drivers/xen/tmem.c

   1 /*
   2  * Xen implementation for transcendent memory (tmem)
   3  *
   4  * Copyright (C) 2009-2011 Oracle Corp.  All rights reserved.
   5  * Author: Dan Magenheimer
   6  */
   7
   8 #include <linux/module.h>
   9 #include <linux/kernel.h>
  10 #include <linux/types.h>
  11 #include <linux/init.h>
  12 #include <linux/pagemap.h>
  13 #include <linux/cleancache.h>
  14
  15 /* temporary ifdef until include/linux/frontswap.h is upstream */
  16 #ifdef CONFIG_FRONTSWAP
  17 #include <linux/frontswap.h>
  18 #endif
  19
  20 #include <xen/xen.h>
  21 #include <xen/interface/xen.h>
  22 #include <asm/xen/hypercall.h>
  23 #include <asm/xen/page.h>
  24 #include <asm/xen/hypervisor.h>
  25 #include <xen/tmem.h>
  26
  27 #define TMEM_CONTROL               0
  28 #define TMEM_NEW_POOL              1
  29 #define TMEM_DESTROY_POOL          2
  30 #define TMEM_NEW_PAGE              3
  31 #define TMEM_PUT_PAGE              4
  32 #define TMEM_GET_PAGE              5
  33 #define TMEM_FLUSH_PAGE            6
  34 #define TMEM_FLUSH_OBJECT          7
  35 #define TMEM_READ                  8
  36 #define TMEM_WRITE                 9
  37 #define TMEM_XCHG                 10
  38
  39 /* Bits for HYPERVISOR_tmem_op(TMEM_NEW_POOL) */
  40 #define TMEM_POOL_PERSIST          1
  41 #define TMEM_POOL_SHARED           2
  42 #define TMEM_POOL_PAGESIZE_SHIFT   4
  43 #define TMEM_VERSION_SHIFT        24
  44
  45
  46 struct tmem_pool_uuid {
  47         u64 uuid_lo;
  48         u64 uuid_hi;
  49 };
  50
  51 struct tmem_oid {
  52         u64 oid[3];
  53 };
  54
  55 #define TMEM_POOL_PRIVATE_UUID  { 0, 0 }
  56
  57 /* flags for tmem_ops.new_pool */
  58 #define TMEM_POOL_PERSIST          1
  59 #define TMEM_POOL_SHARED           2
  60
  61 /* xen tmem foundation ops/hypercalls */
  62
  63 static inline int xen_tmem_op(u32 tmem_cmd, u32 tmem_pool, struct tmem_oid oid,
  64         u32 index, unsigned long gmfn, u32 tmem_offset, u32 pfn_offset, u32 len)
  65 {
  66         struct tmem_op op;
  67         int rc = 0;
  68
  69         op.cmd = tmem_cmd;
  70         op.pool_id = tmem_pool;
  71         op.u.gen.oid[0] = oid.oid[0];
  72         op.u.gen.oid[1] = oid.oid[1];
  73         op.u.gen.oid[2] = oid.oid[2];
  74         op.u.gen.index = index;
  75         op.u.gen.tmem_offset = tmem_offset;
  76         op.u.gen.pfn_offset = pfn_offset;
  77         op.u.gen.len = len;
  78         set_xen_guest_handle(op.u.gen.gmfn, (void *)gmfn);
  79         rc = HYPERVISOR_tmem_op(&op);
  80         return rc;
  81 }
  82
  83 static int xen_tmem_new_pool(struct tmem_pool_uuid uuid,
  84                                 u32 flags, unsigned long pagesize)
  85 {
  86         struct tmem_op op;
  87         int rc = 0, pageshift;
  88
  89         for (pageshift = 0; pagesize != 1; pageshift++)
  90                 pagesize >>= 1;
  91         flags |= (pageshift - 12) << TMEM_POOL_PAGESIZE_SHIFT;
  92         flags |= TMEM_SPEC_VERSION << TMEM_VERSION_SHIFT;
  93         op.cmd = TMEM_NEW_POOL;
  94         op.u.new.uuid[0] = uuid.uuid_lo;
  95         op.u.new.uuid[1] = uuid.uuid_hi;
  96         op.u.new.flags = flags;
  97         rc = HYPERVISOR_tmem_op(&op);
  98         return rc;
  99 }
 100
 101 /* xen generic tmem ops */
 102
 103 static int xen_tmem_put_page(u32 pool_id, struct tmem_oid oid,
 104                              u32 index, unsigned long pfn)
 105 {
 106         unsigned long gmfn = xen_pv_domain() ? pfn_to_mfn(pfn) : pfn;
 107
 108         return xen_tmem_op(TMEM_PUT_PAGE, pool_id, oid, index,
 109                 gmfn, 0, 0, 0);
 110 }
 111
 112 static int xen_tmem_get_page(u32 pool_id, struct tmem_oid oid,
 113                              u32 index, unsigned long pfn)
 114 {
 115         unsigned long gmfn = xen_pv_domain() ? pfn_to_mfn(pfn) : pfn;
 116
 117         return xen_tmem_op(TMEM_GET_PAGE, pool_id, oid, index,
 118                 gmfn, 0, 0, 0);
 119 }
 120
 121 static int xen_tmem_flush_page(u32 pool_id, struct tmem_oid oid, u32 index)
 122 {
 123         return xen_tmem_op(TMEM_FLUSH_PAGE, pool_id, oid, index,
 124                 0, 0, 0, 0);
 125 }
 126
 127 static int xen_tmem_flush_object(u32 pool_id, struct tmem_oid oid)
 128 {
 129         return xen_tmem_op(TMEM_FLUSH_OBJECT, pool_id, oid, 0, 0, 0, 0, 0);
 130 }
 131
 132 #ifndef CONFIG_XEN_TMEM_MODULE
 133 bool __read_mostly tmem_enabled = false;
 134
 135 static int __init enable_tmem(char *s)
 136 {
 137         tmem_enabled = true;
 138         return 1;
 139 }
 140 __setup("tmem", enable_tmem);
 141 #endif
 142
 143 #ifdef CONFIG_CLEANCACHE
 144 static int xen_tmem_destroy_pool(u32 pool_id)
 145 {
 146         struct tmem_oid oid = { { 0 } };
 147
 148         return xen_tmem_op(TMEM_DESTROY_POOL, pool_id, oid, 0, 0, 0, 0, 0);
 149 }
 150
 151 /* cleancache ops */
 152
 153 static void tmem_cleancache_put_page(int pool, struct cleancache_filekey key,
 154                                      pgoff_t index, struct page *page)
 155 {
 156         u32 ind = (u32) index;
 157         struct tmem_oid oid = *(struct tmem_oid *)&key;
 158         unsigned long pfn = page_to_pfn(page);
 159
 160         if (pool < 0)
 161                 return;
 162         if (ind != index)
 163                 return;
 164         mb(); /* ensure page is quiescent; tmem may address it with an alias */
 165         (void)xen_tmem_put_page((u32)pool, oid, ind, pfn);
 166 }
 167
 168 static int tmem_cleancache_get_page(int pool, struct cleancache_filekey key,
 169                                     pgoff_t index, struct page *page)
 170 {
 171         u32 ind = (u32) index;
 172         struct tmem_oid oid = *(struct tmem_oid *)&key;
 173         unsigned long pfn = page_to_pfn(page);
 174         int ret;
 175
 176         /* translate return values to linux semantics */
 177         if (pool < 0)
 178                 return -1;
 179         if (ind != index)
 180                 return -1;
 181         ret = xen_tmem_get_page((u32)pool, oid, ind, pfn);
 182         if (ret == 1)
 183                 return 0;
 184         else
 185                 return -1;
 186 }
 187
 188 static void tmem_cleancache_flush_page(int pool, struct cleancache_filekey key,
 189                                        pgoff_t index)
 190 {
 191         u32 ind = (u32) index;
 192         struct tmem_oid oid = *(struct tmem_oid *)&key;
 193
 194         if (pool < 0)
 195                 return;
 196         if (ind != index)
 197                 return;
 198         (void)xen_tmem_flush_page((u32)pool, oid, ind);
 199 }
 200
 201 static void tmem_cleancache_flush_inode(int pool, struct cleancache_filekey key)
 202 {
 203         struct tmem_oid oid = *(struct tmem_oid *)&key;
 204
 205         if (pool < 0)
 206                 return;
 207         (void)xen_tmem_flush_object((u32)pool, oid);
 208 }
 209
 210 static void tmem_cleancache_flush_fs(int pool)
 211 {
 212         if (pool < 0)
 213                 return;
 214         (void)xen_tmem_destroy_pool((u32)pool);
 215 }
 216
 217 static int tmem_cleancache_init_fs(size_t pagesize)
 218 {
 219         struct tmem_pool_uuid uuid_private = TMEM_POOL_PRIVATE_UUID;
 220
 221         return xen_tmem_new_pool(uuid_private, 0, pagesize);
 222 }
 223
 224 static int tmem_cleancache_init_shared_fs(char *uuid, size_t pagesize)
 225 {
 226         struct tmem_pool_uuid shared_uuid;
 227
 228         shared_uuid.uuid_lo = *(u64 *)uuid;
 229         shared_uuid.uuid_hi = *(u64 *)(&uuid[8]);
 230         return xen_tmem_new_pool(shared_uuid, TMEM_POOL_SHARED, pagesize);
 231 }
 232
 233 static bool disable_cleancache __read_mostly;
 234 static bool disable_selfballooning __read_mostly;
 235 #ifdef CONFIG_XEN_TMEM_MODULE
 236 module_param(disable_cleancache, bool, S_IRUGO);
 237 module_param(disable_selfballooning, bool, S_IRUGO);
 238 #else
 239 static int __init no_cleancache(char *s)
 240 {
 241         disable_cleancache = true;
 242         return 1;
 243 }
 244 __setup("nocleancache", no_cleancache);
 245 #endif
 246
 247 static struct cleancache_ops tmem_cleancache_ops = {
 248         .put_page = tmem_cleancache_put_page,
 249         .get_page = tmem_cleancache_get_page,
 250         .invalidate_page = tmem_cleancache_flush_page,
 251         .invalidate_inode = tmem_cleancache_flush_inode,
 252         .invalidate_fs = tmem_cleancache_flush_fs,
 253         .init_shared_fs = tmem_cleancache_init_shared_fs,
 254         .init_fs = tmem_cleancache_init_fs
 255 };
 256 #endif
 257
 258 #ifdef CONFIG_FRONTSWAP
 259 /* frontswap tmem operations */
 260
 261 /* a single tmem poolid is used for all frontswap "types" (swapfiles) */
 262 static int tmem_frontswap_poolid;
 263
 264 /*
 265  * Swizzling increases objects per swaptype, increasing tmem concurrency
 266  * for heavy swaploads.  Later, larger nr_cpus -> larger SWIZ_BITS
 267  */
 268 #define SWIZ_BITS               4
 269 #define SWIZ_MASK               ((1 << SWIZ_BITS) - 1)
 270 #define _oswiz(_type, _ind)     ((_type << SWIZ_BITS) | (_ind & SWIZ_MASK))
 271 #define iswiz(_ind)             (_ind >> SWIZ_BITS)
 272
 273 static inline struct tmem_oid oswiz(unsigned type, u32 ind)
 274 {
 275         struct tmem_oid oid = { .oid = { 0 } };
 276         oid.oid[0] = _oswiz(type, ind);
 277         return oid;
 278 }
 279
 280 /* returns 0 if the page was successfully put into frontswap, -1 if not */
 281 static int tmem_frontswap_store(unsigned type, pgoff_t offset,
 282                                    struct page *page)
 283 {
 284         u64 ind64 = (u64)offset;
 285         u32 ind = (u32)offset;
 286         unsigned long pfn = page_to_pfn(page);
 287         int pool = tmem_frontswap_poolid;
 288         int ret;
 289
 290         if (pool < 0)
 291                 return -1;
 292         if (ind64 != ind)
 293                 return -1;
 294         mb(); /* ensure page is quiescent; tmem may address it with an alias */
 295         ret = xen_tmem_put_page(pool, oswiz(type, ind), iswiz(ind), pfn);
 296         /* translate Xen tmem return values to linux semantics */
 297         if (ret == 1)
 298                 return 0;
 299         else
 300                 return -1;
 301 }
 302
 303 /*
 304  * returns 0 if the page was successfully gotten from frontswap, -1 if
 305  * was not present (should never happen!)
 306  */
 307 static int tmem_frontswap_load(unsigned type, pgoff_t offset,
 308                                    struct page *page)
 309 {
 310         u64 ind64 = (u64)offset;
 311         u32 ind = (u32)offset;
 312         unsigned long pfn = page_to_pfn(page);
 313         int pool = tmem_frontswap_poolid;
 314         int ret;
 315
 316         if (pool < 0)
 317                 return -1;
 318         if (ind64 != ind)
 319                 return -1;
 320         ret = xen_tmem_get_page(pool, oswiz(type, ind), iswiz(ind), pfn);
 321         /* translate Xen tmem return values to linux semantics */
 322         if (ret == 1)
 323                 return 0;
 324         else
 325                 return -1;
 326 }
 327
 328 /* flush a single page from frontswap */
 329 static void tmem_frontswap_flush_page(unsigned type, pgoff_t offset)
 330 {
 331         u64 ind64 = (u64)offset;
 332         u32 ind = (u32)offset;
 333         int pool = tmem_frontswap_poolid;
 334
 335         if (pool < 0)
 336                 return;
 337         if (ind64 != ind)
 338                 return;
 339         (void) xen_tmem_flush_page(pool, oswiz(type, ind), iswiz(ind));
 340 }
 341
 342 /* flush all pages from the passed swaptype */
 343 static void tmem_frontswap_flush_area(unsigned type)
 344 {
 345         int pool = tmem_frontswap_poolid;
 346         int ind;
 347
 348         if (pool < 0)
 349                 return;
 350         for (ind = SWIZ_MASK; ind >= 0; ind--)
 351                 (void)xen_tmem_flush_object(pool, oswiz(type, ind));
 352 }
 353
 354 static void tmem_frontswap_init(unsigned ignored)
 355 {
 356         struct tmem_pool_uuid private = TMEM_POOL_PRIVATE_UUID;
 357
 358         /* a single tmem poolid is used for all frontswap "types" (swapfiles) */
 359         if (tmem_frontswap_poolid < 0)
 360                 tmem_frontswap_poolid =
 361                     xen_tmem_new_pool(private, TMEM_POOL_PERSIST, PAGE_SIZE);
 362 }
 363
 364 static bool disable_frontswap __read_mostly;
 365 static bool disable_frontswap_selfshrinking __read_mostly;
 366 #ifdef CONFIG_XEN_TMEM_MODULE
 367 module_param(disable_frontswap, bool, S_IRUGO);
 368 module_param(disable_frontswap_selfshrinking, bool, S_IRUGO);
 369 #else
 370 static int __init no_frontswap(char *s)
 371 {
 372         disable_frontswap = true;
 373         return 1;
 374 }
 375 __setup("nofrontswap", no_frontswap);
 376 #endif
 377
 378 static struct frontswap_ops tmem_frontswap_ops = {
 379         .store = tmem_frontswap_store,
 380         .load = tmem_frontswap_load,
 381         .invalidate_page = tmem_frontswap_flush_page,
 382         .invalidate_area = tmem_frontswap_flush_area,
 383         .init = tmem_frontswap_init
 384 };
 385 #else   /* CONFIG_FRONTSWAP */
 386 #define disable_frontswap_selfshrinking 1
 387 #endif
 388
 389 static int xen_tmem_init(void)
 390 {
 391         if (!xen_domain())
 392                 return 0;
 393 #ifdef CONFIG_FRONTSWAP
 394         if (tmem_enabled && !disable_frontswap) {
 395                 char *s = "";
 396                 struct frontswap_ops *old_ops =
 397                         frontswap_register_ops(&tmem_frontswap_ops);
 398
 399                 tmem_frontswap_poolid = -1;
 400                 if (IS_ERR(old_ops) || old_ops) {
 401                         if (IS_ERR(old_ops))
 402                                 return PTR_ERR(old_ops);
 403                         s = " (WARNING: frontswap_ops overridden)";
 404                 }
 405                 printk(KERN_INFO "frontswap enabled, RAM provided by "
 406                                  "Xen Transcendent Memory%s\n", s);
 407         }
 408 #endif
 409 #ifdef CONFIG_CLEANCACHE
 410         BUG_ON(sizeof(struct cleancache_filekey) != sizeof(struct tmem_oid));
 411         if (tmem_enabled && !disable_cleancache) {
 412                 char *s = "";
 413                 struct cleancache_ops *old_ops =
 414                         cleancache_register_ops(&tmem_cleancache_ops);
 415                 if (old_ops)
 416                         s = " (WARNING: cleancache_ops overridden)";
 417                 printk(KERN_INFO "cleancache enabled, RAM provided by "
 418                                  "Xen Transcendent Memory%s\n", s);
 419         }
 420 #endif
 421 #ifdef CONFIG_XEN_SELFBALLOONING
 422         xen_selfballoon_init(!disable_selfballooning,
 423                                 !disable_frontswap_selfshrinking);
 424 #endif
 425         return 0;
 426 }
 427
 428 module_init(xen_tmem_init)
 429 MODULE_LICENSE("GPL");
 430 MODULE_AUTHOR("Dan Magenheimer <dan.magenheimer@oracle.com>");
 431 MODULE_DESCRIPTION("Shim to Xen transcendent memory");