1 /*-
2 * Copyright (c) 2016-2017 Alexander Motin <mav@FreeBSD.org>
3 * Copyright (C) 2013 Intel Corporation
4 * Copyright (C) 2015 EMC Corporation
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29 /*
30 * The Non-Transparent Bridge (NTB) is a device that allows you to connect
31 * two or more systems using a PCI-e links, providing remote memory access.
32 *
33 * This module contains a driver for NTB hardware in Intel Xeon/Atom CPUs.
34 *
35 * NOTE: Much of the code in this module is shared with Linux. Any patches may
36 * be picked up and redistributed in Linux with a dual GPL/BSD license.
37 */
38
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41
42 #include <sys/param.h>
43 #include <sys/kernel.h>
44 #include <sys/systm.h>
45 #include <sys/bus.h>
46 #include <sys/endian.h>
47 #include <sys/interrupt.h>
48 #include <sys/lock.h>
49 #include <sys/malloc.h>
50 #include <sys/module.h>
51 #include <sys/mutex.h>
52 #include <sys/pciio.h>
53 #include <sys/taskqueue.h>
54 #include <sys/tree.h>
55 #include <sys/queue.h>
56 #include <sys/rman.h>
57 #include <sys/sbuf.h>
58 #include <sys/sysctl.h>
59 #include <vm/vm.h>
60 #include <vm/pmap.h>
61 #include <machine/bus.h>
62 #include <machine/intr_machdep.h>
63 #include <machine/resource.h>
64 #include <dev/pci/pcireg.h>
65 #include <dev/pci/pcivar.h>
66 #include <dev/iommu/iommu.h>
67
68 #include "ntb_hw_intel.h"
69 #include "../ntb.h"
70
71 #define MAX_MSIX_INTERRUPTS \
72 MAX(MAX(XEON_DB_COUNT, ATOM_DB_COUNT), XEON_GEN3_DB_COUNT)
73
74 #define NTB_HB_TIMEOUT 1 /* second */
75 #define ATOM_LINK_RECOVERY_TIME 500 /* ms */
76 #define BAR_HIGH_MASK (~((1ull << 12) - 1))
77
78 #define NTB_MSIX_VER_GUARD 0xaabbccdd
79 #define NTB_MSIX_RECEIVED 0xe0f0e0f0
80
81 /*
82 * PCI constants could be somewhere more generic, but aren't defined/used in
83 * pci.c.
84 */
85 #define PCI_MSIX_ENTRY_SIZE 16
86 #define PCI_MSIX_ENTRY_LOWER_ADDR 0
87 #define PCI_MSIX_ENTRY_UPPER_ADDR 4
88 #define PCI_MSIX_ENTRY_DATA 8
89
90 enum ntb_device_type {
91 NTB_XEON_GEN1,
92 NTB_XEON_GEN3,
93 NTB_ATOM
94 };
95
96 /* ntb_conn_type are hardware numbers, cannot change. */
97 enum ntb_conn_type {
98 NTB_CONN_TRANSPARENT = 0,
99 NTB_CONN_B2B = 1,
100 NTB_CONN_RP = 2,
101 };
102
103 enum ntb_b2b_direction {
104 NTB_DEV_USD = 0,
105 NTB_DEV_DSD = 1,
106 };
107
108 enum ntb_bar {
109 NTB_CONFIG_BAR = 0,
110 NTB_B2B_BAR_1,
111 NTB_B2B_BAR_2,
112 NTB_B2B_BAR_3,
113 NTB_MAX_BARS
114 };
115
116 enum {
117 NTB_MSIX_GUARD = 0,
118 NTB_MSIX_DATA0,
119 NTB_MSIX_DATA1,
120 NTB_MSIX_DATA2,
121 NTB_MSIX_OFS0,
122 NTB_MSIX_OFS1,
123 NTB_MSIX_OFS2,
124 NTB_MSIX_DONE,
125 NTB_MAX_MSIX_SPAD
126 };
127
128 /* Device features and workarounds */
129 #define HAS_FEATURE(ntb, feature) \
130 (((ntb)->features & (feature)) != 0)
131
132 struct ntb_hw_info {
133 uint32_t device_id;
134 const char *desc;
135 enum ntb_device_type type;
136 uint32_t features;
137 };
138
139 struct ntb_pci_bar_info {
140 bus_space_tag_t pci_bus_tag;
141 bus_space_handle_t pci_bus_handle;
142 int pci_resource_id;
143 struct resource *pci_resource;
144 vm_paddr_t pbase;
145 caddr_t vbase;
146 vm_size_t size;
147 vm_memattr_t map_mode;
148
149 /* Configuration register offsets */
150 uint32_t psz_off;
151 uint32_t ssz_off;
152 uint32_t pbarxlat_off;
153 };
154
155 struct ntb_int_info {
156 struct resource *res;
157 int rid;
158 void *tag;
159 };
160
161 struct ntb_vec {
162 struct ntb_softc *ntb;
163 uint32_t num;
164 unsigned masked;
165 };
166
167 struct ntb_reg {
168 uint32_t ntb_ctl;
169 uint32_t lnk_sta;
170 uint8_t db_size;
171 unsigned mw_bar[NTB_MAX_BARS];
172 };
173
174 struct ntb_alt_reg {
175 uint32_t db_bell;
176 uint32_t db_mask;
177 uint32_t spad;
178 };
179
180 struct ntb_xlat_reg {
181 uint32_t bar0_base;
182 uint32_t bar2_base;
183 uint32_t bar4_base;
184 uint32_t bar5_base;
185
186 uint32_t bar2_xlat;
187 uint32_t bar4_xlat;
188 uint32_t bar5_xlat;
189
190 uint32_t bar2_limit;
191 uint32_t bar4_limit;
192 uint32_t bar5_limit;
193 };
194
195 struct ntb_b2b_addr {
196 uint64_t bar0_addr;
197 uint64_t bar2_addr64;
198 uint64_t bar4_addr64;
199 uint64_t bar4_addr32;
200 uint64_t bar5_addr32;
201 };
202
203 struct ntb_msix_data {
204 uint32_t nmd_ofs;
205 uint32_t nmd_data;
206 };
207
208 struct ntb_softc {
209 /* ntb.c context. Do not move! Must go first! */
210 void *ntb_store;
211
212 device_t device;
213 enum ntb_device_type type;
214 uint32_t features;
215
216 struct ntb_pci_bar_info bar_info[NTB_MAX_BARS];
217 struct ntb_int_info int_info[MAX_MSIX_INTERRUPTS];
218 uint32_t allocated_interrupts;
219
220 struct ntb_msix_data peer_msix_data[XEON_NONLINK_DB_MSIX_BITS];
221 struct ntb_msix_data msix_data[XEON_NONLINK_DB_MSIX_BITS];
222 bool peer_msix_good;
223 bool peer_msix_done;
224 struct ntb_pci_bar_info *peer_lapic_bar;
225 struct callout peer_msix_work;
226
227 bus_dma_tag_t bar0_dma_tag;
228 bus_dmamap_t bar0_dma_map;
229
230 struct callout heartbeat_timer;
231 struct callout lr_timer;
232
233 struct ntb_vec *msix_vec;
234
235 uint32_t ppd;
236 enum ntb_conn_type conn_type;
237 enum ntb_b2b_direction dev_type;
238
239 /* Offset of peer bar0 in B2B BAR */
240 uint64_t b2b_off;
241 /* Memory window used to access peer bar0 */
242 #define B2B_MW_DISABLED UINT8_MAX
243 uint8_t b2b_mw_idx;
244 uint32_t msix_xlat;
245 uint8_t msix_mw_idx;
246
247 uint8_t mw_count;
248 uint8_t spad_count;
249 uint8_t db_count;
250 uint8_t db_vec_count;
251 uint8_t db_vec_shift;
252
253 /* Protects local db_mask. */
254 #define DB_MASK_LOCK(sc) mtx_lock_spin(&(sc)->db_mask_lock)
255 #define DB_MASK_UNLOCK(sc) mtx_unlock_spin(&(sc)->db_mask_lock)
256 #define DB_MASK_ASSERT(sc,f) mtx_assert(&(sc)->db_mask_lock, (f))
257 struct mtx db_mask_lock;
258
259 volatile uint32_t ntb_ctl;
260 volatile uint32_t lnk_sta;
261
262 uint64_t db_valid_mask;
263 uint64_t db_link_mask;
264 uint64_t db_mask;
265 uint64_t fake_db; /* NTB_SB01BASE_LOCKUP*/
266 uint64_t force_db; /* NTB_SB01BASE_LOCKUP*/
267
268 int last_ts; /* ticks @ last irq */
269
270 const struct ntb_reg *reg;
271 const struct ntb_alt_reg *self_reg;
272 const struct ntb_alt_reg *peer_reg;
273 const struct ntb_xlat_reg *xlat_reg;
274 };
275
276 #ifdef __i386__
277 static __inline uint64_t
278 bus_space_read_8(bus_space_tag_t tag, bus_space_handle_t handle,
279 bus_size_t offset)
280 {
281
282 return (bus_space_read_4(tag, handle, offset) |
283 ((uint64_t)bus_space_read_4(tag, handle, offset + 4)) << 32);
284 }
285
286 static __inline void
287 bus_space_write_8(bus_space_tag_t tag, bus_space_handle_t handle,
288 bus_size_t offset, uint64_t val)
289 {
290
291 bus_space_write_4(tag, handle, offset, val);
292 bus_space_write_4(tag, handle, offset + 4, val >> 32);
293 }
294 #endif
295
296 #define intel_ntb_bar_read(SIZE, bar, offset) \
297 bus_space_read_ ## SIZE (ntb->bar_info[(bar)].pci_bus_tag, \
298 ntb->bar_info[(bar)].pci_bus_handle, (offset))
299 #define intel_ntb_bar_write(SIZE, bar, offset, val) \
300 bus_space_write_ ## SIZE (ntb->bar_info[(bar)].pci_bus_tag, \
301 ntb->bar_info[(bar)].pci_bus_handle, (offset), (val))
302 #define intel_ntb_reg_read(SIZE, offset) \
303 intel_ntb_bar_read(SIZE, NTB_CONFIG_BAR, offset)
304 #define intel_ntb_reg_write(SIZE, offset, val) \
305 intel_ntb_bar_write(SIZE, NTB_CONFIG_BAR, offset, val)
306 #define intel_ntb_mw_read(SIZE, offset) \
307 intel_ntb_bar_read(SIZE, intel_ntb_mw_to_bar(ntb, ntb->b2b_mw_idx), \
308 offset)
309 #define intel_ntb_mw_write(SIZE, offset, val) \
310 intel_ntb_bar_write(SIZE, intel_ntb_mw_to_bar(ntb, ntb->b2b_mw_idx), \
311 offset, val)
312
313 static int intel_ntb_probe(device_t device);
314 static int intel_ntb_attach(device_t device);
315 static int intel_ntb_detach(device_t device);
316 static uint64_t intel_ntb_db_valid_mask(device_t dev);
317 static void intel_ntb_spad_clear(device_t dev);
318 static uint64_t intel_ntb_db_vector_mask(device_t dev, uint32_t vector);
319 static bool intel_ntb_link_is_up(device_t dev, enum ntb_speed *speed,
320 enum ntb_width *width);
321 static int intel_ntb_link_enable(device_t dev, enum ntb_speed speed,
322 enum ntb_width width);
323 static int intel_ntb_link_disable(device_t dev);
324 static int intel_ntb_spad_read(device_t dev, unsigned int idx, uint32_t *val);
325 static int intel_ntb_peer_spad_write(device_t dev, unsigned int idx, uint32_t val);
326
327 static unsigned intel_ntb_user_mw_to_idx(struct ntb_softc *, unsigned uidx);
328 static inline enum ntb_bar intel_ntb_mw_to_bar(struct ntb_softc *, unsigned mw);
329 static inline bool bar_is_64bit(struct ntb_softc *, enum ntb_bar);
330 static inline void bar_get_xlat_params(struct ntb_softc *, enum ntb_bar,
331 uint32_t *base, uint32_t *xlat, uint32_t *lmt);
332 static int intel_ntb_map_pci_bars(struct ntb_softc *ntb);
333 static int intel_ntb_mw_set_wc_internal(struct ntb_softc *, unsigned idx,
334 vm_memattr_t);
335 static void print_map_success(struct ntb_softc *, struct ntb_pci_bar_info *,
336 const char *);
337 static int map_mmr_bar(struct ntb_softc *ntb, struct ntb_pci_bar_info *bar);
338 static int map_memory_window_bar(struct ntb_softc *ntb,
339 struct ntb_pci_bar_info *bar);
340 static void intel_ntb_unmap_pci_bar(struct ntb_softc *ntb);
341 static int intel_ntb_remap_msix(device_t, uint32_t desired, uint32_t avail);
342 static int intel_ntb_init_isr(struct ntb_softc *ntb);
343 static int intel_ntb_xeon_gen3_init_isr(struct ntb_softc *ntb);
344 static int intel_ntb_setup_legacy_interrupt(struct ntb_softc *ntb);
345 static int intel_ntb_setup_msix(struct ntb_softc *ntb, uint32_t num_vectors);
346 static void intel_ntb_teardown_interrupts(struct ntb_softc *ntb);
347 static inline uint64_t intel_ntb_vec_mask(struct ntb_softc *, uint64_t db_vector);
348 static void intel_ntb_interrupt(struct ntb_softc *, uint32_t vec);
349 static void ndev_vec_isr(void *arg);
350 static void ndev_irq_isr(void *arg);
351 static inline uint64_t db_ioread(struct ntb_softc *, uint64_t regoff);
352 static inline void db_iowrite(struct ntb_softc *, uint64_t regoff, uint64_t);
353 static inline void db_iowrite_raw(struct ntb_softc *, uint64_t regoff, uint64_t);
354 static int intel_ntb_create_msix_vec(struct ntb_softc *ntb, uint32_t num_vectors);
355 static void intel_ntb_free_msix_vec(struct ntb_softc *ntb);
356 static void intel_ntb_get_msix_info(struct ntb_softc *ntb);
357 static void intel_ntb_exchange_msix(void *);
358 static struct ntb_hw_info *intel_ntb_get_device_info(uint32_t device_id);
359 static void intel_ntb_detect_max_mw(struct ntb_softc *ntb);
360 static int intel_ntb_detect_xeon(struct ntb_softc *ntb);
361 static int intel_ntb_detect_xeon_gen3(struct ntb_softc *ntb);
362 static int intel_ntb_detect_atom(struct ntb_softc *ntb);
363 static int intel_ntb_xeon_init_dev(struct ntb_softc *ntb);
364 static int intel_ntb_xeon_gen3_init_dev(struct ntb_softc *ntb);
365 static int intel_ntb_atom_init_dev(struct ntb_softc *ntb);
366 static void intel_ntb_teardown_xeon(struct ntb_softc *ntb);
367 static void configure_atom_secondary_side_bars(struct ntb_softc *ntb);
368 static void xeon_reset_sbar_size(struct ntb_softc *, enum ntb_bar idx,
369 enum ntb_bar regbar);
370 static void xeon_set_sbar_base_and_limit(struct ntb_softc *,
371 uint64_t base_addr, enum ntb_bar idx, enum ntb_bar regbar);
372 static void xeon_set_pbar_xlat(struct ntb_softc *, uint64_t base_addr,
373 enum ntb_bar idx);
374 static int xeon_setup_b2b_mw(struct ntb_softc *,
375 const struct ntb_b2b_addr *addr, const struct ntb_b2b_addr *peer_addr);
376 static int xeon_gen3_setup_b2b_mw(struct ntb_softc *);
377 static int intel_ntb_mw_set_trans(device_t dev, unsigned idx, bus_addr_t addr,
378 size_t size);
379 static inline bool link_is_up(struct ntb_softc *ntb);
380 static inline bool _xeon_link_is_up(struct ntb_softc *ntb);
381 static inline bool atom_link_is_err(struct ntb_softc *ntb);
382 static inline enum ntb_speed intel_ntb_link_sta_speed(struct ntb_softc *);
383 static inline enum ntb_width intel_ntb_link_sta_width(struct ntb_softc *);
384 static void atom_link_hb(void *arg);
385 static void recover_atom_link(void *arg);
386 static bool intel_ntb_poll_link(struct ntb_softc *ntb);
387 static void save_bar_parameters(struct ntb_pci_bar_info *bar);
388 static void intel_ntb_sysctl_init(struct ntb_softc *);
389 static int sysctl_handle_features(SYSCTL_HANDLER_ARGS);
390 static int sysctl_handle_link_admin(SYSCTL_HANDLER_ARGS);
391 static int sysctl_handle_link_status_human(SYSCTL_HANDLER_ARGS);
392 static int sysctl_handle_link_status(SYSCTL_HANDLER_ARGS);
393 static int sysctl_handle_register(SYSCTL_HANDLER_ARGS);
394
395 static unsigned g_ntb_hw_debug_level;
396 SYSCTL_UINT(_hw_ntb, OID_AUTO, debug_level, CTLFLAG_RWTUN,
397 &g_ntb_hw_debug_level, 0, "ntb_hw log level -- higher is more verbose");
398 #define intel_ntb_printf(lvl, ...) do { \
399 if ((lvl) <= g_ntb_hw_debug_level) { \
400 device_printf(ntb->device, __VA_ARGS__); \
401 } \
402 } while (0)
403
404 #define _NTB_PAT_UC 0
405 #define _NTB_PAT_WC 1
406 #define _NTB_PAT_WT 4
407 #define _NTB_PAT_WP 5
408 #define _NTB_PAT_WB 6
409 #define _NTB_PAT_UCM 7
410 static unsigned g_ntb_mw_pat = _NTB_PAT_UC;
411 SYSCTL_UINT(_hw_ntb, OID_AUTO, default_mw_pat, CTLFLAG_RDTUN,
412 &g_ntb_mw_pat, 0, "Configure the default memory window cache flags (PAT): "
413 "UC: " __XSTRING(_NTB_PAT_UC) ", "
414 "WC: " __XSTRING(_NTB_PAT_WC) ", "
415 "WT: " __XSTRING(_NTB_PAT_WT) ", "
416 "WP: " __XSTRING(_NTB_PAT_WP) ", "
417 "WB: " __XSTRING(_NTB_PAT_WB) ", "
418 "UC-: " __XSTRING(_NTB_PAT_UCM));
419
420 static inline vm_memattr_t
421 intel_ntb_pat_flags(void)
422 {
423
424 switch (g_ntb_mw_pat) {
425 case _NTB_PAT_WC:
426 return (VM_MEMATTR_WRITE_COMBINING);
427 case _NTB_PAT_WT:
428 return (VM_MEMATTR_WRITE_THROUGH);
429 case _NTB_PAT_WP:
430 return (VM_MEMATTR_WRITE_PROTECTED);
431 case _NTB_PAT_WB:
432 return (VM_MEMATTR_WRITE_BACK);
433 case _NTB_PAT_UCM:
434 return (VM_MEMATTR_WEAK_UNCACHEABLE);
435 case _NTB_PAT_UC:
436 /* FALLTHROUGH */
437 default:
438 return (VM_MEMATTR_UNCACHEABLE);
439 }
440 }
441
442 /*
443 * Well, this obviously doesn't belong here, but it doesn't seem to exist
444 * anywhere better yet.
445 */
446 static inline const char *
447 intel_ntb_vm_memattr_to_str(vm_memattr_t pat)
448 {
449
450 switch (pat) {
451 case VM_MEMATTR_WRITE_COMBINING:
452 return ("WRITE_COMBINING");
453 case VM_MEMATTR_WRITE_THROUGH:
454 return ("WRITE_THROUGH");
455 case VM_MEMATTR_WRITE_PROTECTED:
456 return ("WRITE_PROTECTED");
457 case VM_MEMATTR_WRITE_BACK:
458 return ("WRITE_BACK");
459 case VM_MEMATTR_WEAK_UNCACHEABLE:
460 return ("UNCACHED");
461 case VM_MEMATTR_UNCACHEABLE:
462 return ("UNCACHEABLE");
463 default:
464 return ("UNKNOWN");
465 }
466 }
467
468 static int g_ntb_msix_idx = 1;
469 SYSCTL_INT(_hw_ntb, OID_AUTO, msix_mw_idx, CTLFLAG_RDTUN, &g_ntb_msix_idx,
470 0, "Use this memory window to access the peer MSIX message complex on "
471 "certain Xeon-based NTB systems, as a workaround for a hardware errata. "
472 "Like b2b_mw_idx, negative values index from the last available memory "
473 "window. (Applies on Xeon platforms with SB01BASE_LOCKUP errata.)");
474
475 static int g_ntb_mw_idx = -1;
476 SYSCTL_INT(_hw_ntb, OID_AUTO, b2b_mw_idx, CTLFLAG_RDTUN, &g_ntb_mw_idx,
477 0, "Use this memory window to access the peer NTB registers. A "
478 "non-negative value starts from the first MW index; a negative value "
479 "starts from the last MW index. The default is -1, i.e., the last "
480 "available memory window. Both sides of the NTB MUST set the same "
481 "value here! (Applies on Xeon platforms with SDOORBELL_LOCKUP errata.)");
482
483 /* Hardware owns the low 16 bits of features. */
484 #define NTB_BAR_SIZE_4K (1 << 0)
485 #define NTB_SDOORBELL_LOCKUP (1 << 1)
486 #define NTB_SB01BASE_LOCKUP (1 << 2)
487 #define NTB_B2BDOORBELL_BIT14 (1 << 3)
488 /* Software/configuration owns the top 16 bits. */
489 #define NTB_SPLIT_BAR (1ull << 16)
490 #define NTB_ONE_MSIX (1ull << 17)
491
492 #define NTB_FEATURES_STR \
493 "\2\21SPLIT_BAR4\04B2B_DOORBELL_BIT14\03SB01BASE_LOCKUP" \
494 "\02SDOORBELL_LOCKUP\01BAR_SIZE_4K"
495
496 static struct ntb_hw_info pci_ids[] = {
497 /* XXX: PS/SS IDs left out until they are supported. */
498 { 0x0C4E8086, "BWD Atom Processor S1200 Non-Transparent Bridge B2B",
499 NTB_ATOM, 0 },
500
501 { 0x37258086, "JSF Xeon C35xx/C55xx Non-Transparent Bridge B2B",
502 NTB_XEON_GEN1, NTB_SDOORBELL_LOCKUP | NTB_B2BDOORBELL_BIT14 },
503 { 0x3C0D8086, "SNB Xeon E5/Core i7 Non-Transparent Bridge B2B",
504 NTB_XEON_GEN1, NTB_SDOORBELL_LOCKUP | NTB_B2BDOORBELL_BIT14 },
505 { 0x0E0D8086, "IVT Xeon E5 V2 Non-Transparent Bridge B2B",
506 NTB_XEON_GEN1, NTB_SDOORBELL_LOCKUP | NTB_B2BDOORBELL_BIT14 |
507 NTB_SB01BASE_LOCKUP | NTB_BAR_SIZE_4K },
508 { 0x2F0D8086, "HSX Xeon E5 V3 Non-Transparent Bridge B2B",
509 NTB_XEON_GEN1, NTB_SDOORBELL_LOCKUP | NTB_B2BDOORBELL_BIT14 |
510 NTB_SB01BASE_LOCKUP },
511 { 0x6F0D8086, "BDX Xeon E5 V4 Non-Transparent Bridge B2B",
512 NTB_XEON_GEN1, NTB_SDOORBELL_LOCKUP | NTB_B2BDOORBELL_BIT14 |
513 NTB_SB01BASE_LOCKUP },
514
515 { 0x201C8086, "SKL Xeon E5 V5 Non-Transparent Bridge B2B",
516 NTB_XEON_GEN3, 0 },
517 };
518
519 static const struct ntb_reg atom_reg = {
520 .ntb_ctl = ATOM_NTBCNTL_OFFSET,
521 .lnk_sta = ATOM_LINK_STATUS_OFFSET,
522 .db_size = sizeof(uint64_t),
523 .mw_bar = { NTB_B2B_BAR_1, NTB_B2B_BAR_2 },
524 };
525
526 static const struct ntb_alt_reg atom_pri_reg = {
527 .db_bell = ATOM_PDOORBELL_OFFSET,
528 .db_mask = ATOM_PDBMSK_OFFSET,
529 .spad = ATOM_SPAD_OFFSET,
530 };
531
532 static const struct ntb_alt_reg atom_b2b_reg = {
533 .db_bell = ATOM_B2B_DOORBELL_OFFSET,
534 .spad = ATOM_B2B_SPAD_OFFSET,
535 };
536
537 static const struct ntb_xlat_reg atom_sec_xlat = {
538 #if 0
539 /* "FIXME" says the Linux driver. */
540 .bar0_base = ATOM_SBAR0BASE_OFFSET,
541 .bar2_base = ATOM_SBAR2BASE_OFFSET,
542 .bar4_base = ATOM_SBAR4BASE_OFFSET,
543
544 .bar2_limit = ATOM_SBAR2LMT_OFFSET,
545 .bar4_limit = ATOM_SBAR4LMT_OFFSET,
546 #endif
547
548 .bar2_xlat = ATOM_SBAR2XLAT_OFFSET,
549 .bar4_xlat = ATOM_SBAR4XLAT_OFFSET,
550 };
551
552 static const struct ntb_reg xeon_reg = {
553 .ntb_ctl = XEON_NTBCNTL_OFFSET,
554 .lnk_sta = XEON_LINK_STATUS_OFFSET,
555 .db_size = sizeof(uint16_t),
556 .mw_bar = { NTB_B2B_BAR_1, NTB_B2B_BAR_2, NTB_B2B_BAR_3 },
557 };
558
559 static const struct ntb_alt_reg xeon_pri_reg = {
560 .db_bell = XEON_PDOORBELL_OFFSET,
561 .db_mask = XEON_PDBMSK_OFFSET,
562 .spad = XEON_SPAD_OFFSET,
563 };
564
565 static const struct ntb_alt_reg xeon_b2b_reg = {
566 .db_bell = XEON_B2B_DOORBELL_OFFSET,
567 .spad = XEON_B2B_SPAD_OFFSET,
568 };
569
570 static const struct ntb_xlat_reg xeon_sec_xlat = {
571 .bar0_base = XEON_SBAR0BASE_OFFSET,
572 .bar2_base = XEON_SBAR2BASE_OFFSET,
573 .bar4_base = XEON_SBAR4BASE_OFFSET,
574 .bar5_base = XEON_SBAR5BASE_OFFSET,
575
576 .bar2_limit = XEON_SBAR2LMT_OFFSET,
577 .bar4_limit = XEON_SBAR4LMT_OFFSET,
578 .bar5_limit = XEON_SBAR5LMT_OFFSET,
579
580 .bar2_xlat = XEON_SBAR2XLAT_OFFSET,
581 .bar4_xlat = XEON_SBAR4XLAT_OFFSET,
582 .bar5_xlat = XEON_SBAR5XLAT_OFFSET,
583 };
584
585 static struct ntb_b2b_addr xeon_b2b_usd_addr = {
586 .bar0_addr = XEON_B2B_BAR0_ADDR,
587 .bar2_addr64 = XEON_B2B_BAR2_ADDR64,
588 .bar4_addr64 = XEON_B2B_BAR4_ADDR64,
589 .bar4_addr32 = XEON_B2B_BAR4_ADDR32,
590 .bar5_addr32 = XEON_B2B_BAR5_ADDR32,
591 };
592
593 static struct ntb_b2b_addr xeon_b2b_dsd_addr = {
594 .bar0_addr = XEON_B2B_BAR0_ADDR,
595 .bar2_addr64 = XEON_B2B_BAR2_ADDR64,
596 .bar4_addr64 = XEON_B2B_BAR4_ADDR64,
597 .bar4_addr32 = XEON_B2B_BAR4_ADDR32,
598 .bar5_addr32 = XEON_B2B_BAR5_ADDR32,
599 };
600
601 static const struct ntb_reg xeon_gen3_reg = {
602 .ntb_ctl = XEON_GEN3_REG_IMNTB_CTRL,
603 .lnk_sta = XEON_GEN3_INT_LNK_STS_OFFSET,
604 .db_size = sizeof(uint32_t),
605 .mw_bar = { NTB_B2B_BAR_1, NTB_B2B_BAR_2 },
606 };
607
608 static const struct ntb_alt_reg xeon_gen3_pri_reg = {
609 .db_bell = XEON_GEN3_REG_EMDOORBELL,
610 .db_mask = XEON_GEN3_REG_IMINT_DISABLE,
611 .spad = XEON_GEN3_REG_IMSPAD,
612 };
613
614 static const struct ntb_alt_reg xeon_gen3_b2b_reg = {
615 .db_bell = XEON_GEN3_REG_IMDOORBELL,
616 .db_mask = XEON_GEN3_REG_EMINT_DISABLE,
617 .spad = XEON_GEN3_REG_IMB2B_SSPAD,
618 };
619
620 static const struct ntb_xlat_reg xeon_gen3_sec_xlat = {
621 .bar0_base = XEON_GEN3_EXT_REG_BAR0BASE,
622 .bar2_base = XEON_GEN3_EXT_REG_BAR1BASE,
623 .bar4_base = XEON_GEN3_EXT_REG_BAR2BASE,
624
625 .bar2_limit = XEON_GEN3_REG_IMBAR1XLIMIT,
626 .bar4_limit = XEON_GEN3_REG_IMBAR2XLIMIT,
627
628 .bar2_xlat = XEON_GEN3_REG_IMBAR1XBASE,
629 .bar4_xlat = XEON_GEN3_REG_IMBAR2XBASE,
630 };
631
632 SYSCTL_NODE(_hw_ntb, OID_AUTO, xeon_b2b, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
633 "B2B MW segment overrides -- MUST be the same on both sides");
634
635 SYSCTL_UQUAD(_hw_ntb_xeon_b2b, OID_AUTO, usd_bar2_addr64, CTLFLAG_RDTUN,
636 &xeon_b2b_usd_addr.bar2_addr64, 0, "If using B2B topology on Xeon "
637 "hardware, use this 64-bit address on the bus between the NTB devices for "
638 "the window at BAR2, on the upstream side of the link. MUST be the same "
639 "address on both sides.");
640 SYSCTL_UQUAD(_hw_ntb_xeon_b2b, OID_AUTO, usd_bar4_addr64, CTLFLAG_RDTUN,
641 &xeon_b2b_usd_addr.bar4_addr64, 0, "See usd_bar2_addr64, but BAR4.");
642 SYSCTL_UQUAD(_hw_ntb_xeon_b2b, OID_AUTO, usd_bar4_addr32, CTLFLAG_RDTUN,
643 &xeon_b2b_usd_addr.bar4_addr32, 0, "See usd_bar2_addr64, but BAR4 "
644 "(split-BAR mode).");
645 SYSCTL_UQUAD(_hw_ntb_xeon_b2b, OID_AUTO, usd_bar5_addr32, CTLFLAG_RDTUN,
646 &xeon_b2b_usd_addr.bar5_addr32, 0, "See usd_bar2_addr64, but BAR5 "
647 "(split-BAR mode).");
648
649 SYSCTL_UQUAD(_hw_ntb_xeon_b2b, OID_AUTO, dsd_bar2_addr64, CTLFLAG_RDTUN,
650 &xeon_b2b_dsd_addr.bar2_addr64, 0, "If using B2B topology on Xeon "
651 "hardware, use this 64-bit address on the bus between the NTB devices for "
652 "the window at BAR2, on the downstream side of the link. MUST be the same"
653 " address on both sides.");
654 SYSCTL_UQUAD(_hw_ntb_xeon_b2b, OID_AUTO, dsd_bar4_addr64, CTLFLAG_RDTUN,
655 &xeon_b2b_dsd_addr.bar4_addr64, 0, "See dsd_bar2_addr64, but BAR4.");
656 SYSCTL_UQUAD(_hw_ntb_xeon_b2b, OID_AUTO, dsd_bar4_addr32, CTLFLAG_RDTUN,
657 &xeon_b2b_dsd_addr.bar4_addr32, 0, "See dsd_bar2_addr64, but BAR4 "
658 "(split-BAR mode).");
659 SYSCTL_UQUAD(_hw_ntb_xeon_b2b, OID_AUTO, dsd_bar5_addr32, CTLFLAG_RDTUN,
660 &xeon_b2b_dsd_addr.bar5_addr32, 0, "See dsd_bar2_addr64, but BAR5 "
661 "(split-BAR mode).");
662
663 /*
664 * OS <-> Driver interface structures
665 */
666 MALLOC_DEFINE(M_NTB, "ntb_hw", "ntb_hw driver memory allocations");
667
668 /*
669 * OS <-> Driver linkage functions
670 */
671 static int
672 intel_ntb_probe(device_t device)
673 {
674 struct ntb_hw_info *p;
675
676 p = intel_ntb_get_device_info(pci_get_devid(device));
677 if (p == NULL)
678 return (ENXIO);
679
680 device_set_desc(device, p->desc);
681 return (0);
682 }
683
684 static int
685 intel_ntb_attach(device_t device)
686 {
687 struct ntb_softc *ntb;
688 struct ntb_hw_info *p;
689 int error;
690
691 ntb = device_get_softc(device);
692 p = intel_ntb_get_device_info(pci_get_devid(device));
693
694 ntb->device = device;
695 ntb->type = p->type;
696 ntb->features = p->features;
697 ntb->b2b_mw_idx = B2B_MW_DISABLED;
698 ntb->msix_mw_idx = B2B_MW_DISABLED;
699
700 /* Heartbeat timer for NTB_ATOM since there is no link interrupt */
701 callout_init(&ntb->heartbeat_timer, 1);
702 callout_init(&ntb->lr_timer, 1);
703 callout_init(&ntb->peer_msix_work, 1);
704 mtx_init(&ntb->db_mask_lock, "ntb hw bits", NULL, MTX_SPIN);
705
706 if (ntb->type == NTB_ATOM)
707 error = intel_ntb_detect_atom(ntb);
708 else if (ntb->type == NTB_XEON_GEN3)
709 error = intel_ntb_detect_xeon_gen3(ntb);
710 else
711 error = intel_ntb_detect_xeon(ntb);
712 if (error != 0)
713 goto out;
714
715 intel_ntb_detect_max_mw(ntb);
716
717 pci_enable_busmaster(ntb->device);
718
719 error = intel_ntb_map_pci_bars(ntb);
720 if (error != 0)
721 goto out;
722 if (ntb->type == NTB_ATOM)
723 error = intel_ntb_atom_init_dev(ntb);
724 else if (ntb->type == NTB_XEON_GEN3)
725 error = intel_ntb_xeon_gen3_init_dev(ntb);
726 else
727 error = intel_ntb_xeon_init_dev(ntb);
728 if (error != 0)
729 goto out;
730
731 intel_ntb_spad_clear(device);
732
733 intel_ntb_poll_link(ntb);
734
735 intel_ntb_sysctl_init(ntb);
736
737 /* Attach children to this controller */
738 error = ntb_register_device(device);
739
740 out:
741 if (error != 0)
742 intel_ntb_detach(device);
743 return (error);
744 }
745
746 static int
747 intel_ntb_detach(device_t device)
748 {
749 struct ntb_softc *ntb;
750
751 ntb = device_get_softc(device);
752
753 /* Detach & delete all children */
754 ntb_unregister_device(device);
755
756 if (ntb->self_reg != NULL) {
757 DB_MASK_LOCK(ntb);
758 db_iowrite(ntb, ntb->self_reg->db_mask, ntb->db_valid_mask);
759 DB_MASK_UNLOCK(ntb);
760 }
761 callout_drain(&ntb->heartbeat_timer);
762 callout_drain(&ntb->lr_timer);
763 callout_drain(&ntb->peer_msix_work);
764 pci_disable_busmaster(ntb->device);
765 if (ntb->type == NTB_XEON_GEN1)
766 intel_ntb_teardown_xeon(ntb);
767 intel_ntb_teardown_interrupts(ntb);
768
769 mtx_destroy(&ntb->db_mask_lock);
770
771 intel_ntb_unmap_pci_bar(ntb);
772
773 return (0);
774 }
775
776 /*
777 * Driver internal routines
778 */
779 static inline enum ntb_bar
780 intel_ntb_mw_to_bar(struct ntb_softc *ntb, unsigned mw)
781 {
782
783 KASSERT(mw < ntb->mw_count,
784 ("%s: mw:%u > count:%u", __func__, mw, (unsigned)ntb->mw_count));
785 KASSERT(ntb->reg->mw_bar[mw] != 0, ("invalid mw"));
786
787 return (ntb->reg->mw_bar[mw]);
788 }
789
790 static inline bool
791 bar_is_64bit(struct ntb_softc *ntb, enum ntb_bar bar)
792 {
793 /* XXX This assertion could be stronger. */
794 KASSERT(bar < NTB_MAX_BARS, ("bogus bar"));
795 return (bar < NTB_B2B_BAR_2 || !HAS_FEATURE(ntb, NTB_SPLIT_BAR));
796 }
797
798 static inline void
799 bar_get_xlat_params(struct ntb_softc *ntb, enum ntb_bar bar, uint32_t *base,
800 uint32_t *xlat, uint32_t *lmt)
801 {
802 uint32_t basev, lmtv, xlatv;
803
804 switch (bar) {
805 case NTB_B2B_BAR_1:
806 basev = ntb->xlat_reg->bar2_base;
807 lmtv = ntb->xlat_reg->bar2_limit;
808 xlatv = ntb->xlat_reg->bar2_xlat;
809 break;
810 case NTB_B2B_BAR_2:
811 basev = ntb->xlat_reg->bar4_base;
812 lmtv = ntb->xlat_reg->bar4_limit;
813 xlatv = ntb->xlat_reg->bar4_xlat;
814 break;
815 case NTB_B2B_BAR_3:
816 basev = ntb->xlat_reg->bar5_base;
817 lmtv = ntb->xlat_reg->bar5_limit;
818 xlatv = ntb->xlat_reg->bar5_xlat;
819 break;
820 default:
821 KASSERT(bar >= NTB_B2B_BAR_1 && bar < NTB_MAX_BARS,
822 ("bad bar"));
823 basev = lmtv = xlatv = 0;
824 break;
825 }
826
827 if (base != NULL)
828 *base = basev;
829 if (xlat != NULL)
830 *xlat = xlatv;
831 if (lmt != NULL)
832 *lmt = lmtv;
833 }
834
835 static int
836 intel_ntb_map_pci_bars(struct ntb_softc *ntb)
837 {
838 struct ntb_pci_bar_info *bar;
839 int rc;
840
841 bar = &ntb->bar_info[NTB_CONFIG_BAR];
842 bar->pci_resource_id = PCIR_BAR(0);
843 rc = map_mmr_bar(ntb, bar);
844 if (rc != 0)
845 goto out;
846
847 /*
848 * At least on Xeon v4 NTB device leaks to host some remote side
849 * BAR0 writes supposed to update scratchpad registers. I am not
850 * sure why it happens, but it may be related to the fact that
851 * on a link side BAR0 is 32KB, while on a host side it is 64KB.
852 * Without this hack DMAR blocks those accesses as not allowed.
853 */
854 if (bus_dma_tag_create(bus_get_dma_tag(ntb->device), 1, 0,
855 BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
856 bar->size, 1, bar->size, 0, NULL, NULL, &ntb->bar0_dma_tag)) {
857 device_printf(ntb->device, "Unable to create BAR0 tag\n");
858 return (ENOMEM);
859 }
860 if (bus_dmamap_create(ntb->bar0_dma_tag, 0, &ntb->bar0_dma_map)) {
861 device_printf(ntb->device, "Unable to create BAR0 map\n");
862 return (ENOMEM);
863 }
864 if (bus_dma_iommu_load_ident(ntb->bar0_dma_tag, ntb->bar0_dma_map,
865 bar->pbase, bar->size, 0)) {
866 device_printf(ntb->device, "Unable to load BAR0 map\n");
867 return (ENOMEM);
868 }
869
870 bar = &ntb->bar_info[NTB_B2B_BAR_1];
871 bar->pci_resource_id = PCIR_BAR(2);
872 rc = map_memory_window_bar(ntb, bar);
873 if (rc != 0)
874 goto out;
875 if (ntb->type == NTB_XEON_GEN3) {
876 bar->psz_off = XEON_GEN3_INT_REG_IMBAR1SZ;
877 bar->ssz_off = XEON_GEN3_INT_REG_EMBAR1SZ;
878 bar->pbarxlat_off = XEON_GEN3_REG_EMBAR1XBASE;
879 } else {
880 bar->psz_off = XEON_PBAR23SZ_OFFSET;
881 bar->ssz_off = XEON_SBAR23SZ_OFFSET;
882 bar->pbarxlat_off = XEON_PBAR2XLAT_OFFSET;
883 }
884
885 bar = &ntb->bar_info[NTB_B2B_BAR_2];
886 bar->pci_resource_id = PCIR_BAR(4);
887 rc = map_memory_window_bar(ntb, bar);
888 if (rc != 0)
889 goto out;
890 if (ntb->type == NTB_XEON_GEN3) {
891 bar->psz_off = XEON_GEN3_INT_REG_IMBAR2SZ;
892 bar->ssz_off = XEON_GEN3_INT_REG_EMBAR2SZ;
893 bar->pbarxlat_off = XEON_GEN3_REG_EMBAR2XBASE;
894 } else {
895 bar->psz_off = XEON_PBAR4SZ_OFFSET;
896 bar->ssz_off = XEON_SBAR4SZ_OFFSET;
897 bar->pbarxlat_off = XEON_PBAR4XLAT_OFFSET;
898 }
899
900 if (!HAS_FEATURE(ntb, NTB_SPLIT_BAR))
901 goto out;
902
903 if (ntb->type == NTB_XEON_GEN3) {
904 device_printf(ntb->device, "no split bar support\n");
905 return (ENXIO);
906 }
907
908 bar = &ntb->bar_info[NTB_B2B_BAR_3];
909 bar->pci_resource_id = PCIR_BAR(5);
910 rc = map_memory_window_bar(ntb, bar);
911 bar->psz_off = XEON_PBAR5SZ_OFFSET;
912 bar->ssz_off = XEON_SBAR5SZ_OFFSET;
913 bar->pbarxlat_off = XEON_PBAR5XLAT_OFFSET;
914
915 out:
916 if (rc != 0)
917 device_printf(ntb->device,
918 "unable to allocate pci resource\n");
919 return (rc);
920 }
921
922 static void
923 print_map_success(struct ntb_softc *ntb, struct ntb_pci_bar_info *bar,
924 const char *kind)
925 {
926
927 device_printf(ntb->device,
928 "Mapped BAR%d v:[%p-%p] p:[0x%jx-0x%jx] (0x%jx bytes) (%s)\n",
929 PCI_RID2BAR(bar->pci_resource_id), bar->vbase,
930 (char *)bar->vbase + bar->size - 1,
931 (uintmax_t)bar->pbase, (uintmax_t)(bar->pbase + bar->size - 1),
932 (uintmax_t)bar->size, kind);
933 }
934
935 static int
936 map_mmr_bar(struct ntb_softc *ntb, struct ntb_pci_bar_info *bar)
937 {
938
939 bar->pci_resource = bus_alloc_resource_any(ntb->device, SYS_RES_MEMORY,
940 &bar->pci_resource_id, RF_ACTIVE);
941 if (bar->pci_resource == NULL)
942 return (ENXIO);
943
944 save_bar_parameters(bar);
945 bar->map_mode = VM_MEMATTR_UNCACHEABLE;
946 print_map_success(ntb, bar, "mmr");
947 return (0);
948 }
949
950 static int
951 map_memory_window_bar(struct ntb_softc *ntb, struct ntb_pci_bar_info *bar)
952 {
953 int rc;
954 vm_memattr_t mapmode;
955 uint8_t bar_size_bits = 0;
956
957 bar->pci_resource = bus_alloc_resource_any(ntb->device, SYS_RES_MEMORY,
958 &bar->pci_resource_id, RF_ACTIVE);
959
960 if (bar->pci_resource == NULL)
961 return (ENXIO);
962
963 save_bar_parameters(bar);
964 /*
965 * Ivytown NTB BAR sizes are misreported by the hardware due to a
966 * hardware issue. To work around this, query the size it should be
967 * configured to by the device and modify the resource to correspond to
968 * this new size. The BIOS on systems with this problem is required to
969 * provide enough address space to allow the driver to make this change
970 * safely.
971 *
972 * Ideally I could have just specified the size when I allocated the
973 * resource like:
974 * bus_alloc_resource(ntb->device,
975 * SYS_RES_MEMORY, &bar->pci_resource_id, 0ul, ~0ul,
976 * 1ul << bar_size_bits, RF_ACTIVE);
977 * but the PCI driver does not honor the size in this call, so we have
978 * to modify it after the fact.
979 */
980 if (HAS_FEATURE(ntb, NTB_BAR_SIZE_4K)) {
981 if (bar->pci_resource_id == PCIR_BAR(2))
982 bar_size_bits = pci_read_config(ntb->device,
983 XEON_PBAR23SZ_OFFSET, 1);
984 else
985 bar_size_bits = pci_read_config(ntb->device,
986 XEON_PBAR45SZ_OFFSET, 1);
987
988 rc = bus_adjust_resource(ntb->device, SYS_RES_MEMORY,
989 bar->pci_resource, bar->pbase,
990 bar->pbase + (1ul << bar_size_bits) - 1);
991 if (rc != 0) {
992 device_printf(ntb->device,
993 "unable to resize bar\n");
994 return (rc);
995 }
996
997 save_bar_parameters(bar);
998 }
999
1000 bar->map_mode = VM_MEMATTR_UNCACHEABLE;
1001 print_map_success(ntb, bar, "mw");
1002
1003 /*
1004 * Optionally, mark MW BARs as anything other than UC to improve
1005 * performance.
1006 */
1007 mapmode = intel_ntb_pat_flags();
1008 if (mapmode == bar->map_mode)
1009 return (0);
1010
1011 rc = pmap_change_attr((vm_offset_t)bar->vbase, bar->size, mapmode);
1012 if (rc == 0) {
1013 bar->map_mode = mapmode;
1014 device_printf(ntb->device,
1015 "Marked BAR%d v:[%p-%p] p:[0x%jx-0x%jx] as "
1016 "%s.\n",
1017 PCI_RID2BAR(bar->pci_resource_id), bar->vbase,
1018 (char *)bar->vbase + bar->size - 1,
1019 (uintmax_t)bar->pbase,
1020 (uintmax_t)(bar->pbase + bar->size - 1),
1021 intel_ntb_vm_memattr_to_str(mapmode));
1022 } else
1023 device_printf(ntb->device,
1024 "Unable to mark BAR%d v:[%p-%p] p:[0x%jx-0x%jx] as "
1025 "%s: %d\n",
1026 PCI_RID2BAR(bar->pci_resource_id), bar->vbase,
1027 (char *)bar->vbase + bar->size - 1,
1028 (uintmax_t)bar->pbase,
1029 (uintmax_t)(bar->pbase + bar->size - 1),
1030 intel_ntb_vm_memattr_to_str(mapmode), rc);
1031 /* Proceed anyway */
1032 return (0);
1033 }
1034
1035 static void
1036 intel_ntb_unmap_pci_bar(struct ntb_softc *ntb)
1037 {
1038 struct ntb_pci_bar_info *bar;
1039 int i;
1040
1041 if (ntb->bar0_dma_map != NULL) {
1042 bus_dmamap_unload(ntb->bar0_dma_tag, ntb->bar0_dma_map);
1043 bus_dmamap_destroy(ntb->bar0_dma_tag, ntb->bar0_dma_map);
1044 }
1045 if (ntb->bar0_dma_tag != NULL)
1046 bus_dma_tag_destroy(ntb->bar0_dma_tag);
1047 for (i = 0; i < NTB_MAX_BARS; i++) {
1048 bar = &ntb->bar_info[i];
1049 if (bar->pci_resource != NULL)
1050 bus_release_resource(ntb->device, SYS_RES_MEMORY,
1051 bar->pci_resource_id, bar->pci_resource);
1052 }
1053 }
1054
1055 static int
1056 intel_ntb_setup_msix(struct ntb_softc *ntb, uint32_t num_vectors)
1057 {
1058 uint32_t i;
1059 int rc;
1060
1061 for (i = 0; i < num_vectors; i++) {
1062 ntb->int_info[i].rid = i + 1;
1063 ntb->int_info[i].res = bus_alloc_resource_any(ntb->device,
1064 SYS_RES_IRQ, &ntb->int_info[i].rid, RF_ACTIVE);
1065 if (ntb->int_info[i].res == NULL) {
1066 device_printf(ntb->device,
1067 "bus_alloc_resource failed\n");
1068 return (ENOMEM);
1069 }
1070 ntb->int_info[i].tag = NULL;
1071 ntb->allocated_interrupts++;
1072 rc = bus_setup_intr(ntb->device, ntb->int_info[i].res,
1073 INTR_MPSAFE | INTR_TYPE_MISC, NULL, ndev_vec_isr,
1074 &ntb->msix_vec[i], &ntb->int_info[i].tag);
1075 if (rc != 0) {
1076 device_printf(ntb->device, "bus_setup_intr failed\n");
1077 return (ENXIO);
1078 }
1079 }
1080 return (0);
1081 }
1082
1083 /*
1084 * The Linux NTB driver drops from MSI-X to legacy INTx if a unique vector
1085 * cannot be allocated for each MSI-X message. JHB seems to think remapping
1086 * should be okay. This tunable should enable us to test that hypothesis
1087 * when someone gets their hands on some Xeon hardware.
1088 */
1089 static int ntb_force_remap_mode;
1090 SYSCTL_INT(_hw_ntb, OID_AUTO, force_remap_mode, CTLFLAG_RDTUN,
1091 &ntb_force_remap_mode, 0, "If enabled, force MSI-X messages to be remapped"
1092 " to a smaller number of ithreads, even if the desired number are "
1093 "available");
1094
1095 /*
1096 * In case it is NOT ok, give consumers an abort button.
1097 */
1098 static int ntb_prefer_intx;
1099 SYSCTL_INT(_hw_ntb, OID_AUTO, prefer_intx_to_remap, CTLFLAG_RDTUN,
1100 &ntb_prefer_intx, 0, "If enabled, prefer to use legacy INTx mode rather "
1101 "than remapping MSI-X messages over available slots (match Linux driver "
1102 "behavior)");
1103
1104 /*
1105 * Remap the desired number of MSI-X messages to available ithreads in a simple
1106 * round-robin fashion.
1107 */
1108 static int
1109 intel_ntb_remap_msix(device_t dev, uint32_t desired, uint32_t avail)
1110 {
1111 u_int *vectors;
1112 uint32_t i;
1113 int rc;
1114
1115 if (ntb_prefer_intx != 0)
1116 return (ENXIO);
1117
1118 vectors = malloc(desired * sizeof(*vectors), M_NTB, M_ZERO | M_WAITOK);
1119
1120 for (i = 0; i < desired; i++)
1121 vectors[i] = (i % avail) + 1;
1122
1123 rc = pci_remap_msix(dev, desired, vectors);
1124 free(vectors, M_NTB);
1125 return (rc);
1126 }
1127
1128 static int
1129 intel_ntb_xeon_gen3_init_isr(struct ntb_softc *ntb)
1130 {
1131 uint64_t i, reg;
1132 uint32_t desired_vectors, num_vectors;
1133 int rc;
1134
1135 ntb->allocated_interrupts = 0;
1136 ntb->last_ts = ticks;
1137
1138 /* Mask all the interrupts, including hardware interrupt */
1139 intel_ntb_reg_write(8, XEON_GEN3_REG_IMINT_DISABLE, ~0ULL);
1140
1141 /* Clear Interrupt Status */
1142 reg = intel_ntb_reg_read(8, XEON_GEN3_REG_IMINT_STATUS);
1143 intel_ntb_reg_write(8, XEON_GEN3_REG_IMINT_STATUS, reg);
1144
1145 num_vectors = desired_vectors = MIN(pci_msix_count(ntb->device),
1146 XEON_GEN3_DB_MSIX_VECTOR_COUNT);
1147
1148 rc = pci_alloc_msix(ntb->device, &num_vectors);
1149 if (rc != 0) {
1150 device_printf(ntb->device,
1151 "Interrupt allocation failed %d\n", rc);
1152 return (rc);
1153 }
1154 if (desired_vectors != num_vectors) {
1155 device_printf(ntb->device, "Couldn't get %d vectors\n",
1156 XEON_GEN3_DB_MSIX_VECTOR_COUNT);
1157 return (ENXIO);
1158 }
1159 /* 32 db + 1 hardware */
1160 if (num_vectors == XEON_GEN3_DB_MSIX_VECTOR_COUNT) {
1161 /* Program INTVECXX source register */
1162 for (i = 0; i < XEON_GEN3_DB_MSIX_VECTOR_COUNT; i++) {
1163 /* interrupt source i for vector i */
1164 intel_ntb_reg_write(1, XEON_GEN3_REG_IMINTVEC00 + i, i);
1165 if (i == (XEON_GEN3_DB_MSIX_VECTOR_COUNT - 1)) {
1166 intel_ntb_reg_write(1,
1167 XEON_GEN3_REG_IMINTVEC00 + i,
1168 XEON_GEN3_LINK_VECTOR_INDEX);
1169 }
1170 }
1171
1172 intel_ntb_create_msix_vec(ntb, num_vectors);
1173 rc = intel_ntb_setup_msix(ntb, num_vectors);
1174
1175 /* enable all interrupts */
1176 intel_ntb_reg_write(8, XEON_GEN3_REG_IMINT_DISABLE, 0ULL);
1177 } else {
1178 device_printf(ntb->device, "need to remap interrupts, giving up.\n");
1179 return (ENXIO);
1180 }
1181
1182 return (0);
1183 }
1184
1185 static int
1186 intel_ntb_init_isr(struct ntb_softc *ntb)
1187 {
1188 uint32_t desired_vectors, num_vectors;
1189 int rc;
1190
1191 ntb->allocated_interrupts = 0;
1192 ntb->last_ts = ticks;
1193
1194 /*
1195 * Mask all doorbell interrupts. (Except link events!)
1196 */
1197 DB_MASK_LOCK(ntb);
1198 ntb->db_mask = ntb->db_valid_mask;
1199 db_iowrite(ntb, ntb->self_reg->db_mask, ntb->db_mask);
1200 DB_MASK_UNLOCK(ntb);
1201
1202 num_vectors = desired_vectors = MIN(pci_msix_count(ntb->device),
1203 ntb->db_count);
1204 if (desired_vectors >= 1) {
1205 rc = pci_alloc_msix(ntb->device, &num_vectors);
1206
1207 if (ntb_force_remap_mode != 0 && rc == 0 &&
1208 num_vectors == desired_vectors)
1209 num_vectors--;
1210
1211 if (rc == 0 && num_vectors < desired_vectors) {
1212 rc = intel_ntb_remap_msix(ntb->device, desired_vectors,
1213 num_vectors);
1214 if (rc == 0)
1215 num_vectors = desired_vectors;
1216 else
1217 pci_release_msi(ntb->device);
1218 }
1219 if (rc != 0)
1220 num_vectors = 1;
1221 } else
1222 num_vectors = 1;
1223
1224 if (ntb->type == NTB_XEON_GEN1 && num_vectors < ntb->db_vec_count) {
1225 if (HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP)) {
1226 device_printf(ntb->device,
1227 "Errata workaround does not support MSI or INTX\n");
1228 return (EINVAL);
1229 }
1230
1231 ntb->db_vec_count = 1;
1232 ntb->db_vec_shift = XEON_DB_TOTAL_SHIFT;
1233 rc = intel_ntb_setup_legacy_interrupt(ntb);
1234 } else {
1235 if (num_vectors - 1 != XEON_NONLINK_DB_MSIX_BITS &&
1236 HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP)) {
1237 device_printf(ntb->device,
1238 "Errata workaround expects %d doorbell bits\n",
1239 XEON_NONLINK_DB_MSIX_BITS);
1240 return (EINVAL);
1241 }
1242
1243 intel_ntb_create_msix_vec(ntb, num_vectors);
1244 rc = intel_ntb_setup_msix(ntb, num_vectors);
1245 }
1246 if (rc != 0) {
1247 device_printf(ntb->device,
1248 "Error allocating interrupts: %d\n", rc);
1249 intel_ntb_free_msix_vec(ntb);
1250 }
1251
1252 return (rc);
1253 }
1254
1255 static int
1256 intel_ntb_setup_legacy_interrupt(struct ntb_softc *ntb)
1257 {
1258 int rc;
1259
1260 ntb->int_info[0].rid = 0;
1261 ntb->int_info[0].res = bus_alloc_resource_any(ntb->device, SYS_RES_IRQ,
1262 &ntb->int_info[0].rid, RF_SHAREABLE|RF_ACTIVE);
1263 if (ntb->int_info[0].res == NULL) {
1264 device_printf(ntb->device, "bus_alloc_resource failed\n");
1265 return (ENOMEM);
1266 }
1267
1268 ntb->int_info[0].tag = NULL;
1269 ntb->allocated_interrupts = 1;
1270
1271 rc = bus_setup_intr(ntb->device, ntb->int_info[0].res,
1272 INTR_MPSAFE | INTR_TYPE_MISC, NULL, ndev_irq_isr,
1273 ntb, &ntb->int_info[0].tag);
1274 if (rc != 0) {
1275 device_printf(ntb->device, "bus_setup_intr failed\n");
1276 return (ENXIO);
1277 }
1278
1279 return (0);
1280 }
1281
1282 static void
1283 intel_ntb_teardown_interrupts(struct ntb_softc *ntb)
1284 {
1285 struct ntb_int_info *current_int;
1286 int i;
1287
1288 for (i = 0; i < ntb->allocated_interrupts; i++) {
1289 current_int = &ntb->int_info[i];
1290 if (current_int->tag != NULL)
1291 bus_teardown_intr(ntb->device, current_int->res,
1292 current_int->tag);
1293
1294 if (current_int->res != NULL)
1295 bus_release_resource(ntb->device, SYS_RES_IRQ,
1296 rman_get_rid(current_int->res), current_int->res);
1297 }
1298
1299 intel_ntb_free_msix_vec(ntb);
1300 pci_release_msi(ntb->device);
1301 }
1302
1303 static inline uint64_t
1304 db_ioread(struct ntb_softc *ntb, uint64_t regoff)
1305 {
1306
1307 switch (ntb->type) {
1308 case NTB_ATOM:
1309 case NTB_XEON_GEN3:
1310 return (intel_ntb_reg_read(8, regoff));
1311 case NTB_XEON_GEN1:
1312 return (intel_ntb_reg_read(2, regoff));
1313 }
1314 __assert_unreachable();
1315 }
1316
1317 static inline void
1318 db_iowrite(struct ntb_softc *ntb, uint64_t regoff, uint64_t val)
1319 {
1320
1321 KASSERT((val & ~ntb->db_valid_mask) == 0,
1322 ("%s: Invalid bits 0x%jx (valid: 0x%jx)", __func__,
1323 (uintmax_t)(val & ~ntb->db_valid_mask),
1324 (uintmax_t)ntb->db_valid_mask));
1325
1326 if (regoff == ntb->self_reg->db_mask)
1327 DB_MASK_ASSERT(ntb, MA_OWNED);
1328 db_iowrite_raw(ntb, regoff, val);
1329 }
1330
1331 static inline void
1332 db_iowrite_raw(struct ntb_softc *ntb, uint64_t regoff, uint64_t val)
1333 {
1334
1335 switch (ntb->type) {
1336 case NTB_ATOM:
1337 case NTB_XEON_GEN3:
1338 intel_ntb_reg_write(8, regoff, val);
1339 break;
1340 case NTB_XEON_GEN1:
1341 intel_ntb_reg_write(2, regoff, (uint16_t)val);
1342 break;
1343 }
1344 }
1345
1346 static void
1347 intel_ntb_db_set_mask(device_t dev, uint64_t bits)
1348 {
1349 struct ntb_softc *ntb = device_get_softc(dev);
1350
1351 DB_MASK_LOCK(ntb);
1352 ntb->db_mask |= bits;
1353 if (!HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP))
1354 db_iowrite(ntb, ntb->self_reg->db_mask, ntb->db_mask);
1355 DB_MASK_UNLOCK(ntb);
1356 }
1357
1358 static void
1359 intel_ntb_db_clear_mask(device_t dev, uint64_t bits)
1360 {
1361 struct ntb_softc *ntb = device_get_softc(dev);
1362 uint64_t ibits;
1363 int i;
1364
1365 KASSERT((bits & ~ntb->db_valid_mask) == 0,
1366 ("%s: Invalid bits 0x%jx (valid: 0x%jx)", __func__,
1367 (uintmax_t)(bits & ~ntb->db_valid_mask),
1368 (uintmax_t)ntb->db_valid_mask));
1369
1370 DB_MASK_LOCK(ntb);
1371 ibits = ntb->fake_db & ntb->db_mask & bits;
1372 ntb->db_mask &= ~bits;
1373 if (HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP)) {
1374 /* Simulate fake interrupts if unmasked DB bits are set. */
1375 ntb->force_db |= ibits;
1376 for (i = 0; i < XEON_NONLINK_DB_MSIX_BITS; i++) {
1377 if ((ibits & intel_ntb_db_vector_mask(dev, i)) != 0)
1378 swi_sched(ntb->int_info[i].tag, 0);
1379 }
1380 } else {
1381 db_iowrite(ntb, ntb->self_reg->db_mask, ntb->db_mask);
1382 }
1383 DB_MASK_UNLOCK(ntb);
1384 }
1385
1386 static uint64_t
1387 intel_ntb_db_read(device_t dev)
1388 {
1389 struct ntb_softc *ntb = device_get_softc(dev);
1390
1391 if (HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP))
1392 return (ntb->fake_db);
1393 if (ntb->type == NTB_XEON_GEN3)
1394 return (intel_ntb_reg_read(8, XEON_GEN3_REG_IMINT_STATUS));
1395 else
1396 return (db_ioread(ntb, ntb->self_reg->db_bell));
1397 }
1398
1399 static void
1400 intel_ntb_db_clear(device_t dev, uint64_t bits)
1401 {
1402 struct ntb_softc *ntb = device_get_softc(dev);
1403
1404 KASSERT((bits & ~ntb->db_valid_mask) == 0,
1405 ("%s: Invalid bits 0x%jx (valid: 0x%jx)", __func__,
1406 (uintmax_t)(bits & ~ntb->db_valid_mask),
1407 (uintmax_t)ntb->db_valid_mask));
1408
1409 if (HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP)) {
1410 DB_MASK_LOCK(ntb);
1411 ntb->fake_db &= ~bits;
1412 DB_MASK_UNLOCK(ntb);
1413 return;
1414 }
1415
1416 if (ntb->type == NTB_XEON_GEN3)
1417 intel_ntb_reg_write(4, XEON_GEN3_REG_IMINT_STATUS,
1418 (uint32_t)bits);
1419 else
1420 db_iowrite(ntb, ntb->self_reg->db_bell, bits);
1421 }
1422
1423 static inline uint64_t
1424 intel_ntb_vec_mask(struct ntb_softc *ntb, uint64_t db_vector)
1425 {
1426 uint64_t shift, mask;
1427
1428 if (HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP)) {
1429 /*
1430 * Remap vectors in custom way to make at least first
1431 * three doorbells to not generate stray events.
1432 * This breaks Linux compatibility (if one existed)
1433 * when more then one DB is used (not by if_ntb).
1434 */
1435 if (db_vector < XEON_NONLINK_DB_MSIX_BITS - 1)
1436 return (1 << db_vector);
1437 if (db_vector == XEON_NONLINK_DB_MSIX_BITS - 1)
1438 return (0x7ffc);
1439 }
1440
1441 shift = ntb->db_vec_shift;
1442 mask = (1ull << shift) - 1;
1443 return (mask << (shift * db_vector));
1444 }
1445
1446 static void
1447 intel_ntb_interrupt(struct ntb_softc *ntb, uint32_t vec)
1448 {
1449 uint64_t vec_mask;
1450
1451 ntb->last_ts = ticks;
1452 vec_mask = intel_ntb_vec_mask(ntb, vec);
1453
1454 if (ntb->type == NTB_XEON_GEN3 && vec == XEON_GEN3_LINK_VECTOR_INDEX)
1455 vec_mask |= ntb->db_link_mask;
1456 if ((vec_mask & ntb->db_link_mask) != 0) {
1457 if (intel_ntb_poll_link(ntb))
1458 ntb_link_event(ntb->device);
1459 if (ntb->type == NTB_XEON_GEN3)
1460 intel_ntb_reg_write(8, XEON_GEN3_REG_IMINT_STATUS,
1461 intel_ntb_reg_read(8, XEON_GEN3_REG_IMINT_STATUS));
1462 }
1463
1464 if (HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP) &&
1465 (vec_mask & ntb->db_link_mask) == 0) {
1466 DB_MASK_LOCK(ntb);
1467
1468 /*
1469 * Do not report same DB events again if not cleared yet,
1470 * unless the mask was just cleared for them and this
1471 * interrupt handler call can be the consequence of it.
1472 */
1473 vec_mask &= ~ntb->fake_db | ntb->force_db;
1474 ntb->force_db &= ~vec_mask;
1475
1476 /* Update our internal doorbell register. */
1477 ntb->fake_db |= vec_mask;
1478
1479 /* Do not report masked DB events. */
1480 vec_mask &= ~ntb->db_mask;
1481
1482 DB_MASK_UNLOCK(ntb);
1483 }
1484
1485 if ((vec_mask & ntb->db_valid_mask) != 0)
1486 ntb_db_event(ntb->device, vec);
1487 }
1488
1489 static void
1490 ndev_vec_isr(void *arg)
1491 {
1492 struct ntb_vec *nvec = arg;
1493
1494 intel_ntb_interrupt(nvec->ntb, nvec->num);
1495 }
1496
1497 static void
1498 ndev_irq_isr(void *arg)
1499 {
1500 /* If we couldn't set up MSI-X, we only have the one vector. */
1501 intel_ntb_interrupt(arg, 0);
1502 }
1503
1504 static int
1505 intel_ntb_create_msix_vec(struct ntb_softc *ntb, uint32_t num_vectors)
1506 {
1507 uint32_t i;
1508
1509 ntb->msix_vec = malloc(num_vectors * sizeof(*ntb->msix_vec), M_NTB,
1510 M_ZERO | M_WAITOK);
1511 for (i = 0; i < num_vectors; i++) {
1512 ntb->msix_vec[i].num = i;
1513 ntb->msix_vec[i].ntb = ntb;
1514 }
1515
1516 return (0);
1517 }
1518
1519 static void
1520 intel_ntb_free_msix_vec(struct ntb_softc *ntb)
1521 {
1522
1523 if (ntb->msix_vec == NULL)
1524 return;
1525
1526 free(ntb->msix_vec, M_NTB);
1527 ntb->msix_vec = NULL;
1528 }
1529
1530 static void
1531 intel_ntb_get_msix_info(struct ntb_softc *ntb)
1532 {
1533 struct pci_devinfo *dinfo;
1534 struct pcicfg_msix *msix;
1535 uint32_t laddr, data, i, offset;
1536
1537 dinfo = device_get_ivars(ntb->device);
1538 msix = &dinfo->cfg.msix;
1539
1540 CTASSERT(XEON_NONLINK_DB_MSIX_BITS == nitems(ntb->msix_data));
1541
1542 for (i = 0; i < XEON_NONLINK_DB_MSIX_BITS; i++) {
1543 offset = msix->msix_table_offset + i * PCI_MSIX_ENTRY_SIZE;
1544
1545 laddr = bus_read_4(msix->msix_table_res, offset +
1546 PCI_MSIX_ENTRY_LOWER_ADDR);
1547 intel_ntb_printf(2, "local MSIX addr(%u): 0x%x\n", i, laddr);
1548
1549 KASSERT((laddr & MSI_INTEL_ADDR_BASE) == MSI_INTEL_ADDR_BASE,
1550 ("local MSIX addr 0x%x not in MSI base 0x%x", laddr,
1551 MSI_INTEL_ADDR_BASE));
1552 ntb->msix_data[i].nmd_ofs = laddr;
1553
1554 data = bus_read_4(msix->msix_table_res, offset +
1555 PCI_MSIX_ENTRY_DATA);
1556 intel_ntb_printf(2, "local MSIX data(%u): 0x%x\n", i, data);
1557
1558 ntb->msix_data[i].nmd_data = data;
1559 }
1560 }
1561
1562 static struct ntb_hw_info *
1563 intel_ntb_get_device_info(uint32_t device_id)
1564 {
1565 struct ntb_hw_info *ep;
1566
1567 for (ep = pci_ids; ep < &pci_ids[nitems(pci_ids)]; ep++) {
1568 if (ep->device_id == device_id)
1569 return (ep);
1570 }
1571 return (NULL);
1572 }
1573
1574 static void
1575 intel_ntb_teardown_xeon(struct ntb_softc *ntb)
1576 {
1577
1578 if (ntb->reg != NULL)
1579 intel_ntb_link_disable(ntb->device);
1580 }
1581
1582 static void
1583 intel_ntb_detect_max_mw(struct ntb_softc *ntb)
1584 {
1585
1586 switch (ntb->type) {
1587 case NTB_ATOM:
1588 ntb->mw_count = ATOM_MW_COUNT;
1589 break;
1590 case NTB_XEON_GEN1:
1591 if (HAS_FEATURE(ntb, NTB_SPLIT_BAR))
1592 ntb->mw_count = XEON_HSX_SPLIT_MW_COUNT;
1593 else
1594 ntb->mw_count = XEON_SNB_MW_COUNT;
1595 break;
1596 case NTB_XEON_GEN3:
1597 if (HAS_FEATURE(ntb, NTB_SPLIT_BAR))
1598 ntb->mw_count = XEON_GEN3_SPLIT_MW_COUNT;
1599 else
1600 ntb->mw_count = XEON_GEN3_MW_COUNT;
1601 break;
1602 }
1603 }
1604
1605 static int
1606 intel_ntb_detect_xeon(struct ntb_softc *ntb)
1607 {
1608 uint8_t ppd, conn_type;
1609
1610 ppd = pci_read_config(ntb->device, NTB_PPD_OFFSET, 1);
1611 ntb->ppd = ppd;
1612
1613 if ((ppd & XEON_PPD_DEV_TYPE) != 0)
1614 ntb->dev_type = NTB_DEV_DSD;
1615 else
1616 ntb->dev_type = NTB_DEV_USD;
1617
1618 if ((ppd & XEON_PPD_SPLIT_BAR) != 0)
1619 ntb->features |= NTB_SPLIT_BAR;
1620
1621 if (HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP) &&
1622 !HAS_FEATURE(ntb, NTB_SPLIT_BAR)) {
1623 device_printf(ntb->device,
1624 "Can not apply SB01BASE_LOCKUP workaround "
1625 "with split BARs disabled!\n");
1626 device_printf(ntb->device,
1627 "Expect system hangs under heavy NTB traffic!\n");
1628 ntb->features &= ~NTB_SB01BASE_LOCKUP;
1629 }
1630
1631 /*
1632 * SDOORBELL errata workaround gets in the way of SB01BASE_LOCKUP
1633 * errata workaround; only do one at a time.
1634 */
1635 if (HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP))
1636 ntb->features &= ~NTB_SDOORBELL_LOCKUP;
1637
1638 conn_type = ppd & XEON_PPD_CONN_TYPE;
1639 switch (conn_type) {
1640 case NTB_CONN_B2B:
1641 ntb->conn_type = conn_type;
1642 break;
1643 case NTB_CONN_RP:
1644 case NTB_CONN_TRANSPARENT:
1645 default:
1646 device_printf(ntb->device, "Unsupported connection type: %u\n",
1647 (unsigned)conn_type);
1648 return (ENXIO);
1649 }
1650 return (0);
1651 }
1652
1653 static int
1654 intel_ntb_detect_atom(struct ntb_softc *ntb)
1655 {
1656 uint32_t ppd, conn_type;
1657
1658 ppd = pci_read_config(ntb->device, NTB_PPD_OFFSET, 4);
1659 ntb->ppd = ppd;
1660
1661 if ((ppd & ATOM_PPD_DEV_TYPE) != 0)
1662 ntb->dev_type = NTB_DEV_DSD;
1663 else
1664 ntb->dev_type = NTB_DEV_USD;
1665
1666 conn_type = (ppd & ATOM_PPD_CONN_TYPE) >> 8;
1667 switch (conn_type) {
1668 case NTB_CONN_B2B:
1669 ntb->conn_type = conn_type;
1670 break;
1671 default:
1672 device_printf(ntb->device, "Unsupported NTB configuration\n");
1673 return (ENXIO);
1674 }
1675 return (0);
1676 }
1677
1678 static int
1679 intel_ntb_detect_xeon_gen3(struct ntb_softc *ntb)
1680 {
1681 uint8_t ppd, conn_type;
1682
1683 ppd = pci_read_config(ntb->device, XEON_GEN3_INT_REG_PPD, 1);
1684 ntb->ppd = ppd;
1685
1686 /* check port definition */
1687 conn_type = XEON_GEN3_REG_PPD_PORT_DEF_F(ppd);
1688 switch (conn_type) {
1689 case NTB_CONN_B2B:
1690 ntb->conn_type = conn_type;
1691 break;
1692 default:
1693 device_printf(ntb->device, "Unsupported connection type: %u\n",
1694 conn_type);
1695 return (ENXIO);
1696 }
1697
1698 /* check cross link configuration status */
1699 if (XEON_GEN3_REG_PPD_CONF_STS_F(ppd)) {
1700 /* NTB Port is configured as DSD/USP */
1701 ntb->dev_type = NTB_DEV_DSD;
1702 } else {
1703 /* NTB Port is configured as USD/DSP */
1704 ntb->dev_type = NTB_DEV_USD;
1705 }
1706
1707 if (XEON_GEN3_REG_PPD_ONE_MSIX_F(ppd)) {
1708 /*
1709 * This bit when set, causes only a single MSI-X message to be
1710 * generated if MSI-X is enabled.
1711 */
1712 ntb->features |= NTB_ONE_MSIX;
1713 }
1714
1715 if (XEON_GEN3_REG_PPD_BAR45_SPL_F(ppd)) {
1716 /* BARs 4 and 5 are presented as two 32b non-prefetchable BARs */
1717 ntb->features |= NTB_SPLIT_BAR;
1718 }
1719
1720 device_printf(ntb->device, "conn type 0x%02x, dev type 0x%02x,"
1721 "features 0x%02x\n", ntb->conn_type, ntb->dev_type, ntb->features);
1722
1723 return (0);
1724 }
1725
1726 static int
1727 intel_ntb_xeon_init_dev(struct ntb_softc *ntb)
1728 {
1729 int rc;
1730
1731 ntb->spad_count = XEON_SPAD_COUNT;
1732 ntb->db_count = XEON_DB_COUNT;
1733 ntb->db_link_mask = XEON_DB_LINK_BIT;
1734 ntb->db_vec_count = XEON_DB_MSIX_VECTOR_COUNT;
1735 ntb->db_vec_shift = XEON_DB_MSIX_VECTOR_SHIFT;
1736
1737 if (ntb->conn_type != NTB_CONN_B2B) {
1738 device_printf(ntb->device, "Connection type %d not supported\n",
1739 ntb->conn_type);
1740 return (ENXIO);
1741 }
1742
1743 ntb->reg = &xeon_reg;
1744 ntb->self_reg = &xeon_pri_reg;
1745 ntb->peer_reg = &xeon_b2b_reg;
1746 ntb->xlat_reg = &xeon_sec_xlat;
1747
1748 if (HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP)) {
1749 ntb->force_db = ntb->fake_db = 0;
1750 ntb->msix_mw_idx = (ntb->mw_count + g_ntb_msix_idx) %
1751 ntb->mw_count;
1752 intel_ntb_printf(2, "Setting up MSIX mw idx %d means %u\n",
1753 g_ntb_msix_idx, ntb->msix_mw_idx);
1754 rc = intel_ntb_mw_set_wc_internal(ntb, ntb->msix_mw_idx,
1755 VM_MEMATTR_UNCACHEABLE);
1756 KASSERT(rc == 0, ("shouldn't fail"));
1757 } else if (HAS_FEATURE(ntb, NTB_SDOORBELL_LOCKUP)) {
1758 /*
1759 * There is a Xeon hardware errata related to writes to SDOORBELL or
1760 * B2BDOORBELL in conjunction with inbound access to NTB MMIO space,
1761 * which may hang the system. To workaround this, use a memory
1762 * window to access the interrupt and scratch pad registers on the
1763 * remote system.
1764 */
1765 ntb->b2b_mw_idx = (ntb->mw_count + g_ntb_mw_idx) %
1766 ntb->mw_count;
1767 intel_ntb_printf(2, "Setting up b2b mw idx %d means %u\n",
1768 g_ntb_mw_idx, ntb->b2b_mw_idx);
1769 rc = intel_ntb_mw_set_wc_internal(ntb, ntb->b2b_mw_idx,
1770 VM_MEMATTR_UNCACHEABLE);
1771 KASSERT(rc == 0, ("shouldn't fail"));
1772 } else if (HAS_FEATURE(ntb, NTB_B2BDOORBELL_BIT14))
1773 /*
1774 * HW Errata on bit 14 of b2bdoorbell register. Writes will not be
1775 * mirrored to the remote system. Shrink the number of bits by one,
1776 * since bit 14 is the last bit.
1777 *
1778 * On REGS_THRU_MW errata mode, we don't use the b2bdoorbell register
1779 * anyway. Nor for non-B2B connection types.
1780 */
1781 ntb->db_count = XEON_DB_COUNT - 1;
1782
1783 ntb->db_valid_mask = (1ull << ntb->db_count) - 1;
1784
1785 if (ntb->dev_type == NTB_DEV_USD)
1786 rc = xeon_setup_b2b_mw(ntb, &xeon_b2b_dsd_addr,
1787 &xeon_b2b_usd_addr);
1788 else
1789 rc = xeon_setup_b2b_mw(ntb, &xeon_b2b_usd_addr,
1790 &xeon_b2b_dsd_addr);
1791 if (rc != 0)
1792 return (rc);
1793
1794 /* Enable Bus Master and Memory Space on the secondary side */
1795 intel_ntb_reg_write(2, XEON_SPCICMD_OFFSET,
1796 PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN);
1797
1798 /*
1799 * Mask all doorbell interrupts.
1800 */
1801 DB_MASK_LOCK(ntb);
1802 ntb->db_mask = ntb->db_valid_mask;
1803 db_iowrite(ntb, ntb->self_reg->db_mask, ntb->db_mask);
1804 DB_MASK_UNLOCK(ntb);
1805
1806 rc = intel_ntb_init_isr(ntb);
1807 return (rc);
1808 }
1809
1810 static int
1811 intel_ntb_xeon_gen3_init_dev(struct ntb_softc *ntb)
1812 {
1813 int rc;
1814
1815 ntb->spad_count = XEON_GEN3_SPAD_COUNT;
1816 ntb->db_count = XEON_GEN3_DB_COUNT;
1817 ntb->db_link_mask = XEON_GEN3_DB_LINK_BIT;
1818 ntb->db_vec_count = XEON_GEN3_DB_MSIX_VECTOR_COUNT;
1819 ntb->db_vec_shift = XEON_GEN3_DB_MSIX_VECTOR_SHIFT;
1820
1821 if (ntb->conn_type != NTB_CONN_B2B) {
1822 device_printf(ntb->device, "Connection type %d not supported\n",
1823 ntb->conn_type);
1824 return (ENXIO);
1825 }
1826
1827 ntb->reg = &xeon_gen3_reg;
1828 ntb->self_reg = &xeon_gen3_pri_reg;
1829 ntb->peer_reg = &xeon_gen3_b2b_reg;
1830 ntb->xlat_reg = &xeon_gen3_sec_xlat;
1831
1832 ntb->db_valid_mask = (1ULL << ntb->db_count) - 1;
1833
1834 xeon_gen3_setup_b2b_mw(ntb);
1835
1836 /* Enable Bus Master and Memory Space on the External Side */
1837 intel_ntb_reg_write(2, XEON_GEN3_EXT_REG_PCI_CMD,
1838 PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN);
1839
1840 /* Setup Interrupt */
1841 rc = intel_ntb_xeon_gen3_init_isr(ntb);
1842
1843 return (rc);
1844 }
1845
1846 static int
1847 intel_ntb_atom_init_dev(struct ntb_softc *ntb)
1848 {
1849 int error;
1850
1851 KASSERT(ntb->conn_type == NTB_CONN_B2B,
1852 ("Unsupported NTB configuration (%d)\n", ntb->conn_type));
1853
1854 ntb->spad_count = ATOM_SPAD_COUNT;
1855 ntb->db_count = ATOM_DB_COUNT;
1856 ntb->db_vec_count = ATOM_DB_MSIX_VECTOR_COUNT;
1857 ntb->db_vec_shift = ATOM_DB_MSIX_VECTOR_SHIFT;
1858 ntb->db_valid_mask = (1ull << ntb->db_count) - 1;
1859
1860 ntb->reg = &atom_reg;
1861 ntb->self_reg = &atom_pri_reg;
1862 ntb->peer_reg = &atom_b2b_reg;
1863 ntb->xlat_reg = &atom_sec_xlat;
1864
1865 /*
1866 * FIXME - MSI-X bug on early Atom HW, remove once internal issue is
1867 * resolved. Mask transaction layer internal parity errors.
1868 */
1869 pci_write_config(ntb->device, 0xFC, 0x4, 4);
1870
1871 configure_atom_secondary_side_bars(ntb);
1872
1873 /* Enable Bus Master and Memory Space on the secondary side */
1874 intel_ntb_reg_write(2, ATOM_SPCICMD_OFFSET,
1875 PCIM_CMD_MEMEN | PCIM_CMD_BUSMASTEREN);
1876
1877 error = intel_ntb_init_isr(ntb);
1878 if (error != 0)
1879 return (error);
1880
1881 /* Initiate PCI-E link training */
1882 intel_ntb_link_enable(ntb->device, NTB_SPEED_AUTO, NTB_WIDTH_AUTO);
1883
1884 callout_reset(&ntb->heartbeat_timer, 0, atom_link_hb, ntb);
1885
1886 return (0);
1887 }
1888
1889 /* XXX: Linux driver doesn't seem to do any of this for Atom. */
1890 static void
1891 configure_atom_secondary_side_bars(struct ntb_softc *ntb)
1892 {
1893
1894 if (ntb->dev_type == NTB_DEV_USD) {
1895 intel_ntb_reg_write(8, ATOM_PBAR2XLAT_OFFSET,
1896 XEON_B2B_BAR2_ADDR64);
1897 intel_ntb_reg_write(8, ATOM_PBAR4XLAT_OFFSET,
1898 XEON_B2B_BAR4_ADDR64);
1899 intel_ntb_reg_write(8, ATOM_MBAR23_OFFSET, XEON_B2B_BAR2_ADDR64);
1900 intel_ntb_reg_write(8, ATOM_MBAR45_OFFSET, XEON_B2B_BAR4_ADDR64);
1901 } else {
1902 intel_ntb_reg_write(8, ATOM_PBAR2XLAT_OFFSET,
1903 XEON_B2B_BAR2_ADDR64);
1904 intel_ntb_reg_write(8, ATOM_PBAR4XLAT_OFFSET,
1905 XEON_B2B_BAR4_ADDR64);
1906 intel_ntb_reg_write(8, ATOM_MBAR23_OFFSET, XEON_B2B_BAR2_ADDR64);
1907 intel_ntb_reg_write(8, ATOM_MBAR45_OFFSET, XEON_B2B_BAR4_ADDR64);
1908 }
1909 }
1910
1911 /*
1912 * When working around Xeon SDOORBELL errata by remapping remote registers in a
1913 * MW, limit the B2B MW to half a MW. By sharing a MW, half the shared MW
1914 * remains for use by a higher layer.
1915 *
1916 * Will only be used if working around SDOORBELL errata and the BIOS-configured
1917 * MW size is sufficiently large.
1918 */
1919 static unsigned int ntb_b2b_mw_share;
1920 SYSCTL_UINT(_hw_ntb, OID_AUTO, b2b_mw_share, CTLFLAG_RDTUN, &ntb_b2b_mw_share,
1921 0, "If enabled (non-zero), prefer to share half of the B2B peer register "
1922 "MW with higher level consumers. Both sides of the NTB MUST set the same "
1923 "value here.");
1924
1925 static void
1926 xeon_reset_sbar_size(struct ntb_softc *ntb, enum ntb_bar idx,
1927 enum ntb_bar regbar)
1928 {
1929 struct ntb_pci_bar_info *bar;
1930 uint8_t bar_sz;
1931
1932 if (!HAS_FEATURE(ntb, NTB_SPLIT_BAR) && idx >= NTB_B2B_BAR_3)
1933 return;
1934
1935 bar = &ntb->bar_info[idx];
1936 bar_sz = pci_read_config(ntb->device, bar->psz_off, 1);
1937 if (idx == regbar) {
1938 if (ntb->b2b_off != 0)
1939 bar_sz--;
1940 else
1941 bar_sz = 0;
1942 }
1943 pci_write_config(ntb->device, bar->ssz_off, bar_sz, 1);
1944 bar_sz = pci_read_config(ntb->device, bar->ssz_off, 1);
1945 (void)bar_sz;
1946 }
1947
1948 static void
1949 xeon_set_sbar_base_and_limit(struct ntb_softc *ntb, uint64_t bar_addr,
1950 enum ntb_bar idx, enum ntb_bar regbar)
1951 {
1952 uint64_t reg_val;
1953 uint32_t base_reg, lmt_reg;
1954
1955 bar_get_xlat_params(ntb, idx, &base_reg, NULL, &lmt_reg);
1956 if (idx == regbar) {
1957 if (ntb->b2b_off)
1958 bar_addr += ntb->b2b_off;
1959 else
1960 bar_addr = 0;
1961 }
1962
1963 if (!bar_is_64bit(ntb, idx)) {
1964 intel_ntb_reg_write(4, base_reg, bar_addr);
1965 reg_val = intel_ntb_reg_read(4, base_reg);
1966 (void)reg_val;
1967
1968 intel_ntb_reg_write(4, lmt_reg, bar_addr);
1969 reg_val = intel_ntb_reg_read(4, lmt_reg);
1970 (void)reg_val;
1971 } else {
1972 intel_ntb_reg_write(8, base_reg, bar_addr);
1973 reg_val = intel_ntb_reg_read(8, base_reg);
1974 (void)reg_val;
1975
1976 intel_ntb_reg_write(8, lmt_reg, bar_addr);
1977 reg_val = intel_ntb_reg_read(8, lmt_reg);
1978 (void)reg_val;
1979 }
1980 }
1981
1982 static void
1983 xeon_set_pbar_xlat(struct ntb_softc *ntb, uint64_t base_addr, enum ntb_bar idx)
1984 {
1985 struct ntb_pci_bar_info *bar;
1986
1987 bar = &ntb->bar_info[idx];
1988 if (HAS_FEATURE(ntb, NTB_SPLIT_BAR) && idx >= NTB_B2B_BAR_2) {
1989 intel_ntb_reg_write(4, bar->pbarxlat_off, base_addr);
1990 base_addr = intel_ntb_reg_read(4, bar->pbarxlat_off);
1991 } else {
1992 intel_ntb_reg_write(8, bar->pbarxlat_off, base_addr);
1993 base_addr = intel_ntb_reg_read(8, bar->pbarxlat_off);
1994 }
1995 (void)base_addr;
1996 }
1997
1998 static int
1999 xeon_setup_b2b_mw(struct ntb_softc *ntb, const struct ntb_b2b_addr *addr,
2000 const struct ntb_b2b_addr *peer_addr)
2001 {
2002 struct ntb_pci_bar_info *b2b_bar;
2003 vm_size_t bar_size;
2004 uint64_t bar_addr;
2005 enum ntb_bar b2b_bar_num, i;
2006
2007 if (ntb->b2b_mw_idx == B2B_MW_DISABLED) {
2008 b2b_bar = NULL;
2009 b2b_bar_num = NTB_CONFIG_BAR;
2010 ntb->b2b_off = 0;
2011 } else {
2012 b2b_bar_num = intel_ntb_mw_to_bar(ntb, ntb->b2b_mw_idx);
2013 KASSERT(b2b_bar_num > 0 && b2b_bar_num < NTB_MAX_BARS,
2014 ("invalid b2b mw bar"));
2015
2016 b2b_bar = &ntb->bar_info[b2b_bar_num];
2017 bar_size = b2b_bar->size;
2018
2019 if (ntb_b2b_mw_share != 0 &&
2020 (bar_size >> 1) >= XEON_B2B_MIN_SIZE)
2021 ntb->b2b_off = bar_size >> 1;
2022 else if (bar_size >= XEON_B2B_MIN_SIZE) {
2023 ntb->b2b_off = 0;
2024 } else {
2025 device_printf(ntb->device,
2026 "B2B bar size is too small!\n");
2027 return (EIO);
2028 }
2029 }
2030
2031 /*
2032 * Reset the secondary bar sizes to match the primary bar sizes.
2033 * (Except, disable or halve the size of the B2B secondary bar.)
2034 */
2035 for (i = NTB_B2B_BAR_1; i < NTB_MAX_BARS; i++)
2036 xeon_reset_sbar_size(ntb, i, b2b_bar_num);
2037
2038 bar_addr = 0;
2039 if (b2b_bar_num == NTB_CONFIG_BAR)
2040 bar_addr = addr->bar0_addr;
2041 else if (b2b_bar_num == NTB_B2B_BAR_1)
2042 bar_addr = addr->bar2_addr64;
2043 else if (b2b_bar_num == NTB_B2B_BAR_2 && !HAS_FEATURE(ntb, NTB_SPLIT_BAR))
2044 bar_addr = addr->bar4_addr64;
2045 else if (b2b_bar_num == NTB_B2B_BAR_2)
2046 bar_addr = addr->bar4_addr32;
2047 else if (b2b_bar_num == NTB_B2B_BAR_3)
2048 bar_addr = addr->bar5_addr32;
2049 else
2050 KASSERT(false, ("invalid bar"));
2051
2052 intel_ntb_reg_write(8, XEON_SBAR0BASE_OFFSET, bar_addr);
2053
2054 /*
2055 * Other SBARs are normally hit by the PBAR xlat, except for the b2b
2056 * register BAR. The B2B BAR is either disabled above or configured
2057 * half-size. It starts at PBAR xlat + offset.
2058 *
2059 * Also set up incoming BAR limits == base (zero length window).
2060 */
2061 xeon_set_sbar_base_and_limit(ntb, addr->bar2_addr64, NTB_B2B_BAR_1,
2062 b2b_bar_num);
2063 if (HAS_FEATURE(ntb, NTB_SPLIT_BAR)) {
2064 xeon_set_sbar_base_and_limit(ntb, addr->bar4_addr32,
2065 NTB_B2B_BAR_2, b2b_bar_num);
2066 xeon_set_sbar_base_and_limit(ntb, addr->bar5_addr32,
2067 NTB_B2B_BAR_3, b2b_bar_num);
2068 } else
2069 xeon_set_sbar_base_and_limit(ntb, addr->bar4_addr64,
2070 NTB_B2B_BAR_2, b2b_bar_num);
2071
2072 /* Zero incoming translation addrs */
2073 intel_ntb_reg_write(8, XEON_SBAR2XLAT_OFFSET, 0);
2074 intel_ntb_reg_write(8, XEON_SBAR4XLAT_OFFSET, 0);
2075
2076 if (HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP)) {
2077 uint32_t xlat_reg, lmt_reg;
2078 enum ntb_bar bar_num;
2079
2080 /*
2081 * We point the chosen MSIX MW BAR xlat to remote LAPIC for
2082 * workaround
2083 */
2084 bar_num = intel_ntb_mw_to_bar(ntb, ntb->msix_mw_idx);
2085 bar_get_xlat_params(ntb, bar_num, NULL, &xlat_reg, &lmt_reg);
2086 if (bar_is_64bit(ntb, bar_num)) {
2087 intel_ntb_reg_write(8, xlat_reg, MSI_INTEL_ADDR_BASE);
2088 ntb->msix_xlat = intel_ntb_reg_read(8, xlat_reg);
2089 intel_ntb_reg_write(8, lmt_reg, 0);
2090 } else {
2091 intel_ntb_reg_write(4, xlat_reg, MSI_INTEL_ADDR_BASE);
2092 ntb->msix_xlat = intel_ntb_reg_read(4, xlat_reg);
2093 intel_ntb_reg_write(4, lmt_reg, 0);
2094 }
2095
2096 ntb->peer_lapic_bar = &ntb->bar_info[bar_num];
2097 }
2098 (void)intel_ntb_reg_read(8, XEON_SBAR2XLAT_OFFSET);
2099 (void)intel_ntb_reg_read(8, XEON_SBAR4XLAT_OFFSET);
2100
2101 /* Zero outgoing translation limits (whole bar size windows) */
2102 intel_ntb_reg_write(8, XEON_PBAR2LMT_OFFSET, 0);
2103 intel_ntb_reg_write(8, XEON_PBAR4LMT_OFFSET, 0);
2104
2105 /* Set outgoing translation offsets */
2106 xeon_set_pbar_xlat(ntb, peer_addr->bar2_addr64, NTB_B2B_BAR_1);
2107 if (HAS_FEATURE(ntb, NTB_SPLIT_BAR)) {
2108 xeon_set_pbar_xlat(ntb, peer_addr->bar4_addr32, NTB_B2B_BAR_2);
2109 xeon_set_pbar_xlat(ntb, peer_addr->bar5_addr32, NTB_B2B_BAR_3);
2110 } else
2111 xeon_set_pbar_xlat(ntb, peer_addr->bar4_addr64, NTB_B2B_BAR_2);
2112
2113 /* Set the translation offset for B2B registers */
2114 bar_addr = 0;
2115 if (b2b_bar_num == NTB_CONFIG_BAR)
2116 bar_addr = peer_addr->bar0_addr;
2117 else if (b2b_bar_num == NTB_B2B_BAR_1)
2118 bar_addr = peer_addr->bar2_addr64;
2119 else if (b2b_bar_num == NTB_B2B_BAR_2 && !HAS_FEATURE(ntb, NTB_SPLIT_BAR))
2120 bar_addr = peer_addr->bar4_addr64;
2121 else if (b2b_bar_num == NTB_B2B_BAR_2)
2122 bar_addr = peer_addr->bar4_addr32;
2123 else if (b2b_bar_num == NTB_B2B_BAR_3)
2124 bar_addr = peer_addr->bar5_addr32;
2125 else
2126 KASSERT(false, ("invalid bar"));
2127
2128 /*
2129 * B2B_XLAT_OFFSET is a 64-bit register but can only be written 32 bits
2130 * at a time.
2131 */
2132 intel_ntb_reg_write(4, XEON_B2B_XLAT_OFFSETL, bar_addr & 0xffffffff);
2133 intel_ntb_reg_write(4, XEON_B2B_XLAT_OFFSETU, bar_addr >> 32);
2134 return (0);
2135 }
2136
2137 static int
2138 xeon_gen3_setup_b2b_mw(struct ntb_softc *ntb)
2139 {
2140 uint64_t reg;
2141 uint32_t embarsz, imbarsz;
2142
2143 /* IMBAR1SZ should be equal to EMBAR1SZ */
2144 embarsz = pci_read_config(ntb->device, XEON_GEN3_INT_REG_EMBAR1SZ, 1);
2145 imbarsz = pci_read_config(ntb->device, XEON_GEN3_INT_REG_IMBAR1SZ, 1);
2146 if (embarsz != imbarsz) {
2147 device_printf(ntb->device,
2148 "IMBAR1SZ (%u) should be equal to EMBAR1SZ (%u)\n",
2149 imbarsz, embarsz);
2150 return (EIO);
2151 }
2152
2153 /* IMBAR2SZ should be equal to EMBAR2SZ */
2154 embarsz = pci_read_config(ntb->device, XEON_GEN3_INT_REG_EMBAR2SZ, 1);
2155 imbarsz = pci_read_config(ntb->device, XEON_GEN3_INT_REG_IMBAR2SZ, 1);
2156 if (embarsz != imbarsz) {
2157 device_printf(ntb->device,
2158 "IMBAR2SZ (%u) should be equal to EMBAR2SZ (%u)\n",
2159 imbarsz, embarsz);
2160 return (EIO);
2161 }
2162
2163 /* Client will provide the incoming IMBAR1/2XBASE, zero it for now */
2164 intel_ntb_reg_write(8, XEON_GEN3_REG_IMBAR1XBASE, 0);
2165 intel_ntb_reg_write(8, XEON_GEN3_REG_IMBAR2XBASE, 0);
2166
2167 /*
2168 * If the value in IMBAR1XLIMIT is set equal to the value in IMBAR1XBASE,
2169 * the local memory window exposure from EMBAR1 is disabled.
2170 * Note: It is needed to avoid malicious access.
2171 */
2172 intel_ntb_reg_write(8, XEON_GEN3_REG_IMBAR1XLIMIT, 0);
2173 intel_ntb_reg_write(8, XEON_GEN3_REG_IMBAR2XLIMIT, 0);
2174
2175 /* Config outgoing translation limits (whole bar size windows) */
2176 reg = intel_ntb_reg_read(8, XEON_GEN3_REG_EMBAR1XBASE);
2177 reg += ntb->bar_info[NTB_B2B_BAR_1].size;
2178 intel_ntb_reg_write(8, XEON_GEN3_REG_EMBAR1XLIMIT, reg);
2179
2180 reg = intel_ntb_reg_read(8, XEON_GEN3_REG_EMBAR2XBASE);
2181 reg += ntb->bar_info[NTB_B2B_BAR_2].size;
2182 intel_ntb_reg_write(8, XEON_GEN3_REG_EMBAR2XLIMIT, reg);
2183
2184 return (0);
2185 }
2186
2187 static inline bool
2188 _xeon_link_is_up(struct ntb_softc *ntb)
2189 {
2190
2191 if (ntb->conn_type == NTB_CONN_TRANSPARENT)
2192 return (true);
2193 return ((ntb->lnk_sta & NTB_LINK_STATUS_ACTIVE) != 0);
2194 }
2195
2196 static inline bool
2197 link_is_up(struct ntb_softc *ntb)
2198 {
2199
2200 if (ntb->type == NTB_XEON_GEN1 || ntb->type == NTB_XEON_GEN3)
2201 return (_xeon_link_is_up(ntb) && (ntb->peer_msix_good ||
2202 !HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP)));
2203
2204 KASSERT(ntb->type == NTB_ATOM, ("ntb type"));
2205 return ((ntb->ntb_ctl & ATOM_CNTL_LINK_DOWN) == 0);
2206 }
2207
2208 static inline bool
2209 atom_link_is_err(struct ntb_softc *ntb)
2210 {
2211 uint32_t status;
2212
2213 KASSERT(ntb->type == NTB_ATOM, ("ntb type"));
2214
2215 status = intel_ntb_reg_read(4, ATOM_LTSSMSTATEJMP_OFFSET);
2216 if ((status & ATOM_LTSSMSTATEJMP_FORCEDETECT) != 0)
2217 return (true);
2218
2219 status = intel_ntb_reg_read(4, ATOM_IBSTERRRCRVSTS0_OFFSET);
2220 return ((status & ATOM_IBIST_ERR_OFLOW) != 0);
2221 }
2222
2223 /* Atom does not have link status interrupt, poll on that platform */
2224 static void
2225 atom_link_hb(void *arg)
2226 {
2227 struct ntb_softc *ntb = arg;
2228 sbintime_t timo, poll_ts;
2229
2230 timo = NTB_HB_TIMEOUT * hz;
2231 poll_ts = ntb->last_ts + timo;
2232
2233 /*
2234 * Delay polling the link status if an interrupt was received, unless
2235 * the cached link status says the link is down.
2236 */
2237 if ((sbintime_t)ticks - poll_ts < 0 && link_is_up(ntb)) {
2238 timo = poll_ts - ticks;
2239 goto out;
2240 }
2241
2242 if (intel_ntb_poll_link(ntb))
2243 ntb_link_event(ntb->device);
2244
2245 if (!link_is_up(ntb) && atom_link_is_err(ntb)) {
2246 /* Link is down with error, proceed with recovery */
2247 callout_reset(&ntb->lr_timer, 0, recover_atom_link, ntb);
2248 return;
2249 }
2250
2251 out:
2252 callout_reset(&ntb->heartbeat_timer, timo, atom_link_hb, ntb);
2253 }
2254
2255 static void
2256 atom_perform_link_restart(struct ntb_softc *ntb)
2257 {
2258 uint32_t status;
2259
2260 /* Driver resets the NTB ModPhy lanes - magic! */
2261 intel_ntb_reg_write(1, ATOM_MODPHY_PCSREG6, 0xe0);
2262 intel_ntb_reg_write(1, ATOM_MODPHY_PCSREG4, 0x40);
2263 intel_ntb_reg_write(1, ATOM_MODPHY_PCSREG4, 0x60);
2264 intel_ntb_reg_write(1, ATOM_MODPHY_PCSREG6, 0x60);
2265
2266 /* Driver waits 100ms to allow the NTB ModPhy to settle */
2267 pause("ModPhy", hz / 10);
2268
2269 /* Clear AER Errors, write to clear */
2270 status = intel_ntb_reg_read(4, ATOM_ERRCORSTS_OFFSET);
2271 status &= PCIM_AER_COR_REPLAY_ROLLOVER;
2272 intel_ntb_reg_write(4, ATOM_ERRCORSTS_OFFSET, status);
2273
2274 /* Clear unexpected electrical idle event in LTSSM, write to clear */
2275 status = intel_ntb_reg_read(4, ATOM_LTSSMERRSTS0_OFFSET);
2276 status |= ATOM_LTSSMERRSTS0_UNEXPECTEDEI;
2277 intel_ntb_reg_write(4, ATOM_LTSSMERRSTS0_OFFSET, status);
2278
2279 /* Clear DeSkew Buffer error, write to clear */
2280 status = intel_ntb_reg_read(4, ATOM_DESKEWSTS_OFFSET);
2281 status |= ATOM_DESKEWSTS_DBERR;
2282 intel_ntb_reg_write(4, ATOM_DESKEWSTS_OFFSET, status);
2283
2284 status = intel_ntb_reg_read(4, ATOM_IBSTERRRCRVSTS0_OFFSET);
2285 status &= ATOM_IBIST_ERR_OFLOW;
2286 intel_ntb_reg_write(4, ATOM_IBSTERRRCRVSTS0_OFFSET, status);
2287
2288 /* Releases the NTB state machine to allow the link to retrain */
2289 status = intel_ntb_reg_read(4, ATOM_LTSSMSTATEJMP_OFFSET);
2290 status &= ~ATOM_LTSSMSTATEJMP_FORCEDETECT;
2291 intel_ntb_reg_write(4, ATOM_LTSSMSTATEJMP_OFFSET, status);
2292 }
2293
2294 static int
2295 intel_ntb_port_number(device_t dev)
2296 {
2297 struct ntb_softc *ntb = device_get_softc(dev);
2298
2299 return (ntb->dev_type == NTB_DEV_USD ? 0 : 1);
2300 }
2301
2302 static int
2303 intel_ntb_peer_port_count(device_t dev)
2304 {
2305
2306 return (1);
2307 }
2308
2309 static int
2310 intel_ntb_peer_port_number(device_t dev, int pidx)
2311 {
2312 struct ntb_softc *ntb = device_get_softc(dev);
2313
2314 if (pidx != 0)
2315 return (-EINVAL);
2316
2317 return (ntb->dev_type == NTB_DEV_USD ? 1 : 0);
2318 }
2319
2320 static int
2321 intel_ntb_peer_port_idx(device_t dev, int port)
2322 {
2323 int peer_port;
2324
2325 peer_port = intel_ntb_peer_port_number(dev, 0);
2326 if (peer_port == -EINVAL || port != peer_port)
2327 return (-EINVAL);
2328
2329 return (0);
2330 }
2331
2332 static int
2333 intel_ntb_link_enable(device_t dev, enum ntb_speed speed __unused,
2334 enum ntb_width width __unused)
2335 {
2336 struct ntb_softc *ntb = device_get_softc(dev);
2337 uint32_t cntl;
2338
2339 intel_ntb_printf(2, "%s\n", __func__);
2340
2341 if (ntb->type == NTB_ATOM) {
2342 pci_write_config(ntb->device, NTB_PPD_OFFSET,
2343 ntb->ppd | ATOM_PPD_INIT_LINK, 4);
2344 return (0);
2345 }
2346
2347 if (ntb->conn_type == NTB_CONN_TRANSPARENT) {
2348 ntb_link_event(dev);
2349 return (0);
2350 }
2351
2352 cntl = intel_ntb_reg_read(4, ntb->reg->ntb_ctl);
2353 cntl &= ~(NTB_CNTL_LINK_DISABLE | NTB_CNTL_CFG_LOCK);
2354 cntl |= NTB_CNTL_P2S_BAR23_SNOOP | NTB_CNTL_S2P_BAR23_SNOOP;
2355 cntl |= NTB_CNTL_P2S_BAR4_SNOOP | NTB_CNTL_S2P_BAR4_SNOOP;
2356 if (HAS_FEATURE(ntb, NTB_SPLIT_BAR))
2357 cntl |= NTB_CNTL_P2S_BAR5_SNOOP | NTB_CNTL_S2P_BAR5_SNOOP;
2358 intel_ntb_reg_write(4, ntb->reg->ntb_ctl, cntl);
2359 return (0);
2360 }
2361
2362 static int
2363 intel_ntb_link_disable(device_t dev)
2364 {
2365 struct ntb_softc *ntb = device_get_softc(dev);
2366 uint32_t cntl;
2367
2368 intel_ntb_printf(2, "%s\n", __func__);
2369
2370 if (ntb->conn_type == NTB_CONN_TRANSPARENT) {
2371 ntb_link_event(dev);
2372 return (0);
2373 }
2374
2375 cntl = intel_ntb_reg_read(4, ntb->reg->ntb_ctl);
2376 cntl &= ~(NTB_CNTL_P2S_BAR23_SNOOP | NTB_CNTL_S2P_BAR23_SNOOP);
2377 cntl &= ~(NTB_CNTL_P2S_BAR4_SNOOP | NTB_CNTL_S2P_BAR4_SNOOP);
2378 if (HAS_FEATURE(ntb, NTB_SPLIT_BAR))
2379 cntl &= ~(NTB_CNTL_P2S_BAR5_SNOOP | NTB_CNTL_S2P_BAR5_SNOOP);
2380 cntl |= NTB_CNTL_LINK_DISABLE | NTB_CNTL_CFG_LOCK;
2381 intel_ntb_reg_write(4, ntb->reg->ntb_ctl, cntl);
2382 return (0);
2383 }
2384
2385 static bool
2386 intel_ntb_link_enabled(device_t dev)
2387 {
2388 struct ntb_softc *ntb = device_get_softc(dev);
2389 uint32_t cntl;
2390
2391 if (ntb->type == NTB_ATOM) {
2392 cntl = pci_read_config(ntb->device, NTB_PPD_OFFSET, 4);
2393 return ((cntl & ATOM_PPD_INIT_LINK) != 0);
2394 }
2395
2396 if (ntb->conn_type == NTB_CONN_TRANSPARENT)
2397 return (true);
2398
2399 cntl = intel_ntb_reg_read(4, ntb->reg->ntb_ctl);
2400 return ((cntl & NTB_CNTL_LINK_DISABLE) == 0);
2401 }
2402
2403 static void
2404 recover_atom_link(void *arg)
2405 {
2406 struct ntb_softc *ntb = arg;
2407 unsigned speed, width, oldspeed, oldwidth;
2408 uint32_t status32;
2409
2410 atom_perform_link_restart(ntb);
2411
2412 /*
2413 * There is a potential race between the 2 NTB devices recovering at
2414 * the same time. If the times are the same, the link will not recover
2415 * and the driver will be stuck in this loop forever. Add a random
2416 * interval to the recovery time to prevent this race.
2417 */
2418 status32 = arc4random() % ATOM_LINK_RECOVERY_TIME;
2419 pause("Link", (ATOM_LINK_RECOVERY_TIME + status32) * hz / 1000);
2420
2421 if (atom_link_is_err(ntb))
2422 goto retry;
2423
2424 status32 = intel_ntb_reg_read(4, ntb->reg->ntb_ctl);
2425 if ((status32 & ATOM_CNTL_LINK_DOWN) != 0)
2426 goto out;
2427
2428 status32 = intel_ntb_reg_read(4, ntb->reg->lnk_sta);
2429 width = NTB_LNK_STA_WIDTH(status32);
2430 speed = status32 & NTB_LINK_SPEED_MASK;
2431
2432 oldwidth = NTB_LNK_STA_WIDTH(ntb->lnk_sta);
2433 oldspeed = ntb->lnk_sta & NTB_LINK_SPEED_MASK;
2434 if (oldwidth != width || oldspeed != speed)
2435 goto retry;
2436
2437 out:
2438 callout_reset(&ntb->heartbeat_timer, NTB_HB_TIMEOUT * hz, atom_link_hb,
2439 ntb);
2440 return;
2441
2442 retry:
2443 callout_reset(&ntb->lr_timer, NTB_HB_TIMEOUT * hz, recover_atom_link,
2444 ntb);
2445 }
2446
2447 /*
2448 * Polls the HW link status register(s); returns true if something has changed.
2449 */
2450 static bool
2451 intel_ntb_poll_link(struct ntb_softc *ntb)
2452 {
2453 uint32_t ntb_cntl;
2454 uint16_t reg_val;
2455
2456 if (ntb->type == NTB_ATOM) {
2457 ntb_cntl = intel_ntb_reg_read(4, ntb->reg->ntb_ctl);
2458 if (ntb_cntl == ntb->ntb_ctl)
2459 return (false);
2460
2461 ntb->ntb_ctl = ntb_cntl;
2462 ntb->lnk_sta = intel_ntb_reg_read(4, ntb->reg->lnk_sta);
2463 } else {
2464 if (ntb->type == NTB_XEON_GEN1)
2465 db_iowrite_raw(ntb, ntb->self_reg->db_bell,
2466 ntb->db_link_mask);
2467
2468 reg_val = pci_read_config(ntb->device, ntb->reg->lnk_sta, 2);
2469 if (reg_val == ntb->lnk_sta)
2470 return (false);
2471
2472 ntb->lnk_sta = reg_val;
2473
2474 if (HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP)) {
2475 if (_xeon_link_is_up(ntb)) {
2476 if (!ntb->peer_msix_good) {
2477 callout_reset(&ntb->peer_msix_work, 0,
2478 intel_ntb_exchange_msix, ntb);
2479 return (false);
2480 }
2481 } else {
2482 ntb->peer_msix_good = false;
2483 ntb->peer_msix_done = false;
2484 }
2485 }
2486 }
2487 return (true);
2488 }
2489
2490 static inline enum ntb_speed
2491 intel_ntb_link_sta_speed(struct ntb_softc *ntb)
2492 {
2493
2494 if (!link_is_up(ntb))
2495 return (NTB_SPEED_NONE);
2496 return (ntb->lnk_sta & NTB_LINK_SPEED_MASK);
2497 }
2498
2499 static inline enum ntb_width
2500 intel_ntb_link_sta_width(struct ntb_softc *ntb)
2501 {
2502
2503 if (!link_is_up(ntb))
2504 return (NTB_WIDTH_NONE);
2505 return (NTB_LNK_STA_WIDTH(ntb->lnk_sta));
2506 }
2507
2508 SYSCTL_NODE(_hw_ntb, OID_AUTO, debug_info, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
2509 "Driver state, statistics, and HW registers");
2510
2511 #define NTB_REGSZ_MASK (3ul << 30)
2512 #define NTB_REG_64 (1ul << 30)
2513 #define NTB_REG_32 (2ul << 30)
2514 #define NTB_REG_16 (3ul << 30)
2515 #define NTB_REG_8 (0ul << 30)
2516
2517 #define NTB_DB_READ (1ul << 29)
2518 #define NTB_PCI_REG (1ul << 28)
2519 #define NTB_REGFLAGS_MASK (NTB_REGSZ_MASK | NTB_DB_READ | NTB_PCI_REG)
2520
2521 static void
2522 intel_ntb_sysctl_init(struct ntb_softc *ntb)
2523 {
2524 struct sysctl_oid_list *globals, *tree_par, *regpar, *statpar, *errpar;
2525 struct sysctl_ctx_list *ctx;
2526 struct sysctl_oid *tree, *tmptree;
2527
2528 ctx = device_get_sysctl_ctx(ntb->device);
2529 globals = SYSCTL_CHILDREN(device_get_sysctl_tree(ntb->device));
2530
2531 SYSCTL_ADD_PROC(ctx, globals, OID_AUTO, "link_status",
2532 CTLFLAG_RD | CTLTYPE_STRING | CTLFLAG_MPSAFE, ntb, 0,
2533 sysctl_handle_link_status_human, "A",
2534 "Link status (human readable)");
2535 SYSCTL_ADD_PROC(ctx, globals, OID_AUTO, "active",
2536 CTLFLAG_RD | CTLTYPE_UINT | CTLFLAG_MPSAFE, ntb, 0,
2537 sysctl_handle_link_status, "IU",
2538 "Link status (1=active, 0=inactive)");
2539 SYSCTL_ADD_PROC(ctx, globals, OID_AUTO, "admin_up",
2540 CTLFLAG_RW | CTLTYPE_UINT | CTLFLAG_MPSAFE, ntb, 0,
2541 sysctl_handle_link_admin, "IU",
2542 "Set/get interface status (1=UP, 0=DOWN)");
2543
2544 tree = SYSCTL_ADD_NODE(ctx, globals, OID_AUTO, "debug_info",
2545 CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
2546 "Driver state, statistics, and HW registers");
2547 tree_par = SYSCTL_CHILDREN(tree);
2548
2549 SYSCTL_ADD_UINT(ctx, tree_par, OID_AUTO, "conn_type", CTLFLAG_RD,
2550 &ntb->conn_type, 0, "0 - Transparent; 1 - B2B; 2 - Root Port");
2551 SYSCTL_ADD_UINT(ctx, tree_par, OID_AUTO, "dev_type", CTLFLAG_RD,
2552 &ntb->dev_type, 0, "0 - USD; 1 - DSD");
2553 SYSCTL_ADD_UINT(ctx, tree_par, OID_AUTO, "ppd", CTLFLAG_RD,
2554 &ntb->ppd, 0, "Raw PPD register (cached)");
2555
2556 if (ntb->b2b_mw_idx != B2B_MW_DISABLED) {
2557 SYSCTL_ADD_U8(ctx, tree_par, OID_AUTO, "b2b_idx", CTLFLAG_RD,
2558 &ntb->b2b_mw_idx, 0,
2559 "Index of the MW used for B2B remote register access");
2560 SYSCTL_ADD_UQUAD(ctx, tree_par, OID_AUTO, "b2b_off",
2561 CTLFLAG_RD, &ntb->b2b_off,
2562 "If non-zero, offset of B2B register region in shared MW");
2563 }
2564
2565 SYSCTL_ADD_PROC(ctx, tree_par, OID_AUTO, "features",
2566 CTLFLAG_RD | CTLTYPE_STRING | CTLFLAG_MPSAFE, ntb, 0,
2567 sysctl_handle_features, "A", "Features/errata of this NTB device");
2568
2569 SYSCTL_ADD_UINT(ctx, tree_par, OID_AUTO, "ntb_ctl", CTLFLAG_RD,
2570 __DEVOLATILE(uint32_t *, &ntb->ntb_ctl), 0,
2571 "NTB CTL register (cached)");
2572 SYSCTL_ADD_UINT(ctx, tree_par, OID_AUTO, "lnk_sta", CTLFLAG_RD,
2573 __DEVOLATILE(uint32_t *, &ntb->lnk_sta), 0,
2574 "LNK STA register (cached)");
2575
2576 SYSCTL_ADD_U8(ctx, tree_par, OID_AUTO, "mw_count", CTLFLAG_RD,
2577 &ntb->mw_count, 0, "MW count");
2578 SYSCTL_ADD_U8(ctx, tree_par, OID_AUTO, "spad_count", CTLFLAG_RD,
2579 &ntb->spad_count, 0, "Scratchpad count");
2580 SYSCTL_ADD_U8(ctx, tree_par, OID_AUTO, "db_count", CTLFLAG_RD,
2581 &ntb->db_count, 0, "Doorbell count");
2582 SYSCTL_ADD_U8(ctx, tree_par, OID_AUTO, "db_vec_count", CTLFLAG_RD,
2583 &ntb->db_vec_count, 0, "Doorbell vector count");
2584 SYSCTL_ADD_U8(ctx, tree_par, OID_AUTO, "db_vec_shift", CTLFLAG_RD,
2585 &ntb->db_vec_shift, 0, "Doorbell vector shift");
2586
2587 SYSCTL_ADD_UQUAD(ctx, tree_par, OID_AUTO, "db_valid_mask", CTLFLAG_RD,
2588 &ntb->db_valid_mask, "Doorbell valid mask");
2589 SYSCTL_ADD_UQUAD(ctx, tree_par, OID_AUTO, "db_link_mask", CTLFLAG_RD,
2590 &ntb->db_link_mask, "Doorbell link mask");
2591 SYSCTL_ADD_UQUAD(ctx, tree_par, OID_AUTO, "db_mask", CTLFLAG_RD,
2592 &ntb->db_mask, "Doorbell mask (cached)");
2593
2594 tmptree = SYSCTL_ADD_NODE(ctx, tree_par, OID_AUTO, "registers",
2595 CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
2596 "Raw HW registers (big-endian)");
2597 regpar = SYSCTL_CHILDREN(tmptree);
2598
2599 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "ntbcntl",
2600 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2601 NTB_REG_32 | ntb->reg->ntb_ctl, sysctl_handle_register, "IU",
2602 "NTB Control register");
2603 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "lnkcap",
2604 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2605 NTB_REG_32 | 0x19c, sysctl_handle_register, "IU",
2606 "NTB Link Capabilities");
2607 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "lnkcon",
2608 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2609 NTB_REG_32 | 0x1a0, sysctl_handle_register, "IU",
2610 "NTB Link Control register");
2611
2612 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "db_mask",
2613 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2614 NTB_REG_64 | NTB_DB_READ | ntb->self_reg->db_mask,
2615 sysctl_handle_register, "QU", "Doorbell mask register");
2616 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "db_bell",
2617 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2618 NTB_REG_64 | NTB_DB_READ | ntb->self_reg->db_bell,
2619 sysctl_handle_register, "QU", "Doorbell register");
2620
2621 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "incoming_xlat23",
2622 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2623 NTB_REG_64 | ntb->xlat_reg->bar2_xlat,
2624 sysctl_handle_register, "QU", "Incoming XLAT23 register");
2625 if (HAS_FEATURE(ntb, NTB_SPLIT_BAR)) {
2626 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "incoming_xlat4",
2627 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2628 NTB_REG_32 | ntb->xlat_reg->bar4_xlat,
2629 sysctl_handle_register, "IU", "Incoming XLAT4 register");
2630 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "incoming_xlat5",
2631 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2632 NTB_REG_32 | ntb->xlat_reg->bar5_xlat,
2633 sysctl_handle_register, "IU", "Incoming XLAT5 register");
2634 } else {
2635 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "incoming_xlat45",
2636 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2637 NTB_REG_64 | ntb->xlat_reg->bar4_xlat,
2638 sysctl_handle_register, "QU", "Incoming XLAT45 register");
2639 }
2640
2641 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "incoming_lmt23",
2642 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2643 NTB_REG_64 | ntb->xlat_reg->bar2_limit,
2644 sysctl_handle_register, "QU", "Incoming LMT23 register");
2645 if (HAS_FEATURE(ntb, NTB_SPLIT_BAR)) {
2646 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "incoming_lmt4",
2647 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2648 NTB_REG_32 | ntb->xlat_reg->bar4_limit,
2649 sysctl_handle_register, "IU", "Incoming LMT4 register");
2650 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "incoming_lmt5",
2651 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2652 NTB_REG_32 | ntb->xlat_reg->bar5_limit,
2653 sysctl_handle_register, "IU", "Incoming LMT5 register");
2654 } else {
2655 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "incoming_lmt45",
2656 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2657 NTB_REG_64 | ntb->xlat_reg->bar4_limit,
2658 sysctl_handle_register, "QU", "Incoming LMT45 register");
2659 }
2660
2661 if (ntb->type == NTB_ATOM)
2662 return;
2663
2664 tmptree = SYSCTL_ADD_NODE(ctx, regpar, OID_AUTO, "xeon_stats",
2665 CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Xeon HW statistics");
2666 statpar = SYSCTL_CHILDREN(tmptree);
2667 SYSCTL_ADD_PROC(ctx, statpar, OID_AUTO, "upstream_mem_miss",
2668 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2669 NTB_REG_16 | XEON_USMEMMISS_OFFSET,
2670 sysctl_handle_register, "SU", "Upstream Memory Miss");
2671
2672 tmptree = SYSCTL_ADD_NODE(ctx, regpar, OID_AUTO, "xeon_hw_err",
2673 CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Xeon HW errors");
2674 errpar = SYSCTL_CHILDREN(tmptree);
2675
2676 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "ppd",
2677 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2678 NTB_REG_8 | NTB_PCI_REG | NTB_PPD_OFFSET,
2679 sysctl_handle_register, "CU", "PPD");
2680
2681 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "pbar23_sz",
2682 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2683 NTB_REG_8 | NTB_PCI_REG | XEON_PBAR23SZ_OFFSET,
2684 sysctl_handle_register, "CU", "PBAR23 SZ (log2)");
2685 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "pbar4_sz",
2686 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2687 NTB_REG_8 | NTB_PCI_REG | XEON_PBAR4SZ_OFFSET,
2688 sysctl_handle_register, "CU", "PBAR4 SZ (log2)");
2689 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "pbar5_sz",
2690 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2691 NTB_REG_8 | NTB_PCI_REG | XEON_PBAR5SZ_OFFSET,
2692 sysctl_handle_register, "CU", "PBAR5 SZ (log2)");
2693
2694 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "sbar23_sz",
2695 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2696 NTB_REG_8 | NTB_PCI_REG | XEON_SBAR23SZ_OFFSET,
2697 sysctl_handle_register, "CU", "SBAR23 SZ (log2)");
2698 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "sbar4_sz",
2699 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2700 NTB_REG_8 | NTB_PCI_REG | XEON_SBAR4SZ_OFFSET,
2701 sysctl_handle_register, "CU", "SBAR4 SZ (log2)");
2702 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "sbar5_sz",
2703 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2704 NTB_REG_8 | NTB_PCI_REG | XEON_SBAR5SZ_OFFSET,
2705 sysctl_handle_register, "CU", "SBAR5 SZ (log2)");
2706
2707 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "devsts",
2708 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2709 NTB_REG_16 | NTB_PCI_REG | XEON_DEVSTS_OFFSET,
2710 sysctl_handle_register, "SU", "DEVSTS");
2711 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "lnksts",
2712 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2713 NTB_REG_16 | NTB_PCI_REG | XEON_LINK_STATUS_OFFSET,
2714 sysctl_handle_register, "SU", "LNKSTS");
2715 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "slnksts",
2716 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2717 NTB_REG_16 | NTB_PCI_REG | XEON_SLINK_STATUS_OFFSET,
2718 sysctl_handle_register, "SU", "SLNKSTS");
2719
2720 SYSCTL_ADD_PROC(ctx, errpar, OID_AUTO, "uncerrsts",
2721 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2722 NTB_REG_32 | NTB_PCI_REG | XEON_UNCERRSTS_OFFSET,
2723 sysctl_handle_register, "IU", "UNCERRSTS");
2724 SYSCTL_ADD_PROC(ctx, errpar, OID_AUTO, "corerrsts",
2725 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2726 NTB_REG_32 | NTB_PCI_REG | XEON_CORERRSTS_OFFSET,
2727 sysctl_handle_register, "IU", "CORERRSTS");
2728
2729 if (ntb->conn_type != NTB_CONN_B2B)
2730 return;
2731
2732 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "outgoing_xlat01l",
2733 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2734 NTB_REG_32 | XEON_B2B_XLAT_OFFSETL,
2735 sysctl_handle_register, "IU", "Outgoing XLAT0L register");
2736 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "outgoing_xlat01u",
2737 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2738 NTB_REG_32 | XEON_B2B_XLAT_OFFSETU,
2739 sysctl_handle_register, "IU", "Outgoing XLAT0U register");
2740 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "outgoing_xlat23",
2741 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2742 NTB_REG_64 | ntb->bar_info[NTB_B2B_BAR_1].pbarxlat_off,
2743 sysctl_handle_register, "QU", "Outgoing XLAT23 register");
2744 if (HAS_FEATURE(ntb, NTB_SPLIT_BAR)) {
2745 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "outgoing_xlat4",
2746 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2747 NTB_REG_32 | ntb->bar_info[NTB_B2B_BAR_2].pbarxlat_off,
2748 sysctl_handle_register, "IU", "Outgoing XLAT4 register");
2749 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "outgoing_xlat5",
2750 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2751 NTB_REG_32 | ntb->bar_info[NTB_B2B_BAR_3].pbarxlat_off,
2752 sysctl_handle_register, "IU", "Outgoing XLAT5 register");
2753 } else {
2754 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "outgoing_xlat45",
2755 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2756 NTB_REG_64 | ntb->bar_info[NTB_B2B_BAR_2].pbarxlat_off,
2757 sysctl_handle_register, "QU", "Outgoing XLAT45 register");
2758 }
2759
2760 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "outgoing_lmt23",
2761 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2762 NTB_REG_64 | XEON_PBAR2LMT_OFFSET,
2763 sysctl_handle_register, "QU", "Outgoing LMT23 register");
2764 if (HAS_FEATURE(ntb, NTB_SPLIT_BAR)) {
2765 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "outgoing_lmt4",
2766 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2767 NTB_REG_32 | XEON_PBAR4LMT_OFFSET,
2768 sysctl_handle_register, "IU", "Outgoing LMT4 register");
2769 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "outgoing_lmt5",
2770 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2771 NTB_REG_32 | XEON_PBAR5LMT_OFFSET,
2772 sysctl_handle_register, "IU", "Outgoing LMT5 register");
2773 } else {
2774 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "outgoing_lmt45",
2775 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2776 NTB_REG_64 | XEON_PBAR4LMT_OFFSET,
2777 sysctl_handle_register, "QU", "Outgoing LMT45 register");
2778 }
2779
2780 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "sbar01_base",
2781 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2782 NTB_REG_64 | ntb->xlat_reg->bar0_base,
2783 sysctl_handle_register, "QU", "Secondary BAR01 base register");
2784 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "sbar23_base",
2785 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2786 NTB_REG_64 | ntb->xlat_reg->bar2_base,
2787 sysctl_handle_register, "QU", "Secondary BAR23 base register");
2788 if (HAS_FEATURE(ntb, NTB_SPLIT_BAR)) {
2789 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "sbar4_base",
2790 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2791 NTB_REG_32 | ntb->xlat_reg->bar4_base,
2792 sysctl_handle_register, "IU",
2793 "Secondary BAR4 base register");
2794 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "sbar5_base",
2795 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2796 NTB_REG_32 | ntb->xlat_reg->bar5_base,
2797 sysctl_handle_register, "IU",
2798 "Secondary BAR5 base register");
2799 } else {
2800 SYSCTL_ADD_PROC(ctx, regpar, OID_AUTO, "sbar45_base",
2801 CTLFLAG_RD | CTLTYPE_OPAQUE | CTLFLAG_MPSAFE, ntb,
2802 NTB_REG_64 | ntb->xlat_reg->bar4_base,
2803 sysctl_handle_register, "QU",
2804 "Secondary BAR45 base register");
2805 }
2806 }
2807
2808 static int
2809 sysctl_handle_features(SYSCTL_HANDLER_ARGS)
2810 {
2811 struct ntb_softc *ntb = arg1;
2812 struct sbuf sb;
2813 int error;
2814
2815 sbuf_new_for_sysctl(&sb, NULL, 256, req);
2816
2817 sbuf_printf(&sb, "%b", ntb->features, NTB_FEATURES_STR);
2818 error = sbuf_finish(&sb);
2819 sbuf_delete(&sb);
2820
2821 if (error || !req->newptr)
2822 return (error);
2823 return (EINVAL);
2824 }
2825
2826 static int
2827 sysctl_handle_link_admin(SYSCTL_HANDLER_ARGS)
2828 {
2829 struct ntb_softc *ntb = arg1;
2830 unsigned old, new;
2831 int error;
2832
2833 old = intel_ntb_link_enabled(ntb->device);
2834
2835 error = SYSCTL_OUT(req, &old, sizeof(old));
2836 if (error != 0 || req->newptr == NULL)
2837 return (error);
2838
2839 error = SYSCTL_IN(req, &new, sizeof(new));
2840 if (error != 0)
2841 return (error);
2842
2843 intel_ntb_printf(0, "Admin set interface state to '%sabled'\n",
2844 (new != 0)? "en" : "dis");
2845
2846 if (new != 0)
2847 error = intel_ntb_link_enable(ntb->device, NTB_SPEED_AUTO, NTB_WIDTH_AUTO);
2848 else
2849 error = intel_ntb_link_disable(ntb->device);
2850 return (error);
2851 }
2852
2853 static int
2854 sysctl_handle_link_status_human(SYSCTL_HANDLER_ARGS)
2855 {
2856 struct ntb_softc *ntb = arg1;
2857 struct sbuf sb;
2858 enum ntb_speed speed;
2859 enum ntb_width width;
2860 int error;
2861
2862 sbuf_new_for_sysctl(&sb, NULL, 32, req);
2863
2864 if (intel_ntb_link_is_up(ntb->device, &speed, &width))
2865 sbuf_printf(&sb, "up / PCIe Gen %u / Width x%u",
2866 (unsigned)speed, (unsigned)width);
2867 else
2868 sbuf_printf(&sb, "down");
2869
2870 error = sbuf_finish(&sb);
2871 sbuf_delete(&sb);
2872
2873 if (error || !req->newptr)
2874 return (error);
2875 return (EINVAL);
2876 }
2877
2878 static int
2879 sysctl_handle_link_status(SYSCTL_HANDLER_ARGS)
2880 {
2881 struct ntb_softc *ntb = arg1;
2882 unsigned res;
2883 int error;
2884
2885 res = intel_ntb_link_is_up(ntb->device, NULL, NULL);
2886
2887 error = SYSCTL_OUT(req, &res, sizeof(res));
2888 if (error || !req->newptr)
2889 return (error);
2890 return (EINVAL);
2891 }
2892
2893 static int
2894 sysctl_handle_register(SYSCTL_HANDLER_ARGS)
2895 {
2896 struct ntb_softc *ntb;
2897 const void *outp;
2898 uintptr_t sz;
2899 uint64_t umv;
2900 char be[sizeof(umv)];
2901 size_t outsz;
2902 uint32_t reg;
2903 bool db, pci;
2904 int error;
2905
2906 ntb = arg1;
2907 reg = arg2 & ~NTB_REGFLAGS_MASK;
2908 sz = arg2 & NTB_REGSZ_MASK;
2909 db = (arg2 & NTB_DB_READ) != 0;
2910 pci = (arg2 & NTB_PCI_REG) != 0;
2911
2912 KASSERT(!(db && pci), ("bogus"));
2913
2914 if (db) {
2915 KASSERT(sz == NTB_REG_64, ("bogus"));
2916 umv = db_ioread(ntb, reg);
2917 outsz = sizeof(uint64_t);
2918 } else {
2919 switch (sz) {
2920 case NTB_REG_64:
2921 if (pci)
2922 umv = pci_read_config(ntb->device, reg, 8);
2923 else
2924 umv = intel_ntb_reg_read(8, reg);
2925 outsz = sizeof(uint64_t);
2926 break;
2927 case NTB_REG_32:
2928 if (pci)
2929 umv = pci_read_config(ntb->device, reg, 4);
2930 else
2931 umv = intel_ntb_reg_read(4, reg);
2932 outsz = sizeof(uint32_t);
2933 break;
2934 case NTB_REG_16:
2935 if (pci)
2936 umv = pci_read_config(ntb->device, reg, 2);
2937 else
2938 umv = intel_ntb_reg_read(2, reg);
2939 outsz = sizeof(uint16_t);
2940 break;
2941 case NTB_REG_8:
2942 if (pci)
2943 umv = pci_read_config(ntb->device, reg, 1);
2944 else
2945 umv = intel_ntb_reg_read(1, reg);
2946 outsz = sizeof(uint8_t);
2947 break;
2948 default:
2949 panic("bogus");
2950 break;
2951 }
2952 }
2953
2954 /* Encode bigendian so that sysctl -x is legible. */
2955 be64enc(be, umv);
2956 outp = ((char *)be) + sizeof(umv) - outsz;
2957
2958 error = SYSCTL_OUT(req, outp, outsz);
2959 if (error || !req->newptr)
2960 return (error);
2961 return (EINVAL);
2962 }
2963
2964 static unsigned
2965 intel_ntb_user_mw_to_idx(struct ntb_softc *ntb, unsigned uidx)
2966 {
2967
2968 if ((ntb->b2b_mw_idx != B2B_MW_DISABLED && ntb->b2b_off == 0 &&
2969 uidx >= ntb->b2b_mw_idx) ||
2970 (ntb->msix_mw_idx != B2B_MW_DISABLED && uidx >= ntb->msix_mw_idx))
2971 uidx++;
2972 if ((ntb->b2b_mw_idx != B2B_MW_DISABLED && ntb->b2b_off == 0 &&
2973 uidx >= ntb->b2b_mw_idx) &&
2974 (ntb->msix_mw_idx != B2B_MW_DISABLED && uidx >= ntb->msix_mw_idx))
2975 uidx++;
2976 return (uidx);
2977 }
2978
2979 #ifndef EARLY_AP_STARTUP
2980 static int msix_ready;
2981
2982 static void
2983 intel_ntb_msix_ready(void *arg __unused)
2984 {
2985
2986 msix_ready = 1;
2987 }
2988 SYSINIT(intel_ntb_msix_ready, SI_SUB_SMP, SI_ORDER_ANY,
2989 intel_ntb_msix_ready, NULL);
2990 #endif
2991
2992 static void
2993 intel_ntb_exchange_msix(void *ctx)
2994 {
2995 struct ntb_softc *ntb;
2996 uint32_t val;
2997 unsigned i;
2998
2999 ntb = ctx;
3000
3001 if (ntb->peer_msix_good)
3002 goto msix_good;
3003 if (ntb->peer_msix_done)
3004 goto msix_done;
3005
3006 #ifndef EARLY_AP_STARTUP
3007 /* Block MSIX negotiation until SMP started and IRQ reshuffled. */
3008 if (!msix_ready)
3009 goto reschedule;
3010 #endif
3011
3012 intel_ntb_get_msix_info(ntb);
3013 for (i = 0; i < XEON_NONLINK_DB_MSIX_BITS; i++) {
3014 intel_ntb_peer_spad_write(ntb->device, NTB_MSIX_DATA0 + i,
3015 ntb->msix_data[i].nmd_data);
3016 intel_ntb_peer_spad_write(ntb->device, NTB_MSIX_OFS0 + i,
3017 ntb->msix_data[i].nmd_ofs - ntb->msix_xlat);
3018 }
3019 intel_ntb_peer_spad_write(ntb->device, NTB_MSIX_GUARD, NTB_MSIX_VER_GUARD);
3020
3021 intel_ntb_spad_read(ntb->device, NTB_MSIX_GUARD, &val);
3022 if (val != NTB_MSIX_VER_GUARD)
3023 goto reschedule;
3024
3025 for (i = 0; i < XEON_NONLINK_DB_MSIX_BITS; i++) {
3026 intel_ntb_spad_read(ntb->device, NTB_MSIX_DATA0 + i, &val);
3027 intel_ntb_printf(2, "remote MSIX data(%u): 0x%x\n", i, val);
3028 ntb->peer_msix_data[i].nmd_data = val;
3029 intel_ntb_spad_read(ntb->device, NTB_MSIX_OFS0 + i, &val);
3030 intel_ntb_printf(2, "remote MSIX addr(%u): 0x%x\n", i, val);
3031 ntb->peer_msix_data[i].nmd_ofs = val;
3032 }
3033
3034 ntb->peer_msix_done = true;
3035
3036 msix_done:
3037 intel_ntb_peer_spad_write(ntb->device, NTB_MSIX_DONE, NTB_MSIX_RECEIVED);
3038 intel_ntb_spad_read(ntb->device, NTB_MSIX_DONE, &val);
3039 if (val != NTB_MSIX_RECEIVED)
3040 goto reschedule;
3041
3042 intel_ntb_spad_clear(ntb->device);
3043 ntb->peer_msix_good = true;
3044 /* Give peer time to see our NTB_MSIX_RECEIVED. */
3045 goto reschedule;
3046
3047 msix_good:
3048 intel_ntb_poll_link(ntb);
3049 ntb_link_event(ntb->device);
3050 return;
3051
3052 reschedule:
3053 ntb->lnk_sta = pci_read_config(ntb->device, ntb->reg->lnk_sta, 2);
3054 if (_xeon_link_is_up(ntb)) {
3055 callout_reset(&ntb->peer_msix_work,
3056 hz * (ntb->peer_msix_good ? 2 : 1) / 10,
3057 intel_ntb_exchange_msix, ntb);
3058 } else
3059 intel_ntb_spad_clear(ntb->device);
3060 }
3061
3062 /*
3063 * Public API to the rest of the OS
3064 */
3065
3066 static uint8_t
3067 intel_ntb_spad_count(device_t dev)
3068 {
3069 struct ntb_softc *ntb = device_get_softc(dev);
3070
3071 return (ntb->spad_count);
3072 }
3073
3074 static uint8_t
3075 intel_ntb_mw_count(device_t dev)
3076 {
3077 struct ntb_softc *ntb = device_get_softc(dev);
3078 uint8_t res;
3079
3080 res = ntb->mw_count;
3081 if (ntb->b2b_mw_idx != B2B_MW_DISABLED && ntb->b2b_off == 0)
3082 res--;
3083 if (ntb->msix_mw_idx != B2B_MW_DISABLED)
3084 res--;
3085 return (res);
3086 }
3087
3088 static int
3089 intel_ntb_spad_write(device_t dev, unsigned int idx, uint32_t val)
3090 {
3091 struct ntb_softc *ntb = device_get_softc(dev);
3092
3093 if (idx >= ntb->spad_count)
3094 return (EINVAL);
3095
3096 intel_ntb_reg_write(4, ntb->self_reg->spad + idx * 4, val);
3097
3098 return (0);
3099 }
3100
3101 /*
3102 * Zeros the local scratchpad.
3103 */
3104 static void
3105 intel_ntb_spad_clear(device_t dev)
3106 {
3107 struct ntb_softc *ntb = device_get_softc(dev);
3108 unsigned i;
3109
3110 for (i = 0; i < ntb->spad_count; i++)
3111 intel_ntb_spad_write(dev, i, 0);
3112 }
3113
3114 static int
3115 intel_ntb_spad_read(device_t dev, unsigned int idx, uint32_t *val)
3116 {
3117 struct ntb_softc *ntb = device_get_softc(dev);
3118
3119 if (idx >= ntb->spad_count)
3120 return (EINVAL);
3121
3122 *val = intel_ntb_reg_read(4, ntb->self_reg->spad + idx * 4);
3123
3124 return (0);
3125 }
3126
3127 static int
3128 intel_ntb_peer_spad_write(device_t dev, unsigned int idx, uint32_t val)
3129 {
3130 struct ntb_softc *ntb = device_get_softc(dev);
3131
3132 if (idx >= ntb->spad_count)
3133 return (EINVAL);
3134
3135 if (HAS_FEATURE(ntb, NTB_SDOORBELL_LOCKUP))
3136 intel_ntb_mw_write(4, XEON_SPAD_OFFSET + idx * 4, val);
3137 else
3138 intel_ntb_reg_write(4, ntb->peer_reg->spad + idx * 4, val);
3139
3140 return (0);
3141 }
3142
3143 static int
3144 intel_ntb_peer_spad_read(device_t dev, unsigned int idx, uint32_t *val)
3145 {
3146 struct ntb_softc *ntb = device_get_softc(dev);
3147
3148 if (idx >= ntb->spad_count)
3149 return (EINVAL);
3150
3151 if (HAS_FEATURE(ntb, NTB_SDOORBELL_LOCKUP))
3152 *val = intel_ntb_mw_read(4, XEON_SPAD_OFFSET + idx * 4);
3153 else
3154 *val = intel_ntb_reg_read(4, ntb->peer_reg->spad + idx * 4);
3155
3156 return (0);
3157 }
3158
3159 static int
3160 intel_ntb_mw_get_range(device_t dev, unsigned mw_idx, vm_paddr_t *base,
3161 caddr_t *vbase, size_t *size, size_t *align, size_t *align_size,
3162 bus_addr_t *plimit)
3163 {
3164 struct ntb_softc *ntb = device_get_softc(dev);
3165 struct ntb_pci_bar_info *bar;
3166 bus_addr_t limit;
3167 size_t bar_b2b_off;
3168 enum ntb_bar bar_num;
3169
3170 if (mw_idx >= intel_ntb_mw_count(dev))
3171 return (EINVAL);
3172 mw_idx = intel_ntb_user_mw_to_idx(ntb, mw_idx);
3173
3174 bar_num = intel_ntb_mw_to_bar(ntb, mw_idx);
3175 bar = &ntb->bar_info[bar_num];
3176 bar_b2b_off = 0;
3177 if (mw_idx == ntb->b2b_mw_idx) {
3178 KASSERT(ntb->b2b_off != 0,
3179 ("user shouldn't get non-shared b2b mw"));
3180 bar_b2b_off = ntb->b2b_off;
3181 }
3182
3183 if (bar_is_64bit(ntb, bar_num))
3184 limit = BUS_SPACE_MAXADDR;
3185 else
3186 limit = BUS_SPACE_MAXADDR_32BIT;
3187
3188 if (base != NULL)
3189 *base = bar->pbase + bar_b2b_off;
3190 if (vbase != NULL)
3191 *vbase = bar->vbase + bar_b2b_off;
3192 if (size != NULL)
3193 *size = bar->size - bar_b2b_off;
3194 if (align != NULL)
3195 *align = bar->size;
3196 if (align_size != NULL)
3197 *align_size = 1;
3198 if (plimit != NULL)
3199 *plimit = limit;
3200 return (0);
3201 }
3202
3203 static int
3204 intel_ntb_mw_set_trans(device_t dev, unsigned idx, bus_addr_t addr, size_t size)
3205 {
3206 struct ntb_softc *ntb = device_get_softc(dev);
3207 struct ntb_pci_bar_info *bar;
3208 uint64_t base, limit, reg_val;
3209 size_t bar_size, mw_size;
3210 uint32_t base_reg, xlat_reg, limit_reg;
3211 enum ntb_bar bar_num;
3212
3213 if (idx >= intel_ntb_mw_count(dev))
3214 return (EINVAL);
3215 idx = intel_ntb_user_mw_to_idx(ntb, idx);
3216
3217 bar_num = intel_ntb_mw_to_bar(ntb, idx);
3218 bar = &ntb->bar_info[bar_num];
3219
3220 bar_size = bar->size;
3221 if (idx == ntb->b2b_mw_idx)
3222 mw_size = bar_size - ntb->b2b_off;
3223 else
3224 mw_size = bar_size;
3225
3226 /* Hardware requires that addr is aligned to bar size */
3227 if ((addr & (bar_size - 1)) != 0)
3228 return (EINVAL);
3229
3230 if (size > mw_size)
3231 return (EINVAL);
3232
3233 bar_get_xlat_params(ntb, bar_num, &base_reg, &xlat_reg, &limit_reg);
3234
3235 limit = 0;
3236 if (bar_is_64bit(ntb, bar_num)) {
3237 if (ntb->type == NTB_XEON_GEN3)
3238 base = addr;
3239 else
3240 base = intel_ntb_reg_read(8, base_reg) & BAR_HIGH_MASK;
3241
3242 if (limit_reg != 0 && size != mw_size)
3243 limit = base + size;
3244 else
3245 limit = base + mw_size;
3246
3247 /* Set and verify translation address */
3248 intel_ntb_reg_write(8, xlat_reg, addr);
3249 reg_val = intel_ntb_reg_read(8, xlat_reg) & BAR_HIGH_MASK;
3250 if (reg_val != addr) {
3251 intel_ntb_reg_write(8, xlat_reg, 0);
3252 return (EIO);
3253 }
3254
3255 /* Set and verify the limit */
3256 intel_ntb_reg_write(8, limit_reg, limit);
3257 reg_val = intel_ntb_reg_read(8, limit_reg) & BAR_HIGH_MASK;
3258 if (reg_val != limit) {
3259 intel_ntb_reg_write(8, limit_reg, base);
3260 intel_ntb_reg_write(8, xlat_reg, 0);
3261 return (EIO);
3262 }
3263 } else {
3264 /* Configure 32-bit (split) BAR MW */
3265 if (ntb->type == NTB_XEON_GEN3)
3266 return (EIO);
3267
3268 if ((addr & UINT32_MAX) != addr)
3269 return (ERANGE);
3270 if (((addr + size) & UINT32_MAX) != (addr + size))
3271 return (ERANGE);
3272
3273 base = intel_ntb_reg_read(4, base_reg) & BAR_HIGH_MASK;
3274
3275 if (limit_reg != 0 && size != mw_size)
3276 limit = base + size;
3277
3278 /* Set and verify translation address */
3279 intel_ntb_reg_write(4, xlat_reg, addr);
3280 reg_val = intel_ntb_reg_read(4, xlat_reg) & BAR_HIGH_MASK;
3281 if (reg_val != addr) {
3282 intel_ntb_reg_write(4, xlat_reg, 0);
3283 return (EIO);
3284 }
3285
3286 /* Set and verify the limit */
3287 intel_ntb_reg_write(4, limit_reg, limit);
3288 reg_val = intel_ntb_reg_read(4, limit_reg) & BAR_HIGH_MASK;
3289 if (reg_val != limit) {
3290 intel_ntb_reg_write(4, limit_reg, base);
3291 intel_ntb_reg_write(4, xlat_reg, 0);
3292 return (EIO);
3293 }
3294 }
3295 return (0);
3296 }
3297
3298 static int
3299 intel_ntb_mw_clear_trans(device_t dev, unsigned mw_idx)
3300 {
3301
3302 return (intel_ntb_mw_set_trans(dev, mw_idx, 0, 0));
3303 }
3304
3305 static int
3306 intel_ntb_mw_get_wc(device_t dev, unsigned idx, vm_memattr_t *mode)
3307 {
3308 struct ntb_softc *ntb = device_get_softc(dev);
3309 struct ntb_pci_bar_info *bar;
3310
3311 if (idx >= intel_ntb_mw_count(dev))
3312 return (EINVAL);
3313 idx = intel_ntb_user_mw_to_idx(ntb, idx);
3314
3315 bar = &ntb->bar_info[intel_ntb_mw_to_bar(ntb, idx)];
3316 *mode = bar->map_mode;
3317 return (0);
3318 }
3319
3320 static int
3321 intel_ntb_mw_set_wc(device_t dev, unsigned idx, vm_memattr_t mode)
3322 {
3323 struct ntb_softc *ntb = device_get_softc(dev);
3324
3325 if (idx >= intel_ntb_mw_count(dev))
3326 return (EINVAL);
3327
3328 idx = intel_ntb_user_mw_to_idx(ntb, idx);
3329 return (intel_ntb_mw_set_wc_internal(ntb, idx, mode));
3330 }
3331
3332 static int
3333 intel_ntb_mw_set_wc_internal(struct ntb_softc *ntb, unsigned idx, vm_memattr_t mode)
3334 {
3335 struct ntb_pci_bar_info *bar;
3336 int rc;
3337
3338 bar = &ntb->bar_info[intel_ntb_mw_to_bar(ntb, idx)];
3339 if (bar->map_mode == mode)
3340 return (0);
3341
3342 rc = pmap_change_attr((vm_offset_t)bar->vbase, bar->size, mode);
3343 if (rc == 0)
3344 bar->map_mode = mode;
3345
3346 return (rc);
3347 }
3348
3349 static void
3350 intel_ntb_peer_db_set(device_t dev, uint64_t bits)
3351 {
3352 struct ntb_softc *ntb = device_get_softc(dev);
3353 uint64_t db;
3354
3355 if ((bits & ~ntb->db_valid_mask) != 0) {
3356 device_printf(ntb->device, "Invalid doorbell bits %#jx\n",
3357 (uintmax_t)bits);
3358 return;
3359 }
3360
3361 if (HAS_FEATURE(ntb, NTB_SB01BASE_LOCKUP)) {
3362 struct ntb_pci_bar_info *lapic;
3363 unsigned i;
3364
3365 lapic = ntb->peer_lapic_bar;
3366
3367 for (i = 0; i < XEON_NONLINK_DB_MSIX_BITS; i++) {
3368 if ((bits & intel_ntb_db_vector_mask(dev, i)) != 0)
3369 bus_space_write_4(lapic->pci_bus_tag,
3370 lapic->pci_bus_handle,
3371 ntb->peer_msix_data[i].nmd_ofs,
3372 ntb->peer_msix_data[i].nmd_data);
3373 }
3374 return;
3375 }
3376
3377 if (HAS_FEATURE(ntb, NTB_SDOORBELL_LOCKUP)) {
3378 intel_ntb_mw_write(2, XEON_PDOORBELL_OFFSET, bits);
3379 return;
3380 }
3381
3382 if (ntb->type == NTB_XEON_GEN3) {
3383 while (bits != 0) {
3384 db = ffsll(bits);
3385
3386 intel_ntb_reg_write(1,
3387 ntb->peer_reg->db_bell + (db - 1) * 4, 0x1);
3388
3389 bits = bits & (bits - 1);
3390 }
3391 } else {
3392 db_iowrite(ntb, ntb->peer_reg->db_bell, bits);
3393 }
3394 }
3395
3396 static int
3397 intel_ntb_peer_db_addr(device_t dev, bus_addr_t *db_addr, vm_size_t *db_size)
3398 {
3399 struct ntb_softc *ntb = device_get_softc(dev);
3400 struct ntb_pci_bar_info *bar;
3401 uint64_t regoff;
3402
3403 KASSERT((db_addr != NULL && db_size != NULL), ("must be non-NULL"));
3404
3405 if (!HAS_FEATURE(ntb, NTB_SDOORBELL_LOCKUP)) {
3406 bar = &ntb->bar_info[NTB_CONFIG_BAR];
3407 regoff = ntb->peer_reg->db_bell;
3408 } else {
3409 KASSERT(ntb->b2b_mw_idx != B2B_MW_DISABLED,
3410 ("invalid b2b idx"));
3411
3412 bar = &ntb->bar_info[intel_ntb_mw_to_bar(ntb, ntb->b2b_mw_idx)];
3413 regoff = XEON_PDOORBELL_OFFSET;
3414 }
3415 KASSERT(bar->pci_bus_tag != X86_BUS_SPACE_IO, ("uh oh"));
3416
3417 /* HACK: Specific to current x86 bus implementation. */
3418 *db_addr = ((uint64_t)bar->pci_bus_handle + regoff);
3419 *db_size = ntb->reg->db_size;
3420 return (0);
3421 }
3422
3423 static uint64_t
3424 intel_ntb_db_valid_mask(device_t dev)
3425 {
3426 struct ntb_softc *ntb = device_get_softc(dev);
3427
3428 return (ntb->db_valid_mask);
3429 }
3430
3431 static int
3432 intel_ntb_db_vector_count(device_t dev)
3433 {
3434 struct ntb_softc *ntb = device_get_softc(dev);
3435
3436 return (ntb->db_vec_count);
3437 }
3438
3439 static uint64_t
3440 intel_ntb_db_vector_mask(device_t dev, uint32_t vector)
3441 {
3442 struct ntb_softc *ntb = device_get_softc(dev);
3443
3444 if (vector > ntb->db_vec_count)
3445 return (0);
3446 return (ntb->db_valid_mask & intel_ntb_vec_mask(ntb, vector));
3447 }
3448
3449 static bool
3450 intel_ntb_link_is_up(device_t dev, enum ntb_speed *speed, enum ntb_width *width)
3451 {
3452 struct ntb_softc *ntb = device_get_softc(dev);
3453
3454 if (speed != NULL)
3455 *speed = intel_ntb_link_sta_speed(ntb);
3456 if (width != NULL)
3457 *width = intel_ntb_link_sta_width(ntb);
3458 return (link_is_up(ntb));
3459 }
3460
3461 static void
3462 save_bar_parameters(struct ntb_pci_bar_info *bar)
3463 {
3464
3465 bar->pci_bus_tag = rman_get_bustag(bar->pci_resource);
3466 bar->pci_bus_handle = rman_get_bushandle(bar->pci_resource);
3467 bar->pbase = rman_get_start(bar->pci_resource);
3468 bar->size = rman_get_size(bar->pci_resource);
3469 bar->vbase = rman_get_virtual(bar->pci_resource);
3470 }
3471
3472 static device_method_t ntb_intel_methods[] = {
3473 /* Device interface */
3474 DEVMETHOD(device_probe, intel_ntb_probe),
3475 DEVMETHOD(device_attach, intel_ntb_attach),
3476 DEVMETHOD(device_detach, intel_ntb_detach),
3477 /* Bus interface */
3478 DEVMETHOD(bus_child_location, ntb_child_location),
3479 DEVMETHOD(bus_print_child, ntb_print_child),
3480 DEVMETHOD(bus_get_dma_tag, ntb_get_dma_tag),
3481 /* NTB interface */
3482 DEVMETHOD(ntb_port_number, intel_ntb_port_number),
3483 DEVMETHOD(ntb_peer_port_count, intel_ntb_peer_port_count),
3484 DEVMETHOD(ntb_peer_port_number, intel_ntb_peer_port_number),
3485 DEVMETHOD(ntb_peer_port_idx, intel_ntb_peer_port_idx),
3486 DEVMETHOD(ntb_link_is_up, intel_ntb_link_is_up),
3487 DEVMETHOD(ntb_link_enable, intel_ntb_link_enable),
3488 DEVMETHOD(ntb_link_disable, intel_ntb_link_disable),
3489 DEVMETHOD(ntb_link_enabled, intel_ntb_link_enabled),
3490 DEVMETHOD(ntb_mw_count, intel_ntb_mw_count),
3491 DEVMETHOD(ntb_mw_get_range, intel_ntb_mw_get_range),
3492 DEVMETHOD(ntb_mw_set_trans, intel_ntb_mw_set_trans),
3493 DEVMETHOD(ntb_mw_clear_trans, intel_ntb_mw_clear_trans),
3494 DEVMETHOD(ntb_mw_get_wc, intel_ntb_mw_get_wc),
3495 DEVMETHOD(ntb_mw_set_wc, intel_ntb_mw_set_wc),
3496 DEVMETHOD(ntb_spad_count, intel_ntb_spad_count),
3497 DEVMETHOD(ntb_spad_clear, intel_ntb_spad_clear),
3498 DEVMETHOD(ntb_spad_write, intel_ntb_spad_write),
3499 DEVMETHOD(ntb_spad_read, intel_ntb_spad_read),
3500 DEVMETHOD(ntb_peer_spad_write, intel_ntb_peer_spad_write),
3501 DEVMETHOD(ntb_peer_spad_read, intel_ntb_peer_spad_read),
3502 DEVMETHOD(ntb_db_valid_mask, intel_ntb_db_valid_mask),
3503 DEVMETHOD(ntb_db_vector_count, intel_ntb_db_vector_count),
3504 DEVMETHOD(ntb_db_vector_mask, intel_ntb_db_vector_mask),
3505 DEVMETHOD(ntb_db_clear, intel_ntb_db_clear),
3506 DEVMETHOD(ntb_db_clear_mask, intel_ntb_db_clear_mask),
3507 DEVMETHOD(ntb_db_read, intel_ntb_db_read),
3508 DEVMETHOD(ntb_db_set_mask, intel_ntb_db_set_mask),
3509 DEVMETHOD(ntb_peer_db_addr, intel_ntb_peer_db_addr),
3510 DEVMETHOD(ntb_peer_db_set, intel_ntb_peer_db_set),
3511 DEVMETHOD_END
3512 };
3513
3514 static DEFINE_CLASS_0(ntb_hw, ntb_intel_driver, ntb_intel_methods,
3515 sizeof(struct ntb_softc));
3516 DRIVER_MODULE(ntb_hw_intel, pci, ntb_intel_driver, NULL, NULL);
3517 MODULE_DEPEND(ntb_hw_intel, ntb, 1, 1, 1);
3518 MODULE_VERSION(ntb_hw_intel, 1);
3519 MODULE_PNP_INFO("W32:vendor/device;D:#", pci, ntb_hw_intel, pci_ids,
3520 nitems(pci_ids));
Cache object: 3e1526e1fa5bbbf232c8fd423553b6e3
|