drivers/char/mspec.c

   1 /*
   2  * Copyright (C) 2001-2006 Silicon Graphics, Inc.  All rights
   3  * reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms of version 2 of the GNU General Public License
   7  * as published by the Free Software Foundation.
   8  */
   9
  10 /*
  11  * SN Platform Special Memory (mspec) Support
  12  *
  13  * This driver exports the SN special memory (mspec) facility to user
  14  * processes.
  15  * There are three types of memory made available thru this driver:
  16  * fetchops, uncached and cached.
  17  *
  18  * Fetchops are atomic memory operations that are implemented in the
  19  * memory controller on SGI SN hardware.
  20  *
  21  * Uncached are used for memory write combining feature of the ia64
  22  * cpu.
  23  *
  24  * Cached are used for areas of memory that are used as cached addresses
  25  * on our partition and used as uncached addresses from other partitions.
  26  * Due to a design constraint of the SN2 Shub, you can not have processors
  27  * on the same FSB perform both a cached and uncached reference to the
  28  * same cache line.  These special memory cached regions prevent the
  29  * kernel from ever dropping in a TLB entry and therefore prevent the
  30  * processor from ever speculating a cache line from this page.
  31  */
  32
  33 #include <linux/config.h>
  34 #include <linux/types.h>
  35 #include <linux/kernel.h>
  36 #include <linux/module.h>
  37 #include <linux/init.h>
  38 #include <linux/errno.h>
  39 #include <linux/miscdevice.h>
  40 #include <linux/spinlock.h>
  41 #include <linux/mm.h>
  42 #include <linux/vmalloc.h>
  43 #include <linux/string.h>
  44 #include <linux/slab.h>
  45 #include <linux/numa.h>
  46 #include <asm/page.h>
  47 #include <asm/system.h>
  48 #include <asm/pgtable.h>
  49 #include <asm/atomic.h>
  50 #include <asm/tlbflush.h>
  51 #include <asm/uncached.h>
  52 #include <asm/sn/addrs.h>
  53 #include <asm/sn/arch.h>
  54 #include <asm/sn/mspec.h>
  55 #include <asm/sn/sn_cpuid.h>
  56 #include <asm/sn/io.h>
  57 #include <asm/sn/bte.h>
  58 #include <asm/sn/shubio.h>
  59
  60
  61 #define FETCHOP_ID      "SGI Fetchop,"
  62 #define CACHED_ID       "Cached,"
  63 #define UNCACHED_ID     "Uncached"
  64 #define REVISION        "4.0"
  65 #define MSPEC_BASENAME  "mspec"
  66
  67 /*
  68  * Page types allocated by the device.
  69  */
  70 enum {
  71         MSPEC_FETCHOP = 1,
  72         MSPEC_CACHED,
  73         MSPEC_UNCACHED
  74 };
  75
  76 static int is_sn2;
  77
  78 /*
  79  * One of these structures is allocated when an mspec region is mmaped. The
  80  * structure is pointed to by the vma->vm_private_data field in the vma struct.
  81  * This structure is used to record the addresses of the mspec pages.
  82  */
  83 struct vma_data {
  84         atomic_t refcnt;        /* Number of vmas sharing the data. */
  85         spinlock_t lock;        /* Serialize access to the vma. */
  86         int count;              /* Number of pages allocated. */
  87         int type;               /* Type of pages allocated. */
  88         unsigned long maddr[0]; /* Array of MSPEC addresses. */
  89 };
  90
  91 /* used on shub2 to clear FOP cache in the HUB */
  92 static unsigned long scratch_page[MAX_NUMNODES];
  93 #define SH2_AMO_CACHE_ENTRIES   4
  94
  95 static inline int
  96 mspec_zero_block(unsigned long addr, int len)
  97 {
  98         int status;
  99
 100         if (is_sn2) {
 101                 if (is_shub2()) {
 102                         int nid;
 103                         void *p;
 104                         int i;
 105
 106                         nid = nasid_to_cnodeid(get_node_number(__pa(addr)));
 107                         p = (void *)TO_AMO(scratch_page[nid]);
 108
 109                         for (i=0; i < SH2_AMO_CACHE_ENTRIES; i++) {
 110                                 FETCHOP_LOAD_OP(p, FETCHOP_LOAD);
 111                                 p += FETCHOP_VAR_SIZE;
 112                         }
 113                 }
 114
 115                 status = bte_copy(0, addr & ~__IA64_UNCACHED_OFFSET, len,
 116                                   BTE_WACQUIRE | BTE_ZERO_FILL, NULL);
 117         } else {
 118                 memset((char *) addr, 0, len);
 119                 status = 0;
 120         }
 121         return status;
 122 }
 123
 124 /*
 125  * mspec_open
 126  *
 127  * Called when a device mapping is created by a means other than mmap
 128  * (via fork, etc.).  Increments the reference count on the underlying
 129  * mspec data so it is not freed prematurely.
 130  */
 131 static void
 132 mspec_open(struct vm_area_struct *vma)
 133 {
 134         struct vma_data *vdata;
 135
 136         vdata = vma->vm_private_data;
 137         atomic_inc(&vdata->refcnt);
 138 }
 139
 140 /*
 141  * mspec_close
 142  *
 143  * Called when unmapping a device mapping. Frees all mspec pages
 144  * belonging to the vma.
 145  */
 146 static void
 147 mspec_close(struct vm_area_struct *vma)
 148 {
 149         struct vma_data *vdata;
 150         int i, pages, result, vdata_size;
 151
 152         vdata = vma->vm_private_data;
 153         if (!atomic_dec_and_test(&vdata->refcnt))
 154                 return;
 155
 156         pages = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
 157         vdata_size = sizeof(struct vma_data) + pages * sizeof(long);
 158         for (i = 0; i < pages; i++) {
 159                 if (vdata->maddr[i] == 0)
 160                         continue;
 161                 /*
 162                  * Clear the page before sticking it back
 163                  * into the pool.
 164                  */
 165                 result = mspec_zero_block(vdata->maddr[i], PAGE_SIZE);
 166                 if (!result)
 167                         uncached_free_page(vdata->maddr[i]);
 168                 else
 169                         printk(KERN_WARNING "mspec_close(): "
 170                                "failed to zero page %i\n",
 171                                result);
 172         }
 173
 174         if (vdata_size <= PAGE_SIZE)
 175                 kfree(vdata);
 176         else
 177                 vfree(vdata);
 178 }
 179
 180
 181 /*
 182  * mspec_nopfn
 183  *
 184  * Creates a mspec page and maps it to user space.
 185  */
 186 static unsigned long
 187 mspec_nopfn(struct vm_area_struct *vma, unsigned long address)
 188 {
 189         unsigned long paddr, maddr;
 190         unsigned long pfn;
 191         int index;
 192         struct vma_data *vdata = vma->vm_private_data;
 193
 194         index = (address - vma->vm_start) >> PAGE_SHIFT;
 195         maddr = (volatile unsigned long) vdata->maddr[index];
 196         if (maddr == 0) {
 197                 maddr = uncached_alloc_page(numa_node_id());
 198                 if (maddr == 0)
 199                         return NOPFN_OOM;
 200
 201                 spin_lock(&vdata->lock);
 202                 if (vdata->maddr[index] == 0) {
 203                         vdata->count++;
 204                         vdata->maddr[index] = maddr;
 205                 } else {
 206                         uncached_free_page(maddr);
 207                         maddr = vdata->maddr[index];
 208                 }
 209                 spin_unlock(&vdata->lock);
 210         }
 211
 212         if (vdata->type == MSPEC_FETCHOP)
 213                 paddr = TO_AMO(maddr);
 214         else
 215                 paddr = __pa(TO_CAC(maddr));
 216
 217         pfn = paddr >> PAGE_SHIFT;
 218
 219         return pfn;
 220 }
 221
 222 static struct vm_operations_struct mspec_vm_ops = {
 223         .open = mspec_open,
 224         .close = mspec_close,
 225         .nopfn = mspec_nopfn
 226 };
 227
 228 /*
 229  * mspec_mmap
 230  *
 231  * Called when mmaping the device.  Initializes the vma with a fault handler
 232  * and private data structure necessary to allocate, track, and free the
 233  * underlying pages.
 234  */
 235 static int
 236 mspec_mmap(struct file *file, struct vm_area_struct *vma, int type)
 237 {
 238         struct vma_data *vdata;
 239         int pages, vdata_size;
 240
 241         if (vma->vm_pgoff != 0)
 242                 return -EINVAL;
 243
 244         if ((vma->vm_flags & VM_SHARED) == 0)
 245                 return -EINVAL;
 246
 247         if ((vma->vm_flags & VM_WRITE) == 0)
 248                 return -EPERM;
 249
 250         pages = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
 251         vdata_size = sizeof(struct vma_data) + pages * sizeof(long);
 252         if (vdata_size <= PAGE_SIZE)
 253                 vdata = kmalloc(vdata_size, GFP_KERNEL);
 254         else
 255                 vdata = vmalloc(vdata_size);
 256         if (!vdata)
 257                 return -ENOMEM;
 258         memset(vdata, 0, vdata_size);
 259
 260         vdata->type = type;
 261         spin_lock_init(&vdata->lock);
 262         vdata->refcnt = ATOMIC_INIT(1);
 263         vma->vm_private_data = vdata;
 264
 265         vma->vm_flags |= (VM_IO | VM_LOCKED | VM_RESERVED | VM_PFNMAP);
 266         if (vdata->type == MSPEC_FETCHOP || vdata->type == MSPEC_UNCACHED)
 267                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
 268         vma->vm_ops = &mspec_vm_ops;
 269
 270         return 0;
 271 }
 272
 273 static int
 274 fetchop_mmap(struct file *file, struct vm_area_struct *vma)
 275 {
 276         return mspec_mmap(file, vma, MSPEC_FETCHOP);
 277 }
 278
 279 static int
 280 cached_mmap(struct file *file, struct vm_area_struct *vma)
 281 {
 282         return mspec_mmap(file, vma, MSPEC_CACHED);
 283 }
 284
 285 static int
 286 uncached_mmap(struct file *file, struct vm_area_struct *vma)
 287 {
 288         return mspec_mmap(file, vma, MSPEC_UNCACHED);
 289 }
 290
 291 static struct file_operations fetchop_fops = {
 292         .owner = THIS_MODULE,
 293         .mmap = fetchop_mmap
 294 };
 295
 296 static struct miscdevice fetchop_miscdev = {
 297         .minor = MISC_DYNAMIC_MINOR,
 298         .name = "sgi_fetchop",
 299         .fops = &fetchop_fops
 300 };
 301
 302 static struct file_operations cached_fops = {
 303         .owner = THIS_MODULE,
 304         .mmap = cached_mmap
 305 };
 306
 307 static struct miscdevice cached_miscdev = {
 308         .minor = MISC_DYNAMIC_MINOR,
 309         .name = "mspec_cached",
 310         .fops = &cached_fops
 311 };
 312
 313 static struct file_operations uncached_fops = {
 314         .owner = THIS_MODULE,
 315         .mmap = uncached_mmap
 316 };
 317
 318 static struct miscdevice uncached_miscdev = {
 319         .minor = MISC_DYNAMIC_MINOR,
 320         .name = "mspec_uncached",
 321         .fops = &uncached_fops
 322 };
 323
 324 /*
 325  * mspec_init
 326  *
 327  * Called at boot time to initialize the mspec facility.
 328  */
 329 static int __init
 330 mspec_init(void)
 331 {
 332         int ret;
 333         int nid;
 334
 335         /*
 336          * The fetchop device only works on SN2 hardware, uncached and cached
 337          * memory drivers should both be valid on all ia64 hardware
 338          */
 339         if (ia64_platform_is("sn2")) {
 340                 is_sn2 = 1;
 341                 if (is_shub2()) {
 342                         ret = -ENOMEM;
 343                         for_each_online_node(nid) {
 344                                 int actual_nid;
 345                                 int nasid;
 346                                 unsigned long phys;
 347
 348                                 scratch_page[nid] = uncached_alloc_page(nid);
 349                                 if (scratch_page[nid] == 0)
 350                                         goto free_scratch_pages;
 351                                 phys = __pa(scratch_page[nid]);
 352                                 nasid = get_node_number(phys);
 353                                 actual_nid = nasid_to_cnodeid(nasid);
 354                                 if (actual_nid != nid)
 355                                         goto free_scratch_pages;
 356                         }
 357                 }
 358
 359                 ret = misc_register(&fetchop_miscdev);
 360                 if (ret) {
 361                         printk(KERN_ERR
 362                                "%s: failed to register device %i\n",
 363                                FETCHOP_ID, ret);
 364                         goto free_scratch_pages;
 365                 }
 366         }
 367         ret = misc_register(&cached_miscdev);
 368         if (ret) {
 369                 printk(KERN_ERR "%s: failed to register device %i\n",
 370                        CACHED_ID, ret);
 371                 if (is_sn2)
 372                         misc_deregister(&fetchop_miscdev);
 373                 goto free_scratch_pages;
 374         }
 375         ret = misc_register(&uncached_miscdev);
 376         if (ret) {
 377                 printk(KERN_ERR "%s: failed to register device %i\n",
 378                        UNCACHED_ID, ret);
 379                 misc_deregister(&cached_miscdev);
 380                 if (is_sn2)
 381                         misc_deregister(&fetchop_miscdev);
 382                 goto free_scratch_pages;
 383         }
 384
 385         printk(KERN_INFO "%s %s initialized devices: %s %s %s\n",
 386                MSPEC_BASENAME, REVISION, is_sn2 ? FETCHOP_ID : "",
 387                CACHED_ID, UNCACHED_ID);
 388
 389         return 0;
 390
 391  free_scratch_pages:
 392         for_each_node(nid) {
 393                 if (scratch_page[nid] != 0)
 394                         uncached_free_page(scratch_page[nid]);
 395         }
 396         return ret;
 397 }
 398
 399 static void __exit
 400 mspec_exit(void)
 401 {
 402         int nid;
 403
 404         misc_deregister(&uncached_miscdev);
 405         misc_deregister(&cached_miscdev);
 406         if (is_sn2) {
 407                 misc_deregister(&fetchop_miscdev);
 408
 409                 for_each_node(nid) {
 410                         if (scratch_page[nid] != 0)
 411                                 uncached_free_page(scratch_page[nid]);
 412                 }
 413         }
 414 }
 415
 416 module_init(mspec_init);
 417 module_exit(mspec_exit);
 418
 419 MODULE_AUTHOR("Silicon Graphics, Inc. <linux-altix@sgi.com>");
 420 MODULE_DESCRIPTION("Driver for SGI SN special memory operations");
 421 MODULE_LICENSE("GPL");