osfmk/i386/commpage/commpage.c

   1 /*
   2  * Copyright (c) 2003 Apple Computer, Inc. All rights reserved.
   3  *
   4  * @APPLE_LICENSE_HEADER_START@
   5  *
   6  * The contents of this file constitute Original Code as defined in and
   7  * are subject to the Apple Public Source License Version 1.1 (the
   8  * "License").  You may not use this file except in compliance with the
   9  * License.  Please obtain a copy of the License at
  10  * http://www.apple.com/publicsource and read it before using this file.
  11  *
  12  * This Original Code and all software distributed under the License are
  13  * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  14  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  15  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  16  * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT.  Please see the
  17  * License for the specific language governing rights and limitations
  18  * under the License.
  19  *
  20  * @APPLE_LICENSE_HEADER_END@
  21  */
  22
  23 /*
  24  *      Here's what to do if you want to add a new routine to the comm page:
  25  *
  26  *              1. Add a definition for it's address in osfmk/ppc/cpu_capabilities.h,
  27  *                 being careful to reserve room for future expansion.
  28  *
  29  *              2. Write one or more versions of the routine, each with it's own
  30  *                 commpage_descriptor.  The tricky part is getting the "special",
  31  *                 "musthave", and "canthave" fields right, so that exactly one
  32  *                 version of the routine is selected for every machine.
  33  *                 The source files should be in osfmk/ppc/commpage/.
  34  *
  35  *              3. Add a ptr to your new commpage_descriptor(s) in the "routines"
  36  *                 array in commpage_populate().  Of course, you'll also have to
  37  *                 declare them "extern" in commpage_populate().
  38  *
  39  *              4. Write the code in Libc to use the new routine.
  40  */
  41
  42 #include <mach/mach_types.h>
  43 #include <mach/machine.h>
  44 #include <mach/vm_map.h>
  45 #include <i386/machine_routines.h>
  46 #include <machine/cpu_capabilities.h>
  47 #include <machine/commpage.h>
  48 #include <machine/pmap.h>
  49 #include <vm/vm_kern.h>
  50 #include <vm/vm_map.h>
  51 #include <ipc/ipc_port.h>
  52
  53
  54 extern vm_map_t com_region_map32;       // the shared submap, set up in vm init
  55
  56 static uintptr_t next = 0;              // next available byte in comm page
  57 static int      cur_routine = 0;        // comm page address of "current" routine
  58 static int      matched;                // true if we've found a match for "current" routine
  59
  60 int     _cpu_capabilities = 0;          // define the capability vector
  61
  62 char    *commPagePtr = NULL;            // virtual address of comm page in kernel map
  63
  64 /* Allocate the commpage and add to the shared submap created by vm:
  65  *      1. allocate a page in the kernel map (RW)
  66  *      2. wire it down
  67  *      3. make a memory entry out of it
  68  *      4. map that entry into the shared comm region map (R-only)
  69  */
  70
  71 static  void*
  72 commpage_allocate( void )
  73 {
  74     vm_offset_t         kernel_addr;                // address of commpage in kernel map
  75     vm_offset_t         zero = 0;
  76     vm_size_t           size = _COMM_PAGE_AREA_LENGTH;
  77     vm_map_entry_t      entry;
  78     ipc_port_t          handle;
  79
  80     if (com_region_map32 == NULL)
  81         panic("commpage map is null");
  82
  83     if (vm_allocate(kernel_map,&kernel_addr,_COMM_PAGE_AREA_LENGTH,VM_FLAGS_ANYWHERE))
  84         panic("cannot allocate commpage");
  85
  86     if (vm_map_wire(kernel_map,kernel_addr,kernel_addr+_COMM_PAGE_AREA_LENGTH,VM_PROT_DEFAULT,FALSE))
  87         panic("cannot wire commpage");
  88
  89     /*
  90      * Now that the object is created and wired into the kernel map, mark it so that no delay
  91      * copy-on-write will ever be performed on it as a result of mapping it into user-space.
  92      * If such a delayed copy ever occurred, we could remove the kernel's wired mapping - and
  93      * that would be a real disaster.
  94      *
  95      * JMM - What we really need is a way to create it like this in the first place.
  96      */
  97     if (!vm_map_lookup_entry( kernel_map, vm_map_trunc_page(kernel_addr), &entry) || entry->is_sub_map)
  98         panic("cannot find commpage entry");
  99     entry->object.vm_object->copy_strategy = MEMORY_OBJECT_COPY_NONE;
 100
 101     if (mach_make_memory_entry( kernel_map,         // target map
 102                                 &size,              // size
 103                                 kernel_addr,        // offset (address in kernel map)
 104                                 VM_PROT_DEFAULT,    // map it RW
 105                                 &handle,            // this is the object handle we get
 106                                 NULL ))             // parent_entry (what is this?)
 107         panic("cannot make entry for commpage");
 108
 109     if (vm_map_64(  com_region_map32,               // target map (shared submap)
 110                     &zero,                          // address (map into 1st page in submap)
 111                     _COMM_PAGE_AREA_LENGTH,         // size
 112                     0,                              // mask
 113                     VM_FLAGS_FIXED,                 // flags (it must be 1st page in submap)
 114                     handle,                         // port is the memory entry we just made
 115                     0,                              // offset (map 1st page in memory entry)
 116                     FALSE,                          // copy
 117                     VM_PROT_READ,                   // cur_protection (R-only in user map)
 118                     VM_PROT_READ,                   // max_protection
 119                     VM_INHERIT_SHARE ))             // inheritance
 120         panic("cannot map commpage");
 121
 122     ipc_port_release(handle);
 123
 124     return (void*) kernel_addr;                     // return address in kernel map
 125 }
 126
 127 /* Get address (in kernel map) of a commpage field. */
 128
 129 static void*
 130 commpage_addr_of(
 131     int     addr_at_runtime )
 132 {
 133     return  (void*) ((uintptr_t)commPagePtr + addr_at_runtime - _COMM_PAGE_BASE_ADDRESS);
 134 }
 135
 136 /* Determine number of CPUs on this system.  We cannot rely on
 137  * machine_info.max_cpus this early in the boot.
 138  */
 139 static int
 140 commpage_cpus( void )
 141 {
 142         int cpus;
 143
 144         cpus = ml_get_max_cpus();                   // NB: this call can block
 145
 146         if (cpus == 0)
 147                 panic("commpage cpus==0");
 148         if (cpus > 0xFF)
 149                 cpus = 0xFF;
 150
 151         return cpus;
 152 }
 153
 154 /* Initialize kernel version of _cpu_capabilities vector (used by KEXTs.) */
 155
 156 static void
 157 commpage_init_cpu_capabilities( void )
 158 {
 159         int bits;
 160         int cpus;
 161         ml_cpu_info_t cpu_info;
 162
 163         bits = 0;
 164         ml_cpu_get_info(&cpu_info);
 165
 166         switch (cpu_info.vector_unit) {
 167                 case 5:
 168                         bits |= kHasSSE3;
 169                         /* fall thru */
 170                 case 4:
 171                         bits |= kHasSSE2;
 172                         /* fall thru */
 173                 case 3:
 174                         bits |= kHasSSE;
 175                         /* fall thru */
 176                 case 2:
 177                         bits |= kHasMMX;
 178                 default:
 179                         break;
 180         }
 181         switch (cpu_info.cache_line_size) {
 182                 case 128:
 183                         bits |= kCache128;
 184                         break;
 185                 case 64:
 186                         bits |= kCache64;
 187                         break;
 188                 case 32:
 189                         bits |= kCache32;
 190                         break;
 191                 default:
 192                         break;
 193         }
 194         cpus = commpage_cpus();                 // how many CPUs do we have
 195
 196         if (cpus == 1)
 197                 bits |= kUP;
 198
 199         bits |= (cpus << kNumCPUsShift);
 200
 201         bits |= kFastThreadLocalStorage;        // we use %gs for TLS
 202
 203         _cpu_capabilities = bits;               // set kernel version for use by drivers etc
 204 }
 205
 206 /* Copy data into commpage. */
 207
 208 static void
 209 commpage_stuff(
 210     int         address,
 211     void        *source,
 212     int         length  )
 213 {
 214     void        *dest = commpage_addr_of(address);
 215
 216     if ((uintptr_t)dest < next)
 217         panic("commpage overlap at address 0x%x, 0x%x < 0x%x", address, dest, next);
 218
 219     bcopy(source,dest,length);
 220
 221     next = ((uintptr_t)dest + length);
 222 }
 223
 224
 225 static void
 226 commpage_stuff2(
 227         int address,
 228         void *source,
 229         int length )
 230 {
 231         commpage_stuff(address, source, length);
 232 }
 233
 234 /* Copy a routine into comm page if it matches running machine.
 235  */
 236 static void
 237 commpage_stuff_routine(
 238     commpage_descriptor *rd     )
 239 {
 240     int         must,cant;
 241
 242     if (rd->commpage_address != cur_routine) {
 243         if ((cur_routine!=0) && (matched==0))
 244             panic("commpage no match");
 245         cur_routine = rd->commpage_address;
 246         matched = 0;
 247     }
 248
 249     must = _cpu_capabilities & rd->musthave;
 250     cant = _cpu_capabilities & rd->canthave;
 251
 252     if ((must == rd->musthave) && (cant == 0)) {
 253         if (matched)
 254             panic("commpage duplicate matches");
 255         matched = 1;
 256
 257         commpage_stuff(rd->commpage_address,rd->code_address,rd->code_length);
 258         }
 259 }
 260
 261
 262 #define COMMPAGE_DESC(name)     commpage_ ## name
 263 #define EXTERN_COMMPAGE_DESC(name)                              \
 264         extern commpage_descriptor COMMPAGE_DESC(name)
 265
 266 EXTERN_COMMPAGE_DESC(compare_and_swap32_mp);
 267 EXTERN_COMMPAGE_DESC(compare_and_swap32_up);
 268 EXTERN_COMMPAGE_DESC(compare_and_swap64_mp);
 269 EXTERN_COMMPAGE_DESC(compare_and_swap64_up);
 270 EXTERN_COMMPAGE_DESC(atomic_add32_mp);
 271 EXTERN_COMMPAGE_DESC(atomic_add32_up);
 272 EXTERN_COMMPAGE_DESC(mach_absolute_time);
 273 EXTERN_COMMPAGE_DESC(spin_lock_try_mp);
 274 EXTERN_COMMPAGE_DESC(spin_lock_try_up);
 275 EXTERN_COMMPAGE_DESC(spin_lock_mp);
 276 EXTERN_COMMPAGE_DESC(spin_lock_up);
 277 EXTERN_COMMPAGE_DESC(spin_unlock);
 278 EXTERN_COMMPAGE_DESC(pthread_getspecific);
 279 EXTERN_COMMPAGE_DESC(gettimeofday);
 280 EXTERN_COMMPAGE_DESC(sys_flush_dcache);
 281 EXTERN_COMMPAGE_DESC(sys_icache_invalidate);
 282 EXTERN_COMMPAGE_DESC(pthread_self);
 283 EXTERN_COMMPAGE_DESC(relinquish);
 284 EXTERN_COMMPAGE_DESC(bit_test_and_set_mp);
 285 EXTERN_COMMPAGE_DESC(bit_test_and_set_up);
 286 EXTERN_COMMPAGE_DESC(bit_test_and_clear_mp);
 287 EXTERN_COMMPAGE_DESC(bit_test_and_clear_up);
 288 EXTERN_COMMPAGE_DESC(bzero_scalar);
 289 EXTERN_COMMPAGE_DESC(bcopy_scalar);
 290 EXTERN_COMMPAGE_DESC(nanotime);
 291
 292 static  commpage_descriptor *routines[] = {
 293         &COMMPAGE_DESC(compare_and_swap32_mp),
 294         &COMMPAGE_DESC(compare_and_swap32_up),
 295         &COMMPAGE_DESC(compare_and_swap64_mp),
 296         &COMMPAGE_DESC(compare_and_swap64_up),
 297         &COMMPAGE_DESC(atomic_add32_mp),
 298         &COMMPAGE_DESC(atomic_add32_up),
 299         &COMMPAGE_DESC(mach_absolute_time),
 300         &COMMPAGE_DESC(spin_lock_try_mp),
 301         &COMMPAGE_DESC(spin_lock_try_up),
 302         &COMMPAGE_DESC(spin_lock_mp),
 303         &COMMPAGE_DESC(spin_lock_up),
 304         &COMMPAGE_DESC(spin_unlock),
 305         &COMMPAGE_DESC(pthread_getspecific),
 306         &COMMPAGE_DESC(gettimeofday),
 307         &COMMPAGE_DESC(sys_flush_dcache),
 308         &COMMPAGE_DESC(sys_icache_invalidate),
 309         &COMMPAGE_DESC(pthread_self),
 310         &COMMPAGE_DESC(relinquish),
 311         &COMMPAGE_DESC(bit_test_and_set_mp),
 312         &COMMPAGE_DESC(bit_test_and_set_up),
 313         &COMMPAGE_DESC(bit_test_and_clear_mp),
 314         &COMMPAGE_DESC(bit_test_and_clear_up),
 315         &COMMPAGE_DESC(bzero_scalar),
 316         &COMMPAGE_DESC(bcopy_scalar),
 317         &COMMPAGE_DESC(nanotime),
 318         NULL
 319 };
 320
 321
 322 /* Fill in commpage: called once, during kernel initialization, from the
 323  * startup thread before user-mode code is running.
 324  * See the top of this file for a list of what you have to do to add
 325  * a new routine to the commpage.
 326  */
 327
 328 void
 329 commpage_populate( void )
 330 {
 331         short   c2;
 332         static double   two52 = 1048576.0 * 1048576.0 * 4096.0; // 2**52
 333         static double   ten6 = 1000000.0;                       // 10**6
 334         commpage_descriptor **rd;
 335         short   version = _COMM_PAGE_THIS_VERSION;
 336
 337         commPagePtr = (char *)commpage_allocate();
 338
 339         commpage_init_cpu_capabilities();
 340
 341         /* Stuff in the constants.  We move things into the comm page in strictly
 342         * ascending order, so we can check for overlap and panic if so.
 343         */
 344
 345         commpage_stuff2(_COMM_PAGE_VERSION,&version,sizeof(short));
 346         commpage_stuff(_COMM_PAGE_CPU_CAPABILITIES,&_cpu_capabilities,
 347                 sizeof(int));
 348
 349         if (_cpu_capabilities & kCache32)
 350                 c2 = 32;
 351         else if (_cpu_capabilities & kCache64)
 352                 c2 = 64;
 353         else if (_cpu_capabilities & kCache128)
 354                 c2 = 128;
 355         commpage_stuff(_COMM_PAGE_CACHE_LINESIZE,&c2,2);
 356
 357         c2 = 32;
 358
 359         commpage_stuff2(_COMM_PAGE_2_TO_52,&two52,8);
 360
 361         commpage_stuff2(_COMM_PAGE_10_TO_6,&ten6,8);
 362
 363         for( rd = routines; *rd != NULL ; rd++ )
 364                 commpage_stuff_routine(*rd);
 365
 366         if (!matched)
 367                 panic("commpage no match on last routine");
 368
 369         if (next > (uintptr_t)_COMM_PAGE_END)
 370                 panic("commpage overflow: next = 0x%08x, commPagePtr = 0x%08x", next, (uintptr_t)commPagePtr);
 371
 372
 373         pmap_commpage_init((vm_offset_t) commPagePtr, _COMM_PAGE_BASE_ADDRESS,
 374                            _COMM_PAGE_AREA_LENGTH/INTEL_PGBYTES);
 375 }
 376
 377 /*
 378  * This macro prevents compiler instruction scheduling:
 379  */
 380 #define NO_REORDERING   asm volatile("" : : : "memory")
 381
 382 void
 383 commpage_set_nanotime(commpage_nanotime_t *newp)
 384 {
 385         commpage_nanotime_t     *cnp;
 386
 387         /* Nop if commpage not set up yet */
 388         if (commPagePtr == NULL)
 389                 return;
 390
 391         cnp = (commpage_nanotime_t *)commpage_addr_of(_COMM_PAGE_NANOTIME_INFO);
 392
 393         /*
 394          * Update in reverse order:
 395          * check_tsc first - it's read and compared with base_tsc last.
 396          */
 397         cnp->nt_check_tsc = newp->nt_base_tsc;  NO_REORDERING;
 398         cnp->nt_shift     = newp->nt_shift;     NO_REORDERING;
 399         cnp->nt_scale     = newp->nt_scale;     NO_REORDERING;
 400         cnp->nt_base_ns   = newp->nt_base_ns;   NO_REORDERING;
 401         cnp->nt_base_tsc  = newp->nt_base_tsc;
 402 }
 403