2 * This file is subject to the terms and conditions of the GNU General Public
3 * License. See the file "COPYING" in the main directory of this archive
6 * SGI UV Broadcast Assist Unit definitions
8 * Copyright (C) 2008-2011 Silicon Graphics, Inc. All rights reserved.
11 #ifndef _ASM_X86_UV_UV_BAU_H
12 #define _ASM_X86_UV_UV_BAU_H
14 #include <linux/bitmap.h>
18 * Broadcast Assist Unit messaging structures
20 * Selective Broadcast activations are induced by software action
21 * specifying a particular 8-descriptor "set" via a 6-bit index written
23 * Thus there are 64 unique 512-byte sets of SB descriptors - one set for
24 * each 6-bit index value. These descriptor sets are mapped in sequence
25 * starting with set 0 located at the address specified in the
26 * BAU_SB_DESCRIPTOR_BASE register, set 1 is located at BASE + 512,
27 * set 2 is at BASE + 2*512, set 3 at BASE + 3*512, and so on.
29 * We will use one set for sending BAU messages from each of the
32 * TLB shootdown will use the first of the 8 descriptors of each set.
33 * Each of the descriptors is 64 bytes in size (8*64 = 512 bytes in a set).
36 #define MAX_CPUS_PER_UVHUB 64
37 #define MAX_CPUS_PER_SOCKET 32
38 #define ADP_SZ 64 /* hardware-provided max. */
39 #define UV_CPUS_PER_AS 32 /* hardware-provided max. */
40 #define ITEMS_PER_DESC 8
41 /* the 'throttle' to prevent the hardware stay-busy bug */
42 #define MAX_BAU_CONCURRENT 3
43 #define UV_ACT_STATUS_MASK 0x3
44 #define UV_ACT_STATUS_SIZE 2
45 #define UV_DISTRIBUTION_SIZE 256
46 #define UV_SW_ACK_NPENDING 8
47 #define UV1_NET_ENDPOINT_INTD 0x38
48 #define UV2_NET_ENDPOINT_INTD 0x28
49 #define UV_NET_ENDPOINT_INTD (is_uv1_hub() ? \
50 UV1_NET_ENDPOINT_INTD : UV2_NET_ENDPOINT_INTD)
51 #define UV_DESC_PSHIFT 49
52 #define UV_PAYLOADQ_PNODE_SHIFT 49
53 #define UV_PTC_BASENAME "sgi_uv/ptc_statistics"
54 #define UV_BAU_BASENAME "sgi_uv/bau_tunables"
55 #define UV_BAU_TUNABLES_DIR "sgi_uv"
56 #define UV_BAU_TUNABLES_FILE "bau_tunables"
57 #define WHITESPACE " \t\n"
58 #define uv_mmask ((1UL << uv_hub_info->m_val) - 1)
59 #define uv_physnodeaddr(x) ((__pa((unsigned long)(x)) & uv_mmask))
60 #define cpubit_isset(cpu, bau_local_cpumask) \
61 test_bit((cpu), (bau_local_cpumask).bits)
63 /* [19:16] SOFT_ACK timeout period 19: 1 is urgency 7 17:16 1 is multiplier */
65 * UV2: Bit 19 selects between
66 * (0): 10 microsecond timebase and
67 * (1): 80 microseconds
68 * we're using 655us, similar to UV1: 65 units of 10us
70 #define UV1_INTD_SOFT_ACK_TIMEOUT_PERIOD (9UL)
71 #define UV2_INTD_SOFT_ACK_TIMEOUT_PERIOD (15UL)
73 #define UV_INTD_SOFT_ACK_TIMEOUT_PERIOD (is_uv1_hub() ? \
74 UV1_INTD_SOFT_ACK_TIMEOUT_PERIOD : \
75 UV2_INTD_SOFT_ACK_TIMEOUT_PERIOD)
77 #define BAU_MISC_CONTROL_MULT_MASK 3
79 #define UVH_AGING_PRESCALE_SEL 0x000000b000UL
80 /* [30:28] URGENCY_7 an index into a table of times */
81 #define BAU_URGENCY_7_SHIFT 28
82 #define BAU_URGENCY_7_MASK 7
84 #define UVH_TRANSACTION_TIMEOUT 0x000000b200UL
85 /* [45:40] BAU - BAU transaction timeout select - a multiplier */
86 #define BAU_TRANS_SHIFT 40
87 #define BAU_TRANS_MASK 0x3f
90 * shorten some awkward names
92 #define AS_PUSH_SHIFT UVH_LB_BAU_SB_ACTIVATION_CONTROL_PUSH_SHFT
93 #define SOFTACK_MSHIFT UVH_LB_BAU_MISC_CONTROL_ENABLE_INTD_SOFT_ACK_MODE_SHFT
94 #define SOFTACK_PSHIFT UVH_LB_BAU_MISC_CONTROL_INTD_SOFT_ACK_TIMEOUT_PERIOD_SHFT
95 #define SOFTACK_TIMEOUT_PERIOD UV_INTD_SOFT_ACK_TIMEOUT_PERIOD
96 #define write_gmmr uv_write_global_mmr64
97 #define write_lmmr uv_write_local_mmr
98 #define read_lmmr uv_read_local_mmr
99 #define read_gmmr uv_read_global_mmr64
102 * bits in UVH_LB_BAU_SB_ACTIVATION_STATUS_0/1
106 #define DS_DESTINATION_TIMEOUT 2
107 #define DS_SOURCE_TIMEOUT 3
109 * bits put together from HRP_LB_BAU_SB_ACTIVATION_STATUS_0/1/2
110 * values 1 and 3 will not occur
111 * Decoded meaning ERROR BUSY AUX ERR
112 * ------------------------------- ---- ----- -------
114 * BUSY (active) 0 1 0
115 * SW Ack Timeout (destination) 1 0 0
116 * SW Ack INTD rejected (strong NACK) 1 0 1
117 * Source Side Time Out Detected 1 1 0
118 * Destination Side PUT Failed 1 1 1
120 #define UV2H_DESC_IDLE 0
121 #define UV2H_DESC_BUSY 2
122 #define UV2H_DESC_DEST_TIMEOUT 4
123 #define UV2H_DESC_DEST_STRONG_NACK 5
124 #define UV2H_DESC_SOURCE_TIMEOUT 6
125 #define UV2H_DESC_DEST_PUT_ERR 7
128 * delay for 'plugged' timeout retries, in microseconds
130 #define PLUGGED_DELAY 10
133 * threshholds at which to use IPI to free resources
135 /* after this # consecutive 'plugged' timeouts, use IPI to release resources */
136 #define PLUGSB4RESET 100
137 /* after this many consecutive timeouts, use IPI to release resources */
138 #define TIMEOUTSB4RESET 1
139 /* at this number uses of IPI to release resources, giveup the request */
140 #define IPI_RESET_LIMIT 1
141 /* after this # consecutive successes, bump up the throttle if it was lowered */
142 #define COMPLETE_THRESHOLD 5
144 #define UV_LB_SUBNODEID 0x10
146 /* these two are the same for UV1 and UV2: */
147 #define UV_SA_SHFT UVH_LB_BAU_MISC_CONTROL_INTD_SOFT_ACK_TIMEOUT_PERIOD_SHFT
148 #define UV_SA_MASK UVH_LB_BAU_MISC_CONTROL_INTD_SOFT_ACK_TIMEOUT_PERIOD_MASK
149 /* 4 bits of software ack period */
150 #define UV2_ACK_MASK 0x7UL
151 #define UV2_ACK_UNITS_SHFT 3
152 #define UV2_LEG_SHFT UV2H_LB_BAU_MISC_CONTROL_USE_LEGACY_DESCRIPTOR_FORMATS_SHFT
153 #define UV2_EXT_SHFT UV2H_LB_BAU_MISC_CONTROL_ENABLE_EXTENDED_SB_STATUS_SHFT
156 * number of entries in the destination side payload queue
158 #define DEST_Q_SIZE 20
160 * number of destination side software ack resources
162 #define DEST_NUM_RESOURCES 8
164 * completion statuses for sending a TLB flush message
166 #define FLUSH_RETRY_PLUGGED 1
167 #define FLUSH_RETRY_TIMEOUT 2
168 #define FLUSH_GIVEUP 3
169 #define FLUSH_COMPLETE 4
172 * tuning the action when the numalink network is extremely delayed
174 #define CONGESTED_RESPONSE_US 1000 /* 'long' response time, in
176 #define CONGESTED_REPS 10 /* long delays averaged over
177 this many broadcasts */
178 #define CONGESTED_PERIOD 30 /* time for the bau to be
179 disabled, in seconds */
182 #define MSG_REGULAR 1
186 * Distribution: 32 bytes (256 bits) (bytes 0-0x1f of descriptor)
187 * If the 'multilevel' flag in the header portion of the descriptor
188 * has been set to 0, then endpoint multi-unicast mode is selected.
189 * The distribution specification (32 bytes) is interpreted as a 256-bit
190 * distribution vector. Adjacent bits correspond to consecutive even numbered
191 * nodeIDs. The result of adding the index of a given bit to the 15-bit
192 * 'base_dest_nasid' field of the header corresponds to the
193 * destination nodeID associated with that specified bit.
196 unsigned long bits
[BITS_TO_LONGS(UV_DISTRIBUTION_SIZE
)];
200 * mask of cpu's on a uvhub
201 * (during initialization we need to check that unsigned long has
202 * enough bits for max. cpu's per uvhub)
204 struct bau_local_cpumask
{
209 * Payload: 16 bytes (128 bits) (bytes 0x20-0x2f of descriptor)
210 * only 12 bytes (96 bits) of the payload area are usable.
211 * An additional 3 bytes (bits 27:4) of the header address are carried
212 * to the next bytes of the destination payload queue.
213 * And an additional 2 bytes of the header Suppl_A field are also
214 * carried to the destination payload queue.
215 * But the first byte of the Suppl_A becomes bits 127:120 (the 16th byte)
216 * of the destination payload queue, which is written by the hardware
217 * with the s/w ack resource bit vector.
218 * [ effective message contents (16 bytes (128 bits) maximum), not counting
219 * the s/w ack bit vector ]
223 * The payload is software-defined for INTD transactions
225 struct bau_msg_payload
{
226 unsigned long address
; /* signifies a page or all
229 unsigned short sending_cpu
; /* filled in by sender */
231 unsigned short acknowledge_count
; /* filled in by destination */
233 unsigned int reserved1
:32; /* not usable */
238 * Message header: 16 bytes (128 bits) (bytes 0x30-0x3f of descriptor)
239 * see table 4.2.3.0.1 in broacast_assist spec.
241 struct bau_msg_header
{
242 unsigned int dest_subnodeid
:6; /* must be 0x10, for the LB */
244 unsigned int base_dest_nasid
:15; /* nasid of the first bit */
245 /* bits 20:6 */ /* in uvhub map */
246 unsigned int command
:8; /* message type */
248 /* 0x38: SN3net EndPoint Message */
249 unsigned int rsvd_1
:3; /* must be zero */
251 /* int will align on 32 bits */
252 unsigned int rsvd_2
:9; /* must be zero */
254 /* Suppl_A is 56-41 */
255 unsigned int sequence
:16; /* message sequence number */
256 /* bits 56:41 */ /* becomes bytes 16-17 of msg */
257 /* Address field (96:57) is
258 never used as an address
259 (these are address bits
262 unsigned int rsvd_3
:1; /* must be zero */
264 /* address bits 27:4 are payload */
265 /* these next 24 (58-81) bits become bytes 12-14 of msg */
266 /* bits 65:58 land in byte 12 */
267 unsigned int replied_to
:1; /* sent as 0 by the source to
270 unsigned int msg_type
:3; /* software type of the
273 unsigned int canceled
:1; /* message canceled, resource
276 unsigned int payload_1a
:1; /* not currently used */
278 unsigned int payload_1b
:2; /* not currently used */
281 /* bits 73:66 land in byte 13 */
282 unsigned int payload_1ca
:6; /* not currently used */
284 unsigned int payload_1c
:2; /* not currently used */
287 /* bits 81:74 land in byte 14 */
288 unsigned int payload_1d
:6; /* not currently used */
290 unsigned int payload_1e
:2; /* not currently used */
293 unsigned int rsvd_4
:7; /* must be zero */
295 unsigned int swack_flag
:1; /* software acknowledge flag */
297 /* INTD trasactions at
298 destination are to wait for
299 software acknowledge */
300 unsigned int rsvd_5
:6; /* must be zero */
302 unsigned int rsvd_6
:5; /* must be zero */
304 unsigned int int_both
:1; /* if 1, interrupt both sockets
307 unsigned int fairness
:3; /* usually zero */
309 unsigned int multilevel
:1; /* multi-level multicast
312 /* 0 for TLB: endpoint multi-unicast messages */
313 unsigned int chaining
:1; /* next descriptor is part of
316 unsigned int rsvd_7
:21; /* must be zero */
321 * The activation descriptor:
322 * The format of the message to send, plus all accompanying control
326 struct pnmask distribution
;
328 * message template, consisting of header and payload:
330 struct bau_msg_header header
;
331 struct bau_msg_payload payload
;
334 * -payload-- ---------header------
335 * bytes 0-11 bits 41-56 bits 58-81
338 * A/B/C are moved to:
340 * bytes 0-11 bytes 12-14 bytes 16-17 (byte 15 filled in by hw as vector)
341 * ------------payload queue-----------
345 * The payload queue on the destination side is an array of these.
346 * With BAU_MISC_CONTROL set for software acknowledge mode, the messages
347 * are 32 bytes (2 micropackets) (256 bits) in length, but contain only 17
348 * bytes of usable data, including the sw ack vector in byte 15 (bits 127:120)
349 * (12 bytes come from bau_msg_payload, 3 from payload_1, 2 from
350 * swack_vec and payload_2)
351 * "Enabling Software Acknowledgment mode (see Section 4.3.3 Software
352 * Acknowledge Processing) also selects 32 byte (17 bytes usable) payload
355 struct bau_pq_entry
{
356 unsigned long address
; /* signifies a page or all TLB's
358 /* 64 bits, bytes 0-7 */
359 unsigned short sending_cpu
; /* cpu that sent the message */
360 /* 16 bits, bytes 8-9 */
361 unsigned short acknowledge_count
; /* filled in by destination */
362 /* 16 bits, bytes 10-11 */
363 /* these next 3 bytes come from bits 58-81 of the message header */
364 unsigned short replied_to
:1; /* sent as 0 by the source */
365 unsigned short msg_type
:3; /* software message type */
366 unsigned short canceled
:1; /* sent as 0 by the source */
367 unsigned short unused1
:3; /* not currently using */
369 unsigned char unused2a
; /* not currently using */
371 unsigned char unused2
; /* not currently using */
373 unsigned char swack_vec
; /* filled in by the hardware */
374 /* byte 15 (bits 127:120) */
375 unsigned short sequence
; /* message sequence number */
377 unsigned char unused4
[2]; /* not currently using bytes 18-19 */
379 int number_of_cpus
; /* filled in at destination */
380 /* 32 bits, bytes 20-23 (aligned) */
381 unsigned char unused5
[8]; /* not using */
386 struct bau_pq_entry
*msg
;
389 struct bau_pq_entry
*queue_first
;
390 struct bau_pq_entry
*queue_last
;
398 * This structure is allocated per_cpu for UV TLB shootdown statistics.
401 /* sender statistics */
402 unsigned long s_giveup
; /* number of fall backs to
404 unsigned long s_requestor
; /* number of shootdown
406 unsigned long s_stimeout
; /* source side timeouts */
407 unsigned long s_dtimeout
; /* destination side timeouts */
408 unsigned long s_time
; /* time spent in sending side */
409 unsigned long s_retriesok
; /* successful retries */
410 unsigned long s_ntargcpu
; /* total number of cpu's
412 unsigned long s_ntargself
; /* times the sending cpu was
414 unsigned long s_ntarglocals
; /* targets of cpus on the local
416 unsigned long s_ntargremotes
; /* targets of cpus on remote
418 unsigned long s_ntarglocaluvhub
; /* targets of the local hub */
419 unsigned long s_ntargremoteuvhub
; /* remotes hubs targeted */
420 unsigned long s_ntarguvhub
; /* total number of uvhubs
422 unsigned long s_ntarguvhub16
; /* number of times target
424 unsigned long s_ntarguvhub8
; /* number of times target
426 unsigned long s_ntarguvhub4
; /* number of times target
428 unsigned long s_ntarguvhub2
; /* number of times target
430 unsigned long s_ntarguvhub1
; /* number of times target
432 unsigned long s_resets_plug
; /* ipi-style resets from plug
434 unsigned long s_resets_timeout
; /* ipi-style resets from
436 unsigned long s_busy
; /* status stayed busy past
438 unsigned long s_throttles
; /* waits in throttle */
439 unsigned long s_retry_messages
; /* retry broadcasts */
440 unsigned long s_bau_reenabled
; /* for bau enable/disable */
441 unsigned long s_bau_disabled
; /* for bau enable/disable */
442 /* destination statistics */
443 unsigned long d_alltlb
; /* times all tlb's on this
445 unsigned long d_onetlb
; /* times just one tlb on this
447 unsigned long d_multmsg
; /* interrupts with multiple
449 unsigned long d_nomsg
; /* interrupts with no message */
450 unsigned long d_time
; /* time spent on destination
452 unsigned long d_requestee
; /* number of messages
454 unsigned long d_retries
; /* number of retry messages
456 unsigned long d_canceled
; /* number of messages canceled
458 unsigned long d_nocanceled
; /* retries that found nothing
460 unsigned long d_resets
; /* number of ipi-style requests
462 unsigned long d_rcanceled
; /* number of messages canceled
471 struct hub_and_pnode
{
478 short cpu_number
[MAX_CPUS_PER_SOCKET
];
482 unsigned short socket_mask
;
486 struct socket_desc socket
[2];
490 * one per-cpu; to locate the software tables
493 struct bau_desc
*descriptor_base
;
494 struct bau_pq_entry
*queue_first
;
495 struct bau_pq_entry
*queue_last
;
496 struct bau_pq_entry
*bau_msg_head
;
497 struct bau_control
*uvhub_master
;
498 struct bau_control
*socket_master
;
499 struct ptc_stats
*statp
;
501 unsigned long timeout_interval
;
502 unsigned long set_bau_on_time
;
503 atomic_t active_descriptor_count
;
514 short cpus_in_socket
;
516 short partition_base_pnode
;
517 unsigned short message_number
;
518 unsigned short uvhub_quiesce
;
519 short socket_acknowledge_count
[DEST_Q_SIZE
];
520 cycles_t send_message
;
521 spinlock_t uvhub_lock
;
522 spinlock_t queue_lock
;
525 int max_concurr_const
;
530 int complete_threshold
;
531 int cong_response_us
;
534 cycles_t period_time
;
535 long period_requests
;
536 struct hub_and_pnode
*thp
;
539 static inline unsigned long read_mmr_uv2_status(void)
541 return read_lmmr(UV2H_LB_BAU_SB_ACTIVATION_STATUS_2
);
544 static inline void write_mmr_data_broadcast(int pnode
, unsigned long mmr_image
)
546 write_gmmr(pnode
, UVH_BAU_DATA_BROADCAST
, mmr_image
);
549 static inline void write_mmr_descriptor_base(int pnode
, unsigned long mmr_image
)
551 write_gmmr(pnode
, UVH_LB_BAU_SB_DESCRIPTOR_BASE
, mmr_image
);
554 static inline void write_mmr_activation(unsigned long index
)
556 write_lmmr(UVH_LB_BAU_SB_ACTIVATION_CONTROL
, index
);
559 static inline void write_gmmr_activation(int pnode
, unsigned long mmr_image
)
561 write_gmmr(pnode
, UVH_LB_BAU_SB_ACTIVATION_CONTROL
, mmr_image
);
564 static inline void write_mmr_payload_first(int pnode
, unsigned long mmr_image
)
566 write_gmmr(pnode
, UVH_LB_BAU_INTD_PAYLOAD_QUEUE_FIRST
, mmr_image
);
569 static inline void write_mmr_payload_tail(int pnode
, unsigned long mmr_image
)
571 write_gmmr(pnode
, UVH_LB_BAU_INTD_PAYLOAD_QUEUE_TAIL
, mmr_image
);
574 static inline void write_mmr_payload_last(int pnode
, unsigned long mmr_image
)
576 write_gmmr(pnode
, UVH_LB_BAU_INTD_PAYLOAD_QUEUE_LAST
, mmr_image
);
579 static inline void write_mmr_misc_control(int pnode
, unsigned long mmr_image
)
581 write_gmmr(pnode
, UVH_LB_BAU_MISC_CONTROL
, mmr_image
);
584 static inline unsigned long read_mmr_misc_control(int pnode
)
586 return read_gmmr(pnode
, UVH_LB_BAU_MISC_CONTROL
);
589 static inline void write_mmr_sw_ack(unsigned long mr
)
591 uv_write_local_mmr(UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE_ALIAS
, mr
);
594 static inline unsigned long read_mmr_sw_ack(void)
596 return read_lmmr(UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE
);
599 static inline unsigned long read_gmmr_sw_ack(int pnode
)
601 return read_gmmr(pnode
, UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE
);
604 static inline void write_mmr_data_config(int pnode
, unsigned long mr
)
606 uv_write_global_mmr64(pnode
, UVH_BAU_DATA_CONFIG
, mr
);
609 static inline int bau_uvhub_isset(int uvhub
, struct pnmask
*dstp
)
611 return constant_test_bit(uvhub
, &dstp
->bits
[0]);
613 static inline void bau_uvhub_set(int pnode
, struct pnmask
*dstp
)
615 __set_bit(pnode
, &dstp
->bits
[0]);
617 static inline void bau_uvhubs_clear(struct pnmask
*dstp
,
620 bitmap_zero(&dstp
->bits
[0], nbits
);
622 static inline int bau_uvhub_weight(struct pnmask
*dstp
)
624 return bitmap_weight((unsigned long *)&dstp
->bits
[0],
625 UV_DISTRIBUTION_SIZE
);
628 static inline void bau_cpubits_clear(struct bau_local_cpumask
*dstp
, int nbits
)
630 bitmap_zero(&dstp
->bits
, nbits
);
633 extern void uv_bau_message_intr1(void);
634 extern void uv_bau_timeout_intr1(void);
636 struct atomic_short
{
641 * atomic_read_short - read a short atomic variable
642 * @v: pointer of type atomic_short
644 * Atomically reads the value of @v.
646 static inline int atomic_read_short(const struct atomic_short
*v
)
652 * atom_asr - add and return a short int
653 * @i: short value to add
654 * @v: pointer of type atomic_short
656 * Atomically adds @i to @v and returns @i + @v
658 static inline int atom_asr(short i
, struct atomic_short
*v
)
660 return i
+ xadd(&v
->counter
, i
);
664 * conditionally add 1 to *v, unless *v is >= u
665 * return 0 if we cannot add 1 to *v because it is >= u
666 * return 1 if we can add 1 to *v because it is < u
669 * This is close to atomic_add_unless(), but this allows the 'u' value
670 * to be lowered below the current 'v'. atomic_add_unless can only stop
673 static inline int atomic_inc_unless_ge(spinlock_t
*lock
, atomic_t
*v
, int u
)
676 if (atomic_read(v
) >= u
) {
685 #endif /* _ASM_X86_UV_UV_BAU_H */