drivers/misc/sgi-gru/grutables.h

   1 /*
   2  * SN Platform GRU Driver
   3  *
   4  *            GRU DRIVER TABLES, MACROS, externs, etc
   5  *
   6  *  Copyright (c) 2008 Silicon Graphics, Inc.  All Rights Reserved.
   7  *
   8  *  This program is free software; you can redistribute it and/or modify
   9  *  it under the terms of the GNU General Public License as published by
  10  *  the Free Software Foundation; either version 2 of the License, or
  11  *  (at your option) any later version.
  12  *
  13  *  This program is distributed in the hope that it will be useful,
  14  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  16  *  GNU General Public License for more details.
  17  *
  18  *  You should have received a copy of the GNU General Public License
  19  *  along with this program; if not, write to the Free Software
  20  *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
  21  */
  22
  23 #ifndef __GRUTABLES_H__
  24 #define __GRUTABLES_H__
  25
  26 /*
  27  * GRU Chiplet:
  28  *   The GRU is a user addressible memory accelerator. It provides
  29  *   several forms of load, store, memset, bcopy instructions. In addition, it
  30  *   contains special instructions for AMOs, sending messages to message
  31  *   queues, etc.
  32  *
  33  *   The GRU is an integral part of the node controller. It connects
  34  *   directly to the cpu socket. In its current implementation, there are 2
  35  *   GRU chiplets in the node controller on each blade (~node).
  36  *
  37  *   The entire GRU memory space is fully coherent and cacheable by the cpus.
  38  *
  39  *   Each GRU chiplet has a physical memory map that looks like the following:
  40  *
  41  *      +-----------------+
  42  *      |/////////////////|
  43  *      |/////////////////|
  44  *      |/////////////////|
  45  *      |/////////////////|
  46  *      |/////////////////|
  47  *      |/////////////////|
  48  *      |/////////////////|
  49  *      |/////////////////|
  50  *      +-----------------+
  51  *      |  system control |
  52  *      +-----------------+        _______ +-------------+
  53  *      |/////////////////|       /        |             |
  54  *      |/////////////////|      /         |             |
  55  *      |/////////////////|     /          | instructions|
  56  *      |/////////////////|    /           |             |
  57  *      |/////////////////|   /            |             |
  58  *      |/////////////////|  /             |-------------|
  59  *      |/////////////////| /              |             |
  60  *      +-----------------+                |             |
  61  *      |   context 15    |                |  data       |
  62  *      +-----------------+                |             |
  63  *      |    ......       | \              |             |
  64  *      +-----------------+  \____________ +-------------+
  65  *      |   context 1     |
  66  *      +-----------------+
  67  *      |   context 0     |
  68  *      +-----------------+
  69  *
  70  *   Each of the "contexts" is a chunk of memory that can be mmaped into user
  71  *   space. The context consists of 2 parts:
  72  *
  73  *      - an instruction space that can be directly accessed by the user
  74  *        to issue GRU instructions and to check instruction status.
  75  *
  76  *      - a data area that acts as normal RAM.
  77  *
  78  *   User instructions contain virtual addresses of data to be accessed by the
  79  *   GRU. The GRU contains a TLB that is used to convert these user virtual
  80  *   addresses to physical addresses.
  81  *
  82  *   The "system control" area of the GRU chiplet is used by the kernel driver
  83  *   to manage user contexts and to perform functions such as TLB dropin and
  84  *   purging.
  85  *
  86  *   One context may be reserved for the kernel and used for cross-partition
  87  *   communication. The GRU will also be used to asynchronously zero out
  88  *   large blocks of memory (not currently implemented).
  89  *
  90  *
  91  * Tables:
  92  *
  93  *      VDATA-VMA Data          - Holds a few parameters. Head of linked list of
  94  *                                GTS tables for threads using the GSEG
  95  *      GTS - Gru Thread State  - contains info for managing a GSEG context. A
  96  *                                GTS is allocated for each thread accessing a
  97  *                                GSEG.
  98  *      GTD - GRU Thread Data   - contains shadow copy of GRU data when GSEG is
  99  *                                not loaded into a GRU
 100  *      GMS - GRU Memory Struct - Used to manage TLB shootdowns. Tracks GRUs
 101  *                                where a GSEG has been loaded. Similar to
 102  *                                an mm_struct but for GRU.
 103  *
 104  *      GS  - GRU State         - Used to manage the state of a GRU chiplet
 105  *      BS  - Blade State       - Used to manage state of all GRU chiplets
 106  *                                on a blade
 107  *
 108  *
 109  *  Normal task tables for task using GRU.
 110  *              - 2 threads in process
 111  *              - 2 GSEGs open in process
 112  *              - GSEG1 is being used by both threads
 113  *              - GSEG2 is used only by thread 2
 114  *
 115  *       task -->|
 116  *       task ---+---> mm ->------ (notifier) -------+-> gms
 117  *                     |                             |
 118  *                     |--> vma -> vdata ---> gts--->|          GSEG1 (thread1)
 119  *                     |                  |          |
 120  *                     |                  +-> gts--->|          GSEG1 (thread2)
 121  *                     |                             |
 122  *                     |--> vma -> vdata ---> gts--->|          GSEG2 (thread2)
 123  *                     .
 124  *                     .
 125  *
 126  *  GSEGs are marked DONTCOPY on fork
 127  *
 128  * At open
 129  *      file.private_data -> NULL
 130  *
 131  * At mmap,
 132  *      vma -> vdata
 133  *
 134  * After gseg reference
 135  *      vma -> vdata ->gts
 136  *
 137  * After fork
 138  *   parent
 139  *      vma -> vdata -> gts
 140  *   child
 141  *      (vma is not copied)
 142  *
 143  */
 144
 145 #include <linux/rmap.h>
 146 #include <linux/interrupt.h>
 147 #include <linux/mutex.h>
 148 #include <linux/wait.h>
 149 #include <linux/mmu_notifier.h>
 150 #include "gru.h"
 151 #include "gruhandles.h"
 152
 153 extern struct gru_stats_s gru_stats;
 154 extern struct gru_blade_state *gru_base[];
 155 extern unsigned long gru_start_paddr, gru_end_paddr;
 156 extern unsigned int gru_max_gids;
 157
 158 #define GRU_MAX_BLADES          MAX_NUMNODES
 159 #define GRU_MAX_GRUS            (GRU_MAX_BLADES * GRU_CHIPLETS_PER_BLADE)
 160
 161 #define GRU_DRIVER_ID_STR       "SGI GRU Device Driver"
 162 #define GRU_DRIVER_VERSION_STR  "0.80"
 163
 164 /*
 165  * GRU statistics.
 166  */
 167 struct gru_stats_s {
 168         atomic_long_t vdata_alloc;
 169         atomic_long_t vdata_free;
 170         atomic_long_t gts_alloc;
 171         atomic_long_t gts_free;
 172         atomic_long_t vdata_double_alloc;
 173         atomic_long_t gts_double_allocate;
 174         atomic_long_t assign_context;
 175         atomic_long_t assign_context_failed;
 176         atomic_long_t free_context;
 177         atomic_long_t load_user_context;
 178         atomic_long_t load_kernel_context;
 179         atomic_long_t lock_kernel_context;
 180         atomic_long_t unlock_kernel_context;
 181         atomic_long_t steal_user_context;
 182         atomic_long_t steal_kernel_context;
 183         atomic_long_t steal_context_failed;
 184         atomic_long_t nopfn;
 185         atomic_long_t break_cow;
 186         atomic_long_t asid_new;
 187         atomic_long_t asid_next;
 188         atomic_long_t asid_wrap;
 189         atomic_long_t asid_reuse;
 190         atomic_long_t intr;
 191         atomic_long_t intr_mm_lock_failed;
 192         atomic_long_t call_os;
 193         atomic_long_t call_os_offnode_reference;
 194         atomic_long_t call_os_check_for_bug;
 195         atomic_long_t call_os_wait_queue;
 196         atomic_long_t user_flush_tlb;
 197         atomic_long_t user_unload_context;
 198         atomic_long_t user_exception;
 199         atomic_long_t set_task_slice;
 200         atomic_long_t migrate_check;
 201         atomic_long_t migrated_retarget;
 202         atomic_long_t migrated_unload;
 203         atomic_long_t migrated_unload_delay;
 204         atomic_long_t migrated_nopfn_retarget;
 205         atomic_long_t migrated_nopfn_unload;
 206         atomic_long_t tlb_dropin;
 207         atomic_long_t tlb_dropin_fail_no_asid;
 208         atomic_long_t tlb_dropin_fail_upm;
 209         atomic_long_t tlb_dropin_fail_invalid;
 210         atomic_long_t tlb_dropin_fail_range_active;
 211         atomic_long_t tlb_dropin_fail_idle;
 212         atomic_long_t tlb_dropin_fail_fmm;
 213         atomic_long_t tlb_dropin_fail_no_exception;
 214         atomic_long_t tlb_dropin_fail_no_exception_war;
 215         atomic_long_t mmu_invalidate_range;
 216         atomic_long_t mmu_invalidate_page;
 217         atomic_long_t mmu_clear_flush_young;
 218         atomic_long_t flush_tlb;
 219         atomic_long_t flush_tlb_gru;
 220         atomic_long_t flush_tlb_gru_tgh;
 221         atomic_long_t flush_tlb_gru_zero_asid;
 222
 223         atomic_long_t copy_gpa;
 224
 225         atomic_long_t mesq_receive;
 226         atomic_long_t mesq_receive_none;
 227         atomic_long_t mesq_send;
 228         atomic_long_t mesq_send_failed;
 229         atomic_long_t mesq_noop;
 230         atomic_long_t mesq_send_unexpected_error;
 231         atomic_long_t mesq_send_lb_overflow;
 232         atomic_long_t mesq_send_qlimit_reached;
 233         atomic_long_t mesq_send_amo_nacked;
 234         atomic_long_t mesq_send_put_nacked;
 235         atomic_long_t mesq_qf_not_full;
 236         atomic_long_t mesq_qf_locked;
 237         atomic_long_t mesq_qf_noop_not_full;
 238         atomic_long_t mesq_qf_switch_head_failed;
 239         atomic_long_t mesq_qf_unexpected_error;
 240         atomic_long_t mesq_noop_unexpected_error;
 241         atomic_long_t mesq_noop_lb_overflow;
 242         atomic_long_t mesq_noop_qlimit_reached;
 243         atomic_long_t mesq_noop_amo_nacked;
 244         atomic_long_t mesq_noop_put_nacked;
 245
 246 };
 247
 248 enum mcs_op {cchop_allocate, cchop_start, cchop_interrupt, cchop_interrupt_sync,
 249         cchop_deallocate, tghop_invalidate, mcsop_last};
 250
 251 struct mcs_op_statistic {
 252         atomic_long_t   count;
 253         atomic_long_t   total;
 254         unsigned long   max;
 255 };
 256
 257 extern struct mcs_op_statistic mcs_op_statistics[mcsop_last];
 258
 259 #define OPT_DPRINT      1
 260 #define OPT_STATS       2
 261
 262
 263 #define IRQ_GRU                 110     /* Starting IRQ number for interrupts */
 264
 265 /* Delay in jiffies between attempts to assign a GRU context */
 266 #define GRU_ASSIGN_DELAY        ((HZ * 20) / 1000)
 267
 268 /*
 269  * If a process has it's context stolen, min delay in jiffies before trying to
 270  * steal a context from another process.
 271  */
 272 #define GRU_STEAL_DELAY         ((HZ * 200) / 1000)
 273
 274 #define STAT(id)        do {                                            \
 275                                 if (gru_options & OPT_STATS)            \
 276                                         atomic_long_inc(&gru_stats.id); \
 277                         } while (0)
 278
 279 #ifdef CONFIG_SGI_GRU_DEBUG
 280 #define gru_dbg(dev, fmt, x...)                                         \
 281         do {                                                            \
 282                 if (gru_options & OPT_DPRINT)                           \
 283                         dev_dbg(dev, "%s: " fmt, __func__, x);          \
 284         } while (0)
 285 #else
 286 #define gru_dbg(x...)
 287 #endif
 288
 289 /*-----------------------------------------------------------------------------
 290  * ASID management
 291  */
 292 #define MAX_ASID        0xfffff0
 293 #define MIN_ASID        8
 294 #define ASID_INC        8       /* number of regions */
 295
 296 /* Generate a GRU asid value from a GRU base asid & a virtual address. */
 297 #if defined CONFIG_IA64
 298 #define VADDR_HI_BIT            64
 299 #elif defined CONFIG_X86_64
 300 #define VADDR_HI_BIT            48
 301 #else
 302 #error "Unsupported architecture"
 303 #endif
 304 #define GRUREGION(addr)         ((addr) >> (VADDR_HI_BIT - 3) & 3)
 305 #define GRUASID(asid, addr)     ((asid) + GRUREGION(addr))
 306
 307 /*------------------------------------------------------------------------------
 308  *  File & VMS Tables
 309  */
 310
 311 struct gru_state;
 312
 313 /*
 314  * This structure is pointed to from the mmstruct via the notifier pointer.
 315  * There is one of these per address space.
 316  */
 317 struct gru_mm_tracker {                         /* pack to reduce size */
 318         unsigned int            mt_asid_gen:24; /* ASID wrap count */
 319         unsigned int            mt_asid:24;     /* current base ASID for gru */
 320         unsigned short          mt_ctxbitmap:16;/* bitmap of contexts using
 321                                                    asid */
 322 } __attribute__ ((packed));
 323
 324 struct gru_mm_struct {
 325         struct mmu_notifier     ms_notifier;
 326         atomic_t                ms_refcnt;
 327         spinlock_t              ms_asid_lock;   /* protects ASID assignment */
 328         atomic_t                ms_range_active;/* num range_invals active */
 329         char                    ms_released;
 330         wait_queue_head_t       ms_wait_queue;
 331         DECLARE_BITMAP(ms_asidmap, GRU_MAX_GRUS);
 332         struct gru_mm_tracker   ms_asids[GRU_MAX_GRUS];
 333 };
 334
 335 /*
 336  * One of these structures is allocated when a GSEG is mmaped. The
 337  * structure is pointed to by the vma->vm_private_data field in the vma struct.
 338  */
 339 struct gru_vma_data {
 340         spinlock_t              vd_lock;        /* Serialize access to vma */
 341         struct list_head        vd_head;        /* head of linked list of gts */
 342         long                    vd_user_options;/* misc user option flags */
 343         int                     vd_cbr_au_count;
 344         int                     vd_dsr_au_count;
 345 };
 346
 347 /*
 348  * One of these is allocated for each thread accessing a mmaped GRU. A linked
 349  * list of these structure is hung off the struct gru_vma_data in the mm_struct.
 350  */
 351 struct gru_thread_state {
 352         struct list_head        ts_next;        /* list - head at vma-private */
 353         struct mutex            ts_ctxlock;     /* load/unload CTX lock */
 354         struct mm_struct        *ts_mm;         /* mm currently mapped to
 355                                                    context */
 356         struct vm_area_struct   *ts_vma;        /* vma of GRU context */
 357         struct gru_state        *ts_gru;        /* GRU where the context is
 358                                                    loaded */
 359         struct gru_mm_struct    *ts_gms;        /* asid & ioproc struct */
 360         unsigned long           ts_cbr_map;     /* map of allocated CBRs */
 361         unsigned long           ts_dsr_map;     /* map of allocated DATA
 362                                                    resources */
 363         unsigned long           ts_steal_jiffies;/* jiffies when context last
 364                                                     stolen */
 365         long                    ts_user_options;/* misc user option flags */
 366         pid_t                   ts_tgid_owner;  /* task that is using the
 367                                                    context - for migration */
 368         unsigned short          ts_sizeavail;   /* Pagesizes in use */
 369         int                     ts_tsid;        /* thread that owns the
 370                                                    structure */
 371         int                     ts_tlb_int_select;/* target cpu if interrupts
 372                                                      enabled */
 373         int                     ts_ctxnum;      /* context number where the
 374                                                    context is loaded */
 375         atomic_t                ts_refcnt;      /* reference count GTS */
 376         unsigned char           ts_dsr_au_count;/* Number of DSR resources
 377                                                    required for contest */
 378         unsigned char           ts_cbr_au_count;/* Number of CBR resources
 379                                                    required for contest */
 380         char                    ts_blade;       /* If >= 0, migrate context if
 381                                                    ref from diferent blade */
 382         char                    ts_force_cch_reload;
 383         char                    ts_force_unload;/* force context to be unloaded
 384                                                    after migration */
 385         char                    ts_cbr_idx[GRU_CBR_AU];/* CBR numbers of each
 386                                                           allocated CB */
 387         int                     ts_data_valid;  /* Indicates if ts_gdata has
 388                                                    valid data */
 389         unsigned long           ts_gdata[0];    /* save area for GRU data (CB,
 390                                                    DS, CBE) */
 391 };
 392
 393 /*
 394  * Threaded programs actually allocate an array of GSEGs when a context is
 395  * created. Each thread uses a separate GSEG. TSID is the index into the GSEG
 396  * array.
 397  */
 398 #define TSID(a, v)              (((a) - (v)->vm_start) / GRU_GSEG_PAGESIZE)
 399 #define UGRUADDR(gts)           ((gts)->ts_vma->vm_start +              \
 400                                         (gts)->ts_tsid * GRU_GSEG_PAGESIZE)
 401
 402 #define NULLCTX                 (-1)    /* if context not loaded into GRU */
 403
 404 /*-----------------------------------------------------------------------------
 405  *  GRU State Tables
 406  */
 407
 408 /*
 409  * One of these exists for each GRU chiplet.
 410  */
 411 struct gru_state {
 412         struct gru_blade_state  *gs_blade;              /* GRU state for entire
 413                                                            blade */
 414         unsigned long           gs_gru_base_paddr;      /* Physical address of
 415                                                            gru segments (64) */
 416         void                    *gs_gru_base_vaddr;     /* Virtual address of
 417                                                            gru segments (64) */
 418         unsigned short          gs_gid;                 /* unique GRU number */
 419         unsigned short          gs_blade_id;            /* blade of GRU */
 420         unsigned char           gs_tgh_local_shift;     /* used to pick TGH for
 421                                                            local flush */
 422         unsigned char           gs_tgh_first_remote;    /* starting TGH# for
 423                                                            remote flush */
 424         spinlock_t              gs_asid_lock;           /* lock used for
 425                                                            assigning asids */
 426         spinlock_t              gs_lock;                /* lock used for
 427                                                            assigning contexts */
 428
 429         /* -- the following are protected by the gs_asid_lock spinlock ---- */
 430         unsigned int            gs_asid;                /* Next availe ASID */
 431         unsigned int            gs_asid_limit;          /* Limit of available
 432                                                            ASIDs */
 433         unsigned int            gs_asid_gen;            /* asid generation.
 434                                                            Inc on wrap */
 435
 436         /* --- the following fields are protected by the gs_lock spinlock --- */
 437         unsigned long           gs_context_map;         /* bitmap to manage
 438                                                            contexts in use */
 439         unsigned long           gs_cbr_map;             /* bitmap to manage CB
 440                                                            resources */
 441         unsigned long           gs_dsr_map;             /* bitmap used to manage
 442                                                            DATA resources */
 443         unsigned int            gs_reserved_cbrs;       /* Number of kernel-
 444                                                            reserved cbrs */
 445         unsigned int            gs_reserved_dsr_bytes;  /* Bytes of kernel-
 446                                                            reserved dsrs */
 447         unsigned short          gs_active_contexts;     /* number of contexts
 448                                                            in use */
 449         struct gru_thread_state *gs_gts[GRU_NUM_CCH];   /* GTS currently using
 450                                                            the context */
 451 };
 452
 453 /*
 454  * This structure contains the GRU state for all the GRUs on a blade.
 455  */
 456 struct gru_blade_state {
 457         void                    *kernel_cb;             /* First kernel
 458                                                            reserved cb */
 459         void                    *kernel_dsr;            /* First kernel
 460                                                            reserved DSR */
 461         struct rw_semaphore     bs_kgts_sema;           /* lock for kgts */
 462         struct gru_thread_state *bs_kgts;               /* GTS for kernel use */
 463
 464         /* ---- the following are used for managing kernel async GRU CBRs --- */
 465         int                     bs_async_dsr_bytes;     /* DSRs for async */
 466         int                     bs_async_cbrs;          /* CBRs AU for async */
 467         struct completion       *bs_async_wq;
 468
 469         /* ---- the following are protected by the bs_lock spinlock ---- */
 470         spinlock_t              bs_lock;                /* lock used for
 471                                                            stealing contexts */
 472         int                     bs_lru_ctxnum;          /* STEAL - last context
 473                                                            stolen */
 474         struct gru_state        *bs_lru_gru;            /* STEAL - last gru
 475                                                            stolen */
 476
 477         struct gru_state        bs_grus[GRU_CHIPLETS_PER_BLADE];
 478 };
 479
 480 /*-----------------------------------------------------------------------------
 481  * Address Primitives
 482  */
 483 #define get_tfm_for_cpu(g, c)                                           \
 484         ((struct gru_tlb_fault_map *)get_tfm((g)->gs_gru_base_vaddr, (c)))
 485 #define get_tfh_by_index(g, i)                                          \
 486         ((struct gru_tlb_fault_handle *)get_tfh((g)->gs_gru_base_vaddr, (i)))
 487 #define get_tgh_by_index(g, i)                                          \
 488         ((struct gru_tlb_global_handle *)get_tgh((g)->gs_gru_base_vaddr, (i)))
 489 #define get_cbe_by_index(g, i)                                          \
 490         ((struct gru_control_block_extended *)get_cbe((g)->gs_gru_base_vaddr,\
 491                         (i)))
 492
 493 /*-----------------------------------------------------------------------------
 494  * Useful Macros
 495  */
 496
 497 /* Given a blade# & chiplet#, get a pointer to the GRU */
 498 #define get_gru(b, c)           (&gru_base[b]->bs_grus[c])
 499
 500 /* Number of bytes to save/restore when unloading/loading GRU contexts */
 501 #define DSR_BYTES(dsr)          ((dsr) * GRU_DSR_AU_BYTES)
 502 #define CBR_BYTES(cbr)          ((cbr) * GRU_HANDLE_BYTES * GRU_CBR_AU_SIZE * 2)
 503
 504 /* Convert a user CB number to the actual CBRNUM */
 505 #define thread_cbr_number(gts, n) ((gts)->ts_cbr_idx[(n) / GRU_CBR_AU_SIZE] \
 506                                   * GRU_CBR_AU_SIZE + (n) % GRU_CBR_AU_SIZE)
 507
 508 /* Convert a gid to a pointer to the GRU */
 509 #define GID_TO_GRU(gid)                                                 \
 510         (gru_base[(gid) / GRU_CHIPLETS_PER_BLADE] ?                     \
 511                 (&gru_base[(gid) / GRU_CHIPLETS_PER_BLADE]->            \
 512                         bs_grus[(gid) % GRU_CHIPLETS_PER_BLADE]) :      \
 513          NULL)
 514
 515 /* Scan all active GRUs in a GRU bitmap */
 516 #define for_each_gru_in_bitmap(gid, map)                                \
 517         for ((gid) = find_first_bit((map), GRU_MAX_GRUS); (gid) < GRU_MAX_GRUS;\
 518                 (gid)++, (gid) = find_next_bit((map), GRU_MAX_GRUS, (gid)))
 519
 520 /* Scan all active GRUs on a specific blade */
 521 #define for_each_gru_on_blade(gru, nid, i)                              \
 522         for ((gru) = gru_base[nid]->bs_grus, (i) = 0;                   \
 523                         (i) < GRU_CHIPLETS_PER_BLADE;                   \
 524                         (i)++, (gru)++)
 525
 526 /* Scan all GRUs */
 527 #define foreach_gid(gid)                                                \
 528         for ((gid) = 0; (gid) < gru_max_gids; (gid)++)
 529
 530 /* Scan all active GTSs on a gru. Note: must hold ss_lock to use this macro. */
 531 #define for_each_gts_on_gru(gts, gru, ctxnum)                           \
 532         for ((ctxnum) = 0; (ctxnum) < GRU_NUM_CCH; (ctxnum)++)          \
 533                 if (((gts) = (gru)->gs_gts[ctxnum]))
 534
 535 /* Scan each CBR whose bit is set in a TFM (or copy of) */
 536 #define for_each_cbr_in_tfm(i, map)                                     \
 537         for ((i) = find_first_bit(map, GRU_NUM_CBE);                    \
 538                         (i) < GRU_NUM_CBE;                              \
 539                         (i)++, (i) = find_next_bit(map, GRU_NUM_CBE, i))
 540
 541 /* Scan each CBR in a CBR bitmap. Note: multiple CBRs in an allocation unit */
 542 #define for_each_cbr_in_allocation_map(i, map, k)                       \
 543         for ((k) = find_first_bit(map, GRU_CBR_AU); (k) < GRU_CBR_AU;   \
 544                         (k) = find_next_bit(map, GRU_CBR_AU, (k) + 1))  \
 545                 for ((i) = (k)*GRU_CBR_AU_SIZE;                         \
 546                                 (i) < ((k) + 1) * GRU_CBR_AU_SIZE; (i)++)
 547
 548 /* Scan each DSR in a DSR bitmap. Note: multiple DSRs in an allocation unit */
 549 #define for_each_dsr_in_allocation_map(i, map, k)                       \
 550         for ((k) = find_first_bit((const unsigned long *)map, GRU_DSR_AU);\
 551                         (k) < GRU_DSR_AU;                               \
 552                         (k) = find_next_bit((const unsigned long *)map, \
 553                                           GRU_DSR_AU, (k) + 1))         \
 554                 for ((i) = (k) * GRU_DSR_AU_CL;                         \
 555                                 (i) < ((k) + 1) * GRU_DSR_AU_CL; (i)++)
 556
 557 #define gseg_physical_address(gru, ctxnum)                              \
 558                 ((gru)->gs_gru_base_paddr + ctxnum * GRU_GSEG_STRIDE)
 559 #define gseg_virtual_address(gru, ctxnum)                               \
 560                 ((gru)->gs_gru_base_vaddr + ctxnum * GRU_GSEG_STRIDE)
 561
 562 /*-----------------------------------------------------------------------------
 563  * Lock / Unlock GRU handles
 564  *      Use the "delresp" bit in the handle as a "lock" bit.
 565  */
 566
 567 /* Lock hierarchy checking enabled only in emulator */
 568
 569 /* 0 = lock failed, 1 = locked */
 570 static inline int __trylock_handle(void *h)
 571 {
 572         return !test_and_set_bit(1, h);
 573 }
 574
 575 static inline void __lock_handle(void *h)
 576 {
 577         while (test_and_set_bit(1, h))
 578                 cpu_relax();
 579 }
 580
 581 static inline void __unlock_handle(void *h)
 582 {
 583         clear_bit(1, h);
 584 }
 585
 586 static inline int trylock_cch_handle(struct gru_context_configuration_handle *cch)
 587 {
 588         return __trylock_handle(cch);
 589 }
 590
 591 static inline void lock_cch_handle(struct gru_context_configuration_handle *cch)
 592 {
 593         __lock_handle(cch);
 594 }
 595
 596 static inline void unlock_cch_handle(struct gru_context_configuration_handle
 597                                      *cch)
 598 {
 599         __unlock_handle(cch);
 600 }
 601
 602 static inline void lock_tgh_handle(struct gru_tlb_global_handle *tgh)
 603 {
 604         __lock_handle(tgh);
 605 }
 606
 607 static inline void unlock_tgh_handle(struct gru_tlb_global_handle *tgh)
 608 {
 609         __unlock_handle(tgh);
 610 }
 611
 612 static inline int is_kernel_context(struct gru_thread_state *gts)
 613 {
 614         return !gts->ts_mm;
 615 }
 616
 617 /*-----------------------------------------------------------------------------
 618  * Function prototypes & externs
 619  */
 620 struct gru_unload_context_req;
 621
 622 extern struct vm_operations_struct gru_vm_ops;
 623 extern struct device *grudev;
 624
 625 extern struct gru_vma_data *gru_alloc_vma_data(struct vm_area_struct *vma,
 626                                 int tsid);
 627 extern struct gru_thread_state *gru_find_thread_state(struct vm_area_struct
 628                                 *vma, int tsid);
 629 extern struct gru_thread_state *gru_alloc_thread_state(struct vm_area_struct
 630                                 *vma, int tsid);
 631 extern struct gru_state *gru_assign_gru_context(struct gru_thread_state *gts,
 632                 int blade);
 633 extern void gru_load_context(struct gru_thread_state *gts);
 634 extern void gru_steal_context(struct gru_thread_state *gts, int blade_id);
 635 extern void gru_unload_context(struct gru_thread_state *gts, int savestate);
 636 extern int gru_update_cch(struct gru_thread_state *gts, int force_unload);
 637 extern void gts_drop(struct gru_thread_state *gts);
 638 extern void gru_tgh_flush_init(struct gru_state *gru);
 639 extern int gru_kservices_init(struct gru_state *gru);
 640 extern void gru_kservices_exit(struct gru_state *gru);
 641 extern int gru_dump_chiplet_request(unsigned long arg);
 642 extern irqreturn_t gru_intr(int irq, void *dev_id);
 643 extern int gru_handle_user_call_os(unsigned long address);
 644 extern int gru_user_flush_tlb(unsigned long arg);
 645 extern int gru_user_unload_context(unsigned long arg);
 646 extern int gru_get_exception_detail(unsigned long arg);
 647 extern int gru_set_task_slice(long address);
 648 extern int gru_cpu_fault_map_id(void);
 649 extern struct vm_area_struct *gru_find_vma(unsigned long vaddr);
 650 extern void gru_flush_all_tlb(struct gru_state *gru);
 651 extern int gru_proc_init(void);
 652 extern void gru_proc_exit(void);
 653
 654 extern struct gru_thread_state *gru_alloc_gts(struct vm_area_struct *vma,
 655                 int cbr_au_count, int dsr_au_count, int options, int tsid);
 656 extern unsigned long gru_reserve_cb_resources(struct gru_state *gru,
 657                 int cbr_au_count, char *cbmap);
 658 extern unsigned long gru_reserve_ds_resources(struct gru_state *gru,
 659                 int dsr_au_count, char *dsmap);
 660 extern int gru_fault(struct vm_area_struct *, struct vm_fault *vmf);
 661 extern struct gru_mm_struct *gru_register_mmu_notifier(void);
 662 extern void gru_drop_mmu_notifier(struct gru_mm_struct *gms);
 663
 664 extern int gru_ktest(unsigned long arg);
 665 extern void gru_flush_tlb_range(struct gru_mm_struct *gms, unsigned long start,
 666                                         unsigned long len);
 667
 668 extern unsigned long gru_options;
 669
 670 #endif /* __GRUTABLES_H__ */