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 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 UV_ADP_SIZE 64 /* hardware-provided max. */
39 #define UV_CPUS_PER_ACT_STATUS 32 /* hardware-provided max. */
40 #define UV_ITEMS_PER_DESCRIPTOR 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 UV_NET_ENDPOINT_INTD 0x38
48 #define UV_DESC_BASE_PNODE_SHIFT 49
49 #define UV_PAYLOADQ_PNODE_SHIFT 49
50 #define UV_PTC_BASENAME "sgi_uv/ptc_statistics"
51 #define UV_BAU_BASENAME "sgi_uv/bau_tunables"
52 #define UV_BAU_TUNABLES_DIR "sgi_uv"
53 #define UV_BAU_TUNABLES_FILE "bau_tunables"
54 #define WHITESPACE " \t\n"
55 #define uv_physnodeaddr(x) ((__pa((unsigned long)(x)) & uv_mmask))
56 #define UV_ENABLE_INTD_SOFT_ACK_MODE_SHIFT 15
57 #define UV_INTD_SOFT_ACK_TIMEOUT_PERIOD_SHIFT 16
58 #define UV_INTD_SOFT_ACK_TIMEOUT_PERIOD 0x0000000009UL
59 /* [19:16] SOFT_ACK timeout period 19: 1 is urgency 7 17:16 1 is multiplier */
60 #define BAU_MISC_CONTROL_MULT_MASK 3
62 #define UVH_AGING_PRESCALE_SEL 0x000000b000UL
63 /* [30:28] URGENCY_7 an index into a table of times */
64 #define BAU_URGENCY_7_SHIFT 28
65 #define BAU_URGENCY_7_MASK 7
67 #define UVH_TRANSACTION_TIMEOUT 0x000000b200UL
68 /* [45:40] BAU - BAU transaction timeout select - a multiplier */
69 #define BAU_TRANS_SHIFT 40
70 #define BAU_TRANS_MASK 0x3f
73 * bits in UVH_LB_BAU_SB_ACTIVATION_STATUS_0/1
75 #define DESC_STATUS_IDLE 0
76 #define DESC_STATUS_ACTIVE 1
77 #define DESC_STATUS_DESTINATION_TIMEOUT 2
78 #define DESC_STATUS_SOURCE_TIMEOUT 3
81 * delay for 'plugged' timeout retries, in microseconds
83 #define PLUGGED_DELAY 10
86 * threshholds at which to use IPI to free resources
88 /* after this # consecutive 'plugged' timeouts, use IPI to release resources */
89 #define PLUGSB4RESET 100
90 /* after this many consecutive timeouts, use IPI to release resources */
91 #define TIMEOUTSB4RESET 1
92 /* at this number uses of IPI to release resources, giveup the request */
93 #define IPI_RESET_LIMIT 1
94 /* after this # consecutive successes, bump up the throttle if it was lowered */
95 #define COMPLETE_THRESHOLD 5
97 #define UV_LB_SUBNODEID 0x10
100 * number of entries in the destination side payload queue
102 #define DEST_Q_SIZE 20
104 * number of destination side software ack resources
106 #define DEST_NUM_RESOURCES 8
108 * completion statuses for sending a TLB flush message
110 #define FLUSH_RETRY_PLUGGED 1
111 #define FLUSH_RETRY_TIMEOUT 2
112 #define FLUSH_GIVEUP 3
113 #define FLUSH_COMPLETE 4
116 * tuning the action when the numalink network is extremely delayed
118 #define CONGESTED_RESPONSE_US 1000 /* 'long' response time, in microseconds */
119 #define CONGESTED_REPS 10 /* long delays averaged over this many broadcasts */
120 #define CONGESTED_PERIOD 30 /* time for the bau to be disabled, in seconds */
123 * Distribution: 32 bytes (256 bits) (bytes 0-0x1f of descriptor)
124 * If the 'multilevel' flag in the header portion of the descriptor
125 * has been set to 0, then endpoint multi-unicast mode is selected.
126 * The distribution specification (32 bytes) is interpreted as a 256-bit
127 * distribution vector. Adjacent bits correspond to consecutive even numbered
128 * nodeIDs. The result of adding the index of a given bit to the 15-bit
129 * 'base_dest_nasid' field of the header corresponds to the
130 * destination nodeID associated with that specified bit.
132 struct bau_target_uvhubmask {
133 unsigned long bits[BITS_TO_LONGS(UV_DISTRIBUTION_SIZE)];
137 * mask of cpu's on a uvhub
138 * (during initialization we need to check that unsigned long has
139 * enough bits for max. cpu's per uvhub)
141 struct bau_local_cpumask {
146 * Payload: 16 bytes (128 bits) (bytes 0x20-0x2f of descriptor)
147 * only 12 bytes (96 bits) of the payload area are usable.
148 * An additional 3 bytes (bits 27:4) of the header address are carried
149 * to the next bytes of the destination payload queue.
150 * And an additional 2 bytes of the header Suppl_A field are also
151 * carried to the destination payload queue.
152 * But the first byte of the Suppl_A becomes bits 127:120 (the 16th byte)
153 * of the destination payload queue, which is written by the hardware
154 * with the s/w ack resource bit vector.
155 * [ effective message contents (16 bytes (128 bits) maximum), not counting
156 * the s/w ack bit vector ]
160 * The payload is software-defined for INTD transactions
162 struct bau_msg_payload {
163 unsigned long address; /* signifies a page or all TLB's
166 unsigned short sending_cpu; /* filled in by sender */
168 unsigned short acknowledge_count;/* filled in by destination */
170 unsigned int reserved1:32; /* not usable */
175 * Message header: 16 bytes (128 bits) (bytes 0x30-0x3f of descriptor)
176 * see table 4.2.3.0.1 in broacast_assist spec.
178 struct bau_msg_header {
179 unsigned int dest_subnodeid:6; /* must be 0x10, for the LB */
181 unsigned int base_dest_nasid:15; /* nasid of the */
182 /* bits 20:6 */ /* first bit in uvhub map */
183 unsigned int command:8; /* message type */
185 /* 0x38: SN3net EndPoint Message */
186 unsigned int rsvd_1:3; /* must be zero */
188 /* int will align on 32 bits */
189 unsigned int rsvd_2:9; /* must be zero */
191 /* Suppl_A is 56-41 */
192 unsigned int sequence:16;/* message sequence number */
193 /* bits 56:41 */ /* becomes bytes 16-17 of msg */
194 /* Address field (96:57) is never used as an
195 address (these are address bits 42:3) */
197 unsigned int rsvd_3:1; /* must be zero */
199 /* address bits 27:4 are payload */
200 /* these next 24 (58-81) bits become bytes 12-14 of msg */
202 /* bits 65:58 land in byte 12 */
203 unsigned int replied_to:1;/* sent as 0 by the source to byte 12 */
205 unsigned int msg_type:3; /* software type of the message*/
207 unsigned int canceled:1; /* message canceled, resource to be freed*/
209 unsigned int payload_1a:1;/* not currently used */
211 unsigned int payload_1b:2;/* not currently used */
214 /* bits 73:66 land in byte 13 */
215 unsigned int payload_1ca:6;/* not currently used */
217 unsigned int payload_1c:2;/* not currently used */
220 /* bits 81:74 land in byte 14 */
221 unsigned int payload_1d:6;/* not currently used */
223 unsigned int payload_1e:2;/* not currently used */
226 unsigned int rsvd_4:7; /* must be zero */
228 unsigned int sw_ack_flag:1;/* software acknowledge flag */
230 /* INTD trasactions at destination are to
231 wait for software acknowledge */
232 unsigned int rsvd_5:6; /* must be zero */
234 unsigned int rsvd_6:5; /* must be zero */
236 unsigned int int_both:1;/* if 1, interrupt both sockets on the uvhub */
238 unsigned int fairness:3;/* usually zero */
240 unsigned int multilevel:1; /* multi-level multicast format */
242 /* 0 for TLB: endpoint multi-unicast messages */
243 unsigned int chaining:1;/* next descriptor is part of this activation*/
245 unsigned int rsvd_7:21; /* must be zero */
251 #define MSG_REGULAR 1
255 * The activation descriptor:
256 * The format of the message to send, plus all accompanying control
260 struct bau_target_uvhubmask distribution;
262 * message template, consisting of header and payload:
264 struct bau_msg_header header;
265 struct bau_msg_payload payload;
268 * -payload-- ---------header------
269 * bytes 0-11 bits 41-56 bits 58-81
272 * A/B/C are moved to:
274 * bytes 0-11 bytes 12-14 bytes 16-17 (byte 15 filled in by hw as vector)
275 * ------------payload queue-----------
279 * The payload queue on the destination side is an array of these.
280 * With BAU_MISC_CONTROL set for software acknowledge mode, the messages
281 * are 32 bytes (2 micropackets) (256 bits) in length, but contain only 17
282 * bytes of usable data, including the sw ack vector in byte 15 (bits 127:120)
283 * (12 bytes come from bau_msg_payload, 3 from payload_1, 2 from
284 * sw_ack_vector and payload_2)
285 * "Enabling Software Acknowledgment mode (see Section 4.3.3 Software
286 * Acknowledge Processing) also selects 32 byte (17 bytes usable) payload
289 struct bau_payload_queue_entry {
290 unsigned long address; /* signifies a page or all TLB's
292 /* 64 bits, bytes 0-7 */
294 unsigned short sending_cpu; /* cpu that sent the message */
295 /* 16 bits, bytes 8-9 */
297 unsigned short acknowledge_count; /* filled in by destination */
298 /* 16 bits, bytes 10-11 */
300 /* these next 3 bytes come from bits 58-81 of the message header */
301 unsigned short replied_to:1; /* sent as 0 by the source */
302 unsigned short msg_type:3; /* software message type */
303 unsigned short canceled:1; /* sent as 0 by the source */
304 unsigned short unused1:3; /* not currently using */
307 unsigned char unused2a; /* not currently using */
309 unsigned char unused2; /* not currently using */
312 unsigned char sw_ack_vector; /* filled in by the hardware */
313 /* byte 15 (bits 127:120) */
315 unsigned short sequence; /* message sequence number */
317 unsigned char unused4[2]; /* not currently using bytes 18-19 */
320 int number_of_cpus; /* filled in at destination */
321 /* 32 bits, bytes 20-23 (aligned) */
323 unsigned char unused5[8]; /* not using */
328 struct bau_payload_queue_entry *msg;
331 struct bau_payload_queue_entry *va_queue_first;
332 struct bau_payload_queue_entry *va_queue_last;
340 * This structure is allocated per_cpu for UV TLB shootdown statistics.
343 /* sender statistics */
344 unsigned long s_giveup; /* number of fall backs to IPI-style flushes */
345 unsigned long s_requestor; /* number of shootdown requests */
346 unsigned long s_stimeout; /* source side timeouts */
347 unsigned long s_dtimeout; /* destination side timeouts */
348 unsigned long s_time; /* time spent in sending side */
349 unsigned long s_retriesok; /* successful retries */
350 unsigned long s_ntargcpu; /* total number of cpu's targeted */
351 unsigned long s_ntargself; /* times the sending cpu was targeted */
352 unsigned long s_ntarglocals; /* targets of cpus on the local blade */
353 unsigned long s_ntargremotes; /* targets of cpus on remote blades */
354 unsigned long s_ntarglocaluvhub; /* targets of the local hub */
355 unsigned long s_ntargremoteuvhub; /* remotes hubs targeted */
356 unsigned long s_ntarguvhub; /* total number of uvhubs targeted */
357 unsigned long s_ntarguvhub16; /* number of times target hubs >= 16*/
358 unsigned long s_ntarguvhub8; /* number of times target hubs >= 8 */
359 unsigned long s_ntarguvhub4; /* number of times target hubs >= 4 */
360 unsigned long s_ntarguvhub2; /* number of times target hubs >= 2 */
361 unsigned long s_ntarguvhub1; /* number of times target hubs == 1 */
362 unsigned long s_resets_plug; /* ipi-style resets from plug state */
363 unsigned long s_resets_timeout; /* ipi-style resets from timeouts */
364 unsigned long s_busy; /* status stayed busy past s/w timer */
365 unsigned long s_throttles; /* waits in throttle */
366 unsigned long s_retry_messages; /* retry broadcasts */
367 unsigned long s_bau_reenabled; /* for bau enable/disable */
368 unsigned long s_bau_disabled; /* for bau enable/disable */
369 /* destination statistics */
370 unsigned long d_alltlb; /* times all tlb's on this cpu were flushed */
371 unsigned long d_onetlb; /* times just one tlb on this cpu was flushed */
372 unsigned long d_multmsg; /* interrupts with multiple messages */
373 unsigned long d_nomsg; /* interrupts with no message */
374 unsigned long d_time; /* time spent on destination side */
375 unsigned long d_requestee; /* number of messages processed */
376 unsigned long d_retries; /* number of retry messages processed */
377 unsigned long d_canceled; /* number of messages canceled by retries */
378 unsigned long d_nocanceled; /* retries that found nothing to cancel */
379 unsigned long d_resets; /* number of ipi-style requests processed */
380 unsigned long d_rcanceled; /* number of messages canceled by resets */
383 struct hub_and_pnode {
388 * one per-cpu; to locate the software tables
391 struct bau_desc *descriptor_base;
392 struct bau_payload_queue_entry *va_queue_first;
393 struct bau_payload_queue_entry *va_queue_last;
394 struct bau_payload_queue_entry *bau_msg_head;
395 struct bau_control *uvhub_master;
396 struct bau_control *socket_master;
397 struct ptc_stats *statp;
398 unsigned long timeout_interval;
399 unsigned long set_bau_on_time;
400 atomic_t active_descriptor_count;
411 short cpus_in_socket;
413 short partition_base_pnode;
414 unsigned short message_number;
415 unsigned short uvhub_quiesce;
416 short socket_acknowledge_count[DEST_Q_SIZE];
417 cycles_t send_message;
418 spinlock_t uvhub_lock;
419 spinlock_t queue_lock;
421 int max_bau_concurrent;
422 int max_bau_concurrent_constant;
427 int complete_threshold;
428 int congested_response_us;
430 int congested_period;
431 cycles_t period_time;
432 long period_requests;
433 struct hub_and_pnode *target_hub_and_pnode;
436 static inline int bau_uvhub_isset(int uvhub, struct bau_target_uvhubmask *dstp)
438 return constant_test_bit(uvhub, &dstp->bits[0]);
440 static inline void bau_uvhub_set(int pnode, struct bau_target_uvhubmask *dstp)
442 __set_bit(pnode, &dstp->bits[0]);
444 static inline void bau_uvhubs_clear(struct bau_target_uvhubmask *dstp,
447 bitmap_zero(&dstp->bits[0], nbits);
449 static inline int bau_uvhub_weight(struct bau_target_uvhubmask *dstp)
451 return bitmap_weight((unsigned long *)&dstp->bits[0],
452 UV_DISTRIBUTION_SIZE);
455 static inline void bau_cpubits_clear(struct bau_local_cpumask *dstp, int nbits)
457 bitmap_zero(&dstp->bits, nbits);
460 #define cpubit_isset(cpu, bau_local_cpumask) \
461 test_bit((cpu), (bau_local_cpumask).bits)
463 extern void uv_bau_message_intr1(void);
464 extern void uv_bau_timeout_intr1(void);
466 struct atomic_short {
471 * atomic_read_short - read a short atomic variable
472 * @v: pointer of type atomic_short
474 * Atomically reads the value of @v.
476 static inline int atomic_read_short(const struct atomic_short *v)
482 * atomic_add_short_return - add and return a short int
483 * @i: short value to add
484 * @v: pointer of type atomic_short
486 * Atomically adds @i to @v and returns @i + @v
488 static inline int atomic_add_short_return(short i, struct atomic_short *v)
491 asm volatile(LOCK_PREFIX "xaddw %0, %1"
492 : "+r" (i), "+m" (v->counter)
497 #endif /* _ASM_X86_UV_UV_BAU_H */