linux/arch/x86/include/asm/uv/uv_bau.h
<<
>>
Prefs
   1/*
   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
   4 * for more details.
   5 *
   6 * SGI UV Broadcast Assist Unit definitions
   7 *
   8 * Copyright (C) 2008-2011 Silicon Graphics, Inc. All rights reserved.
   9 */
  10
  11#ifndef _ASM_X86_UV_UV_BAU_H
  12#define _ASM_X86_UV_UV_BAU_H
  13
  14#include <linux/bitmap.h>
  15#define BITSPERBYTE 8
  16
  17/*
  18 * Broadcast Assist Unit messaging structures
  19 *
  20 * Selective Broadcast activations are induced by software action
  21 * specifying a particular 8-descriptor "set" via a 6-bit index written
  22 * to an MMR.
  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.
  28 *
  29 * We will use one set for sending BAU messages from each of the
  30 * cpu's on the uvhub.
  31 *
  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).
  34 */
  35
  36#define MAX_CPUS_PER_UVHUB              128
  37#define MAX_CPUS_PER_SOCKET             64
  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_PAYLOADQ_GNODE_SHIFT         49
  52#define UV_PTC_BASENAME                 "sgi_uv/ptc_statistics"
  53#define UV_BAU_BASENAME                 "sgi_uv/bau_tunables"
  54#define UV_BAU_TUNABLES_DIR             "sgi_uv"
  55#define UV_BAU_TUNABLES_FILE            "bau_tunables"
  56#define WHITESPACE                      " \t\n"
  57#define cpubit_isset(cpu, bau_local_cpumask) \
  58        test_bit((cpu), (bau_local_cpumask).bits)
  59
  60/* [19:16] SOFT_ACK timeout period  19: 1 is urgency 7  17:16 1 is multiplier */
  61/*
  62 * UV2: Bit 19 selects between
  63 *  (0): 10 microsecond timebase and
  64 *  (1): 80 microseconds
  65 *  we're using 560us, similar to UV1: 65 units of 10us
  66 */
  67#define UV1_INTD_SOFT_ACK_TIMEOUT_PERIOD (9UL)
  68#define UV2_INTD_SOFT_ACK_TIMEOUT_PERIOD (15UL)
  69
  70#define UV_INTD_SOFT_ACK_TIMEOUT_PERIOD (is_uv1_hub() ?                 \
  71                UV1_INTD_SOFT_ACK_TIMEOUT_PERIOD :                      \
  72                UV2_INTD_SOFT_ACK_TIMEOUT_PERIOD)
  73/* assuming UV3 is the same */
  74
  75#define BAU_MISC_CONTROL_MULT_MASK      3
  76
  77#define UVH_AGING_PRESCALE_SEL          0x000000b000UL
  78/* [30:28] URGENCY_7  an index into a table of times */
  79#define BAU_URGENCY_7_SHIFT             28
  80#define BAU_URGENCY_7_MASK              7
  81
  82#define UVH_TRANSACTION_TIMEOUT         0x000000b200UL
  83/* [45:40] BAU - BAU transaction timeout select - a multiplier */
  84#define BAU_TRANS_SHIFT                 40
  85#define BAU_TRANS_MASK                  0x3f
  86
  87/*
  88 * shorten some awkward names
  89 */
  90#define AS_PUSH_SHIFT UVH_LB_BAU_SB_ACTIVATION_CONTROL_PUSH_SHFT
  91#define SOFTACK_MSHIFT UVH_LB_BAU_MISC_CONTROL_ENABLE_INTD_SOFT_ACK_MODE_SHFT
  92#define SOFTACK_PSHIFT UVH_LB_BAU_MISC_CONTROL_INTD_SOFT_ACK_TIMEOUT_PERIOD_SHFT
  93#define SOFTACK_TIMEOUT_PERIOD UV_INTD_SOFT_ACK_TIMEOUT_PERIOD
  94#define PREFETCH_HINT_SHFT UV3H_LB_BAU_MISC_CONTROL_ENABLE_INTD_PREFETCH_HINT_SHFT
  95#define SB_STATUS_SHFT UV3H_LB_BAU_MISC_CONTROL_ENABLE_EXTENDED_SB_STATUS_SHFT
  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
 100
 101/*
 102 * bits in UVH_LB_BAU_SB_ACTIVATION_STATUS_0/1
 103 */
 104#define DS_IDLE                         0
 105#define DS_ACTIVE                       1
 106#define DS_DESTINATION_TIMEOUT          2
 107#define DS_SOURCE_TIMEOUT               3
 108/*
 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 * -------------------------------     ----   -----   -------
 113 * IDLE                                 0       0        0
 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
 119 */
 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
 126
 127/*
 128 * delay for 'plugged' timeout retries, in microseconds
 129 */
 130#define PLUGGED_DELAY                   10
 131
 132/*
 133 * threshholds at which to use IPI to free resources
 134 */
 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
 143/* after this # of giveups (fall back to kernel IPI's) disable the use of
 144   the BAU for a period of time */
 145#define GIVEUP_LIMIT                    100
 146
 147#define UV_LB_SUBNODEID                 0x10
 148
 149/* these two are the same for UV1 and UV2: */
 150#define UV_SA_SHFT UVH_LB_BAU_MISC_CONTROL_INTD_SOFT_ACK_TIMEOUT_PERIOD_SHFT
 151#define UV_SA_MASK UVH_LB_BAU_MISC_CONTROL_INTD_SOFT_ACK_TIMEOUT_PERIOD_MASK
 152/* 4 bits of software ack period */
 153#define UV2_ACK_MASK                    0x7UL
 154#define UV2_ACK_UNITS_SHFT              3
 155#define UV2_EXT_SHFT UV2H_LB_BAU_MISC_CONTROL_ENABLE_EXTENDED_SB_STATUS_SHFT
 156
 157/*
 158 * number of entries in the destination side payload queue
 159 */
 160#define DEST_Q_SIZE                     20
 161/*
 162 * number of destination side software ack resources
 163 */
 164#define DEST_NUM_RESOURCES              8
 165/*
 166 * completion statuses for sending a TLB flush message
 167 */
 168#define FLUSH_RETRY_PLUGGED             1
 169#define FLUSH_RETRY_TIMEOUT             2
 170#define FLUSH_GIVEUP                    3
 171#define FLUSH_COMPLETE                  4
 172
 173/*
 174 * tuning the action when the numalink network is extremely delayed
 175 */
 176#define CONGESTED_RESPONSE_US           1000    /* 'long' response time, in
 177                                                   microseconds */
 178#define CONGESTED_REPS                  10      /* long delays averaged over
 179                                                   this many broadcasts */
 180#define DISABLED_PERIOD                 10      /* time for the bau to be
 181                                                   disabled, in seconds */
 182/* see msg_type: */
 183#define MSG_NOOP                        0
 184#define MSG_REGULAR                     1
 185#define MSG_RETRY                       2
 186
 187#define BAU_DESC_QUALIFIER              0x534749
 188
 189enum uv_bau_version {
 190        UV_BAU_V1 = 1,
 191        UV_BAU_V2,
 192        UV_BAU_V3,
 193        UV_BAU_V4,
 194};
 195
 196/*
 197 * Distribution: 32 bytes (256 bits) (bytes 0-0x1f of descriptor)
 198 * If the 'multilevel' flag in the header portion of the descriptor
 199 * has been set to 0, then endpoint multi-unicast mode is selected.
 200 * The distribution specification (32 bytes) is interpreted as a 256-bit
 201 * distribution vector. Adjacent bits correspond to consecutive even numbered
 202 * nodeIDs. The result of adding the index of a given bit to the 15-bit
 203 * 'base_dest_nasid' field of the header corresponds to the
 204 * destination nodeID associated with that specified bit.
 205 */
 206struct pnmask {
 207        unsigned long           bits[BITS_TO_LONGS(UV_DISTRIBUTION_SIZE)];
 208};
 209
 210/*
 211 * mask of cpu's on a uvhub
 212 * (during initialization we need to check that unsigned long has
 213 *  enough bits for max. cpu's per uvhub)
 214 */
 215struct bau_local_cpumask {
 216        unsigned long           bits;
 217};
 218
 219/*
 220 * Payload: 16 bytes (128 bits) (bytes 0x20-0x2f of descriptor)
 221 * only 12 bytes (96 bits) of the payload area are usable.
 222 * An additional 3 bytes (bits 27:4) of the header address are carried
 223 * to the next bytes of the destination payload queue.
 224 * And an additional 2 bytes of the header Suppl_A field are also
 225 * carried to the destination payload queue.
 226 * But the first byte of the Suppl_A becomes bits 127:120 (the 16th byte)
 227 * of the destination payload queue, which is written by the hardware
 228 * with the s/w ack resource bit vector.
 229 * [ effective message contents (16 bytes (128 bits) maximum), not counting
 230 *   the s/w ack bit vector  ]
 231 */
 232
 233/**
 234 * struct uv1_2_3_bau_msg_payload - defines payload for INTD transactions
 235 * @address:            Signifies a page or all TLB's of the cpu
 236 * @sending_cpu:        CPU from which the message originates
 237 * @acknowledge_count:  CPUs on the destination Hub that received the interrupt
 238 */
 239struct uv1_2_3_bau_msg_payload {
 240        u64 address;
 241        u16 sending_cpu;
 242        u16 acknowledge_count;
 243};
 244
 245/**
 246 * struct uv4_bau_msg_payload - defines payload for INTD transactions
 247 * @address:            Signifies a page or all TLB's of the cpu
 248 * @sending_cpu:        CPU from which the message originates
 249 * @acknowledge_count:  CPUs on the destination Hub that received the interrupt
 250 * @qualifier:          Set by source to verify origin of INTD broadcast
 251 */
 252struct uv4_bau_msg_payload {
 253        u64 address;
 254        u16 sending_cpu;
 255        u16 acknowledge_count;
 256        u32 reserved:8;
 257        u32 qualifier:24;
 258};
 259
 260/*
 261 * UV1 Message header:  16 bytes (128 bits) (bytes 0x30-0x3f of descriptor)
 262 * see table 4.2.3.0.1 in broacast_assist spec.
 263 */
 264struct uv1_bau_msg_header {
 265        unsigned int    dest_subnodeid:6;       /* must be 0x10, for the LB */
 266        /* bits 5:0 */
 267        unsigned int    base_dest_nasid:15;     /* nasid of the first bit */
 268        /* bits 20:6 */                         /* in uvhub map */
 269        unsigned int    command:8;              /* message type */
 270        /* bits 28:21 */
 271        /* 0x38: SN3net EndPoint Message */
 272        unsigned int    rsvd_1:3;               /* must be zero */
 273        /* bits 31:29 */
 274        /* int will align on 32 bits */
 275        unsigned int    rsvd_2:9;               /* must be zero */
 276        /* bits 40:32 */
 277        /* Suppl_A is 56-41 */
 278        unsigned int    sequence:16;            /* message sequence number */
 279        /* bits 56:41 */                        /* becomes bytes 16-17 of msg */
 280                                                /* Address field (96:57) is
 281                                                   never used as an address
 282                                                   (these are address bits
 283                                                   42:3) */
 284
 285        unsigned int    rsvd_3:1;               /* must be zero */
 286        /* bit 57 */
 287        /* address bits 27:4 are payload */
 288        /* these next 24  (58-81) bits become bytes 12-14 of msg */
 289        /* bits 65:58 land in byte 12 */
 290        unsigned int    replied_to:1;           /* sent as 0 by the source to
 291                                                   byte 12 */
 292        /* bit 58 */
 293        unsigned int    msg_type:3;             /* software type of the
 294                                                   message */
 295        /* bits 61:59 */
 296        unsigned int    canceled:1;             /* message canceled, resource
 297                                                   is to be freed*/
 298        /* bit 62 */
 299        unsigned int    payload_1a:1;           /* not currently used */
 300        /* bit 63 */
 301        unsigned int    payload_1b:2;           /* not currently used */
 302        /* bits 65:64 */
 303
 304        /* bits 73:66 land in byte 13 */
 305        unsigned int    payload_1ca:6;          /* not currently used */
 306        /* bits 71:66 */
 307        unsigned int    payload_1c:2;           /* not currently used */
 308        /* bits 73:72 */
 309
 310        /* bits 81:74 land in byte 14 */
 311        unsigned int    payload_1d:6;           /* not currently used */
 312        /* bits 79:74 */
 313        unsigned int    payload_1e:2;           /* not currently used */
 314        /* bits 81:80 */
 315
 316        unsigned int    rsvd_4:7;               /* must be zero */
 317        /* bits 88:82 */
 318        unsigned int    swack_flag:1;           /* software acknowledge flag */
 319        /* bit 89 */
 320                                                /* INTD trasactions at
 321                                                   destination are to wait for
 322                                                   software acknowledge */
 323        unsigned int    rsvd_5:6;               /* must be zero */
 324        /* bits 95:90 */
 325        unsigned int    rsvd_6:5;               /* must be zero */
 326        /* bits 100:96 */
 327        unsigned int    int_both:1;             /* if 1, interrupt both sockets
 328                                                   on the uvhub */
 329        /* bit 101*/
 330        unsigned int    fairness:3;             /* usually zero */
 331        /* bits 104:102 */
 332        unsigned int    multilevel:1;           /* multi-level multicast
 333                                                   format */
 334        /* bit 105 */
 335        /* 0 for TLB: endpoint multi-unicast messages */
 336        unsigned int    chaining:1;             /* next descriptor is part of
 337                                                   this activation*/
 338        /* bit 106 */
 339        unsigned int    rsvd_7:21;              /* must be zero */
 340        /* bits 127:107 */
 341};
 342
 343/*
 344 * UV2 Message header:  16 bytes (128 bits) (bytes 0x30-0x3f of descriptor)
 345 * see figure 9-2 of harp_sys.pdf
 346 * assuming UV3 is the same
 347 */
 348struct uv2_3_bau_msg_header {
 349        unsigned int    base_dest_nasid:15;     /* nasid of the first bit */
 350        /* bits 14:0 */                         /* in uvhub map */
 351        unsigned int    dest_subnodeid:5;       /* must be 0x10, for the LB */
 352        /* bits 19:15 */
 353        unsigned int    rsvd_1:1;               /* must be zero */
 354        /* bit 20 */
 355        /* Address bits 59:21 */
 356        /* bits 25:2 of address (44:21) are payload */
 357        /* these next 24 bits become bytes 12-14 of msg */
 358        /* bits 28:21 land in byte 12 */
 359        unsigned int    replied_to:1;           /* sent as 0 by the source to
 360                                                   byte 12 */
 361        /* bit 21 */
 362        unsigned int    msg_type:3;             /* software type of the
 363                                                   message */
 364        /* bits 24:22 */
 365        unsigned int    canceled:1;             /* message canceled, resource
 366                                                   is to be freed*/
 367        /* bit 25 */
 368        unsigned int    payload_1:3;            /* not currently used */
 369        /* bits 28:26 */
 370
 371        /* bits 36:29 land in byte 13 */
 372        unsigned int    payload_2a:3;           /* not currently used */
 373        unsigned int    payload_2b:5;           /* not currently used */
 374        /* bits 36:29 */
 375
 376        /* bits 44:37 land in byte 14 */
 377        unsigned int    payload_3:8;            /* not currently used */
 378        /* bits 44:37 */
 379
 380        unsigned int    rsvd_2:7;               /* reserved */
 381        /* bits 51:45 */
 382        unsigned int    swack_flag:1;           /* software acknowledge flag */
 383        /* bit 52 */
 384        unsigned int    rsvd_3a:3;              /* must be zero */
 385        unsigned int    rsvd_3b:8;              /* must be zero */
 386        unsigned int    rsvd_3c:8;              /* must be zero */
 387        unsigned int    rsvd_3d:3;              /* must be zero */
 388        /* bits 74:53 */
 389        unsigned int    fairness:3;             /* usually zero */
 390        /* bits 77:75 */
 391
 392        unsigned int    sequence:16;            /* message sequence number */
 393        /* bits 93:78  Suppl_A  */
 394        unsigned int    chaining:1;             /* next descriptor is part of
 395                                                   this activation*/
 396        /* bit 94 */
 397        unsigned int    multilevel:1;           /* multi-level multicast
 398                                                   format */
 399        /* bit 95 */
 400        unsigned int    rsvd_4:24;              /* ordered / source node /
 401                                                   source subnode / aging
 402                                                   must be zero */
 403        /* bits 119:96 */
 404        unsigned int    command:8;              /* message type */
 405        /* bits 127:120 */
 406};
 407
 408/*
 409 * The activation descriptor:
 410 * The format of the message to send, plus all accompanying control
 411 * Should be 64 bytes
 412 */
 413struct bau_desc {
 414        struct pnmask                           distribution;
 415        /*
 416         * message template, consisting of header and payload:
 417         */
 418        union bau_msg_header {
 419                struct uv1_bau_msg_header       uv1_hdr;
 420                struct uv2_3_bau_msg_header     uv2_3_hdr;
 421        } header;
 422
 423        union bau_payload_header {
 424                struct uv1_2_3_bau_msg_payload  uv1_2_3;
 425                struct uv4_bau_msg_payload      uv4;
 426        } payload;
 427};
 428/* UV1:
 429 *   -payload--    ---------header------
 430 *   bytes 0-11    bits 41-56  bits 58-81
 431 *       A           B  (2)      C (3)
 432 *
 433 *            A/B/C are moved to:
 434 *       A            C          B
 435 *   bytes 0-11  bytes 12-14  bytes 16-17  (byte 15 filled in by hw as vector)
 436 *   ------------payload queue-----------
 437 */
 438/* UV2:
 439 *   -payload--    ---------header------
 440 *   bytes 0-11    bits 70-78  bits 21-44
 441 *       A           B  (2)      C (3)
 442 *
 443 *            A/B/C are moved to:
 444 *       A            C          B
 445 *   bytes 0-11  bytes 12-14  bytes 16-17  (byte 15 filled in by hw as vector)
 446 *   ------------payload queue-----------
 447 */
 448
 449/*
 450 * The payload queue on the destination side is an array of these.
 451 * With BAU_MISC_CONTROL set for software acknowledge mode, the messages
 452 * are 32 bytes (2 micropackets) (256 bits) in length, but contain only 17
 453 * bytes of usable data, including the sw ack vector in byte 15 (bits 127:120)
 454 * (12 bytes come from bau_msg_payload, 3 from payload_1, 2 from
 455 *  swack_vec and payload_2)
 456 * "Enabling Software Acknowledgment mode (see Section 4.3.3 Software
 457 *  Acknowledge Processing) also selects 32 byte (17 bytes usable) payload
 458 *  operation."
 459 */
 460struct bau_pq_entry {
 461        unsigned long   address;        /* signifies a page or all TLB's
 462                                           of the cpu */
 463        /* 64 bits, bytes 0-7 */
 464        unsigned short  sending_cpu;    /* cpu that sent the message */
 465        /* 16 bits, bytes 8-9 */
 466        unsigned short  acknowledge_count; /* filled in by destination */
 467        /* 16 bits, bytes 10-11 */
 468        /* these next 3 bytes come from bits 58-81 of the message header */
 469        unsigned short  replied_to:1;   /* sent as 0 by the source */
 470        unsigned short  msg_type:3;     /* software message type */
 471        unsigned short  canceled:1;     /* sent as 0 by the source */
 472        unsigned short  unused1:3;      /* not currently using */
 473        /* byte 12 */
 474        unsigned char   unused2a;       /* not currently using */
 475        /* byte 13 */
 476        unsigned char   unused2;        /* not currently using */
 477        /* byte 14 */
 478        unsigned char   swack_vec;      /* filled in by the hardware */
 479        /* byte 15 (bits 127:120) */
 480        unsigned short  sequence;       /* message sequence number */
 481        /* bytes 16-17 */
 482        unsigned char   unused4[2];     /* not currently using bytes 18-19 */
 483        /* bytes 18-19 */
 484        int             number_of_cpus; /* filled in at destination */
 485        /* 32 bits, bytes 20-23 (aligned) */
 486        unsigned char   unused5[8];     /* not using */
 487        /* bytes 24-31 */
 488};
 489
 490struct msg_desc {
 491        struct bau_pq_entry     *msg;
 492        int                     msg_slot;
 493        struct bau_pq_entry     *queue_first;
 494        struct bau_pq_entry     *queue_last;
 495};
 496
 497struct reset_args {
 498        int                     sender;
 499};
 500
 501/*
 502 * This structure is allocated per_cpu for UV TLB shootdown statistics.
 503 */
 504struct ptc_stats {
 505        /* sender statistics */
 506        unsigned long   s_giveup;               /* number of fall backs to
 507                                                   IPI-style flushes */
 508        unsigned long   s_requestor;            /* number of shootdown
 509                                                   requests */
 510        unsigned long   s_stimeout;             /* source side timeouts */
 511        unsigned long   s_dtimeout;             /* destination side timeouts */
 512        unsigned long   s_strongnacks;          /* number of strong nack's */
 513        unsigned long   s_time;                 /* time spent in sending side */
 514        unsigned long   s_retriesok;            /* successful retries */
 515        unsigned long   s_ntargcpu;             /* total number of cpu's
 516                                                   targeted */
 517        unsigned long   s_ntargself;            /* times the sending cpu was
 518                                                   targeted */
 519        unsigned long   s_ntarglocals;          /* targets of cpus on the local
 520                                                   blade */
 521        unsigned long   s_ntargremotes;         /* targets of cpus on remote
 522                                                   blades */
 523        unsigned long   s_ntarglocaluvhub;      /* targets of the local hub */
 524        unsigned long   s_ntargremoteuvhub;     /* remotes hubs targeted */
 525        unsigned long   s_ntarguvhub;           /* total number of uvhubs
 526                                                   targeted */
 527        unsigned long   s_ntarguvhub16;         /* number of times target
 528                                                   hubs >= 16*/
 529        unsigned long   s_ntarguvhub8;          /* number of times target
 530                                                   hubs >= 8 */
 531        unsigned long   s_ntarguvhub4;          /* number of times target
 532                                                   hubs >= 4 */
 533        unsigned long   s_ntarguvhub2;          /* number of times target
 534                                                   hubs >= 2 */
 535        unsigned long   s_ntarguvhub1;          /* number of times target
 536                                                   hubs == 1 */
 537        unsigned long   s_resets_plug;          /* ipi-style resets from plug
 538                                                   state */
 539        unsigned long   s_resets_timeout;       /* ipi-style resets from
 540                                                   timeouts */
 541        unsigned long   s_busy;                 /* status stayed busy past
 542                                                   s/w timer */
 543        unsigned long   s_throttles;            /* waits in throttle */
 544        unsigned long   s_retry_messages;       /* retry broadcasts */
 545        unsigned long   s_bau_reenabled;        /* for bau enable/disable */
 546        unsigned long   s_bau_disabled;         /* for bau enable/disable */
 547        unsigned long   s_uv2_wars;             /* uv2 workaround, perm. busy */
 548        unsigned long   s_uv2_wars_hw;          /* uv2 workaround, hiwater */
 549        unsigned long   s_uv2_war_waits;        /* uv2 workaround, long waits */
 550        unsigned long   s_overipilimit;         /* over the ipi reset limit */
 551        unsigned long   s_giveuplimit;          /* disables, over giveup limit*/
 552        unsigned long   s_enters;               /* entries to the driver */
 553        unsigned long   s_ipifordisabled;       /* fall back to IPI; disabled */
 554        unsigned long   s_plugged;              /* plugged by h/w bug*/
 555        unsigned long   s_congested;            /* giveup on long wait */
 556        /* destination statistics */
 557        unsigned long   d_alltlb;               /* times all tlb's on this
 558                                                   cpu were flushed */
 559        unsigned long   d_onetlb;               /* times just one tlb on this
 560                                                   cpu was flushed */
 561        unsigned long   d_multmsg;              /* interrupts with multiple
 562                                                   messages */
 563        unsigned long   d_nomsg;                /* interrupts with no message */
 564        unsigned long   d_time;                 /* time spent on destination
 565                                                   side */
 566        unsigned long   d_requestee;            /* number of messages
 567                                                   processed */
 568        unsigned long   d_retries;              /* number of retry messages
 569                                                   processed */
 570        unsigned long   d_canceled;             /* number of messages canceled
 571                                                   by retries */
 572        unsigned long   d_nocanceled;           /* retries that found nothing
 573                                                   to cancel */
 574        unsigned long   d_resets;               /* number of ipi-style requests
 575                                                   processed */
 576        unsigned long   d_rcanceled;            /* number of messages canceled
 577                                                   by resets */
 578};
 579
 580struct tunables {
 581        int                     *tunp;
 582        int                     deflt;
 583};
 584
 585struct hub_and_pnode {
 586        short                   uvhub;
 587        short                   pnode;
 588};
 589
 590struct socket_desc {
 591        short                   num_cpus;
 592        short                   cpu_number[MAX_CPUS_PER_SOCKET];
 593};
 594
 595struct uvhub_desc {
 596        unsigned short          socket_mask;
 597        short                   num_cpus;
 598        short                   uvhub;
 599        short                   pnode;
 600        struct socket_desc      socket[2];
 601};
 602
 603/**
 604 * struct bau_control
 605 * @status_mmr: location of status mmr, determined by uvhub_cpu
 606 * @status_index: index of ERR|BUSY bits in status mmr, determined by uvhub_cpu
 607 *
 608 * Per-cpu control struct containing CPU topology information and BAU tuneables.
 609 */
 610struct bau_control {
 611        struct bau_desc         *descriptor_base;
 612        struct bau_pq_entry     *queue_first;
 613        struct bau_pq_entry     *queue_last;
 614        struct bau_pq_entry     *bau_msg_head;
 615        struct bau_control      *uvhub_master;
 616        struct bau_control      *socket_master;
 617        struct ptc_stats        *statp;
 618        cpumask_t               *cpumask;
 619        unsigned long           timeout_interval;
 620        unsigned long           set_bau_on_time;
 621        atomic_t                active_descriptor_count;
 622        int                     plugged_tries;
 623        int                     timeout_tries;
 624        int                     ipi_attempts;
 625        int                     conseccompletes;
 626        u64                     status_mmr;
 627        int                     status_index;
 628        bool                    nobau;
 629        short                   baudisabled;
 630        short                   cpu;
 631        short                   osnode;
 632        short                   uvhub_cpu;
 633        short                   uvhub;
 634        short                   uvhub_version;
 635        short                   cpus_in_socket;
 636        short                   cpus_in_uvhub;
 637        short                   partition_base_pnode;
 638        short                   busy;       /* all were busy (war) */
 639        unsigned short          message_number;
 640        unsigned short          uvhub_quiesce;
 641        short                   socket_acknowledge_count[DEST_Q_SIZE];
 642        cycles_t                send_message;
 643        cycles_t                period_end;
 644        cycles_t                period_time;
 645        spinlock_t              uvhub_lock;
 646        spinlock_t              queue_lock;
 647        spinlock_t              disable_lock;
 648        /* tunables */
 649        int                     max_concurr;
 650        int                     max_concurr_const;
 651        int                     plugged_delay;
 652        int                     plugsb4reset;
 653        int                     timeoutsb4reset;
 654        int                     ipi_reset_limit;
 655        int                     complete_threshold;
 656        int                     cong_response_us;
 657        int                     cong_reps;
 658        cycles_t                disabled_period;
 659        int                     period_giveups;
 660        int                     giveup_limit;
 661        long                    period_requests;
 662        struct hub_and_pnode    *thp;
 663};
 664
 665/* Abstracted BAU functions */
 666struct bau_operations {
 667        unsigned long   (*read_l_sw_ack)(void);
 668        unsigned long   (*read_g_sw_ack)(int pnode);
 669        unsigned long   (*bau_gpa_to_offset)(unsigned long vaddr);
 670        void            (*write_l_sw_ack)(unsigned long mmr);
 671        void            (*write_g_sw_ack)(int pnode, unsigned long mmr);
 672        void            (*write_payload_first)(int pnode, unsigned long mmr);
 673        void            (*write_payload_last)(int pnode, unsigned long mmr);
 674        int             (*wait_completion)(struct bau_desc*,
 675                                struct bau_control*, long try);
 676};
 677
 678static inline void write_mmr_data_broadcast(int pnode, unsigned long mmr_image)
 679{
 680        write_gmmr(pnode, UVH_BAU_DATA_BROADCAST, mmr_image);
 681}
 682
 683static inline void write_mmr_descriptor_base(int pnode, unsigned long mmr_image)
 684{
 685        write_gmmr(pnode, UVH_LB_BAU_SB_DESCRIPTOR_BASE, mmr_image);
 686}
 687
 688static inline void write_mmr_activation(unsigned long index)
 689{
 690        write_lmmr(UVH_LB_BAU_SB_ACTIVATION_CONTROL, index);
 691}
 692
 693static inline void write_gmmr_activation(int pnode, unsigned long mmr_image)
 694{
 695        write_gmmr(pnode, UVH_LB_BAU_SB_ACTIVATION_CONTROL, mmr_image);
 696}
 697
 698static inline void write_mmr_proc_payload_first(int pnode, unsigned long mmr_image)
 699{
 700        write_gmmr(pnode, UV4H_LB_PROC_INTD_QUEUE_FIRST, mmr_image);
 701}
 702
 703static inline void write_mmr_proc_payload_last(int pnode, unsigned long mmr_image)
 704{
 705        write_gmmr(pnode, UV4H_LB_PROC_INTD_QUEUE_LAST, mmr_image);
 706}
 707
 708static inline void write_mmr_payload_first(int pnode, unsigned long mmr_image)
 709{
 710        write_gmmr(pnode, UVH_LB_BAU_INTD_PAYLOAD_QUEUE_FIRST, mmr_image);
 711}
 712
 713static inline void write_mmr_payload_tail(int pnode, unsigned long mmr_image)
 714{
 715        write_gmmr(pnode, UVH_LB_BAU_INTD_PAYLOAD_QUEUE_TAIL, mmr_image);
 716}
 717
 718static inline void write_mmr_payload_last(int pnode, unsigned long mmr_image)
 719{
 720        write_gmmr(pnode, UVH_LB_BAU_INTD_PAYLOAD_QUEUE_LAST, mmr_image);
 721}
 722
 723static inline void write_mmr_misc_control(int pnode, unsigned long mmr_image)
 724{
 725        write_gmmr(pnode, UVH_LB_BAU_MISC_CONTROL, mmr_image);
 726}
 727
 728static inline unsigned long read_mmr_misc_control(int pnode)
 729{
 730        return read_gmmr(pnode, UVH_LB_BAU_MISC_CONTROL);
 731}
 732
 733static inline void write_mmr_sw_ack(unsigned long mr)
 734{
 735        uv_write_local_mmr(UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE_ALIAS, mr);
 736}
 737
 738static inline void write_gmmr_sw_ack(int pnode, unsigned long mr)
 739{
 740        write_gmmr(pnode, UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE_ALIAS, mr);
 741}
 742
 743static inline unsigned long read_mmr_sw_ack(void)
 744{
 745        return read_lmmr(UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE);
 746}
 747
 748static inline unsigned long read_gmmr_sw_ack(int pnode)
 749{
 750        return read_gmmr(pnode, UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE);
 751}
 752
 753static inline void write_mmr_proc_sw_ack(unsigned long mr)
 754{
 755        uv_write_local_mmr(UV4H_LB_PROC_INTD_SOFT_ACK_CLEAR, mr);
 756}
 757
 758static inline void write_gmmr_proc_sw_ack(int pnode, unsigned long mr)
 759{
 760        write_gmmr(pnode, UV4H_LB_PROC_INTD_SOFT_ACK_CLEAR, mr);
 761}
 762
 763static inline unsigned long read_mmr_proc_sw_ack(void)
 764{
 765        return read_lmmr(UV4H_LB_PROC_INTD_SOFT_ACK_PENDING);
 766}
 767
 768static inline unsigned long read_gmmr_proc_sw_ack(int pnode)
 769{
 770        return read_gmmr(pnode, UV4H_LB_PROC_INTD_SOFT_ACK_PENDING);
 771}
 772
 773static inline void write_mmr_data_config(int pnode, unsigned long mr)
 774{
 775        uv_write_global_mmr64(pnode, UVH_BAU_DATA_CONFIG, mr);
 776}
 777
 778static inline int bau_uvhub_isset(int uvhub, struct pnmask *dstp)
 779{
 780        return constant_test_bit(uvhub, &dstp->bits[0]);
 781}
 782static inline void bau_uvhub_set(int pnode, struct pnmask *dstp)
 783{
 784        __set_bit(pnode, &dstp->bits[0]);
 785}
 786static inline void bau_uvhubs_clear(struct pnmask *dstp,
 787                                    int nbits)
 788{
 789        bitmap_zero(&dstp->bits[0], nbits);
 790}
 791static inline int bau_uvhub_weight(struct pnmask *dstp)
 792{
 793        return bitmap_weight((unsigned long *)&dstp->bits[0],
 794                                UV_DISTRIBUTION_SIZE);
 795}
 796
 797static inline void bau_cpubits_clear(struct bau_local_cpumask *dstp, int nbits)
 798{
 799        bitmap_zero(&dstp->bits, nbits);
 800}
 801
 802extern void uv_bau_message_intr1(void);
 803#ifdef CONFIG_TRACING
 804#define trace_uv_bau_message_intr1 uv_bau_message_intr1
 805#endif
 806extern void uv_bau_timeout_intr1(void);
 807
 808struct atomic_short {
 809        short counter;
 810};
 811
 812/*
 813 * atomic_read_short - read a short atomic variable
 814 * @v: pointer of type atomic_short
 815 *
 816 * Atomically reads the value of @v.
 817 */
 818static inline int atomic_read_short(const struct atomic_short *v)
 819{
 820        return v->counter;
 821}
 822
 823/*
 824 * atom_asr - add and return a short int
 825 * @i: short value to add
 826 * @v: pointer of type atomic_short
 827 *
 828 * Atomically adds @i to @v and returns @i + @v
 829 */
 830static inline int atom_asr(short i, struct atomic_short *v)
 831{
 832        short __i = i;
 833        asm volatile(LOCK_PREFIX "xaddw %0, %1"
 834                        : "+r" (i), "+m" (v->counter)
 835                        : : "memory");
 836        return i + __i;
 837}
 838
 839/*
 840 * conditionally add 1 to *v, unless *v is >= u
 841 * return 0 if we cannot add 1 to *v because it is >= u
 842 * return 1 if we can add 1 to *v because it is < u
 843 * the add is atomic
 844 *
 845 * This is close to atomic_add_unless(), but this allows the 'u' value
 846 * to be lowered below the current 'v'.  atomic_add_unless can only stop
 847 * on equal.
 848 */
 849static inline int atomic_inc_unless_ge(spinlock_t *lock, atomic_t *v, int u)
 850{
 851        spin_lock(lock);
 852        if (atomic_read(v) >= u) {
 853                spin_unlock(lock);
 854                return 0;
 855        }
 856        atomic_inc(v);
 857        spin_unlock(lock);
 858        return 1;
 859}
 860
 861#endif /* _ASM_X86_UV_UV_BAU_H */
 862