xref: /freebsd/sys/arm64/arm64/gicv3_its.c (revision 9768746b)
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 __FBSDID("$FreeBSD$");
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/bus.h>
42 #include <sys/cpuset.h>
43 #include <sys/domainset.h>
44 #include <sys/endian.h>
45 #include <sys/kernel.h>
46 #include <sys/lock.h>
47 #include <sys/malloc.h>
48 #include <sys/module.h>
49 #include <sys/mutex.h>
50 #include <sys/proc.h>
51 #include <sys/taskqueue.h>
52 #include <sys/tree.h>
53 #include <sys/queue.h>
54 #include <sys/rman.h>
55 #include <sys/sbuf.h>
56 #include <sys/smp.h>
57 #include <sys/sysctl.h>
58 #include <sys/vmem.h>
59 
60 #include <vm/vm.h>
61 #include <vm/pmap.h>
62 #include <vm/vm_page.h>
63 
64 #include <machine/bus.h>
65 #include <machine/intr.h>
66 
67 #include <arm/arm/gic_common.h>
68 #include <arm64/arm64/gic_v3_reg.h>
69 #include <arm64/arm64/gic_v3_var.h>
70 
71 #ifdef FDT
72 #include <dev/ofw/openfirm.h>
73 #include <dev/ofw/ofw_bus.h>
74 #include <dev/ofw/ofw_bus_subr.h>
75 #endif
76 #include <dev/pci/pcireg.h>
77 #include <dev/pci/pcivar.h>
78 
79 #ifdef IOMMU
80 #include <dev/iommu/iommu.h>
81 #include <dev/iommu/iommu_gas.h>
82 #endif
83 
84 #include "pcib_if.h"
85 #include "pic_if.h"
86 #include "msi_if.h"
87 
88 MALLOC_DEFINE(M_GICV3_ITS, "GICv3 ITS",
89     "ARM GICv3 Interrupt Translation Service");
90 
91 #define	LPI_NIRQS		(64 * 1024)
92 
93 /* The size and alignment of the command circular buffer */
94 #define	ITS_CMDQ_SIZE		(64 * 1024)	/* Must be a multiple of 4K */
95 #define	ITS_CMDQ_ALIGN		(64 * 1024)
96 
97 #define	LPI_CONFTAB_SIZE	LPI_NIRQS
98 #define	LPI_CONFTAB_ALIGN	(64 * 1024)
99 #define	LPI_CONFTAB_MAX_ADDR	((1ul << 48) - 1) /* We need a 47 bit PA */
100 
101 /* 1 bit per SPI, PPI, and SGI (8k), and 1 bit per LPI (LPI_CONFTAB_SIZE) */
102 #define	LPI_PENDTAB_SIZE	((LPI_NIRQS + GIC_FIRST_LPI) / 8)
103 #define	LPI_PENDTAB_ALIGN	(64 * 1024)
104 #define	LPI_PENDTAB_MAX_ADDR	((1ul << 48) - 1) /* We need a 47 bit PA */
105 
106 #define	LPI_INT_TRANS_TAB_ALIGN	256
107 #define	LPI_INT_TRANS_TAB_MAX_ADDR ((1ul << 48) - 1)
108 
109 /* ITS commands encoding */
110 #define	ITS_CMD_MOVI		(0x01)
111 #define	ITS_CMD_SYNC		(0x05)
112 #define	ITS_CMD_MAPD		(0x08)
113 #define	ITS_CMD_MAPC		(0x09)
114 #define	ITS_CMD_MAPTI		(0x0a)
115 #define	ITS_CMD_MAPI		(0x0b)
116 #define	ITS_CMD_INV		(0x0c)
117 #define	ITS_CMD_INVALL		(0x0d)
118 /* Command */
119 #define	CMD_COMMAND_MASK	(0xFFUL)
120 /* PCI device ID */
121 #define	CMD_DEVID_SHIFT		(32)
122 #define	CMD_DEVID_MASK		(0xFFFFFFFFUL << CMD_DEVID_SHIFT)
123 /* Size of IRQ ID bitfield */
124 #define	CMD_SIZE_MASK		(0xFFUL)
125 /* Virtual LPI ID */
126 #define	CMD_ID_MASK		(0xFFFFFFFFUL)
127 /* Physical LPI ID */
128 #define	CMD_PID_SHIFT		(32)
129 #define	CMD_PID_MASK		(0xFFFFFFFFUL << CMD_PID_SHIFT)
130 /* Collection */
131 #define	CMD_COL_MASK		(0xFFFFUL)
132 /* Target (CPU or Re-Distributor) */
133 #define	CMD_TARGET_SHIFT	(16)
134 #define	CMD_TARGET_MASK		(0xFFFFFFFFUL << CMD_TARGET_SHIFT)
135 /* Interrupt Translation Table address */
136 #define	CMD_ITT_MASK		(0xFFFFFFFFFF00UL)
137 /* Valid command bit */
138 #define	CMD_VALID_SHIFT		(63)
139 #define	CMD_VALID_MASK		(1UL << CMD_VALID_SHIFT)
140 
141 #define	ITS_TARGET_NONE		0xFBADBEEF
142 
143 /* LPI chunk owned by ITS device */
144 struct lpi_chunk {
145 	u_int	lpi_base;
146 	u_int	lpi_free;	/* First free LPI in set */
147 	u_int	lpi_num;	/* Total number of LPIs in chunk */
148 	u_int	lpi_busy;	/* Number of busy LPIs in chink */
149 };
150 
151 /* ITS device */
152 struct its_dev {
153 	TAILQ_ENTRY(its_dev)	entry;
154 	/* PCI device */
155 	device_t		pci_dev;
156 	/* Device ID (i.e. PCI device ID) */
157 	uint32_t		devid;
158 	/* List of assigned LPIs */
159 	struct lpi_chunk	lpis;
160 	/* Virtual address of ITT */
161 	vm_offset_t		itt;
162 	size_t			itt_size;
163 };
164 
165 /*
166  * ITS command descriptor.
167  * Idea for command description passing taken from Linux.
168  */
169 struct its_cmd_desc {
170 	uint8_t cmd_type;
171 
172 	union {
173 		struct {
174 			struct its_dev *its_dev;
175 			struct its_col *col;
176 			uint32_t id;
177 		} cmd_desc_movi;
178 
179 		struct {
180 			struct its_col *col;
181 		} cmd_desc_sync;
182 
183 		struct {
184 			struct its_col *col;
185 			uint8_t valid;
186 		} cmd_desc_mapc;
187 
188 		struct {
189 			struct its_dev *its_dev;
190 			struct its_col *col;
191 			uint32_t pid;
192 			uint32_t id;
193 		} cmd_desc_mapvi;
194 
195 		struct {
196 			struct its_dev *its_dev;
197 			struct its_col *col;
198 			uint32_t pid;
199 		} cmd_desc_mapi;
200 
201 		struct {
202 			struct its_dev *its_dev;
203 			uint8_t valid;
204 		} cmd_desc_mapd;
205 
206 		struct {
207 			struct its_dev *its_dev;
208 			struct its_col *col;
209 			uint32_t pid;
210 		} cmd_desc_inv;
211 
212 		struct {
213 			struct its_col *col;
214 		} cmd_desc_invall;
215 	};
216 };
217 
218 /* ITS command. Each command is 32 bytes long */
219 struct its_cmd {
220 	uint64_t	cmd_dword[4];	/* ITS command double word */
221 };
222 
223 /* An ITS private table */
224 struct its_ptable {
225 	vm_offset_t	ptab_vaddr;
226 	unsigned long	ptab_size;
227 };
228 
229 /* ITS collection description. */
230 struct its_col {
231 	uint64_t	col_target;	/* Target Re-Distributor */
232 	uint64_t	col_id;		/* Collection ID */
233 };
234 
235 struct gicv3_its_irqsrc {
236 	struct intr_irqsrc	gi_isrc;
237 	u_int			gi_id;
238 	u_int			gi_lpi;
239 	struct its_dev		*gi_its_dev;
240 	TAILQ_ENTRY(gicv3_its_irqsrc) gi_link;
241 };
242 
243 struct gicv3_its_softc {
244 	device_t	dev;
245 	struct intr_pic *sc_pic;
246 	struct resource *sc_its_res;
247 
248 	cpuset_t	sc_cpus;
249 	struct domainset *sc_ds;
250 	u_int		gic_irq_cpu;
251 
252 	struct its_ptable sc_its_ptab[GITS_BASER_NUM];
253 	struct its_col *sc_its_cols[MAXCPU];	/* Per-CPU collections */
254 
255 	/*
256 	 * TODO: We should get these from the parent as we only want a
257 	 * single copy of each across the interrupt controller.
258 	 */
259 	uint8_t		*sc_conf_base;
260 	vm_offset_t sc_pend_base[MAXCPU];
261 
262 	/* Command handling */
263 	struct mtx sc_its_cmd_lock;
264 	struct its_cmd *sc_its_cmd_base; /* Command circular buffer address */
265 	size_t sc_its_cmd_next_idx;
266 
267 	vmem_t *sc_irq_alloc;
268 	struct gicv3_its_irqsrc	**sc_irqs;
269 	u_int	sc_irq_base;
270 	u_int	sc_irq_length;
271 	u_int	sc_irq_count;
272 
273 	struct mtx sc_its_dev_lock;
274 	TAILQ_HEAD(its_dev_list, its_dev) sc_its_dev_list;
275 	TAILQ_HEAD(free_irqs, gicv3_its_irqsrc) sc_free_irqs;
276 
277 #define	ITS_FLAGS_CMDQ_FLUSH		0x00000001
278 #define	ITS_FLAGS_LPI_CONF_FLUSH	0x00000002
279 #define	ITS_FLAGS_ERRATA_CAVIUM_22375	0x00000004
280 	u_int sc_its_flags;
281 	bool	trace_enable;
282 	vm_page_t ma; /* fake msi page */
283 };
284 
285 static void *conf_base;
286 
287 typedef void (its_quirk_func_t)(device_t);
288 static its_quirk_func_t its_quirk_cavium_22375;
289 
290 static const struct {
291 	const char *desc;
292 	uint32_t iidr;
293 	uint32_t iidr_mask;
294 	its_quirk_func_t *func;
295 } its_quirks[] = {
296 	{
297 		/* Cavium ThunderX Pass 1.x */
298 		.desc = "Cavium ThunderX errata: 22375, 24313",
299 		.iidr = GITS_IIDR_RAW(GITS_IIDR_IMPL_CAVIUM,
300 		    GITS_IIDR_PROD_THUNDER, GITS_IIDR_VAR_THUNDER_1, 0),
301 		.iidr_mask = ~GITS_IIDR_REVISION_MASK,
302 		.func = its_quirk_cavium_22375,
303 	},
304 };
305 
306 #define	gic_its_read_4(sc, reg)			\
307     bus_read_4((sc)->sc_its_res, (reg))
308 #define	gic_its_read_8(sc, reg)			\
309     bus_read_8((sc)->sc_its_res, (reg))
310 
311 #define	gic_its_write_4(sc, reg, val)		\
312     bus_write_4((sc)->sc_its_res, (reg), (val))
313 #define	gic_its_write_8(sc, reg, val)		\
314     bus_write_8((sc)->sc_its_res, (reg), (val))
315 
316 static device_attach_t gicv3_its_attach;
317 static device_detach_t gicv3_its_detach;
318 
319 static pic_disable_intr_t gicv3_its_disable_intr;
320 static pic_enable_intr_t gicv3_its_enable_intr;
321 static pic_map_intr_t gicv3_its_map_intr;
322 static pic_setup_intr_t gicv3_its_setup_intr;
323 static pic_post_filter_t gicv3_its_post_filter;
324 static pic_post_ithread_t gicv3_its_post_ithread;
325 static pic_pre_ithread_t gicv3_its_pre_ithread;
326 static pic_bind_intr_t gicv3_its_bind_intr;
327 #ifdef SMP
328 static pic_init_secondary_t gicv3_its_init_secondary;
329 #endif
330 static msi_alloc_msi_t gicv3_its_alloc_msi;
331 static msi_release_msi_t gicv3_its_release_msi;
332 static msi_alloc_msix_t gicv3_its_alloc_msix;
333 static msi_release_msix_t gicv3_its_release_msix;
334 static msi_map_msi_t gicv3_its_map_msi;
335 #ifdef IOMMU
336 static msi_iommu_init_t gicv3_iommu_init;
337 static msi_iommu_deinit_t gicv3_iommu_deinit;
338 #endif
339 
340 static void its_cmd_movi(device_t, struct gicv3_its_irqsrc *);
341 static void its_cmd_mapc(device_t, struct its_col *, uint8_t);
342 static void its_cmd_mapti(device_t, struct gicv3_its_irqsrc *);
343 static void its_cmd_mapd(device_t, struct its_dev *, uint8_t);
344 static void its_cmd_inv(device_t, struct its_dev *, struct gicv3_its_irqsrc *);
345 static void its_cmd_invall(device_t, struct its_col *);
346 
347 static device_method_t gicv3_its_methods[] = {
348 	/* Device interface */
349 	DEVMETHOD(device_detach,	gicv3_its_detach),
350 
351 	/* Interrupt controller interface */
352 	DEVMETHOD(pic_disable_intr,	gicv3_its_disable_intr),
353 	DEVMETHOD(pic_enable_intr,	gicv3_its_enable_intr),
354 	DEVMETHOD(pic_map_intr,		gicv3_its_map_intr),
355 	DEVMETHOD(pic_setup_intr,	gicv3_its_setup_intr),
356 	DEVMETHOD(pic_post_filter,	gicv3_its_post_filter),
357 	DEVMETHOD(pic_post_ithread,	gicv3_its_post_ithread),
358 	DEVMETHOD(pic_pre_ithread,	gicv3_its_pre_ithread),
359 #ifdef SMP
360 	DEVMETHOD(pic_bind_intr,	gicv3_its_bind_intr),
361 	DEVMETHOD(pic_init_secondary,	gicv3_its_init_secondary),
362 #endif
363 
364 	/* MSI/MSI-X */
365 	DEVMETHOD(msi_alloc_msi,	gicv3_its_alloc_msi),
366 	DEVMETHOD(msi_release_msi,	gicv3_its_release_msi),
367 	DEVMETHOD(msi_alloc_msix,	gicv3_its_alloc_msix),
368 	DEVMETHOD(msi_release_msix,	gicv3_its_release_msix),
369 	DEVMETHOD(msi_map_msi,		gicv3_its_map_msi),
370 #ifdef IOMMU
371 	DEVMETHOD(msi_iommu_init,	gicv3_iommu_init),
372 	DEVMETHOD(msi_iommu_deinit,	gicv3_iommu_deinit),
373 #endif
374 
375 	/* End */
376 	DEVMETHOD_END
377 };
378 
379 static DEFINE_CLASS_0(gic, gicv3_its_driver, gicv3_its_methods,
380     sizeof(struct gicv3_its_softc));
381 
382 static void
383 gicv3_its_cmdq_init(struct gicv3_its_softc *sc)
384 {
385 	vm_paddr_t cmd_paddr;
386 	uint64_t reg, tmp;
387 
388 	/* Set up the command circular buffer */
389 	sc->sc_its_cmd_base = contigmalloc_domainset(ITS_CMDQ_SIZE, M_GICV3_ITS,
390 	    sc->sc_ds, M_WAITOK | M_ZERO, 0, (1ul << 48) - 1, ITS_CMDQ_ALIGN,
391 	    0);
392 	sc->sc_its_cmd_next_idx = 0;
393 
394 	cmd_paddr = vtophys(sc->sc_its_cmd_base);
395 
396 	/* Set the base of the command buffer */
397 	reg = GITS_CBASER_VALID |
398 	    (GITS_CBASER_CACHE_NIWAWB << GITS_CBASER_CACHE_SHIFT) |
399 	    cmd_paddr | (GITS_CBASER_SHARE_IS << GITS_CBASER_SHARE_SHIFT) |
400 	    (ITS_CMDQ_SIZE / 4096 - 1);
401 	gic_its_write_8(sc, GITS_CBASER, reg);
402 
403 	/* Read back to check for fixed value fields */
404 	tmp = gic_its_read_8(sc, GITS_CBASER);
405 
406 	if ((tmp & GITS_CBASER_SHARE_MASK) !=
407 	    (GITS_CBASER_SHARE_IS << GITS_CBASER_SHARE_SHIFT)) {
408 		/* Check if the hardware reported non-shareable */
409 		if ((tmp & GITS_CBASER_SHARE_MASK) ==
410 		    (GITS_CBASER_SHARE_NS << GITS_CBASER_SHARE_SHIFT)) {
411 			/* If so remove the cache attribute */
412 			reg &= ~GITS_CBASER_CACHE_MASK;
413 			reg &= ~GITS_CBASER_SHARE_MASK;
414 			/* Set to Non-cacheable, Non-shareable */
415 			reg |= GITS_CBASER_CACHE_NIN << GITS_CBASER_CACHE_SHIFT;
416 			reg |= GITS_CBASER_SHARE_NS << GITS_CBASER_SHARE_SHIFT;
417 
418 			gic_its_write_8(sc, GITS_CBASER, reg);
419 		}
420 
421 		/* The command queue has to be flushed after each command */
422 		sc->sc_its_flags |= ITS_FLAGS_CMDQ_FLUSH;
423 	}
424 
425 	/* Get the next command from the start of the buffer */
426 	gic_its_write_8(sc, GITS_CWRITER, 0x0);
427 }
428 
429 static int
430 gicv3_its_table_init(device_t dev, struct gicv3_its_softc *sc)
431 {
432 	vm_offset_t table;
433 	vm_paddr_t paddr;
434 	uint64_t cache, reg, share, tmp, type;
435 	size_t esize, its_tbl_size, nidents, nitspages, npages;
436 	int i, page_size;
437 	int devbits;
438 
439 	if ((sc->sc_its_flags & ITS_FLAGS_ERRATA_CAVIUM_22375) != 0) {
440 		/*
441 		 * GITS_TYPER[17:13] of ThunderX reports that device IDs
442 		 * are to be 21 bits in length. The entry size of the ITS
443 		 * table can be read from GITS_BASERn[52:48] and on ThunderX
444 		 * is supposed to be 8 bytes in length (for device table).
445 		 * Finally the page size that is to be used by ITS to access
446 		 * this table will be set to 64KB.
447 		 *
448 		 * This gives 0x200000 entries of size 0x8 bytes covered by
449 		 * 256 pages each of which 64KB in size. The number of pages
450 		 * (minus 1) should then be written to GITS_BASERn[7:0]. In
451 		 * that case this value would be 0xFF but on ThunderX the
452 		 * maximum value that HW accepts is 0xFD.
453 		 *
454 		 * Set an arbitrary number of device ID bits to 20 in order
455 		 * to limit the number of entries in ITS device table to
456 		 * 0x100000 and the table size to 8MB.
457 		 */
458 		devbits = 20;
459 		cache = 0;
460 	} else {
461 		devbits = GITS_TYPER_DEVB(gic_its_read_8(sc, GITS_TYPER));
462 		cache = GITS_BASER_CACHE_WAWB;
463 	}
464 	share = GITS_BASER_SHARE_IS;
465 	page_size = PAGE_SIZE_64K;
466 
467 	for (i = 0; i < GITS_BASER_NUM; i++) {
468 		reg = gic_its_read_8(sc, GITS_BASER(i));
469 		/* The type of table */
470 		type = GITS_BASER_TYPE(reg);
471 		/* The table entry size */
472 		esize = GITS_BASER_ESIZE(reg);
473 
474 		switch(type) {
475 		case GITS_BASER_TYPE_DEV:
476 			nidents = (1 << devbits);
477 			its_tbl_size = esize * nidents;
478 			its_tbl_size = roundup2(its_tbl_size, PAGE_SIZE_64K);
479 			break;
480 		case GITS_BASER_TYPE_VP:
481 		case GITS_BASER_TYPE_PP: /* Undocumented? */
482 		case GITS_BASER_TYPE_IC:
483 			its_tbl_size = page_size;
484 			break;
485 		default:
486 			continue;
487 		}
488 		npages = howmany(its_tbl_size, PAGE_SIZE);
489 
490 		/* Allocate the table */
491 		table = (vm_offset_t)contigmalloc_domainset(npages * PAGE_SIZE,
492 		    M_GICV3_ITS, sc->sc_ds, M_WAITOK | M_ZERO, 0,
493 		    (1ul << 48) - 1, PAGE_SIZE_64K, 0);
494 
495 		sc->sc_its_ptab[i].ptab_vaddr = table;
496 		sc->sc_its_ptab[i].ptab_size = npages * PAGE_SIZE;
497 
498 		paddr = vtophys(table);
499 
500 		while (1) {
501 			nitspages = howmany(its_tbl_size, page_size);
502 
503 			/* Clear the fields we will be setting */
504 			reg &= ~(GITS_BASER_VALID | GITS_BASER_INDIRECT |
505 			    GITS_BASER_CACHE_MASK | GITS_BASER_TYPE_MASK |
506 			    GITS_BASER_ESIZE_MASK | GITS_BASER_PA_MASK |
507 			    GITS_BASER_SHARE_MASK | GITS_BASER_PSZ_MASK |
508 			    GITS_BASER_SIZE_MASK);
509 			/* Set the new values */
510 			reg |= GITS_BASER_VALID |
511 			    (cache << GITS_BASER_CACHE_SHIFT) |
512 			    (type << GITS_BASER_TYPE_SHIFT) |
513 			    ((esize - 1) << GITS_BASER_ESIZE_SHIFT) |
514 			    paddr | (share << GITS_BASER_SHARE_SHIFT) |
515 			    (nitspages - 1);
516 
517 			switch (page_size) {
518 			case PAGE_SIZE_4K:	/* 4KB */
519 				reg |=
520 				    GITS_BASER_PSZ_4K << GITS_BASER_PSZ_SHIFT;
521 				break;
522 			case PAGE_SIZE_16K:	/* 16KB */
523 				reg |=
524 				    GITS_BASER_PSZ_16K << GITS_BASER_PSZ_SHIFT;
525 				break;
526 			case PAGE_SIZE_64K:	/* 64KB */
527 				reg |=
528 				    GITS_BASER_PSZ_64K << GITS_BASER_PSZ_SHIFT;
529 				break;
530 			}
531 
532 			gic_its_write_8(sc, GITS_BASER(i), reg);
533 
534 			/* Read back to check */
535 			tmp = gic_its_read_8(sc, GITS_BASER(i));
536 
537 			/* Do the shareability masks line up? */
538 			if ((tmp & GITS_BASER_SHARE_MASK) !=
539 			    (reg & GITS_BASER_SHARE_MASK)) {
540 				share = (tmp & GITS_BASER_SHARE_MASK) >>
541 				    GITS_BASER_SHARE_SHIFT;
542 				continue;
543 			}
544 
545 			if ((tmp & GITS_BASER_PSZ_MASK) !=
546 			    (reg & GITS_BASER_PSZ_MASK)) {
547 				switch (page_size) {
548 				case PAGE_SIZE_16K:
549 					page_size = PAGE_SIZE_4K;
550 					continue;
551 				case PAGE_SIZE_64K:
552 					page_size = PAGE_SIZE_16K;
553 					continue;
554 				}
555 			}
556 
557 			if (tmp != reg) {
558 				device_printf(dev, "GITS_BASER%d: "
559 				    "unable to be updated: %lx != %lx\n",
560 				    i, reg, tmp);
561 				return (ENXIO);
562 			}
563 
564 			/* We should have made all needed changes */
565 			break;
566 		}
567 	}
568 
569 	return (0);
570 }
571 
572 static void
573 gicv3_its_conftable_init(struct gicv3_its_softc *sc)
574 {
575 	void *conf_table;
576 
577 	conf_table = atomic_load_ptr(&conf_base);
578 	if (conf_table == NULL) {
579 		conf_table = contigmalloc(LPI_CONFTAB_SIZE,
580 		    M_GICV3_ITS, M_WAITOK, 0, LPI_CONFTAB_MAX_ADDR,
581 		    LPI_CONFTAB_ALIGN, 0);
582 
583 		if (atomic_cmpset_ptr((uintptr_t *)&conf_base,
584 		    (uintptr_t)NULL, (uintptr_t)conf_table) == 0) {
585 			contigfree(conf_table, LPI_CONFTAB_SIZE, M_GICV3_ITS);
586 			conf_table = atomic_load_ptr(&conf_base);
587 		}
588 	}
589 	sc->sc_conf_base = conf_table;
590 
591 	/* Set the default configuration */
592 	memset(sc->sc_conf_base, GIC_PRIORITY_MAX | LPI_CONF_GROUP1,
593 	    LPI_CONFTAB_SIZE);
594 
595 	/* Flush the table to memory */
596 	cpu_dcache_wb_range((vm_offset_t)sc->sc_conf_base, LPI_CONFTAB_SIZE);
597 }
598 
599 static void
600 gicv3_its_pendtables_init(struct gicv3_its_softc *sc)
601 {
602 	int i;
603 
604 	for (i = 0; i <= mp_maxid; i++) {
605 		if (CPU_ISSET(i, &sc->sc_cpus) == 0)
606 			continue;
607 
608 		sc->sc_pend_base[i] = (vm_offset_t)contigmalloc(
609 		    LPI_PENDTAB_SIZE, M_GICV3_ITS, M_WAITOK | M_ZERO,
610 		    0, LPI_PENDTAB_MAX_ADDR, LPI_PENDTAB_ALIGN, 0);
611 
612 		/* Flush so the ITS can see the memory */
613 		cpu_dcache_wb_range((vm_offset_t)sc->sc_pend_base[i],
614 		    LPI_PENDTAB_SIZE);
615 	}
616 }
617 
618 static void
619 its_init_cpu_lpi(device_t dev, struct gicv3_its_softc *sc)
620 {
621 	device_t gicv3;
622 	uint64_t xbaser, tmp;
623 	uint32_t ctlr;
624 	u_int cpuid;
625 
626 	gicv3 = device_get_parent(dev);
627 	cpuid = PCPU_GET(cpuid);
628 
629 	/* Disable LPIs */
630 	ctlr = gic_r_read_4(gicv3, GICR_CTLR);
631 	ctlr &= ~GICR_CTLR_LPI_ENABLE;
632 	gic_r_write_4(gicv3, GICR_CTLR, ctlr);
633 
634 	/* Make sure changes are observable my the GIC */
635 	dsb(sy);
636 
637 	/*
638 	 * Set the redistributor base
639 	 */
640 	xbaser = vtophys(sc->sc_conf_base) |
641 	    (GICR_PROPBASER_SHARE_IS << GICR_PROPBASER_SHARE_SHIFT) |
642 	    (GICR_PROPBASER_CACHE_NIWAWB << GICR_PROPBASER_CACHE_SHIFT) |
643 	    (flsl(LPI_CONFTAB_SIZE | GIC_FIRST_LPI) - 1);
644 	gic_r_write_8(gicv3, GICR_PROPBASER, xbaser);
645 
646 	/* Check the cache attributes we set */
647 	tmp = gic_r_read_8(gicv3, GICR_PROPBASER);
648 
649 	if ((tmp & GICR_PROPBASER_SHARE_MASK) !=
650 	    (xbaser & GICR_PROPBASER_SHARE_MASK)) {
651 		if ((tmp & GICR_PROPBASER_SHARE_MASK) ==
652 		    (GICR_PROPBASER_SHARE_NS << GICR_PROPBASER_SHARE_SHIFT)) {
653 			/* We need to mark as non-cacheable */
654 			xbaser &= ~(GICR_PROPBASER_SHARE_MASK |
655 			    GICR_PROPBASER_CACHE_MASK);
656 			/* Non-cacheable */
657 			xbaser |= GICR_PROPBASER_CACHE_NIN <<
658 			    GICR_PROPBASER_CACHE_SHIFT;
659 			/* Non-shareable */
660 			xbaser |= GICR_PROPBASER_SHARE_NS <<
661 			    GICR_PROPBASER_SHARE_SHIFT;
662 			gic_r_write_8(gicv3, GICR_PROPBASER, xbaser);
663 		}
664 		sc->sc_its_flags |= ITS_FLAGS_LPI_CONF_FLUSH;
665 	}
666 
667 	/*
668 	 * Set the LPI pending table base
669 	 */
670 	xbaser = vtophys(sc->sc_pend_base[cpuid]) |
671 	    (GICR_PENDBASER_CACHE_NIWAWB << GICR_PENDBASER_CACHE_SHIFT) |
672 	    (GICR_PENDBASER_SHARE_IS << GICR_PENDBASER_SHARE_SHIFT);
673 
674 	gic_r_write_8(gicv3, GICR_PENDBASER, xbaser);
675 
676 	tmp = gic_r_read_8(gicv3, GICR_PENDBASER);
677 
678 	if ((tmp & GICR_PENDBASER_SHARE_MASK) ==
679 	    (GICR_PENDBASER_SHARE_NS << GICR_PENDBASER_SHARE_SHIFT)) {
680 		/* Clear the cahce and shareability bits */
681 		xbaser &= ~(GICR_PENDBASER_CACHE_MASK |
682 		    GICR_PENDBASER_SHARE_MASK);
683 		/* Mark as non-shareable */
684 		xbaser |= GICR_PENDBASER_SHARE_NS << GICR_PENDBASER_SHARE_SHIFT;
685 		/* And non-cacheable */
686 		xbaser |= GICR_PENDBASER_CACHE_NIN <<
687 		    GICR_PENDBASER_CACHE_SHIFT;
688 	}
689 
690 	/* Enable LPIs */
691 	ctlr = gic_r_read_4(gicv3, GICR_CTLR);
692 	ctlr |= GICR_CTLR_LPI_ENABLE;
693 	gic_r_write_4(gicv3, GICR_CTLR, ctlr);
694 
695 	/* Make sure the GIC has seen everything */
696 	dsb(sy);
697 }
698 
699 static int
700 its_init_cpu(device_t dev, struct gicv3_its_softc *sc)
701 {
702 	device_t gicv3;
703 	vm_paddr_t target;
704 	u_int cpuid;
705 	struct redist_pcpu *rpcpu;
706 
707 	gicv3 = device_get_parent(dev);
708 	cpuid = PCPU_GET(cpuid);
709 	if (!CPU_ISSET(cpuid, &sc->sc_cpus))
710 		return (0);
711 
712 	/* Check if the ITS is enabled on this CPU */
713 	if ((gic_r_read_8(gicv3, GICR_TYPER) & GICR_TYPER_PLPIS) == 0)
714 		return (ENXIO);
715 
716 	rpcpu = gicv3_get_redist(dev);
717 
718 	/* Do per-cpu LPI init once */
719 	if (!rpcpu->lpi_enabled) {
720 		its_init_cpu_lpi(dev, sc);
721 		rpcpu->lpi_enabled = true;
722 	}
723 
724 	if ((gic_its_read_8(sc, GITS_TYPER) & GITS_TYPER_PTA) != 0) {
725 		/* This ITS wants the redistributor physical address */
726 		target = vtophys(rman_get_virtual(&rpcpu->res));
727 	} else {
728 		/* This ITS wants the unique processor number */
729 		target = GICR_TYPER_CPUNUM(gic_r_read_8(gicv3, GICR_TYPER)) <<
730 		    CMD_TARGET_SHIFT;
731 	}
732 
733 	sc->sc_its_cols[cpuid]->col_target = target;
734 	sc->sc_its_cols[cpuid]->col_id = cpuid;
735 
736 	its_cmd_mapc(dev, sc->sc_its_cols[cpuid], 1);
737 	its_cmd_invall(dev, sc->sc_its_cols[cpuid]);
738 
739 	return (0);
740 }
741 
742 static int
743 gicv3_its_sysctl_trace_enable(SYSCTL_HANDLER_ARGS)
744 {
745 	struct gicv3_its_softc *sc;
746 	int rv;
747 
748 	sc = arg1;
749 
750 	rv = sysctl_handle_bool(oidp, &sc->trace_enable, 0, req);
751 	if (rv != 0 || req->newptr == NULL)
752 		return (rv);
753 	if (sc->trace_enable)
754 		gic_its_write_8(sc, GITS_TRKCTLR, 3);
755 	else
756 		gic_its_write_8(sc, GITS_TRKCTLR, 0);
757 
758 	return (0);
759 }
760 
761 static int
762 gicv3_its_sysctl_trace_regs(SYSCTL_HANDLER_ARGS)
763 {
764 	struct gicv3_its_softc *sc;
765 	struct sbuf *sb;
766 	int err;
767 
768 	sc = arg1;
769 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
770 	if (sb == NULL) {
771 		device_printf(sc->dev, "Could not allocate sbuf for output.\n");
772 		return (ENOMEM);
773 	}
774 	sbuf_cat(sb, "\n");
775 	sbuf_printf(sb, "GITS_TRKCTLR: 0x%08X\n",
776 	    gic_its_read_4(sc, GITS_TRKCTLR));
777 	sbuf_printf(sb, "GITS_TRKR:    0x%08X\n",
778 	    gic_its_read_4(sc, GITS_TRKR));
779 	sbuf_printf(sb, "GITS_TRKDIDR: 0x%08X\n",
780 	    gic_its_read_4(sc, GITS_TRKDIDR));
781 	sbuf_printf(sb, "GITS_TRKPIDR: 0x%08X\n",
782 	    gic_its_read_4(sc, GITS_TRKPIDR));
783 	sbuf_printf(sb, "GITS_TRKVIDR: 0x%08X\n",
784 	    gic_its_read_4(sc, GITS_TRKVIDR));
785 	sbuf_printf(sb, "GITS_TRKTGTR: 0x%08X\n",
786 	   gic_its_read_4(sc, GITS_TRKTGTR));
787 
788 	err = sbuf_finish(sb);
789 	if (err)
790 		device_printf(sc->dev, "Error finishing sbuf: %d\n", err);
791 	sbuf_delete(sb);
792 	return(err);
793 }
794 
795 static int
796 gicv3_its_init_sysctl(struct gicv3_its_softc *sc)
797 {
798 	struct sysctl_oid *oid, *child;
799 	struct sysctl_ctx_list *ctx_list;
800 
801 	ctx_list = device_get_sysctl_ctx(sc->dev);
802 	child = device_get_sysctl_tree(sc->dev);
803 	oid = SYSCTL_ADD_NODE(ctx_list,
804 	    SYSCTL_CHILDREN(child), OID_AUTO, "tracing",
805 	    CTLFLAG_RD| CTLFLAG_MPSAFE, NULL, "Messages tracing");
806 	if (oid == NULL)
807 		return (ENXIO);
808 
809 	/* Add registers */
810 	SYSCTL_ADD_PROC(ctx_list,
811 	    SYSCTL_CHILDREN(oid), OID_AUTO, "enable",
812 	    CTLTYPE_U8 | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
813 	    gicv3_its_sysctl_trace_enable, "CU", "Enable tracing");
814 	SYSCTL_ADD_PROC(ctx_list,
815 	    SYSCTL_CHILDREN(oid), OID_AUTO, "capture",
816 	    CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
817 	    gicv3_its_sysctl_trace_regs, "", "Captured tracing registers.");
818 
819 	return (0);
820 }
821 
822 static int
823 gicv3_its_attach(device_t dev)
824 {
825 	struct gicv3_its_softc *sc;
826 	int domain, err, i, rid;
827 	uint64_t phys;
828 	uint32_t ctlr, iidr;
829 
830 	sc = device_get_softc(dev);
831 
832 	sc->sc_irq_length = gicv3_get_nirqs(dev);
833 	sc->sc_irq_base = GIC_FIRST_LPI;
834 	sc->sc_irq_base += device_get_unit(dev) * sc->sc_irq_length;
835 
836 	rid = 0;
837 	sc->sc_its_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid,
838 	    RF_ACTIVE);
839 	if (sc->sc_its_res == NULL) {
840 		device_printf(dev, "Could not allocate memory\n");
841 		return (ENXIO);
842 	}
843 
844 	phys = rounddown2(vtophys(rman_get_virtual(sc->sc_its_res)) +
845 	    GITS_TRANSLATER, PAGE_SIZE);
846 	sc->ma = malloc(sizeof(struct vm_page), M_DEVBUF, M_WAITOK | M_ZERO);
847 	vm_page_initfake(sc->ma, phys, VM_MEMATTR_DEFAULT);
848 
849 	CPU_COPY(&all_cpus, &sc->sc_cpus);
850 	iidr = gic_its_read_4(sc, GITS_IIDR);
851 	for (i = 0; i < nitems(its_quirks); i++) {
852 		if ((iidr & its_quirks[i].iidr_mask) == its_quirks[i].iidr) {
853 			if (bootverbose) {
854 				device_printf(dev, "Applying %s\n",
855 				    its_quirks[i].desc);
856 			}
857 			its_quirks[i].func(dev);
858 			break;
859 		}
860 	}
861 
862 	if (bus_get_domain(dev, &domain) == 0 && domain < MAXMEMDOM) {
863 		sc->sc_ds = DOMAINSET_PREF(domain);
864 	} else {
865 		sc->sc_ds = DOMAINSET_RR();
866 	}
867 
868 	/*
869 	 * GIT_CTLR_EN is mandated to reset to 0 on a Warm reset, but we may be
870 	 * coming in via, for instance, a kexec/kboot style setup where a
871 	 * previous kernel has configured then relinquished control.  Clear it
872 	 * so that we can reconfigure GITS_BASER*.
873 	 */
874 	ctlr = gic_its_read_4(sc, GITS_CTLR);
875 	if ((ctlr & GITS_CTLR_EN) != 0) {
876 		ctlr &= ~GITS_CTLR_EN;
877 		gic_its_write_4(sc, GITS_CTLR, ctlr);
878 	}
879 
880 	/* Allocate the private tables */
881 	err = gicv3_its_table_init(dev, sc);
882 	if (err != 0)
883 		return (err);
884 
885 	/* Protects access to the device list */
886 	mtx_init(&sc->sc_its_dev_lock, "ITS device lock", NULL, MTX_SPIN);
887 
888 	/* Protects access to the ITS command circular buffer. */
889 	mtx_init(&sc->sc_its_cmd_lock, "ITS cmd lock", NULL, MTX_SPIN);
890 
891 	/* Allocate the command circular buffer */
892 	gicv3_its_cmdq_init(sc);
893 
894 	/* Allocate the per-CPU collections */
895 	for (int cpu = 0; cpu <= mp_maxid; cpu++)
896 		if (CPU_ISSET(cpu, &sc->sc_cpus) != 0)
897 			sc->sc_its_cols[cpu] = malloc_domainset(
898 			    sizeof(*sc->sc_its_cols[0]), M_GICV3_ITS,
899 			    DOMAINSET_PREF(pcpu_find(cpu)->pc_domain),
900 			    M_WAITOK | M_ZERO);
901 		else
902 			sc->sc_its_cols[cpu] = NULL;
903 
904 	/* Enable the ITS */
905 	gic_its_write_4(sc, GITS_CTLR, ctlr | GITS_CTLR_EN);
906 
907 	/* Create the LPI configuration table */
908 	gicv3_its_conftable_init(sc);
909 
910 	/* And the pending tebles */
911 	gicv3_its_pendtables_init(sc);
912 
913 	/* Enable LPIs on this CPU */
914 	its_init_cpu(dev, sc);
915 
916 	TAILQ_INIT(&sc->sc_its_dev_list);
917 	TAILQ_INIT(&sc->sc_free_irqs);
918 
919 	/*
920 	 * Create the vmem object to allocate INTRNG IRQs from. We try to
921 	 * use all IRQs not already used by the GICv3.
922 	 * XXX: This assumes there are no other interrupt controllers in the
923 	 * system.
924 	 */
925 	sc->sc_irq_alloc = vmem_create(device_get_nameunit(dev), 0,
926 	    gicv3_get_nirqs(dev), 1, 0, M_FIRSTFIT | M_WAITOK);
927 
928 	sc->sc_irqs = malloc(sizeof(*sc->sc_irqs) * sc->sc_irq_length,
929 	    M_GICV3_ITS, M_WAITOK | M_ZERO);
930 
931 	/* For GIC-500 install tracking sysctls. */
932 	if ((iidr & (GITS_IIDR_PRODUCT_MASK | GITS_IIDR_IMPLEMENTOR_MASK)) ==
933 	    GITS_IIDR_RAW(GITS_IIDR_IMPL_ARM, GITS_IIDR_PROD_GIC500, 0, 0))
934 		gicv3_its_init_sysctl(sc);
935 
936 	return (0);
937 }
938 
939 static int
940 gicv3_its_detach(device_t dev)
941 {
942 
943 	return (ENXIO);
944 }
945 
946 static void
947 its_quirk_cavium_22375(device_t dev)
948 {
949 	struct gicv3_its_softc *sc;
950 	int domain;
951 
952 	sc = device_get_softc(dev);
953 	sc->sc_its_flags |= ITS_FLAGS_ERRATA_CAVIUM_22375;
954 
955 	/*
956 	 * We need to limit which CPUs we send these interrupts to on
957 	 * the original dual socket ThunderX as it is unable to
958 	 * forward them between the two sockets.
959 	 */
960 	if (bus_get_domain(dev, &domain) == 0) {
961 		if (domain < MAXMEMDOM) {
962 			CPU_COPY(&cpuset_domain[domain], &sc->sc_cpus);
963 		} else {
964 			CPU_ZERO(&sc->sc_cpus);
965 		}
966 	}
967 }
968 
969 static void
970 gicv3_its_disable_intr(device_t dev, struct intr_irqsrc *isrc)
971 {
972 	struct gicv3_its_softc *sc;
973 	struct gicv3_its_irqsrc *girq;
974 	uint8_t *conf;
975 
976 	sc = device_get_softc(dev);
977 	girq = (struct gicv3_its_irqsrc *)isrc;
978 	conf = sc->sc_conf_base;
979 
980 	conf[girq->gi_lpi] &= ~LPI_CONF_ENABLE;
981 
982 	if ((sc->sc_its_flags & ITS_FLAGS_LPI_CONF_FLUSH) != 0) {
983 		/* Clean D-cache under command. */
984 		cpu_dcache_wb_range((vm_offset_t)&conf[girq->gi_lpi], 1);
985 	} else {
986 		/* DSB inner shareable, store */
987 		dsb(ishst);
988 	}
989 
990 	its_cmd_inv(dev, girq->gi_its_dev, girq);
991 }
992 
993 static void
994 gicv3_its_enable_intr(device_t dev, struct intr_irqsrc *isrc)
995 {
996 	struct gicv3_its_softc *sc;
997 	struct gicv3_its_irqsrc *girq;
998 	uint8_t *conf;
999 
1000 	sc = device_get_softc(dev);
1001 	girq = (struct gicv3_its_irqsrc *)isrc;
1002 	conf = sc->sc_conf_base;
1003 
1004 	conf[girq->gi_lpi] |= LPI_CONF_ENABLE;
1005 
1006 	if ((sc->sc_its_flags & ITS_FLAGS_LPI_CONF_FLUSH) != 0) {
1007 		/* Clean D-cache under command. */
1008 		cpu_dcache_wb_range((vm_offset_t)&conf[girq->gi_lpi], 1);
1009 	} else {
1010 		/* DSB inner shareable, store */
1011 		dsb(ishst);
1012 	}
1013 
1014 	its_cmd_inv(dev, girq->gi_its_dev, girq);
1015 }
1016 
1017 static int
1018 gicv3_its_intr(void *arg, uintptr_t irq)
1019 {
1020 	struct gicv3_its_softc *sc = arg;
1021 	struct gicv3_its_irqsrc *girq;
1022 	struct trapframe *tf;
1023 
1024 	irq -= sc->sc_irq_base;
1025 	girq = sc->sc_irqs[irq];
1026 	if (girq == NULL)
1027 		panic("gicv3_its_intr: Invalid interrupt %ld",
1028 		    irq + sc->sc_irq_base);
1029 
1030 	tf = curthread->td_intr_frame;
1031 	intr_isrc_dispatch(&girq->gi_isrc, tf);
1032 	return (FILTER_HANDLED);
1033 }
1034 
1035 static void
1036 gicv3_its_pre_ithread(device_t dev, struct intr_irqsrc *isrc)
1037 {
1038 	struct gicv3_its_irqsrc *girq;
1039 
1040 	girq = (struct gicv3_its_irqsrc *)isrc;
1041 	gic_icc_write(EOIR1, girq->gi_lpi + GIC_FIRST_LPI);
1042 }
1043 
1044 static void
1045 gicv3_its_post_ithread(device_t dev, struct intr_irqsrc *isrc)
1046 {
1047 
1048 }
1049 
1050 static void
1051 gicv3_its_post_filter(device_t dev, struct intr_irqsrc *isrc)
1052 {
1053 	struct gicv3_its_irqsrc *girq;
1054 
1055 	girq = (struct gicv3_its_irqsrc *)isrc;
1056 	gic_icc_write(EOIR1, girq->gi_lpi + GIC_FIRST_LPI);
1057 }
1058 
1059 static int
1060 gicv3_its_select_cpu(device_t dev, struct intr_irqsrc *isrc)
1061 {
1062 	struct gicv3_its_softc *sc;
1063 
1064 	sc = device_get_softc(dev);
1065 	if (CPU_EMPTY(&isrc->isrc_cpu)) {
1066 		sc->gic_irq_cpu = intr_irq_next_cpu(sc->gic_irq_cpu,
1067 		    &sc->sc_cpus);
1068 		CPU_SETOF(sc->gic_irq_cpu, &isrc->isrc_cpu);
1069 	}
1070 
1071 	return (0);
1072 }
1073 
1074 static int
1075 gicv3_its_bind_intr(device_t dev, struct intr_irqsrc *isrc)
1076 {
1077 	struct gicv3_its_irqsrc *girq;
1078 
1079 	gicv3_its_select_cpu(dev, isrc);
1080 
1081 	girq = (struct gicv3_its_irqsrc *)isrc;
1082 	its_cmd_movi(dev, girq);
1083 	return (0);
1084 }
1085 
1086 static int
1087 gicv3_its_map_intr(device_t dev, struct intr_map_data *data,
1088     struct intr_irqsrc **isrcp)
1089 {
1090 
1091 	/*
1092 	 * This should never happen, we only call this function to map
1093 	 * interrupts found before the controller driver is ready.
1094 	 */
1095 	panic("gicv3_its_map_intr: Unable to map a MSI interrupt");
1096 }
1097 
1098 static int
1099 gicv3_its_setup_intr(device_t dev, struct intr_irqsrc *isrc,
1100     struct resource *res, struct intr_map_data *data)
1101 {
1102 
1103 	/* Bind the interrupt to a CPU */
1104 	gicv3_its_bind_intr(dev, isrc);
1105 
1106 	return (0);
1107 }
1108 
1109 #ifdef SMP
1110 static void
1111 gicv3_its_init_secondary(device_t dev)
1112 {
1113 	struct gicv3_its_softc *sc;
1114 
1115 	sc = device_get_softc(dev);
1116 
1117 	/*
1118 	 * This is fatal as otherwise we may bind interrupts to this CPU.
1119 	 * We need a way to tell the interrupt framework to only bind to a
1120 	 * subset of given CPUs when it performs the shuffle.
1121 	 */
1122 	if (its_init_cpu(dev, sc) != 0)
1123 		panic("gicv3_its_init_secondary: No usable ITS on CPU%d",
1124 		    PCPU_GET(cpuid));
1125 }
1126 #endif
1127 
1128 static uint32_t
1129 its_get_devid(device_t pci_dev)
1130 {
1131 	uintptr_t id;
1132 
1133 	if (pci_get_id(pci_dev, PCI_ID_MSI, &id) != 0)
1134 		panic("%s: %s: Unable to get the MSI DeviceID", __func__,
1135 		    device_get_nameunit(pci_dev));
1136 
1137 	return (id);
1138 }
1139 
1140 static struct its_dev *
1141 its_device_find(device_t dev, device_t child)
1142 {
1143 	struct gicv3_its_softc *sc;
1144 	struct its_dev *its_dev = NULL;
1145 
1146 	sc = device_get_softc(dev);
1147 
1148 	mtx_lock_spin(&sc->sc_its_dev_lock);
1149 	TAILQ_FOREACH(its_dev, &sc->sc_its_dev_list, entry) {
1150 		if (its_dev->pci_dev == child)
1151 			break;
1152 	}
1153 	mtx_unlock_spin(&sc->sc_its_dev_lock);
1154 
1155 	return (its_dev);
1156 }
1157 
1158 static struct its_dev *
1159 its_device_get(device_t dev, device_t child, u_int nvecs)
1160 {
1161 	struct gicv3_its_softc *sc;
1162 	struct its_dev *its_dev;
1163 	vmem_addr_t irq_base;
1164 	size_t esize;
1165 
1166 	sc = device_get_softc(dev);
1167 
1168 	its_dev = its_device_find(dev, child);
1169 	if (its_dev != NULL)
1170 		return (its_dev);
1171 
1172 	its_dev = malloc(sizeof(*its_dev), M_GICV3_ITS, M_NOWAIT | M_ZERO);
1173 	if (its_dev == NULL)
1174 		return (NULL);
1175 
1176 	its_dev->pci_dev = child;
1177 	its_dev->devid = its_get_devid(child);
1178 
1179 	its_dev->lpis.lpi_busy = 0;
1180 	its_dev->lpis.lpi_num = nvecs;
1181 	its_dev->lpis.lpi_free = nvecs;
1182 
1183 	if (vmem_alloc(sc->sc_irq_alloc, nvecs, M_FIRSTFIT | M_NOWAIT,
1184 	    &irq_base) != 0) {
1185 		free(its_dev, M_GICV3_ITS);
1186 		return (NULL);
1187 	}
1188 	its_dev->lpis.lpi_base = irq_base;
1189 
1190 	/* Get ITT entry size */
1191 	esize = GITS_TYPER_ITTES(gic_its_read_8(sc, GITS_TYPER));
1192 
1193 	/*
1194 	 * Allocate ITT for this device.
1195 	 * PA has to be 256 B aligned. At least two entries for device.
1196 	 */
1197 	its_dev->itt_size = roundup2(MAX(nvecs, 2) * esize, 256);
1198 	its_dev->itt = (vm_offset_t)contigmalloc_domainset(its_dev->itt_size,
1199 	    M_GICV3_ITS, sc->sc_ds, M_NOWAIT | M_ZERO, 0,
1200 	    LPI_INT_TRANS_TAB_MAX_ADDR, LPI_INT_TRANS_TAB_ALIGN, 0);
1201 	if (its_dev->itt == 0) {
1202 		vmem_free(sc->sc_irq_alloc, its_dev->lpis.lpi_base, nvecs);
1203 		free(its_dev, M_GICV3_ITS);
1204 		return (NULL);
1205 	}
1206 
1207 	/* Make sure device sees zeroed ITT. */
1208 	if ((sc->sc_its_flags & ITS_FLAGS_CMDQ_FLUSH) != 0)
1209 		cpu_dcache_wb_range(its_dev->itt, its_dev->itt_size);
1210 
1211 	mtx_lock_spin(&sc->sc_its_dev_lock);
1212 	TAILQ_INSERT_TAIL(&sc->sc_its_dev_list, its_dev, entry);
1213 	mtx_unlock_spin(&sc->sc_its_dev_lock);
1214 
1215 	/* Map device to its ITT */
1216 	its_cmd_mapd(dev, its_dev, 1);
1217 
1218 	return (its_dev);
1219 }
1220 
1221 static void
1222 its_device_release(device_t dev, struct its_dev *its_dev)
1223 {
1224 	struct gicv3_its_softc *sc;
1225 
1226 	KASSERT(its_dev->lpis.lpi_busy == 0,
1227 	    ("its_device_release: Trying to release an inuse ITS device"));
1228 
1229 	/* Unmap device in ITS */
1230 	its_cmd_mapd(dev, its_dev, 0);
1231 
1232 	sc = device_get_softc(dev);
1233 
1234 	/* Remove the device from the list of devices */
1235 	mtx_lock_spin(&sc->sc_its_dev_lock);
1236 	TAILQ_REMOVE(&sc->sc_its_dev_list, its_dev, entry);
1237 	mtx_unlock_spin(&sc->sc_its_dev_lock);
1238 
1239 	/* Free ITT */
1240 	KASSERT(its_dev->itt != 0, ("Invalid ITT in valid ITS device"));
1241 	contigfree((void *)its_dev->itt, its_dev->itt_size, M_GICV3_ITS);
1242 
1243 	/* Free the IRQ allocation */
1244 	vmem_free(sc->sc_irq_alloc, its_dev->lpis.lpi_base,
1245 	    its_dev->lpis.lpi_num);
1246 
1247 	free(its_dev, M_GICV3_ITS);
1248 }
1249 
1250 static struct gicv3_its_irqsrc *
1251 gicv3_its_alloc_irqsrc(device_t dev, struct gicv3_its_softc *sc, u_int irq)
1252 {
1253 	struct gicv3_its_irqsrc *girq = NULL;
1254 
1255 	KASSERT(sc->sc_irqs[irq] == NULL,
1256 	    ("%s: Interrupt %u already allocated", __func__, irq));
1257 	mtx_lock_spin(&sc->sc_its_dev_lock);
1258 	if (!TAILQ_EMPTY(&sc->sc_free_irqs)) {
1259 		girq = TAILQ_FIRST(&sc->sc_free_irqs);
1260 		TAILQ_REMOVE(&sc->sc_free_irqs, girq, gi_link);
1261 	}
1262 	mtx_unlock_spin(&sc->sc_its_dev_lock);
1263 	if (girq == NULL) {
1264 		girq = malloc(sizeof(*girq), M_GICV3_ITS,
1265 		    M_NOWAIT | M_ZERO);
1266 		if (girq == NULL)
1267 			return (NULL);
1268 		girq->gi_id = -1;
1269 		if (intr_isrc_register(&girq->gi_isrc, dev, 0,
1270 		    "%s,%u", device_get_nameunit(dev), irq) != 0) {
1271 			free(girq, M_GICV3_ITS);
1272 			return (NULL);
1273 		}
1274 	}
1275 	girq->gi_lpi = irq + sc->sc_irq_base - GIC_FIRST_LPI;
1276 	sc->sc_irqs[irq] = girq;
1277 
1278 	return (girq);
1279 }
1280 
1281 static void
1282 gicv3_its_release_irqsrc(struct gicv3_its_softc *sc,
1283     struct gicv3_its_irqsrc *girq)
1284 {
1285 	u_int irq;
1286 
1287 	mtx_assert(&sc->sc_its_dev_lock, MA_OWNED);
1288 
1289 	irq = girq->gi_lpi + GIC_FIRST_LPI - sc->sc_irq_base;
1290 	sc->sc_irqs[irq] = NULL;
1291 
1292 	girq->gi_id = -1;
1293 	girq->gi_its_dev = NULL;
1294 	TAILQ_INSERT_TAIL(&sc->sc_free_irqs, girq, gi_link);
1295 }
1296 
1297 static int
1298 gicv3_its_alloc_msi(device_t dev, device_t child, int count, int maxcount,
1299     device_t *pic, struct intr_irqsrc **srcs)
1300 {
1301 	struct gicv3_its_softc *sc;
1302 	struct gicv3_its_irqsrc *girq;
1303 	struct its_dev *its_dev;
1304 	u_int irq;
1305 	int i;
1306 
1307 	its_dev = its_device_get(dev, child, count);
1308 	if (its_dev == NULL)
1309 		return (ENXIO);
1310 
1311 	KASSERT(its_dev->lpis.lpi_free >= count,
1312 	    ("gicv3_its_alloc_msi: No free LPIs"));
1313 	sc = device_get_softc(dev);
1314 	irq = its_dev->lpis.lpi_base + its_dev->lpis.lpi_num -
1315 	    its_dev->lpis.lpi_free;
1316 
1317 	/* Allocate the irqsrc for each MSI */
1318 	for (i = 0; i < count; i++, irq++) {
1319 		its_dev->lpis.lpi_free--;
1320 		srcs[i] = (struct intr_irqsrc *)gicv3_its_alloc_irqsrc(dev,
1321 		    sc, irq);
1322 		if (srcs[i] == NULL)
1323 			break;
1324 	}
1325 
1326 	/* The allocation failed, release them */
1327 	if (i != count) {
1328 		mtx_lock_spin(&sc->sc_its_dev_lock);
1329 		for (i = 0; i < count; i++) {
1330 			girq = (struct gicv3_its_irqsrc *)srcs[i];
1331 			if (girq == NULL)
1332 				break;
1333 			gicv3_its_release_irqsrc(sc, girq);
1334 			srcs[i] = NULL;
1335 		}
1336 		mtx_unlock_spin(&sc->sc_its_dev_lock);
1337 		return (ENXIO);
1338 	}
1339 
1340 	/* Finish the allocation now we have all MSI irqsrcs */
1341 	for (i = 0; i < count; i++) {
1342 		girq = (struct gicv3_its_irqsrc *)srcs[i];
1343 		girq->gi_id = i;
1344 		girq->gi_its_dev = its_dev;
1345 
1346 		/* Map the message to the given IRQ */
1347 		gicv3_its_select_cpu(dev, (struct intr_irqsrc *)girq);
1348 		its_cmd_mapti(dev, girq);
1349 	}
1350 	its_dev->lpis.lpi_busy += count;
1351 	*pic = dev;
1352 
1353 	return (0);
1354 }
1355 
1356 static int
1357 gicv3_its_release_msi(device_t dev, device_t child, int count,
1358     struct intr_irqsrc **isrc)
1359 {
1360 	struct gicv3_its_softc *sc;
1361 	struct gicv3_its_irqsrc *girq;
1362 	struct its_dev *its_dev;
1363 	int i;
1364 
1365 	its_dev = its_device_find(dev, child);
1366 
1367 	KASSERT(its_dev != NULL,
1368 	    ("gicv3_its_release_msi: Releasing a MSI interrupt with "
1369 	     "no ITS device"));
1370 	KASSERT(its_dev->lpis.lpi_busy >= count,
1371 	    ("gicv3_its_release_msi: Releasing more interrupts than "
1372 	     "were allocated: releasing %d, allocated %d", count,
1373 	     its_dev->lpis.lpi_busy));
1374 
1375 	sc = device_get_softc(dev);
1376 	mtx_lock_spin(&sc->sc_its_dev_lock);
1377 	for (i = 0; i < count; i++) {
1378 		girq = (struct gicv3_its_irqsrc *)isrc[i];
1379 		gicv3_its_release_irqsrc(sc, girq);
1380 	}
1381 	mtx_unlock_spin(&sc->sc_its_dev_lock);
1382 	its_dev->lpis.lpi_busy -= count;
1383 
1384 	if (its_dev->lpis.lpi_busy == 0)
1385 		its_device_release(dev, its_dev);
1386 
1387 	return (0);
1388 }
1389 
1390 static int
1391 gicv3_its_alloc_msix(device_t dev, device_t child, device_t *pic,
1392     struct intr_irqsrc **isrcp)
1393 {
1394 	struct gicv3_its_softc *sc;
1395 	struct gicv3_its_irqsrc *girq;
1396 	struct its_dev *its_dev;
1397 	u_int nvecs, irq;
1398 
1399 	nvecs = pci_msix_count(child);
1400 	its_dev = its_device_get(dev, child, nvecs);
1401 	if (its_dev == NULL)
1402 		return (ENXIO);
1403 
1404 	KASSERT(its_dev->lpis.lpi_free > 0,
1405 	    ("gicv3_its_alloc_msix: No free LPIs"));
1406 	sc = device_get_softc(dev);
1407 	irq = its_dev->lpis.lpi_base + its_dev->lpis.lpi_num -
1408 	    its_dev->lpis.lpi_free;
1409 
1410 	girq = gicv3_its_alloc_irqsrc(dev, sc, irq);
1411 	if (girq == NULL)
1412 		return (ENXIO);
1413 	girq->gi_id = its_dev->lpis.lpi_busy;
1414 	girq->gi_its_dev = its_dev;
1415 
1416 	its_dev->lpis.lpi_free--;
1417 	its_dev->lpis.lpi_busy++;
1418 
1419 	/* Map the message to the given IRQ */
1420 	gicv3_its_select_cpu(dev, (struct intr_irqsrc *)girq);
1421 	its_cmd_mapti(dev, girq);
1422 
1423 	*pic = dev;
1424 	*isrcp = (struct intr_irqsrc *)girq;
1425 
1426 	return (0);
1427 }
1428 
1429 static int
1430 gicv3_its_release_msix(device_t dev, device_t child, struct intr_irqsrc *isrc)
1431 {
1432 	struct gicv3_its_softc *sc;
1433 	struct gicv3_its_irqsrc *girq;
1434 	struct its_dev *its_dev;
1435 
1436 	its_dev = its_device_find(dev, child);
1437 
1438 	KASSERT(its_dev != NULL,
1439 	    ("gicv3_its_release_msix: Releasing a MSI-X interrupt with "
1440 	     "no ITS device"));
1441 	KASSERT(its_dev->lpis.lpi_busy > 0,
1442 	    ("gicv3_its_release_msix: Releasing more interrupts than "
1443 	     "were allocated: allocated %d", its_dev->lpis.lpi_busy));
1444 
1445 	sc = device_get_softc(dev);
1446 	girq = (struct gicv3_its_irqsrc *)isrc;
1447 	mtx_lock_spin(&sc->sc_its_dev_lock);
1448 	gicv3_its_release_irqsrc(sc, girq);
1449 	mtx_unlock_spin(&sc->sc_its_dev_lock);
1450 	its_dev->lpis.lpi_busy--;
1451 
1452 	if (its_dev->lpis.lpi_busy == 0)
1453 		its_device_release(dev, its_dev);
1454 
1455 	return (0);
1456 }
1457 
1458 static int
1459 gicv3_its_map_msi(device_t dev, device_t child, struct intr_irqsrc *isrc,
1460     uint64_t *addr, uint32_t *data)
1461 {
1462 	struct gicv3_its_softc *sc;
1463 	struct gicv3_its_irqsrc *girq;
1464 
1465 	sc = device_get_softc(dev);
1466 	girq = (struct gicv3_its_irqsrc *)isrc;
1467 
1468 	*addr = vtophys(rman_get_virtual(sc->sc_its_res)) + GITS_TRANSLATER;
1469 	*data = girq->gi_id;
1470 
1471 	return (0);
1472 }
1473 
1474 #ifdef IOMMU
1475 static int
1476 gicv3_iommu_init(device_t dev, device_t child, struct iommu_domain **domain)
1477 {
1478 	struct gicv3_its_softc *sc;
1479 	struct iommu_ctx *ctx;
1480 	int error;
1481 
1482 	sc = device_get_softc(dev);
1483 	ctx = iommu_get_dev_ctx(child);
1484 	if (ctx == NULL)
1485 		return (ENXIO);
1486 	/* Map the page containing the GITS_TRANSLATER register. */
1487 	error = iommu_map_msi(ctx, PAGE_SIZE, 0,
1488 	    IOMMU_MAP_ENTRY_WRITE, IOMMU_MF_CANWAIT, &sc->ma);
1489 	*domain = iommu_get_ctx_domain(ctx);
1490 
1491 	return (error);
1492 }
1493 
1494 static void
1495 gicv3_iommu_deinit(device_t dev, device_t child)
1496 {
1497 	struct iommu_ctx *ctx;
1498 
1499 	ctx = iommu_get_dev_ctx(child);
1500 	if (ctx == NULL)
1501 		return;
1502 
1503 	iommu_unmap_msi(ctx);
1504 }
1505 #endif
1506 
1507 /*
1508  * Commands handling.
1509  */
1510 
1511 static __inline void
1512 cmd_format_command(struct its_cmd *cmd, uint8_t cmd_type)
1513 {
1514 	/* Command field: DW0 [7:0] */
1515 	cmd->cmd_dword[0] &= htole64(~CMD_COMMAND_MASK);
1516 	cmd->cmd_dword[0] |= htole64(cmd_type);
1517 }
1518 
1519 static __inline void
1520 cmd_format_devid(struct its_cmd *cmd, uint32_t devid)
1521 {
1522 	/* Device ID field: DW0 [63:32] */
1523 	cmd->cmd_dword[0] &= htole64(~CMD_DEVID_MASK);
1524 	cmd->cmd_dword[0] |= htole64((uint64_t)devid << CMD_DEVID_SHIFT);
1525 }
1526 
1527 static __inline void
1528 cmd_format_size(struct its_cmd *cmd, uint16_t size)
1529 {
1530 	/* Size field: DW1 [4:0] */
1531 	cmd->cmd_dword[1] &= htole64(~CMD_SIZE_MASK);
1532 	cmd->cmd_dword[1] |= htole64((size & CMD_SIZE_MASK));
1533 }
1534 
1535 static __inline void
1536 cmd_format_id(struct its_cmd *cmd, uint32_t id)
1537 {
1538 	/* ID field: DW1 [31:0] */
1539 	cmd->cmd_dword[1] &= htole64(~CMD_ID_MASK);
1540 	cmd->cmd_dword[1] |= htole64(id);
1541 }
1542 
1543 static __inline void
1544 cmd_format_pid(struct its_cmd *cmd, uint32_t pid)
1545 {
1546 	/* Physical ID field: DW1 [63:32] */
1547 	cmd->cmd_dword[1] &= htole64(~CMD_PID_MASK);
1548 	cmd->cmd_dword[1] |= htole64((uint64_t)pid << CMD_PID_SHIFT);
1549 }
1550 
1551 static __inline void
1552 cmd_format_col(struct its_cmd *cmd, uint16_t col_id)
1553 {
1554 	/* Collection field: DW2 [16:0] */
1555 	cmd->cmd_dword[2] &= htole64(~CMD_COL_MASK);
1556 	cmd->cmd_dword[2] |= htole64(col_id);
1557 }
1558 
1559 static __inline void
1560 cmd_format_target(struct its_cmd *cmd, uint64_t target)
1561 {
1562 	/* Target Address field: DW2 [47:16] */
1563 	cmd->cmd_dword[2] &= htole64(~CMD_TARGET_MASK);
1564 	cmd->cmd_dword[2] |= htole64(target & CMD_TARGET_MASK);
1565 }
1566 
1567 static __inline void
1568 cmd_format_itt(struct its_cmd *cmd, uint64_t itt)
1569 {
1570 	/* ITT Address field: DW2 [47:8] */
1571 	cmd->cmd_dword[2] &= htole64(~CMD_ITT_MASK);
1572 	cmd->cmd_dword[2] |= htole64(itt & CMD_ITT_MASK);
1573 }
1574 
1575 static __inline void
1576 cmd_format_valid(struct its_cmd *cmd, uint8_t valid)
1577 {
1578 	/* Valid field: DW2 [63] */
1579 	cmd->cmd_dword[2] &= htole64(~CMD_VALID_MASK);
1580 	cmd->cmd_dword[2] |= htole64((uint64_t)valid << CMD_VALID_SHIFT);
1581 }
1582 
1583 static inline bool
1584 its_cmd_queue_full(struct gicv3_its_softc *sc)
1585 {
1586 	size_t read_idx, next_write_idx;
1587 
1588 	/* Get the index of the next command */
1589 	next_write_idx = (sc->sc_its_cmd_next_idx + 1) %
1590 	    (ITS_CMDQ_SIZE / sizeof(struct its_cmd));
1591 	/* And the index of the current command being read */
1592 	read_idx = gic_its_read_4(sc, GITS_CREADR) / sizeof(struct its_cmd);
1593 
1594 	/*
1595 	 * The queue is full when the write offset points
1596 	 * at the command before the current read offset.
1597 	 */
1598 	return (next_write_idx == read_idx);
1599 }
1600 
1601 static inline void
1602 its_cmd_sync(struct gicv3_its_softc *sc, struct its_cmd *cmd)
1603 {
1604 
1605 	if ((sc->sc_its_flags & ITS_FLAGS_CMDQ_FLUSH) != 0) {
1606 		/* Clean D-cache under command. */
1607 		cpu_dcache_wb_range((vm_offset_t)cmd, sizeof(*cmd));
1608 	} else {
1609 		/* DSB inner shareable, store */
1610 		dsb(ishst);
1611 	}
1612 
1613 }
1614 
1615 static inline uint64_t
1616 its_cmd_cwriter_offset(struct gicv3_its_softc *sc, struct its_cmd *cmd)
1617 {
1618 	uint64_t off;
1619 
1620 	off = (cmd - sc->sc_its_cmd_base) * sizeof(*cmd);
1621 
1622 	return (off);
1623 }
1624 
1625 static void
1626 its_cmd_wait_completion(device_t dev, struct its_cmd *cmd_first,
1627     struct its_cmd *cmd_last)
1628 {
1629 	struct gicv3_its_softc *sc;
1630 	uint64_t first, last, read;
1631 	size_t us_left;
1632 
1633 	sc = device_get_softc(dev);
1634 
1635 	/*
1636 	 * XXX ARM64TODO: This is obviously a significant delay.
1637 	 * The reason for that is that currently the time frames for
1638 	 * the command to complete are not known.
1639 	 */
1640 	us_left = 1000000;
1641 
1642 	first = its_cmd_cwriter_offset(sc, cmd_first);
1643 	last = its_cmd_cwriter_offset(sc, cmd_last);
1644 
1645 	for (;;) {
1646 		read = gic_its_read_8(sc, GITS_CREADR);
1647 		if (first < last) {
1648 			if (read < first || read >= last)
1649 				break;
1650 		} else if (read < first && read >= last)
1651 			break;
1652 
1653 		if (us_left-- == 0) {
1654 			/* This means timeout */
1655 			device_printf(dev,
1656 			    "Timeout while waiting for CMD completion.\n");
1657 			return;
1658 		}
1659 		DELAY(1);
1660 	}
1661 }
1662 
1663 static struct its_cmd *
1664 its_cmd_alloc_locked(device_t dev)
1665 {
1666 	struct gicv3_its_softc *sc;
1667 	struct its_cmd *cmd;
1668 	size_t us_left;
1669 
1670 	sc = device_get_softc(dev);
1671 
1672 	/*
1673 	 * XXX ARM64TODO: This is obviously a significant delay.
1674 	 * The reason for that is that currently the time frames for
1675 	 * the command to complete (and therefore free the descriptor)
1676 	 * are not known.
1677 	 */
1678 	us_left = 1000000;
1679 
1680 	mtx_assert(&sc->sc_its_cmd_lock, MA_OWNED);
1681 	while (its_cmd_queue_full(sc)) {
1682 		if (us_left-- == 0) {
1683 			/* Timeout while waiting for free command */
1684 			device_printf(dev,
1685 			    "Timeout while waiting for free command\n");
1686 			return (NULL);
1687 		}
1688 		DELAY(1);
1689 	}
1690 
1691 	cmd = &sc->sc_its_cmd_base[sc->sc_its_cmd_next_idx];
1692 	sc->sc_its_cmd_next_idx++;
1693 	sc->sc_its_cmd_next_idx %= ITS_CMDQ_SIZE / sizeof(struct its_cmd);
1694 
1695 	return (cmd);
1696 }
1697 
1698 static uint64_t
1699 its_cmd_prepare(struct its_cmd *cmd, struct its_cmd_desc *desc)
1700 {
1701 	uint64_t target;
1702 	uint8_t cmd_type;
1703 	u_int size;
1704 
1705 	cmd_type = desc->cmd_type;
1706 	target = ITS_TARGET_NONE;
1707 
1708 	switch (cmd_type) {
1709 	case ITS_CMD_MOVI:	/* Move interrupt ID to another collection */
1710 		target = desc->cmd_desc_movi.col->col_target;
1711 		cmd_format_command(cmd, ITS_CMD_MOVI);
1712 		cmd_format_id(cmd, desc->cmd_desc_movi.id);
1713 		cmd_format_col(cmd, desc->cmd_desc_movi.col->col_id);
1714 		cmd_format_devid(cmd, desc->cmd_desc_movi.its_dev->devid);
1715 		break;
1716 	case ITS_CMD_SYNC:	/* Wait for previous commands completion */
1717 		target = desc->cmd_desc_sync.col->col_target;
1718 		cmd_format_command(cmd, ITS_CMD_SYNC);
1719 		cmd_format_target(cmd, target);
1720 		break;
1721 	case ITS_CMD_MAPD:	/* Assign ITT to device */
1722 		cmd_format_command(cmd, ITS_CMD_MAPD);
1723 		cmd_format_itt(cmd, vtophys(desc->cmd_desc_mapd.its_dev->itt));
1724 		/*
1725 		 * Size describes number of bits to encode interrupt IDs
1726 		 * supported by the device minus one.
1727 		 * When V (valid) bit is zero, this field should be written
1728 		 * as zero.
1729 		 */
1730 		if (desc->cmd_desc_mapd.valid != 0) {
1731 			size = fls(desc->cmd_desc_mapd.its_dev->lpis.lpi_num);
1732 			size = MAX(1, size) - 1;
1733 		} else
1734 			size = 0;
1735 
1736 		cmd_format_size(cmd, size);
1737 		cmd_format_devid(cmd, desc->cmd_desc_mapd.its_dev->devid);
1738 		cmd_format_valid(cmd, desc->cmd_desc_mapd.valid);
1739 		break;
1740 	case ITS_CMD_MAPC:	/* Map collection to Re-Distributor */
1741 		target = desc->cmd_desc_mapc.col->col_target;
1742 		cmd_format_command(cmd, ITS_CMD_MAPC);
1743 		cmd_format_col(cmd, desc->cmd_desc_mapc.col->col_id);
1744 		cmd_format_valid(cmd, desc->cmd_desc_mapc.valid);
1745 		cmd_format_target(cmd, target);
1746 		break;
1747 	case ITS_CMD_MAPTI:
1748 		target = desc->cmd_desc_mapvi.col->col_target;
1749 		cmd_format_command(cmd, ITS_CMD_MAPTI);
1750 		cmd_format_devid(cmd, desc->cmd_desc_mapvi.its_dev->devid);
1751 		cmd_format_id(cmd, desc->cmd_desc_mapvi.id);
1752 		cmd_format_pid(cmd, desc->cmd_desc_mapvi.pid);
1753 		cmd_format_col(cmd, desc->cmd_desc_mapvi.col->col_id);
1754 		break;
1755 	case ITS_CMD_MAPI:
1756 		target = desc->cmd_desc_mapi.col->col_target;
1757 		cmd_format_command(cmd, ITS_CMD_MAPI);
1758 		cmd_format_devid(cmd, desc->cmd_desc_mapi.its_dev->devid);
1759 		cmd_format_id(cmd, desc->cmd_desc_mapi.pid);
1760 		cmd_format_col(cmd, desc->cmd_desc_mapi.col->col_id);
1761 		break;
1762 	case ITS_CMD_INV:
1763 		target = desc->cmd_desc_inv.col->col_target;
1764 		cmd_format_command(cmd, ITS_CMD_INV);
1765 		cmd_format_devid(cmd, desc->cmd_desc_inv.its_dev->devid);
1766 		cmd_format_id(cmd, desc->cmd_desc_inv.pid);
1767 		break;
1768 	case ITS_CMD_INVALL:
1769 		cmd_format_command(cmd, ITS_CMD_INVALL);
1770 		cmd_format_col(cmd, desc->cmd_desc_invall.col->col_id);
1771 		break;
1772 	default:
1773 		panic("its_cmd_prepare: Invalid command: %x", cmd_type);
1774 	}
1775 
1776 	return (target);
1777 }
1778 
1779 static int
1780 its_cmd_send(device_t dev, struct its_cmd_desc *desc)
1781 {
1782 	struct gicv3_its_softc *sc;
1783 	struct its_cmd *cmd, *cmd_sync, *cmd_write;
1784 	struct its_col col_sync;
1785 	struct its_cmd_desc desc_sync;
1786 	uint64_t target, cwriter;
1787 
1788 	sc = device_get_softc(dev);
1789 	mtx_lock_spin(&sc->sc_its_cmd_lock);
1790 	cmd = its_cmd_alloc_locked(dev);
1791 	if (cmd == NULL) {
1792 		device_printf(dev, "could not allocate ITS command\n");
1793 		mtx_unlock_spin(&sc->sc_its_cmd_lock);
1794 		return (EBUSY);
1795 	}
1796 
1797 	target = its_cmd_prepare(cmd, desc);
1798 	its_cmd_sync(sc, cmd);
1799 
1800 	if (target != ITS_TARGET_NONE) {
1801 		cmd_sync = its_cmd_alloc_locked(dev);
1802 		if (cmd_sync != NULL) {
1803 			desc_sync.cmd_type = ITS_CMD_SYNC;
1804 			col_sync.col_target = target;
1805 			desc_sync.cmd_desc_sync.col = &col_sync;
1806 			its_cmd_prepare(cmd_sync, &desc_sync);
1807 			its_cmd_sync(sc, cmd_sync);
1808 		}
1809 	}
1810 
1811 	/* Update GITS_CWRITER */
1812 	cwriter = sc->sc_its_cmd_next_idx * sizeof(struct its_cmd);
1813 	gic_its_write_8(sc, GITS_CWRITER, cwriter);
1814 	cmd_write = &sc->sc_its_cmd_base[sc->sc_its_cmd_next_idx];
1815 	mtx_unlock_spin(&sc->sc_its_cmd_lock);
1816 
1817 	its_cmd_wait_completion(dev, cmd, cmd_write);
1818 
1819 	return (0);
1820 }
1821 
1822 /* Handlers to send commands */
1823 static void
1824 its_cmd_movi(device_t dev, struct gicv3_its_irqsrc *girq)
1825 {
1826 	struct gicv3_its_softc *sc;
1827 	struct its_cmd_desc desc;
1828 	struct its_col *col;
1829 
1830 	sc = device_get_softc(dev);
1831 	col = sc->sc_its_cols[CPU_FFS(&girq->gi_isrc.isrc_cpu) - 1];
1832 
1833 	desc.cmd_type = ITS_CMD_MOVI;
1834 	desc.cmd_desc_movi.its_dev = girq->gi_its_dev;
1835 	desc.cmd_desc_movi.col = col;
1836 	desc.cmd_desc_movi.id = girq->gi_id;
1837 
1838 	its_cmd_send(dev, &desc);
1839 }
1840 
1841 static void
1842 its_cmd_mapc(device_t dev, struct its_col *col, uint8_t valid)
1843 {
1844 	struct its_cmd_desc desc;
1845 
1846 	desc.cmd_type = ITS_CMD_MAPC;
1847 	desc.cmd_desc_mapc.col = col;
1848 	/*
1849 	 * Valid bit set - map the collection.
1850 	 * Valid bit cleared - unmap the collection.
1851 	 */
1852 	desc.cmd_desc_mapc.valid = valid;
1853 
1854 	its_cmd_send(dev, &desc);
1855 }
1856 
1857 static void
1858 its_cmd_mapti(device_t dev, struct gicv3_its_irqsrc *girq)
1859 {
1860 	struct gicv3_its_softc *sc;
1861 	struct its_cmd_desc desc;
1862 	struct its_col *col;
1863 	u_int col_id;
1864 
1865 	sc = device_get_softc(dev);
1866 
1867 	col_id = CPU_FFS(&girq->gi_isrc.isrc_cpu) - 1;
1868 	col = sc->sc_its_cols[col_id];
1869 
1870 	desc.cmd_type = ITS_CMD_MAPTI;
1871 	desc.cmd_desc_mapvi.its_dev = girq->gi_its_dev;
1872 	desc.cmd_desc_mapvi.col = col;
1873 	/* The EventID sent to the device */
1874 	desc.cmd_desc_mapvi.id = girq->gi_id;
1875 	/* The physical interrupt presented to softeware */
1876 	desc.cmd_desc_mapvi.pid = girq->gi_lpi + GIC_FIRST_LPI;
1877 
1878 	its_cmd_send(dev, &desc);
1879 }
1880 
1881 static void
1882 its_cmd_mapd(device_t dev, struct its_dev *its_dev, uint8_t valid)
1883 {
1884 	struct its_cmd_desc desc;
1885 
1886 	desc.cmd_type = ITS_CMD_MAPD;
1887 	desc.cmd_desc_mapd.its_dev = its_dev;
1888 	desc.cmd_desc_mapd.valid = valid;
1889 
1890 	its_cmd_send(dev, &desc);
1891 }
1892 
1893 static void
1894 its_cmd_inv(device_t dev, struct its_dev *its_dev,
1895     struct gicv3_its_irqsrc *girq)
1896 {
1897 	struct gicv3_its_softc *sc;
1898 	struct its_cmd_desc desc;
1899 	struct its_col *col;
1900 
1901 	sc = device_get_softc(dev);
1902 	col = sc->sc_its_cols[CPU_FFS(&girq->gi_isrc.isrc_cpu) - 1];
1903 
1904 	desc.cmd_type = ITS_CMD_INV;
1905 	/* The EventID sent to the device */
1906 	desc.cmd_desc_inv.pid = girq->gi_id;
1907 	desc.cmd_desc_inv.its_dev = its_dev;
1908 	desc.cmd_desc_inv.col = col;
1909 
1910 	its_cmd_send(dev, &desc);
1911 }
1912 
1913 static void
1914 its_cmd_invall(device_t dev, struct its_col *col)
1915 {
1916 	struct its_cmd_desc desc;
1917 
1918 	desc.cmd_type = ITS_CMD_INVALL;
1919 	desc.cmd_desc_invall.col = col;
1920 
1921 	its_cmd_send(dev, &desc);
1922 }
1923 
1924 #ifdef FDT
1925 static device_probe_t gicv3_its_fdt_probe;
1926 static device_attach_t gicv3_its_fdt_attach;
1927 
1928 static device_method_t gicv3_its_fdt_methods[] = {
1929 	/* Device interface */
1930 	DEVMETHOD(device_probe,		gicv3_its_fdt_probe),
1931 	DEVMETHOD(device_attach,	gicv3_its_fdt_attach),
1932 
1933 	/* End */
1934 	DEVMETHOD_END
1935 };
1936 
1937 #define its_baseclasses its_fdt_baseclasses
1938 DEFINE_CLASS_1(its, gicv3_its_fdt_driver, gicv3_its_fdt_methods,
1939     sizeof(struct gicv3_its_softc), gicv3_its_driver);
1940 #undef its_baseclasses
1941 
1942 EARLY_DRIVER_MODULE(its_fdt, gic, gicv3_its_fdt_driver, 0, 0,
1943     BUS_PASS_INTERRUPT + BUS_PASS_ORDER_MIDDLE);
1944 
1945 static int
1946 gicv3_its_fdt_probe(device_t dev)
1947 {
1948 
1949 	if (!ofw_bus_status_okay(dev))
1950 		return (ENXIO);
1951 
1952 	if (!ofw_bus_is_compatible(dev, "arm,gic-v3-its"))
1953 		return (ENXIO);
1954 
1955 	device_set_desc(dev, "ARM GIC Interrupt Translation Service");
1956 	return (BUS_PROBE_DEFAULT);
1957 }
1958 
1959 static int
1960 gicv3_its_fdt_attach(device_t dev)
1961 {
1962 	struct gicv3_its_softc *sc;
1963 	phandle_t xref;
1964 	int err;
1965 
1966 	sc = device_get_softc(dev);
1967 	sc->dev = dev;
1968 	err = gicv3_its_attach(dev);
1969 	if (err != 0)
1970 		return (err);
1971 
1972 	/* Register this device as a interrupt controller */
1973 	xref = OF_xref_from_node(ofw_bus_get_node(dev));
1974 	sc->sc_pic = intr_pic_register(dev, xref);
1975 	err = intr_pic_add_handler(device_get_parent(dev), sc->sc_pic,
1976 	    gicv3_its_intr, sc, sc->sc_irq_base, sc->sc_irq_length);
1977 	if (err != 0) {
1978 		device_printf(dev, "Failed to add PIC handler: %d\n", err);
1979 		return (err);
1980 	}
1981 
1982 	/* Register this device to handle MSI interrupts */
1983 	err = intr_msi_register(dev, xref);
1984 	if (err != 0) {
1985 		device_printf(dev, "Failed to register for MSIs: %d\n", err);
1986 		return (err);
1987 	}
1988 
1989 	return (0);
1990 }
1991 #endif
1992 
1993 #ifdef DEV_ACPI
1994 static device_probe_t gicv3_its_acpi_probe;
1995 static device_attach_t gicv3_its_acpi_attach;
1996 
1997 static device_method_t gicv3_its_acpi_methods[] = {
1998 	/* Device interface */
1999 	DEVMETHOD(device_probe,		gicv3_its_acpi_probe),
2000 	DEVMETHOD(device_attach,	gicv3_its_acpi_attach),
2001 
2002 	/* End */
2003 	DEVMETHOD_END
2004 };
2005 
2006 #define its_baseclasses its_acpi_baseclasses
2007 DEFINE_CLASS_1(its, gicv3_its_acpi_driver, gicv3_its_acpi_methods,
2008     sizeof(struct gicv3_its_softc), gicv3_its_driver);
2009 #undef its_baseclasses
2010 
2011 EARLY_DRIVER_MODULE(its_acpi, gic, gicv3_its_acpi_driver, 0, 0,
2012     BUS_PASS_INTERRUPT + BUS_PASS_ORDER_MIDDLE);
2013 
2014 static int
2015 gicv3_its_acpi_probe(device_t dev)
2016 {
2017 
2018 	if (gic_get_bus(dev) != GIC_BUS_ACPI)
2019 		return (EINVAL);
2020 
2021 	if (gic_get_hw_rev(dev) < 3)
2022 		return (EINVAL);
2023 
2024 	device_set_desc(dev, "ARM GIC Interrupt Translation Service");
2025 	return (BUS_PROBE_DEFAULT);
2026 }
2027 
2028 static int
2029 gicv3_its_acpi_attach(device_t dev)
2030 {
2031 	struct gicv3_its_softc *sc;
2032 	struct gic_v3_devinfo *di;
2033 	int err;
2034 
2035 	sc = device_get_softc(dev);
2036 	sc->dev = dev;
2037 	err = gicv3_its_attach(dev);
2038 	if (err != 0)
2039 		return (err);
2040 
2041 	di = device_get_ivars(dev);
2042 	sc->sc_pic = intr_pic_register(dev, di->msi_xref);
2043 	err = intr_pic_add_handler(device_get_parent(dev), sc->sc_pic,
2044 	    gicv3_its_intr, sc, sc->sc_irq_base, sc->sc_irq_length);
2045 	if (err != 0) {
2046 		device_printf(dev, "Failed to add PIC handler: %d\n", err);
2047 		return (err);
2048 	}
2049 
2050 	/* Register this device to handle MSI interrupts */
2051 	err = intr_msi_register(dev, di->msi_xref);
2052 	if (err != 0) {
2053 		device_printf(dev, "Failed to register for MSIs: %d\n", err);
2054 		return (err);
2055 	}
2056 
2057 	return (0);
2058 }
2059 #endif
2060