1 /* SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause */
3 * Copyright(c) 2020 Cornelis Networks, Inc.
4 * Copyright(c) 2015-2020 Intel Corporation.
10 #include <linux/refcount.h>
11 #include <linux/interrupt.h>
12 #include <linux/pci.h>
13 #include <linux/dma-mapping.h>
14 #include <linux/mutex.h>
15 #include <linux/list.h>
16 #include <linux/scatterlist.h>
17 #include <linux/slab.h>
20 #include <linux/completion.h>
21 #include <linux/kref.h>
22 #include <linux/sched.h>
23 #include <linux/cdev.h>
24 #include <linux/delay.h>
25 #include <linux/kthread.h>
26 #include <linux/i2c.h>
27 #include <linux/i2c-algo-bit.h>
28 #include <linux/xarray.h>
29 #include <rdma/ib_hdrs.h>
30 #include <rdma/opa_addr.h>
31 #include <linux/rhashtable.h>
32 #include <rdma/rdma_vt.h>
34 #include "chip_registers.h"
46 /* bumped 1 from s/w major version of TrueScale */
47 #define HFI1_CHIP_VERS_MAJ 3U
49 /* don't care about this except printing */
50 #define HFI1_CHIP_VERS_MIN 0U
52 /* The Organization Unique Identifier (Mfg code), and its position in GUID */
53 #define HFI1_OUI 0x001175
54 #define HFI1_OUI_LSB 40
56 #define DROP_PACKET_OFF 0
57 #define DROP_PACKET_ON 1
59 #define NEIGHBOR_TYPE_HFI 0
60 #define NEIGHBOR_TYPE_SWITCH 1
62 #define HFI1_MAX_ACTIVE_WORKQUEUE_ENTRIES 5
64 extern unsigned long hfi1_cap_mask;
65 #define HFI1_CAP_KGET_MASK(mask, cap) ((mask) & HFI1_CAP_##cap)
66 #define HFI1_CAP_UGET_MASK(mask, cap) \
67 (((mask) >> HFI1_CAP_USER_SHIFT) & HFI1_CAP_##cap)
68 #define HFI1_CAP_KGET(cap) (HFI1_CAP_KGET_MASK(hfi1_cap_mask, cap))
69 #define HFI1_CAP_UGET(cap) (HFI1_CAP_UGET_MASK(hfi1_cap_mask, cap))
70 #define HFI1_CAP_IS_KSET(cap) (!!HFI1_CAP_KGET(cap))
71 #define HFI1_CAP_IS_USET(cap) (!!HFI1_CAP_UGET(cap))
72 #define HFI1_MISC_GET() ((hfi1_cap_mask >> HFI1_CAP_MISC_SHIFT) & \
74 /* Offline Disabled Reason is 4-bits */
75 #define HFI1_ODR_MASK(rsn) ((rsn) & OPA_PI_MASK_OFFLINE_REASON)
78 * Control context is always 0 and handles the error packets.
79 * It also handles the VL15 and multicast packets.
81 #define HFI1_CTRL_CTXT 0
84 * Driver context will store software counters for each of the events
85 * associated with these status registers
87 #define NUM_CCE_ERR_STATUS_COUNTERS 41
88 #define NUM_RCV_ERR_STATUS_COUNTERS 64
89 #define NUM_MISC_ERR_STATUS_COUNTERS 13
90 #define NUM_SEND_PIO_ERR_STATUS_COUNTERS 36
91 #define NUM_SEND_DMA_ERR_STATUS_COUNTERS 4
92 #define NUM_SEND_EGRESS_ERR_STATUS_COUNTERS 64
93 #define NUM_SEND_ERR_STATUS_COUNTERS 3
94 #define NUM_SEND_CTXT_ERR_STATUS_COUNTERS 5
95 #define NUM_SEND_DMA_ENG_ERR_STATUS_COUNTERS 24
98 * per driver stats, either not device nor port-specific, or
99 * summed over all of the devices and ports.
100 * They are described by name via ipathfs filesystem, so layout
101 * and number of elements can change without breaking compatibility.
102 * If members are added or deleted hfi1_statnames[] in debugfs.c must
105 struct hfi1_ib_stats {
106 __u64 sps_ints; /* number of interrupts handled */
107 __u64 sps_errints; /* number of error interrupts */
108 __u64 sps_txerrs; /* tx-related packet errors */
109 __u64 sps_rcverrs; /* non-crc rcv packet errors */
110 __u64 sps_hwerrs; /* hardware errors reported (parity, etc.) */
111 __u64 sps_nopiobufs; /* no pio bufs avail from kernel */
112 __u64 sps_ctxts; /* number of contexts currently open */
113 __u64 sps_lenerrs; /* number of kernel packets where RHF != LRH len */
118 extern struct hfi1_ib_stats hfi1_stats;
119 extern const struct pci_error_handlers hfi1_pci_err_handler;
121 extern int num_driver_cntrs;
124 * First-cut criterion for "device is active" is
125 * two thousand dwords combined Tx, Rx traffic per
126 * 5-second interval. SMA packets are 64 dwords,
127 * and occur "a few per second", presumably each way.
129 #define HFI1_TRAFFIC_ACTIVE_THRESHOLD (2000)
132 * Below contains all data related to a single context (formerly called port).
135 struct hfi1_opcode_stats_perctx;
137 struct ctxt_eager_bufs {
138 struct eager_buffer {
147 u32 size; /* total size of eager buffers */
148 u32 rcvtid_size; /* size of each eager rcv tid */
149 u16 count; /* size of buffers array */
150 u16 numbufs; /* number of buffers allocated */
151 u16 alloced; /* number of rcvarray entries used */
152 u16 threshold; /* head update threshold */
156 struct list_head list;
160 struct hfi1_ctxtdata;
161 typedef int (*intr_handler)(struct hfi1_ctxtdata *rcd, int data);
162 typedef void (*rhf_rcv_function_ptr)(struct hfi1_packet *packet);
165 struct list_head queue_head;
166 /* queue head for QP TID resource waiters */
167 u32 enqueue; /* count of tid enqueues */
168 u32 dequeue; /* count of tid dequeues */
171 struct hfi1_ctxtdata {
172 /* rcvhdrq base, needs mmap before useful */
174 /* kernel virtual address where hdrqtail is updated */
175 volatile __le64 *rcvhdrtail_kvaddr;
176 /* so functions that need physical port can get it easily */
177 struct hfi1_pportdata *ppd;
178 /* so file ops can get at unit */
179 struct hfi1_devdata *dd;
180 /* this receive context's assigned PIO ACK send context */
181 struct send_context *sc;
182 /* per context recv functions */
183 const rhf_rcv_function_ptr *rhf_rcv_function_map;
185 * The interrupt handler for a particular receive context can vary
186 * throughout it's lifetime. This is not a lock protected data member so
187 * it must be updated atomically and the prev and new value must always
188 * be valid. Worst case is we process an extra interrupt and up to 64
189 * packets with the wrong interrupt handler.
191 intr_handler do_interrupt;
192 /** fast handler after autoactive */
193 intr_handler fast_handler;
195 intr_handler slow_handler;
196 /* napi pointer assiociated with netdev */
197 struct napi_struct *napi;
198 /* verbs rx_stats per rcd */
199 struct hfi1_opcode_stats_perctx *opstats;
200 /* clear interrupt mask */
202 /* ctxt rcvhdrq head offset */
204 /* number of rcvhdrq entries */
206 u8 ireg; /* clear interrupt register */
207 /* receive packet sequence counter */
209 /* size of each of the rcvhdrq entries */
211 /* offset of RHF within receive header entry */
213 /* dynamic receive available interrupt timeout */
215 /* Indicates that this is vnic context */
217 /* vnic queue index this context is mapped to */
219 /* Is ASPM interrupt supported for this context */
220 bool aspm_intr_supported;
221 /* ASPM state (enabled/disabled) for this context */
223 /* Is ASPM processing enabled for this context (in intr context) */
224 bool aspm_intr_enable;
225 struct ctxt_eager_bufs egrbufs;
226 /* QPs waiting for context processing */
227 struct list_head qp_wait_list;
228 /* tid allocation lists */
229 struct exp_tid_set tid_group_list;
230 struct exp_tid_set tid_used_list;
231 struct exp_tid_set tid_full_list;
233 /* Timer for re-enabling ASPM if interrupt activity quiets down */
234 struct timer_list aspm_timer;
235 /* per-context configuration flags */
237 /* array of tid_groups */
238 struct tid_group *groups;
239 /* mmap of hdrq, must fit in 44 bits */
240 dma_addr_t rcvhdrq_dma;
241 dma_addr_t rcvhdrqtailaddr_dma;
242 /* Last interrupt timestamp */
243 ktime_t aspm_ts_last_intr;
244 /* Last timestamp at which we scheduled a timer for this context */
245 ktime_t aspm_ts_timer_sched;
246 /* Lock to serialize between intr, timer intr and user threads */
247 spinlock_t aspm_lock;
248 /* Reference count the base context usage */
250 /* numa node of this context */
252 /* associated msix interrupt. */
256 /* number of RcvArray groups for this context. */
257 u16 rcv_array_groups;
258 /* index of first eager TID entry. */
260 /* number of expected TID entries */
262 /* index of first expected TID entry. */
264 /* Device context index */
267 /* PSM Specific fields */
268 /* lock protecting all Expected TID data */
269 struct mutex exp_mutex;
270 /* lock protecting all Expected TID data of kernel contexts */
272 /* Queue for QP's waiting for HW TID flows */
273 struct tid_queue flow_queue;
274 /* Queue for QP's waiting for HW receive array entries */
275 struct tid_queue rarr_queue;
276 /* when waiting for rcv or pioavail */
277 wait_queue_head_t wait;
280 /* same size as task_struct .comm[], command that opened context */
281 char comm[TASK_COMM_LEN];
282 /* Bitmask of in use context(s) */
283 DECLARE_BITMAP(in_use_ctxts, HFI1_MAX_SHARED_CTXTS);
284 /* per-context event flags for fileops/intr communication */
285 unsigned long event_flags;
286 /* A page of memory for rcvhdrhead, rcvegrhead, rcvegrtail * N */
287 void *subctxt_uregbase;
288 /* An array of pages for the eager receive buffers * N */
289 void *subctxt_rcvegrbuf;
290 /* An array of pages for the eager header queue entries * N */
291 void *subctxt_rcvhdr_base;
292 /* total number of polled urgent packets */
294 /* saved total number of polled urgent packets for poll edge trigger */
296 /* Type of packets or conditions we want to poll for */
298 /* non-zero if ctxt is being shared. */
300 /* The version of the library which opened this ctxt */
303 * non-zero if ctxt can be shared, and defines the maximum number of
304 * sub-contexts for this device context.
308 /* Bit mask to track free TID RDMA HW flows */
309 unsigned long flow_mask;
310 struct tid_flow_state flows[RXE_NUM_TID_FLOWS];
314 * rcvhdrq_size - return total size in bytes for header queue
315 * @rcd: the receive context
317 * rcvhdrqentsize is in DWs, so we have to convert to bytes
320 static inline u32 rcvhdrq_size(struct hfi1_ctxtdata *rcd)
322 return PAGE_ALIGN(rcd->rcvhdrq_cnt *
323 rcd->rcvhdrqentsize * sizeof(u32));
327 * Represents a single packet at a high level. Put commonly computed things in
328 * here so we do not have to keep doing them over and over. The rule of thumb is
329 * if something is used one time to derive some value, store that something in
330 * here. If it is used multiple times, then store the result of that derivation
337 struct hfi1_ctxtdata *rcd;
340 struct ib_other_headers *ohdr;
342 struct opa_16b_mgmt *mgmt;
365 #define HFI1_PKT_TYPE_9B 0
366 #define HFI1_PKT_TYPE_16B 1
371 #define OPA_16B_L4_MASK 0xFFull
372 #define OPA_16B_SC_MASK 0x1F00000ull
373 #define OPA_16B_SC_SHIFT 20
374 #define OPA_16B_LID_MASK 0xFFFFFull
375 #define OPA_16B_DLID_MASK 0xF000ull
376 #define OPA_16B_DLID_SHIFT 20
377 #define OPA_16B_DLID_HIGH_SHIFT 12
378 #define OPA_16B_SLID_MASK 0xF00ull
379 #define OPA_16B_SLID_SHIFT 20
380 #define OPA_16B_SLID_HIGH_SHIFT 8
381 #define OPA_16B_BECN_MASK 0x80000000ull
382 #define OPA_16B_BECN_SHIFT 31
383 #define OPA_16B_FECN_MASK 0x10000000ull
384 #define OPA_16B_FECN_SHIFT 28
385 #define OPA_16B_L2_MASK 0x60000000ull
386 #define OPA_16B_L2_SHIFT 29
387 #define OPA_16B_PKEY_MASK 0xFFFF0000ull
388 #define OPA_16B_PKEY_SHIFT 16
389 #define OPA_16B_LEN_MASK 0x7FF00000ull
390 #define OPA_16B_LEN_SHIFT 20
391 #define OPA_16B_RC_MASK 0xE000000ull
392 #define OPA_16B_RC_SHIFT 25
393 #define OPA_16B_AGE_MASK 0xFF0000ull
394 #define OPA_16B_AGE_SHIFT 16
395 #define OPA_16B_ENTROPY_MASK 0xFFFFull
398 * OPA 16B L2/L4 Encodings
400 #define OPA_16B_L4_9B 0x00
401 #define OPA_16B_L2_TYPE 0x02
402 #define OPA_16B_L4_FM 0x08
403 #define OPA_16B_L4_IB_LOCAL 0x09
404 #define OPA_16B_L4_IB_GLOBAL 0x0A
405 #define OPA_16B_L4_ETHR OPA_VNIC_L4_ETHR
410 #define OPA_16B_L4_FM_PAD 3 /* fixed 3B pad */
411 #define OPA_16B_L4_FM_HLEN 24 /* 16B(16) + L4_FM(8) */
413 static inline u8 hfi1_16B_get_l4(struct hfi1_16b_header *hdr)
415 return (u8)(hdr->lrh[2] & OPA_16B_L4_MASK);
418 static inline u8 hfi1_16B_get_sc(struct hfi1_16b_header *hdr)
420 return (u8)((hdr->lrh[1] & OPA_16B_SC_MASK) >> OPA_16B_SC_SHIFT);
423 static inline u32 hfi1_16B_get_dlid(struct hfi1_16b_header *hdr)
425 return (u32)((hdr->lrh[1] & OPA_16B_LID_MASK) |
426 (((hdr->lrh[2] & OPA_16B_DLID_MASK) >>
427 OPA_16B_DLID_HIGH_SHIFT) << OPA_16B_DLID_SHIFT));
430 static inline u32 hfi1_16B_get_slid(struct hfi1_16b_header *hdr)
432 return (u32)((hdr->lrh[0] & OPA_16B_LID_MASK) |
433 (((hdr->lrh[2] & OPA_16B_SLID_MASK) >>
434 OPA_16B_SLID_HIGH_SHIFT) << OPA_16B_SLID_SHIFT));
437 static inline u8 hfi1_16B_get_becn(struct hfi1_16b_header *hdr)
439 return (u8)((hdr->lrh[0] & OPA_16B_BECN_MASK) >> OPA_16B_BECN_SHIFT);
442 static inline u8 hfi1_16B_get_fecn(struct hfi1_16b_header *hdr)
444 return (u8)((hdr->lrh[1] & OPA_16B_FECN_MASK) >> OPA_16B_FECN_SHIFT);
447 static inline u8 hfi1_16B_get_l2(struct hfi1_16b_header *hdr)
449 return (u8)((hdr->lrh[1] & OPA_16B_L2_MASK) >> OPA_16B_L2_SHIFT);
452 static inline u16 hfi1_16B_get_pkey(struct hfi1_16b_header *hdr)
454 return (u16)((hdr->lrh[2] & OPA_16B_PKEY_MASK) >> OPA_16B_PKEY_SHIFT);
457 static inline u8 hfi1_16B_get_rc(struct hfi1_16b_header *hdr)
459 return (u8)((hdr->lrh[1] & OPA_16B_RC_MASK) >> OPA_16B_RC_SHIFT);
462 static inline u8 hfi1_16B_get_age(struct hfi1_16b_header *hdr)
464 return (u8)((hdr->lrh[3] & OPA_16B_AGE_MASK) >> OPA_16B_AGE_SHIFT);
467 static inline u16 hfi1_16B_get_len(struct hfi1_16b_header *hdr)
469 return (u16)((hdr->lrh[0] & OPA_16B_LEN_MASK) >> OPA_16B_LEN_SHIFT);
472 static inline u16 hfi1_16B_get_entropy(struct hfi1_16b_header *hdr)
474 return (u16)(hdr->lrh[3] & OPA_16B_ENTROPY_MASK);
477 #define OPA_16B_MAKE_QW(low_dw, high_dw) (((u64)(high_dw) << 32) | (low_dw))
482 #define OPA_16B_BTH_PAD_MASK 7
483 static inline u8 hfi1_16B_bth_get_pad(struct ib_other_headers *ohdr)
485 return (u8)((be32_to_cpu(ohdr->bth[0]) >> IB_BTH_PAD_SHIFT) &
486 OPA_16B_BTH_PAD_MASK);
492 #define OPA_16B_MGMT_QPN_MASK 0xFFFFFF
493 static inline u32 hfi1_16B_get_dest_qpn(struct opa_16b_mgmt *mgmt)
495 return be32_to_cpu(mgmt->dest_qpn) & OPA_16B_MGMT_QPN_MASK;
498 static inline u32 hfi1_16B_get_src_qpn(struct opa_16b_mgmt *mgmt)
500 return be32_to_cpu(mgmt->src_qpn) & OPA_16B_MGMT_QPN_MASK;
503 static inline void hfi1_16B_set_qpn(struct opa_16b_mgmt *mgmt,
504 u32 dest_qp, u32 src_qp)
506 mgmt->dest_qpn = cpu_to_be32(dest_qp & OPA_16B_MGMT_QPN_MASK);
507 mgmt->src_qpn = cpu_to_be32(src_qp & OPA_16B_MGMT_QPN_MASK);
511 * hfi1_get_rc_ohdr - get extended header
512 * @opah - the opaheader
514 static inline struct ib_other_headers *
515 hfi1_get_rc_ohdr(struct hfi1_opa_header *opah)
517 struct ib_other_headers *ohdr;
518 struct ib_header *hdr = NULL;
519 struct hfi1_16b_header *hdr_16b = NULL;
521 /* Find out where the BTH is */
522 if (opah->hdr_type == HFI1_PKT_TYPE_9B) {
524 if (ib_get_lnh(hdr) == HFI1_LRH_BTH)
527 ohdr = &hdr->u.l.oth;
531 hdr_16b = &opah->opah;
532 l4 = hfi1_16B_get_l4(hdr_16b);
533 if (l4 == OPA_16B_L4_IB_LOCAL)
534 ohdr = &hdr_16b->u.oth;
536 ohdr = &hdr_16b->u.l.oth;
541 struct rvt_sge_state;
544 * Get/Set IB link-level config parameters for f_get/set_ib_cfg()
545 * Mostly for MADs that set or query link parameters, also ipath
548 #define HFI1_IB_CFG_LIDLMC 0 /* LID (LS16b) and Mask (MS16b) */
549 #define HFI1_IB_CFG_LWID_DG_ENB 1 /* allowed Link-width downgrade */
550 #define HFI1_IB_CFG_LWID_ENB 2 /* allowed Link-width */
551 #define HFI1_IB_CFG_LWID 3 /* currently active Link-width */
552 #define HFI1_IB_CFG_SPD_ENB 4 /* allowed Link speeds */
553 #define HFI1_IB_CFG_SPD 5 /* current Link spd */
554 #define HFI1_IB_CFG_RXPOL_ENB 6 /* Auto-RX-polarity enable */
555 #define HFI1_IB_CFG_LREV_ENB 7 /* Auto-Lane-reversal enable */
556 #define HFI1_IB_CFG_LINKLATENCY 8 /* Link Latency (IB1.2 only) */
557 #define HFI1_IB_CFG_HRTBT 9 /* IB heartbeat off/enable/auto; DDR/QDR only */
558 #define HFI1_IB_CFG_OP_VLS 10 /* operational VLs */
559 #define HFI1_IB_CFG_VL_HIGH_CAP 11 /* num of VL high priority weights */
560 #define HFI1_IB_CFG_VL_LOW_CAP 12 /* num of VL low priority weights */
561 #define HFI1_IB_CFG_OVERRUN_THRESH 13 /* IB overrun threshold */
562 #define HFI1_IB_CFG_PHYERR_THRESH 14 /* IB PHY error threshold */
563 #define HFI1_IB_CFG_LINKDEFAULT 15 /* IB link default (sleep/poll) */
564 #define HFI1_IB_CFG_PKEYS 16 /* update partition keys */
565 #define HFI1_IB_CFG_MTU 17 /* update MTU in IBC */
566 #define HFI1_IB_CFG_VL_HIGH_LIMIT 19
567 #define HFI1_IB_CFG_PMA_TICKS 20 /* PMA sample tick resolution */
568 #define HFI1_IB_CFG_PORT 21 /* switch port we are connected to */
571 * HFI or Host Link States
573 * These describe the states the driver thinks the logical and physical
574 * states are in. Used as an argument to set_link_state(). Implemented
575 * as bits for easy multi-state checking. The actual state can only be
578 #define __HLS_UP_INIT_BP 0
579 #define __HLS_UP_ARMED_BP 1
580 #define __HLS_UP_ACTIVE_BP 2
581 #define __HLS_DN_DOWNDEF_BP 3 /* link down default */
582 #define __HLS_DN_POLL_BP 4
583 #define __HLS_DN_DISABLE_BP 5
584 #define __HLS_DN_OFFLINE_BP 6
585 #define __HLS_VERIFY_CAP_BP 7
586 #define __HLS_GOING_UP_BP 8
587 #define __HLS_GOING_OFFLINE_BP 9
588 #define __HLS_LINK_COOLDOWN_BP 10
590 #define HLS_UP_INIT BIT(__HLS_UP_INIT_BP)
591 #define HLS_UP_ARMED BIT(__HLS_UP_ARMED_BP)
592 #define HLS_UP_ACTIVE BIT(__HLS_UP_ACTIVE_BP)
593 #define HLS_DN_DOWNDEF BIT(__HLS_DN_DOWNDEF_BP) /* link down default */
594 #define HLS_DN_POLL BIT(__HLS_DN_POLL_BP)
595 #define HLS_DN_DISABLE BIT(__HLS_DN_DISABLE_BP)
596 #define HLS_DN_OFFLINE BIT(__HLS_DN_OFFLINE_BP)
597 #define HLS_VERIFY_CAP BIT(__HLS_VERIFY_CAP_BP)
598 #define HLS_GOING_UP BIT(__HLS_GOING_UP_BP)
599 #define HLS_GOING_OFFLINE BIT(__HLS_GOING_OFFLINE_BP)
600 #define HLS_LINK_COOLDOWN BIT(__HLS_LINK_COOLDOWN_BP)
602 #define HLS_UP (HLS_UP_INIT | HLS_UP_ARMED | HLS_UP_ACTIVE)
603 #define HLS_DOWN ~(HLS_UP)
605 #define HLS_DEFAULT HLS_DN_POLL
607 /* use this MTU size if none other is given */
608 #define HFI1_DEFAULT_ACTIVE_MTU 10240
609 /* use this MTU size as the default maximum */
610 #define HFI1_DEFAULT_MAX_MTU 10240
611 /* default partition key */
612 #define DEFAULT_PKEY 0xffff
615 * Possible fabric manager config parameters for fm_{get,set}_table()
617 #define FM_TBL_VL_HIGH_ARB 1 /* Get/set VL high prio weights */
618 #define FM_TBL_VL_LOW_ARB 2 /* Get/set VL low prio weights */
619 #define FM_TBL_BUFFER_CONTROL 3 /* Get/set Buffer Control */
620 #define FM_TBL_SC2VLNT 4 /* Get/set SC->VLnt */
621 #define FM_TBL_VL_PREEMPT_ELEMS 5 /* Get (no set) VL preempt elems */
622 #define FM_TBL_VL_PREEMPT_MATRIX 6 /* Get (no set) VL preempt matrix */
625 * Possible "operations" for f_rcvctrl(ppd, op, ctxt)
626 * these are bits so they can be combined, e.g.
627 * HFI1_RCVCTRL_INTRAVAIL_ENB | HFI1_RCVCTRL_CTXT_ENB
629 #define HFI1_RCVCTRL_TAILUPD_ENB 0x01
630 #define HFI1_RCVCTRL_TAILUPD_DIS 0x02
631 #define HFI1_RCVCTRL_CTXT_ENB 0x04
632 #define HFI1_RCVCTRL_CTXT_DIS 0x08
633 #define HFI1_RCVCTRL_INTRAVAIL_ENB 0x10
634 #define HFI1_RCVCTRL_INTRAVAIL_DIS 0x20
635 #define HFI1_RCVCTRL_PKEY_ENB 0x40 /* Note, default is enabled */
636 #define HFI1_RCVCTRL_PKEY_DIS 0x80
637 #define HFI1_RCVCTRL_TIDFLOW_ENB 0x0400
638 #define HFI1_RCVCTRL_TIDFLOW_DIS 0x0800
639 #define HFI1_RCVCTRL_ONE_PKT_EGR_ENB 0x1000
640 #define HFI1_RCVCTRL_ONE_PKT_EGR_DIS 0x2000
641 #define HFI1_RCVCTRL_NO_RHQ_DROP_ENB 0x4000
642 #define HFI1_RCVCTRL_NO_RHQ_DROP_DIS 0x8000
643 #define HFI1_RCVCTRL_NO_EGR_DROP_ENB 0x10000
644 #define HFI1_RCVCTRL_NO_EGR_DROP_DIS 0x20000
645 #define HFI1_RCVCTRL_URGENT_ENB 0x40000
646 #define HFI1_RCVCTRL_URGENT_DIS 0x80000
648 /* partition enforcement flags */
649 #define HFI1_PART_ENFORCE_IN 0x1
650 #define HFI1_PART_ENFORCE_OUT 0x2
652 /* how often we check for synthetic counter wrap around */
653 #define SYNTH_CNT_TIME 3
656 #define CNTR_NORMAL 0x0 /* Normal counters, just read register */
657 #define CNTR_SYNTH 0x1 /* Synthetic counters, saturate at all 1s */
658 #define CNTR_DISABLED 0x2 /* Disable this counter */
659 #define CNTR_32BIT 0x4 /* Simulate 64 bits for this counter */
660 #define CNTR_VL 0x8 /* Per VL counter */
661 #define CNTR_SDMA 0x10
662 #define CNTR_INVALID_VL -1 /* Specifies invalid VL */
663 #define CNTR_MODE_W 0x0
664 #define CNTR_MODE_R 0x1
666 /* VLs Supported/Operational */
667 #define HFI1_MIN_VLS_SUPPORTED 1
668 #define HFI1_MAX_VLS_SUPPORTED 8
670 #define HFI1_GUIDS_PER_PORT 5
671 #define HFI1_PORT_GUID_INDEX 0
673 static inline void incr_cntr64(u64 *cntr)
675 if (*cntr < (u64)-1LL)
679 #define MAX_NAME_SIZE 64
680 struct hfi1_msix_entry {
685 struct irq_affinity_notify notify;
688 struct hfi1_msix_info {
689 /* lock to synchronize in_use_msix access */
690 spinlock_t msix_lock;
691 DECLARE_BITMAP(in_use_msix, CCE_NUM_MSIX_VECTORS);
692 struct hfi1_msix_entry *msix_entries;
696 /* per-SL CCA information */
698 struct hrtimer hrtimer;
699 struct hfi1_pportdata *ppd; /* read-only */
700 int sl; /* read-only */
701 u16 ccti; /* read/write - current value of CCTI */
704 struct link_down_reason {
706 * SMA-facing value. Should be set from .latest when
707 * HLS_UP_* -> HLS_DN_* transition actually occurs.
719 struct vl_arb_cache {
720 /* protect vl arb cache */
722 struct ib_vl_weight_elem table[VL_ARB_TABLE_SIZE];
726 * The structure below encapsulates data relevant to a physical IB Port.
727 * Current chips support only one such port, but the separation
728 * clarifies things a bit. Note that to conform to IB conventions,
729 * port-numbers are one-based. The first or only port is port1.
731 struct hfi1_pportdata {
732 struct hfi1_ibport ibport_data;
734 struct hfi1_devdata *dd;
737 struct qsfp_data qsfp_info;
738 /* Values for SI tuning of SerDes */
748 /* did we read platform config from scratch registers? */
749 bool config_from_scratch;
751 /* GUIDs for this interface, in host order, guids[0] is a port guid */
752 u64 guids[HFI1_GUIDS_PER_PORT];
754 /* GUID for peer interface, in host order */
757 /* up or down physical link state */
761 * this address is mapped read-only into user processes so they can
762 * get status cheaply, whenever they want. One qword of status per port
766 /* SendDMA related entries */
768 struct workqueue_struct *hfi1_wq;
769 struct workqueue_struct *link_wq;
771 /* move out of interrupt context */
772 struct work_struct link_vc_work;
773 struct work_struct link_up_work;
774 struct work_struct link_down_work;
775 struct work_struct sma_message_work;
776 struct work_struct freeze_work;
777 struct work_struct link_downgrade_work;
778 struct work_struct link_bounce_work;
779 struct delayed_work start_link_work;
780 /* host link state variables */
781 struct mutex hls_lock;
784 /* these are the "32 bit" regs */
786 u32 ibmtu; /* The MTU programmed for this unit */
788 * Current max size IB packet (in bytes) including IB headers, that
789 * we can send. Changes when ibmtu changes.
792 u32 current_egress_rate; /* units [10^6 bits/sec] */
793 /* LID programmed for this instance */
795 /* list of pkeys programmed; 0 if not set */
796 u16 pkeys[MAX_PKEY_VALUES];
797 u16 link_width_supported;
798 u16 link_width_downgrade_supported;
799 u16 link_speed_supported;
800 u16 link_width_enabled;
801 u16 link_width_downgrade_enabled;
802 u16 link_speed_enabled;
803 u16 link_width_active;
804 u16 link_width_downgrade_tx_active;
805 u16 link_width_downgrade_rx_active;
806 u16 link_speed_active;
809 u8 actual_vls_operational;
810 /* LID mask control */
812 /* Rx Polarity inversion (compensate for ~tx on partner) */
815 u8 hw_pidx; /* physical port index */
816 u32 port; /* IB port number and index into dd->pports - 1 */
817 /* type of neighbor node */
820 u8 neighbor_fm_security; /* 1 if firmware checking is disabled */
821 u8 neighbor_port_number;
822 u8 is_sm_config_started;
823 u8 offline_disabled_reason;
824 u8 is_active_optimize_enabled;
825 u8 driver_link_ready; /* driver ready for active link */
826 u8 link_enabled; /* link enabled? */
828 u8 local_tx_rate; /* rate given to 8051 firmware */
831 /* placeholders for IB MAD packet settings */
832 u8 overrun_threshold;
833 u8 phy_error_threshold;
834 unsigned int is_link_down_queued;
836 /* Used to override LED behavior for things like maintenance beaconing*/
838 * Alternates per phase of blink
839 * [0] holds LED off duration, [1] holds LED on duration
841 unsigned long led_override_vals[2];
842 u8 led_override_phase; /* LSB picks from vals[] */
843 atomic_t led_override_timer_active;
844 /* Used to flash LEDs in override mode */
845 struct timer_list led_override_timer;
851 * cca_timer_lock protects access to the per-SL cca_timer
852 * structures (specifically the ccti member).
854 spinlock_t cca_timer_lock ____cacheline_aligned_in_smp;
855 struct cca_timer cca_timer[OPA_MAX_SLS];
857 /* List of congestion control table entries */
858 struct ib_cc_table_entry_shadow ccti_entries[CC_TABLE_SHADOW_MAX];
860 /* congestion entries, each entry corresponding to a SL */
861 struct opa_congestion_setting_entry_shadow
862 congestion_entries[OPA_MAX_SLS];
865 * cc_state_lock protects (write) access to the per-port
868 spinlock_t cc_state_lock ____cacheline_aligned_in_smp;
870 struct cc_state __rcu *cc_state;
872 /* Total number of congestion control table entries */
875 /* Bit map identifying service level */
876 u32 cc_sl_control_map;
878 /* CA's max number of 64 entry units in the congestion control table */
879 u8 cc_max_table_entries;
882 * begin congestion log related entries
883 * cc_log_lock protects all congestion log related data
885 spinlock_t cc_log_lock ____cacheline_aligned_in_smp;
886 u8 threshold_cong_event_map[OPA_MAX_SLS / 8];
887 u16 threshold_event_counter;
888 struct opa_hfi1_cong_log_event_internal cc_events[OPA_CONG_LOG_ELEMS];
889 int cc_log_idx; /* index for logging events */
890 int cc_mad_idx; /* index for reporting events */
891 /* end congestion log related entries */
893 struct vl_arb_cache vl_arb_cache[MAX_PRIO_TABLE];
895 /* port relative counter buffer */
897 /* port relative synthetic counter buffer */
899 /* port_xmit_discards are synthesized from different egress errors */
900 u64 port_xmit_discards;
901 u64 port_xmit_discards_vl[C_VL_COUNT];
902 u64 port_xmit_constraint_errors;
903 u64 port_rcv_constraint_errors;
904 /* count of 'link_err' interrupts from DC */
906 /* number of times link retrained successfully */
908 /* number of times a link unknown frame was reported */
909 u64 unknown_frame_count;
910 /* port_ltp_crc_mode is returned in 'portinfo' MADs */
911 u16 port_ltp_crc_mode;
912 /* port_crc_mode_enabled is the crc we support */
913 u8 port_crc_mode_enabled;
914 /* mgmt_allowed is also returned in 'portinfo' MADs */
916 u8 part_enforce; /* partition enforcement flags */
917 struct link_down_reason local_link_down_reason;
918 struct link_down_reason neigh_link_down_reason;
919 /* Value to be sent to link peer on LinkDown .*/
920 u8 remote_link_down_reason;
921 /* Error events that will cause a port bounce. */
922 u32 port_error_action;
923 struct work_struct linkstate_active_work;
924 /* Does this port need to prescan for FECNs */
927 * Sample sendWaitCnt & sendWaitVlCnt during link transition
928 * and counter request.
930 u64 port_vl_xmit_wait_last[C_VL_COUNT + 1];
932 u64 vl_xmit_flit_cnt[C_VL_COUNT + 1];
935 typedef void (*opcode_handler)(struct hfi1_packet *packet);
936 typedef void (*hfi1_make_req)(struct rvt_qp *qp,
937 struct hfi1_pkt_state *ps,
938 struct rvt_swqe *wqe);
939 extern const rhf_rcv_function_ptr normal_rhf_rcv_functions[];
940 extern const rhf_rcv_function_ptr netdev_rhf_rcv_functions[];
942 /* return values for the RHF receive functions */
943 #define RHF_RCV_CONTINUE 0 /* keep going */
944 #define RHF_RCV_DONE 1 /* stop, this packet processed */
945 #define RHF_RCV_REPROCESS 2 /* stop. retain this packet */
947 struct rcv_array_data {
955 struct send_context *sc;
958 /* 16 to directly index */
959 #define PER_VL_SEND_CONTEXTS 16
961 struct err_info_rcvport {
967 struct err_info_constraint {
974 unsigned int curr; /* current temperature */
975 unsigned int lo_lim; /* low temperature limit */
976 unsigned int hi_lim; /* high temperature limit */
977 unsigned int crit_lim; /* critical temperature limit */
978 u8 triggers; /* temperature triggers */
981 struct hfi1_i2c_bus {
982 struct hfi1_devdata *controlling_dd; /* current controlling device */
983 struct i2c_adapter adapter; /* bus details */
984 struct i2c_algo_bit_data algo; /* bus algorithm details */
985 int num; /* bus number, 0 or 1 */
988 /* common data between shared ASIC HFIs */
989 struct hfi1_asic_data {
990 struct hfi1_devdata *dds[2]; /* back pointers */
991 struct mutex asic_resource_mutex;
992 struct hfi1_i2c_bus *i2c_bus0;
993 struct hfi1_i2c_bus *i2c_bus1;
996 /* sizes for both the QP and RSM map tables */
997 #define NUM_MAP_ENTRIES 256
998 #define NUM_MAP_REGS 32
1000 /* Virtual NIC information */
1001 struct hfi1_vnic_data {
1002 struct kmem_cache *txreq_cache;
1006 struct hfi1_vnic_vport_info;
1008 /* device data struct now contains only "general per-device" info.
1009 * fields related to a physical IB port are in a hfi1_pportdata struct.
1014 #define BOARD_VERS_MAX 96 /* how long the version string can be */
1015 #define SERIAL_MAX 16 /* length of the serial number */
1017 typedef int (*send_routine)(struct rvt_qp *, struct hfi1_pkt_state *, u64);
1018 struct hfi1_netdev_rx;
1019 struct hfi1_devdata {
1020 struct hfi1_ibdev verbs_dev; /* must be first */
1021 /* pointers to related structs for this device */
1022 /* pci access data structure */
1023 struct pci_dev *pcidev;
1024 struct cdev user_cdev;
1025 struct cdev diag_cdev;
1026 struct cdev ui_cdev;
1027 struct device *user_device;
1028 struct device *diag_device;
1029 struct device *ui_device;
1031 /* first mapping up to RcvArray */
1032 u8 __iomem *kregbase1;
1033 resource_size_t physaddr;
1035 /* second uncached mapping from RcvArray to pio send buffers */
1036 u8 __iomem *kregbase2;
1037 /* for detecting offset above kregbase2 address */
1040 /* Per VL data. Enough for all VLs but not all elements are set/used. */
1041 struct per_vl_data vld[PER_VL_SEND_CONTEXTS];
1042 /* send context data */
1043 struct send_context_info *send_contexts;
1044 /* map hardware send contexts to software index */
1046 /* spinlock for allocating and releasing send context resources */
1048 /* lock for pio_map */
1049 spinlock_t pio_map_lock;
1050 /* Send Context initialization lock. */
1051 spinlock_t sc_init_lock;
1052 /* lock for sdma_map */
1053 spinlock_t sde_map_lock;
1054 /* array of kernel send contexts */
1055 struct send_context **kernel_send_context;
1056 /* array of vl maps */
1057 struct pio_vl_map __rcu *pio_map;
1058 /* default flags to last descriptor */
1061 /* fields common to all SDMA engines */
1063 volatile __le64 *sdma_heads_dma; /* DMA'ed by chip */
1064 dma_addr_t sdma_heads_phys;
1065 void *sdma_pad_dma; /* DMA'ed by chip */
1066 dma_addr_t sdma_pad_phys;
1067 /* for deallocation */
1068 size_t sdma_heads_size;
1071 /* array of engines sized by num_sdma */
1072 struct sdma_engine *per_sdma;
1073 /* array of vl maps */
1074 struct sdma_vl_map __rcu *sdma_map;
1075 /* SPC freeze waitqueue and variable */
1076 wait_queue_head_t sdma_unfreeze_wq;
1077 atomic_t sdma_unfreeze_count;
1079 u32 lcb_access_count; /* count of LCB users */
1081 /* common data between shared ASIC HFIs in this OS */
1082 struct hfi1_asic_data *asic_data;
1084 /* mem-mapped pointer to base of PIO buffers */
1085 void __iomem *piobase;
1087 * write-combining mem-mapped pointer to base of RcvArray
1090 void __iomem *rcvarray_wc;
1092 * credit return base - a per-NUMA range of DMA address that
1093 * the chip will use to update the per-context free counter
1095 struct credit_return_base *cr_base;
1097 /* send context numbers and sizes for each type */
1098 struct sc_config_sizes sc_sizes[SC_MAX];
1100 char *boardname; /* human readable board info */
1107 u64 z_send_schedule;
1109 u64 __percpu *send_schedule;
1110 /* number of reserved contexts for netdev usage */
1111 u16 num_netdev_contexts;
1112 /* number of receive contexts in use by the driver */
1113 u32 num_rcv_contexts;
1114 /* number of pio send contexts in use by the driver */
1115 u32 num_send_contexts;
1117 * number of ctxts available for PSM open
1120 /* total number of available user/PSM contexts */
1121 u32 num_user_contexts;
1122 /* base receive interrupt timeout, in CSR units */
1123 u32 rcv_intr_timeout_csr;
1125 spinlock_t sendctrl_lock; /* protect changes to SendCtrl */
1126 spinlock_t rcvctrl_lock; /* protect changes to RcvCtrl */
1127 spinlock_t uctxt_lock; /* protect rcd changes */
1128 struct mutex dc8051_lock; /* exclusive access to 8051 */
1129 struct workqueue_struct *update_cntr_wq;
1130 struct work_struct update_cntr_work;
1131 /* exclusive access to 8051 memory */
1132 spinlock_t dc8051_memlock;
1133 int dc8051_timed_out; /* remember if the 8051 timed out */
1135 * A page that will hold event notification bitmaps for all
1136 * contexts. This page will be mapped into all processes.
1138 unsigned long *events;
1140 * per unit status, see also portdata statusp
1141 * mapped read-only into user processes so they can get unit and
1142 * IB link status cheaply
1144 struct hfi1_status *status;
1146 /* revision register shadow */
1148 /* Base GUID for device (network order) */
1151 /* both sides of the PCIe link are gen3 capable */
1152 u8 link_gen3_capable;
1154 /* localbus width (1, 2,4,8,16,32) from config space */
1156 /* localbus speed in MHz */
1158 int unit; /* unit # of this chip */
1159 int node; /* home node of this chip */
1161 /* save these PCI fields to restore after a reset */
1173 * ASCII serial number, from flash, large enough for original
1174 * all digit strings, and longer serial number format
1176 u8 serial[SERIAL_MAX];
1177 /* human readable board version */
1178 u8 boardversion[BOARD_VERS_MAX];
1179 u8 lbus_info[32]; /* human readable localbus info */
1180 /* chip major rev, from CceRevision */
1182 /* chip minor rev, from CceRevision */
1186 /* implementation code */
1188 /* vAU of this device */
1190 /* vCU of this device */
1192 /* link credits of this device */
1194 /* initial vl15 credits to use */
1198 * Cached value for vl15buf, read during verify cap interrupt. VL15
1199 * credits are to be kept at 0 and set when handling the link-up
1200 * interrupt. This removes the possibility of receiving VL15 MAD
1201 * packets before this HFI is ready.
1205 /* Misc small ints */
1209 u16 irev; /* implementation revision */
1210 u32 dc8051_ver; /* 8051 firmware version */
1212 spinlock_t hfi1_diag_trans_lock; /* protect diag observer ops */
1213 struct platform_config platform_config;
1214 struct platform_config_cache pcfg_cache;
1216 struct diag_client *diag_client;
1218 /* general interrupt: mask of handled interrupts */
1219 u64 gi_mask[CCE_NUM_INT_CSRS];
1221 struct rcv_array_data rcv_entries;
1223 /* cycle length of PS* counters in HW (in picoseconds) */
1224 u16 psxmitwait_check_rate;
1227 * 64 bit synthetic counters
1229 struct timer_list synth_stats_timer;
1231 /* MSI-X information */
1232 struct hfi1_msix_info msix_info;
1238 size_t cntrnameslen;
1244 * remembered values for synthetic counters
1253 char *portcntrnames;
1254 size_t portcntrnameslen;
1256 struct err_info_rcvport err_info_rcvport;
1257 struct err_info_constraint err_info_rcv_constraint;
1258 struct err_info_constraint err_info_xmit_constraint;
1260 atomic_t drop_packet;
1262 u8 err_info_uncorrectable;
1263 u8 err_info_fmconfig;
1266 * Software counters for the status bits defined by the
1267 * associated error status registers
1269 u64 cce_err_status_cnt[NUM_CCE_ERR_STATUS_COUNTERS];
1270 u64 rcv_err_status_cnt[NUM_RCV_ERR_STATUS_COUNTERS];
1271 u64 misc_err_status_cnt[NUM_MISC_ERR_STATUS_COUNTERS];
1272 u64 send_pio_err_status_cnt[NUM_SEND_PIO_ERR_STATUS_COUNTERS];
1273 u64 send_dma_err_status_cnt[NUM_SEND_DMA_ERR_STATUS_COUNTERS];
1274 u64 send_egress_err_status_cnt[NUM_SEND_EGRESS_ERR_STATUS_COUNTERS];
1275 u64 send_err_status_cnt[NUM_SEND_ERR_STATUS_COUNTERS];
1277 /* Software counter that spans all contexts */
1278 u64 sw_ctxt_err_status_cnt[NUM_SEND_CTXT_ERR_STATUS_COUNTERS];
1279 /* Software counter that spans all DMA engines */
1280 u64 sw_send_dma_eng_err_status_cnt[
1281 NUM_SEND_DMA_ENG_ERR_STATUS_COUNTERS];
1282 /* Software counter that aggregates all cce_err_status errors */
1283 u64 sw_cce_err_status_aggregate;
1284 /* Software counter that aggregates all bypass packet rcv errors */
1285 u64 sw_rcv_bypass_packet_errors;
1287 /* Save the enabled LCB error bits */
1289 struct cpu_mask_set *comp_vect;
1290 int *comp_vect_mappings;
1291 u32 comp_vect_possible_cpus;
1294 * Capability to have different send engines simply by changing a
1297 send_routine process_pio_send ____cacheline_aligned_in_smp;
1298 send_routine process_dma_send;
1299 void (*pio_inline_send)(struct hfi1_devdata *dd, struct pio_buf *pbuf,
1300 u64 pbc, const void *from, size_t count);
1301 int (*process_vnic_dma_send)(struct hfi1_devdata *dd, u8 q_idx,
1302 struct hfi1_vnic_vport_info *vinfo,
1303 struct sk_buff *skb, u64 pbc, u8 plen);
1304 /* hfi1_pportdata, points to array of (physical) port-specific
1305 * data structs, indexed by pidx (0..n-1)
1307 struct hfi1_pportdata *pport;
1308 /* receive context data */
1309 struct hfi1_ctxtdata **rcd;
1310 u64 __percpu *int_counter;
1311 /* verbs tx opcode stats */
1312 struct hfi1_opcode_stats_perctx __percpu *tx_opstats;
1313 /* device (not port) flags, basically device capabilities */
1315 /* Number of physical ports available */
1317 /* Lowest context number which can be used by user processes or VNIC */
1318 u8 first_dyn_alloc_ctxt;
1319 /* adding a new field here would make it part of this cacheline */
1321 /* seqlock for sc2vl */
1322 seqlock_t sc2vl_lock ____cacheline_aligned_in_smp;
1324 u64 __percpu *rcv_limit;
1325 /* adding a new field here would make it part of this cacheline */
1327 /* OUI comes from the HW. Used everywhere as 3 separate bytes. */
1332 /* Timer and counter used to detect RcvBufOvflCnt changes */
1333 struct timer_list rcverr_timer;
1335 wait_queue_head_t event_queue;
1337 /* receive context tail dummy address */
1338 __le64 *rcvhdrtail_dummy_kvaddr;
1339 dma_addr_t rcvhdrtail_dummy_dma;
1342 /* Serialize ASPM enable/disable between multiple verbs contexts */
1343 spinlock_t aspm_lock;
1344 /* Number of verbs contexts which have disabled ASPM */
1345 atomic_t aspm_disabled_cnt;
1346 /* Keeps track of user space clients */
1347 refcount_t user_refcount;
1348 /* Used to wait for outstanding user space clients before dev removal */
1349 struct completion user_comp;
1351 bool eprom_available; /* true if EPROM is available for this device */
1352 bool aspm_supported; /* Does HW support ASPM */
1353 bool aspm_enabled; /* ASPM state: enabled/disabled */
1354 struct rhashtable *sdma_rht;
1357 struct hfi1_vnic_data vnic;
1358 /* Lock to protect IRQ SRC register access */
1359 spinlock_t irq_src_lock;
1360 int vnic_num_vports;
1361 struct hfi1_netdev_rx *netdev_rx;
1362 struct hfi1_affinity_node *affinity_entry;
1364 /* Keeps track of IPoIB RSM rule users */
1365 atomic_t ipoib_rsm_usr_num;
1368 /* 8051 firmware version helper */
1369 #define dc8051_ver(a, b, c) ((a) << 16 | (b) << 8 | (c))
1370 #define dc8051_ver_maj(a) (((a) & 0xff0000) >> 16)
1371 #define dc8051_ver_min(a) (((a) & 0x00ff00) >> 8)
1372 #define dc8051_ver_patch(a) ((a) & 0x0000ff)
1374 /* f_put_tid types */
1375 #define PT_EXPECTED 0
1377 #define PT_INVALID_FLUSH 2
1378 #define PT_INVALID 3
1382 struct mmu_rb_handler;
1384 /* Private data for file operations */
1385 struct hfi1_filedata {
1386 struct srcu_struct pq_srcu;
1387 struct hfi1_devdata *dd;
1388 struct hfi1_ctxtdata *uctxt;
1389 struct hfi1_user_sdma_comp_q *cq;
1390 /* update side lock for SRCU */
1391 spinlock_t pq_rcu_lock;
1392 struct hfi1_user_sdma_pkt_q __rcu *pq;
1394 /* for cpu affinity; -1 if none */
1398 struct tid_rb_node **entry_to_rb;
1399 spinlock_t tid_lock; /* protect tid_[limit,used] counters */
1403 u32 invalid_tid_idx;
1404 /* protect invalid_tids array and invalid_tid_idx */
1405 spinlock_t invalid_lock;
1408 extern struct xarray hfi1_dev_table;
1409 struct hfi1_devdata *hfi1_lookup(int unit);
1411 static inline unsigned long uctxt_offset(struct hfi1_ctxtdata *uctxt)
1413 return (uctxt->ctxt - uctxt->dd->first_dyn_alloc_ctxt) *
1414 HFI1_MAX_SHARED_CTXTS;
1417 int hfi1_init(struct hfi1_devdata *dd, int reinit);
1418 int hfi1_count_active_units(void);
1420 int hfi1_diag_add(struct hfi1_devdata *dd);
1421 void hfi1_diag_remove(struct hfi1_devdata *dd);
1422 void handle_linkup_change(struct hfi1_devdata *dd, u32 linkup);
1424 void handle_user_interrupt(struct hfi1_ctxtdata *rcd);
1426 int hfi1_create_rcvhdrq(struct hfi1_devdata *dd, struct hfi1_ctxtdata *rcd);
1427 int hfi1_setup_eagerbufs(struct hfi1_ctxtdata *rcd);
1428 int hfi1_create_kctxts(struct hfi1_devdata *dd);
1429 int hfi1_create_ctxtdata(struct hfi1_pportdata *ppd, int numa,
1430 struct hfi1_ctxtdata **rcd);
1431 void hfi1_free_ctxt(struct hfi1_ctxtdata *rcd);
1432 void hfi1_init_pportdata(struct pci_dev *pdev, struct hfi1_pportdata *ppd,
1433 struct hfi1_devdata *dd, u8 hw_pidx, u32 port);
1434 void hfi1_free_ctxtdata(struct hfi1_devdata *dd, struct hfi1_ctxtdata *rcd);
1435 int hfi1_rcd_put(struct hfi1_ctxtdata *rcd);
1436 int hfi1_rcd_get(struct hfi1_ctxtdata *rcd);
1437 struct hfi1_ctxtdata *hfi1_rcd_get_by_index_safe(struct hfi1_devdata *dd,
1439 struct hfi1_ctxtdata *hfi1_rcd_get_by_index(struct hfi1_devdata *dd, u16 ctxt);
1440 int handle_receive_interrupt(struct hfi1_ctxtdata *rcd, int thread);
1441 int handle_receive_interrupt_nodma_rtail(struct hfi1_ctxtdata *rcd, int thread);
1442 int handle_receive_interrupt_dma_rtail(struct hfi1_ctxtdata *rcd, int thread);
1443 int handle_receive_interrupt_napi_fp(struct hfi1_ctxtdata *rcd, int budget);
1444 int handle_receive_interrupt_napi_sp(struct hfi1_ctxtdata *rcd, int budget);
1445 void set_all_slowpath(struct hfi1_devdata *dd);
1447 extern const struct pci_device_id hfi1_pci_tbl[];
1448 void hfi1_make_ud_req_9B(struct rvt_qp *qp,
1449 struct hfi1_pkt_state *ps,
1450 struct rvt_swqe *wqe);
1452 void hfi1_make_ud_req_16B(struct rvt_qp *qp,
1453 struct hfi1_pkt_state *ps,
1454 struct rvt_swqe *wqe);
1456 /* receive packet handler dispositions */
1457 #define RCV_PKT_OK 0x0 /* keep going */
1458 #define RCV_PKT_LIMIT 0x1 /* stop, hit limit, start thread */
1459 #define RCV_PKT_DONE 0x2 /* stop, no more packets detected */
1462 * hfi1_rcd_head - add accessor for rcd head
1465 static inline u32 hfi1_rcd_head(struct hfi1_ctxtdata *rcd)
1471 * hfi1_set_rcd_head - add accessor for rcd head
1473 * @head: the new head
1475 static inline void hfi1_set_rcd_head(struct hfi1_ctxtdata *rcd, u32 head)
1480 /* calculate the current RHF address */
1481 static inline __le32 *get_rhf_addr(struct hfi1_ctxtdata *rcd)
1483 return (__le32 *)rcd->rcvhdrq + rcd->head + rcd->rhf_offset;
1486 /* return DMA_RTAIL configuration */
1487 static inline bool get_dma_rtail_setting(struct hfi1_ctxtdata *rcd)
1489 return !!HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL);
1493 * hfi1_seq_incr_wrap - wrapping increment for sequence
1494 * @seq: the current sequence number
1496 * Returns: the incremented seq
1498 static inline u8 hfi1_seq_incr_wrap(u8 seq)
1500 if (++seq > RHF_MAX_SEQ)
1506 * hfi1_seq_cnt - return seq_cnt member
1507 * @rcd: the receive context
1509 * Return seq_cnt member
1511 static inline u8 hfi1_seq_cnt(struct hfi1_ctxtdata *rcd)
1513 return rcd->seq_cnt;
1517 * hfi1_set_seq_cnt - return seq_cnt member
1518 * @rcd: the receive context
1520 * Return seq_cnt member
1522 static inline void hfi1_set_seq_cnt(struct hfi1_ctxtdata *rcd, u8 cnt)
1528 * last_rcv_seq - is last
1529 * @rcd: the receive context
1532 * return true if last packet
1534 static inline bool last_rcv_seq(struct hfi1_ctxtdata *rcd, u32 seq)
1536 return seq != rcd->seq_cnt;
1540 * rcd_seq_incr - increment context sequence number
1541 * @rcd: the receive context
1542 * @seq: the current sequence number
1544 * Returns: true if the this was the last packet
1546 static inline bool hfi1_seq_incr(struct hfi1_ctxtdata *rcd, u32 seq)
1548 rcd->seq_cnt = hfi1_seq_incr_wrap(rcd->seq_cnt);
1549 return last_rcv_seq(rcd, seq);
1553 * get_hdrqentsize - return hdrq entry size
1554 * @rcd: the receive context
1556 static inline u8 get_hdrqentsize(struct hfi1_ctxtdata *rcd)
1558 return rcd->rcvhdrqentsize;
1562 * get_hdrq_cnt - return hdrq count
1563 * @rcd: the receive context
1565 static inline u16 get_hdrq_cnt(struct hfi1_ctxtdata *rcd)
1567 return rcd->rcvhdrq_cnt;
1571 * hfi1_is_slowpath - check if this context is slow path
1572 * @rcd: the receive context
1574 static inline bool hfi1_is_slowpath(struct hfi1_ctxtdata *rcd)
1576 return rcd->do_interrupt == rcd->slow_handler;
1580 * hfi1_is_fastpath - check if this context is fast path
1581 * @rcd: the receive context
1583 static inline bool hfi1_is_fastpath(struct hfi1_ctxtdata *rcd)
1585 if (rcd->ctxt == HFI1_CTRL_CTXT)
1588 return rcd->do_interrupt == rcd->fast_handler;
1592 * hfi1_set_fast - change to the fast handler
1593 * @rcd: the receive context
1595 static inline void hfi1_set_fast(struct hfi1_ctxtdata *rcd)
1599 if (unlikely(!hfi1_is_fastpath(rcd)))
1600 rcd->do_interrupt = rcd->fast_handler;
1603 int hfi1_reset_device(int);
1605 void receive_interrupt_work(struct work_struct *work);
1607 /* extract service channel from header and rhf */
1608 static inline int hfi1_9B_get_sc5(struct ib_header *hdr, u64 rhf)
1610 return ib_get_sc(hdr) | ((!!(rhf_dc_info(rhf))) << 4);
1613 #define HFI1_JKEY_WIDTH 16
1614 #define HFI1_JKEY_MASK (BIT(16) - 1)
1615 #define HFI1_ADMIN_JKEY_RANGE 32
1618 * J_KEYs are split and allocated in the following groups:
1619 * 0 - 31 - users with administrator privileges
1620 * 32 - 63 - kernel protocols using KDETH packets
1621 * 64 - 65535 - all other users using KDETH packets
1623 static inline u16 generate_jkey(kuid_t uid)
1625 u16 jkey = from_kuid(current_user_ns(), uid) & HFI1_JKEY_MASK;
1627 if (capable(CAP_SYS_ADMIN))
1628 jkey &= HFI1_ADMIN_JKEY_RANGE - 1;
1630 jkey |= BIT(HFI1_JKEY_WIDTH - 1);
1636 * active_egress_rate
1638 * returns the active egress rate in units of [10^6 bits/sec]
1640 static inline u32 active_egress_rate(struct hfi1_pportdata *ppd)
1642 u16 link_speed = ppd->link_speed_active;
1643 u16 link_width = ppd->link_width_active;
1646 if (link_speed == OPA_LINK_SPEED_25G)
1647 egress_rate = 25000;
1648 else /* assume OPA_LINK_SPEED_12_5G */
1649 egress_rate = 12500;
1651 switch (link_width) {
1652 case OPA_LINK_WIDTH_4X:
1655 case OPA_LINK_WIDTH_3X:
1658 case OPA_LINK_WIDTH_2X:
1662 /* assume IB_WIDTH_1X */
1672 * Returns the number of 'fabric clock cycles' to egress a packet
1673 * of length 'len' bytes, at 'rate' Mbit/s. Since the fabric clock
1674 * rate is (approximately) 805 MHz, the units of the returned value
1677 static inline u32 egress_cycles(u32 len, u32 rate)
1684 * (length) [bits] / (rate) [bits/sec]
1685 * ---------------------------------------------------
1686 * fabric_clock_period == 1 /(805 * 10^6) [cycles/sec]
1689 cycles = len * 8; /* bits */
1696 void set_link_ipg(struct hfi1_pportdata *ppd);
1697 void process_becn(struct hfi1_pportdata *ppd, u8 sl, u32 rlid, u32 lqpn,
1698 u32 rqpn, u8 svc_type);
1699 void return_cnp(struct hfi1_ibport *ibp, struct rvt_qp *qp, u32 remote_qpn,
1700 u16 pkey, u32 slid, u32 dlid, u8 sc5,
1701 const struct ib_grh *old_grh);
1702 void return_cnp_16B(struct hfi1_ibport *ibp, struct rvt_qp *qp,
1703 u32 remote_qpn, u16 pkey, u32 slid, u32 dlid,
1704 u8 sc5, const struct ib_grh *old_grh);
1705 typedef void (*hfi1_handle_cnp)(struct hfi1_ibport *ibp, struct rvt_qp *qp,
1706 u32 remote_qpn, u16 pkey, u32 slid, u32 dlid,
1707 u8 sc5, const struct ib_grh *old_grh);
1709 #define PKEY_CHECK_INVALID -1
1710 int egress_pkey_check(struct hfi1_pportdata *ppd, u32 slid, u16 pkey,
1711 u8 sc5, int8_t s_pkey_index);
1713 #define PACKET_EGRESS_TIMEOUT 350
1714 static inline void pause_for_credit_return(struct hfi1_devdata *dd)
1716 /* Pause at least 1us, to ensure chip returns all credits */
1717 u32 usec = cclock_to_ns(dd, PACKET_EGRESS_TIMEOUT) / 1000;
1719 udelay(usec ? usec : 1);
1723 * sc_to_vlt() - reverse lookup sc to vl
1727 static inline u8 sc_to_vlt(struct hfi1_devdata *dd, u8 sc5)
1732 if (sc5 >= OPA_MAX_SCS)
1736 seq = read_seqbegin(&dd->sc2vl_lock);
1737 rval = *(((u8 *)dd->sc2vl) + sc5);
1738 } while (read_seqretry(&dd->sc2vl_lock, seq));
1743 #define PKEY_MEMBER_MASK 0x8000
1744 #define PKEY_LOW_15_MASK 0x7fff
1747 * ingress_pkey_matches_entry - return 1 if the pkey matches ent (ent
1748 * being an entry from the ingress partition key table), return 0
1749 * otherwise. Use the matching criteria for ingress partition keys
1750 * specified in the OPAv1 spec., section 9.10.14.
1752 static inline int ingress_pkey_matches_entry(u16 pkey, u16 ent)
1754 u16 mkey = pkey & PKEY_LOW_15_MASK;
1755 u16 ment = ent & PKEY_LOW_15_MASK;
1759 * If pkey[15] is clear (limited partition member),
1760 * is bit 15 in the corresponding table element
1761 * clear (limited member)?
1763 if (!(pkey & PKEY_MEMBER_MASK))
1764 return !!(ent & PKEY_MEMBER_MASK);
1771 * ingress_pkey_table_search - search the entire pkey table for
1772 * an entry which matches 'pkey'. return 0 if a match is found,
1775 static int ingress_pkey_table_search(struct hfi1_pportdata *ppd, u16 pkey)
1779 for (i = 0; i < MAX_PKEY_VALUES; i++) {
1780 if (ingress_pkey_matches_entry(pkey, ppd->pkeys[i]))
1787 * ingress_pkey_table_fail - record a failure of ingress pkey validation,
1788 * i.e., increment port_rcv_constraint_errors for the port, and record
1789 * the 'error info' for this failure.
1791 static void ingress_pkey_table_fail(struct hfi1_pportdata *ppd, u16 pkey,
1794 struct hfi1_devdata *dd = ppd->dd;
1796 incr_cntr64(&ppd->port_rcv_constraint_errors);
1797 if (!(dd->err_info_rcv_constraint.status & OPA_EI_STATUS_SMASK)) {
1798 dd->err_info_rcv_constraint.status |= OPA_EI_STATUS_SMASK;
1799 dd->err_info_rcv_constraint.slid = slid;
1800 dd->err_info_rcv_constraint.pkey = pkey;
1805 * ingress_pkey_check - Return 0 if the ingress pkey is valid, return 1
1806 * otherwise. Use the criteria in the OPAv1 spec, section 9.10.14. idx
1807 * is a hint as to the best place in the partition key table to begin
1808 * searching. This function should not be called on the data path because
1809 * of performance reasons. On datapath pkey check is expected to be done
1810 * by HW and rcv_pkey_check function should be called instead.
1812 static inline int ingress_pkey_check(struct hfi1_pportdata *ppd, u16 pkey,
1813 u8 sc5, u8 idx, u32 slid, bool force)
1815 if (!(force) && !(ppd->part_enforce & HFI1_PART_ENFORCE_IN))
1818 /* If SC15, pkey[0:14] must be 0x7fff */
1819 if ((sc5 == 0xf) && ((pkey & PKEY_LOW_15_MASK) != PKEY_LOW_15_MASK))
1822 /* Is the pkey = 0x0, or 0x8000? */
1823 if ((pkey & PKEY_LOW_15_MASK) == 0)
1826 /* The most likely matching pkey has index 'idx' */
1827 if (ingress_pkey_matches_entry(pkey, ppd->pkeys[idx]))
1830 /* no match - try the whole table */
1831 if (!ingress_pkey_table_search(ppd, pkey))
1835 ingress_pkey_table_fail(ppd, pkey, slid);
1840 * rcv_pkey_check - Return 0 if the ingress pkey is valid, return 1
1841 * otherwise. It only ensures pkey is vlid for QP0. This function
1842 * should be called on the data path instead of ingress_pkey_check
1843 * as on data path, pkey check is done by HW (except for QP0).
1845 static inline int rcv_pkey_check(struct hfi1_pportdata *ppd, u16 pkey,
1848 if (!(ppd->part_enforce & HFI1_PART_ENFORCE_IN))
1851 /* If SC15, pkey[0:14] must be 0x7fff */
1852 if ((sc5 == 0xf) && ((pkey & PKEY_LOW_15_MASK) != PKEY_LOW_15_MASK))
1857 ingress_pkey_table_fail(ppd, pkey, slid);
1863 /* MTU enumeration, 256-4k match IB */
1865 #define OPA_MTU_256 1
1866 #define OPA_MTU_512 2
1867 #define OPA_MTU_1024 3
1868 #define OPA_MTU_2048 4
1869 #define OPA_MTU_4096 5
1871 u32 lrh_max_header_bytes(struct hfi1_devdata *dd);
1872 int mtu_to_enum(u32 mtu, int default_if_bad);
1873 u16 enum_to_mtu(int mtu);
1874 static inline int valid_ib_mtu(unsigned int mtu)
1876 return mtu == 256 || mtu == 512 ||
1877 mtu == 1024 || mtu == 2048 ||
1881 static inline int valid_opa_max_mtu(unsigned int mtu)
1883 return mtu >= 2048 &&
1884 (valid_ib_mtu(mtu) || mtu == 8192 || mtu == 10240);
1887 int set_mtu(struct hfi1_pportdata *ppd);
1889 int hfi1_set_lid(struct hfi1_pportdata *ppd, u32 lid, u8 lmc);
1890 void hfi1_disable_after_error(struct hfi1_devdata *dd);
1891 int hfi1_set_uevent_bits(struct hfi1_pportdata *ppd, const int evtbit);
1892 int hfi1_rcvbuf_validate(u32 size, u8 type, u16 *encode);
1894 int fm_get_table(struct hfi1_pportdata *ppd, int which, void *t);
1895 int fm_set_table(struct hfi1_pportdata *ppd, int which, void *t);
1897 void set_up_vau(struct hfi1_devdata *dd, u8 vau);
1898 void set_up_vl15(struct hfi1_devdata *dd, u16 vl15buf);
1899 void reset_link_credits(struct hfi1_devdata *dd);
1900 void assign_remote_cm_au_table(struct hfi1_devdata *dd, u8 vcu);
1902 int set_buffer_control(struct hfi1_pportdata *ppd, struct buffer_control *bc);
1904 static inline struct hfi1_devdata *dd_from_ppd(struct hfi1_pportdata *ppd)
1909 static inline struct hfi1_devdata *dd_from_dev(struct hfi1_ibdev *dev)
1911 return container_of(dev, struct hfi1_devdata, verbs_dev);
1914 static inline struct hfi1_devdata *dd_from_ibdev(struct ib_device *ibdev)
1916 return dd_from_dev(to_idev(ibdev));
1919 static inline struct hfi1_pportdata *ppd_from_ibp(struct hfi1_ibport *ibp)
1921 return container_of(ibp, struct hfi1_pportdata, ibport_data);
1924 static inline struct hfi1_ibdev *dev_from_rdi(struct rvt_dev_info *rdi)
1926 return container_of(rdi, struct hfi1_ibdev, rdi);
1929 static inline struct hfi1_ibport *to_iport(struct ib_device *ibdev, u32 port)
1931 struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
1932 u32 pidx = port - 1; /* IB number port from 1, hdw from 0 */
1934 WARN_ON(pidx >= dd->num_pports);
1935 return &dd->pport[pidx].ibport_data;
1938 static inline struct hfi1_ibport *rcd_to_iport(struct hfi1_ctxtdata *rcd)
1940 return &rcd->ppd->ibport_data;
1944 * hfi1_may_ecn - Check whether FECN or BECN processing should be done
1945 * @pkt: the packet to be evaluated
1947 * Check whether the FECN or BECN bits in the packet's header are
1948 * enabled, depending on packet type.
1950 * This function only checks for FECN and BECN bits. Additional checks
1951 * are done in the slowpath (hfi1_process_ecn_slowpath()) in order to
1952 * ensure correct handling.
1954 static inline bool hfi1_may_ecn(struct hfi1_packet *pkt)
1958 if (pkt->etype == RHF_RCV_TYPE_BYPASS) {
1959 fecn = hfi1_16B_get_fecn(pkt->hdr);
1960 becn = hfi1_16B_get_becn(pkt->hdr);
1962 fecn = ib_bth_get_fecn(pkt->ohdr);
1963 becn = ib_bth_get_becn(pkt->ohdr);
1965 return fecn || becn;
1968 bool hfi1_process_ecn_slowpath(struct rvt_qp *qp, struct hfi1_packet *pkt,
1970 static inline bool process_ecn(struct rvt_qp *qp, struct hfi1_packet *pkt)
1974 do_work = hfi1_may_ecn(pkt);
1975 if (unlikely(do_work))
1976 return hfi1_process_ecn_slowpath(qp, pkt, false);
1981 * Return the indexed PKEY from the port PKEY table.
1983 static inline u16 hfi1_get_pkey(struct hfi1_ibport *ibp, unsigned index)
1985 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
1988 if (index >= ARRAY_SIZE(ppd->pkeys))
1991 ret = ppd->pkeys[index];
1997 * Return the indexed GUID from the port GUIDs table.
1999 static inline __be64 get_sguid(struct hfi1_ibport *ibp, unsigned int index)
2001 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
2003 WARN_ON(index >= HFI1_GUIDS_PER_PORT);
2004 return cpu_to_be64(ppd->guids[index]);
2008 * Called by readers of cc_state only, must call under rcu_read_lock().
2010 static inline struct cc_state *get_cc_state(struct hfi1_pportdata *ppd)
2012 return rcu_dereference(ppd->cc_state);
2016 * Called by writers of cc_state only, must call under cc_state_lock.
2019 struct cc_state *get_cc_state_protected(struct hfi1_pportdata *ppd)
2021 return rcu_dereference_protected(ppd->cc_state,
2022 lockdep_is_held(&ppd->cc_state_lock));
2026 * values for dd->flags (_device_ related flags)
2028 #define HFI1_INITTED 0x1 /* chip and driver up and initted */
2029 #define HFI1_PRESENT 0x2 /* chip accesses can be done */
2030 #define HFI1_FROZEN 0x4 /* chip in SPC freeze */
2031 #define HFI1_HAS_SDMA_TIMEOUT 0x8
2032 #define HFI1_HAS_SEND_DMA 0x10 /* Supports Send DMA */
2033 #define HFI1_FORCED_FREEZE 0x80 /* driver forced freeze mode */
2034 #define HFI1_SHUTDOWN 0x100 /* device is shutting down */
2036 /* IB dword length mask in PBC (lower 11 bits); same for all chips */
2037 #define HFI1_PBC_LENGTH_MASK ((1 << 11) - 1)
2039 /* ctxt_flag bit offsets */
2040 /* base context has not finished initializing */
2041 #define HFI1_CTXT_BASE_UNINIT 1
2042 /* base context initaliation failed */
2043 #define HFI1_CTXT_BASE_FAILED 2
2044 /* waiting for a packet to arrive */
2045 #define HFI1_CTXT_WAITING_RCV 3
2046 /* waiting for an urgent packet to arrive */
2047 #define HFI1_CTXT_WAITING_URG 4
2049 /* free up any allocated data at closes */
2050 int hfi1_init_dd(struct hfi1_devdata *dd);
2051 void hfi1_free_devdata(struct hfi1_devdata *dd);
2053 /* LED beaconing functions */
2054 void hfi1_start_led_override(struct hfi1_pportdata *ppd, unsigned int timeon,
2055 unsigned int timeoff);
2056 void shutdown_led_override(struct hfi1_pportdata *ppd);
2058 #define HFI1_CREDIT_RETURN_RATE (100)
2061 * The number of words for the KDETH protocol field. If this is
2062 * larger then the actual field used, then part of the payload
2063 * will be in the header.
2065 * Optimally, we want this sized so that a typical case will
2066 * use full cache lines. The typical local KDETH header would
2077 * For a 64-byte cache line, KDETH would need to be 36 bytes or 9 DWORDS
2079 #define DEFAULT_RCVHDRSIZE 9
2082 * Maximal header byte count:
2093 * We also want to maintain a cache line alignment to assist DMA'ing
2094 * of the header bytes. Round up to a good size.
2096 #define DEFAULT_RCVHDR_ENTSIZE 32
2098 bool hfi1_can_pin_pages(struct hfi1_devdata *dd, struct mm_struct *mm,
2099 u32 nlocked, u32 npages);
2100 int hfi1_acquire_user_pages(struct mm_struct *mm, unsigned long vaddr,
2101 size_t npages, bool writable, struct page **pages);
2102 void hfi1_release_user_pages(struct mm_struct *mm, struct page **p,
2103 size_t npages, bool dirty);
2106 * hfi1_rcvhdrtail_kvaddr - return tail kvaddr
2107 * @rcd - the receive context
2109 static inline __le64 *hfi1_rcvhdrtail_kvaddr(const struct hfi1_ctxtdata *rcd)
2111 return (__le64 *)rcd->rcvhdrtail_kvaddr;
2114 static inline void clear_rcvhdrtail(const struct hfi1_ctxtdata *rcd)
2116 u64 *kv = (u64 *)hfi1_rcvhdrtail_kvaddr(rcd);
2122 static inline u32 get_rcvhdrtail(const struct hfi1_ctxtdata *rcd)
2125 * volatile because it's a DMA target from the chip, routine is
2126 * inlined, and don't want register caching or reordering.
2128 return (u32)le64_to_cpu(*hfi1_rcvhdrtail_kvaddr(rcd));
2131 static inline bool hfi1_packet_present(struct hfi1_ctxtdata *rcd)
2133 if (likely(!rcd->rcvhdrtail_kvaddr)) {
2134 u32 seq = rhf_rcv_seq(rhf_to_cpu(get_rhf_addr(rcd)));
2136 return !last_rcv_seq(rcd, seq);
2138 return hfi1_rcd_head(rcd) != get_rcvhdrtail(rcd);
2145 extern const char ib_hfi1_version[];
2146 extern const struct attribute_group ib_hfi1_attr_group;
2147 extern const struct attribute_group *hfi1_attr_port_groups[];
2149 int hfi1_device_create(struct hfi1_devdata *dd);
2150 void hfi1_device_remove(struct hfi1_devdata *dd);
2152 int hfi1_verbs_register_sysfs(struct hfi1_devdata *dd);
2153 void hfi1_verbs_unregister_sysfs(struct hfi1_devdata *dd);
2154 /* Hook for sysfs read of QSFP */
2155 int qsfp_dump(struct hfi1_pportdata *ppd, char *buf, int len);
2157 int hfi1_pcie_init(struct hfi1_devdata *dd);
2158 void hfi1_pcie_cleanup(struct pci_dev *pdev);
2159 int hfi1_pcie_ddinit(struct hfi1_devdata *dd, struct pci_dev *pdev);
2160 void hfi1_pcie_ddcleanup(struct hfi1_devdata *);
2161 int pcie_speeds(struct hfi1_devdata *dd);
2162 int restore_pci_variables(struct hfi1_devdata *dd);
2163 int save_pci_variables(struct hfi1_devdata *dd);
2164 int do_pcie_gen3_transition(struct hfi1_devdata *dd);
2165 void tune_pcie_caps(struct hfi1_devdata *dd);
2166 int parse_platform_config(struct hfi1_devdata *dd);
2167 int get_platform_config_field(struct hfi1_devdata *dd,
2168 enum platform_config_table_type_encoding
2169 table_type, int table_index, int field_index,
2170 u32 *data, u32 len);
2172 struct pci_dev *get_pci_dev(struct rvt_dev_info *rdi);
2175 * Flush write combining store buffers (if present) and perform a write
2178 static inline void flush_wc(void)
2180 asm volatile("sfence" : : : "memory");
2183 void handle_eflags(struct hfi1_packet *packet);
2184 void seqfile_dump_rcd(struct seq_file *s, struct hfi1_ctxtdata *rcd);
2186 /* global module parameter variables */
2187 extern unsigned int hfi1_max_mtu;
2188 extern unsigned int hfi1_cu;
2189 extern unsigned int user_credit_return_threshold;
2190 extern int num_user_contexts;
2191 extern unsigned long n_krcvqs;
2192 extern uint krcvqs[];
2193 extern int krcvqsset;
2194 extern uint loopback;
2195 extern uint quick_linkup;
2196 extern uint rcv_intr_timeout;
2197 extern uint rcv_intr_count;
2198 extern uint rcv_intr_dynamic;
2199 extern ushort link_crc_mask;
2201 extern struct mutex hfi1_mutex;
2203 /* Number of seconds before our card status check... */
2204 #define STATUS_TIMEOUT 60
2206 #define DRIVER_NAME "hfi1"
2207 #define HFI1_USER_MINOR_BASE 0
2208 #define HFI1_TRACE_MINOR 127
2209 #define HFI1_NMINORS 255
2211 #define PCI_VENDOR_ID_INTEL 0x8086
2212 #define PCI_DEVICE_ID_INTEL0 0x24f0
2213 #define PCI_DEVICE_ID_INTEL1 0x24f1
2215 #define HFI1_PKT_USER_SC_INTEGRITY \
2216 (SEND_CTXT_CHECK_ENABLE_DISALLOW_NON_KDETH_PACKETS_SMASK \
2217 | SEND_CTXT_CHECK_ENABLE_DISALLOW_KDETH_PACKETS_SMASK \
2218 | SEND_CTXT_CHECK_ENABLE_DISALLOW_BYPASS_SMASK \
2219 | SEND_CTXT_CHECK_ENABLE_DISALLOW_GRH_SMASK)
2221 #define HFI1_PKT_KERNEL_SC_INTEGRITY \
2222 (SEND_CTXT_CHECK_ENABLE_DISALLOW_KDETH_PACKETS_SMASK)
2224 static inline u64 hfi1_pkt_default_send_ctxt_mask(struct hfi1_devdata *dd,
2227 u64 base_sc_integrity;
2229 /* No integrity checks if HFI1_CAP_NO_INTEGRITY is set */
2230 if (HFI1_CAP_IS_KSET(NO_INTEGRITY))
2234 SEND_CTXT_CHECK_ENABLE_DISALLOW_BYPASS_BAD_PKT_LEN_SMASK
2235 | SEND_CTXT_CHECK_ENABLE_DISALLOW_PBC_STATIC_RATE_CONTROL_SMASK
2236 | SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_LONG_BYPASS_PACKETS_SMASK
2237 | SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_LONG_IB_PACKETS_SMASK
2238 | SEND_CTXT_CHECK_ENABLE_DISALLOW_BAD_PKT_LEN_SMASK
2239 #ifndef CONFIG_FAULT_INJECTION
2240 | SEND_CTXT_CHECK_ENABLE_DISALLOW_PBC_TEST_SMASK
2242 | SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_SMALL_BYPASS_PACKETS_SMASK
2243 | SEND_CTXT_CHECK_ENABLE_DISALLOW_TOO_SMALL_IB_PACKETS_SMASK
2244 | SEND_CTXT_CHECK_ENABLE_DISALLOW_RAW_IPV6_SMASK
2245 | SEND_CTXT_CHECK_ENABLE_DISALLOW_RAW_SMASK
2246 | SEND_CTXT_CHECK_ENABLE_CHECK_BYPASS_VL_MAPPING_SMASK
2247 | SEND_CTXT_CHECK_ENABLE_CHECK_VL_MAPPING_SMASK
2248 | SEND_CTXT_CHECK_ENABLE_CHECK_OPCODE_SMASK
2249 | SEND_CTXT_CHECK_ENABLE_CHECK_SLID_SMASK
2250 | SEND_CTXT_CHECK_ENABLE_CHECK_VL_SMASK
2251 | SEND_CTXT_CHECK_ENABLE_CHECK_ENABLE_SMASK;
2253 if (ctxt_type == SC_USER)
2254 base_sc_integrity |=
2255 #ifndef CONFIG_FAULT_INJECTION
2256 SEND_CTXT_CHECK_ENABLE_DISALLOW_PBC_TEST_SMASK |
2258 HFI1_PKT_USER_SC_INTEGRITY;
2259 else if (ctxt_type != SC_KERNEL)
2260 base_sc_integrity |= HFI1_PKT_KERNEL_SC_INTEGRITY;
2262 /* turn on send-side job key checks if !A0 */
2264 base_sc_integrity |= SEND_CTXT_CHECK_ENABLE_CHECK_JOB_KEY_SMASK;
2266 return base_sc_integrity;
2269 static inline u64 hfi1_pkt_base_sdma_integrity(struct hfi1_devdata *dd)
2271 u64 base_sdma_integrity;
2273 /* No integrity checks if HFI1_CAP_NO_INTEGRITY is set */
2274 if (HFI1_CAP_IS_KSET(NO_INTEGRITY))
2277 base_sdma_integrity =
2278 SEND_DMA_CHECK_ENABLE_DISALLOW_BYPASS_BAD_PKT_LEN_SMASK
2279 | SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_LONG_BYPASS_PACKETS_SMASK
2280 | SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_LONG_IB_PACKETS_SMASK
2281 | SEND_DMA_CHECK_ENABLE_DISALLOW_BAD_PKT_LEN_SMASK
2282 | SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_SMALL_BYPASS_PACKETS_SMASK
2283 | SEND_DMA_CHECK_ENABLE_DISALLOW_TOO_SMALL_IB_PACKETS_SMASK
2284 | SEND_DMA_CHECK_ENABLE_DISALLOW_RAW_IPV6_SMASK
2285 | SEND_DMA_CHECK_ENABLE_DISALLOW_RAW_SMASK
2286 | SEND_DMA_CHECK_ENABLE_CHECK_BYPASS_VL_MAPPING_SMASK
2287 | SEND_DMA_CHECK_ENABLE_CHECK_VL_MAPPING_SMASK
2288 | SEND_DMA_CHECK_ENABLE_CHECK_OPCODE_SMASK
2289 | SEND_DMA_CHECK_ENABLE_CHECK_SLID_SMASK
2290 | SEND_DMA_CHECK_ENABLE_CHECK_VL_SMASK
2291 | SEND_DMA_CHECK_ENABLE_CHECK_ENABLE_SMASK;
2293 if (!HFI1_CAP_IS_KSET(STATIC_RATE_CTRL))
2294 base_sdma_integrity |=
2295 SEND_DMA_CHECK_ENABLE_DISALLOW_PBC_STATIC_RATE_CONTROL_SMASK;
2297 /* turn on send-side job key checks if !A0 */
2299 base_sdma_integrity |=
2300 SEND_DMA_CHECK_ENABLE_CHECK_JOB_KEY_SMASK;
2302 return base_sdma_integrity;
2305 #define dd_dev_emerg(dd, fmt, ...) \
2306 dev_emerg(&(dd)->pcidev->dev, "%s: " fmt, \
2307 rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), ##__VA_ARGS__)
2309 #define dd_dev_err(dd, fmt, ...) \
2310 dev_err(&(dd)->pcidev->dev, "%s: " fmt, \
2311 rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), ##__VA_ARGS__)
2313 #define dd_dev_err_ratelimited(dd, fmt, ...) \
2314 dev_err_ratelimited(&(dd)->pcidev->dev, "%s: " fmt, \
2315 rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), \
2318 #define dd_dev_warn(dd, fmt, ...) \
2319 dev_warn(&(dd)->pcidev->dev, "%s: " fmt, \
2320 rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), ##__VA_ARGS__)
2322 #define dd_dev_warn_ratelimited(dd, fmt, ...) \
2323 dev_warn_ratelimited(&(dd)->pcidev->dev, "%s: " fmt, \
2324 rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), \
2327 #define dd_dev_info(dd, fmt, ...) \
2328 dev_info(&(dd)->pcidev->dev, "%s: " fmt, \
2329 rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), ##__VA_ARGS__)
2331 #define dd_dev_info_ratelimited(dd, fmt, ...) \
2332 dev_info_ratelimited(&(dd)->pcidev->dev, "%s: " fmt, \
2333 rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), \
2336 #define dd_dev_dbg(dd, fmt, ...) \
2337 dev_dbg(&(dd)->pcidev->dev, "%s: " fmt, \
2338 rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), ##__VA_ARGS__)
2340 #define hfi1_dev_porterr(dd, port, fmt, ...) \
2341 dev_err(&(dd)->pcidev->dev, "%s: port %u: " fmt, \
2342 rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), (port), ##__VA_ARGS__)
2345 * this is used for formatting hw error messages...
2347 struct hfi1_hwerror_msgs {
2354 void hfi1_format_hwerrors(u64 hwerrs,
2355 const struct hfi1_hwerror_msgs *hwerrmsgs,
2356 size_t nhwerrmsgs, char *msg, size_t lmsg);
2358 #define USER_OPCODE_CHECK_VAL 0xC0
2359 #define USER_OPCODE_CHECK_MASK 0xC0
2360 #define OPCODE_CHECK_VAL_DISABLED 0x0
2361 #define OPCODE_CHECK_MASK_DISABLED 0x0
2363 static inline void hfi1_reset_cpu_counters(struct hfi1_devdata *dd)
2365 struct hfi1_pportdata *ppd;
2368 dd->z_int_counter = get_all_cpu_total(dd->int_counter);
2369 dd->z_rcv_limit = get_all_cpu_total(dd->rcv_limit);
2370 dd->z_send_schedule = get_all_cpu_total(dd->send_schedule);
2372 ppd = (struct hfi1_pportdata *)(dd + 1);
2373 for (i = 0; i < dd->num_pports; i++, ppd++) {
2374 ppd->ibport_data.rvp.z_rc_acks =
2375 get_all_cpu_total(ppd->ibport_data.rvp.rc_acks);
2376 ppd->ibport_data.rvp.z_rc_qacks =
2377 get_all_cpu_total(ppd->ibport_data.rvp.rc_qacks);
2381 /* Control LED state */
2382 static inline void setextled(struct hfi1_devdata *dd, u32 on)
2385 write_csr(dd, DCC_CFG_LED_CNTRL, 0x1F);
2387 write_csr(dd, DCC_CFG_LED_CNTRL, 0x10);
2390 /* return the i2c resource given the target */
2391 static inline u32 i2c_target(u32 target)
2393 return target ? CR_I2C2 : CR_I2C1;
2396 /* return the i2c chain chip resource that this HFI uses for QSFP */
2397 static inline u32 qsfp_resource(struct hfi1_devdata *dd)
2399 return i2c_target(dd->hfi1_id);
2402 /* Is this device integrated or discrete? */
2403 static inline bool is_integrated(struct hfi1_devdata *dd)
2405 return dd->pcidev->device == PCI_DEVICE_ID_INTEL1;
2409 * hfi1_need_drop - detect need for drop
2412 * In some cases, the first packet needs to be dropped.
2414 * Return true is the current packet needs to be dropped and false otherwise.
2416 static inline bool hfi1_need_drop(struct hfi1_devdata *dd)
2418 if (unlikely(dd->do_drop &&
2419 atomic_xchg(&dd->drop_packet, DROP_PACKET_OFF) ==
2421 dd->do_drop = false;
2427 int hfi1_tempsense_rd(struct hfi1_devdata *dd, struct hfi1_temp *temp);
2429 #define DD_DEV_ENTRY(dd) __string(dev, dev_name(&(dd)->pcidev->dev))
2430 #define DD_DEV_ASSIGN(dd) __assign_str(dev, dev_name(&(dd)->pcidev->dev))
2432 static inline void hfi1_update_ah_attr(struct ib_device *ibdev,
2433 struct rdma_ah_attr *attr)
2435 struct hfi1_pportdata *ppd;
2436 struct hfi1_ibport *ibp;
2437 u32 dlid = rdma_ah_get_dlid(attr);
2440 * Kernel clients may not have setup GRH information
2443 ibp = to_iport(ibdev, rdma_ah_get_port_num(attr));
2444 ppd = ppd_from_ibp(ibp);
2445 if ((((dlid >= be16_to_cpu(IB_MULTICAST_LID_BASE)) ||
2446 (ppd->lid >= be16_to_cpu(IB_MULTICAST_LID_BASE))) &&
2447 (dlid != be32_to_cpu(OPA_LID_PERMISSIVE)) &&
2448 (dlid != be16_to_cpu(IB_LID_PERMISSIVE)) &&
2449 (!(rdma_ah_get_ah_flags(attr) & IB_AH_GRH))) ||
2450 (rdma_ah_get_make_grd(attr))) {
2451 rdma_ah_set_ah_flags(attr, IB_AH_GRH);
2452 rdma_ah_set_interface_id(attr, OPA_MAKE_ID(dlid));
2453 rdma_ah_set_subnet_prefix(attr, ibp->rvp.gid_prefix);
2458 * hfi1_check_mcast- Check if the given lid is
2459 * in the OPA multicast range.
2461 * The LID might either reside in ah.dlid or might be
2462 * in the GRH of the address handle as DGID if extended
2463 * addresses are in use.
2465 static inline bool hfi1_check_mcast(u32 lid)
2467 return ((lid >= opa_get_mcast_base(OPA_MCAST_NR)) &&
2468 (lid != be32_to_cpu(OPA_LID_PERMISSIVE)));
2471 #define opa_get_lid(lid, format) \
2472 __opa_get_lid(lid, OPA_PORT_PACKET_FORMAT_##format)
2474 /* Convert a lid to a specific lid space */
2475 static inline u32 __opa_get_lid(u32 lid, u8 format)
2477 bool is_mcast = hfi1_check_mcast(lid);
2480 case OPA_PORT_PACKET_FORMAT_8B:
2481 case OPA_PORT_PACKET_FORMAT_10B:
2483 return (lid - opa_get_mcast_base(OPA_MCAST_NR) +
2485 return lid & 0xFFFFF;
2486 case OPA_PORT_PACKET_FORMAT_16B:
2488 return (lid - opa_get_mcast_base(OPA_MCAST_NR) +
2490 return lid & 0xFFFFFF;
2491 case OPA_PORT_PACKET_FORMAT_9B:
2494 opa_get_mcast_base(OPA_MCAST_NR) +
2495 be16_to_cpu(IB_MULTICAST_LID_BASE));
2497 return lid & 0xFFFF;
2503 /* Return true if the given lid is the OPA 16B multicast range */
2504 static inline bool hfi1_is_16B_mcast(u32 lid)
2507 opa_get_lid(opa_get_mcast_base(OPA_MCAST_NR), 16B)) &&
2508 (lid != opa_get_lid(be32_to_cpu(OPA_LID_PERMISSIVE), 16B)));
2511 static inline void hfi1_make_opa_lid(struct rdma_ah_attr *attr)
2513 const struct ib_global_route *grh = rdma_ah_read_grh(attr);
2514 u32 dlid = rdma_ah_get_dlid(attr);
2516 /* Modify ah_attr.dlid to be in the 32 bit LID space.
2517 * This is how the address will be laid out:
2518 * Assuming MCAST_NR to be 4,
2519 * 32 bit permissive LID = 0xFFFFFFFF
2520 * Multicast LID range = 0xFFFFFFFE to 0xF0000000
2521 * Unicast LID range = 0xEFFFFFFF to 1
2524 if (ib_is_opa_gid(&grh->dgid))
2525 dlid = opa_get_lid_from_gid(&grh->dgid);
2526 else if ((dlid >= be16_to_cpu(IB_MULTICAST_LID_BASE)) &&
2527 (dlid != be16_to_cpu(IB_LID_PERMISSIVE)) &&
2528 (dlid != be32_to_cpu(OPA_LID_PERMISSIVE)))
2529 dlid = dlid - be16_to_cpu(IB_MULTICAST_LID_BASE) +
2530 opa_get_mcast_base(OPA_MCAST_NR);
2531 else if (dlid == be16_to_cpu(IB_LID_PERMISSIVE))
2532 dlid = be32_to_cpu(OPA_LID_PERMISSIVE);
2534 rdma_ah_set_dlid(attr, dlid);
2537 static inline u8 hfi1_get_packet_type(u32 lid)
2539 /* 9B if lid > 0xF0000000 */
2540 if (lid >= opa_get_mcast_base(OPA_MCAST_NR))
2541 return HFI1_PKT_TYPE_9B;
2543 /* 16B if lid > 0xC000 */
2544 if (lid >= opa_get_lid(opa_get_mcast_base(OPA_MCAST_NR), 9B))
2545 return HFI1_PKT_TYPE_16B;
2547 return HFI1_PKT_TYPE_9B;
2550 static inline bool hfi1_get_hdr_type(u32 lid, struct rdma_ah_attr *attr)
2553 * If there was an incoming 16B packet with permissive
2554 * LIDs, OPA GIDs would have been programmed when those
2555 * packets were received. A 16B packet will have to
2556 * be sent in response to that packet. Return a 16B
2557 * header type if that's the case.
2559 if (rdma_ah_get_dlid(attr) == be32_to_cpu(OPA_LID_PERMISSIVE))
2560 return (ib_is_opa_gid(&rdma_ah_read_grh(attr)->dgid)) ?
2561 HFI1_PKT_TYPE_16B : HFI1_PKT_TYPE_9B;
2564 * Return a 16B header type if either the destination
2565 * or source lid is extended.
2567 if (hfi1_get_packet_type(rdma_ah_get_dlid(attr)) == HFI1_PKT_TYPE_16B)
2568 return HFI1_PKT_TYPE_16B;
2570 return hfi1_get_packet_type(lid);
2573 static inline void hfi1_make_ext_grh(struct hfi1_packet *packet,
2574 struct ib_grh *grh, u32 slid,
2577 struct hfi1_ibport *ibp = &packet->rcd->ppd->ibport_data;
2578 struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
2584 grh->sgid.global.subnet_prefix = ibp->rvp.gid_prefix;
2585 if (slid == opa_get_lid(be32_to_cpu(OPA_LID_PERMISSIVE), 16B))
2586 grh->sgid.global.interface_id =
2587 OPA_MAKE_ID(be32_to_cpu(OPA_LID_PERMISSIVE));
2589 grh->sgid.global.interface_id = OPA_MAKE_ID(slid);
2592 * Upper layers (like mad) may compare the dgid in the
2593 * wc that is obtained here with the sgid_index in
2594 * the wr. Since sgid_index in wr is always 0 for
2595 * extended lids, set the dgid here to the default
2598 grh->dgid.global.subnet_prefix = ibp->rvp.gid_prefix;
2599 grh->dgid.global.interface_id =
2600 cpu_to_be64(ppd->guids[HFI1_PORT_GUID_INDEX]);
2603 static inline int hfi1_get_16b_padding(u32 hdr_size, u32 payload)
2605 return -(hdr_size + payload + (SIZE_OF_CRC << 2) +
2609 static inline void hfi1_make_ib_hdr(struct ib_header *hdr,
2613 hdr->lrh[0] = cpu_to_be16(lrh0);
2614 hdr->lrh[1] = cpu_to_be16(dlid);
2615 hdr->lrh[2] = cpu_to_be16(len);
2616 hdr->lrh[3] = cpu_to_be16(slid);
2619 static inline void hfi1_make_16b_hdr(struct hfi1_16b_header *hdr,
2622 bool becn, bool fecn, u8 l4,
2626 u32 lrh1 = 0x40000000;
2630 lrh0 = (lrh0 & ~OPA_16B_BECN_MASK) | (becn << OPA_16B_BECN_SHIFT);
2631 lrh0 = (lrh0 & ~OPA_16B_LEN_MASK) | (len << OPA_16B_LEN_SHIFT);
2632 lrh0 = (lrh0 & ~OPA_16B_LID_MASK) | (slid & OPA_16B_LID_MASK);
2633 lrh1 = (lrh1 & ~OPA_16B_FECN_MASK) | (fecn << OPA_16B_FECN_SHIFT);
2634 lrh1 = (lrh1 & ~OPA_16B_SC_MASK) | (sc << OPA_16B_SC_SHIFT);
2635 lrh1 = (lrh1 & ~OPA_16B_LID_MASK) | (dlid & OPA_16B_LID_MASK);
2636 lrh2 = (lrh2 & ~OPA_16B_SLID_MASK) |
2637 ((slid >> OPA_16B_SLID_SHIFT) << OPA_16B_SLID_HIGH_SHIFT);
2638 lrh2 = (lrh2 & ~OPA_16B_DLID_MASK) |
2639 ((dlid >> OPA_16B_DLID_SHIFT) << OPA_16B_DLID_HIGH_SHIFT);
2640 lrh2 = (lrh2 & ~OPA_16B_PKEY_MASK) | ((u32)pkey << OPA_16B_PKEY_SHIFT);
2641 lrh2 = (lrh2 & ~OPA_16B_L4_MASK) | l4;
2648 #endif /* _HFI1_KERNEL_H */