arch/x86/include/asm/uv/uv_bau.h

   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 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              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 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_BASE_PNODE_SHIFT        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
  60
  61 /* [19:16] SOFT_ACK timeout period  19: 1 is urgency 7  17:16 1 is multiplier */
  62 /*
  63  * UV2: Bit 19 selects between
  64  *  (0): 10 microsecond timebase and
  65  *  (1): 80 microseconds
  66  *  we're using 655us, similar to UV1: 65 units of 10us
  67  */
  68 #define UV1_INTD_SOFT_ACK_TIMEOUT_PERIOD (9UL)
  69 #define UV2_INTD_SOFT_ACK_TIMEOUT_PERIOD (65*10UL)
  70
  71 #define UV_INTD_SOFT_ACK_TIMEOUT_PERIOD (is_uv1_hub() ?                 \
  72                 UV1_INTD_SOFT_ACK_TIMEOUT_PERIOD :                      \
  73                 UV2_INTD_SOFT_ACK_TIMEOUT_PERIOD)
  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  * bits in UVH_LB_BAU_SB_ACTIVATION_STATUS_0/1
  89  */
  90 #define DESC_STATUS_IDLE                0
  91 #define DESC_STATUS_ACTIVE              1
  92 #define DESC_STATUS_DESTINATION_TIMEOUT 2
  93 #define DESC_STATUS_SOURCE_TIMEOUT      3
  94 /*
  95  * bits put together from HRP_LB_BAU_SB_ACTIVATION_STATUS_0/1/2
  96  * values 1 and 5 will not occur
  97  */
  98 #define UV2H_DESC_IDLE                  0
  99 #define UV2H_DESC_DEST_TIMEOUT          2
 100 #define UV2H_DESC_DEST_STRONG_NACK      3
 101 #define UV2H_DESC_BUSY                  4
 102 #define UV2H_DESC_SOURCE_TIMEOUT        6
 103 #define UV2H_DESC_DEST_PUT_ERR          7
 104
 105 /*
 106  * delay for 'plugged' timeout retries, in microseconds
 107  */
 108 #define PLUGGED_DELAY                   10
 109
 110 /*
 111  * threshholds at which to use IPI to free resources
 112  */
 113 /* after this # consecutive 'plugged' timeouts, use IPI to release resources */
 114 #define PLUGSB4RESET 100
 115 /* after this many consecutive timeouts, use IPI to release resources */
 116 #define TIMEOUTSB4RESET 1
 117 /* at this number uses of IPI to release resources, giveup the request */
 118 #define IPI_RESET_LIMIT 1
 119 /* after this # consecutive successes, bump up the throttle if it was lowered */
 120 #define COMPLETE_THRESHOLD 5
 121
 122 #define UV_LB_SUBNODEID 0x10
 123
 124 /* these two are the same for UV1 and UV2: */
 125 #define UV_SA_SHFT UVH_LB_BAU_MISC_CONTROL_INTD_SOFT_ACK_TIMEOUT_PERIOD_SHFT
 126 #define UV_SA_MASK UVH_LB_BAU_MISC_CONTROL_INTD_SOFT_ACK_TIMEOUT_PERIOD_MASK
 127 /* 4 bits of software ack period */
 128 #define UV2_ACK_MASK 0x7UL
 129 #define UV2_ACK_UNITS_SHFT 3
 130 #define UV2_LEG_SHFT UV2H_LB_BAU_MISC_CONTROL_USE_LEGACY_DESCRIPTOR_FORMATS_SHFT
 131 #define UV2_EXT_SHFT UV2H_LB_BAU_MISC_CONTROL_ENABLE_EXTENDED_SB_STATUS_SHFT
 132
 133 /*
 134  * number of entries in the destination side payload queue
 135  */
 136 #define DEST_Q_SIZE                     20
 137 /*
 138  * number of destination side software ack resources
 139  */
 140 #define DEST_NUM_RESOURCES              8
 141 /*
 142  * completion statuses for sending a TLB flush message
 143  */
 144 #define FLUSH_RETRY_PLUGGED             1
 145 #define FLUSH_RETRY_TIMEOUT             2
 146 #define FLUSH_GIVEUP                    3
 147 #define FLUSH_COMPLETE                  4
 148
 149 /*
 150  * tuning the action when the numalink network is extremely delayed
 151  */
 152 #define CONGESTED_RESPONSE_US 1000 /* 'long' response time, in microseconds */
 153 #define CONGESTED_REPS 10 /* long delays averaged over this many broadcasts */
 154 #define CONGESTED_PERIOD 30 /* time for the bau to be disabled, in seconds */
 155
 156 /*
 157  * Distribution: 32 bytes (256 bits) (bytes 0-0x1f of descriptor)
 158  * If the 'multilevel' flag in the header portion of the descriptor
 159  * has been set to 0, then endpoint multi-unicast mode is selected.
 160  * The distribution specification (32 bytes) is interpreted as a 256-bit
 161  * distribution vector. Adjacent bits correspond to consecutive even numbered
 162  * nodeIDs. The result of adding the index of a given bit to the 15-bit
 163  * 'base_dest_nasid' field of the header corresponds to the
 164  * destination nodeID associated with that specified bit.
 165  */
 166 struct bau_target_uvhubmask {
 167         unsigned long bits[BITS_TO_LONGS(UV_DISTRIBUTION_SIZE)];
 168 };
 169
 170 /*
 171  * mask of cpu's on a uvhub
 172  * (during initialization we need to check that unsigned long has
 173  *  enough bits for max. cpu's per uvhub)
 174  */
 175 struct bau_local_cpumask {
 176         unsigned long bits;
 177 };
 178
 179 /*
 180  * Payload: 16 bytes (128 bits) (bytes 0x20-0x2f of descriptor)
 181  * only 12 bytes (96 bits) of the payload area are usable.
 182  * An additional 3 bytes (bits 27:4) of the header address are carried
 183  * to the next bytes of the destination payload queue.
 184  * And an additional 2 bytes of the header Suppl_A field are also
 185  * carried to the destination payload queue.
 186  * But the first byte of the Suppl_A becomes bits 127:120 (the 16th byte)
 187  * of the destination payload queue, which is written by the hardware
 188  * with the s/w ack resource bit vector.
 189  * [ effective message contents (16 bytes (128 bits) maximum), not counting
 190  *   the s/w ack bit vector  ]
 191  */
 192
 193 /*
 194  * The payload is software-defined for INTD transactions
 195  */
 196 struct bau_msg_payload {
 197         unsigned long address;          /* signifies a page or all TLB's
 198                                                 of the cpu */
 199         /* 64 bits */
 200         unsigned short sending_cpu;     /* filled in by sender */
 201         /* 16 bits */
 202         unsigned short acknowledge_count;/* filled in by destination */
 203         /* 16 bits */
 204         unsigned int reserved1:32;      /* not usable */
 205 };
 206
 207
 208 /*
 209  * Message header:  16 bytes (128 bits) (bytes 0x30-0x3f of descriptor)
 210  * see table 4.2.3.0.1 in broacast_assist spec.
 211  */
 212 struct bau_msg_header {
 213         unsigned int dest_subnodeid:6;  /* must be 0x10, for the LB */
 214         /* bits 5:0 */
 215         unsigned int base_dest_nasid:15; /* nasid of the */
 216         /* bits 20:6 */                   /* first bit in uvhub map */
 217         unsigned int command:8; /* message type */
 218         /* bits 28:21 */
 219                                 /* 0x38: SN3net EndPoint Message */
 220         unsigned int rsvd_1:3;  /* must be zero */
 221         /* bits 31:29 */
 222                                 /* int will align on 32 bits */
 223         unsigned int rsvd_2:9;  /* must be zero */
 224         /* bits 40:32 */
 225                                 /* Suppl_A is 56-41 */
 226         unsigned int sequence:16;/* message sequence number */
 227         /* bits 56:41 */        /* becomes bytes 16-17 of msg */
 228                                 /* Address field (96:57) is never used as an
 229                                    address (these are address bits 42:3) */
 230
 231         unsigned int rsvd_3:1;  /* must be zero */
 232         /* bit 57 */
 233                                 /* address bits 27:4 are payload */
 234         /* these next 24  (58-81) bits become bytes 12-14 of msg */
 235
 236         /* bits 65:58 land in byte 12 */
 237         unsigned int replied_to:1;/* sent as 0 by the source to byte 12 */
 238         /* bit 58 */
 239         unsigned int msg_type:3; /* software type of the message*/
 240         /* bits 61:59 */
 241         unsigned int canceled:1; /* message canceled, resource to be freed*/
 242         /* bit 62 */
 243         unsigned int payload_1a:1;/* not currently used */
 244         /* bit 63 */
 245         unsigned int payload_1b:2;/* not currently used */
 246         /* bits 65:64 */
 247
 248         /* bits 73:66 land in byte 13 */
 249         unsigned int payload_1ca:6;/* not currently used */
 250         /* bits 71:66 */
 251         unsigned int payload_1c:2;/* not currently used */
 252         /* bits 73:72 */
 253
 254         /* bits 81:74 land in byte 14 */
 255         unsigned int payload_1d:6;/* not currently used */
 256         /* bits 79:74 */
 257         unsigned int payload_1e:2;/* not currently used */
 258         /* bits 81:80 */
 259
 260         unsigned int rsvd_4:7;  /* must be zero */
 261         /* bits 88:82 */
 262         unsigned int sw_ack_flag:1;/* software acknowledge flag */
 263         /* bit 89 */
 264                                 /* INTD trasactions at destination are to
 265                                    wait for software acknowledge */
 266         unsigned int rsvd_5:6;  /* must be zero */
 267         /* bits 95:90 */
 268         unsigned int rsvd_6:5;  /* must be zero */
 269         /* bits 100:96 */
 270         unsigned int int_both:1;/* if 1, interrupt both sockets on the uvhub */
 271         /* bit 101*/
 272         unsigned int fairness:3;/* usually zero */
 273         /* bits 104:102 */
 274         unsigned int multilevel:1;      /* multi-level multicast format */
 275         /* bit 105 */
 276                                 /* 0 for TLB: endpoint multi-unicast messages */
 277         unsigned int chaining:1;/* next descriptor is part of this activation*/
 278         /* bit 106 */
 279         unsigned int rsvd_7:21; /* must be zero */
 280         /* bits 127:107 */
 281 };
 282
 283 /* see msg_type: */
 284 #define MSG_NOOP 0
 285 #define MSG_REGULAR 1
 286 #define MSG_RETRY 2
 287
 288 /*
 289  * The activation descriptor:
 290  * The format of the message to send, plus all accompanying control
 291  * Should be 64 bytes
 292  */
 293 struct bau_desc {
 294         struct bau_target_uvhubmask distribution;
 295         /*
 296          * message template, consisting of header and payload:
 297          */
 298         struct bau_msg_header header;
 299         struct bau_msg_payload payload;
 300 };
 301 /*
 302  *   -payload--    ---------header------
 303  *   bytes 0-11    bits 41-56  bits 58-81
 304  *       A           B  (2)      C (3)
 305  *
 306  *            A/B/C are moved to:
 307  *       A            C          B
 308  *   bytes 0-11  bytes 12-14  bytes 16-17  (byte 15 filled in by hw as vector)
 309  *   ------------payload queue-----------
 310  */
 311
 312 /*
 313  * The payload queue on the destination side is an array of these.
 314  * With BAU_MISC_CONTROL set for software acknowledge mode, the messages
 315  * are 32 bytes (2 micropackets) (256 bits) in length, but contain only 17
 316  * bytes of usable data, including the sw ack vector in byte 15 (bits 127:120)
 317  * (12 bytes come from bau_msg_payload, 3 from payload_1, 2 from
 318  *  sw_ack_vector and payload_2)
 319  * "Enabling Software Acknowledgment mode (see Section 4.3.3 Software
 320  *  Acknowledge Processing) also selects 32 byte (17 bytes usable) payload
 321  *  operation."
 322  */
 323 struct bau_payload_queue_entry {
 324         unsigned long address;          /* signifies a page or all TLB's
 325                                                 of the cpu */
 326         /* 64 bits, bytes 0-7 */
 327
 328         unsigned short sending_cpu;     /* cpu that sent the message */
 329         /* 16 bits, bytes 8-9 */
 330
 331         unsigned short acknowledge_count; /* filled in by destination */
 332         /* 16 bits, bytes 10-11 */
 333
 334         /* these next 3 bytes come from bits 58-81 of the message header */
 335         unsigned short replied_to:1;    /* sent as 0 by the source */
 336         unsigned short msg_type:3;      /* software message type */
 337         unsigned short canceled:1;      /* sent as 0 by the source */
 338         unsigned short unused1:3;       /* not currently using */
 339         /* byte 12 */
 340
 341         unsigned char unused2a;         /* not currently using */
 342         /* byte 13 */
 343         unsigned char unused2;          /* not currently using */
 344         /* byte 14 */
 345
 346         unsigned char sw_ack_vector;    /* filled in by the hardware */
 347         /* byte 15 (bits 127:120) */
 348
 349         unsigned short sequence;        /* message sequence number */
 350         /* bytes 16-17 */
 351         unsigned char unused4[2];       /* not currently using bytes 18-19 */
 352         /* bytes 18-19 */
 353
 354         int number_of_cpus;             /* filled in at destination */
 355         /* 32 bits, bytes 20-23 (aligned) */
 356
 357         unsigned char unused5[8];       /* not using */
 358         /* bytes 24-31 */
 359 };
 360
 361 struct msg_desc {
 362         struct bau_payload_queue_entry *msg;
 363         int msg_slot;
 364         int sw_ack_slot;
 365         struct bau_payload_queue_entry *va_queue_first;
 366         struct bau_payload_queue_entry *va_queue_last;
 367 };
 368
 369 struct reset_args {
 370         int sender;
 371 };
 372
 373 /*
 374  * This structure is allocated per_cpu for UV TLB shootdown statistics.
 375  */
 376 struct ptc_stats {
 377         /* sender statistics */
 378         unsigned long s_giveup; /* number of fall backs to IPI-style flushes */
 379         unsigned long s_requestor; /* number of shootdown requests */
 380         unsigned long s_stimeout; /* source side timeouts */
 381         unsigned long s_dtimeout; /* destination side timeouts */
 382         unsigned long s_time; /* time spent in sending side */
 383         unsigned long s_retriesok; /* successful retries */
 384         unsigned long s_ntargcpu; /* total number of cpu's targeted */
 385         unsigned long s_ntargself; /* times the sending cpu was targeted */
 386         unsigned long s_ntarglocals; /* targets of cpus on the local blade */
 387         unsigned long s_ntargremotes; /* targets of cpus on remote blades */
 388         unsigned long s_ntarglocaluvhub; /* targets of the local hub */
 389         unsigned long s_ntargremoteuvhub; /* remotes hubs targeted */
 390         unsigned long s_ntarguvhub; /* total number of uvhubs targeted */
 391         unsigned long s_ntarguvhub16; /* number of times target hubs >= 16*/
 392         unsigned long s_ntarguvhub8; /* number of times target hubs >= 8 */
 393         unsigned long s_ntarguvhub4; /* number of times target hubs >= 4 */
 394         unsigned long s_ntarguvhub2; /* number of times target hubs >= 2 */
 395         unsigned long s_ntarguvhub1; /* number of times target hubs == 1 */
 396         unsigned long s_resets_plug; /* ipi-style resets from plug state */
 397         unsigned long s_resets_timeout; /* ipi-style resets from timeouts */
 398         unsigned long s_busy; /* status stayed busy past s/w timer */
 399         unsigned long s_throttles; /* waits in throttle */
 400         unsigned long s_retry_messages; /* retry broadcasts */
 401         unsigned long s_bau_reenabled; /* for bau enable/disable */
 402         unsigned long s_bau_disabled; /* for bau enable/disable */
 403         /* destination statistics */
 404         unsigned long d_alltlb; /* times all tlb's on this cpu were flushed */
 405         unsigned long d_onetlb; /* times just one tlb on this cpu was flushed */
 406         unsigned long d_multmsg; /* interrupts with multiple messages */
 407         unsigned long d_nomsg; /* interrupts with no message */
 408         unsigned long d_time; /* time spent on destination side */
 409         unsigned long d_requestee; /* number of messages processed */
 410         unsigned long d_retries; /* number of retry messages processed */
 411         unsigned long d_canceled; /* number of messages canceled by retries */
 412         unsigned long d_nocanceled; /* retries that found nothing to cancel */
 413         unsigned long d_resets; /* number of ipi-style requests processed */
 414         unsigned long d_rcanceled; /* number of messages canceled by resets */
 415 };
 416
 417 struct hub_and_pnode {
 418         short uvhub;
 419         short pnode;
 420 };
 421 /*
 422  * one per-cpu; to locate the software tables
 423  */
 424 struct bau_control {
 425         struct bau_desc *descriptor_base;
 426         struct bau_payload_queue_entry *va_queue_first;
 427         struct bau_payload_queue_entry *va_queue_last;
 428         struct bau_payload_queue_entry *bau_msg_head;
 429         struct bau_control *uvhub_master;
 430         struct bau_control *socket_master;
 431         struct ptc_stats *statp;
 432         unsigned long timeout_interval;
 433         unsigned long set_bau_on_time;
 434         atomic_t active_descriptor_count;
 435         int plugged_tries;
 436         int timeout_tries;
 437         int ipi_attempts;
 438         int conseccompletes;
 439         int baudisabled;
 440         int set_bau_off;
 441         short cpu;
 442         short osnode;
 443         short uvhub_cpu;
 444         short uvhub;
 445         short cpus_in_socket;
 446         short cpus_in_uvhub;
 447         short partition_base_pnode;
 448         unsigned short message_number;
 449         unsigned short uvhub_quiesce;
 450         short socket_acknowledge_count[DEST_Q_SIZE];
 451         cycles_t send_message;
 452         spinlock_t uvhub_lock;
 453         spinlock_t queue_lock;
 454         /* tunables */
 455         int max_bau_concurrent;
 456         int max_bau_concurrent_constant;
 457         int plugged_delay;
 458         int plugsb4reset;
 459         int timeoutsb4reset;
 460         int ipi_reset_limit;
 461         int complete_threshold;
 462         int congested_response_us;
 463         int congested_reps;
 464         int congested_period;
 465         cycles_t period_time;
 466         long period_requests;
 467         struct hub_and_pnode *target_hub_and_pnode;
 468 };
 469
 470 static inline int bau_uvhub_isset(int uvhub, struct bau_target_uvhubmask *dstp)
 471 {
 472         return constant_test_bit(uvhub, &dstp->bits[0]);
 473 }
 474 static inline void bau_uvhub_set(int pnode, struct bau_target_uvhubmask *dstp)
 475 {
 476         __set_bit(pnode, &dstp->bits[0]);
 477 }
 478 static inline void bau_uvhubs_clear(struct bau_target_uvhubmask *dstp,
 479                                     int nbits)
 480 {
 481         bitmap_zero(&dstp->bits[0], nbits);
 482 }
 483 static inline int bau_uvhub_weight(struct bau_target_uvhubmask *dstp)
 484 {
 485         return bitmap_weight((unsigned long *)&dstp->bits[0],
 486                                 UV_DISTRIBUTION_SIZE);
 487 }
 488
 489 static inline void bau_cpubits_clear(struct bau_local_cpumask *dstp, int nbits)
 490 {
 491         bitmap_zero(&dstp->bits, nbits);
 492 }
 493
 494 #define cpubit_isset(cpu, bau_local_cpumask) \
 495         test_bit((cpu), (bau_local_cpumask).bits)
 496
 497 extern void uv_bau_message_intr1(void);
 498 extern void uv_bau_timeout_intr1(void);
 499
 500 struct atomic_short {
 501         short counter;
 502 };
 503
 504 /**
 505  * atomic_read_short - read a short atomic variable
 506  * @v: pointer of type atomic_short
 507  *
 508  * Atomically reads the value of @v.
 509  */
 510 static inline int atomic_read_short(const struct atomic_short *v)
 511 {
 512         return v->counter;
 513 }
 514
 515 /**
 516  * atomic_add_short_return - add and return a short int
 517  * @i: short value to add
 518  * @v: pointer of type atomic_short
 519  *
 520  * Atomically adds @i to @v and returns @i + @v
 521  */
 522 static inline int atomic_add_short_return(short i, struct atomic_short *v)
 523 {
 524         short __i = i;
 525         asm volatile(LOCK_PREFIX "xaddw %0, %1"
 526                         : "+r" (i), "+m" (v->counter)
 527                         : : "memory");
 528         return i + __i;
 529 }
 530
 531 #endif /* _ASM_X86_UV_UV_BAU_H */