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_physnodeaddr(x) ((__pa((unsigned long)(x)) & uv_mmask))
59 #define cpubit_isset(cpu, bau_local_cpumask) \
60 test_bit((cpu), (bau_local_cpumask).bits)
62 /* [19:16] SOFT_ACK timeout period 19: 1 is urgency 7 17:16 1 is multiplier */
64 * UV2: Bit 19 selects between
65 * (0): 10 microsecond timebase and
66 * (1): 80 microseconds
67 * we're using 655us, similar to UV1: 65 units of 10us
69 #define UV1_INTD_SOFT_ACK_TIMEOUT_PERIOD (9UL)
70 #define UV2_INTD_SOFT_ACK_TIMEOUT_PERIOD (15UL)
72 #define UV_INTD_SOFT_ACK_TIMEOUT_PERIOD (is_uv1_hub() ? \
73 UV1_INTD_SOFT_ACK_TIMEOUT_PERIOD : \
74 UV2_INTD_SOFT_ACK_TIMEOUT_PERIOD)
76 #define BAU_MISC_CONTROL_MULT_MASK 3
78 #define UVH_AGING_PRESCALE_SEL 0x000000b000UL
79 /* [30:28] URGENCY_7 an index into a table of times */
80 #define BAU_URGENCY_7_SHIFT 28
81 #define BAU_URGENCY_7_MASK 7
83 #define UVH_TRANSACTION_TIMEOUT 0x000000b200UL
84 /* [45:40] BAU - BAU transaction timeout select - a multiplier */
85 #define BAU_TRANS_SHIFT 40
86 #define BAU_TRANS_MASK 0x3f
89 * shorten some awkward names
91 #define AS_PUSH_SHIFT UVH_LB_BAU_SB_ACTIVATION_CONTROL_PUSH_SHFT
92 #define SOFTACK_MSHIFT UVH_LB_BAU_MISC_CONTROL_ENABLE_INTD_SOFT_ACK_MODE_SHFT
93 #define SOFTACK_PSHIFT UVH_LB_BAU_MISC_CONTROL_INTD_SOFT_ACK_TIMEOUT_PERIOD_SHFT
94 #define SOFTACK_TIMEOUT_PERIOD UV_INTD_SOFT_ACK_TIMEOUT_PERIOD
95 #define write_gmmr uv_write_global_mmr64
96 #define write_lmmr uv_write_local_mmr
97 #define read_lmmr uv_read_local_mmr
98 #define read_gmmr uv_read_global_mmr64
101 * bits in UVH_LB_BAU_SB_ACTIVATION_STATUS_0/1
105 #define DS_DESTINATION_TIMEOUT 2
106 #define DS_SOURCE_TIMEOUT 3
108 * bits put together from HRP_LB_BAU_SB_ACTIVATION_STATUS_0/1/2
109 * values 1 and 3 will not occur
110 * Decoded meaning ERROR BUSY AUX ERR
111 * ------------------------------- ---- ----- -------
113 * BUSY (active) 0 1 0
114 * SW Ack Timeout (destination) 1 0 0
115 * SW Ack INTD rejected (strong NACK) 1 0 1
116 * Source Side Time Out Detected 1 1 0
117 * Destination Side PUT Failed 1 1 1
119 #define UV2H_DESC_IDLE 0
120 #define UV2H_DESC_BUSY 2
121 #define UV2H_DESC_DEST_TIMEOUT 4
122 #define UV2H_DESC_DEST_STRONG_NACK 5
123 #define UV2H_DESC_SOURCE_TIMEOUT 6
124 #define UV2H_DESC_DEST_PUT_ERR 7
127 * delay for 'plugged' timeout retries, in microseconds
129 #define PLUGGED_DELAY 10
132 * threshholds at which to use IPI to free resources
134 /* after this # consecutive 'plugged' timeouts, use IPI to release resources */
135 #define PLUGSB4RESET 100
136 /* after this many consecutive timeouts, use IPI to release resources */
137 #define TIMEOUTSB4RESET 1
138 /* at this number uses of IPI to release resources, giveup the request */
139 #define IPI_RESET_LIMIT 1
140 /* after this # consecutive successes, bump up the throttle if it was lowered */
141 #define COMPLETE_THRESHOLD 5
143 #define UV_LB_SUBNODEID 0x10
145 /* these two are the same for UV1 and UV2: */
146 #define UV_SA_SHFT UVH_LB_BAU_MISC_CONTROL_INTD_SOFT_ACK_TIMEOUT_PERIOD_SHFT
147 #define UV_SA_MASK UVH_LB_BAU_MISC_CONTROL_INTD_SOFT_ACK_TIMEOUT_PERIOD_MASK
148 /* 4 bits of software ack period */
149 #define UV2_ACK_MASK 0x7UL
150 #define UV2_ACK_UNITS_SHFT 3
151 #define UV2_LEG_SHFT UV2H_LB_BAU_MISC_CONTROL_USE_LEGACY_DESCRIPTOR_FORMATS_SHFT
152 #define UV2_EXT_SHFT UV2H_LB_BAU_MISC_CONTROL_ENABLE_EXTENDED_SB_STATUS_SHFT
155 * number of entries in the destination side payload queue
157 #define DEST_Q_SIZE 20
159 * number of destination side software ack resources
161 #define DEST_NUM_RESOURCES 8
163 * completion statuses for sending a TLB flush message
165 #define FLUSH_RETRY_PLUGGED 1
166 #define FLUSH_RETRY_TIMEOUT 2
167 #define FLUSH_GIVEUP 3
168 #define FLUSH_COMPLETE 4
171 * tuning the action when the numalink network is extremely delayed
173 #define CONGESTED_RESPONSE_US 1000 /* 'long' response time, in
175 #define CONGESTED_REPS 10 /* long delays averaged over
176 this many broadcasts */
177 #define CONGESTED_PERIOD 30 /* time for the bau to be
178 disabled, in seconds */
181 #define MSG_REGULAR 1
185 * Distribution: 32 bytes (256 bits) (bytes 0-0x1f of descriptor)
186 * If the 'multilevel' flag in the header portion of the descriptor
187 * has been set to 0, then endpoint multi-unicast mode is selected.
188 * The distribution specification (32 bytes) is interpreted as a 256-bit
189 * distribution vector. Adjacent bits correspond to consecutive even numbered
190 * nodeIDs. The result of adding the index of a given bit to the 15-bit
191 * 'base_dest_nasid' field of the header corresponds to the
192 * destination nodeID associated with that specified bit.
195 unsigned long bits
[BITS_TO_LONGS(UV_DISTRIBUTION_SIZE
)];
199 * mask of cpu's on a uvhub
200 * (during initialization we need to check that unsigned long has
201 * enough bits for max. cpu's per uvhub)
203 struct bau_local_cpumask
{
208 * Payload: 16 bytes (128 bits) (bytes 0x20-0x2f of descriptor)
209 * only 12 bytes (96 bits) of the payload area are usable.
210 * An additional 3 bytes (bits 27:4) of the header address are carried
211 * to the next bytes of the destination payload queue.
212 * And an additional 2 bytes of the header Suppl_A field are also
213 * carried to the destination payload queue.
214 * But the first byte of the Suppl_A becomes bits 127:120 (the 16th byte)
215 * of the destination payload queue, which is written by the hardware
216 * with the s/w ack resource bit vector.
217 * [ effective message contents (16 bytes (128 bits) maximum), not counting
218 * the s/w ack bit vector ]
222 * The payload is software-defined for INTD transactions
224 struct bau_msg_payload
{
225 unsigned long address
; /* signifies a page or all
228 unsigned short sending_cpu
; /* filled in by sender */
230 unsigned short acknowledge_count
; /* filled in by destination */
232 unsigned int reserved1
:32; /* not usable */
237 * Message header: 16 bytes (128 bits) (bytes 0x30-0x3f of descriptor)
238 * see table 4.2.3.0.1 in broacast_assist spec.
240 struct bau_msg_header
{
241 unsigned int dest_subnodeid
:6; /* must be 0x10, for the LB */
243 unsigned int base_dest_nasid
:15; /* nasid of the first bit */
244 /* bits 20:6 */ /* in uvhub map */
245 unsigned int command
:8; /* message type */
247 /* 0x38: SN3net EndPoint Message */
248 unsigned int rsvd_1
:3; /* must be zero */
250 /* int will align on 32 bits */
251 unsigned int rsvd_2
:9; /* must be zero */
253 /* Suppl_A is 56-41 */
254 unsigned int sequence
:16; /* message sequence number */
255 /* bits 56:41 */ /* becomes bytes 16-17 of msg */
256 /* Address field (96:57) is
257 never used as an address
258 (these are address bits
261 unsigned int rsvd_3
:1; /* must be zero */
263 /* address bits 27:4 are payload */
264 /* these next 24 (58-81) bits become bytes 12-14 of msg */
265 /* bits 65:58 land in byte 12 */
266 unsigned int replied_to
:1; /* sent as 0 by the source to
269 unsigned int msg_type
:3; /* software type of the
272 unsigned int canceled
:1; /* message canceled, resource
275 unsigned int payload_1a
:1; /* not currently used */
277 unsigned int payload_1b
:2; /* not currently used */
280 /* bits 73:66 land in byte 13 */
281 unsigned int payload_1ca
:6; /* not currently used */
283 unsigned int payload_1c
:2; /* not currently used */
286 /* bits 81:74 land in byte 14 */
287 unsigned int payload_1d
:6; /* not currently used */
289 unsigned int payload_1e
:2; /* not currently used */
292 unsigned int rsvd_4
:7; /* must be zero */
294 unsigned int swack_flag
:1; /* software acknowledge flag */
296 /* INTD trasactions at
297 destination are to wait for
298 software acknowledge */
299 unsigned int rsvd_5
:6; /* must be zero */
301 unsigned int rsvd_6
:5; /* must be zero */
303 unsigned int int_both
:1; /* if 1, interrupt both sockets
306 unsigned int fairness
:3; /* usually zero */
308 unsigned int multilevel
:1; /* multi-level multicast
311 /* 0 for TLB: endpoint multi-unicast messages */
312 unsigned int chaining
:1; /* next descriptor is part of
315 unsigned int rsvd_7
:21; /* must be zero */
320 * The activation descriptor:
321 * The format of the message to send, plus all accompanying control
325 struct pnmask distribution
;
327 * message template, consisting of header and payload:
329 struct bau_msg_header header
;
330 struct bau_msg_payload payload
;
333 * -payload-- ---------header------
334 * bytes 0-11 bits 41-56 bits 58-81
337 * A/B/C are moved to:
339 * bytes 0-11 bytes 12-14 bytes 16-17 (byte 15 filled in by hw as vector)
340 * ------------payload queue-----------
344 * The payload queue on the destination side is an array of these.
345 * With BAU_MISC_CONTROL set for software acknowledge mode, the messages
346 * are 32 bytes (2 micropackets) (256 bits) in length, but contain only 17
347 * bytes of usable data, including the sw ack vector in byte 15 (bits 127:120)
348 * (12 bytes come from bau_msg_payload, 3 from payload_1, 2 from
349 * swack_vec and payload_2)
350 * "Enabling Software Acknowledgment mode (see Section 4.3.3 Software
351 * Acknowledge Processing) also selects 32 byte (17 bytes usable) payload
354 struct bau_pq_entry
{
355 unsigned long address
; /* signifies a page or all TLB's
357 /* 64 bits, bytes 0-7 */
358 unsigned short sending_cpu
; /* cpu that sent the message */
359 /* 16 bits, bytes 8-9 */
360 unsigned short acknowledge_count
; /* filled in by destination */
361 /* 16 bits, bytes 10-11 */
362 /* these next 3 bytes come from bits 58-81 of the message header */
363 unsigned short replied_to
:1; /* sent as 0 by the source */
364 unsigned short msg_type
:3; /* software message type */
365 unsigned short canceled
:1; /* sent as 0 by the source */
366 unsigned short unused1
:3; /* not currently using */
368 unsigned char unused2a
; /* not currently using */
370 unsigned char unused2
; /* not currently using */
372 unsigned char swack_vec
; /* filled in by the hardware */
373 /* byte 15 (bits 127:120) */
374 unsigned short sequence
; /* message sequence number */
376 unsigned char unused4
[2]; /* not currently using bytes 18-19 */
378 int number_of_cpus
; /* filled in at destination */
379 /* 32 bits, bytes 20-23 (aligned) */
380 unsigned char unused5
[8]; /* not using */
385 struct bau_pq_entry
*msg
;
388 struct bau_pq_entry
*queue_first
;
389 struct bau_pq_entry
*queue_last
;
397 * This structure is allocated per_cpu for UV TLB shootdown statistics.
400 /* sender statistics */
401 unsigned long s_giveup
; /* number of fall backs to
403 unsigned long s_requestor
; /* number of shootdown
405 unsigned long s_stimeout
; /* source side timeouts */
406 unsigned long s_dtimeout
; /* destination side timeouts */
407 unsigned long s_time
; /* time spent in sending side */
408 unsigned long s_retriesok
; /* successful retries */
409 unsigned long s_ntargcpu
; /* total number of cpu's
411 unsigned long s_ntargself
; /* times the sending cpu was
413 unsigned long s_ntarglocals
; /* targets of cpus on the local
415 unsigned long s_ntargremotes
; /* targets of cpus on remote
417 unsigned long s_ntarglocaluvhub
; /* targets of the local hub */
418 unsigned long s_ntargremoteuvhub
; /* remotes hubs targeted */
419 unsigned long s_ntarguvhub
; /* total number of uvhubs
421 unsigned long s_ntarguvhub16
; /* number of times target
423 unsigned long s_ntarguvhub8
; /* number of times target
425 unsigned long s_ntarguvhub4
; /* number of times target
427 unsigned long s_ntarguvhub2
; /* number of times target
429 unsigned long s_ntarguvhub1
; /* number of times target
431 unsigned long s_resets_plug
; /* ipi-style resets from plug
433 unsigned long s_resets_timeout
; /* ipi-style resets from
435 unsigned long s_busy
; /* status stayed busy past
437 unsigned long s_throttles
; /* waits in throttle */
438 unsigned long s_retry_messages
; /* retry broadcasts */
439 unsigned long s_bau_reenabled
; /* for bau enable/disable */
440 unsigned long s_bau_disabled
; /* for bau enable/disable */
441 /* destination statistics */
442 unsigned long d_alltlb
; /* times all tlb's on this
444 unsigned long d_onetlb
; /* times just one tlb on this
446 unsigned long d_multmsg
; /* interrupts with multiple
448 unsigned long d_nomsg
; /* interrupts with no message */
449 unsigned long d_time
; /* time spent on destination
451 unsigned long d_requestee
; /* number of messages
453 unsigned long d_retries
; /* number of retry messages
455 unsigned long d_canceled
; /* number of messages canceled
457 unsigned long d_nocanceled
; /* retries that found nothing
459 unsigned long d_resets
; /* number of ipi-style requests
461 unsigned long d_rcanceled
; /* number of messages canceled
470 struct hub_and_pnode
{
477 short cpu_number
[MAX_CPUS_PER_SOCKET
];
481 unsigned short socket_mask
;
485 struct socket_desc socket
[2];
489 * one per-cpu; to locate the software tables
492 struct bau_desc
*descriptor_base
;
493 struct bau_pq_entry
*queue_first
;
494 struct bau_pq_entry
*queue_last
;
495 struct bau_pq_entry
*bau_msg_head
;
496 struct bau_control
*uvhub_master
;
497 struct bau_control
*socket_master
;
498 struct ptc_stats
*statp
;
500 unsigned long timeout_interval
;
501 unsigned long set_bau_on_time
;
502 atomic_t active_descriptor_count
;
513 short cpus_in_socket
;
515 short partition_base_pnode
;
516 unsigned short message_number
;
517 unsigned short uvhub_quiesce
;
518 short socket_acknowledge_count
[DEST_Q_SIZE
];
519 cycles_t send_message
;
520 spinlock_t uvhub_lock
;
521 spinlock_t queue_lock
;
524 int max_concurr_const
;
529 int complete_threshold
;
530 int cong_response_us
;
533 cycles_t period_time
;
534 long period_requests
;
535 struct hub_and_pnode
*thp
;
538 static inline unsigned long read_mmr_uv2_status(void)
540 return read_lmmr(UV2H_LB_BAU_SB_ACTIVATION_STATUS_2
);
543 static inline void write_mmr_data_broadcast(int pnode
, unsigned long mmr_image
)
545 write_gmmr(pnode
, UVH_BAU_DATA_BROADCAST
, mmr_image
);
548 static inline void write_mmr_descriptor_base(int pnode
, unsigned long mmr_image
)
550 write_gmmr(pnode
, UVH_LB_BAU_SB_DESCRIPTOR_BASE
, mmr_image
);
553 static inline void write_mmr_activation(unsigned long index
)
555 write_lmmr(UVH_LB_BAU_SB_ACTIVATION_CONTROL
, index
);
558 static inline void write_gmmr_activation(int pnode
, unsigned long mmr_image
)
560 write_gmmr(pnode
, UVH_LB_BAU_SB_ACTIVATION_CONTROL
, mmr_image
);
563 static inline void write_mmr_payload_first(int pnode
, unsigned long mmr_image
)
565 write_gmmr(pnode
, UVH_LB_BAU_INTD_PAYLOAD_QUEUE_FIRST
, mmr_image
);
568 static inline void write_mmr_payload_tail(int pnode
, unsigned long mmr_image
)
570 write_gmmr(pnode
, UVH_LB_BAU_INTD_PAYLOAD_QUEUE_TAIL
, mmr_image
);
573 static inline void write_mmr_payload_last(int pnode
, unsigned long mmr_image
)
575 write_gmmr(pnode
, UVH_LB_BAU_INTD_PAYLOAD_QUEUE_LAST
, mmr_image
);
578 static inline void write_mmr_misc_control(int pnode
, unsigned long mmr_image
)
580 write_gmmr(pnode
, UVH_LB_BAU_MISC_CONTROL
, mmr_image
);
583 static inline unsigned long read_mmr_misc_control(int pnode
)
585 return read_gmmr(pnode
, UVH_LB_BAU_MISC_CONTROL
);
588 static inline void write_mmr_sw_ack(unsigned long mr
)
590 uv_write_local_mmr(UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE_ALIAS
, mr
);
593 static inline unsigned long read_mmr_sw_ack(void)
595 return read_lmmr(UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE
);
598 static inline unsigned long read_gmmr_sw_ack(int pnode
)
600 return read_gmmr(pnode
, UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE
);
603 static inline void write_mmr_data_config(int pnode
, unsigned long mr
)
605 uv_write_global_mmr64(pnode
, UVH_BAU_DATA_CONFIG
, mr
);
608 static inline int bau_uvhub_isset(int uvhub
, struct pnmask
*dstp
)
610 return constant_test_bit(uvhub
, &dstp
->bits
[0]);
612 static inline void bau_uvhub_set(int pnode
, struct pnmask
*dstp
)
614 __set_bit(pnode
, &dstp
->bits
[0]);
616 static inline void bau_uvhubs_clear(struct pnmask
*dstp
,
619 bitmap_zero(&dstp
->bits
[0], nbits
);
621 static inline int bau_uvhub_weight(struct pnmask
*dstp
)
623 return bitmap_weight((unsigned long *)&dstp
->bits
[0],
624 UV_DISTRIBUTION_SIZE
);
627 static inline void bau_cpubits_clear(struct bau_local_cpumask
*dstp
, int nbits
)
629 bitmap_zero(&dstp
->bits
, nbits
);
632 extern void uv_bau_message_intr1(void);
633 extern void uv_bau_timeout_intr1(void);
635 struct atomic_short
{
640 * atomic_read_short - read a short atomic variable
641 * @v: pointer of type atomic_short
643 * Atomically reads the value of @v.
645 static inline int atomic_read_short(const struct atomic_short
*v
)
651 * atom_asr - add and return a short int
652 * @i: short value to add
653 * @v: pointer of type atomic_short
655 * Atomically adds @i to @v and returns @i + @v
657 static inline int atom_asr(short i
, struct atomic_short
*v
)
660 asm volatile(LOCK_PREFIX
"xaddw %0, %1"
661 : "+r" (i
), "+m" (v
->counter
)
667 * conditionally add 1 to *v, unless *v is >= u
668 * return 0 if we cannot add 1 to *v because it is >= u
669 * return 1 if we can add 1 to *v because it is < u
672 * This is close to atomic_add_unless(), but this allows the 'u' value
673 * to be lowered below the current 'v'. atomic_add_unless can only stop
676 static inline int atomic_inc_unless_ge(spinlock_t
*lock
, atomic_t
*v
, int u
)
679 if (atomic_read(v
) >= u
) {
688 #endif /* _ASM_X86_UV_UV_BAU_H */