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 node.
  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 UV_CPUS_PER_ACT_STATUS          32
  38 #define UV_ACT_STATUS_MASK              0x3
  39 #define UV_ACT_STATUS_SIZE              2
  40 #define UV_ADP_SIZE                     32
  41 #define UV_DISTRIBUTION_SIZE            256
  42 #define UV_SW_ACK_NPENDING              8
  43 #define UV_NET_ENDPOINT_INTD            0x38
  44 #define UV_DESC_BASE_PNODE_SHIFT        49
  45 #define UV_PAYLOADQ_PNODE_SHIFT         49
  46 #define UV_PTC_BASENAME                 "sgi_uv/ptc_statistics"
  47 #define uv_physnodeaddr(x)              ((__pa((unsigned long)(x)) & uv_mmask))
  48
  49 /*
  50  * bits in UVH_LB_BAU_SB_ACTIVATION_STATUS_0/1
  51  */
  52 #define DESC_STATUS_IDLE                0
  53 #define DESC_STATUS_ACTIVE              1
  54 #define DESC_STATUS_DESTINATION_TIMEOUT 2
  55 #define DESC_STATUS_SOURCE_TIMEOUT      3
  56
  57 /*
  58  * source side threshholds at which message retries print a warning
  59  */
  60 #define SOURCE_TIMEOUT_LIMIT            20
  61 #define DESTINATION_TIMEOUT_LIMIT       20
  62
  63 /*
  64  * number of entries in the destination side payload queue
  65  */
  66 #define DEST_Q_SIZE                     17
  67 /*
  68  * number of destination side software ack resources
  69  */
  70 #define DEST_NUM_RESOURCES              8
  71 #define MAX_CPUS_PER_NODE               32
  72 /*
  73  * completion statuses for sending a TLB flush message
  74  */
  75 #define FLUSH_RETRY                     1
  76 #define FLUSH_GIVEUP                    2
  77 #define FLUSH_COMPLETE                  3
  78
  79 /*
  80  * Distribution: 32 bytes (256 bits) (bytes 0-0x1f of descriptor)
  81  * If the 'multilevel' flag in the header portion of the descriptor
  82  * has been set to 0, then endpoint multi-unicast mode is selected.
  83  * The distribution specification (32 bytes) is interpreted as a 256-bit
  84  * distribution vector. Adjacent bits correspond to consecutive even numbered
  85  * nodeIDs. The result of adding the index of a given bit to the 15-bit
  86  * 'base_dest_nodeid' field of the header corresponds to the
  87  * destination nodeID associated with that specified bit.
  88  */
  89 struct bau_target_nodemask {
  90         unsigned long bits[BITS_TO_LONGS(256)];
  91 };
  92
  93 /*
  94  * mask of cpu's on a node
  95  * (during initialization we need to check that unsigned long has
  96  *  enough bits for max. cpu's per node)
  97  */
  98 struct bau_local_cpumask {
  99         unsigned long bits;
 100 };
 101
 102 /*
 103  * Payload: 16 bytes (128 bits) (bytes 0x20-0x2f of descriptor)
 104  * only 12 bytes (96 bits) of the payload area are usable.
 105  * An additional 3 bytes (bits 27:4) of the header address are carried
 106  * to the next bytes of the destination payload queue.
 107  * And an additional 2 bytes of the header Suppl_A field are also
 108  * carried to the destination payload queue.
 109  * But the first byte of the Suppl_A becomes bits 127:120 (the 16th byte)
 110  * of the destination payload queue, which is written by the hardware
 111  * with the s/w ack resource bit vector.
 112  * [ effective message contents (16 bytes (128 bits) maximum), not counting
 113  *   the s/w ack bit vector  ]
 114  */
 115
 116 /*
 117  * The payload is software-defined for INTD transactions
 118  */
 119 struct bau_msg_payload {
 120         unsigned long address;          /* signifies a page or all TLB's
 121                                                 of the cpu */
 122         /* 64 bits */
 123         unsigned short sending_cpu;     /* filled in by sender */
 124         /* 16 bits */
 125         unsigned short acknowledge_count;/* filled in by destination */
 126         /* 16 bits */
 127         unsigned int reserved1:32;      /* not usable */
 128 };
 129
 130
 131 /*
 132  * Message header:  16 bytes (128 bits) (bytes 0x30-0x3f of descriptor)
 133  * see table 4.2.3.0.1 in broacast_assist spec.
 134  */
 135 struct bau_msg_header {
 136         unsigned int dest_subnodeid:6;  /* must be 0x10, for the LB */
 137         /* bits 5:0 */
 138         unsigned int base_dest_nodeid:15; /* nasid>>1 (pnode) of */
 139         /* bits 20:6 */                   /* first bit in node_map */
 140         unsigned int command:8; /* message type */
 141         /* bits 28:21 */
 142                                 /* 0x38: SN3net EndPoint Message */
 143         unsigned int rsvd_1:3;  /* must be zero */
 144         /* bits 31:29 */
 145                                 /* int will align on 32 bits */
 146         unsigned int rsvd_2:9;  /* must be zero */
 147         /* bits 40:32 */
 148                                 /* Suppl_A is 56-41 */
 149         unsigned int payload_2a:8;/* becomes byte 16 of msg */
 150         /* bits 48:41 */        /* not currently using */
 151         unsigned int payload_2b:8;/* becomes byte 17 of msg */
 152         /* bits 56:49 */        /* not currently using */
 153                                 /* Address field (96:57) is never used as an
 154                                    address (these are address bits 42:3) */
 155         unsigned int rsvd_3:1;  /* must be zero */
 156         /* bit 57 */
 157                                 /* address bits 27:4 are payload */
 158                                 /* these 24 bits become bytes 12-14 of msg */
 159         unsigned int replied_to:1;/* sent as 0 by the source to byte 12 */
 160         /* bit 58 */
 161
 162         unsigned int payload_1a:5;/* not currently used */
 163         /* bits 63:59 */
 164         unsigned int payload_1b:8;/* not currently used */
 165         /* bits 71:64 */
 166         unsigned int payload_1c:8;/* not currently used */
 167         /* bits 79:72 */
 168         unsigned int payload_1d:2;/* not currently used */
 169         /* bits 81:80 */
 170
 171         unsigned int rsvd_4:7;  /* must be zero */
 172         /* bits 88:82 */
 173         unsigned int sw_ack_flag:1;/* software acknowledge flag */
 174         /* bit 89 */
 175                                 /* INTD trasactions at destination are to
 176                                    wait for software acknowledge */
 177         unsigned int rsvd_5:6;  /* must be zero */
 178         /* bits 95:90 */
 179         unsigned int rsvd_6:5;  /* must be zero */
 180         /* bits 100:96 */
 181         unsigned int int_both:1;/* if 1, interrupt both sockets on the blade */
 182         /* bit 101*/
 183         unsigned int fairness:3;/* usually zero */
 184         /* bits 104:102 */
 185         unsigned int multilevel:1;      /* multi-level multicast format */
 186         /* bit 105 */
 187                                 /* 0 for TLB: endpoint multi-unicast messages */
 188         unsigned int chaining:1;/* next descriptor is part of this activation*/
 189         /* bit 106 */
 190         unsigned int rsvd_7:21; /* must be zero */
 191         /* bits 127:107 */
 192 };
 193
 194 /*
 195  * The activation descriptor:
 196  * The format of the message to send, plus all accompanying control
 197  * Should be 64 bytes
 198  */
 199 struct bau_desc {
 200         struct bau_target_nodemask distribution;
 201         /*
 202          * message template, consisting of header and payload:
 203          */
 204         struct bau_msg_header header;
 205         struct bau_msg_payload payload;
 206 };
 207 /*
 208  *   -payload--    ---------header------
 209  *   bytes 0-11    bits 41-56  bits 58-81
 210  *       A           B  (2)      C (3)
 211  *
 212  *            A/B/C are moved to:
 213  *       A            C          B
 214  *   bytes 0-11  bytes 12-14  bytes 16-17  (byte 15 filled in by hw as vector)
 215  *   ------------payload queue-----------
 216  */
 217
 218 /*
 219  * The payload queue on the destination side is an array of these.
 220  * With BAU_MISC_CONTROL set for software acknowledge mode, the messages
 221  * are 32 bytes (2 micropackets) (256 bits) in length, but contain only 17
 222  * bytes of usable data, including the sw ack vector in byte 15 (bits 127:120)
 223  * (12 bytes come from bau_msg_payload, 3 from payload_1, 2 from
 224  *  sw_ack_vector and payload_2)
 225  * "Enabling Software Acknowledgment mode (see Section 4.3.3 Software
 226  *  Acknowledge Processing) also selects 32 byte (17 bytes usable) payload
 227  *  operation."
 228  */
 229 struct bau_payload_queue_entry {
 230         unsigned long address;          /* signifies a page or all TLB's
 231                                                 of the cpu */
 232         /* 64 bits, bytes 0-7 */
 233
 234         unsigned short sending_cpu;     /* cpu that sent the message */
 235         /* 16 bits, bytes 8-9 */
 236
 237         unsigned short acknowledge_count; /* filled in by destination */
 238         /* 16 bits, bytes 10-11 */
 239
 240         unsigned short replied_to:1;    /* sent as 0 by the source */
 241         /* 1 bit */
 242         unsigned short unused1:7;       /* not currently using */
 243         /* 7 bits: byte 12) */
 244
 245         unsigned char unused2[2];       /* not currently using */
 246         /* bytes 13-14 */
 247
 248         unsigned char sw_ack_vector;    /* filled in by the hardware */
 249         /* byte 15 (bits 127:120) */
 250
 251         unsigned char unused4[3];       /* not currently using bytes 17-19 */
 252         /* bytes 17-19 */
 253
 254         int number_of_cpus;             /* filled in at destination */
 255         /* 32 bits, bytes 20-23 (aligned) */
 256
 257         unsigned char unused5[8];       /* not using */
 258         /* bytes 24-31 */
 259 };
 260
 261 /*
 262  * one for every slot in the destination payload queue
 263  */
 264 struct bau_msg_status {
 265         struct bau_local_cpumask seen_by;       /* map of cpu's */
 266 };
 267
 268 /*
 269  * one for every slot in the destination software ack resources
 270  */
 271 struct bau_sw_ack_status {
 272         struct bau_payload_queue_entry *msg;    /* associated message */
 273         int watcher;                            /* cpu monitoring, or -1 */
 274 };
 275
 276 /*
 277  * one on every node and per-cpu; to locate the software tables
 278  */
 279 struct bau_control {
 280         struct bau_desc *descriptor_base;
 281         struct bau_payload_queue_entry *bau_msg_head;
 282         struct bau_payload_queue_entry *va_queue_first;
 283         struct bau_payload_queue_entry *va_queue_last;
 284         struct bau_msg_status *msg_statuses;
 285         int *watching; /* pointer to array */
 286 };
 287
 288 /*
 289  * This structure is allocated per_cpu for UV TLB shootdown statistics.
 290  */
 291 struct ptc_stats {
 292         unsigned long ptc_i;    /* number of IPI-style flushes */
 293         unsigned long requestor;        /* number of nodes this cpu sent to */
 294         unsigned long requestee;        /* times cpu was remotely requested */
 295         unsigned long alltlb;   /* times all tlb's on this cpu were flushed */
 296         unsigned long onetlb;   /* times just one tlb on this cpu was flushed */
 297         unsigned long s_retry;  /* retries on source side timeouts */
 298         unsigned long d_retry;  /* retries on destination side timeouts */
 299         unsigned long sflush;   /* cycles spent in uv_flush_tlb_others */
 300         unsigned long dflush;   /* cycles spent on destination side */
 301         unsigned long retriesok; /* successes on retries */
 302         unsigned long nomsg;    /* interrupts with no message */
 303         unsigned long multmsg;  /* interrupts with multiple messages */
 304         unsigned long ntargeted;/* nodes targeted */
 305 };
 306
 307 static inline int bau_node_isset(int node, struct bau_target_nodemask *dstp)
 308 {
 309         return constant_test_bit(node, &dstp->bits[0]);
 310 }
 311 static inline void bau_node_set(int node, struct bau_target_nodemask *dstp)
 312 {
 313         __set_bit(node, &dstp->bits[0]);
 314 }
 315 static inline void bau_nodes_clear(struct bau_target_nodemask *dstp, int nbits)
 316 {
 317         bitmap_zero(&dstp->bits[0], nbits);
 318 }
 319
 320 static inline void bau_cpubits_clear(struct bau_local_cpumask *dstp, int nbits)
 321 {
 322         bitmap_zero(&dstp->bits, nbits);
 323 }
 324
 325 #define cpubit_isset(cpu, bau_local_cpumask) \
 326         test_bit((cpu), (bau_local_cpumask).bits)
 327
 328 extern void uv_bau_message_intr1(void);
 329 extern void uv_bau_timeout_intr1(void);
 330
 331 #endif /* _ASM_X86_UV_UV_BAU_H */