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 31 sets, one for sending BAU messages from each of the 32
  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 UV_ITEMS_PER_DESCRIPTOR         8
  37 #define MAX_BAU_CONCURRENT              3
  38 #define UV_CPUS_PER_ACT_STATUS          32
  39 #define UV_ACT_STATUS_MASK              0x3
  40 #define UV_ACT_STATUS_SIZE              2
  41 #define UV_ADP_SIZE                     32
  42 #define UV_DISTRIBUTION_SIZE            256
  43 #define UV_SW_ACK_NPENDING              8
  44 #define UV_NET_ENDPOINT_INTD            0x38
  45 #define UV_DESC_BASE_PNODE_SHIFT        49
  46 #define UV_PAYLOADQ_PNODE_SHIFT         49
  47 #define UV_PTC_BASENAME                 "sgi_uv/ptc_statistics"
  48 #define uv_physnodeaddr(x)              ((__pa((unsigned long)(x)) & uv_mmask))
  49 #define UV_ENABLE_INTD_SOFT_ACK_MODE_SHIFT 15
  50 #define UV_INTD_SOFT_ACK_TIMEOUT_PERIOD_SHIFT 16
  51 #define UV_INTD_SOFT_ACK_TIMEOUT_PERIOD 0x000000000bUL
  52
  53 /*
  54  * bits in UVH_LB_BAU_SB_ACTIVATION_STATUS_0/1
  55  */
  56 #define DESC_STATUS_IDLE                0
  57 #define DESC_STATUS_ACTIVE              1
  58 #define DESC_STATUS_DESTINATION_TIMEOUT 2
  59 #define DESC_STATUS_SOURCE_TIMEOUT      3
  60
  61 /*
  62  * source side threshholds at which message retries print a warning
  63  */
  64 #define SOURCE_TIMEOUT_LIMIT            20
  65 #define DESTINATION_TIMEOUT_LIMIT       20
  66
  67 /*
  68  * misc. delays, in microseconds
  69  */
  70 #define THROTTLE_DELAY                  10
  71 #define TIMEOUT_DELAY                   10
  72 #define BIOS_TO                         1000
  73 /* BIOS is assumed to set the destination timeout to 1003520 nanoseconds */
  74
  75 /*
  76  * threshholds at which to use IPI to free resources
  77  */
  78 #define PLUGSB4RESET 100
  79 #define TIMEOUTSB4RESET 100
  80
  81 /*
  82  * number of entries in the destination side payload queue
  83  */
  84 #define DEST_Q_SIZE                     20
  85 /*
  86  * number of destination side software ack resources
  87  */
  88 #define DEST_NUM_RESOURCES              8
  89 #define MAX_CPUS_PER_NODE               32
  90 /*
  91  * completion statuses for sending a TLB flush message
  92  */
  93 #define FLUSH_RETRY_PLUGGED             1
  94 #define FLUSH_RETRY_TIMEOUT             2
  95 #define FLUSH_GIVEUP                    3
  96 #define FLUSH_COMPLETE                  4
  97
  98 /*
  99  * Distribution: 32 bytes (256 bits) (bytes 0-0x1f of descriptor)
 100  * If the 'multilevel' flag in the header portion of the descriptor
 101  * has been set to 0, then endpoint multi-unicast mode is selected.
 102  * The distribution specification (32 bytes) is interpreted as a 256-bit
 103  * distribution vector. Adjacent bits correspond to consecutive even numbered
 104  * nodeIDs. The result of adding the index of a given bit to the 15-bit
 105  * 'base_dest_nodeid' field of the header corresponds to the
 106  * destination nodeID associated with that specified bit.
 107  */
 108 struct bau_target_uvhubmask {
 109         unsigned long bits[BITS_TO_LONGS(UV_DISTRIBUTION_SIZE)];
 110 };
 111
 112 /*
 113  * mask of cpu's on a uvhub
 114  * (during initialization we need to check that unsigned long has
 115  *  enough bits for max. cpu's per uvhub)
 116  */
 117 struct bau_local_cpumask {
 118         unsigned long bits;
 119 };
 120
 121 /*
 122  * Payload: 16 bytes (128 bits) (bytes 0x20-0x2f of descriptor)
 123  * only 12 bytes (96 bits) of the payload area are usable.
 124  * An additional 3 bytes (bits 27:4) of the header address are carried
 125  * to the next bytes of the destination payload queue.
 126  * And an additional 2 bytes of the header Suppl_A field are also
 127  * carried to the destination payload queue.
 128  * But the first byte of the Suppl_A becomes bits 127:120 (the 16th byte)
 129  * of the destination payload queue, which is written by the hardware
 130  * with the s/w ack resource bit vector.
 131  * [ effective message contents (16 bytes (128 bits) maximum), not counting
 132  *   the s/w ack bit vector  ]
 133  */
 134
 135 /*
 136  * The payload is software-defined for INTD transactions
 137  */
 138 struct bau_msg_payload {
 139         unsigned long address;          /* signifies a page or all TLB's
 140                                                 of the cpu */
 141         /* 64 bits */
 142         unsigned short sending_cpu;     /* filled in by sender */
 143         /* 16 bits */
 144         unsigned short acknowledge_count;/* filled in by destination */
 145         /* 16 bits */
 146         unsigned int reserved1:32;      /* not usable */
 147 };
 148
 149
 150 /*
 151  * Message header:  16 bytes (128 bits) (bytes 0x30-0x3f of descriptor)
 152  * see table 4.2.3.0.1 in broacast_assist spec.
 153  */
 154 struct bau_msg_header {
 155         unsigned int dest_subnodeid:6;  /* must be 0x10, for the LB */
 156         /* bits 5:0 */
 157         unsigned int base_dest_nodeid:15; /* nasid (pnode<<1) of */
 158         /* bits 20:6 */                   /* first bit in uvhub map */
 159         unsigned int command:8; /* message type */
 160         /* bits 28:21 */
 161                                 /* 0x38: SN3net EndPoint Message */
 162         unsigned int rsvd_1:3;  /* must be zero */
 163         /* bits 31:29 */
 164                                 /* int will align on 32 bits */
 165         unsigned int rsvd_2:9;  /* must be zero */
 166         /* bits 40:32 */
 167                                 /* Suppl_A is 56-41 */
 168         unsigned int sequence:16;/* message sequence number */
 169         /* bits 56:41 */        /* becomes bytes 16-17 of msg */
 170                                 /* Address field (96:57) is never used as an
 171                                    address (these are address bits 42:3) */
 172
 173         unsigned int rsvd_3:1;  /* must be zero */
 174         /* bit 57 */
 175                                 /* address bits 27:4 are payload */
 176         /* these next 24  (58-81) bits become bytes 12-14 of msg */
 177
 178         /* bits 65:58 land in byte 12 */
 179         unsigned int replied_to:1;/* sent as 0 by the source to byte 12 */
 180         /* bit 58 */
 181         unsigned int msg_type:3; /* software type of the message*/
 182         /* bits 61:59 */
 183         unsigned int canceled:1; /* message canceled, resource to be freed*/
 184         /* bit 62 */
 185         unsigned int payload_1a:1;/* not currently used */
 186         /* bit 63 */
 187         unsigned int payload_1b:2;/* not currently used */
 188         /* bits 65:64 */
 189
 190         /* bits 73:66 land in byte 13 */
 191         unsigned int payload_1ca:6;/* not currently used */
 192         /* bits 71:66 */
 193         unsigned int payload_1c:2;/* not currently used */
 194         /* bits 73:72 */
 195
 196         /* bits 81:74 land in byte 14 */
 197         unsigned int payload_1d:6;/* not currently used */
 198         /* bits 79:74 */
 199         unsigned int payload_1e:2;/* not currently used */
 200         /* bits 81:80 */
 201
 202         unsigned int rsvd_4:7;  /* must be zero */
 203         /* bits 88:82 */
 204         unsigned int sw_ack_flag:1;/* software acknowledge flag */
 205         /* bit 89 */
 206                                 /* INTD trasactions at destination are to
 207                                    wait for software acknowledge */
 208         unsigned int rsvd_5:6;  /* must be zero */
 209         /* bits 95:90 */
 210         unsigned int rsvd_6:5;  /* must be zero */
 211         /* bits 100:96 */
 212         unsigned int int_both:1;/* if 1, interrupt both sockets on the uvhub */
 213         /* bit 101*/
 214         unsigned int fairness:3;/* usually zero */
 215         /* bits 104:102 */
 216         unsigned int multilevel:1;      /* multi-level multicast format */
 217         /* bit 105 */
 218                                 /* 0 for TLB: endpoint multi-unicast messages */
 219         unsigned int chaining:1;/* next descriptor is part of this activation*/
 220         /* bit 106 */
 221         unsigned int rsvd_7:21; /* must be zero */
 222         /* bits 127:107 */
 223 };
 224
 225 /* see msg_type: */
 226 #define MSG_NOOP 0
 227 #define MSG_REGULAR 1
 228 #define MSG_RETRY 2
 229
 230 /*
 231  * The activation descriptor:
 232  * The format of the message to send, plus all accompanying control
 233  * Should be 64 bytes
 234  */
 235 struct bau_desc {
 236         struct bau_target_uvhubmask distribution;
 237         /*
 238          * message template, consisting of header and payload:
 239          */
 240         struct bau_msg_header header;
 241         struct bau_msg_payload payload;
 242 };
 243 /*
 244  *   -payload--    ---------header------
 245  *   bytes 0-11    bits 41-56  bits 58-81
 246  *       A           B  (2)      C (3)
 247  *
 248  *            A/B/C are moved to:
 249  *       A            C          B
 250  *   bytes 0-11  bytes 12-14  bytes 16-17  (byte 15 filled in by hw as vector)
 251  *   ------------payload queue-----------
 252  */
 253
 254 /*
 255  * The payload queue on the destination side is an array of these.
 256  * With BAU_MISC_CONTROL set for software acknowledge mode, the messages
 257  * are 32 bytes (2 micropackets) (256 bits) in length, but contain only 17
 258  * bytes of usable data, including the sw ack vector in byte 15 (bits 127:120)
 259  * (12 bytes come from bau_msg_payload, 3 from payload_1, 2 from
 260  *  sw_ack_vector and payload_2)
 261  * "Enabling Software Acknowledgment mode (see Section 4.3.3 Software
 262  *  Acknowledge Processing) also selects 32 byte (17 bytes usable) payload
 263  *  operation."
 264  */
 265 struct bau_payload_queue_entry {
 266         unsigned long address;          /* signifies a page or all TLB's
 267                                                 of the cpu */
 268         /* 64 bits, bytes 0-7 */
 269
 270         unsigned short sending_cpu;     /* cpu that sent the message */
 271         /* 16 bits, bytes 8-9 */
 272
 273         unsigned short acknowledge_count; /* filled in by destination */
 274         /* 16 bits, bytes 10-11 */
 275
 276         /* these next 3 bytes come from bits 58-81 of the message header */
 277         unsigned short replied_to:1;    /* sent as 0 by the source */
 278         unsigned short msg_type:3;      /* software message type */
 279         unsigned short canceled:1;      /* sent as 0 by the source */
 280         unsigned short unused1:3;       /* not currently using */
 281         /* byte 12 */
 282
 283         unsigned char unused2a;         /* not currently using */
 284         /* byte 13 */
 285         unsigned char unused2;          /* not currently using */
 286         /* byte 14 */
 287
 288         unsigned char sw_ack_vector;    /* filled in by the hardware */
 289         /* byte 15 (bits 127:120) */
 290
 291         unsigned short sequence;        /* message sequence number */
 292         /* bytes 16-17 */
 293         unsigned char unused4[2];       /* not currently using bytes 18-19 */
 294         /* bytes 18-19 */
 295
 296         int number_of_cpus;             /* filled in at destination */
 297         /* 32 bits, bytes 20-23 (aligned) */
 298
 299         unsigned char unused5[8];       /* not using */
 300         /* bytes 24-31 */
 301 };
 302
 303 /*
 304  * one per-cpu; to locate the software tables
 305  */
 306 struct bau_control {
 307         struct bau_desc *descriptor_base;
 308         struct bau_payload_queue_entry *va_queue_first;
 309         struct bau_payload_queue_entry *va_queue_last;
 310         struct bau_payload_queue_entry *bau_msg_head;
 311         struct bau_control *uvhub_master;
 312         struct bau_control *socket_master;
 313         unsigned long timeout_interval;
 314         atomic_t active_descriptor_count;
 315         int max_concurrent;
 316         int max_concurrent_constant;
 317         int retry_message_scans;
 318         int plugged_tries;
 319         int timeout_tries;
 320         int ipi_attempts;
 321         int conseccompletes;
 322         short cpu;
 323         short uvhub_cpu;
 324         short uvhub;
 325         short cpus_in_socket;
 326         short cpus_in_uvhub;
 327         unsigned short message_number;
 328         unsigned short uvhub_quiesce;
 329         short socket_acknowledge_count[DEST_Q_SIZE];
 330         cycles_t send_message;
 331         spinlock_t masks_lock;
 332         spinlock_t uvhub_lock;
 333         spinlock_t queue_lock;
 334 };
 335
 336 /*
 337  * This structure is allocated per_cpu for UV TLB shootdown statistics.
 338  */
 339 struct ptc_stats {
 340         /* sender statistics */
 341         unsigned long s_giveup; /* number of fall backs to IPI-style flushes */
 342         unsigned long s_requestor; /* number of shootdown requests */
 343         unsigned long s_stimeout; /* source side timeouts */
 344         unsigned long s_dtimeout; /* destination side timeouts */
 345         unsigned long s_time; /* time spent in sending side */
 346         unsigned long s_retriesok; /* successful retries */
 347         unsigned long s_ntargcpu; /* number of cpus targeted */
 348         unsigned long s_ntarguvhub; /* number of uvhubs targeted */
 349         unsigned long s_ntarguvhub16; /* number of times >= 16 target hubs */
 350         unsigned long s_ntarguvhub8; /* number of times >= 8 target hubs */
 351         unsigned long s_ntarguvhub4; /* number of times >= 4 target hubs */
 352         unsigned long s_ntarguvhub2; /* number of times >= 2 target hubs */
 353         unsigned long s_ntarguvhub1; /* number of times == 1 target hub */
 354         unsigned long s_resets_plug; /* ipi-style resets from plug state */
 355         unsigned long s_resets_timeout; /* ipi-style resets from timeouts */
 356         unsigned long s_busy; /* status stayed busy past s/w timer */
 357         unsigned long s_throttles; /* waits in throttle */
 358         unsigned long s_retry_messages; /* retry broadcasts */
 359         /* destination statistics */
 360         unsigned long d_alltlb; /* times all tlb's on this cpu were flushed */
 361         unsigned long d_onetlb; /* times just one tlb on this cpu was flushed */
 362         unsigned long d_multmsg; /* interrupts with multiple messages */
 363         unsigned long d_nomsg; /* interrupts with no message */
 364         unsigned long d_time; /* time spent on destination side */
 365         unsigned long d_requestee; /* number of messages processed */
 366         unsigned long d_retries; /* number of retry messages processed */
 367         unsigned long d_canceled; /* number of messages canceled by retries */
 368         unsigned long d_nocanceled; /* retries that found nothing to cancel */
 369         unsigned long d_resets; /* number of ipi-style requests processed */
 370         unsigned long d_rcanceled; /* number of messages canceled by resets */
 371 };
 372
 373 static inline int bau_uvhub_isset(int uvhub, struct bau_target_uvhubmask *dstp)
 374 {
 375         return constant_test_bit(uvhub, &dstp->bits[0]);
 376 }
 377 static inline void bau_uvhub_set(int uvhub, struct bau_target_uvhubmask *dstp)
 378 {
 379         __set_bit(uvhub, &dstp->bits[0]);
 380 }
 381 static inline void bau_uvhubs_clear(struct bau_target_uvhubmask *dstp,
 382                                     int nbits)
 383 {
 384         bitmap_zero(&dstp->bits[0], nbits);
 385 }
 386 static inline int bau_uvhub_weight(struct bau_target_uvhubmask *dstp)
 387 {
 388         return bitmap_weight((unsigned long *)&dstp->bits[0],
 389                                 UV_DISTRIBUTION_SIZE);
 390 }
 391
 392 static inline void bau_cpubits_clear(struct bau_local_cpumask *dstp, int nbits)
 393 {
 394         bitmap_zero(&dstp->bits, nbits);
 395 }
 396
 397 #define cpubit_isset(cpu, bau_local_cpumask) \
 398         test_bit((cpu), (bau_local_cpumask).bits)
 399
 400 extern void uv_bau_message_intr1(void);
 401 extern void uv_bau_timeout_intr1(void);
 402
 403 struct atomic_short {
 404         short counter;
 405 };
 406
 407 /**
 408  * atomic_read_short - read a short atomic variable
 409  * @v: pointer of type atomic_short
 410  *
 411  * Atomically reads the value of @v.
 412  */
 413 static inline int atomic_read_short(const struct atomic_short *v)
 414 {
 415         return v->counter;
 416 }
 417
 418 /**
 419  * atomic_add_short_return - add and return a short int
 420  * @i: short value to add
 421  * @v: pointer of type atomic_short
 422  *
 423  * Atomically adds @i to @v and returns @i + @v
 424  */
 425 static inline int atomic_add_short_return(short i, struct atomic_short *v)
 426 {
 427         short __i = i;
 428         asm volatile(LOCK_PREFIX "xaddw %0, %1"
 429                         : "+r" (i), "+m" (v->counter)
 430                         : : "memory");
 431         return i + __i;
 432 }
 433
 434 #endif /* _ASM_X86_UV_UV_BAU_H */