drivers/misc/sgi-gru/grufile.c

   1 /*
   2  * SN Platform GRU Driver
   3  *
   4  *              FILE OPERATIONS & DRIVER INITIALIZATION
   5  *
   6  * This file supports the user system call for file open, close, mmap, etc.
   7  * This also incudes the driver initialization code.
   8  *
   9  *  Copyright (c) 2008 Silicon Graphics, Inc.  All Rights Reserved.
  10  *
  11  *  This program is free software; you can redistribute it and/or modify
  12  *  it under the terms of the GNU General Public License as published by
  13  *  the Free Software Foundation; either version 2 of the License, or
  14  *  (at your option) any later version.
  15  *
  16  *  This program is distributed in the hope that it will be useful,
  17  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  19  *  GNU General Public License for more details.
  20  *
  21  *  You should have received a copy of the GNU General Public License
  22  *  along with this program; if not, write to the Free Software
  23  *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
  24  */
  25
  26 #include <linux/module.h>
  27 #include <linux/kernel.h>
  28 #include <linux/errno.h>
  29 #include <linux/slab.h>
  30 #include <linux/mm.h>
  31 #include <linux/io.h>
  32 #include <linux/smp_lock.h>
  33 #include <linux/spinlock.h>
  34 #include <linux/device.h>
  35 #include <linux/miscdevice.h>
  36 #include <linux/interrupt.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/uaccess.h>
  39 #include <asm/uv/uv.h>
  40 #include "gru.h"
  41 #include "grulib.h"
  42 #include "grutables.h"
  43
  44 #include <asm/uv/uv_hub.h>
  45 #include <asm/uv/uv_mmrs.h>
  46
  47 struct gru_blade_state *gru_base[GRU_MAX_BLADES] __read_mostly;
  48 unsigned long gru_start_paddr __read_mostly;
  49 unsigned long gru_end_paddr __read_mostly;
  50 unsigned int gru_max_gids __read_mostly;
  51 struct gru_stats_s gru_stats;
  52
  53 /* Guaranteed user available resources on each node */
  54 static int max_user_cbrs, max_user_dsr_bytes;
  55
  56 static struct file_operations gru_fops;
  57 static struct miscdevice gru_miscdev;
  58
  59
  60 /*
  61  * gru_vma_close
  62  *
  63  * Called when unmapping a device mapping. Frees all gru resources
  64  * and tables belonging to the vma.
  65  */
  66 static void gru_vma_close(struct vm_area_struct *vma)
  67 {
  68         struct gru_vma_data *vdata;
  69         struct gru_thread_state *gts;
  70         struct list_head *entry, *next;
  71
  72         if (!vma->vm_private_data)
  73                 return;
  74
  75         vdata = vma->vm_private_data;
  76         vma->vm_private_data = NULL;
  77         gru_dbg(grudev, "vma %p, file %p, vdata %p\n", vma, vma->vm_file,
  78                                 vdata);
  79         list_for_each_safe(entry, next, &vdata->vd_head) {
  80                 gts =
  81                     list_entry(entry, struct gru_thread_state, ts_next);
  82                 list_del(&gts->ts_next);
  83                 mutex_lock(&gts->ts_ctxlock);
  84                 if (gts->ts_gru)
  85                         gru_unload_context(gts, 0);
  86                 mutex_unlock(&gts->ts_ctxlock);
  87                 gts_drop(gts);
  88         }
  89         kfree(vdata);
  90         STAT(vdata_free);
  91 }
  92
  93 /*
  94  * gru_file_mmap
  95  *
  96  * Called when mmaping the device.  Initializes the vma with a fault handler
  97  * and private data structure necessary to allocate, track, and free the
  98  * underlying pages.
  99  */
 100 static int gru_file_mmap(struct file *file, struct vm_area_struct *vma)
 101 {
 102         if ((vma->vm_flags & (VM_SHARED | VM_WRITE)) != (VM_SHARED | VM_WRITE))
 103                 return -EPERM;
 104
 105         if (vma->vm_start & (GRU_GSEG_PAGESIZE - 1) ||
 106                                 vma->vm_end & (GRU_GSEG_PAGESIZE - 1))
 107                 return -EINVAL;
 108
 109         vma->vm_flags |=
 110             (VM_IO | VM_DONTCOPY | VM_LOCKED | VM_DONTEXPAND | VM_PFNMAP |
 111                         VM_RESERVED);
 112         vma->vm_page_prot = PAGE_SHARED;
 113         vma->vm_ops = &gru_vm_ops;
 114
 115         vma->vm_private_data = gru_alloc_vma_data(vma, 0);
 116         if (!vma->vm_private_data)
 117                 return -ENOMEM;
 118
 119         gru_dbg(grudev, "file %p, vaddr 0x%lx, vma %p, vdata %p\n",
 120                 file, vma->vm_start, vma, vma->vm_private_data);
 121         return 0;
 122 }
 123
 124 /*
 125  * Create a new GRU context
 126  */
 127 static int gru_create_new_context(unsigned long arg)
 128 {
 129         struct gru_create_context_req req;
 130         struct vm_area_struct *vma;
 131         struct gru_vma_data *vdata;
 132         int ret = -EINVAL;
 133
 134
 135         if (copy_from_user(&req, (void __user *)arg, sizeof(req)))
 136                 return -EFAULT;
 137
 138         if (req.data_segment_bytes > max_user_dsr_bytes)
 139                 return -EINVAL;
 140         if (req.control_blocks > max_user_cbrs || !req.maximum_thread_count)
 141                 return -EINVAL;
 142
 143         if (!(req.options & GRU_OPT_MISS_MASK))
 144                 req.options |= GRU_OPT_MISS_FMM_INTR;
 145
 146         down_write(&current->mm->mmap_sem);
 147         vma = gru_find_vma(req.gseg);
 148         if (vma) {
 149                 vdata = vma->vm_private_data;
 150                 vdata->vd_user_options = req.options;
 151                 vdata->vd_dsr_au_count =
 152                     GRU_DS_BYTES_TO_AU(req.data_segment_bytes);
 153                 vdata->vd_cbr_au_count = GRU_CB_COUNT_TO_AU(req.control_blocks);
 154                 ret = 0;
 155         }
 156         up_write(&current->mm->mmap_sem);
 157
 158         return ret;
 159 }
 160
 161 /*
 162  * Get GRU configuration info (temp - for emulator testing)
 163  */
 164 static long gru_get_config_info(unsigned long arg)
 165 {
 166         struct gru_config_info info;
 167         int nodesperblade;
 168
 169         if (num_online_nodes() > 1 &&
 170                         (uv_node_to_blade_id(1) == uv_node_to_blade_id(0)))
 171                 nodesperblade = 2;
 172         else
 173                 nodesperblade = 1;
 174         info.cpus = num_online_cpus();
 175         info.nodes = num_online_nodes();
 176         info.blades = info.nodes / nodesperblade;
 177         info.chiplets = GRU_CHIPLETS_PER_BLADE * info.blades;
 178
 179         if (copy_to_user((void __user *)arg, &info, sizeof(info)))
 180                 return -EFAULT;
 181         return 0;
 182 }
 183
 184 /*
 185  * Get GRU chiplet status
 186  */
 187 static long gru_get_chiplet_status(unsigned long arg)
 188 {
 189         struct gru_state *gru;
 190         struct gru_chiplet_info info;
 191
 192         if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
 193                 return -EFAULT;
 194
 195         if (info.node == -1)
 196                 info.node = numa_node_id();
 197         if (info.node >= num_possible_nodes() ||
 198                         info.chiplet >= GRU_CHIPLETS_PER_HUB ||
 199                         info.node < 0 || info.chiplet < 0)
 200                 return -EINVAL;
 201
 202         info.blade = uv_node_to_blade_id(info.node);
 203         gru = get_gru(info.blade, info.chiplet);
 204
 205         info.total_dsr_bytes = GRU_NUM_DSR_BYTES;
 206         info.total_cbr = GRU_NUM_CB;
 207         info.total_user_dsr_bytes = GRU_NUM_DSR_BYTES -
 208                 gru->gs_reserved_dsr_bytes;
 209         info.total_user_cbr = GRU_NUM_CB - gru->gs_reserved_cbrs;
 210         info.free_user_dsr_bytes = hweight64(gru->gs_dsr_map) *
 211                         GRU_DSR_AU_BYTES;
 212         info.free_user_cbr = hweight64(gru->gs_cbr_map) * GRU_CBR_AU_SIZE;
 213
 214         if (copy_to_user((void __user *)arg, &info, sizeof(info)))
 215                 return -EFAULT;
 216         return 0;
 217 }
 218
 219 /*
 220  * gru_file_unlocked_ioctl
 221  *
 222  * Called to update file attributes via IOCTL calls.
 223  */
 224 static long gru_file_unlocked_ioctl(struct file *file, unsigned int req,
 225                                     unsigned long arg)
 226 {
 227         int err = -EBADRQC;
 228
 229         gru_dbg(grudev, "file %p\n", file);
 230
 231         switch (req) {
 232         case GRU_CREATE_CONTEXT:
 233                 err = gru_create_new_context(arg);
 234                 break;
 235         case GRU_SET_TASK_SLICE:
 236                 err = gru_set_task_slice(arg);
 237                 break;
 238         case GRU_USER_GET_EXCEPTION_DETAIL:
 239                 err = gru_get_exception_detail(arg);
 240                 break;
 241         case GRU_USER_UNLOAD_CONTEXT:
 242                 err = gru_user_unload_context(arg);
 243                 break;
 244         case GRU_GET_CHIPLET_STATUS:
 245                 err = gru_get_chiplet_status(arg);
 246                 break;
 247         case GRU_USER_FLUSH_TLB:
 248                 err = gru_user_flush_tlb(arg);
 249                 break;
 250         case GRU_USER_CALL_OS:
 251                 err = gru_handle_user_call_os(arg);
 252                 break;
 253         case GRU_KTEST:
 254                 err = gru_ktest(arg);
 255                 break;
 256         case GRU_GET_CONFIG_INFO:
 257                 err = gru_get_config_info(arg);
 258                 break;
 259         case GRU_DUMP_CHIPLET_STATE:
 260                 err = gru_dump_chiplet_request(arg);
 261                 break;
 262         }
 263         return err;
 264 }
 265
 266 /*
 267  * Called at init time to build tables for all GRUs that are present in the
 268  * system.
 269  */
 270 static void gru_init_chiplet(struct gru_state *gru, unsigned long paddr,
 271                              void *vaddr, int nid, int bid, int grunum)
 272 {
 273         spin_lock_init(&gru->gs_lock);
 274         spin_lock_init(&gru->gs_asid_lock);
 275         gru->gs_gru_base_paddr = paddr;
 276         gru->gs_gru_base_vaddr = vaddr;
 277         gru->gs_gid = bid * GRU_CHIPLETS_PER_BLADE + grunum;
 278         gru->gs_blade = gru_base[bid];
 279         gru->gs_blade_id = bid;
 280         gru->gs_cbr_map = (GRU_CBR_AU == 64) ? ~0 : (1UL << GRU_CBR_AU) - 1;
 281         gru->gs_dsr_map = (1UL << GRU_DSR_AU) - 1;
 282         gru->gs_asid_limit = MAX_ASID;
 283         gru_tgh_flush_init(gru);
 284         if (gru->gs_gid >= gru_max_gids)
 285                 gru_max_gids = gru->gs_gid + 1;
 286         gru_dbg(grudev, "bid %d, nid %d, gid %d, vaddr %p (0x%lx)\n",
 287                 bid, nid, gru->gs_gid, gru->gs_gru_base_vaddr,
 288                 gru->gs_gru_base_paddr);
 289         gru_kservices_init(gru);
 290 }
 291
 292 static int gru_init_tables(unsigned long gru_base_paddr, void *gru_base_vaddr)
 293 {
 294         int pnode, nid, bid, chip;
 295         int cbrs, dsrbytes, n;
 296         int order = get_order(sizeof(struct gru_blade_state));
 297         struct page *page;
 298         struct gru_state *gru;
 299         unsigned long paddr;
 300         void *vaddr;
 301
 302         max_user_cbrs = GRU_NUM_CB;
 303         max_user_dsr_bytes = GRU_NUM_DSR_BYTES;
 304         for_each_online_node(nid) {
 305                 bid = uv_node_to_blade_id(nid);
 306                 pnode = uv_node_to_pnode(nid);
 307                 if (bid < 0 || gru_base[bid])
 308                         continue;
 309                 page = alloc_pages_exact_node(nid, GFP_KERNEL, order);
 310                 if (!page)
 311                         goto fail;
 312                 gru_base[bid] = page_address(page);
 313                 memset(gru_base[bid], 0, sizeof(struct gru_blade_state));
 314                 gru_base[bid]->bs_lru_gru = &gru_base[bid]->bs_grus[0];
 315                 spin_lock_init(&gru_base[bid]->bs_lock);
 316
 317                 dsrbytes = 0;
 318                 cbrs = 0;
 319                 for (gru = gru_base[bid]->bs_grus, chip = 0;
 320                                 chip < GRU_CHIPLETS_PER_BLADE;
 321                                 chip++, gru++) {
 322                         paddr = gru_chiplet_paddr(gru_base_paddr, pnode, chip);
 323                         vaddr = gru_chiplet_vaddr(gru_base_vaddr, pnode, chip);
 324                         gru_init_chiplet(gru, paddr, vaddr, nid, bid, chip);
 325                         n = hweight64(gru->gs_cbr_map) * GRU_CBR_AU_SIZE;
 326                         cbrs = max(cbrs, n);
 327                         n = hweight64(gru->gs_dsr_map) * GRU_DSR_AU_BYTES;
 328                         dsrbytes = max(dsrbytes, n);
 329                 }
 330                 max_user_cbrs = min(max_user_cbrs, cbrs);
 331                 max_user_dsr_bytes = min(max_user_dsr_bytes, dsrbytes);
 332         }
 333
 334         return 0;
 335
 336 fail:
 337         for (nid--; nid >= 0; nid--)
 338                 free_pages((unsigned long)gru_base[nid], order);
 339         return -ENOMEM;
 340 }
 341
 342 #ifdef CONFIG_IA64
 343
 344 static int get_base_irq(void)
 345 {
 346         return IRQ_GRU;
 347 }
 348
 349 #elif defined CONFIG_X86_64
 350
 351 static void noop(unsigned int irq)
 352 {
 353 }
 354
 355 static struct irq_chip gru_chip = {
 356         .name           = "gru",
 357         .mask           = noop,
 358         .unmask         = noop,
 359         .ack            = noop,
 360 };
 361
 362 static int get_base_irq(void)
 363 {
 364         set_irq_chip(IRQ_GRU, &gru_chip);
 365         set_irq_chip(IRQ_GRU + 1, &gru_chip);
 366         return IRQ_GRU;
 367 }
 368 #endif
 369
 370 /*
 371  * gru_init
 372  *
 373  * Called at boot or module load time to initialize the GRUs.
 374  */
 375 static int __init gru_init(void)
 376 {
 377         int ret, irq, chip;
 378         char id[10];
 379         void *gru_start_vaddr;
 380
 381         if (!is_uv_system())
 382                 return 0;
 383
 384 #if defined CONFIG_IA64
 385         gru_start_paddr = 0xd000000000UL; /* ZZZZZZZZZZZZZZZZZZZ fixme */
 386 #else
 387         gru_start_paddr = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR) &
 388                                 0x7fffffffffffUL;
 389 #endif
 390         gru_start_vaddr = __va(gru_start_paddr);
 391         gru_end_paddr = gru_start_paddr + GRU_MAX_BLADES * GRU_SIZE;
 392         printk(KERN_INFO "GRU space: 0x%lx - 0x%lx\n",
 393                gru_start_paddr, gru_end_paddr);
 394         irq = get_base_irq();
 395         for (chip = 0; chip < GRU_CHIPLETS_PER_BLADE; chip++) {
 396                 ret = request_irq(irq + chip, gru_intr, 0, id, NULL);
 397                 /* TODO: fix irq handling on x86. For now ignore failure because
 398                  * interrupts are not required & not yet fully supported */
 399                 if (ret) {
 400                         printk(KERN_WARNING
 401                                "!!!WARNING: GRU ignoring request failure!!!\n");
 402                         ret = 0;
 403                 }
 404                 if (ret) {
 405                         printk(KERN_ERR "%s: request_irq failed\n",
 406                                GRU_DRIVER_ID_STR);
 407                         goto exit1;
 408                 }
 409         }
 410
 411         ret = misc_register(&gru_miscdev);
 412         if (ret) {
 413                 printk(KERN_ERR "%s: misc_register failed\n",
 414                        GRU_DRIVER_ID_STR);
 415                 goto exit1;
 416         }
 417
 418         ret = gru_proc_init();
 419         if (ret) {
 420                 printk(KERN_ERR "%s: proc init failed\n", GRU_DRIVER_ID_STR);
 421                 goto exit2;
 422         }
 423
 424         ret = gru_init_tables(gru_start_paddr, gru_start_vaddr);
 425         if (ret) {
 426                 printk(KERN_ERR "%s: init tables failed\n", GRU_DRIVER_ID_STR);
 427                 goto exit3;
 428         }
 429
 430         printk(KERN_INFO "%s: v%s\n", GRU_DRIVER_ID_STR,
 431                GRU_DRIVER_VERSION_STR);
 432         return 0;
 433
 434 exit3:
 435         gru_proc_exit();
 436 exit2:
 437         misc_deregister(&gru_miscdev);
 438 exit1:
 439         for (--chip; chip >= 0; chip--)
 440                 free_irq(irq + chip, NULL);
 441         return ret;
 442
 443 }
 444
 445 static void __exit gru_exit(void)
 446 {
 447         int i, bid, gid;
 448         int order = get_order(sizeof(struct gru_state) *
 449                               GRU_CHIPLETS_PER_BLADE);
 450
 451         if (!is_uv_system())
 452                 return;
 453
 454         for (i = 0; i < GRU_CHIPLETS_PER_BLADE; i++)
 455                 free_irq(IRQ_GRU + i, NULL);
 456
 457         foreach_gid(gid)
 458                 gru_kservices_exit(GID_TO_GRU(gid));
 459
 460         for (bid = 0; bid < GRU_MAX_BLADES; bid++)
 461                 free_pages((unsigned long)gru_base[bid], order);
 462
 463         misc_deregister(&gru_miscdev);
 464         gru_proc_exit();
 465 }
 466
 467 static struct file_operations gru_fops = {
 468         .owner          = THIS_MODULE,
 469         .unlocked_ioctl = gru_file_unlocked_ioctl,
 470         .mmap           = gru_file_mmap,
 471 };
 472
 473 static struct miscdevice gru_miscdev = {
 474         .minor          = MISC_DYNAMIC_MINOR,
 475         .name           = "gru",
 476         .fops           = &gru_fops,
 477 };
 478
 479 struct vm_operations_struct gru_vm_ops = {
 480         .close          = gru_vma_close,
 481         .fault          = gru_fault,
 482 };
 483
 484 #ifndef MODULE
 485 fs_initcall(gru_init);
 486 #else
 487 module_init(gru_init);
 488 #endif
 489 module_exit(gru_exit);
 490
 491 module_param(gru_options, ulong, 0644);
 492 MODULE_PARM_DESC(gru_options, "Various debug options");
 493
 494 MODULE_AUTHOR("Silicon Graphics, Inc.");
 495 MODULE_LICENSE("GPL");
 496 MODULE_DESCRIPTION(GRU_DRIVER_ID_STR GRU_DRIVER_VERSION_STR);
 497 MODULE_VERSION(GRU_DRIVER_VERSION_STR);
 498