xref: /freebsd/sys/arm64/arm64/gicv3_its.c (revision 1d386b48)
1 /*-
2  * Copyright (c) 2015-2016 The FreeBSD Foundation
3  *
4  * This software was developed by Andrew Turner under
5  * the sponsorship of the FreeBSD Foundation.
6  *
7  * This software was developed by Semihalf under
8  * the sponsorship of the FreeBSD Foundation.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 #include "opt_acpi.h"
33 #include "opt_platform.h"
34 #include "opt_iommu.h"
35 
36 #include <sys/cdefs.h>
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/bus.h>
40 #include <sys/cpuset.h>
41 #include <sys/domainset.h>
42 #include <sys/endian.h>
43 #include <sys/kernel.h>
44 #include <sys/lock.h>
45 #include <sys/malloc.h>
46 #include <sys/module.h>
47 #include <sys/mutex.h>
48 #include <sys/proc.h>
49 #include <sys/taskqueue.h>
50 #include <sys/tree.h>
51 #include <sys/queue.h>
52 #include <sys/rman.h>
53 #include <sys/sbuf.h>
54 #include <sys/smp.h>
55 #include <sys/sysctl.h>
56 #include <sys/vmem.h>
57 
58 #include <vm/vm.h>
59 #include <vm/pmap.h>
60 #include <vm/vm_page.h>
61 
62 #include <machine/bus.h>
63 #include <machine/intr.h>
64 
65 #include <arm/arm/gic_common.h>
66 #include <arm64/arm64/gic_v3_reg.h>
67 #include <arm64/arm64/gic_v3_var.h>
68 
69 #ifdef FDT
70 #include <dev/ofw/openfirm.h>
71 #include <dev/ofw/ofw_bus.h>
72 #include <dev/ofw/ofw_bus_subr.h>
73 #endif
74 #include <dev/pci/pcireg.h>
75 #include <dev/pci/pcivar.h>
76 
77 #ifdef IOMMU
78 #include <dev/iommu/iommu.h>
79 #include <dev/iommu/iommu_gas.h>
80 #endif
81 
82 #include "pcib_if.h"
83 #include "pic_if.h"
84 #include "msi_if.h"
85 
86 MALLOC_DEFINE(M_GICV3_ITS, "GICv3 ITS",
87     "ARM GICv3 Interrupt Translation Service");
88 
89 #define	LPI_NIRQS		(64 * 1024)
90 
91 /* The size and alignment of the command circular buffer */
92 #define	ITS_CMDQ_SIZE		(64 * 1024)	/* Must be a multiple of 4K */
93 #define	ITS_CMDQ_ALIGN		(64 * 1024)
94 
95 #define	LPI_CONFTAB_SIZE	LPI_NIRQS
96 #define	LPI_CONFTAB_ALIGN	(64 * 1024)
97 #define	LPI_CONFTAB_MAX_ADDR	((1ul << 48) - 1) /* We need a 47 bit PA */
98 
99 /* 1 bit per SPI, PPI, and SGI (8k), and 1 bit per LPI (LPI_CONFTAB_SIZE) */
100 #define	LPI_PENDTAB_SIZE	((LPI_NIRQS + GIC_FIRST_LPI) / 8)
101 #define	LPI_PENDTAB_ALIGN	(64 * 1024)
102 #define	LPI_PENDTAB_MAX_ADDR	((1ul << 48) - 1) /* We need a 47 bit PA */
103 
104 #define	LPI_INT_TRANS_TAB_ALIGN	256
105 #define	LPI_INT_TRANS_TAB_MAX_ADDR ((1ul << 48) - 1)
106 
107 /* ITS commands encoding */
108 #define	ITS_CMD_MOVI		(0x01)
109 #define	ITS_CMD_SYNC		(0x05)
110 #define	ITS_CMD_MAPD		(0x08)
111 #define	ITS_CMD_MAPC		(0x09)
112 #define	ITS_CMD_MAPTI		(0x0a)
113 #define	ITS_CMD_MAPI		(0x0b)
114 #define	ITS_CMD_INV		(0x0c)
115 #define	ITS_CMD_INVALL		(0x0d)
116 /* Command */
117 #define	CMD_COMMAND_MASK	(0xFFUL)
118 /* PCI device ID */
119 #define	CMD_DEVID_SHIFT		(32)
120 #define	CMD_DEVID_MASK		(0xFFFFFFFFUL << CMD_DEVID_SHIFT)
121 /* Size of IRQ ID bitfield */
122 #define	CMD_SIZE_MASK		(0xFFUL)
123 /* Virtual LPI ID */
124 #define	CMD_ID_MASK		(0xFFFFFFFFUL)
125 /* Physical LPI ID */
126 #define	CMD_PID_SHIFT		(32)
127 #define	CMD_PID_MASK		(0xFFFFFFFFUL << CMD_PID_SHIFT)
128 /* Collection */
129 #define	CMD_COL_MASK		(0xFFFFUL)
130 /* Target (CPU or Re-Distributor) */
131 #define	CMD_TARGET_SHIFT	(16)
132 #define	CMD_TARGET_MASK		(0xFFFFFFFFUL << CMD_TARGET_SHIFT)
133 /* Interrupt Translation Table address */
134 #define	CMD_ITT_MASK		(0xFFFFFFFFFF00UL)
135 /* Valid command bit */
136 #define	CMD_VALID_SHIFT		(63)
137 #define	CMD_VALID_MASK		(1UL << CMD_VALID_SHIFT)
138 
139 #define	ITS_TARGET_NONE		0xFBADBEEF
140 
141 /* LPI chunk owned by ITS device */
142 struct lpi_chunk {
143 	u_int	lpi_base;
144 	u_int	lpi_free;	/* First free LPI in set */
145 	u_int	lpi_num;	/* Total number of LPIs in chunk */
146 	u_int	lpi_busy;	/* Number of busy LPIs in chink */
147 };
148 
149 /* ITS device */
150 struct its_dev {
151 	TAILQ_ENTRY(its_dev)	entry;
152 	/* PCI device */
153 	device_t		pci_dev;
154 	/* Device ID (i.e. PCI device ID) */
155 	uint32_t		devid;
156 	/* List of assigned LPIs */
157 	struct lpi_chunk	lpis;
158 	/* Virtual address of ITT */
159 	vm_offset_t		itt;
160 	size_t			itt_size;
161 };
162 
163 /*
164  * ITS command descriptor.
165  * Idea for command description passing taken from Linux.
166  */
167 struct its_cmd_desc {
168 	uint8_t cmd_type;
169 
170 	union {
171 		struct {
172 			struct its_dev *its_dev;
173 			struct its_col *col;
174 			uint32_t id;
175 		} cmd_desc_movi;
176 
177 		struct {
178 			struct its_col *col;
179 		} cmd_desc_sync;
180 
181 		struct {
182 			struct its_col *col;
183 			uint8_t valid;
184 		} cmd_desc_mapc;
185 
186 		struct {
187 			struct its_dev *its_dev;
188 			struct its_col *col;
189 			uint32_t pid;
190 			uint32_t id;
191 		} cmd_desc_mapvi;
192 
193 		struct {
194 			struct its_dev *its_dev;
195 			struct its_col *col;
196 			uint32_t pid;
197 		} cmd_desc_mapi;
198 
199 		struct {
200 			struct its_dev *its_dev;
201 			uint8_t valid;
202 		} cmd_desc_mapd;
203 
204 		struct {
205 			struct its_dev *its_dev;
206 			struct its_col *col;
207 			uint32_t pid;
208 		} cmd_desc_inv;
209 
210 		struct {
211 			struct its_col *col;
212 		} cmd_desc_invall;
213 	};
214 };
215 
216 /* ITS command. Each command is 32 bytes long */
217 struct its_cmd {
218 	uint64_t	cmd_dword[4];	/* ITS command double word */
219 };
220 
221 /* An ITS private table */
222 struct its_ptable {
223 	vm_offset_t	ptab_vaddr;
224 	unsigned long	ptab_size;
225 };
226 
227 /* ITS collection description. */
228 struct its_col {
229 	uint64_t	col_target;	/* Target Re-Distributor */
230 	uint64_t	col_id;		/* Collection ID */
231 };
232 
233 struct gicv3_its_irqsrc {
234 	struct intr_irqsrc	gi_isrc;
235 	u_int			gi_id;
236 	u_int			gi_lpi;
237 	struct its_dev		*gi_its_dev;
238 	TAILQ_ENTRY(gicv3_its_irqsrc) gi_link;
239 };
240 
241 struct gicv3_its_softc {
242 	device_t	dev;
243 	struct intr_pic *sc_pic;
244 	struct resource *sc_its_res;
245 
246 	cpuset_t	sc_cpus;
247 	struct domainset *sc_ds;
248 	u_int		gic_irq_cpu;
249 
250 	struct its_ptable sc_its_ptab[GITS_BASER_NUM];
251 	struct its_col *sc_its_cols[MAXCPU];	/* Per-CPU collections */
252 
253 	/*
254 	 * TODO: We should get these from the parent as we only want a
255 	 * single copy of each across the interrupt controller.
256 	 */
257 	uint8_t		*sc_conf_base;
258 	vm_offset_t sc_pend_base[MAXCPU];
259 
260 	/* Command handling */
261 	struct mtx sc_its_cmd_lock;
262 	struct its_cmd *sc_its_cmd_base; /* Command circular buffer address */
263 	size_t sc_its_cmd_next_idx;
264 
265 	vmem_t *sc_irq_alloc;
266 	struct gicv3_its_irqsrc	**sc_irqs;
267 	u_int	sc_irq_base;
268 	u_int	sc_irq_length;
269 	u_int	sc_irq_count;
270 
271 	struct mtx sc_its_dev_lock;
272 	TAILQ_HEAD(its_dev_list, its_dev) sc_its_dev_list;
273 	TAILQ_HEAD(free_irqs, gicv3_its_irqsrc) sc_free_irqs;
274 
275 #define	ITS_FLAGS_CMDQ_FLUSH		0x00000001
276 #define	ITS_FLAGS_LPI_CONF_FLUSH	0x00000002
277 #define	ITS_FLAGS_ERRATA_CAVIUM_22375	0x00000004
278 	u_int sc_its_flags;
279 	bool	trace_enable;
280 	vm_page_t ma; /* fake msi page */
281 };
282 
283 static void *conf_base;
284 
285 typedef void (its_quirk_func_t)(device_t);
286 static its_quirk_func_t its_quirk_cavium_22375;
287 
288 static const struct {
289 	const char *desc;
290 	uint32_t iidr;
291 	uint32_t iidr_mask;
292 	its_quirk_func_t *func;
293 } its_quirks[] = {
294 	{
295 		/* Cavium ThunderX Pass 1.x */
296 		.desc = "Cavium ThunderX errata: 22375, 24313",
297 		.iidr = GITS_IIDR_RAW(GITS_IIDR_IMPL_CAVIUM,
298 		    GITS_IIDR_PROD_THUNDER, GITS_IIDR_VAR_THUNDER_1, 0),
299 		.iidr_mask = ~GITS_IIDR_REVISION_MASK,
300 		.func = its_quirk_cavium_22375,
301 	},
302 };
303 
304 #define	gic_its_read_4(sc, reg)			\
305     bus_read_4((sc)->sc_its_res, (reg))
306 #define	gic_its_read_8(sc, reg)			\
307     bus_read_8((sc)->sc_its_res, (reg))
308 
309 #define	gic_its_write_4(sc, reg, val)		\
310     bus_write_4((sc)->sc_its_res, (reg), (val))
311 #define	gic_its_write_8(sc, reg, val)		\
312     bus_write_8((sc)->sc_its_res, (reg), (val))
313 
314 static device_attach_t gicv3_its_attach;
315 static device_detach_t gicv3_its_detach;
316 
317 static pic_disable_intr_t gicv3_its_disable_intr;
318 static pic_enable_intr_t gicv3_its_enable_intr;
319 static pic_map_intr_t gicv3_its_map_intr;
320 static pic_setup_intr_t gicv3_its_setup_intr;
321 static pic_post_filter_t gicv3_its_post_filter;
322 static pic_post_ithread_t gicv3_its_post_ithread;
323 static pic_pre_ithread_t gicv3_its_pre_ithread;
324 static pic_bind_intr_t gicv3_its_bind_intr;
325 #ifdef SMP
326 static pic_init_secondary_t gicv3_its_init_secondary;
327 #endif
328 static msi_alloc_msi_t gicv3_its_alloc_msi;
329 static msi_release_msi_t gicv3_its_release_msi;
330 static msi_alloc_msix_t gicv3_its_alloc_msix;
331 static msi_release_msix_t gicv3_its_release_msix;
332 static msi_map_msi_t gicv3_its_map_msi;
333 #ifdef IOMMU
334 static msi_iommu_init_t gicv3_iommu_init;
335 static msi_iommu_deinit_t gicv3_iommu_deinit;
336 #endif
337 
338 static void its_cmd_movi(device_t, struct gicv3_its_irqsrc *);
339 static void its_cmd_mapc(device_t, struct its_col *, uint8_t);
340 static void its_cmd_mapti(device_t, struct gicv3_its_irqsrc *);
341 static void its_cmd_mapd(device_t, struct its_dev *, uint8_t);
342 static void its_cmd_inv(device_t, struct its_dev *, struct gicv3_its_irqsrc *);
343 static void its_cmd_invall(device_t, struct its_col *);
344 
345 static device_method_t gicv3_its_methods[] = {
346 	/* Device interface */
347 	DEVMETHOD(device_detach,	gicv3_its_detach),
348 
349 	/* Interrupt controller interface */
350 	DEVMETHOD(pic_disable_intr,	gicv3_its_disable_intr),
351 	DEVMETHOD(pic_enable_intr,	gicv3_its_enable_intr),
352 	DEVMETHOD(pic_map_intr,		gicv3_its_map_intr),
353 	DEVMETHOD(pic_setup_intr,	gicv3_its_setup_intr),
354 	DEVMETHOD(pic_post_filter,	gicv3_its_post_filter),
355 	DEVMETHOD(pic_post_ithread,	gicv3_its_post_ithread),
356 	DEVMETHOD(pic_pre_ithread,	gicv3_its_pre_ithread),
357 #ifdef SMP
358 	DEVMETHOD(pic_bind_intr,	gicv3_its_bind_intr),
359 	DEVMETHOD(pic_init_secondary,	gicv3_its_init_secondary),
360 #endif
361 
362 	/* MSI/MSI-X */
363 	DEVMETHOD(msi_alloc_msi,	gicv3_its_alloc_msi),
364 	DEVMETHOD(msi_release_msi,	gicv3_its_release_msi),
365 	DEVMETHOD(msi_alloc_msix,	gicv3_its_alloc_msix),
366 	DEVMETHOD(msi_release_msix,	gicv3_its_release_msix),
367 	DEVMETHOD(msi_map_msi,		gicv3_its_map_msi),
368 #ifdef IOMMU
369 	DEVMETHOD(msi_iommu_init,	gicv3_iommu_init),
370 	DEVMETHOD(msi_iommu_deinit,	gicv3_iommu_deinit),
371 #endif
372 
373 	/* End */
374 	DEVMETHOD_END
375 };
376 
377 static DEFINE_CLASS_0(gic, gicv3_its_driver, gicv3_its_methods,
378     sizeof(struct gicv3_its_softc));
379 
380 static void
381 gicv3_its_cmdq_init(struct gicv3_its_softc *sc)
382 {
383 	vm_paddr_t cmd_paddr;
384 	uint64_t reg, tmp;
385 
386 	/* Set up the command circular buffer */
387 	sc->sc_its_cmd_base = contigmalloc_domainset(ITS_CMDQ_SIZE, M_GICV3_ITS,
388 	    sc->sc_ds, M_WAITOK | M_ZERO, 0, (1ul << 48) - 1, ITS_CMDQ_ALIGN,
389 	    0);
390 	sc->sc_its_cmd_next_idx = 0;
391 
392 	cmd_paddr = vtophys(sc->sc_its_cmd_base);
393 
394 	/* Set the base of the command buffer */
395 	reg = GITS_CBASER_VALID |
396 	    (GITS_CBASER_CACHE_NIWAWB << GITS_CBASER_CACHE_SHIFT) |
397 	    cmd_paddr | (GITS_CBASER_SHARE_IS << GITS_CBASER_SHARE_SHIFT) |
398 	    (ITS_CMDQ_SIZE / 4096 - 1);
399 	gic_its_write_8(sc, GITS_CBASER, reg);
400 
401 	/* Read back to check for fixed value fields */
402 	tmp = gic_its_read_8(sc, GITS_CBASER);
403 
404 	if ((tmp & GITS_CBASER_SHARE_MASK) !=
405 	    (GITS_CBASER_SHARE_IS << GITS_CBASER_SHARE_SHIFT)) {
406 		/* Check if the hardware reported non-shareable */
407 		if ((tmp & GITS_CBASER_SHARE_MASK) ==
408 		    (GITS_CBASER_SHARE_NS << GITS_CBASER_SHARE_SHIFT)) {
409 			/* If so remove the cache attribute */
410 			reg &= ~GITS_CBASER_CACHE_MASK;
411 			reg &= ~GITS_CBASER_SHARE_MASK;
412 			/* Set to Non-cacheable, Non-shareable */
413 			reg |= GITS_CBASER_CACHE_NIN << GITS_CBASER_CACHE_SHIFT;
414 			reg |= GITS_CBASER_SHARE_NS << GITS_CBASER_SHARE_SHIFT;
415 
416 			gic_its_write_8(sc, GITS_CBASER, reg);
417 		}
418 
419 		/* The command queue has to be flushed after each command */
420 		sc->sc_its_flags |= ITS_FLAGS_CMDQ_FLUSH;
421 	}
422 
423 	/* Get the next command from the start of the buffer */
424 	gic_its_write_8(sc, GITS_CWRITER, 0x0);
425 }
426 
427 static int
428 gicv3_its_table_init(device_t dev, struct gicv3_its_softc *sc)
429 {
430 	vm_offset_t table;
431 	vm_paddr_t paddr;
432 	uint64_t cache, reg, share, tmp, type;
433 	size_t esize, its_tbl_size, nidents, nitspages, npages;
434 	int i, page_size;
435 	int devbits;
436 
437 	if ((sc->sc_its_flags & ITS_FLAGS_ERRATA_CAVIUM_22375) != 0) {
438 		/*
439 		 * GITS_TYPER[17:13] of ThunderX reports that device IDs
440 		 * are to be 21 bits in length. The entry size of the ITS
441 		 * table can be read from GITS_BASERn[52:48] and on ThunderX
442 		 * is supposed to be 8 bytes in length (for device table).
443 		 * Finally the page size that is to be used by ITS to access
444 		 * this table will be set to 64KB.
445 		 *
446 		 * This gives 0x200000 entries of size 0x8 bytes covered by
447 		 * 256 pages each of which 64KB in size. The number of pages
448 		 * (minus 1) should then be written to GITS_BASERn[7:0]. In
449 		 * that case this value would be 0xFF but on ThunderX the
450 		 * maximum value that HW accepts is 0xFD.
451 		 *
452 		 * Set an arbitrary number of device ID bits to 20 in order
453 		 * to limit the number of entries in ITS device table to
454 		 * 0x100000 and the table size to 8MB.
455 		 */
456 		devbits = 20;
457 		cache = 0;
458 	} else {
459 		devbits = GITS_TYPER_DEVB(gic_its_read_8(sc, GITS_TYPER));
460 		cache = GITS_BASER_CACHE_WAWB;
461 	}
462 	share = GITS_BASER_SHARE_IS;
463 	page_size = PAGE_SIZE_64K;
464 
465 	for (i = 0; i < GITS_BASER_NUM; i++) {
466 		reg = gic_its_read_8(sc, GITS_BASER(i));
467 		/* The type of table */
468 		type = GITS_BASER_TYPE(reg);
469 		/* The table entry size */
470 		esize = GITS_BASER_ESIZE(reg);
471 
472 		switch(type) {
473 		case GITS_BASER_TYPE_DEV:
474 			nidents = (1 << devbits);
475 			its_tbl_size = esize * nidents;
476 			its_tbl_size = roundup2(its_tbl_size, PAGE_SIZE_64K);
477 			break;
478 		case GITS_BASER_TYPE_VP:
479 		case GITS_BASER_TYPE_PP: /* Undocumented? */
480 		case GITS_BASER_TYPE_IC:
481 			its_tbl_size = page_size;
482 			break;
483 		default:
484 			continue;
485 		}
486 		npages = howmany(its_tbl_size, PAGE_SIZE);
487 
488 		/* Allocate the table */
489 		table = (vm_offset_t)contigmalloc_domainset(npages * PAGE_SIZE,
490 		    M_GICV3_ITS, sc->sc_ds, M_WAITOK | M_ZERO, 0,
491 		    (1ul << 48) - 1, PAGE_SIZE_64K, 0);
492 
493 		sc->sc_its_ptab[i].ptab_vaddr = table;
494 		sc->sc_its_ptab[i].ptab_size = npages * PAGE_SIZE;
495 
496 		paddr = vtophys(table);
497 
498 		while (1) {
499 			nitspages = howmany(its_tbl_size, page_size);
500 
501 			/* Clear the fields we will be setting */
502 			reg &= ~(GITS_BASER_VALID | GITS_BASER_INDIRECT |
503 			    GITS_BASER_CACHE_MASK | GITS_BASER_TYPE_MASK |
504 			    GITS_BASER_ESIZE_MASK | GITS_BASER_PA_MASK |
505 			    GITS_BASER_SHARE_MASK | GITS_BASER_PSZ_MASK |
506 			    GITS_BASER_SIZE_MASK);
507 			/* Set the new values */
508 			reg |= GITS_BASER_VALID |
509 			    (cache << GITS_BASER_CACHE_SHIFT) |
510 			    (type << GITS_BASER_TYPE_SHIFT) |
511 			    ((esize - 1) << GITS_BASER_ESIZE_SHIFT) |
512 			    paddr | (share << GITS_BASER_SHARE_SHIFT) |
513 			    (nitspages - 1);
514 
515 			switch (page_size) {
516 			case PAGE_SIZE_4K:	/* 4KB */
517 				reg |=
518 				    GITS_BASER_PSZ_4K << GITS_BASER_PSZ_SHIFT;
519 				break;
520 			case PAGE_SIZE_16K:	/* 16KB */
521 				reg |=
522 				    GITS_BASER_PSZ_16K << GITS_BASER_PSZ_SHIFT;
523 				break;
524 			case PAGE_SIZE_64K:	/* 64KB */
525 				reg |=
526 				    GITS_BASER_PSZ_64K << GITS_BASER_PSZ_SHIFT;
527 				break;
528 			}
529 
530 			gic_its_write_8(sc, GITS_BASER(i), reg);
531 
532 			/* Read back to check */
533 			tmp = gic_its_read_8(sc, GITS_BASER(i));
534 
535 			/* Do the shareability masks line up? */
536 			if ((tmp & GITS_BASER_SHARE_MASK) !=
537 			    (reg & GITS_BASER_SHARE_MASK)) {
538 				share = (tmp & GITS_BASER_SHARE_MASK) >>
539 				    GITS_BASER_SHARE_SHIFT;
540 				continue;
541 			}
542 
543 			if ((tmp & GITS_BASER_PSZ_MASK) !=
544 			    (reg & GITS_BASER_PSZ_MASK)) {
545 				switch (page_size) {
546 				case PAGE_SIZE_16K:
547 					page_size = PAGE_SIZE_4K;
548 					continue;
549 				case PAGE_SIZE_64K:
550 					page_size = PAGE_SIZE_16K;
551 					continue;
552 				}
553 			}
554 
555 			if (tmp != reg) {
556 				device_printf(dev, "GITS_BASER%d: "
557 				    "unable to be updated: %lx != %lx\n",
558 				    i, reg, tmp);
559 				return (ENXIO);
560 			}
561 
562 			/* We should have made all needed changes */
563 			break;
564 		}
565 	}
566 
567 	return (0);
568 }
569 
570 static void
571 gicv3_its_conftable_init(struct gicv3_its_softc *sc)
572 {
573 	void *conf_table;
574 
575 	conf_table = atomic_load_ptr(&conf_base);
576 	if (conf_table == NULL) {
577 		conf_table = contigmalloc(LPI_CONFTAB_SIZE,
578 		    M_GICV3_ITS, M_WAITOK, 0, LPI_CONFTAB_MAX_ADDR,
579 		    LPI_CONFTAB_ALIGN, 0);
580 
581 		if (atomic_cmpset_ptr((uintptr_t *)&conf_base,
582 		    (uintptr_t)NULL, (uintptr_t)conf_table) == 0) {
583 			contigfree(conf_table, LPI_CONFTAB_SIZE, M_GICV3_ITS);
584 			conf_table = atomic_load_ptr(&conf_base);
585 		}
586 	}
587 	sc->sc_conf_base = conf_table;
588 
589 	/* Set the default configuration */
590 	memset(sc->sc_conf_base, GIC_PRIORITY_MAX | LPI_CONF_GROUP1,
591 	    LPI_CONFTAB_SIZE);
592 
593 	/* Flush the table to memory */
594 	cpu_dcache_wb_range((vm_offset_t)sc->sc_conf_base, LPI_CONFTAB_SIZE);
595 }
596 
597 static void
598 gicv3_its_pendtables_init(struct gicv3_its_softc *sc)
599 {
600 	int i;
601 
602 	for (i = 0; i <= mp_maxid; i++) {
603 		if (CPU_ISSET(i, &sc->sc_cpus) == 0)
604 			continue;
605 
606 		sc->sc_pend_base[i] = (vm_offset_t)contigmalloc(
607 		    LPI_PENDTAB_SIZE, M_GICV3_ITS, M_WAITOK | M_ZERO,
608 		    0, LPI_PENDTAB_MAX_ADDR, LPI_PENDTAB_ALIGN, 0);
609 
610 		/* Flush so the ITS can see the memory */
611 		cpu_dcache_wb_range((vm_offset_t)sc->sc_pend_base[i],
612 		    LPI_PENDTAB_SIZE);
613 	}
614 }
615 
616 static void
617 its_init_cpu_lpi(device_t dev, struct gicv3_its_softc *sc)
618 {
619 	device_t gicv3;
620 	uint64_t xbaser, tmp;
621 	uint32_t ctlr;
622 	u_int cpuid;
623 
624 	gicv3 = device_get_parent(dev);
625 	cpuid = PCPU_GET(cpuid);
626 
627 	/* Disable LPIs */
628 	ctlr = gic_r_read_4(gicv3, GICR_CTLR);
629 	ctlr &= ~GICR_CTLR_LPI_ENABLE;
630 	gic_r_write_4(gicv3, GICR_CTLR, ctlr);
631 
632 	/* Make sure changes are observable my the GIC */
633 	dsb(sy);
634 
635 	/*
636 	 * Set the redistributor base
637 	 */
638 	xbaser = vtophys(sc->sc_conf_base) |
639 	    (GICR_PROPBASER_SHARE_IS << GICR_PROPBASER_SHARE_SHIFT) |
640 	    (GICR_PROPBASER_CACHE_NIWAWB << GICR_PROPBASER_CACHE_SHIFT) |
641 	    (flsl(LPI_CONFTAB_SIZE | GIC_FIRST_LPI) - 1);
642 	gic_r_write_8(gicv3, GICR_PROPBASER, xbaser);
643 
644 	/* Check the cache attributes we set */
645 	tmp = gic_r_read_8(gicv3, GICR_PROPBASER);
646 
647 	if ((tmp & GICR_PROPBASER_SHARE_MASK) !=
648 	    (xbaser & GICR_PROPBASER_SHARE_MASK)) {
649 		if ((tmp & GICR_PROPBASER_SHARE_MASK) ==
650 		    (GICR_PROPBASER_SHARE_NS << GICR_PROPBASER_SHARE_SHIFT)) {
651 			/* We need to mark as non-cacheable */
652 			xbaser &= ~(GICR_PROPBASER_SHARE_MASK |
653 			    GICR_PROPBASER_CACHE_MASK);
654 			/* Non-cacheable */
655 			xbaser |= GICR_PROPBASER_CACHE_NIN <<
656 			    GICR_PROPBASER_CACHE_SHIFT;
657 			/* Non-shareable */
658 			xbaser |= GICR_PROPBASER_SHARE_NS <<
659 			    GICR_PROPBASER_SHARE_SHIFT;
660 			gic_r_write_8(gicv3, GICR_PROPBASER, xbaser);
661 		}
662 		sc->sc_its_flags |= ITS_FLAGS_LPI_CONF_FLUSH;
663 	}
664 
665 	/*
666 	 * Set the LPI pending table base
667 	 */
668 	xbaser = vtophys(sc->sc_pend_base[cpuid]) |
669 	    (GICR_PENDBASER_CACHE_NIWAWB << GICR_PENDBASER_CACHE_SHIFT) |
670 	    (GICR_PENDBASER_SHARE_IS << GICR_PENDBASER_SHARE_SHIFT);
671 
672 	gic_r_write_8(gicv3, GICR_PENDBASER, xbaser);
673 
674 	tmp = gic_r_read_8(gicv3, GICR_PENDBASER);
675 
676 	if ((tmp & GICR_PENDBASER_SHARE_MASK) ==
677 	    (GICR_PENDBASER_SHARE_NS << GICR_PENDBASER_SHARE_SHIFT)) {
678 		/* Clear the cahce and shareability bits */
679 		xbaser &= ~(GICR_PENDBASER_CACHE_MASK |
680 		    GICR_PENDBASER_SHARE_MASK);
681 		/* Mark as non-shareable */
682 		xbaser |= GICR_PENDBASER_SHARE_NS << GICR_PENDBASER_SHARE_SHIFT;
683 		/* And non-cacheable */
684 		xbaser |= GICR_PENDBASER_CACHE_NIN <<
685 		    GICR_PENDBASER_CACHE_SHIFT;
686 	}
687 
688 	/* Enable LPIs */
689 	ctlr = gic_r_read_4(gicv3, GICR_CTLR);
690 	ctlr |= GICR_CTLR_LPI_ENABLE;
691 	gic_r_write_4(gicv3, GICR_CTLR, ctlr);
692 
693 	/* Make sure the GIC has seen everything */
694 	dsb(sy);
695 }
696 
697 static int
698 its_init_cpu(device_t dev, struct gicv3_its_softc *sc)
699 {
700 	device_t gicv3;
701 	vm_paddr_t target;
702 	u_int cpuid;
703 	struct redist_pcpu *rpcpu;
704 
705 	gicv3 = device_get_parent(dev);
706 	cpuid = PCPU_GET(cpuid);
707 	if (!CPU_ISSET(cpuid, &sc->sc_cpus))
708 		return (0);
709 
710 	/* Check if the ITS is enabled on this CPU */
711 	if ((gic_r_read_8(gicv3, GICR_TYPER) & GICR_TYPER_PLPIS) == 0)
712 		return (ENXIO);
713 
714 	rpcpu = gicv3_get_redist(dev);
715 
716 	/* Do per-cpu LPI init once */
717 	if (!rpcpu->lpi_enabled) {
718 		its_init_cpu_lpi(dev, sc);
719 		rpcpu->lpi_enabled = true;
720 	}
721 
722 	if ((gic_its_read_8(sc, GITS_TYPER) & GITS_TYPER_PTA) != 0) {
723 		/* This ITS wants the redistributor physical address */
724 		target = vtophys((vm_offset_t)rman_get_virtual(rpcpu->res) +
725 		    rpcpu->offset);
726 	} else {
727 		/* This ITS wants the unique processor number */
728 		target = GICR_TYPER_CPUNUM(gic_r_read_8(gicv3, GICR_TYPER)) <<
729 		    CMD_TARGET_SHIFT;
730 	}
731 
732 	sc->sc_its_cols[cpuid]->col_target = target;
733 	sc->sc_its_cols[cpuid]->col_id = cpuid;
734 
735 	its_cmd_mapc(dev, sc->sc_its_cols[cpuid], 1);
736 	its_cmd_invall(dev, sc->sc_its_cols[cpuid]);
737 
738 	return (0);
739 }
740 
741 static int
742 gicv3_its_sysctl_trace_enable(SYSCTL_HANDLER_ARGS)
743 {
744 	struct gicv3_its_softc *sc;
745 	int rv;
746 
747 	sc = arg1;
748 
749 	rv = sysctl_handle_bool(oidp, &sc->trace_enable, 0, req);
750 	if (rv != 0 || req->newptr == NULL)
751 		return (rv);
752 	if (sc->trace_enable)
753 		gic_its_write_8(sc, GITS_TRKCTLR, 3);
754 	else
755 		gic_its_write_8(sc, GITS_TRKCTLR, 0);
756 
757 	return (0);
758 }
759 
760 static int
761 gicv3_its_sysctl_trace_regs(SYSCTL_HANDLER_ARGS)
762 {
763 	struct gicv3_its_softc *sc;
764 	struct sbuf *sb;
765 	int err;
766 
767 	sc = arg1;
768 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
769 	if (sb == NULL) {
770 		device_printf(sc->dev, "Could not allocate sbuf for output.\n");
771 		return (ENOMEM);
772 	}
773 	sbuf_cat(sb, "\n");
774 	sbuf_printf(sb, "GITS_TRKCTLR: 0x%08X\n",
775 	    gic_its_read_4(sc, GITS_TRKCTLR));
776 	sbuf_printf(sb, "GITS_TRKR:    0x%08X\n",
777 	    gic_its_read_4(sc, GITS_TRKR));
778 	sbuf_printf(sb, "GITS_TRKDIDR: 0x%08X\n",
779 	    gic_its_read_4(sc, GITS_TRKDIDR));
780 	sbuf_printf(sb, "GITS_TRKPIDR: 0x%08X\n",
781 	    gic_its_read_4(sc, GITS_TRKPIDR));
782 	sbuf_printf(sb, "GITS_TRKVIDR: 0x%08X\n",
783 	    gic_its_read_4(sc, GITS_TRKVIDR));
784 	sbuf_printf(sb, "GITS_TRKTGTR: 0x%08X\n",
785 	   gic_its_read_4(sc, GITS_TRKTGTR));
786 
787 	err = sbuf_finish(sb);
788 	if (err)
789 		device_printf(sc->dev, "Error finishing sbuf: %d\n", err);
790 	sbuf_delete(sb);
791 	return(err);
792 }
793 
794 static int
795 gicv3_its_init_sysctl(struct gicv3_its_softc *sc)
796 {
797 	struct sysctl_oid *oid, *child;
798 	struct sysctl_ctx_list *ctx_list;
799 
800 	ctx_list = device_get_sysctl_ctx(sc->dev);
801 	child = device_get_sysctl_tree(sc->dev);
802 	oid = SYSCTL_ADD_NODE(ctx_list,
803 	    SYSCTL_CHILDREN(child), OID_AUTO, "tracing",
804 	    CTLFLAG_RD| CTLFLAG_MPSAFE, NULL, "Messages tracing");
805 	if (oid == NULL)
806 		return (ENXIO);
807 
808 	/* Add registers */
809 	SYSCTL_ADD_PROC(ctx_list,
810 	    SYSCTL_CHILDREN(oid), OID_AUTO, "enable",
811 	    CTLTYPE_U8 | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
812 	    gicv3_its_sysctl_trace_enable, "CU", "Enable tracing");
813 	SYSCTL_ADD_PROC(ctx_list,
814 	    SYSCTL_CHILDREN(oid), OID_AUTO, "capture",
815 	    CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
816 	    gicv3_its_sysctl_trace_regs, "", "Captured tracing registers.");
817 
818 	return (0);
819 }
820 
821 static int
822 gicv3_its_attach(device_t dev)
823 {
824 	struct gicv3_its_softc *sc;
825 	int domain, err, i, rid;
826 	uint64_t phys;
827 	uint32_t ctlr, iidr;
828 
829 	sc = device_get_softc(dev);
830 
831 	sc->sc_irq_length = gicv3_get_nirqs(dev);
832 	sc->sc_irq_base = GIC_FIRST_LPI;
833 	sc->sc_irq_base += device_get_unit(dev) * sc->sc_irq_length;
834 
835 	rid = 0;
836 	sc->sc_its_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid,
837 	    RF_ACTIVE);
838 	if (sc->sc_its_res == NULL) {
839 		device_printf(dev, "Could not allocate memory\n");
840 		return (ENXIO);
841 	}
842 
843 	phys = rounddown2(vtophys(rman_get_virtual(sc->sc_its_res)) +
844 	    GITS_TRANSLATER, PAGE_SIZE);
845 	sc->ma = malloc(sizeof(struct vm_page), M_DEVBUF, M_WAITOK | M_ZERO);
846 	vm_page_initfake(sc->ma, phys, VM_MEMATTR_DEFAULT);
847 
848 	CPU_COPY(&all_cpus, &sc->sc_cpus);
849 	iidr = gic_its_read_4(sc, GITS_IIDR);
850 	for (i = 0; i < nitems(its_quirks); i++) {
851 		if ((iidr & its_quirks[i].iidr_mask) == its_quirks[i].iidr) {
852 			if (bootverbose) {
853 				device_printf(dev, "Applying %s\n",
854 				    its_quirks[i].desc);
855 			}
856 			its_quirks[i].func(dev);
857 			break;
858 		}
859 	}
860 
861 	if (bus_get_domain(dev, &domain) == 0 && domain < MAXMEMDOM) {
862 		sc->sc_ds = DOMAINSET_PREF(domain);
863 	} else {
864 		sc->sc_ds = DOMAINSET_RR();
865 	}
866 
867 	/*
868 	 * GIT_CTLR_EN is mandated to reset to 0 on a Warm reset, but we may be
869 	 * coming in via, for instance, a kexec/kboot style setup where a
870 	 * previous kernel has configured then relinquished control.  Clear it
871 	 * so that we can reconfigure GITS_BASER*.
872 	 */
873 	ctlr = gic_its_read_4(sc, GITS_CTLR);
874 	if ((ctlr & GITS_CTLR_EN) != 0) {
875 		ctlr &= ~GITS_CTLR_EN;
876 		gic_its_write_4(sc, GITS_CTLR, ctlr);
877 	}
878 
879 	/* Allocate the private tables */
880 	err = gicv3_its_table_init(dev, sc);
881 	if (err != 0)
882 		return (err);
883 
884 	/* Protects access to the device list */
885 	mtx_init(&sc->sc_its_dev_lock, "ITS device lock", NULL, MTX_SPIN);
886 
887 	/* Protects access to the ITS command circular buffer. */
888 	mtx_init(&sc->sc_its_cmd_lock, "ITS cmd lock", NULL, MTX_SPIN);
889 
890 	/* Allocate the command circular buffer */
891 	gicv3_its_cmdq_init(sc);
892 
893 	/* Allocate the per-CPU collections */
894 	for (int cpu = 0; cpu <= mp_maxid; cpu++)
895 		if (CPU_ISSET(cpu, &sc->sc_cpus) != 0)
896 			sc->sc_its_cols[cpu] = malloc_domainset(
897 			    sizeof(*sc->sc_its_cols[0]), M_GICV3_ITS,
898 			    DOMAINSET_PREF(pcpu_find(cpu)->pc_domain),
899 			    M_WAITOK | M_ZERO);
900 		else
901 			sc->sc_its_cols[cpu] = NULL;
902 
903 	/* Enable the ITS */
904 	gic_its_write_4(sc, GITS_CTLR, ctlr | GITS_CTLR_EN);
905 
906 	/* Create the LPI configuration table */
907 	gicv3_its_conftable_init(sc);
908 
909 	/* And the pending tebles */
910 	gicv3_its_pendtables_init(sc);
911 
912 	/* Enable LPIs on this CPU */
913 	its_init_cpu(dev, sc);
914 
915 	TAILQ_INIT(&sc->sc_its_dev_list);
916 	TAILQ_INIT(&sc->sc_free_irqs);
917 
918 	/*
919 	 * Create the vmem object to allocate INTRNG IRQs from. We try to
920 	 * use all IRQs not already used by the GICv3.
921 	 * XXX: This assumes there are no other interrupt controllers in the
922 	 * system.
923 	 */
924 	sc->sc_irq_alloc = vmem_create(device_get_nameunit(dev), 0,
925 	    gicv3_get_nirqs(dev), 1, 0, M_FIRSTFIT | M_WAITOK);
926 
927 	sc->sc_irqs = malloc(sizeof(*sc->sc_irqs) * sc->sc_irq_length,
928 	    M_GICV3_ITS, M_WAITOK | M_ZERO);
929 
930 	/* For GIC-500 install tracking sysctls. */
931 	if ((iidr & (GITS_IIDR_PRODUCT_MASK | GITS_IIDR_IMPLEMENTOR_MASK)) ==
932 	    GITS_IIDR_RAW(GITS_IIDR_IMPL_ARM, GITS_IIDR_PROD_GIC500, 0, 0))
933 		gicv3_its_init_sysctl(sc);
934 
935 	return (0);
936 }
937 
938 static int
939 gicv3_its_detach(device_t dev)
940 {
941 
942 	return (ENXIO);
943 }
944 
945 static void
946 its_quirk_cavium_22375(device_t dev)
947 {
948 	struct gicv3_its_softc *sc;
949 	int domain;
950 
951 	sc = device_get_softc(dev);
952 	sc->sc_its_flags |= ITS_FLAGS_ERRATA_CAVIUM_22375;
953 
954 	/*
955 	 * We need to limit which CPUs we send these interrupts to on
956 	 * the original dual socket ThunderX as it is unable to
957 	 * forward them between the two sockets.
958 	 */
959 	if (bus_get_domain(dev, &domain) == 0) {
960 		if (domain < MAXMEMDOM) {
961 			CPU_COPY(&cpuset_domain[domain], &sc->sc_cpus);
962 		} else {
963 			CPU_ZERO(&sc->sc_cpus);
964 		}
965 	}
966 }
967 
968 static void
969 gicv3_its_disable_intr(device_t dev, struct intr_irqsrc *isrc)
970 {
971 	struct gicv3_its_softc *sc;
972 	struct gicv3_its_irqsrc *girq;
973 	uint8_t *conf;
974 
975 	sc = device_get_softc(dev);
976 	girq = (struct gicv3_its_irqsrc *)isrc;
977 	conf = sc->sc_conf_base;
978 
979 	conf[girq->gi_lpi] &= ~LPI_CONF_ENABLE;
980 
981 	if ((sc->sc_its_flags & ITS_FLAGS_LPI_CONF_FLUSH) != 0) {
982 		/* Clean D-cache under command. */
983 		cpu_dcache_wb_range((vm_offset_t)&conf[girq->gi_lpi], 1);
984 	} else {
985 		/* DSB inner shareable, store */
986 		dsb(ishst);
987 	}
988 
989 	its_cmd_inv(dev, girq->gi_its_dev, girq);
990 }
991 
992 static void
993 gicv3_its_enable_intr(device_t dev, struct intr_irqsrc *isrc)
994 {
995 	struct gicv3_its_softc *sc;
996 	struct gicv3_its_irqsrc *girq;
997 	uint8_t *conf;
998 
999 	sc = device_get_softc(dev);
1000 	girq = (struct gicv3_its_irqsrc *)isrc;
1001 	conf = sc->sc_conf_base;
1002 
1003 	conf[girq->gi_lpi] |= LPI_CONF_ENABLE;
1004 
1005 	if ((sc->sc_its_flags & ITS_FLAGS_LPI_CONF_FLUSH) != 0) {
1006 		/* Clean D-cache under command. */
1007 		cpu_dcache_wb_range((vm_offset_t)&conf[girq->gi_lpi], 1);
1008 	} else {
1009 		/* DSB inner shareable, store */
1010 		dsb(ishst);
1011 	}
1012 
1013 	its_cmd_inv(dev, girq->gi_its_dev, girq);
1014 }
1015 
1016 static int
1017 gicv3_its_intr(void *arg, uintptr_t irq)
1018 {
1019 	struct gicv3_its_softc *sc = arg;
1020 	struct gicv3_its_irqsrc *girq;
1021 	struct trapframe *tf;
1022 
1023 	irq -= sc->sc_irq_base;
1024 	girq = sc->sc_irqs[irq];
1025 	if (girq == NULL)
1026 		panic("gicv3_its_intr: Invalid interrupt %ld",
1027 		    irq + sc->sc_irq_base);
1028 
1029 	tf = curthread->td_intr_frame;
1030 	intr_isrc_dispatch(&girq->gi_isrc, tf);
1031 	return (FILTER_HANDLED);
1032 }
1033 
1034 static void
1035 gicv3_its_pre_ithread(device_t dev, struct intr_irqsrc *isrc)
1036 {
1037 	struct gicv3_its_irqsrc *girq;
1038 
1039 	girq = (struct gicv3_its_irqsrc *)isrc;
1040 	gic_icc_write(EOIR1, girq->gi_lpi + GIC_FIRST_LPI);
1041 }
1042 
1043 static void
1044 gicv3_its_post_ithread(device_t dev, struct intr_irqsrc *isrc)
1045 {
1046 
1047 }
1048 
1049 static void
1050 gicv3_its_post_filter(device_t dev, struct intr_irqsrc *isrc)
1051 {
1052 	struct gicv3_its_irqsrc *girq;
1053 
1054 	girq = (struct gicv3_its_irqsrc *)isrc;
1055 	gic_icc_write(EOIR1, girq->gi_lpi + GIC_FIRST_LPI);
1056 }
1057 
1058 static int
1059 gicv3_its_select_cpu(device_t dev, struct intr_irqsrc *isrc)
1060 {
1061 	struct gicv3_its_softc *sc;
1062 
1063 	sc = device_get_softc(dev);
1064 	if (CPU_EMPTY(&isrc->isrc_cpu)) {
1065 		sc->gic_irq_cpu = intr_irq_next_cpu(sc->gic_irq_cpu,
1066 		    &sc->sc_cpus);
1067 		CPU_SETOF(sc->gic_irq_cpu, &isrc->isrc_cpu);
1068 	}
1069 
1070 	return (0);
1071 }
1072 
1073 static int
1074 gicv3_its_bind_intr(device_t dev, struct intr_irqsrc *isrc)
1075 {
1076 	struct gicv3_its_irqsrc *girq;
1077 
1078 	gicv3_its_select_cpu(dev, isrc);
1079 
1080 	girq = (struct gicv3_its_irqsrc *)isrc;
1081 	its_cmd_movi(dev, girq);
1082 	return (0);
1083 }
1084 
1085 static int
1086 gicv3_its_map_intr(device_t dev, struct intr_map_data *data,
1087     struct intr_irqsrc **isrcp)
1088 {
1089 
1090 	/*
1091 	 * This should never happen, we only call this function to map
1092 	 * interrupts found before the controller driver is ready.
1093 	 */
1094 	panic("gicv3_its_map_intr: Unable to map a MSI interrupt");
1095 }
1096 
1097 static int
1098 gicv3_its_setup_intr(device_t dev, struct intr_irqsrc *isrc,
1099     struct resource *res, struct intr_map_data *data)
1100 {
1101 
1102 	/* Bind the interrupt to a CPU */
1103 	gicv3_its_bind_intr(dev, isrc);
1104 
1105 	return (0);
1106 }
1107 
1108 #ifdef SMP
1109 static void
1110 gicv3_its_init_secondary(device_t dev)
1111 {
1112 	struct gicv3_its_softc *sc;
1113 
1114 	sc = device_get_softc(dev);
1115 
1116 	/*
1117 	 * This is fatal as otherwise we may bind interrupts to this CPU.
1118 	 * We need a way to tell the interrupt framework to only bind to a
1119 	 * subset of given CPUs when it performs the shuffle.
1120 	 */
1121 	if (its_init_cpu(dev, sc) != 0)
1122 		panic("gicv3_its_init_secondary: No usable ITS on CPU%d",
1123 		    PCPU_GET(cpuid));
1124 }
1125 #endif
1126 
1127 static uint32_t
1128 its_get_devid(device_t pci_dev)
1129 {
1130 	uintptr_t id;
1131 
1132 	if (pci_get_id(pci_dev, PCI_ID_MSI, &id) != 0)
1133 		panic("%s: %s: Unable to get the MSI DeviceID", __func__,
1134 		    device_get_nameunit(pci_dev));
1135 
1136 	return (id);
1137 }
1138 
1139 static struct its_dev *
1140 its_device_find(device_t dev, device_t child)
1141 {
1142 	struct gicv3_its_softc *sc;
1143 	struct its_dev *its_dev = NULL;
1144 
1145 	sc = device_get_softc(dev);
1146 
1147 	mtx_lock_spin(&sc->sc_its_dev_lock);
1148 	TAILQ_FOREACH(its_dev, &sc->sc_its_dev_list, entry) {
1149 		if (its_dev->pci_dev == child)
1150 			break;
1151 	}
1152 	mtx_unlock_spin(&sc->sc_its_dev_lock);
1153 
1154 	return (its_dev);
1155 }
1156 
1157 static struct its_dev *
1158 its_device_get(device_t dev, device_t child, u_int nvecs)
1159 {
1160 	struct gicv3_its_softc *sc;
1161 	struct its_dev *its_dev;
1162 	vmem_addr_t irq_base;
1163 	size_t esize;
1164 
1165 	sc = device_get_softc(dev);
1166 
1167 	its_dev = its_device_find(dev, child);
1168 	if (its_dev != NULL)
1169 		return (its_dev);
1170 
1171 	its_dev = malloc(sizeof(*its_dev), M_GICV3_ITS, M_NOWAIT | M_ZERO);
1172 	if (its_dev == NULL)
1173 		return (NULL);
1174 
1175 	its_dev->pci_dev = child;
1176 	its_dev->devid = its_get_devid(child);
1177 
1178 	its_dev->lpis.lpi_busy = 0;
1179 	its_dev->lpis.lpi_num = nvecs;
1180 	its_dev->lpis.lpi_free = nvecs;
1181 
1182 	if (vmem_alloc(sc->sc_irq_alloc, nvecs, M_FIRSTFIT | M_NOWAIT,
1183 	    &irq_base) != 0) {
1184 		free(its_dev, M_GICV3_ITS);
1185 		return (NULL);
1186 	}
1187 	its_dev->lpis.lpi_base = irq_base;
1188 
1189 	/* Get ITT entry size */
1190 	esize = GITS_TYPER_ITTES(gic_its_read_8(sc, GITS_TYPER));
1191 
1192 	/*
1193 	 * Allocate ITT for this device.
1194 	 * PA has to be 256 B aligned. At least two entries for device.
1195 	 */
1196 	its_dev->itt_size = roundup2(MAX(nvecs, 2) * esize, 256);
1197 	its_dev->itt = (vm_offset_t)contigmalloc_domainset(its_dev->itt_size,
1198 	    M_GICV3_ITS, sc->sc_ds, M_NOWAIT | M_ZERO, 0,
1199 	    LPI_INT_TRANS_TAB_MAX_ADDR, LPI_INT_TRANS_TAB_ALIGN, 0);
1200 	if (its_dev->itt == 0) {
1201 		vmem_free(sc->sc_irq_alloc, its_dev->lpis.lpi_base, nvecs);
1202 		free(its_dev, M_GICV3_ITS);
1203 		return (NULL);
1204 	}
1205 
1206 	/* Make sure device sees zeroed ITT. */
1207 	if ((sc->sc_its_flags & ITS_FLAGS_CMDQ_FLUSH) != 0)
1208 		cpu_dcache_wb_range(its_dev->itt, its_dev->itt_size);
1209 
1210 	mtx_lock_spin(&sc->sc_its_dev_lock);
1211 	TAILQ_INSERT_TAIL(&sc->sc_its_dev_list, its_dev, entry);
1212 	mtx_unlock_spin(&sc->sc_its_dev_lock);
1213 
1214 	/* Map device to its ITT */
1215 	its_cmd_mapd(dev, its_dev, 1);
1216 
1217 	return (its_dev);
1218 }
1219 
1220 static void
1221 its_device_release(device_t dev, struct its_dev *its_dev)
1222 {
1223 	struct gicv3_its_softc *sc;
1224 
1225 	KASSERT(its_dev->lpis.lpi_busy == 0,
1226 	    ("its_device_release: Trying to release an inuse ITS device"));
1227 
1228 	/* Unmap device in ITS */
1229 	its_cmd_mapd(dev, its_dev, 0);
1230 
1231 	sc = device_get_softc(dev);
1232 
1233 	/* Remove the device from the list of devices */
1234 	mtx_lock_spin(&sc->sc_its_dev_lock);
1235 	TAILQ_REMOVE(&sc->sc_its_dev_list, its_dev, entry);
1236 	mtx_unlock_spin(&sc->sc_its_dev_lock);
1237 
1238 	/* Free ITT */
1239 	KASSERT(its_dev->itt != 0, ("Invalid ITT in valid ITS device"));
1240 	contigfree((void *)its_dev->itt, its_dev->itt_size, M_GICV3_ITS);
1241 
1242 	/* Free the IRQ allocation */
1243 	vmem_free(sc->sc_irq_alloc, its_dev->lpis.lpi_base,
1244 	    its_dev->lpis.lpi_num);
1245 
1246 	free(its_dev, M_GICV3_ITS);
1247 }
1248 
1249 static struct gicv3_its_irqsrc *
1250 gicv3_its_alloc_irqsrc(device_t dev, struct gicv3_its_softc *sc, u_int irq)
1251 {
1252 	struct gicv3_its_irqsrc *girq = NULL;
1253 
1254 	KASSERT(sc->sc_irqs[irq] == NULL,
1255 	    ("%s: Interrupt %u already allocated", __func__, irq));
1256 	mtx_lock_spin(&sc->sc_its_dev_lock);
1257 	if (!TAILQ_EMPTY(&sc->sc_free_irqs)) {
1258 		girq = TAILQ_FIRST(&sc->sc_free_irqs);
1259 		TAILQ_REMOVE(&sc->sc_free_irqs, girq, gi_link);
1260 	}
1261 	mtx_unlock_spin(&sc->sc_its_dev_lock);
1262 	if (girq == NULL) {
1263 		girq = malloc(sizeof(*girq), M_GICV3_ITS,
1264 		    M_NOWAIT | M_ZERO);
1265 		if (girq == NULL)
1266 			return (NULL);
1267 		girq->gi_id = -1;
1268 		if (intr_isrc_register(&girq->gi_isrc, dev, 0,
1269 		    "%s,%u", device_get_nameunit(dev), irq) != 0) {
1270 			free(girq, M_GICV3_ITS);
1271 			return (NULL);
1272 		}
1273 	}
1274 	girq->gi_lpi = irq + sc->sc_irq_base - GIC_FIRST_LPI;
1275 	sc->sc_irqs[irq] = girq;
1276 
1277 	return (girq);
1278 }
1279 
1280 static void
1281 gicv3_its_release_irqsrc(struct gicv3_its_softc *sc,
1282     struct gicv3_its_irqsrc *girq)
1283 {
1284 	u_int irq;
1285 
1286 	mtx_assert(&sc->sc_its_dev_lock, MA_OWNED);
1287 
1288 	irq = girq->gi_lpi + GIC_FIRST_LPI - sc->sc_irq_base;
1289 	sc->sc_irqs[irq] = NULL;
1290 
1291 	girq->gi_id = -1;
1292 	girq->gi_its_dev = NULL;
1293 	TAILQ_INSERT_TAIL(&sc->sc_free_irqs, girq, gi_link);
1294 }
1295 
1296 static int
1297 gicv3_its_alloc_msi(device_t dev, device_t child, int count, int maxcount,
1298     device_t *pic, struct intr_irqsrc **srcs)
1299 {
1300 	struct gicv3_its_softc *sc;
1301 	struct gicv3_its_irqsrc *girq;
1302 	struct its_dev *its_dev;
1303 	u_int irq;
1304 	int i;
1305 
1306 	its_dev = its_device_get(dev, child, count);
1307 	if (its_dev == NULL)
1308 		return (ENXIO);
1309 
1310 	KASSERT(its_dev->lpis.lpi_free >= count,
1311 	    ("gicv3_its_alloc_msi: No free LPIs"));
1312 	sc = device_get_softc(dev);
1313 	irq = its_dev->lpis.lpi_base + its_dev->lpis.lpi_num -
1314 	    its_dev->lpis.lpi_free;
1315 
1316 	/* Allocate the irqsrc for each MSI */
1317 	for (i = 0; i < count; i++, irq++) {
1318 		its_dev->lpis.lpi_free--;
1319 		srcs[i] = (struct intr_irqsrc *)gicv3_its_alloc_irqsrc(dev,
1320 		    sc, irq);
1321 		if (srcs[i] == NULL)
1322 			break;
1323 	}
1324 
1325 	/* The allocation failed, release them */
1326 	if (i != count) {
1327 		mtx_lock_spin(&sc->sc_its_dev_lock);
1328 		for (i = 0; i < count; i++) {
1329 			girq = (struct gicv3_its_irqsrc *)srcs[i];
1330 			if (girq == NULL)
1331 				break;
1332 			gicv3_its_release_irqsrc(sc, girq);
1333 			srcs[i] = NULL;
1334 		}
1335 		mtx_unlock_spin(&sc->sc_its_dev_lock);
1336 		return (ENXIO);
1337 	}
1338 
1339 	/* Finish the allocation now we have all MSI irqsrcs */
1340 	for (i = 0; i < count; i++) {
1341 		girq = (struct gicv3_its_irqsrc *)srcs[i];
1342 		girq->gi_id = i;
1343 		girq->gi_its_dev = its_dev;
1344 
1345 		/* Map the message to the given IRQ */
1346 		gicv3_its_select_cpu(dev, (struct intr_irqsrc *)girq);
1347 		its_cmd_mapti(dev, girq);
1348 	}
1349 	its_dev->lpis.lpi_busy += count;
1350 	*pic = dev;
1351 
1352 	return (0);
1353 }
1354 
1355 static int
1356 gicv3_its_release_msi(device_t dev, device_t child, int count,
1357     struct intr_irqsrc **isrc)
1358 {
1359 	struct gicv3_its_softc *sc;
1360 	struct gicv3_its_irqsrc *girq;
1361 	struct its_dev *its_dev;
1362 	int i;
1363 
1364 	its_dev = its_device_find(dev, child);
1365 
1366 	KASSERT(its_dev != NULL,
1367 	    ("gicv3_its_release_msi: Releasing a MSI interrupt with "
1368 	     "no ITS device"));
1369 	KASSERT(its_dev->lpis.lpi_busy >= count,
1370 	    ("gicv3_its_release_msi: Releasing more interrupts than "
1371 	     "were allocated: releasing %d, allocated %d", count,
1372 	     its_dev->lpis.lpi_busy));
1373 
1374 	sc = device_get_softc(dev);
1375 	mtx_lock_spin(&sc->sc_its_dev_lock);
1376 	for (i = 0; i < count; i++) {
1377 		girq = (struct gicv3_its_irqsrc *)isrc[i];
1378 		gicv3_its_release_irqsrc(sc, girq);
1379 	}
1380 	mtx_unlock_spin(&sc->sc_its_dev_lock);
1381 	its_dev->lpis.lpi_busy -= count;
1382 
1383 	if (its_dev->lpis.lpi_busy == 0)
1384 		its_device_release(dev, its_dev);
1385 
1386 	return (0);
1387 }
1388 
1389 static int
1390 gicv3_its_alloc_msix(device_t dev, device_t child, device_t *pic,
1391     struct intr_irqsrc **isrcp)
1392 {
1393 	struct gicv3_its_softc *sc;
1394 	struct gicv3_its_irqsrc *girq;
1395 	struct its_dev *its_dev;
1396 	u_int nvecs, irq;
1397 
1398 	nvecs = pci_msix_count(child);
1399 	its_dev = its_device_get(dev, child, nvecs);
1400 	if (its_dev == NULL)
1401 		return (ENXIO);
1402 
1403 	KASSERT(its_dev->lpis.lpi_free > 0,
1404 	    ("gicv3_its_alloc_msix: No free LPIs"));
1405 	sc = device_get_softc(dev);
1406 	irq = its_dev->lpis.lpi_base + its_dev->lpis.lpi_num -
1407 	    its_dev->lpis.lpi_free;
1408 
1409 	girq = gicv3_its_alloc_irqsrc(dev, sc, irq);
1410 	if (girq == NULL)
1411 		return (ENXIO);
1412 	girq->gi_id = its_dev->lpis.lpi_busy;
1413 	girq->gi_its_dev = its_dev;
1414 
1415 	its_dev->lpis.lpi_free--;
1416 	its_dev->lpis.lpi_busy++;
1417 
1418 	/* Map the message to the given IRQ */
1419 	gicv3_its_select_cpu(dev, (struct intr_irqsrc *)girq);
1420 	its_cmd_mapti(dev, girq);
1421 
1422 	*pic = dev;
1423 	*isrcp = (struct intr_irqsrc *)girq;
1424 
1425 	return (0);
1426 }
1427 
1428 static int
1429 gicv3_its_release_msix(device_t dev, device_t child, struct intr_irqsrc *isrc)
1430 {
1431 	struct gicv3_its_softc *sc;
1432 	struct gicv3_its_irqsrc *girq;
1433 	struct its_dev *its_dev;
1434 
1435 	its_dev = its_device_find(dev, child);
1436 
1437 	KASSERT(its_dev != NULL,
1438 	    ("gicv3_its_release_msix: Releasing a MSI-X interrupt with "
1439 	     "no ITS device"));
1440 	KASSERT(its_dev->lpis.lpi_busy > 0,
1441 	    ("gicv3_its_release_msix: Releasing more interrupts than "
1442 	     "were allocated: allocated %d", its_dev->lpis.lpi_busy));
1443 
1444 	sc = device_get_softc(dev);
1445 	girq = (struct gicv3_its_irqsrc *)isrc;
1446 	mtx_lock_spin(&sc->sc_its_dev_lock);
1447 	gicv3_its_release_irqsrc(sc, girq);
1448 	mtx_unlock_spin(&sc->sc_its_dev_lock);
1449 	its_dev->lpis.lpi_busy--;
1450 
1451 	if (its_dev->lpis.lpi_busy == 0)
1452 		its_device_release(dev, its_dev);
1453 
1454 	return (0);
1455 }
1456 
1457 static int
1458 gicv3_its_map_msi(device_t dev, device_t child, struct intr_irqsrc *isrc,
1459     uint64_t *addr, uint32_t *data)
1460 {
1461 	struct gicv3_its_softc *sc;
1462 	struct gicv3_its_irqsrc *girq;
1463 
1464 	sc = device_get_softc(dev);
1465 	girq = (struct gicv3_its_irqsrc *)isrc;
1466 
1467 	*addr = vtophys(rman_get_virtual(sc->sc_its_res)) + GITS_TRANSLATER;
1468 	*data = girq->gi_id;
1469 
1470 	return (0);
1471 }
1472 
1473 #ifdef IOMMU
1474 static int
1475 gicv3_iommu_init(device_t dev, device_t child, struct iommu_domain **domain)
1476 {
1477 	struct gicv3_its_softc *sc;
1478 	struct iommu_ctx *ctx;
1479 	int error;
1480 
1481 	sc = device_get_softc(dev);
1482 	ctx = iommu_get_dev_ctx(child);
1483 	if (ctx == NULL)
1484 		return (ENXIO);
1485 	/* Map the page containing the GITS_TRANSLATER register. */
1486 	error = iommu_map_msi(ctx, PAGE_SIZE, 0,
1487 	    IOMMU_MAP_ENTRY_WRITE, IOMMU_MF_CANWAIT, &sc->ma);
1488 	*domain = iommu_get_ctx_domain(ctx);
1489 
1490 	return (error);
1491 }
1492 
1493 static void
1494 gicv3_iommu_deinit(device_t dev, device_t child)
1495 {
1496 	struct iommu_ctx *ctx;
1497 
1498 	ctx = iommu_get_dev_ctx(child);
1499 	if (ctx == NULL)
1500 		return;
1501 
1502 	iommu_unmap_msi(ctx);
1503 }
1504 #endif
1505 
1506 /*
1507  * Commands handling.
1508  */
1509 
1510 static __inline void
1511 cmd_format_command(struct its_cmd *cmd, uint8_t cmd_type)
1512 {
1513 	/* Command field: DW0 [7:0] */
1514 	cmd->cmd_dword[0] &= htole64(~CMD_COMMAND_MASK);
1515 	cmd->cmd_dword[0] |= htole64(cmd_type);
1516 }
1517 
1518 static __inline void
1519 cmd_format_devid(struct its_cmd *cmd, uint32_t devid)
1520 {
1521 	/* Device ID field: DW0 [63:32] */
1522 	cmd->cmd_dword[0] &= htole64(~CMD_DEVID_MASK);
1523 	cmd->cmd_dword[0] |= htole64((uint64_t)devid << CMD_DEVID_SHIFT);
1524 }
1525 
1526 static __inline void
1527 cmd_format_size(struct its_cmd *cmd, uint16_t size)
1528 {
1529 	/* Size field: DW1 [4:0] */
1530 	cmd->cmd_dword[1] &= htole64(~CMD_SIZE_MASK);
1531 	cmd->cmd_dword[1] |= htole64((size & CMD_SIZE_MASK));
1532 }
1533 
1534 static __inline void
1535 cmd_format_id(struct its_cmd *cmd, uint32_t id)
1536 {
1537 	/* ID field: DW1 [31:0] */
1538 	cmd->cmd_dword[1] &= htole64(~CMD_ID_MASK);
1539 	cmd->cmd_dword[1] |= htole64(id);
1540 }
1541 
1542 static __inline void
1543 cmd_format_pid(struct its_cmd *cmd, uint32_t pid)
1544 {
1545 	/* Physical ID field: DW1 [63:32] */
1546 	cmd->cmd_dword[1] &= htole64(~CMD_PID_MASK);
1547 	cmd->cmd_dword[1] |= htole64((uint64_t)pid << CMD_PID_SHIFT);
1548 }
1549 
1550 static __inline void
1551 cmd_format_col(struct its_cmd *cmd, uint16_t col_id)
1552 {
1553 	/* Collection field: DW2 [16:0] */
1554 	cmd->cmd_dword[2] &= htole64(~CMD_COL_MASK);
1555 	cmd->cmd_dword[2] |= htole64(col_id);
1556 }
1557 
1558 static __inline void
1559 cmd_format_target(struct its_cmd *cmd, uint64_t target)
1560 {
1561 	/* Target Address field: DW2 [47:16] */
1562 	cmd->cmd_dword[2] &= htole64(~CMD_TARGET_MASK);
1563 	cmd->cmd_dword[2] |= htole64(target & CMD_TARGET_MASK);
1564 }
1565 
1566 static __inline void
1567 cmd_format_itt(struct its_cmd *cmd, uint64_t itt)
1568 {
1569 	/* ITT Address field: DW2 [47:8] */
1570 	cmd->cmd_dword[2] &= htole64(~CMD_ITT_MASK);
1571 	cmd->cmd_dword[2] |= htole64(itt & CMD_ITT_MASK);
1572 }
1573 
1574 static __inline void
1575 cmd_format_valid(struct its_cmd *cmd, uint8_t valid)
1576 {
1577 	/* Valid field: DW2 [63] */
1578 	cmd->cmd_dword[2] &= htole64(~CMD_VALID_MASK);
1579 	cmd->cmd_dword[2] |= htole64((uint64_t)valid << CMD_VALID_SHIFT);
1580 }
1581 
1582 static inline bool
1583 its_cmd_queue_full(struct gicv3_its_softc *sc)
1584 {
1585 	size_t read_idx, next_write_idx;
1586 
1587 	/* Get the index of the next command */
1588 	next_write_idx = (sc->sc_its_cmd_next_idx + 1) %
1589 	    (ITS_CMDQ_SIZE / sizeof(struct its_cmd));
1590 	/* And the index of the current command being read */
1591 	read_idx = gic_its_read_4(sc, GITS_CREADR) / sizeof(struct its_cmd);
1592 
1593 	/*
1594 	 * The queue is full when the write offset points
1595 	 * at the command before the current read offset.
1596 	 */
1597 	return (next_write_idx == read_idx);
1598 }
1599 
1600 static inline void
1601 its_cmd_sync(struct gicv3_its_softc *sc, struct its_cmd *cmd)
1602 {
1603 
1604 	if ((sc->sc_its_flags & ITS_FLAGS_CMDQ_FLUSH) != 0) {
1605 		/* Clean D-cache under command. */
1606 		cpu_dcache_wb_range((vm_offset_t)cmd, sizeof(*cmd));
1607 	} else {
1608 		/* DSB inner shareable, store */
1609 		dsb(ishst);
1610 	}
1611 
1612 }
1613 
1614 static inline uint64_t
1615 its_cmd_cwriter_offset(struct gicv3_its_softc *sc, struct its_cmd *cmd)
1616 {
1617 	uint64_t off;
1618 
1619 	off = (cmd - sc->sc_its_cmd_base) * sizeof(*cmd);
1620 
1621 	return (off);
1622 }
1623 
1624 static void
1625 its_cmd_wait_completion(device_t dev, struct its_cmd *cmd_first,
1626     struct its_cmd *cmd_last)
1627 {
1628 	struct gicv3_its_softc *sc;
1629 	uint64_t first, last, read;
1630 	size_t us_left;
1631 
1632 	sc = device_get_softc(dev);
1633 
1634 	/*
1635 	 * XXX ARM64TODO: This is obviously a significant delay.
1636 	 * The reason for that is that currently the time frames for
1637 	 * the command to complete are not known.
1638 	 */
1639 	us_left = 1000000;
1640 
1641 	first = its_cmd_cwriter_offset(sc, cmd_first);
1642 	last = its_cmd_cwriter_offset(sc, cmd_last);
1643 
1644 	for (;;) {
1645 		read = gic_its_read_8(sc, GITS_CREADR);
1646 		if (first < last) {
1647 			if (read < first || read >= last)
1648 				break;
1649 		} else if (read < first && read >= last)
1650 			break;
1651 
1652 		if (us_left-- == 0) {
1653 			/* This means timeout */
1654 			device_printf(dev,
1655 			    "Timeout while waiting for CMD completion.\n");
1656 			return;
1657 		}
1658 		DELAY(1);
1659 	}
1660 }
1661 
1662 static struct its_cmd *
1663 its_cmd_alloc_locked(device_t dev)
1664 {
1665 	struct gicv3_its_softc *sc;
1666 	struct its_cmd *cmd;
1667 	size_t us_left;
1668 
1669 	sc = device_get_softc(dev);
1670 
1671 	/*
1672 	 * XXX ARM64TODO: This is obviously a significant delay.
1673 	 * The reason for that is that currently the time frames for
1674 	 * the command to complete (and therefore free the descriptor)
1675 	 * are not known.
1676 	 */
1677 	us_left = 1000000;
1678 
1679 	mtx_assert(&sc->sc_its_cmd_lock, MA_OWNED);
1680 	while (its_cmd_queue_full(sc)) {
1681 		if (us_left-- == 0) {
1682 			/* Timeout while waiting for free command */
1683 			device_printf(dev,
1684 			    "Timeout while waiting for free command\n");
1685 			return (NULL);
1686 		}
1687 		DELAY(1);
1688 	}
1689 
1690 	cmd = &sc->sc_its_cmd_base[sc->sc_its_cmd_next_idx];
1691 	sc->sc_its_cmd_next_idx++;
1692 	sc->sc_its_cmd_next_idx %= ITS_CMDQ_SIZE / sizeof(struct its_cmd);
1693 
1694 	return (cmd);
1695 }
1696 
1697 static uint64_t
1698 its_cmd_prepare(struct its_cmd *cmd, struct its_cmd_desc *desc)
1699 {
1700 	uint64_t target;
1701 	uint8_t cmd_type;
1702 	u_int size;
1703 
1704 	cmd_type = desc->cmd_type;
1705 	target = ITS_TARGET_NONE;
1706 
1707 	switch (cmd_type) {
1708 	case ITS_CMD_MOVI:	/* Move interrupt ID to another collection */
1709 		target = desc->cmd_desc_movi.col->col_target;
1710 		cmd_format_command(cmd, ITS_CMD_MOVI);
1711 		cmd_format_id(cmd, desc->cmd_desc_movi.id);
1712 		cmd_format_col(cmd, desc->cmd_desc_movi.col->col_id);
1713 		cmd_format_devid(cmd, desc->cmd_desc_movi.its_dev->devid);
1714 		break;
1715 	case ITS_CMD_SYNC:	/* Wait for previous commands completion */
1716 		target = desc->cmd_desc_sync.col->col_target;
1717 		cmd_format_command(cmd, ITS_CMD_SYNC);
1718 		cmd_format_target(cmd, target);
1719 		break;
1720 	case ITS_CMD_MAPD:	/* Assign ITT to device */
1721 		cmd_format_command(cmd, ITS_CMD_MAPD);
1722 		cmd_format_itt(cmd, vtophys(desc->cmd_desc_mapd.its_dev->itt));
1723 		/*
1724 		 * Size describes number of bits to encode interrupt IDs
1725 		 * supported by the device minus one.
1726 		 * When V (valid) bit is zero, this field should be written
1727 		 * as zero.
1728 		 */
1729 		if (desc->cmd_desc_mapd.valid != 0) {
1730 			size = fls(desc->cmd_desc_mapd.its_dev->lpis.lpi_num);
1731 			size = MAX(1, size) - 1;
1732 		} else
1733 			size = 0;
1734 
1735 		cmd_format_size(cmd, size);
1736 		cmd_format_devid(cmd, desc->cmd_desc_mapd.its_dev->devid);
1737 		cmd_format_valid(cmd, desc->cmd_desc_mapd.valid);
1738 		break;
1739 	case ITS_CMD_MAPC:	/* Map collection to Re-Distributor */
1740 		target = desc->cmd_desc_mapc.col->col_target;
1741 		cmd_format_command(cmd, ITS_CMD_MAPC);
1742 		cmd_format_col(cmd, desc->cmd_desc_mapc.col->col_id);
1743 		cmd_format_valid(cmd, desc->cmd_desc_mapc.valid);
1744 		cmd_format_target(cmd, target);
1745 		break;
1746 	case ITS_CMD_MAPTI:
1747 		target = desc->cmd_desc_mapvi.col->col_target;
1748 		cmd_format_command(cmd, ITS_CMD_MAPTI);
1749 		cmd_format_devid(cmd, desc->cmd_desc_mapvi.its_dev->devid);
1750 		cmd_format_id(cmd, desc->cmd_desc_mapvi.id);
1751 		cmd_format_pid(cmd, desc->cmd_desc_mapvi.pid);
1752 		cmd_format_col(cmd, desc->cmd_desc_mapvi.col->col_id);
1753 		break;
1754 	case ITS_CMD_MAPI:
1755 		target = desc->cmd_desc_mapi.col->col_target;
1756 		cmd_format_command(cmd, ITS_CMD_MAPI);
1757 		cmd_format_devid(cmd, desc->cmd_desc_mapi.its_dev->devid);
1758 		cmd_format_id(cmd, desc->cmd_desc_mapi.pid);
1759 		cmd_format_col(cmd, desc->cmd_desc_mapi.col->col_id);
1760 		break;
1761 	case ITS_CMD_INV:
1762 		target = desc->cmd_desc_inv.col->col_target;
1763 		cmd_format_command(cmd, ITS_CMD_INV);
1764 		cmd_format_devid(cmd, desc->cmd_desc_inv.its_dev->devid);
1765 		cmd_format_id(cmd, desc->cmd_desc_inv.pid);
1766 		break;
1767 	case ITS_CMD_INVALL:
1768 		cmd_format_command(cmd, ITS_CMD_INVALL);
1769 		cmd_format_col(cmd, desc->cmd_desc_invall.col->col_id);
1770 		break;
1771 	default:
1772 		panic("its_cmd_prepare: Invalid command: %x", cmd_type);
1773 	}
1774 
1775 	return (target);
1776 }
1777 
1778 static int
1779 its_cmd_send(device_t dev, struct its_cmd_desc *desc)
1780 {
1781 	struct gicv3_its_softc *sc;
1782 	struct its_cmd *cmd, *cmd_sync, *cmd_write;
1783 	struct its_col col_sync;
1784 	struct its_cmd_desc desc_sync;
1785 	uint64_t target, cwriter;
1786 
1787 	sc = device_get_softc(dev);
1788 	mtx_lock_spin(&sc->sc_its_cmd_lock);
1789 	cmd = its_cmd_alloc_locked(dev);
1790 	if (cmd == NULL) {
1791 		device_printf(dev, "could not allocate ITS command\n");
1792 		mtx_unlock_spin(&sc->sc_its_cmd_lock);
1793 		return (EBUSY);
1794 	}
1795 
1796 	target = its_cmd_prepare(cmd, desc);
1797 	its_cmd_sync(sc, cmd);
1798 
1799 	if (target != ITS_TARGET_NONE) {
1800 		cmd_sync = its_cmd_alloc_locked(dev);
1801 		if (cmd_sync != NULL) {
1802 			desc_sync.cmd_type = ITS_CMD_SYNC;
1803 			col_sync.col_target = target;
1804 			desc_sync.cmd_desc_sync.col = &col_sync;
1805 			its_cmd_prepare(cmd_sync, &desc_sync);
1806 			its_cmd_sync(sc, cmd_sync);
1807 		}
1808 	}
1809 
1810 	/* Update GITS_CWRITER */
1811 	cwriter = sc->sc_its_cmd_next_idx * sizeof(struct its_cmd);
1812 	gic_its_write_8(sc, GITS_CWRITER, cwriter);
1813 	cmd_write = &sc->sc_its_cmd_base[sc->sc_its_cmd_next_idx];
1814 	mtx_unlock_spin(&sc->sc_its_cmd_lock);
1815 
1816 	its_cmd_wait_completion(dev, cmd, cmd_write);
1817 
1818 	return (0);
1819 }
1820 
1821 /* Handlers to send commands */
1822 static void
1823 its_cmd_movi(device_t dev, struct gicv3_its_irqsrc *girq)
1824 {
1825 	struct gicv3_its_softc *sc;
1826 	struct its_cmd_desc desc;
1827 	struct its_col *col;
1828 
1829 	sc = device_get_softc(dev);
1830 	col = sc->sc_its_cols[CPU_FFS(&girq->gi_isrc.isrc_cpu) - 1];
1831 
1832 	desc.cmd_type = ITS_CMD_MOVI;
1833 	desc.cmd_desc_movi.its_dev = girq->gi_its_dev;
1834 	desc.cmd_desc_movi.col = col;
1835 	desc.cmd_desc_movi.id = girq->gi_id;
1836 
1837 	its_cmd_send(dev, &desc);
1838 }
1839 
1840 static void
1841 its_cmd_mapc(device_t dev, struct its_col *col, uint8_t valid)
1842 {
1843 	struct its_cmd_desc desc;
1844 
1845 	desc.cmd_type = ITS_CMD_MAPC;
1846 	desc.cmd_desc_mapc.col = col;
1847 	/*
1848 	 * Valid bit set - map the collection.
1849 	 * Valid bit cleared - unmap the collection.
1850 	 */
1851 	desc.cmd_desc_mapc.valid = valid;
1852 
1853 	its_cmd_send(dev, &desc);
1854 }
1855 
1856 static void
1857 its_cmd_mapti(device_t dev, struct gicv3_its_irqsrc *girq)
1858 {
1859 	struct gicv3_its_softc *sc;
1860 	struct its_cmd_desc desc;
1861 	struct its_col *col;
1862 	u_int col_id;
1863 
1864 	sc = device_get_softc(dev);
1865 
1866 	col_id = CPU_FFS(&girq->gi_isrc.isrc_cpu) - 1;
1867 	col = sc->sc_its_cols[col_id];
1868 
1869 	desc.cmd_type = ITS_CMD_MAPTI;
1870 	desc.cmd_desc_mapvi.its_dev = girq->gi_its_dev;
1871 	desc.cmd_desc_mapvi.col = col;
1872 	/* The EventID sent to the device */
1873 	desc.cmd_desc_mapvi.id = girq->gi_id;
1874 	/* The physical interrupt presented to softeware */
1875 	desc.cmd_desc_mapvi.pid = girq->gi_lpi + GIC_FIRST_LPI;
1876 
1877 	its_cmd_send(dev, &desc);
1878 }
1879 
1880 static void
1881 its_cmd_mapd(device_t dev, struct its_dev *its_dev, uint8_t valid)
1882 {
1883 	struct its_cmd_desc desc;
1884 
1885 	desc.cmd_type = ITS_CMD_MAPD;
1886 	desc.cmd_desc_mapd.its_dev = its_dev;
1887 	desc.cmd_desc_mapd.valid = valid;
1888 
1889 	its_cmd_send(dev, &desc);
1890 }
1891 
1892 static void
1893 its_cmd_inv(device_t dev, struct its_dev *its_dev,
1894     struct gicv3_its_irqsrc *girq)
1895 {
1896 	struct gicv3_its_softc *sc;
1897 	struct its_cmd_desc desc;
1898 	struct its_col *col;
1899 
1900 	sc = device_get_softc(dev);
1901 	col = sc->sc_its_cols[CPU_FFS(&girq->gi_isrc.isrc_cpu) - 1];
1902 
1903 	desc.cmd_type = ITS_CMD_INV;
1904 	/* The EventID sent to the device */
1905 	desc.cmd_desc_inv.pid = girq->gi_id;
1906 	desc.cmd_desc_inv.its_dev = its_dev;
1907 	desc.cmd_desc_inv.col = col;
1908 
1909 	its_cmd_send(dev, &desc);
1910 }
1911 
1912 static void
1913 its_cmd_invall(device_t dev, struct its_col *col)
1914 {
1915 	struct its_cmd_desc desc;
1916 
1917 	desc.cmd_type = ITS_CMD_INVALL;
1918 	desc.cmd_desc_invall.col = col;
1919 
1920 	its_cmd_send(dev, &desc);
1921 }
1922 
1923 #ifdef FDT
1924 static device_probe_t gicv3_its_fdt_probe;
1925 static device_attach_t gicv3_its_fdt_attach;
1926 
1927 static device_method_t gicv3_its_fdt_methods[] = {
1928 	/* Device interface */
1929 	DEVMETHOD(device_probe,		gicv3_its_fdt_probe),
1930 	DEVMETHOD(device_attach,	gicv3_its_fdt_attach),
1931 
1932 	/* End */
1933 	DEVMETHOD_END
1934 };
1935 
1936 #define its_baseclasses its_fdt_baseclasses
1937 DEFINE_CLASS_1(its, gicv3_its_fdt_driver, gicv3_its_fdt_methods,
1938     sizeof(struct gicv3_its_softc), gicv3_its_driver);
1939 #undef its_baseclasses
1940 
1941 EARLY_DRIVER_MODULE(its_fdt, gic, gicv3_its_fdt_driver, 0, 0,
1942     BUS_PASS_INTERRUPT + BUS_PASS_ORDER_MIDDLE);
1943 
1944 static int
1945 gicv3_its_fdt_probe(device_t dev)
1946 {
1947 
1948 	if (!ofw_bus_status_okay(dev))
1949 		return (ENXIO);
1950 
1951 	if (!ofw_bus_is_compatible(dev, "arm,gic-v3-its"))
1952 		return (ENXIO);
1953 
1954 	device_set_desc(dev, "ARM GIC Interrupt Translation Service");
1955 	return (BUS_PROBE_DEFAULT);
1956 }
1957 
1958 static int
1959 gicv3_its_fdt_attach(device_t dev)
1960 {
1961 	struct gicv3_its_softc *sc;
1962 	phandle_t xref;
1963 	int err;
1964 
1965 	sc = device_get_softc(dev);
1966 	sc->dev = dev;
1967 	err = gicv3_its_attach(dev);
1968 	if (err != 0)
1969 		return (err);
1970 
1971 	/* Register this device as a interrupt controller */
1972 	xref = OF_xref_from_node(ofw_bus_get_node(dev));
1973 	sc->sc_pic = intr_pic_register(dev, xref);
1974 	err = intr_pic_add_handler(device_get_parent(dev), sc->sc_pic,
1975 	    gicv3_its_intr, sc, sc->sc_irq_base, sc->sc_irq_length);
1976 	if (err != 0) {
1977 		device_printf(dev, "Failed to add PIC handler: %d\n", err);
1978 		return (err);
1979 	}
1980 
1981 	/* Register this device to handle MSI interrupts */
1982 	err = intr_msi_register(dev, xref);
1983 	if (err != 0) {
1984 		device_printf(dev, "Failed to register for MSIs: %d\n", err);
1985 		return (err);
1986 	}
1987 
1988 	return (0);
1989 }
1990 #endif
1991 
1992 #ifdef DEV_ACPI
1993 static device_probe_t gicv3_its_acpi_probe;
1994 static device_attach_t gicv3_its_acpi_attach;
1995 
1996 static device_method_t gicv3_its_acpi_methods[] = {
1997 	/* Device interface */
1998 	DEVMETHOD(device_probe,		gicv3_its_acpi_probe),
1999 	DEVMETHOD(device_attach,	gicv3_its_acpi_attach),
2000 
2001 	/* End */
2002 	DEVMETHOD_END
2003 };
2004 
2005 #define its_baseclasses its_acpi_baseclasses
2006 DEFINE_CLASS_1(its, gicv3_its_acpi_driver, gicv3_its_acpi_methods,
2007     sizeof(struct gicv3_its_softc), gicv3_its_driver);
2008 #undef its_baseclasses
2009 
2010 EARLY_DRIVER_MODULE(its_acpi, gic, gicv3_its_acpi_driver, 0, 0,
2011     BUS_PASS_INTERRUPT + BUS_PASS_ORDER_MIDDLE);
2012 
2013 static int
2014 gicv3_its_acpi_probe(device_t dev)
2015 {
2016 
2017 	if (gic_get_bus(dev) != GIC_BUS_ACPI)
2018 		return (EINVAL);
2019 
2020 	if (gic_get_hw_rev(dev) < 3)
2021 		return (EINVAL);
2022 
2023 	device_set_desc(dev, "ARM GIC Interrupt Translation Service");
2024 	return (BUS_PROBE_DEFAULT);
2025 }
2026 
2027 static int
2028 gicv3_its_acpi_attach(device_t dev)
2029 {
2030 	struct gicv3_its_softc *sc;
2031 	struct gic_v3_devinfo *di;
2032 	int err;
2033 
2034 	sc = device_get_softc(dev);
2035 	sc->dev = dev;
2036 	err = gicv3_its_attach(dev);
2037 	if (err != 0)
2038 		return (err);
2039 
2040 	di = device_get_ivars(dev);
2041 	sc->sc_pic = intr_pic_register(dev, di->msi_xref);
2042 	err = intr_pic_add_handler(device_get_parent(dev), sc->sc_pic,
2043 	    gicv3_its_intr, sc, sc->sc_irq_base, sc->sc_irq_length);
2044 	if (err != 0) {
2045 		device_printf(dev, "Failed to add PIC handler: %d\n", err);
2046 		return (err);
2047 	}
2048 
2049 	/* Register this device to handle MSI interrupts */
2050 	err = intr_msi_register(dev, di->msi_xref);
2051 	if (err != 0) {
2052 		device_printf(dev, "Failed to register for MSIs: %d\n", err);
2053 		return (err);
2054 	}
2055 
2056 	return (0);
2057 }
2058 #endif
2059