ntb_hw.c revision 291031
1/*-
2 * Copyright (C) 2013 Intel Corporation
3 * Copyright (C) 2015 EMC Corporation
4 * All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 *    notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 *    notice, this list of conditions and the following disclaimer in the
13 *    documentation and/or other materials provided with the distribution.
14 *
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25 * SUCH DAMAGE.
26 */
27
28#include <sys/cdefs.h>
29__FBSDID("$FreeBSD: head/sys/dev/ntb/ntb_hw/ntb_hw.c 291031 2015-11-18 22:20:21Z cem $");
30
31#include <sys/param.h>
32#include <sys/kernel.h>
33#include <sys/systm.h>
34#include <sys/bus.h>
35#include <sys/endian.h>
36#include <sys/malloc.h>
37#include <sys/module.h>
38#include <sys/queue.h>
39#include <sys/rman.h>
40#include <sys/sbuf.h>
41#include <sys/sysctl.h>
42#include <vm/vm.h>
43#include <vm/pmap.h>
44#include <machine/bus.h>
45#include <machine/pmap.h>
46#include <machine/resource.h>
47#include <dev/pci/pcireg.h>
48#include <dev/pci/pcivar.h>
49
50#include "ntb_regs.h"
51#include "ntb_hw.h"
52
53/*
54 * The Non-Transparent Bridge (NTB) is a device on some Intel processors that
55 * allows you to connect two systems using a PCI-e link.
56 *
57 * This module contains the hardware abstraction layer for the NTB. It allows
58 * you to send and recieve interrupts, map the memory windows and send and
59 * receive messages in the scratch-pad registers.
60 *
61 * NOTE: Much of the code in this module is shared with Linux. Any patches may
62 * be picked up and redistributed in Linux with a dual GPL/BSD license.
63 */
64
65#define MAX_MSIX_INTERRUPTS MAX(XEON_DB_COUNT, ATOM_DB_COUNT)
66
67#define NTB_HB_TIMEOUT		1 /* second */
68#define ATOM_LINK_RECOVERY_TIME	500 /* ms */
69
70#define DEVICE2SOFTC(dev) ((struct ntb_softc *) device_get_softc(dev))
71
72enum ntb_device_type {
73	NTB_XEON,
74	NTB_ATOM
75};
76
77/* ntb_conn_type are hardware numbers, cannot change. */
78enum ntb_conn_type {
79	NTB_CONN_TRANSPARENT = 0,
80	NTB_CONN_B2B = 1,
81	NTB_CONN_RP = 2,
82};
83
84enum ntb_b2b_direction {
85	NTB_DEV_USD = 0,
86	NTB_DEV_DSD = 1,
87};
88
89enum ntb_bar {
90	NTB_CONFIG_BAR = 0,
91	NTB_B2B_BAR_1,
92	NTB_B2B_BAR_2,
93	NTB_B2B_BAR_3,
94	NTB_MAX_BARS
95};
96
97/* Device features and workarounds */
98#define HAS_FEATURE(feature)	\
99	((ntb->features & (feature)) != 0)
100
101struct ntb_hw_info {
102	uint32_t		device_id;
103	const char		*desc;
104	enum ntb_device_type	type;
105	uint32_t		features;
106};
107
108struct ntb_pci_bar_info {
109	bus_space_tag_t		pci_bus_tag;
110	bus_space_handle_t	pci_bus_handle;
111	int			pci_resource_id;
112	struct resource		*pci_resource;
113	vm_paddr_t		pbase;
114	caddr_t			vbase;
115	vm_size_t		size;
116	bool			mapped_wc : 1;
117
118	/* Configuration register offsets */
119	uint32_t		psz_off;
120	uint32_t		ssz_off;
121	uint32_t		pbarxlat_off;
122};
123
124struct ntb_int_info {
125	struct resource	*res;
126	int		rid;
127	void		*tag;
128};
129
130struct ntb_vec {
131	struct ntb_softc	*ntb;
132	uint32_t		num;
133};
134
135struct ntb_reg {
136	uint32_t	ntb_ctl;
137	uint32_t	lnk_sta;
138	uint8_t		db_size;
139	unsigned	mw_bar[NTB_MAX_BARS];
140};
141
142struct ntb_alt_reg {
143	uint32_t	db_bell;
144	uint32_t	db_mask;
145	uint32_t	spad;
146};
147
148struct ntb_xlat_reg {
149	uint32_t	bar0_base;
150	uint32_t	bar2_base;
151	uint32_t	bar4_base;
152	uint32_t	bar5_base;
153
154	uint32_t	bar2_xlat;
155	uint32_t	bar4_xlat;
156	uint32_t	bar5_xlat;
157
158	uint32_t	bar2_limit;
159	uint32_t	bar4_limit;
160	uint32_t	bar5_limit;
161};
162
163struct ntb_b2b_addr {
164	uint64_t	bar0_addr;
165	uint64_t	bar2_addr64;
166	uint64_t	bar4_addr64;
167	uint64_t	bar4_addr32;
168	uint64_t	bar5_addr32;
169};
170
171struct ntb_softc {
172	device_t		device;
173	enum ntb_device_type	type;
174	uint32_t		features;
175
176	struct ntb_pci_bar_info	bar_info[NTB_MAX_BARS];
177	struct ntb_int_info	int_info[MAX_MSIX_INTERRUPTS];
178	uint32_t		allocated_interrupts;
179
180	struct callout		heartbeat_timer;
181	struct callout		lr_timer;
182
183	void			*ntb_ctx;
184	const struct ntb_ctx_ops *ctx_ops;
185	struct ntb_vec		*msix_vec;
186#define CTX_LOCK(sc)		mtx_lock(&(sc)->ctx_lock)
187#define CTX_UNLOCK(sc)		mtx_unlock(&(sc)->ctx_lock)
188#define CTX_ASSERT(sc,f)	mtx_assert(&(sc)->ctx_lock, (f))
189	struct mtx		ctx_lock;
190
191	uint32_t		ppd;
192	enum ntb_conn_type	conn_type;
193	enum ntb_b2b_direction	dev_type;
194
195	/* Offset of peer bar0 in B2B BAR */
196	uint64_t			b2b_off;
197	/* Memory window used to access peer bar0 */
198#define B2B_MW_DISABLED			UINT8_MAX
199	uint8_t				b2b_mw_idx;
200
201	uint8_t				mw_count;
202	uint8_t				spad_count;
203	uint8_t				db_count;
204	uint8_t				db_vec_count;
205	uint8_t				db_vec_shift;
206
207	/* Protects local db_mask. */
208#define DB_MASK_LOCK(sc)	mtx_lock_spin(&(sc)->db_mask_lock)
209#define DB_MASK_UNLOCK(sc)	mtx_unlock_spin(&(sc)->db_mask_lock)
210#define DB_MASK_ASSERT(sc,f)	mtx_assert(&(sc)->db_mask_lock, (f))
211	struct mtx			db_mask_lock;
212
213	volatile uint32_t		ntb_ctl;
214	volatile uint32_t		lnk_sta;
215
216	uint64_t			db_valid_mask;
217	uint64_t			db_link_mask;
218	uint64_t			db_mask;
219
220	int				last_ts;	/* ticks @ last irq */
221
222	const struct ntb_reg		*reg;
223	const struct ntb_alt_reg	*self_reg;
224	const struct ntb_alt_reg	*peer_reg;
225	const struct ntb_xlat_reg	*xlat_reg;
226};
227
228#ifdef __i386__
229static __inline uint64_t
230bus_space_read_8(bus_space_tag_t tag, bus_space_handle_t handle,
231    bus_size_t offset)
232{
233
234	return (bus_space_read_4(tag, handle, offset) |
235	    ((uint64_t)bus_space_read_4(tag, handle, offset + 4)) << 32);
236}
237
238static __inline void
239bus_space_write_8(bus_space_tag_t tag, bus_space_handle_t handle,
240    bus_size_t offset, uint64_t val)
241{
242
243	bus_space_write_4(tag, handle, offset, val);
244	bus_space_write_4(tag, handle, offset + 4, val >> 32);
245}
246#endif
247
248#define ntb_bar_read(SIZE, bar, offset) \
249	    bus_space_read_ ## SIZE (ntb->bar_info[(bar)].pci_bus_tag, \
250	    ntb->bar_info[(bar)].pci_bus_handle, (offset))
251#define ntb_bar_write(SIZE, bar, offset, val) \
252	    bus_space_write_ ## SIZE (ntb->bar_info[(bar)].pci_bus_tag, \
253	    ntb->bar_info[(bar)].pci_bus_handle, (offset), (val))
254#define ntb_reg_read(SIZE, offset) ntb_bar_read(SIZE, NTB_CONFIG_BAR, offset)
255#define ntb_reg_write(SIZE, offset, val) \
256	    ntb_bar_write(SIZE, NTB_CONFIG_BAR, offset, val)
257#define ntb_mw_read(SIZE, offset) \
258	    ntb_bar_read(SIZE, ntb_mw_to_bar(ntb, ntb->b2b_mw_idx), offset)
259#define ntb_mw_write(SIZE, offset, val) \
260	    ntb_bar_write(SIZE, ntb_mw_to_bar(ntb, ntb->b2b_mw_idx), \
261		offset, val)
262
263static int ntb_probe(device_t device);
264static int ntb_attach(device_t device);
265static int ntb_detach(device_t device);
266static inline enum ntb_bar ntb_mw_to_bar(struct ntb_softc *, unsigned mw);
267static inline bool bar_is_64bit(struct ntb_softc *, enum ntb_bar);
268static inline void bar_get_xlat_params(struct ntb_softc *, enum ntb_bar,
269    uint32_t *base, uint32_t *xlat, uint32_t *lmt);
270static int ntb_map_pci_bars(struct ntb_softc *ntb);
271static void print_map_success(struct ntb_softc *, struct ntb_pci_bar_info *,
272    const char *);
273static int map_mmr_bar(struct ntb_softc *ntb, struct ntb_pci_bar_info *bar);
274static int map_memory_window_bar(struct ntb_softc *ntb,
275    struct ntb_pci_bar_info *bar);
276static void ntb_unmap_pci_bar(struct ntb_softc *ntb);
277static int ntb_remap_msix(device_t, uint32_t desired, uint32_t avail);
278static int ntb_init_isr(struct ntb_softc *ntb);
279static int ntb_setup_legacy_interrupt(struct ntb_softc *ntb);
280static int ntb_setup_msix(struct ntb_softc *ntb, uint32_t num_vectors);
281static void ntb_teardown_interrupts(struct ntb_softc *ntb);
282static inline uint64_t ntb_vec_mask(struct ntb_softc *, uint64_t db_vector);
283static void ntb_interrupt(struct ntb_softc *, uint32_t vec);
284static void ndev_vec_isr(void *arg);
285static void ndev_irq_isr(void *arg);
286static inline uint64_t db_ioread(struct ntb_softc *, uint64_t regoff);
287static inline void db_iowrite(struct ntb_softc *, uint64_t regoff, uint64_t);
288static inline void db_iowrite_raw(struct ntb_softc *, uint64_t regoff, uint64_t);
289static int ntb_create_msix_vec(struct ntb_softc *ntb, uint32_t num_vectors);
290static void ntb_free_msix_vec(struct ntb_softc *ntb);
291static struct ntb_hw_info *ntb_get_device_info(uint32_t device_id);
292static void ntb_detect_max_mw(struct ntb_softc *ntb);
293static int ntb_detect_xeon(struct ntb_softc *ntb);
294static int ntb_detect_atom(struct ntb_softc *ntb);
295static int ntb_xeon_init_dev(struct ntb_softc *ntb);
296static int ntb_atom_init_dev(struct ntb_softc *ntb);
297static void ntb_teardown_xeon(struct ntb_softc *ntb);
298static void configure_atom_secondary_side_bars(struct ntb_softc *ntb);
299static void xeon_reset_sbar_size(struct ntb_softc *, enum ntb_bar idx,
300    enum ntb_bar regbar);
301static void xeon_set_sbar_base_and_limit(struct ntb_softc *,
302    uint64_t base_addr, enum ntb_bar idx, enum ntb_bar regbar);
303static void xeon_set_pbar_xlat(struct ntb_softc *, uint64_t base_addr,
304    enum ntb_bar idx);
305static int xeon_setup_b2b_mw(struct ntb_softc *,
306    const struct ntb_b2b_addr *addr, const struct ntb_b2b_addr *peer_addr);
307static inline bool link_is_up(struct ntb_softc *ntb);
308static inline bool atom_link_is_err(struct ntb_softc *ntb);
309static inline enum ntb_speed ntb_link_sta_speed(struct ntb_softc *);
310static inline enum ntb_width ntb_link_sta_width(struct ntb_softc *);
311static void atom_link_hb(void *arg);
312static void ntb_db_event(struct ntb_softc *ntb, uint32_t vec);
313static void recover_atom_link(void *arg);
314static bool ntb_poll_link(struct ntb_softc *ntb);
315static void save_bar_parameters(struct ntb_pci_bar_info *bar);
316static void ntb_sysctl_init(struct ntb_softc *);
317static int sysctl_handle_features(SYSCTL_HANDLER_ARGS);
318static int sysctl_handle_link_status(SYSCTL_HANDLER_ARGS);
319static int sysctl_handle_register(SYSCTL_HANDLER_ARGS);
320
321static unsigned g_ntb_hw_debug_level;
322SYSCTL_UINT(_hw_ntb, OID_AUTO, debug_level, CTLFLAG_RWTUN,
323    &g_ntb_hw_debug_level, 0, "ntb_hw log level -- higher is more verbose");
324#define ntb_printf(lvl, ...) do {				\
325	if ((lvl) <= g_ntb_hw_debug_level) {			\
326		device_printf(ntb->device, __VA_ARGS__);	\
327	}							\
328} while (0)
329
330static unsigned g_ntb_enable_wc = 1;
331SYSCTL_UINT(_hw_ntb, OID_AUTO, enable_writecombine, CTLFLAG_RDTUN,
332    &g_ntb_enable_wc, 0, "Set to 1 to map memory windows write combining");
333
334static struct ntb_hw_info pci_ids[] = {
335	/* XXX: PS/SS IDs left out until they are supported. */
336	{ 0x0C4E8086, "BWD Atom Processor S1200 Non-Transparent Bridge B2B",
337		NTB_ATOM, 0 },
338
339	{ 0x37258086, "JSF Xeon C35xx/C55xx Non-Transparent Bridge B2B",
340		NTB_XEON, NTB_SDOORBELL_LOCKUP | NTB_B2BDOORBELL_BIT14 },
341	{ 0x3C0D8086, "SNB Xeon E5/Core i7 Non-Transparent Bridge B2B",
342		NTB_XEON, NTB_SDOORBELL_LOCKUP | NTB_B2BDOORBELL_BIT14 },
343	{ 0x0E0D8086, "IVT Xeon E5 V2 Non-Transparent Bridge B2B", NTB_XEON,
344		NTB_SDOORBELL_LOCKUP | NTB_B2BDOORBELL_BIT14 |
345		    NTB_SB01BASE_LOCKUP | NTB_BAR_SIZE_4K },
346	{ 0x2F0D8086, "HSX Xeon E5 V3 Non-Transparent Bridge B2B", NTB_XEON,
347		NTB_SDOORBELL_LOCKUP | NTB_B2BDOORBELL_BIT14 |
348		    NTB_SB01BASE_LOCKUP },
349	{ 0x6F0D8086, "BDX Xeon E5 V4 Non-Transparent Bridge B2B", NTB_XEON,
350		NTB_SDOORBELL_LOCKUP | NTB_B2BDOORBELL_BIT14 |
351		    NTB_SB01BASE_LOCKUP },
352
353	{ 0x00000000, NULL, NTB_ATOM, 0 }
354};
355
356static const struct ntb_reg atom_reg = {
357	.ntb_ctl = ATOM_NTBCNTL_OFFSET,
358	.lnk_sta = ATOM_LINK_STATUS_OFFSET,
359	.db_size = sizeof(uint64_t),
360	.mw_bar = { NTB_B2B_BAR_1, NTB_B2B_BAR_2 },
361};
362
363static const struct ntb_alt_reg atom_pri_reg = {
364	.db_bell = ATOM_PDOORBELL_OFFSET,
365	.db_mask = ATOM_PDBMSK_OFFSET,
366	.spad = ATOM_SPAD_OFFSET,
367};
368
369static const struct ntb_alt_reg atom_b2b_reg = {
370	.db_bell = ATOM_B2B_DOORBELL_OFFSET,
371	.spad = ATOM_B2B_SPAD_OFFSET,
372};
373
374static const struct ntb_xlat_reg atom_sec_xlat = {
375#if 0
376	/* "FIXME" says the Linux driver. */
377	.bar0_base = ATOM_SBAR0BASE_OFFSET,
378	.bar2_base = ATOM_SBAR2BASE_OFFSET,
379	.bar4_base = ATOM_SBAR4BASE_OFFSET,
380
381	.bar2_limit = ATOM_SBAR2LMT_OFFSET,
382	.bar4_limit = ATOM_SBAR4LMT_OFFSET,
383#endif
384
385	.bar2_xlat = ATOM_SBAR2XLAT_OFFSET,
386	.bar4_xlat = ATOM_SBAR4XLAT_OFFSET,
387};
388
389static const struct ntb_reg xeon_reg = {
390	.ntb_ctl = XEON_NTBCNTL_OFFSET,
391	.lnk_sta = XEON_LINK_STATUS_OFFSET,
392	.db_size = sizeof(uint16_t),
393	.mw_bar = { NTB_B2B_BAR_1, NTB_B2B_BAR_2, NTB_B2B_BAR_3 },
394};
395
396static const struct ntb_alt_reg xeon_pri_reg = {
397	.db_bell = XEON_PDOORBELL_OFFSET,
398	.db_mask = XEON_PDBMSK_OFFSET,
399	.spad = XEON_SPAD_OFFSET,
400};
401
402static const struct ntb_alt_reg xeon_b2b_reg = {
403	.db_bell = XEON_B2B_DOORBELL_OFFSET,
404	.spad = XEON_B2B_SPAD_OFFSET,
405};
406
407static const struct ntb_xlat_reg xeon_sec_xlat = {
408	.bar0_base = XEON_SBAR0BASE_OFFSET,
409	.bar2_base = XEON_SBAR2BASE_OFFSET,
410	.bar4_base = XEON_SBAR4BASE_OFFSET,
411	.bar5_base = XEON_SBAR5BASE_OFFSET,
412
413	.bar2_limit = XEON_SBAR2LMT_OFFSET,
414	.bar4_limit = XEON_SBAR4LMT_OFFSET,
415	.bar5_limit = XEON_SBAR5LMT_OFFSET,
416
417	.bar2_xlat = XEON_SBAR2XLAT_OFFSET,
418	.bar4_xlat = XEON_SBAR4XLAT_OFFSET,
419	.bar5_xlat = XEON_SBAR5XLAT_OFFSET,
420};
421
422static struct ntb_b2b_addr xeon_b2b_usd_addr = {
423	.bar0_addr = XEON_B2B_BAR0_ADDR,
424	.bar2_addr64 = XEON_B2B_BAR2_ADDR64,
425	.bar4_addr64 = XEON_B2B_BAR4_ADDR64,
426	.bar4_addr32 = XEON_B2B_BAR4_ADDR32,
427	.bar5_addr32 = XEON_B2B_BAR5_ADDR32,
428};
429
430static struct ntb_b2b_addr xeon_b2b_dsd_addr = {
431	.bar0_addr = XEON_B2B_BAR0_ADDR,
432	.bar2_addr64 = XEON_B2B_BAR2_ADDR64,
433	.bar4_addr64 = XEON_B2B_BAR4_ADDR64,
434	.bar4_addr32 = XEON_B2B_BAR4_ADDR32,
435	.bar5_addr32 = XEON_B2B_BAR5_ADDR32,
436};
437
438SYSCTL_NODE(_hw_ntb, OID_AUTO, xeon_b2b, CTLFLAG_RW, 0,
439    "B2B MW segment overrides -- MUST be the same on both sides");
440
441SYSCTL_UQUAD(_hw_ntb_xeon_b2b, OID_AUTO, usd_bar2_addr64, CTLFLAG_RDTUN,
442    &xeon_b2b_usd_addr.bar2_addr64, 0, "If using B2B topology on Xeon "
443    "hardware, use this 64-bit address on the bus between the NTB devices for "
444    "the window at BAR2, on the upstream side of the link.  MUST be the same "
445    "address on both sides.");
446SYSCTL_UQUAD(_hw_ntb_xeon_b2b, OID_AUTO, usd_bar4_addr64, CTLFLAG_RDTUN,
447    &xeon_b2b_usd_addr.bar4_addr64, 0, "See usd_bar2_addr64, but BAR4.");
448SYSCTL_UQUAD(_hw_ntb_xeon_b2b, OID_AUTO, usd_bar4_addr32, CTLFLAG_RDTUN,
449    &xeon_b2b_usd_addr.bar4_addr32, 0, "See usd_bar2_addr64, but BAR4 "
450    "(split-BAR mode).");
451SYSCTL_UQUAD(_hw_ntb_xeon_b2b, OID_AUTO, usd_bar5_addr32, CTLFLAG_RDTUN,
452    &xeon_b2b_usd_addr.bar5_addr32, 0, "See usd_bar2_addr64, but BAR5 "
453    "(split-BAR mode).");
454
455SYSCTL_UQUAD(_hw_ntb_xeon_b2b, OID_AUTO, dsd_bar2_addr64, CTLFLAG_RDTUN,
456    &xeon_b2b_dsd_addr.bar2_addr64, 0, "If using B2B topology on Xeon "
457    "hardware, use this 64-bit address on the bus between the NTB devices for "
458    "the window at BAR2, on the downstream side of the link.  MUST be the same"
459    " address on both sides.");
460SYSCTL_UQUAD(_hw_ntb_xeon_b2b, OID_AUTO, dsd_bar4_addr64, CTLFLAG_RDTUN,
461    &xeon_b2b_dsd_addr.bar4_addr64, 0, "See dsd_bar2_addr64, but BAR4.");
462SYSCTL_UQUAD(_hw_ntb_xeon_b2b, OID_AUTO, dsd_bar4_addr32, CTLFLAG_RDTUN,
463    &xeon_b2b_dsd_addr.bar4_addr32, 0, "See dsd_bar2_addr64, but BAR4 "
464    "(split-BAR mode).");
465SYSCTL_UQUAD(_hw_ntb_xeon_b2b, OID_AUTO, dsd_bar5_addr32, CTLFLAG_RDTUN,
466    &xeon_b2b_dsd_addr.bar5_addr32, 0, "See dsd_bar2_addr64, but BAR5 "
467    "(split-BAR mode).");
468
469/*
470 * OS <-> Driver interface structures
471 */
472MALLOC_DEFINE(M_NTB, "ntb_hw", "ntb_hw driver memory allocations");
473
474static device_method_t ntb_pci_methods[] = {
475	/* Device interface */
476	DEVMETHOD(device_probe,     ntb_probe),
477	DEVMETHOD(device_attach,    ntb_attach),
478	DEVMETHOD(device_detach,    ntb_detach),
479	DEVMETHOD_END
480};
481
482static driver_t ntb_pci_driver = {
483	"ntb_hw",
484	ntb_pci_methods,
485	sizeof(struct ntb_softc),
486};
487
488static devclass_t ntb_devclass;
489DRIVER_MODULE(ntb_hw, pci, ntb_pci_driver, ntb_devclass, NULL, NULL);
490MODULE_VERSION(ntb_hw, 1);
491
492SYSCTL_NODE(_hw, OID_AUTO, ntb, CTLFLAG_RW, 0, "NTB sysctls");
493
494/*
495 * OS <-> Driver linkage functions
496 */
497static int
498ntb_probe(device_t device)
499{
500	struct ntb_hw_info *p;
501
502	p = ntb_get_device_info(pci_get_devid(device));
503	if (p == NULL)
504		return (ENXIO);
505
506	device_set_desc(device, p->desc);
507	return (0);
508}
509
510static int
511ntb_attach(device_t device)
512{
513	struct ntb_softc *ntb;
514	struct ntb_hw_info *p;
515	int error;
516
517	ntb = DEVICE2SOFTC(device);
518	p = ntb_get_device_info(pci_get_devid(device));
519
520	ntb->device = device;
521	ntb->type = p->type;
522	ntb->features = p->features;
523	ntb->b2b_mw_idx = B2B_MW_DISABLED;
524
525	/* Heartbeat timer for NTB_ATOM since there is no link interrupt */
526	callout_init(&ntb->heartbeat_timer, 1);
527	callout_init(&ntb->lr_timer, 1);
528	mtx_init(&ntb->db_mask_lock, "ntb hw bits", NULL, MTX_SPIN);
529	mtx_init(&ntb->ctx_lock, "ntb ctx", NULL, MTX_DEF);
530
531	if (ntb->type == NTB_ATOM)
532		error = ntb_detect_atom(ntb);
533	else
534		error = ntb_detect_xeon(ntb);
535	if (error != 0)
536		goto out;
537
538	ntb_detect_max_mw(ntb);
539
540	pci_enable_busmaster(ntb->device);
541
542	error = ntb_map_pci_bars(ntb);
543	if (error != 0)
544		goto out;
545	if (ntb->type == NTB_ATOM)
546		error = ntb_atom_init_dev(ntb);
547	else
548		error = ntb_xeon_init_dev(ntb);
549	if (error != 0)
550		goto out;
551
552	ntb_poll_link(ntb);
553
554	ntb_sysctl_init(ntb);
555
556out:
557	if (error != 0)
558		ntb_detach(device);
559	return (error);
560}
561
562static int
563ntb_detach(device_t device)
564{
565	struct ntb_softc *ntb;
566
567	ntb = DEVICE2SOFTC(device);
568
569	if (ntb->self_reg != NULL)
570		ntb_db_set_mask(ntb, ntb->db_valid_mask);
571	callout_drain(&ntb->heartbeat_timer);
572	callout_drain(&ntb->lr_timer);
573	pci_disable_busmaster(ntb->device);
574	if (ntb->type == NTB_XEON)
575		ntb_teardown_xeon(ntb);
576	ntb_teardown_interrupts(ntb);
577
578	mtx_destroy(&ntb->db_mask_lock);
579	mtx_destroy(&ntb->ctx_lock);
580
581	/*
582	 * Redetect total MWs so we unmap properly -- in case we lowered the
583	 * maximum to work around Xeon errata.
584	 */
585	ntb_detect_max_mw(ntb);
586	ntb_unmap_pci_bar(ntb);
587
588	return (0);
589}
590
591/*
592 * Driver internal routines
593 */
594static inline enum ntb_bar
595ntb_mw_to_bar(struct ntb_softc *ntb, unsigned mw)
596{
597
598	KASSERT(mw < ntb->mw_count ||
599	    (mw != B2B_MW_DISABLED && mw == ntb->b2b_mw_idx),
600	    ("%s: mw:%u > count:%u", __func__, mw, (unsigned)ntb->mw_count));
601	KASSERT(ntb->reg->mw_bar[mw] != 0, ("invalid mw"));
602
603	return (ntb->reg->mw_bar[mw]);
604}
605
606static inline bool
607bar_is_64bit(struct ntb_softc *ntb, enum ntb_bar bar)
608{
609	/* XXX This assertion could be stronger. */
610	KASSERT(bar < NTB_MAX_BARS, ("bogus bar"));
611	return (bar < NTB_B2B_BAR_2 || !HAS_FEATURE(NTB_SPLIT_BAR));
612}
613
614static inline void
615bar_get_xlat_params(struct ntb_softc *ntb, enum ntb_bar bar, uint32_t *base,
616    uint32_t *xlat, uint32_t *lmt)
617{
618	uint32_t basev, lmtv, xlatv;
619
620	switch (bar) {
621	case NTB_B2B_BAR_1:
622		basev = ntb->xlat_reg->bar2_base;
623		lmtv = ntb->xlat_reg->bar2_limit;
624		xlatv = ntb->xlat_reg->bar2_xlat;
625		break;
626	case NTB_B2B_BAR_2:
627		basev = ntb->xlat_reg->bar4_base;
628		lmtv = ntb->xlat_reg->bar4_limit;
629		xlatv = ntb->xlat_reg->bar4_xlat;
630		break;
631	case NTB_B2B_BAR_3:
632		basev = ntb->xlat_reg->bar5_base;
633		lmtv = ntb->xlat_reg->bar5_limit;
634		xlatv = ntb->xlat_reg->bar5_xlat;
635		break;
636	default:
637		KASSERT(bar >= NTB_B2B_BAR_1 && bar < NTB_MAX_BARS,
638		    ("bad bar"));
639		basev = lmtv = xlatv = 0;
640		break;
641	}
642
643	if (base != NULL)
644		*base = basev;
645	if (xlat != NULL)
646		*xlat = xlatv;
647	if (lmt != NULL)
648		*lmt = lmtv;
649}
650
651static int
652ntb_map_pci_bars(struct ntb_softc *ntb)
653{
654	int rc;
655
656	ntb->bar_info[NTB_CONFIG_BAR].pci_resource_id = PCIR_BAR(0);
657	rc = map_mmr_bar(ntb, &ntb->bar_info[NTB_CONFIG_BAR]);
658	if (rc != 0)
659		goto out;
660
661	ntb->bar_info[NTB_B2B_BAR_1].pci_resource_id = PCIR_BAR(2);
662	rc = map_memory_window_bar(ntb, &ntb->bar_info[NTB_B2B_BAR_1]);
663	if (rc != 0)
664		goto out;
665	ntb->bar_info[NTB_B2B_BAR_1].psz_off = XEON_PBAR23SZ_OFFSET;
666	ntb->bar_info[NTB_B2B_BAR_1].ssz_off = XEON_SBAR23SZ_OFFSET;
667	ntb->bar_info[NTB_B2B_BAR_1].pbarxlat_off = XEON_PBAR2XLAT_OFFSET;
668
669	ntb->bar_info[NTB_B2B_BAR_2].pci_resource_id = PCIR_BAR(4);
670	/* XXX Are shared MW B2Bs write-combining? */
671	if (HAS_FEATURE(NTB_SDOORBELL_LOCKUP) && !HAS_FEATURE(NTB_SPLIT_BAR))
672		rc = map_mmr_bar(ntb, &ntb->bar_info[NTB_B2B_BAR_2]);
673	else
674		rc = map_memory_window_bar(ntb, &ntb->bar_info[NTB_B2B_BAR_2]);
675	ntb->bar_info[NTB_B2B_BAR_2].psz_off = XEON_PBAR4SZ_OFFSET;
676	ntb->bar_info[NTB_B2B_BAR_2].ssz_off = XEON_SBAR4SZ_OFFSET;
677	ntb->bar_info[NTB_B2B_BAR_2].pbarxlat_off = XEON_PBAR4XLAT_OFFSET;
678
679	if (!HAS_FEATURE(NTB_SPLIT_BAR))
680		goto out;
681
682	ntb->bar_info[NTB_B2B_BAR_3].pci_resource_id = PCIR_BAR(5);
683	if (HAS_FEATURE(NTB_SDOORBELL_LOCKUP))
684		rc = map_mmr_bar(ntb, &ntb->bar_info[NTB_B2B_BAR_3]);
685	else
686		rc = map_memory_window_bar(ntb, &ntb->bar_info[NTB_B2B_BAR_3]);
687	ntb->bar_info[NTB_B2B_BAR_3].psz_off = XEON_PBAR5SZ_OFFSET;
688	ntb->bar_info[NTB_B2B_BAR_3].ssz_off = XEON_SBAR5SZ_OFFSET;
689	ntb->bar_info[NTB_B2B_BAR_3].pbarxlat_off = XEON_PBAR5XLAT_OFFSET;
690
691out:
692	if (rc != 0)
693		device_printf(ntb->device,
694		    "unable to allocate pci resource\n");
695	return (rc);
696}
697
698static void
699print_map_success(struct ntb_softc *ntb, struct ntb_pci_bar_info *bar,
700    const char *kind)
701{
702
703	device_printf(ntb->device,
704	    "Mapped BAR%d v:[%p-%p] p:[%p-%p] (0x%jx bytes) (%s)\n",
705	    PCI_RID2BAR(bar->pci_resource_id), bar->vbase,
706	    (char *)bar->vbase + bar->size - 1,
707	    (void *)bar->pbase, (void *)(bar->pbase + bar->size - 1),
708	    (uintmax_t)bar->size, kind);
709}
710
711static int
712map_mmr_bar(struct ntb_softc *ntb, struct ntb_pci_bar_info *bar)
713{
714
715	bar->pci_resource = bus_alloc_resource_any(ntb->device, SYS_RES_MEMORY,
716	    &bar->pci_resource_id, RF_ACTIVE);
717	if (bar->pci_resource == NULL)
718		return (ENXIO);
719
720	save_bar_parameters(bar);
721	print_map_success(ntb, bar, "mmr");
722	return (0);
723}
724
725static int
726map_memory_window_bar(struct ntb_softc *ntb, struct ntb_pci_bar_info *bar)
727{
728	int rc;
729	uint8_t bar_size_bits = 0;
730
731	bar->pci_resource = bus_alloc_resource_any(ntb->device, SYS_RES_MEMORY,
732	    &bar->pci_resource_id, RF_ACTIVE);
733
734	if (bar->pci_resource == NULL)
735		return (ENXIO);
736
737	save_bar_parameters(bar);
738	/*
739	 * Ivytown NTB BAR sizes are misreported by the hardware due to a
740	 * hardware issue. To work around this, query the size it should be
741	 * configured to by the device and modify the resource to correspond to
742	 * this new size. The BIOS on systems with this problem is required to
743	 * provide enough address space to allow the driver to make this change
744	 * safely.
745	 *
746	 * Ideally I could have just specified the size when I allocated the
747	 * resource like:
748	 *  bus_alloc_resource(ntb->device,
749	 *	SYS_RES_MEMORY, &bar->pci_resource_id, 0ul, ~0ul,
750	 *	1ul << bar_size_bits, RF_ACTIVE);
751	 * but the PCI driver does not honor the size in this call, so we have
752	 * to modify it after the fact.
753	 */
754	if (HAS_FEATURE(NTB_BAR_SIZE_4K)) {
755		if (bar->pci_resource_id == PCIR_BAR(2))
756			bar_size_bits = pci_read_config(ntb->device,
757			    XEON_PBAR23SZ_OFFSET, 1);
758		else
759			bar_size_bits = pci_read_config(ntb->device,
760			    XEON_PBAR45SZ_OFFSET, 1);
761
762		rc = bus_adjust_resource(ntb->device, SYS_RES_MEMORY,
763		    bar->pci_resource, bar->pbase,
764		    bar->pbase + (1ul << bar_size_bits) - 1);
765		if (rc != 0) {
766			device_printf(ntb->device,
767			    "unable to resize bar\n");
768			return (rc);
769		}
770
771		save_bar_parameters(bar);
772	}
773
774	print_map_success(ntb, bar, "mw");
775	if (g_ntb_enable_wc == 0)
776		return (0);
777
778	/* Mark bar region as write combining to improve performance. */
779	rc = pmap_change_attr((vm_offset_t)bar->vbase, bar->size,
780	    VM_MEMATTR_WRITE_COMBINING);
781	if (rc == 0) {
782		bar->mapped_wc = true;
783		device_printf(ntb->device,
784		    "Marked BAR%d v:[%p-%p] p:[%p-%p] as "
785		    "WRITE_COMBINING.\n",
786		    PCI_RID2BAR(bar->pci_resource_id), bar->vbase,
787		    (char *)bar->vbase + bar->size - 1,
788		    (void *)bar->pbase, (void *)(bar->pbase + bar->size - 1));
789	} else
790		device_printf(ntb->device,
791		    "Unable to mark BAR%d v:[%p-%p] p:[%p-%p] as "
792		    "WRITE_COMBINING: %d\n",
793		    PCI_RID2BAR(bar->pci_resource_id), bar->vbase,
794		    (char *)bar->vbase + bar->size - 1,
795		    (void *)bar->pbase, (void *)(bar->pbase + bar->size - 1),
796		    rc);
797		/* Proceed anyway */
798	return (0);
799}
800
801static void
802ntb_unmap_pci_bar(struct ntb_softc *ntb)
803{
804	struct ntb_pci_bar_info *current_bar;
805	int i;
806
807	for (i = 0; i < NTB_MAX_BARS; i++) {
808		current_bar = &ntb->bar_info[i];
809		if (current_bar->pci_resource != NULL)
810			bus_release_resource(ntb->device, SYS_RES_MEMORY,
811			    current_bar->pci_resource_id,
812			    current_bar->pci_resource);
813	}
814}
815
816static int
817ntb_setup_msix(struct ntb_softc *ntb, uint32_t num_vectors)
818{
819	uint32_t i;
820	int rc;
821
822	for (i = 0; i < num_vectors; i++) {
823		ntb->int_info[i].rid = i + 1;
824		ntb->int_info[i].res = bus_alloc_resource_any(ntb->device,
825		    SYS_RES_IRQ, &ntb->int_info[i].rid, RF_ACTIVE);
826		if (ntb->int_info[i].res == NULL) {
827			device_printf(ntb->device,
828			    "bus_alloc_resource failed\n");
829			return (ENOMEM);
830		}
831		ntb->int_info[i].tag = NULL;
832		ntb->allocated_interrupts++;
833		rc = bus_setup_intr(ntb->device, ntb->int_info[i].res,
834		    INTR_MPSAFE | INTR_TYPE_MISC, NULL, ndev_vec_isr,
835		    &ntb->msix_vec[i], &ntb->int_info[i].tag);
836		if (rc != 0) {
837			device_printf(ntb->device, "bus_setup_intr failed\n");
838			return (ENXIO);
839		}
840	}
841	return (0);
842}
843
844/*
845 * The Linux NTB driver drops from MSI-X to legacy INTx if a unique vector
846 * cannot be allocated for each MSI-X message.  JHB seems to think remapping
847 * should be okay.  This tunable should enable us to test that hypothesis
848 * when someone gets their hands on some Xeon hardware.
849 */
850static int ntb_force_remap_mode;
851SYSCTL_INT(_hw_ntb, OID_AUTO, force_remap_mode, CTLFLAG_RDTUN,
852    &ntb_force_remap_mode, 0, "If enabled, force MSI-X messages to be remapped"
853    " to a smaller number of ithreads, even if the desired number are "
854    "available");
855
856/*
857 * In case it is NOT ok, give consumers an abort button.
858 */
859static int ntb_prefer_intx;
860SYSCTL_INT(_hw_ntb, OID_AUTO, prefer_intx_to_remap, CTLFLAG_RDTUN,
861    &ntb_prefer_intx, 0, "If enabled, prefer to use legacy INTx mode rather "
862    "than remapping MSI-X messages over available slots (match Linux driver "
863    "behavior)");
864
865/*
866 * Remap the desired number of MSI-X messages to available ithreads in a simple
867 * round-robin fashion.
868 */
869static int
870ntb_remap_msix(device_t dev, uint32_t desired, uint32_t avail)
871{
872	u_int *vectors;
873	uint32_t i;
874	int rc;
875
876	if (ntb_prefer_intx != 0)
877		return (ENXIO);
878
879	vectors = malloc(desired * sizeof(*vectors), M_NTB, M_ZERO | M_WAITOK);
880
881	for (i = 0; i < desired; i++)
882		vectors[i] = (i % avail) + 1;
883
884	rc = pci_remap_msix(dev, desired, vectors);
885	free(vectors, M_NTB);
886	return (rc);
887}
888
889static int
890ntb_init_isr(struct ntb_softc *ntb)
891{
892	uint32_t desired_vectors, num_vectors;
893	int rc;
894
895	ntb->allocated_interrupts = 0;
896	ntb->last_ts = ticks;
897
898	/*
899	 * Mask all doorbell interrupts.
900	 */
901	ntb_db_set_mask(ntb, ntb->db_valid_mask);
902
903	num_vectors = desired_vectors = MIN(pci_msix_count(ntb->device),
904	    ntb->db_count);
905	if (desired_vectors >= 1) {
906		rc = pci_alloc_msix(ntb->device, &num_vectors);
907
908		if (ntb_force_remap_mode != 0 && rc == 0 &&
909		    num_vectors == desired_vectors)
910			num_vectors--;
911
912		if (rc == 0 && num_vectors < desired_vectors) {
913			rc = ntb_remap_msix(ntb->device, desired_vectors,
914			    num_vectors);
915			if (rc == 0)
916				num_vectors = desired_vectors;
917			else
918				pci_release_msi(ntb->device);
919		}
920		if (rc != 0)
921			num_vectors = 1;
922	} else
923		num_vectors = 1;
924
925	if (ntb->type == NTB_XEON && num_vectors < ntb->db_vec_count) {
926		ntb->db_vec_count = 1;
927		ntb->db_vec_shift = XEON_DB_TOTAL_SHIFT;
928		rc = ntb_setup_legacy_interrupt(ntb);
929	} else {
930		ntb_create_msix_vec(ntb, num_vectors);
931		rc = ntb_setup_msix(ntb, num_vectors);
932	}
933	if (rc != 0) {
934		device_printf(ntb->device,
935		    "Error allocating interrupts: %d\n", rc);
936		ntb_free_msix_vec(ntb);
937	}
938
939	return (rc);
940}
941
942static int
943ntb_setup_legacy_interrupt(struct ntb_softc *ntb)
944{
945	int rc;
946
947	ntb->int_info[0].rid = 0;
948	ntb->int_info[0].res = bus_alloc_resource_any(ntb->device, SYS_RES_IRQ,
949	    &ntb->int_info[0].rid, RF_SHAREABLE|RF_ACTIVE);
950	if (ntb->int_info[0].res == NULL) {
951		device_printf(ntb->device, "bus_alloc_resource failed\n");
952		return (ENOMEM);
953	}
954
955	ntb->int_info[0].tag = NULL;
956	ntb->allocated_interrupts = 1;
957
958	rc = bus_setup_intr(ntb->device, ntb->int_info[0].res,
959	    INTR_MPSAFE | INTR_TYPE_MISC, NULL, ndev_irq_isr,
960	    ntb, &ntb->int_info[0].tag);
961	if (rc != 0) {
962		device_printf(ntb->device, "bus_setup_intr failed\n");
963		return (ENXIO);
964	}
965
966	return (0);
967}
968
969static void
970ntb_teardown_interrupts(struct ntb_softc *ntb)
971{
972	struct ntb_int_info *current_int;
973	int i;
974
975	for (i = 0; i < ntb->allocated_interrupts; i++) {
976		current_int = &ntb->int_info[i];
977		if (current_int->tag != NULL)
978			bus_teardown_intr(ntb->device, current_int->res,
979			    current_int->tag);
980
981		if (current_int->res != NULL)
982			bus_release_resource(ntb->device, SYS_RES_IRQ,
983			    rman_get_rid(current_int->res), current_int->res);
984	}
985
986	ntb_free_msix_vec(ntb);
987	pci_release_msi(ntb->device);
988}
989
990/*
991 * Doorbell register and mask are 64-bit on Atom, 16-bit on Xeon.  Abstract it
992 * out to make code clearer.
993 */
994static inline uint64_t
995db_ioread(struct ntb_softc *ntb, uint64_t regoff)
996{
997
998	if (ntb->type == NTB_ATOM)
999		return (ntb_reg_read(8, regoff));
1000
1001	KASSERT(ntb->type == NTB_XEON, ("bad ntb type"));
1002
1003	return (ntb_reg_read(2, regoff));
1004}
1005
1006static inline void
1007db_iowrite(struct ntb_softc *ntb, uint64_t regoff, uint64_t val)
1008{
1009
1010	KASSERT((val & ~ntb->db_valid_mask) == 0,
1011	    ("%s: Invalid bits 0x%jx (valid: 0x%jx)", __func__,
1012	     (uintmax_t)(val & ~ntb->db_valid_mask),
1013	     (uintmax_t)ntb->db_valid_mask));
1014
1015	if (regoff == ntb->self_reg->db_mask)
1016		DB_MASK_ASSERT(ntb, MA_OWNED);
1017	db_iowrite_raw(ntb, regoff, val);
1018}
1019
1020static inline void
1021db_iowrite_raw(struct ntb_softc *ntb, uint64_t regoff, uint64_t val)
1022{
1023
1024	if (ntb->type == NTB_ATOM) {
1025		ntb_reg_write(8, regoff, val);
1026		return;
1027	}
1028
1029	KASSERT(ntb->type == NTB_XEON, ("bad ntb type"));
1030	ntb_reg_write(2, regoff, (uint16_t)val);
1031}
1032
1033void
1034ntb_db_set_mask(struct ntb_softc *ntb, uint64_t bits)
1035{
1036
1037	DB_MASK_LOCK(ntb);
1038	ntb->db_mask |= bits;
1039	db_iowrite(ntb, ntb->self_reg->db_mask, ntb->db_mask);
1040	DB_MASK_UNLOCK(ntb);
1041}
1042
1043void
1044ntb_db_clear_mask(struct ntb_softc *ntb, uint64_t bits)
1045{
1046
1047	KASSERT((bits & ~ntb->db_valid_mask) == 0,
1048	    ("%s: Invalid bits 0x%jx (valid: 0x%jx)", __func__,
1049	     (uintmax_t)(bits & ~ntb->db_valid_mask),
1050	     (uintmax_t)ntb->db_valid_mask));
1051
1052	DB_MASK_LOCK(ntb);
1053	ntb->db_mask &= ~bits;
1054	db_iowrite(ntb, ntb->self_reg->db_mask, ntb->db_mask);
1055	DB_MASK_UNLOCK(ntb);
1056}
1057
1058uint64_t
1059ntb_db_read(struct ntb_softc *ntb)
1060{
1061
1062	return (db_ioread(ntb, ntb->self_reg->db_bell));
1063}
1064
1065void
1066ntb_db_clear(struct ntb_softc *ntb, uint64_t bits)
1067{
1068
1069	KASSERT((bits & ~ntb->db_valid_mask) == 0,
1070	    ("%s: Invalid bits 0x%jx (valid: 0x%jx)", __func__,
1071	     (uintmax_t)(bits & ~ntb->db_valid_mask),
1072	     (uintmax_t)ntb->db_valid_mask));
1073
1074	db_iowrite(ntb, ntb->self_reg->db_bell, bits);
1075}
1076
1077static inline uint64_t
1078ntb_vec_mask(struct ntb_softc *ntb, uint64_t db_vector)
1079{
1080	uint64_t shift, mask;
1081
1082	shift = ntb->db_vec_shift;
1083	mask = (1ull << shift) - 1;
1084	return (mask << (shift * db_vector));
1085}
1086
1087static void
1088ntb_interrupt(struct ntb_softc *ntb, uint32_t vec)
1089{
1090	uint64_t vec_mask;
1091
1092	ntb->last_ts = ticks;
1093	vec_mask = ntb_vec_mask(ntb, vec);
1094
1095	if ((vec_mask & ntb->db_link_mask) != 0) {
1096		if (ntb_poll_link(ntb))
1097			ntb_link_event(ntb);
1098	}
1099
1100	if ((vec_mask & ntb->db_valid_mask) != 0)
1101		ntb_db_event(ntb, vec);
1102}
1103
1104static void
1105ndev_vec_isr(void *arg)
1106{
1107	struct ntb_vec *nvec = arg;
1108
1109	ntb_interrupt(nvec->ntb, nvec->num);
1110}
1111
1112static void
1113ndev_irq_isr(void *arg)
1114{
1115	/* If we couldn't set up MSI-X, we only have the one vector. */
1116	ntb_interrupt(arg, 0);
1117}
1118
1119static int
1120ntb_create_msix_vec(struct ntb_softc *ntb, uint32_t num_vectors)
1121{
1122	uint32_t i;
1123
1124	ntb->msix_vec = malloc(num_vectors * sizeof(*ntb->msix_vec), M_NTB,
1125	    M_ZERO | M_WAITOK);
1126	for (i = 0; i < num_vectors; i++) {
1127		ntb->msix_vec[i].num = i;
1128		ntb->msix_vec[i].ntb = ntb;
1129	}
1130
1131	return (0);
1132}
1133
1134static void
1135ntb_free_msix_vec(struct ntb_softc *ntb)
1136{
1137
1138	if (ntb->msix_vec == NULL)
1139		return;
1140
1141	free(ntb->msix_vec, M_NTB);
1142	ntb->msix_vec = NULL;
1143}
1144
1145static struct ntb_hw_info *
1146ntb_get_device_info(uint32_t device_id)
1147{
1148	struct ntb_hw_info *ep = pci_ids;
1149
1150	while (ep->device_id) {
1151		if (ep->device_id == device_id)
1152			return (ep);
1153		++ep;
1154	}
1155	return (NULL);
1156}
1157
1158static void
1159ntb_teardown_xeon(struct ntb_softc *ntb)
1160{
1161
1162	if (ntb->reg != NULL)
1163		ntb_link_disable(ntb);
1164}
1165
1166static void
1167ntb_detect_max_mw(struct ntb_softc *ntb)
1168{
1169
1170	if (ntb->type == NTB_ATOM) {
1171		ntb->mw_count = ATOM_MW_COUNT;
1172		return;
1173	}
1174
1175	if (HAS_FEATURE(NTB_SPLIT_BAR))
1176		ntb->mw_count = XEON_HSX_SPLIT_MW_COUNT;
1177	else
1178		ntb->mw_count = XEON_SNB_MW_COUNT;
1179}
1180
1181static int
1182ntb_detect_xeon(struct ntb_softc *ntb)
1183{
1184	uint8_t ppd, conn_type;
1185
1186	ppd = pci_read_config(ntb->device, NTB_PPD_OFFSET, 1);
1187	ntb->ppd = ppd;
1188
1189	if ((ppd & XEON_PPD_DEV_TYPE) != 0)
1190		ntb->dev_type = NTB_DEV_DSD;
1191	else
1192		ntb->dev_type = NTB_DEV_USD;
1193
1194	if ((ppd & XEON_PPD_SPLIT_BAR) != 0)
1195		ntb->features |= NTB_SPLIT_BAR;
1196
1197	/* SB01BASE_LOCKUP errata is a superset of SDOORBELL errata */
1198	if (HAS_FEATURE(NTB_SB01BASE_LOCKUP))
1199		ntb->features |= NTB_SDOORBELL_LOCKUP;
1200
1201	conn_type = ppd & XEON_PPD_CONN_TYPE;
1202	switch (conn_type) {
1203	case NTB_CONN_B2B:
1204		ntb->conn_type = conn_type;
1205		break;
1206	case NTB_CONN_RP:
1207	case NTB_CONN_TRANSPARENT:
1208	default:
1209		device_printf(ntb->device, "Unsupported connection type: %u\n",
1210		    (unsigned)conn_type);
1211		return (ENXIO);
1212	}
1213	return (0);
1214}
1215
1216static int
1217ntb_detect_atom(struct ntb_softc *ntb)
1218{
1219	uint32_t ppd, conn_type;
1220
1221	ppd = pci_read_config(ntb->device, NTB_PPD_OFFSET, 4);
1222	ntb->ppd = ppd;
1223
1224	if ((ppd & ATOM_PPD_DEV_TYPE) != 0)
1225		ntb->dev_type = NTB_DEV_DSD;
1226	else
1227		ntb->dev_type = NTB_DEV_USD;
1228
1229	conn_type = (ppd & ATOM_PPD_CONN_TYPE) >> 8;
1230	switch (conn_type) {
1231	case NTB_CONN_B2B:
1232		ntb->conn_type = conn_type;
1233		break;
1234	default:
1235		device_printf(ntb->device, "Unsupported NTB configuration\n");
1236		return (ENXIO);
1237	}
1238	return (0);
1239}
1240
1241static int
1242ntb_xeon_init_dev(struct ntb_softc *ntb)
1243{
1244	int rc;
1245
1246	ntb->spad_count		= XEON_SPAD_COUNT;
1247	ntb->db_count		= XEON_DB_COUNT;
1248	ntb->db_link_mask	= XEON_DB_LINK_BIT;
1249	ntb->db_vec_count	= XEON_DB_MSIX_VECTOR_COUNT;
1250	ntb->db_vec_shift	= XEON_DB_MSIX_VECTOR_SHIFT;
1251
1252	if (ntb->conn_type != NTB_CONN_B2B) {
1253		device_printf(ntb->device, "Connection type %d not supported\n",
1254		    ntb->conn_type);
1255		return (ENXIO);
1256	}
1257
1258	ntb->reg = &xeon_reg;
1259	ntb->self_reg = &xeon_pri_reg;
1260	ntb->peer_reg = &xeon_b2b_reg;
1261	ntb->xlat_reg = &xeon_sec_xlat;
1262
1263	/*
1264	 * There is a Xeon hardware errata related to writes to SDOORBELL or
1265	 * B2BDOORBELL in conjunction with inbound access to NTB MMIO space,
1266	 * which may hang the system.  To workaround this use the second memory
1267	 * window to access the interrupt and scratch pad registers on the
1268	 * remote system.
1269	 */
1270	if (HAS_FEATURE(NTB_SDOORBELL_LOCKUP))
1271		/* Use the last MW for mapping remote spad */
1272		ntb->b2b_mw_idx = ntb->mw_count - 1;
1273	else if (HAS_FEATURE(NTB_B2BDOORBELL_BIT14))
1274		/*
1275		 * HW Errata on bit 14 of b2bdoorbell register.  Writes will not be
1276		 * mirrored to the remote system.  Shrink the number of bits by one,
1277		 * since bit 14 is the last bit.
1278		 *
1279		 * On REGS_THRU_MW errata mode, we don't use the b2bdoorbell register
1280		 * anyway.  Nor for non-B2B connection types.
1281		 */
1282		ntb->db_count = XEON_DB_COUNT - 1;
1283
1284	ntb->db_valid_mask = (1ull << ntb->db_count) - 1;
1285
1286	if (ntb->dev_type == NTB_DEV_USD)
1287		rc = xeon_setup_b2b_mw(ntb, &xeon_b2b_dsd_addr,
1288		    &xeon_b2b_usd_addr);
1289	else
1290		rc = xeon_setup_b2b_mw(ntb, &xeon_b2b_usd_addr,
1291		    &xeon_b2b_dsd_addr);
1292	if (rc != 0)
1293		return (rc);
1294
1295	/* Enable Bus Master and Memory Space on the secondary side */
1296	ntb_reg_write(2, XEON_SPCICMD_OFFSET,
1297	    PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN);
1298
1299	/*
1300	 * Mask all doorbell interrupts.
1301	 */
1302	ntb_db_set_mask(ntb, ntb->db_valid_mask);
1303
1304	rc = ntb_init_isr(ntb);
1305	return (rc);
1306}
1307
1308static int
1309ntb_atom_init_dev(struct ntb_softc *ntb)
1310{
1311	int error;
1312
1313	KASSERT(ntb->conn_type == NTB_CONN_B2B,
1314	    ("Unsupported NTB configuration (%d)\n", ntb->conn_type));
1315
1316	ntb->spad_count		 = ATOM_SPAD_COUNT;
1317	ntb->db_count		 = ATOM_DB_COUNT;
1318	ntb->db_vec_count	 = ATOM_DB_MSIX_VECTOR_COUNT;
1319	ntb->db_vec_shift	 = ATOM_DB_MSIX_VECTOR_SHIFT;
1320	ntb->db_valid_mask	 = (1ull << ntb->db_count) - 1;
1321
1322	ntb->reg = &atom_reg;
1323	ntb->self_reg = &atom_pri_reg;
1324	ntb->peer_reg = &atom_b2b_reg;
1325	ntb->xlat_reg = &atom_sec_xlat;
1326
1327	/*
1328	 * FIXME - MSI-X bug on early Atom HW, remove once internal issue is
1329	 * resolved.  Mask transaction layer internal parity errors.
1330	 */
1331	pci_write_config(ntb->device, 0xFC, 0x4, 4);
1332
1333	configure_atom_secondary_side_bars(ntb);
1334
1335	/* Enable Bus Master and Memory Space on the secondary side */
1336	ntb_reg_write(2, ATOM_SPCICMD_OFFSET,
1337	    PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN);
1338
1339	error = ntb_init_isr(ntb);
1340	if (error != 0)
1341		return (error);
1342
1343	/* Initiate PCI-E link training */
1344	ntb_link_enable(ntb, NTB_SPEED_AUTO, NTB_WIDTH_AUTO);
1345
1346	callout_reset(&ntb->heartbeat_timer, 0, atom_link_hb, ntb);
1347
1348	return (0);
1349}
1350
1351/* XXX: Linux driver doesn't seem to do any of this for Atom. */
1352static void
1353configure_atom_secondary_side_bars(struct ntb_softc *ntb)
1354{
1355
1356	if (ntb->dev_type == NTB_DEV_USD) {
1357		ntb_reg_write(8, ATOM_PBAR2XLAT_OFFSET,
1358		    XEON_B2B_BAR2_ADDR64);
1359		ntb_reg_write(8, ATOM_PBAR4XLAT_OFFSET,
1360		    XEON_B2B_BAR4_ADDR64);
1361		ntb_reg_write(8, ATOM_MBAR23_OFFSET, XEON_B2B_BAR2_ADDR64);
1362		ntb_reg_write(8, ATOM_MBAR45_OFFSET, XEON_B2B_BAR4_ADDR64);
1363	} else {
1364		ntb_reg_write(8, ATOM_PBAR2XLAT_OFFSET,
1365		    XEON_B2B_BAR2_ADDR64);
1366		ntb_reg_write(8, ATOM_PBAR4XLAT_OFFSET,
1367		    XEON_B2B_BAR4_ADDR64);
1368		ntb_reg_write(8, ATOM_MBAR23_OFFSET, XEON_B2B_BAR2_ADDR64);
1369		ntb_reg_write(8, ATOM_MBAR45_OFFSET, XEON_B2B_BAR4_ADDR64);
1370	}
1371}
1372
1373
1374/*
1375 * When working around Xeon SDOORBELL errata by remapping remote registers in a
1376 * MW, limit the B2B MW to half a MW.  By sharing a MW, half the shared MW
1377 * remains for use by a higher layer.
1378 *
1379 * Will only be used if working around SDOORBELL errata and the BIOS-configured
1380 * MW size is sufficiently large.
1381 */
1382static unsigned int ntb_b2b_mw_share;
1383SYSCTL_UINT(_hw_ntb, OID_AUTO, b2b_mw_share, CTLFLAG_RDTUN, &ntb_b2b_mw_share,
1384    0, "If enabled (non-zero), prefer to share half of the B2B peer register "
1385    "MW with higher level consumers.  Both sides of the NTB MUST set the same "
1386    "value here.");
1387
1388static void
1389xeon_reset_sbar_size(struct ntb_softc *ntb, enum ntb_bar idx,
1390    enum ntb_bar regbar)
1391{
1392	struct ntb_pci_bar_info *bar;
1393	uint8_t bar_sz;
1394
1395	if (!HAS_FEATURE(NTB_SPLIT_BAR) && idx >= NTB_B2B_BAR_3)
1396		return;
1397
1398	bar = &ntb->bar_info[idx];
1399	bar_sz = pci_read_config(ntb->device, bar->psz_off, 1);
1400	if (idx == regbar) {
1401		if (ntb->b2b_off != 0)
1402			bar_sz--;
1403		else
1404			bar_sz = 0;
1405	}
1406	pci_write_config(ntb->device, bar->ssz_off, bar_sz, 1);
1407	bar_sz = pci_read_config(ntb->device, bar->ssz_off, 1);
1408	(void)bar_sz;
1409}
1410
1411static void
1412xeon_set_sbar_base_and_limit(struct ntb_softc *ntb, uint64_t bar_addr,
1413    enum ntb_bar idx, enum ntb_bar regbar)
1414{
1415	uint64_t reg_val;
1416	uint32_t base_reg, lmt_reg;
1417
1418	bar_get_xlat_params(ntb, idx, &base_reg, NULL, &lmt_reg);
1419	if (idx == regbar)
1420		bar_addr += ntb->b2b_off;
1421
1422	if (!bar_is_64bit(ntb, idx)) {
1423		ntb_reg_write(4, base_reg, bar_addr);
1424		reg_val = ntb_reg_read(4, base_reg);
1425		(void)reg_val;
1426
1427		ntb_reg_write(4, lmt_reg, bar_addr);
1428		reg_val = ntb_reg_read(4, lmt_reg);
1429		(void)reg_val;
1430	} else {
1431		ntb_reg_write(8, base_reg, bar_addr);
1432		reg_val = ntb_reg_read(8, base_reg);
1433		(void)reg_val;
1434
1435		ntb_reg_write(8, lmt_reg, bar_addr);
1436		reg_val = ntb_reg_read(8, lmt_reg);
1437		(void)reg_val;
1438	}
1439}
1440
1441static void
1442xeon_set_pbar_xlat(struct ntb_softc *ntb, uint64_t base_addr, enum ntb_bar idx)
1443{
1444	struct ntb_pci_bar_info *bar;
1445
1446	bar = &ntb->bar_info[idx];
1447	if (HAS_FEATURE(NTB_SPLIT_BAR) && idx >= NTB_B2B_BAR_2) {
1448		ntb_reg_write(4, bar->pbarxlat_off, base_addr);
1449		base_addr = ntb_reg_read(4, bar->pbarxlat_off);
1450	} else {
1451		ntb_reg_write(8, bar->pbarxlat_off, base_addr);
1452		base_addr = ntb_reg_read(8, bar->pbarxlat_off);
1453	}
1454	(void)base_addr;
1455}
1456
1457static int
1458xeon_setup_b2b_mw(struct ntb_softc *ntb, const struct ntb_b2b_addr *addr,
1459    const struct ntb_b2b_addr *peer_addr)
1460{
1461	struct ntb_pci_bar_info *b2b_bar;
1462	vm_size_t bar_size;
1463	uint64_t bar_addr;
1464	enum ntb_bar b2b_bar_num, i;
1465
1466	if (ntb->b2b_mw_idx == B2B_MW_DISABLED) {
1467		b2b_bar = NULL;
1468		b2b_bar_num = NTB_CONFIG_BAR;
1469		ntb->b2b_off = 0;
1470	} else {
1471		b2b_bar_num = ntb_mw_to_bar(ntb, ntb->b2b_mw_idx);
1472		KASSERT(b2b_bar_num > 0 && b2b_bar_num < NTB_MAX_BARS,
1473		    ("invalid b2b mw bar"));
1474
1475		b2b_bar = &ntb->bar_info[b2b_bar_num];
1476		bar_size = b2b_bar->size;
1477
1478		if (ntb_b2b_mw_share != 0 &&
1479		    (bar_size >> 1) >= XEON_B2B_MIN_SIZE)
1480			ntb->b2b_off = bar_size >> 1;
1481		else if (bar_size >= XEON_B2B_MIN_SIZE) {
1482			ntb->b2b_off = 0;
1483			ntb->mw_count--;
1484		} else {
1485			device_printf(ntb->device,
1486			    "B2B bar size is too small!\n");
1487			return (EIO);
1488		}
1489	}
1490
1491	/*
1492	 * Reset the secondary bar sizes to match the primary bar sizes.
1493	 * (Except, disable or halve the size of the B2B secondary bar.)
1494	 */
1495	for (i = NTB_B2B_BAR_1; i < NTB_MAX_BARS; i++)
1496		xeon_reset_sbar_size(ntb, i, b2b_bar_num);
1497
1498	bar_addr = 0;
1499	if (b2b_bar_num == NTB_CONFIG_BAR)
1500		bar_addr = addr->bar0_addr;
1501	else if (b2b_bar_num == NTB_B2B_BAR_1)
1502		bar_addr = addr->bar2_addr64;
1503	else if (b2b_bar_num == NTB_B2B_BAR_2 && !HAS_FEATURE(NTB_SPLIT_BAR))
1504		bar_addr = addr->bar4_addr64;
1505	else if (b2b_bar_num == NTB_B2B_BAR_2)
1506		bar_addr = addr->bar4_addr32;
1507	else if (b2b_bar_num == NTB_B2B_BAR_3)
1508		bar_addr = addr->bar5_addr32;
1509	else
1510		KASSERT(false, ("invalid bar"));
1511
1512	ntb_reg_write(8, XEON_SBAR0BASE_OFFSET, bar_addr);
1513
1514	/*
1515	 * Other SBARs are normally hit by the PBAR xlat, except for the b2b
1516	 * register BAR.  The B2B BAR is either disabled above or configured
1517	 * half-size.  It starts at PBAR xlat + offset.
1518	 *
1519	 * Also set up incoming BAR limits == base (zero length window).
1520	 */
1521	xeon_set_sbar_base_and_limit(ntb, addr->bar2_addr64, NTB_B2B_BAR_1,
1522	    b2b_bar_num);
1523	if (HAS_FEATURE(NTB_SPLIT_BAR)) {
1524		xeon_set_sbar_base_and_limit(ntb, addr->bar4_addr32,
1525		    NTB_B2B_BAR_2, b2b_bar_num);
1526		xeon_set_sbar_base_and_limit(ntb, addr->bar5_addr32,
1527		    NTB_B2B_BAR_3, b2b_bar_num);
1528	} else
1529		xeon_set_sbar_base_and_limit(ntb, addr->bar4_addr64,
1530		    NTB_B2B_BAR_2, b2b_bar_num);
1531
1532	/* Zero incoming translation addrs */
1533	ntb_reg_write(8, XEON_SBAR2XLAT_OFFSET, 0);
1534	ntb_reg_write(8, XEON_SBAR4XLAT_OFFSET, 0);
1535
1536	/* Zero outgoing translation limits (whole bar size windows) */
1537	ntb_reg_write(8, XEON_PBAR2LMT_OFFSET, 0);
1538	ntb_reg_write(8, XEON_PBAR4LMT_OFFSET, 0);
1539
1540	/* Set outgoing translation offsets */
1541	xeon_set_pbar_xlat(ntb, peer_addr->bar2_addr64, NTB_B2B_BAR_1);
1542	if (HAS_FEATURE(NTB_SPLIT_BAR)) {
1543		xeon_set_pbar_xlat(ntb, peer_addr->bar4_addr32, NTB_B2B_BAR_2);
1544		xeon_set_pbar_xlat(ntb, peer_addr->bar5_addr32, NTB_B2B_BAR_3);
1545	} else
1546		xeon_set_pbar_xlat(ntb, peer_addr->bar4_addr64, NTB_B2B_BAR_2);
1547
1548	/* Set the translation offset for B2B registers */
1549	bar_addr = 0;
1550	if (b2b_bar_num == NTB_CONFIG_BAR)
1551		bar_addr = peer_addr->bar0_addr;
1552	else if (b2b_bar_num == NTB_B2B_BAR_1)
1553		bar_addr = peer_addr->bar2_addr64;
1554	else if (b2b_bar_num == NTB_B2B_BAR_2 && !HAS_FEATURE(NTB_SPLIT_BAR))
1555		bar_addr = peer_addr->bar4_addr64;
1556	else if (b2b_bar_num == NTB_B2B_BAR_2)
1557		bar_addr = peer_addr->bar4_addr32;
1558	else if (b2b_bar_num == NTB_B2B_BAR_3)
1559		bar_addr = peer_addr->bar5_addr32;
1560	else
1561		KASSERT(false, ("invalid bar"));
1562
1563	/*
1564	 * B2B_XLAT_OFFSET is a 64-bit register but can only be written 32 bits
1565	 * at a time.
1566	 */
1567	ntb_reg_write(4, XEON_B2B_XLAT_OFFSETL, bar_addr & 0xffffffff);
1568	ntb_reg_write(4, XEON_B2B_XLAT_OFFSETU, bar_addr >> 32);
1569	return (0);
1570}
1571
1572static inline bool
1573link_is_up(struct ntb_softc *ntb)
1574{
1575
1576	if (ntb->type == NTB_XEON) {
1577		if (ntb->conn_type == NTB_CONN_TRANSPARENT)
1578			return (true);
1579		return ((ntb->lnk_sta & NTB_LINK_STATUS_ACTIVE) != 0);
1580	}
1581
1582	KASSERT(ntb->type == NTB_ATOM, ("ntb type"));
1583	return ((ntb->ntb_ctl & ATOM_CNTL_LINK_DOWN) == 0);
1584}
1585
1586static inline bool
1587atom_link_is_err(struct ntb_softc *ntb)
1588{
1589	uint32_t status;
1590
1591	KASSERT(ntb->type == NTB_ATOM, ("ntb type"));
1592
1593	status = ntb_reg_read(4, ATOM_LTSSMSTATEJMP_OFFSET);
1594	if ((status & ATOM_LTSSMSTATEJMP_FORCEDETECT) != 0)
1595		return (true);
1596
1597	status = ntb_reg_read(4, ATOM_IBSTERRRCRVSTS0_OFFSET);
1598	return ((status & ATOM_IBIST_ERR_OFLOW) != 0);
1599}
1600
1601/* Atom does not have link status interrupt, poll on that platform */
1602static void
1603atom_link_hb(void *arg)
1604{
1605	struct ntb_softc *ntb = arg;
1606	sbintime_t timo, poll_ts;
1607
1608	timo = NTB_HB_TIMEOUT * hz;
1609	poll_ts = ntb->last_ts + timo;
1610
1611	/*
1612	 * Delay polling the link status if an interrupt was received, unless
1613	 * the cached link status says the link is down.
1614	 */
1615	if ((sbintime_t)ticks - poll_ts < 0 && link_is_up(ntb)) {
1616		timo = poll_ts - ticks;
1617		goto out;
1618	}
1619
1620	if (ntb_poll_link(ntb))
1621		ntb_link_event(ntb);
1622
1623	if (!link_is_up(ntb) && atom_link_is_err(ntb)) {
1624		/* Link is down with error, proceed with recovery */
1625		callout_reset(&ntb->lr_timer, 0, recover_atom_link, ntb);
1626		return;
1627	}
1628
1629out:
1630	callout_reset(&ntb->heartbeat_timer, timo, atom_link_hb, ntb);
1631}
1632
1633static void
1634atom_perform_link_restart(struct ntb_softc *ntb)
1635{
1636	uint32_t status;
1637
1638	/* Driver resets the NTB ModPhy lanes - magic! */
1639	ntb_reg_write(1, ATOM_MODPHY_PCSREG6, 0xe0);
1640	ntb_reg_write(1, ATOM_MODPHY_PCSREG4, 0x40);
1641	ntb_reg_write(1, ATOM_MODPHY_PCSREG4, 0x60);
1642	ntb_reg_write(1, ATOM_MODPHY_PCSREG6, 0x60);
1643
1644	/* Driver waits 100ms to allow the NTB ModPhy to settle */
1645	pause("ModPhy", hz / 10);
1646
1647	/* Clear AER Errors, write to clear */
1648	status = ntb_reg_read(4, ATOM_ERRCORSTS_OFFSET);
1649	status &= PCIM_AER_COR_REPLAY_ROLLOVER;
1650	ntb_reg_write(4, ATOM_ERRCORSTS_OFFSET, status);
1651
1652	/* Clear unexpected electrical idle event in LTSSM, write to clear */
1653	status = ntb_reg_read(4, ATOM_LTSSMERRSTS0_OFFSET);
1654	status |= ATOM_LTSSMERRSTS0_UNEXPECTEDEI;
1655	ntb_reg_write(4, ATOM_LTSSMERRSTS0_OFFSET, status);
1656
1657	/* Clear DeSkew Buffer error, write to clear */
1658	status = ntb_reg_read(4, ATOM_DESKEWSTS_OFFSET);
1659	status |= ATOM_DESKEWSTS_DBERR;
1660	ntb_reg_write(4, ATOM_DESKEWSTS_OFFSET, status);
1661
1662	status = ntb_reg_read(4, ATOM_IBSTERRRCRVSTS0_OFFSET);
1663	status &= ATOM_IBIST_ERR_OFLOW;
1664	ntb_reg_write(4, ATOM_IBSTERRRCRVSTS0_OFFSET, status);
1665
1666	/* Releases the NTB state machine to allow the link to retrain */
1667	status = ntb_reg_read(4, ATOM_LTSSMSTATEJMP_OFFSET);
1668	status &= ~ATOM_LTSSMSTATEJMP_FORCEDETECT;
1669	ntb_reg_write(4, ATOM_LTSSMSTATEJMP_OFFSET, status);
1670}
1671
1672/*
1673 * ntb_set_ctx() - associate a driver context with an ntb device
1674 * @ntb:        NTB device context
1675 * @ctx:        Driver context
1676 * @ctx_ops:    Driver context operations
1677 *
1678 * Associate a driver context and operations with a ntb device.  The context is
1679 * provided by the client driver, and the driver may associate a different
1680 * context with each ntb device.
1681 *
1682 * Return: Zero if the context is associated, otherwise an error number.
1683 */
1684int
1685ntb_set_ctx(struct ntb_softc *ntb, void *ctx, const struct ntb_ctx_ops *ops)
1686{
1687
1688	if (ctx == NULL || ops == NULL)
1689		return (EINVAL);
1690	if (ntb->ctx_ops != NULL)
1691		return (EINVAL);
1692
1693	CTX_LOCK(ntb);
1694	if (ntb->ctx_ops != NULL) {
1695		CTX_UNLOCK(ntb);
1696		return (EINVAL);
1697	}
1698	ntb->ntb_ctx = ctx;
1699	ntb->ctx_ops = ops;
1700	CTX_UNLOCK(ntb);
1701
1702	return (0);
1703}
1704
1705/*
1706 * It is expected that this will only be used from contexts where the ctx_lock
1707 * is not needed to protect ntb_ctx lifetime.
1708 */
1709void *
1710ntb_get_ctx(struct ntb_softc *ntb, const struct ntb_ctx_ops **ops)
1711{
1712
1713	KASSERT(ntb->ntb_ctx != NULL && ntb->ctx_ops != NULL, ("bogus"));
1714	if (ops != NULL)
1715		*ops = ntb->ctx_ops;
1716	return (ntb->ntb_ctx);
1717}
1718
1719/*
1720 * ntb_clear_ctx() - disassociate any driver context from an ntb device
1721 * @ntb:        NTB device context
1722 *
1723 * Clear any association that may exist between a driver context and the ntb
1724 * device.
1725 */
1726void
1727ntb_clear_ctx(struct ntb_softc *ntb)
1728{
1729
1730	CTX_LOCK(ntb);
1731	ntb->ntb_ctx = NULL;
1732	ntb->ctx_ops = NULL;
1733	CTX_UNLOCK(ntb);
1734}
1735
1736/*
1737 * ntb_link_event() - notify driver context of a change in link status
1738 * @ntb:        NTB device context
1739 *
1740 * Notify the driver context that the link status may have changed.  The driver
1741 * should call ntb_link_is_up() to get the current status.
1742 */
1743void
1744ntb_link_event(struct ntb_softc *ntb)
1745{
1746
1747	CTX_LOCK(ntb);
1748	if (ntb->ctx_ops != NULL && ntb->ctx_ops->link_event != NULL)
1749		ntb->ctx_ops->link_event(ntb->ntb_ctx);
1750	CTX_UNLOCK(ntb);
1751}
1752
1753/*
1754 * ntb_db_event() - notify driver context of a doorbell event
1755 * @ntb:        NTB device context
1756 * @vector:     Interrupt vector number
1757 *
1758 * Notify the driver context of a doorbell event.  If hardware supports
1759 * multiple interrupt vectors for doorbells, the vector number indicates which
1760 * vector received the interrupt.  The vector number is relative to the first
1761 * vector used for doorbells, starting at zero, and must be less than
1762 * ntb_db_vector_count().  The driver may call ntb_db_read() to check which
1763 * doorbell bits need service, and ntb_db_vector_mask() to determine which of
1764 * those bits are associated with the vector number.
1765 */
1766static void
1767ntb_db_event(struct ntb_softc *ntb, uint32_t vec)
1768{
1769
1770	CTX_LOCK(ntb);
1771	if (ntb->ctx_ops != NULL && ntb->ctx_ops->db_event != NULL)
1772		ntb->ctx_ops->db_event(ntb->ntb_ctx, vec);
1773	CTX_UNLOCK(ntb);
1774}
1775
1776/*
1777 * ntb_link_enable() - enable the link on the secondary side of the ntb
1778 * @ntb:        NTB device context
1779 * @max_speed:  The maximum link speed expressed as PCIe generation number[0]
1780 * @max_width:  The maximum link width expressed as the number of PCIe lanes[0]
1781 *
1782 * Enable the link on the secondary side of the ntb.  This can only be done
1783 * from the primary side of the ntb in primary or b2b topology.  The ntb device
1784 * should train the link to its maximum speed and width, or the requested speed
1785 * and width, whichever is smaller, if supported.
1786 *
1787 * Return: Zero on success, otherwise an error number.
1788 *
1789 * [0]: Only NTB_SPEED_AUTO and NTB_WIDTH_AUTO are valid inputs; other speed
1790 *      and width input will be ignored.
1791 */
1792int
1793ntb_link_enable(struct ntb_softc *ntb, enum ntb_speed s __unused,
1794    enum ntb_width w __unused)
1795{
1796	uint32_t cntl;
1797
1798	if (ntb->type == NTB_ATOM) {
1799		pci_write_config(ntb->device, NTB_PPD_OFFSET,
1800		    ntb->ppd | ATOM_PPD_INIT_LINK, 4);
1801		return (0);
1802	}
1803
1804	if (ntb->conn_type == NTB_CONN_TRANSPARENT) {
1805		ntb_link_event(ntb);
1806		return (0);
1807	}
1808
1809	cntl = ntb_reg_read(4, ntb->reg->ntb_ctl);
1810	cntl &= ~(NTB_CNTL_LINK_DISABLE | NTB_CNTL_CFG_LOCK);
1811	cntl |= NTB_CNTL_P2S_BAR23_SNOOP | NTB_CNTL_S2P_BAR23_SNOOP;
1812	cntl |= NTB_CNTL_P2S_BAR4_SNOOP | NTB_CNTL_S2P_BAR4_SNOOP;
1813	if (HAS_FEATURE(NTB_SPLIT_BAR))
1814		cntl |= NTB_CNTL_P2S_BAR5_SNOOP | NTB_CNTL_S2P_BAR5_SNOOP;
1815	ntb_reg_write(4, ntb->reg->ntb_ctl, cntl);
1816	return (0);
1817}
1818
1819/*
1820 * ntb_link_disable() - disable the link on the secondary side of the ntb
1821 * @ntb:        NTB device context
1822 *
1823 * Disable the link on the secondary side of the ntb.  This can only be done
1824 * from the primary side of the ntb in primary or b2b topology.  The ntb device
1825 * should disable the link.  Returning from this call must indicate that a
1826 * barrier has passed, though with no more writes may pass in either direction
1827 * across the link, except if this call returns an error number.
1828 *
1829 * Return: Zero on success, otherwise an error number.
1830 */
1831int
1832ntb_link_disable(struct ntb_softc *ntb)
1833{
1834	uint32_t cntl;
1835
1836	if (ntb->conn_type == NTB_CONN_TRANSPARENT) {
1837		ntb_link_event(ntb);
1838		return (0);
1839	}
1840
1841	cntl = ntb_reg_read(4, ntb->reg->ntb_ctl);
1842	cntl &= ~(NTB_CNTL_P2S_BAR23_SNOOP | NTB_CNTL_S2P_BAR23_SNOOP);
1843	cntl &= ~(NTB_CNTL_P2S_BAR4_SNOOP | NTB_CNTL_S2P_BAR4_SNOOP);
1844	if (HAS_FEATURE(NTB_SPLIT_BAR))
1845		cntl &= ~(NTB_CNTL_P2S_BAR5_SNOOP | NTB_CNTL_S2P_BAR5_SNOOP);
1846	cntl |= NTB_CNTL_LINK_DISABLE | NTB_CNTL_CFG_LOCK;
1847	ntb_reg_write(4, ntb->reg->ntb_ctl, cntl);
1848	return (0);
1849}
1850
1851static void
1852recover_atom_link(void *arg)
1853{
1854	struct ntb_softc *ntb = arg;
1855	unsigned speed, width, oldspeed, oldwidth;
1856	uint32_t status32;
1857
1858	atom_perform_link_restart(ntb);
1859
1860	/*
1861	 * There is a potential race between the 2 NTB devices recovering at
1862	 * the same time.  If the times are the same, the link will not recover
1863	 * and the driver will be stuck in this loop forever.  Add a random
1864	 * interval to the recovery time to prevent this race.
1865	 */
1866	status32 = arc4random() % ATOM_LINK_RECOVERY_TIME;
1867	pause("Link", (ATOM_LINK_RECOVERY_TIME + status32) * hz / 1000);
1868
1869	if (atom_link_is_err(ntb))
1870		goto retry;
1871
1872	status32 = ntb_reg_read(4, ntb->reg->ntb_ctl);
1873	if ((status32 & ATOM_CNTL_LINK_DOWN) != 0)
1874		goto out;
1875
1876	status32 = ntb_reg_read(4, ntb->reg->lnk_sta);
1877	width = NTB_LNK_STA_WIDTH(status32);
1878	speed = status32 & NTB_LINK_SPEED_MASK;
1879
1880	oldwidth = NTB_LNK_STA_WIDTH(ntb->lnk_sta);
1881	oldspeed = ntb->lnk_sta & NTB_LINK_SPEED_MASK;
1882	if (oldwidth != width || oldspeed != speed)
1883		goto retry;
1884
1885out:
1886	callout_reset(&ntb->heartbeat_timer, NTB_HB_TIMEOUT * hz, atom_link_hb,
1887	    ntb);
1888	return;
1889
1890retry:
1891	callout_reset(&ntb->lr_timer, NTB_HB_TIMEOUT * hz, recover_atom_link,
1892	    ntb);
1893}
1894
1895/*
1896 * Polls the HW link status register(s); returns true if something has changed.
1897 */
1898static bool
1899ntb_poll_link(struct ntb_softc *ntb)
1900{
1901	uint32_t ntb_cntl;
1902	uint16_t reg_val;
1903
1904	if (ntb->type == NTB_ATOM) {
1905		ntb_cntl = ntb_reg_read(4, ntb->reg->ntb_ctl);
1906		if (ntb_cntl == ntb->ntb_ctl)
1907			return (false);
1908
1909		ntb->ntb_ctl = ntb_cntl;
1910		ntb->lnk_sta = ntb_reg_read(4, ntb->reg->lnk_sta);
1911	} else {
1912		db_iowrite_raw(ntb, ntb->self_reg->db_bell, ntb->db_link_mask);
1913
1914		reg_val = pci_read_config(ntb->device, ntb->reg->lnk_sta, 2);
1915		if (reg_val == ntb->lnk_sta)
1916			return (false);
1917
1918		ntb->lnk_sta = reg_val;
1919	}
1920	return (true);
1921}
1922
1923static inline enum ntb_speed
1924ntb_link_sta_speed(struct ntb_softc *ntb)
1925{
1926
1927	if (!link_is_up(ntb))
1928		return (NTB_SPEED_NONE);
1929	return (ntb->lnk_sta & NTB_LINK_SPEED_MASK);
1930}
1931
1932static inline enum ntb_width
1933ntb_link_sta_width(struct ntb_softc *ntb)
1934{
1935
1936	if (!link_is_up(ntb))
1937		return (NTB_WIDTH_NONE);
1938	return (NTB_LNK_STA_WIDTH(ntb->lnk_sta));
1939}
1940
1941SYSCTL_NODE(_hw_ntb, OID_AUTO, debug_info, CTLFLAG_RW, 0,
1942    "Driver state, statistics, and HW registers");
1943
1944#define NTB_REGSZ_MASK	(3ul << 30)
1945#define NTB_REG_64	(1ul << 30)
1946#define NTB_REG_32	(2ul << 30)
1947#define NTB_REG_16	(3ul << 30)
1948#define NTB_REG_8	(0ul << 30)
1949
1950#define NTB_DB_READ	(1ul << 29)
1951#define NTB_PCI_REG	(1ul << 28)
1952#define NTB_REGFLAGS_MASK	(NTB_REGSZ_MASK | NTB_DB_READ | NTB_PCI_REG)
1953
1954static void
1955ntb_sysctl_init(struct ntb_softc *ntb)
1956{
1957	struct sysctl_oid_list *tree_par, *regpar, *statpar, *errpar;
1958	struct sysctl_ctx_list *ctx;
1959	struct sysctl_oid *tree, *tmptree;
1960
1961	ctx = device_get_sysctl_ctx(ntb->device);
1962
1963	tree = SYSCTL_ADD_NODE(ctx,
1964	    SYSCTL_CHILDREN(device_get_sysctl_tree(ntb->device)), OID_AUTO,
1965	    "debug_info", CTLFLAG_RD, NULL,
1966	    "Driver state, statistics, and HW registers");
1967	tree_par = SYSCTL_CHILDREN(tree);
1968
1969	SYSCTL_ADD_UINT(ctx, tree_par, OID_AUTO, "conn_type", CTLFLAG_RD,
1970	    &ntb->conn_type, 0, "0 - Transparent; 1 - B2B; 2 - Root Port");
1971	SYSCTL_ADD_UINT(ctx, tree_par, OID_AUTO, "dev_type", CTLFLAG_RD,
1972	    &ntb->dev_type, 0, "0 - USD; 1 - DSD");
1973	SYSCTL_ADD_UINT(ctx, tree_par, OID_AUTO, "ppd", CTLFLAG_RD,
1974	    &ntb->ppd, 0, "Raw PPD register (cached)");
1975
1976	if (ntb->b2b_mw_idx != B2B_MW_DISABLED) {
1977		SYSCTL_ADD_U8(ctx, tree_par, OID_AUTO, "b2b_idx", CTLFLAG_RD,
1978		    &ntb->b2b_mw_idx, 0,
1979		    "Index of the MW used for B2B remote register access");
1980		SYSCTL_ADD_UQUAD(ctx, tree_par, OID_AUTO, "b2b_off",
1981		    CTLFLAG_RD, &ntb->b2b_off,
1982		    "If non-zero, offset of B2B register region in shared MW");
1983	}
1984
1985	SYSCTL_ADD_PROC(ctx, tree_par, OID_AUTO, "features",
1986	    CTLFLAG_RD | CTLTYPE_STRING, ntb, 0, sysctl_handle_features, "A",
1987	    "Features/errata of this NTB device");
1988
1989	SYSCTL_ADD_UINT(ctx, tree_par, OID_AUTO, "ntb_ctl", CTLFLAG_RD,
1990	    __DEVOLATILE(uint32_t *, &ntb->ntb_ctl), 0,
1991	    "NTB CTL register (cached)");
1992	SYSCTL_ADD_UINT(ctx, tree_par, OID_AUTO, "lnk_sta", CTLFLAG_RD,
1993	    __DEVOLATILE(uint32_t *, &ntb->lnk_sta), 0,
1994	    "LNK STA register (cached)");
1995
1996	SYSCTL_ADD_PROC(ctx, tree_par, OID_AUTO, "link_status",
1997	    CTLFLAG_RD | CTLTYPE_STRING, ntb, 0, sysctl_handle_link_status,
1998	    "A", "Link status");
1999
2000	SYSCTL_ADD_U8(ctx, tree_par, OID_AUTO, "mw_count", CTLFLAG_RD,
2001	    &ntb->mw_count, 0, "MW count (excl. non-shared B2B register BAR)");
2002	SYSCTL_ADD_U8(ctx, tree_par, OID_AUTO, "spad_count", CTLFLAG_RD,
2003	    &ntb->spad_count, 0, "Scratchpad count");
2004	SYSCTL_ADD_U8(ctx, tree_par, OID_AUTO, "db_count", CTLFLAG_RD,
2005	    &ntb->db_count, 0, "Doorbell count");
2006	SYSCTL_ADD_U8(ctx, tree_par, OID_AUTO, "db_vec_count", CTLFLAG_RD,
2007	    &ntb->db_vec_count, 0, "Doorbell vector count");
2008	SYSCTL_ADD_U8(ctx, tree_par, OID_AUTO, "db_vec_shift", CTLFLAG_RD,
2009	    &ntb->db_vec_shift, 0, "Doorbell vector shift");
2010
2011	SYSCTL_ADD_UQUAD(ctx, tree_par, OID_AUTO, "db_valid_mask", CTLFLAG_RD,
2012	    &ntb->db_valid_mask, "Doorbell valid mask");
2013	SYSCTL_ADD_UQUAD(ctx, tree_par, OID_AUTO, "db_link_mask", CTLFLAG_RD,
2014	    &ntb->db_link_mask, "Doorbell link mask");
2015	SYSCTL_ADD_UQUAD(ctx, tree_par, OID_AUTO, "db_mask", CTLFLAG_RD,
2016	    &ntb->db_mask, "Doorbell mask (cached)");
2017
2018	tmptree = SYSCTL_ADD_NODE(ctx, tree_par, OID_AUTO, "registers",
2019	    CTLFLAG_RD, NULL, "Raw HW registers (big-endian)");
2020	regpar = SYSCTL_CHILDREN(tmptree);
2021
2022	SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "ntbcntl",
2023	    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb, NTB_REG_32 |
2024	    ntb->reg->ntb_ctl, sysctl_handle_register, "IU",
2025	    "NTB Control register");
2026	SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "lnkcap",
2027	    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb, NTB_REG_32 |
2028	    0x19c, sysctl_handle_register, "IU",
2029	    "NTB Link Capabilities");
2030	SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "lnkcon",
2031	    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb, NTB_REG_32 |
2032	    0x1a0, sysctl_handle_register, "IU",
2033	    "NTB Link Control register");
2034
2035	SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "db_mask",
2036	    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2037	    NTB_REG_64 | NTB_DB_READ | ntb->self_reg->db_mask,
2038	    sysctl_handle_register, "QU", "Doorbell mask register");
2039	SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "db_bell",
2040	    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2041	    NTB_REG_64 | NTB_DB_READ | ntb->self_reg->db_bell,
2042	    sysctl_handle_register, "QU", "Doorbell register");
2043
2044	SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "incoming_xlat23",
2045	    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2046	    NTB_REG_64 | ntb->xlat_reg->bar2_xlat,
2047	    sysctl_handle_register, "QU", "Incoming XLAT23 register");
2048	if (HAS_FEATURE(NTB_SPLIT_BAR)) {
2049		SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "incoming_xlat4",
2050		    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2051		    NTB_REG_32 | ntb->xlat_reg->bar4_xlat,
2052		    sysctl_handle_register, "IU", "Incoming XLAT4 register");
2053		SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "incoming_xlat5",
2054		    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2055		    NTB_REG_32 | ntb->xlat_reg->bar5_xlat,
2056		    sysctl_handle_register, "IU", "Incoming XLAT5 register");
2057	} else {
2058		SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "incoming_xlat45",
2059		    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2060		    NTB_REG_64 | ntb->xlat_reg->bar4_xlat,
2061		    sysctl_handle_register, "QU", "Incoming XLAT45 register");
2062	}
2063
2064	SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "incoming_lmt23",
2065	    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2066	    NTB_REG_64 | ntb->xlat_reg->bar2_limit,
2067	    sysctl_handle_register, "QU", "Incoming LMT23 register");
2068	if (HAS_FEATURE(NTB_SPLIT_BAR)) {
2069		SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "incoming_lmt4",
2070		    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2071		    NTB_REG_32 | ntb->xlat_reg->bar4_limit,
2072		    sysctl_handle_register, "IU", "Incoming LMT4 register");
2073		SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "incoming_lmt5",
2074		    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2075		    NTB_REG_32 | ntb->xlat_reg->bar5_limit,
2076		    sysctl_handle_register, "IU", "Incoming LMT5 register");
2077	} else {
2078		SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "incoming_lmt45",
2079		    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2080		    NTB_REG_64 | ntb->xlat_reg->bar4_limit,
2081		    sysctl_handle_register, "QU", "Incoming LMT45 register");
2082	}
2083
2084	if (ntb->type == NTB_ATOM)
2085		return;
2086
2087	tmptree = SYSCTL_ADD_NODE(ctx, regpar, OID_AUTO, "xeon_stats",
2088	    CTLFLAG_RD, NULL, "Xeon HW statistics");
2089	statpar = SYSCTL_CHILDREN(tmptree);
2090	SYSCTL_ADD_PROC(ctx, statpar, OID_AUTO, "upstream_mem_miss",
2091	    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2092	    NTB_REG_16 | XEON_USMEMMISS_OFFSET,
2093	    sysctl_handle_register, "SU", "Upstream Memory Miss");
2094
2095	tmptree = SYSCTL_ADD_NODE(ctx, regpar, OID_AUTO, "xeon_hw_err",
2096	    CTLFLAG_RD, NULL, "Xeon HW errors");
2097	errpar = SYSCTL_CHILDREN(tmptree);
2098
2099	SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "ppd",
2100	    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2101	    NTB_REG_8 | NTB_PCI_REG | NTB_PPD_OFFSET,
2102	    sysctl_handle_register, "CU", "PPD");
2103
2104	SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "pbar23_sz",
2105	    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2106	    NTB_REG_8 | NTB_PCI_REG | XEON_PBAR23SZ_OFFSET,
2107	    sysctl_handle_register, "CU", "PBAR23 SZ (log2)");
2108	SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "pbar4_sz",
2109	    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2110	    NTB_REG_8 | NTB_PCI_REG | XEON_PBAR4SZ_OFFSET,
2111	    sysctl_handle_register, "CU", "PBAR4 SZ (log2)");
2112	SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "pbar5_sz",
2113	    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2114	    NTB_REG_8 | NTB_PCI_REG | XEON_PBAR5SZ_OFFSET,
2115	    sysctl_handle_register, "CU", "PBAR5 SZ (log2)");
2116
2117	SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "sbar23_sz",
2118	    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2119	    NTB_REG_8 | NTB_PCI_REG | XEON_SBAR23SZ_OFFSET,
2120	    sysctl_handle_register, "CU", "SBAR23 SZ (log2)");
2121	SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "sbar4_sz",
2122	    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2123	    NTB_REG_8 | NTB_PCI_REG | XEON_SBAR4SZ_OFFSET,
2124	    sysctl_handle_register, "CU", "SBAR4 SZ (log2)");
2125	SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "sbar5_sz",
2126	    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2127	    NTB_REG_8 | NTB_PCI_REG | XEON_SBAR5SZ_OFFSET,
2128	    sysctl_handle_register, "CU", "SBAR5 SZ (log2)");
2129
2130	SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "devsts",
2131	    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2132	    NTB_REG_16 | NTB_PCI_REG | XEON_DEVSTS_OFFSET,
2133	    sysctl_handle_register, "SU", "DEVSTS");
2134	SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "lnksts",
2135	    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2136	    NTB_REG_16 | NTB_PCI_REG | XEON_LINK_STATUS_OFFSET,
2137	    sysctl_handle_register, "SU", "LNKSTS");
2138	SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "slnksts",
2139	    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2140	    NTB_REG_16 | NTB_PCI_REG | XEON_SLINK_STATUS_OFFSET,
2141	    sysctl_handle_register, "SU", "SLNKSTS");
2142
2143	SYSCTL_ADD_PROC(ctx, errpar, OID_AUTO, "uncerrsts",
2144	    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2145	    NTB_REG_32 | NTB_PCI_REG | XEON_UNCERRSTS_OFFSET,
2146	    sysctl_handle_register, "IU", "UNCERRSTS");
2147	SYSCTL_ADD_PROC(ctx, errpar, OID_AUTO, "corerrsts",
2148	    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2149	    NTB_REG_32 | NTB_PCI_REG | XEON_CORERRSTS_OFFSET,
2150	    sysctl_handle_register, "IU", "CORERRSTS");
2151
2152	if (ntb->conn_type != NTB_CONN_B2B)
2153		return;
2154
2155	SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "outgoing_xlat23",
2156	    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2157	    NTB_REG_64 | ntb->bar_info[NTB_B2B_BAR_1].pbarxlat_off,
2158	    sysctl_handle_register, "QU", "Outgoing XLAT23 register");
2159	if (HAS_FEATURE(NTB_SPLIT_BAR)) {
2160		SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "outgoing_xlat4",
2161		    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2162		    NTB_REG_32 | ntb->bar_info[NTB_B2B_BAR_2].pbarxlat_off,
2163		    sysctl_handle_register, "IU", "Outgoing XLAT4 register");
2164		SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "outgoing_xlat5",
2165		    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2166		    NTB_REG_32 | ntb->bar_info[NTB_B2B_BAR_3].pbarxlat_off,
2167		    sysctl_handle_register, "IU", "Outgoing XLAT5 register");
2168	} else {
2169		SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "outgoing_xlat45",
2170		    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2171		    NTB_REG_64 | ntb->bar_info[NTB_B2B_BAR_2].pbarxlat_off,
2172		    sysctl_handle_register, "QU", "Outgoing XLAT45 register");
2173	}
2174
2175	SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "outgoing_lmt23",
2176	    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2177	    NTB_REG_64 | XEON_PBAR2LMT_OFFSET,
2178	    sysctl_handle_register, "QU", "Outgoing LMT23 register");
2179	if (HAS_FEATURE(NTB_SPLIT_BAR)) {
2180		SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "outgoing_lmt4",
2181		    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2182		    NTB_REG_32 | XEON_PBAR4LMT_OFFSET,
2183		    sysctl_handle_register, "IU", "Outgoing LMT4 register");
2184		SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "outgoing_lmt5",
2185		    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2186		    NTB_REG_32 | XEON_PBAR5LMT_OFFSET,
2187		    sysctl_handle_register, "IU", "Outgoing LMT5 register");
2188	} else {
2189		SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "outgoing_lmt45",
2190		    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2191		    NTB_REG_64 | XEON_PBAR4LMT_OFFSET,
2192		    sysctl_handle_register, "QU", "Outgoing LMT45 register");
2193	}
2194
2195	SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "sbar01_base",
2196	    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2197	    NTB_REG_64 | ntb->xlat_reg->bar0_base,
2198	    sysctl_handle_register, "QU", "Secondary BAR01 base register");
2199	SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "sbar23_base",
2200	    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2201	    NTB_REG_64 | ntb->xlat_reg->bar2_base,
2202	    sysctl_handle_register, "QU", "Secondary BAR23 base register");
2203	if (HAS_FEATURE(NTB_SPLIT_BAR)) {
2204		SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "sbar4_base",
2205		    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2206		    NTB_REG_32 | ntb->xlat_reg->bar4_base,
2207		    sysctl_handle_register, "IU",
2208		    "Secondary BAR4 base register");
2209		SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "sbar5_base",
2210		    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2211		    NTB_REG_32 | ntb->xlat_reg->bar5_base,
2212		    sysctl_handle_register, "IU",
2213		    "Secondary BAR5 base register");
2214	} else {
2215		SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "sbar45_base",
2216		    CTLFLAG_RD | CTLTYPE_OPAQUE, ntb,
2217		    NTB_REG_64 | ntb->xlat_reg->bar4_base,
2218		    sysctl_handle_register, "QU",
2219		    "Secondary BAR45 base register");
2220	}
2221}
2222
2223static int
2224sysctl_handle_features(SYSCTL_HANDLER_ARGS)
2225{
2226	struct ntb_softc *ntb;
2227	struct sbuf sb;
2228	int error;
2229
2230	error = 0;
2231	ntb = arg1;
2232
2233	sbuf_new_for_sysctl(&sb, NULL, 256, req);
2234
2235	sbuf_printf(&sb, "%b", ntb->features, NTB_FEATURES_STR);
2236	error = sbuf_finish(&sb);
2237	sbuf_delete(&sb);
2238
2239	if (error || !req->newptr)
2240		return (error);
2241	return (EINVAL);
2242}
2243
2244static int
2245sysctl_handle_link_status(SYSCTL_HANDLER_ARGS)
2246{
2247	struct ntb_softc *ntb;
2248	struct sbuf sb;
2249	enum ntb_speed speed;
2250	enum ntb_width width;
2251	int error;
2252
2253	error = 0;
2254	ntb = arg1;
2255
2256	sbuf_new_for_sysctl(&sb, NULL, 32, req);
2257
2258	if (ntb_link_is_up(ntb, &speed, &width))
2259		sbuf_printf(&sb, "up / PCIe Gen %u / Width x%u",
2260		    (unsigned)speed, (unsigned)width);
2261	else
2262		sbuf_printf(&sb, "down");
2263
2264	error = sbuf_finish(&sb);
2265	sbuf_delete(&sb);
2266
2267	if (error || !req->newptr)
2268		return (error);
2269	return (EINVAL);
2270}
2271
2272static int
2273sysctl_handle_register(SYSCTL_HANDLER_ARGS)
2274{
2275	struct ntb_softc *ntb;
2276	const void *outp;
2277	uintptr_t sz;
2278	uint64_t umv;
2279	char be[sizeof(umv)];
2280	size_t outsz;
2281	uint32_t reg;
2282	bool db, pci;
2283	int error;
2284
2285	ntb = arg1;
2286	reg = arg2 & ~NTB_REGFLAGS_MASK;
2287	sz = arg2 & NTB_REGSZ_MASK;
2288	db = (arg2 & NTB_DB_READ) != 0;
2289	pci = (arg2 & NTB_PCI_REG) != 0;
2290
2291	KASSERT(!(db && pci), ("bogus"));
2292
2293	if (db) {
2294		KASSERT(sz == NTB_REG_64, ("bogus"));
2295		umv = db_ioread(ntb, reg);
2296		outsz = sizeof(uint64_t);
2297	} else {
2298		switch (sz) {
2299		case NTB_REG_64:
2300			if (pci)
2301				umv = pci_read_config(ntb->device, reg, 8);
2302			else
2303				umv = ntb_reg_read(8, reg);
2304			outsz = sizeof(uint64_t);
2305			break;
2306		case NTB_REG_32:
2307			if (pci)
2308				umv = pci_read_config(ntb->device, reg, 4);
2309			else
2310				umv = ntb_reg_read(4, reg);
2311			outsz = sizeof(uint32_t);
2312			break;
2313		case NTB_REG_16:
2314			if (pci)
2315				umv = pci_read_config(ntb->device, reg, 2);
2316			else
2317				umv = ntb_reg_read(2, reg);
2318			outsz = sizeof(uint16_t);
2319			break;
2320		case NTB_REG_8:
2321			if (pci)
2322				umv = pci_read_config(ntb->device, reg, 1);
2323			else
2324				umv = ntb_reg_read(1, reg);
2325			outsz = sizeof(uint8_t);
2326			break;
2327		default:
2328			panic("bogus");
2329			break;
2330		}
2331	}
2332
2333	/* Encode bigendian so that sysctl -x is legible. */
2334	be64enc(be, umv);
2335	outp = ((char *)be) + sizeof(umv) - outsz;
2336
2337	error = SYSCTL_OUT(req, outp, outsz);
2338	if (error || !req->newptr)
2339		return (error);
2340	return (EINVAL);
2341}
2342
2343/*
2344 * Public API to the rest of the OS
2345 */
2346
2347/**
2348 * ntb_get_max_spads() - get the total scratch regs usable
2349 * @ntb: pointer to ntb_softc instance
2350 *
2351 * This function returns the max 32bit scratchpad registers usable by the
2352 * upper layer.
2353 *
2354 * RETURNS: total number of scratch pad registers available
2355 */
2356uint8_t
2357ntb_get_max_spads(struct ntb_softc *ntb)
2358{
2359
2360	return (ntb->spad_count);
2361}
2362
2363uint8_t
2364ntb_mw_count(struct ntb_softc *ntb)
2365{
2366
2367	return (ntb->mw_count);
2368}
2369
2370/**
2371 * ntb_spad_write() - write to the secondary scratchpad register
2372 * @ntb: pointer to ntb_softc instance
2373 * @idx: index to the scratchpad register, 0 based
2374 * @val: the data value to put into the register
2375 *
2376 * This function allows writing of a 32bit value to the indexed scratchpad
2377 * register. The register resides on the secondary (external) side.
2378 *
2379 * RETURNS: An appropriate ERRNO error value on error, or zero for success.
2380 */
2381int
2382ntb_spad_write(struct ntb_softc *ntb, unsigned int idx, uint32_t val)
2383{
2384
2385	if (idx >= ntb->spad_count)
2386		return (EINVAL);
2387
2388	ntb_reg_write(4, ntb->self_reg->spad + idx * 4, val);
2389
2390	return (0);
2391}
2392
2393/**
2394 * ntb_spad_read() - read from the primary scratchpad register
2395 * @ntb: pointer to ntb_softc instance
2396 * @idx: index to scratchpad register, 0 based
2397 * @val: pointer to 32bit integer for storing the register value
2398 *
2399 * This function allows reading of the 32bit scratchpad register on
2400 * the primary (internal) side.
2401 *
2402 * RETURNS: An appropriate ERRNO error value on error, or zero for success.
2403 */
2404int
2405ntb_spad_read(struct ntb_softc *ntb, unsigned int idx, uint32_t *val)
2406{
2407
2408	if (idx >= ntb->spad_count)
2409		return (EINVAL);
2410
2411	*val = ntb_reg_read(4, ntb->self_reg->spad + idx * 4);
2412
2413	return (0);
2414}
2415
2416/**
2417 * ntb_peer_spad_write() - write to the secondary scratchpad register
2418 * @ntb: pointer to ntb_softc instance
2419 * @idx: index to the scratchpad register, 0 based
2420 * @val: the data value to put into the register
2421 *
2422 * This function allows writing of a 32bit value to the indexed scratchpad
2423 * register. The register resides on the secondary (external) side.
2424 *
2425 * RETURNS: An appropriate ERRNO error value on error, or zero for success.
2426 */
2427int
2428ntb_peer_spad_write(struct ntb_softc *ntb, unsigned int idx, uint32_t val)
2429{
2430
2431	if (idx >= ntb->spad_count)
2432		return (EINVAL);
2433
2434	if (HAS_FEATURE(NTB_SDOORBELL_LOCKUP))
2435		ntb_mw_write(4, XEON_SPAD_OFFSET + idx * 4, val);
2436	else
2437		ntb_reg_write(4, ntb->peer_reg->spad + idx * 4, val);
2438
2439	return (0);
2440}
2441
2442/**
2443 * ntb_peer_spad_read() - read from the primary scratchpad register
2444 * @ntb: pointer to ntb_softc instance
2445 * @idx: index to scratchpad register, 0 based
2446 * @val: pointer to 32bit integer for storing the register value
2447 *
2448 * This function allows reading of the 32bit scratchpad register on
2449 * the primary (internal) side.
2450 *
2451 * RETURNS: An appropriate ERRNO error value on error, or zero for success.
2452 */
2453int
2454ntb_peer_spad_read(struct ntb_softc *ntb, unsigned int idx, uint32_t *val)
2455{
2456
2457	if (idx >= ntb->spad_count)
2458		return (EINVAL);
2459
2460	if (HAS_FEATURE(NTB_SDOORBELL_LOCKUP))
2461		*val = ntb_mw_read(4, XEON_SPAD_OFFSET + idx * 4);
2462	else
2463		*val = ntb_reg_read(4, ntb->peer_reg->spad + idx * 4);
2464
2465	return (0);
2466}
2467
2468/*
2469 * ntb_mw_get_range() - get the range of a memory window
2470 * @ntb:        NTB device context
2471 * @idx:        Memory window number
2472 * @base:       OUT - the base address for mapping the memory window
2473 * @size:       OUT - the size for mapping the memory window
2474 * @align:      OUT - the base alignment for translating the memory window
2475 * @align_size: OUT - the size alignment for translating the memory window
2476 *
2477 * Get the range of a memory window.  NULL may be given for any output
2478 * parameter if the value is not needed.  The base and size may be used for
2479 * mapping the memory window, to access the peer memory.  The alignment and
2480 * size may be used for translating the memory window, for the peer to access
2481 * memory on the local system.
2482 *
2483 * Return: Zero on success, otherwise an error number.
2484 */
2485int
2486ntb_mw_get_range(struct ntb_softc *ntb, unsigned mw_idx, vm_paddr_t *base,
2487    caddr_t *vbase, size_t *size, size_t *align, size_t *align_size)
2488{
2489	struct ntb_pci_bar_info *bar;
2490	size_t bar_b2b_off;
2491
2492	if (mw_idx >= ntb_mw_count(ntb))
2493		return (EINVAL);
2494
2495	bar = &ntb->bar_info[ntb_mw_to_bar(ntb, mw_idx)];
2496	bar_b2b_off = 0;
2497	if (mw_idx == ntb->b2b_mw_idx) {
2498		KASSERT(ntb->b2b_off != 0,
2499		    ("user shouldn't get non-shared b2b mw"));
2500		bar_b2b_off = ntb->b2b_off;
2501	}
2502
2503	if (base != NULL)
2504		*base = bar->pbase + bar_b2b_off;
2505	if (vbase != NULL)
2506		*vbase = bar->vbase + bar_b2b_off;
2507	if (size != NULL)
2508		*size = bar->size - bar_b2b_off;
2509	if (align != NULL)
2510		*align = bar->size;
2511	if (align_size != NULL)
2512		*align_size = 1;
2513	return (0);
2514}
2515
2516/*
2517 * ntb_mw_set_trans() - set the translation of a memory window
2518 * @ntb:        NTB device context
2519 * @idx:        Memory window number
2520 * @addr:       The dma address local memory to expose to the peer
2521 * @size:       The size of the local memory to expose to the peer
2522 *
2523 * Set the translation of a memory window.  The peer may access local memory
2524 * through the window starting at the address, up to the size.  The address
2525 * must be aligned to the alignment specified by ntb_mw_get_range().  The size
2526 * must be aligned to the size alignment specified by ntb_mw_get_range().
2527 *
2528 * Return: Zero on success, otherwise an error number.
2529 */
2530int
2531ntb_mw_set_trans(struct ntb_softc *ntb, unsigned idx, bus_addr_t addr,
2532    size_t size)
2533{
2534	struct ntb_pci_bar_info *bar;
2535	uint64_t base, limit, reg_val;
2536	size_t bar_size, mw_size;
2537	uint32_t base_reg, xlat_reg, limit_reg;
2538	enum ntb_bar bar_num;
2539
2540	if (idx >= ntb_mw_count(ntb))
2541		return (EINVAL);
2542
2543	bar_num = ntb_mw_to_bar(ntb, idx);
2544	bar = &ntb->bar_info[bar_num];
2545
2546	bar_size = bar->size;
2547	if (idx == ntb->b2b_mw_idx)
2548		mw_size = bar_size - ntb->b2b_off;
2549	else
2550		mw_size = bar_size;
2551
2552	/* Hardware requires that addr is aligned to bar size */
2553	if ((addr & (bar_size - 1)) != 0)
2554		return (EINVAL);
2555
2556	if (size > mw_size)
2557		return (EINVAL);
2558
2559	bar_get_xlat_params(ntb, bar_num, &base_reg, &xlat_reg, &limit_reg);
2560
2561	limit = 0;
2562	if (bar_is_64bit(ntb, bar_num)) {
2563		base = ntb_reg_read(8, base_reg);
2564
2565		if (limit_reg != 0 && size != mw_size)
2566			limit = base + size;
2567
2568		/* Set and verify translation address */
2569		ntb_reg_write(8, xlat_reg, addr);
2570		reg_val = ntb_reg_read(8, xlat_reg);
2571		if (reg_val != addr) {
2572			ntb_reg_write(8, xlat_reg, 0);
2573			return (EIO);
2574		}
2575
2576		/* Set and verify the limit */
2577		ntb_reg_write(8, limit_reg, limit);
2578		reg_val = ntb_reg_read(8, limit_reg);
2579		if (reg_val != limit) {
2580			ntb_reg_write(8, limit_reg, base);
2581			ntb_reg_write(8, xlat_reg, 0);
2582			return (EIO);
2583		}
2584	} else {
2585		/* Configure 32-bit (split) BAR MW */
2586
2587		if ((addr & UINT32_MAX) != addr)
2588			return (EINVAL);
2589		if (((addr + size) & UINT32_MAX) != (addr + size))
2590			return (EINVAL);
2591
2592		base = ntb_reg_read(4, base_reg);
2593
2594		if (limit_reg != 0 && size != mw_size)
2595			limit = base + size;
2596
2597		/* Set and verify translation address */
2598		ntb_reg_write(4, xlat_reg, addr);
2599		reg_val = ntb_reg_read(4, xlat_reg);
2600		if (reg_val != addr) {
2601			ntb_reg_write(4, xlat_reg, 0);
2602			return (EIO);
2603		}
2604
2605		/* Set and verify the limit */
2606		ntb_reg_write(4, limit_reg, limit);
2607		reg_val = ntb_reg_read(4, limit_reg);
2608		if (reg_val != limit) {
2609			ntb_reg_write(4, limit_reg, base);
2610			ntb_reg_write(4, xlat_reg, 0);
2611			return (EIO);
2612		}
2613	}
2614	return (0);
2615}
2616
2617/*
2618 * ntb_mw_clear_trans() - clear the translation of a memory window
2619 * @ntb:	NTB device context
2620 * @idx:	Memory window number
2621 *
2622 * Clear the translation of a memory window.  The peer may no longer access
2623 * local memory through the window.
2624 *
2625 * Return: Zero on success, otherwise an error number.
2626 */
2627int
2628ntb_mw_clear_trans(struct ntb_softc *ntb, unsigned mw_idx)
2629{
2630
2631	return (ntb_mw_set_trans(ntb, mw_idx, 0, 0));
2632}
2633
2634/*
2635 * ntb_mw_get_wc - Get the write-combine status of a memory window
2636 *
2637 * Returns:  Zero on success, setting *wc; otherwise an error number (e.g. if
2638 * idx is an invalid memory window).
2639 */
2640int
2641ntb_mw_get_wc(struct ntb_softc *ntb, unsigned idx, bool *wc)
2642{
2643	struct ntb_pci_bar_info *bar;
2644
2645	if (idx >= ntb_mw_count(ntb))
2646		return (EINVAL);
2647
2648	bar = &ntb->bar_info[ntb_mw_to_bar(ntb, idx)];
2649	*wc = bar->mapped_wc;
2650	return (0);
2651}
2652
2653/*
2654 * ntb_mw_set_wc - Set the write-combine status of a memory window
2655 *
2656 * If 'wc' matches the current status, this does nothing and succeeds.
2657 *
2658 * Returns:  Zero on success, setting the caching attribute on the virtual
2659 * mapping of the BAR; otherwise an error number (e.g. if idx is an invalid
2660 * memory window, or if changing the caching attribute fails).
2661 */
2662int
2663ntb_mw_set_wc(struct ntb_softc *ntb, unsigned idx, bool wc)
2664{
2665	struct ntb_pci_bar_info *bar;
2666	vm_memattr_t attr;
2667	int rc;
2668
2669	if (idx >= ntb_mw_count(ntb))
2670		return (EINVAL);
2671
2672	bar = &ntb->bar_info[ntb_mw_to_bar(ntb, idx)];
2673	if (bar->mapped_wc == wc)
2674		return (0);
2675
2676	if (wc)
2677		attr = VM_MEMATTR_WRITE_COMBINING;
2678	else
2679		attr = VM_MEMATTR_DEFAULT;
2680
2681	rc = pmap_change_attr((vm_offset_t)bar->vbase, bar->size, attr);
2682	if (rc == 0)
2683		bar->mapped_wc = wc;
2684
2685	return (rc);
2686}
2687
2688/**
2689 * ntb_peer_db_set() - Set the doorbell on the secondary/external side
2690 * @ntb: pointer to ntb_softc instance
2691 * @bit: doorbell bits to ring
2692 *
2693 * This function allows triggering of a doorbell on the secondary/external
2694 * side that will initiate an interrupt on the remote host
2695 */
2696void
2697ntb_peer_db_set(struct ntb_softc *ntb, uint64_t bit)
2698{
2699
2700	if (HAS_FEATURE(NTB_SDOORBELL_LOCKUP)) {
2701		ntb_mw_write(2, XEON_PDOORBELL_OFFSET, bit);
2702		return;
2703	}
2704
2705	db_iowrite(ntb, ntb->peer_reg->db_bell, bit);
2706}
2707
2708/*
2709 * ntb_get_peer_db_addr() - Return the address of the remote doorbell register,
2710 * as well as the size of the register (via *sz_out).
2711 *
2712 * This function allows a caller using I/OAT DMA to chain the remote doorbell
2713 * ring to its memory window write.
2714 *
2715 * Note that writing the peer doorbell via a memory window will *not* generate
2716 * an interrupt on the remote host; that must be done seperately.
2717 */
2718bus_addr_t
2719ntb_get_peer_db_addr(struct ntb_softc *ntb, vm_size_t *sz_out)
2720{
2721	struct ntb_pci_bar_info *bar;
2722	uint64_t regoff;
2723
2724	KASSERT(sz_out != NULL, ("must be non-NULL"));
2725
2726	if (!HAS_FEATURE(NTB_SDOORBELL_LOCKUP)) {
2727		bar = &ntb->bar_info[NTB_CONFIG_BAR];
2728		regoff = ntb->peer_reg->db_bell;
2729	} else {
2730		KASSERT((HAS_FEATURE(NTB_SPLIT_BAR) && ntb->mw_count == 2) ||
2731		    (!HAS_FEATURE(NTB_SPLIT_BAR) && ntb->mw_count == 1),
2732		    ("mw_count invalid after setup"));
2733		KASSERT(ntb->b2b_mw_idx != B2B_MW_DISABLED,
2734		    ("invalid b2b idx"));
2735
2736		bar = &ntb->bar_info[ntb_mw_to_bar(ntb, ntb->b2b_mw_idx)];
2737		regoff = XEON_PDOORBELL_OFFSET;
2738	}
2739	KASSERT(bar->pci_bus_tag != X86_BUS_SPACE_IO, ("uh oh"));
2740
2741	*sz_out = ntb->reg->db_size;
2742	/* HACK: Specific to current x86 bus implementation. */
2743	return ((uint64_t)bar->pci_bus_handle + regoff);
2744}
2745
2746/*
2747 * ntb_db_valid_mask() - get a mask of doorbell bits supported by the ntb
2748 * @ntb:	NTB device context
2749 *
2750 * Hardware may support different number or arrangement of doorbell bits.
2751 *
2752 * Return: A mask of doorbell bits supported by the ntb.
2753 */
2754uint64_t
2755ntb_db_valid_mask(struct ntb_softc *ntb)
2756{
2757
2758	return (ntb->db_valid_mask);
2759}
2760
2761/*
2762 * ntb_db_vector_mask() - get a mask of doorbell bits serviced by a vector
2763 * @ntb:	NTB device context
2764 * @vector:	Doorbell vector number
2765 *
2766 * Each interrupt vector may have a different number or arrangement of bits.
2767 *
2768 * Return: A mask of doorbell bits serviced by a vector.
2769 */
2770uint64_t
2771ntb_db_vector_mask(struct ntb_softc *ntb, uint32_t vector)
2772{
2773
2774	if (vector > ntb->db_vec_count)
2775		return (0);
2776	return (ntb->db_valid_mask & ntb_vec_mask(ntb, vector));
2777}
2778
2779/**
2780 * ntb_link_is_up() - get the current ntb link state
2781 * @ntb:        NTB device context
2782 * @speed:      OUT - The link speed expressed as PCIe generation number
2783 * @width:      OUT - The link width expressed as the number of PCIe lanes
2784 *
2785 * RETURNS: true or false based on the hardware link state
2786 */
2787bool
2788ntb_link_is_up(struct ntb_softc *ntb, enum ntb_speed *speed,
2789    enum ntb_width *width)
2790{
2791
2792	if (speed != NULL)
2793		*speed = ntb_link_sta_speed(ntb);
2794	if (width != NULL)
2795		*width = ntb_link_sta_width(ntb);
2796	return (link_is_up(ntb));
2797}
2798
2799static void
2800save_bar_parameters(struct ntb_pci_bar_info *bar)
2801{
2802
2803	bar->pci_bus_tag = rman_get_bustag(bar->pci_resource);
2804	bar->pci_bus_handle = rman_get_bushandle(bar->pci_resource);
2805	bar->pbase = rman_get_start(bar->pci_resource);
2806	bar->size = rman_get_size(bar->pci_resource);
2807	bar->vbase = rman_get_virtual(bar->pci_resource);
2808}
2809
2810device_t
2811ntb_get_device(struct ntb_softc *ntb)
2812{
2813
2814	return (ntb->device);
2815}
2816
2817/* Export HW-specific errata information. */
2818bool
2819ntb_has_feature(struct ntb_softc *ntb, uint32_t feature)
2820{
2821
2822	return (HAS_FEATURE(feature));
2823}
2824