3 ITS domain token (ITS ID %d) to IORT\n",
1836 &res.start, its_entry->translation_id);
1840 err = its_probe_one(&res, dom_handle, NUMA_NO_NODE);
1844 iort_deregister_domain_token(its_entry->translation_id);
1846 irq_domain_free_fwnode(dom_handle);
1850 static void __init its_acpi_probe(void)
1852 acpi_table_parse_madt(ACPI_MADT_TYPE_GENERIC_TRANSLATOR,
1853 gic_acpi_parse_madt_its, 0);
1856 static void __init its_acpi_probe(void) { }
1859 int __init its_init(struct fwnode_handle *handle, struct rdists *rdists,
1860 struct irq_domain *parent_domain)
1862 struct device_node *of_node;
1864 its_parent = parent_domain;
1865 of_node = to_of_node(handle);
1867 its_of_probe(of_node);
1871 if (list_empty(&its_nodes)) {
1872 pr_warn("ITS: No ITS available, not enabling LPIs\n");
1876 gic_rdists = rdists;
1877 its_alloc_lpi_tables();
1878 its_lpi_init(rdists->id_bits);