]> git.saurik.com Git - apple/xnu.git/blob - osfmk/ppc/pmap.c
875ee6912b1da84204222abbafdc76679b5ad17e
[apple/xnu.git] / osfmk / ppc / pmap.c
1 /*
2 * Copyright (c) 2000-2005 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 * @OSF_COPYRIGHT@
24 */
25 /*
26 * Mach Operating System
27 * Copyright (c) 1990,1991,1992 The University of Utah and
28 * the Center for Software Science (CSS).
29 * Copyright (c) 1991,1987 Carnegie Mellon University.
30 * All rights reserved.
31 *
32 * Permission to use, copy, modify and distribute this software and its
33 * documentation is hereby granted, provided that both the copyright
34 * notice and this permission notice appear in all copies of the
35 * software, derivative works or modified versions, and any portions
36 * thereof, and that both notices appear in supporting documentation,
37 * and that all advertising materials mentioning features or use of
38 * this software display the following acknowledgement: ``This product
39 * includes software developed by the Center for Software Science at
40 * the University of Utah.''
41 *
42 * CARNEGIE MELLON, THE UNIVERSITY OF UTAH AND CSS ALLOW FREE USE OF
43 * THIS SOFTWARE IN ITS "AS IS" CONDITION, AND DISCLAIM ANY LIABILITY
44 * OF ANY KIND FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF
45 * THIS SOFTWARE.
46 *
47 * CSS requests users of this software to return to css-dist@cs.utah.edu any
48 * improvements that they make and grant CSS redistribution rights.
49 *
50 * Carnegie Mellon requests users of this software to return to
51 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
52 * School of Computer Science
53 * Carnegie Mellon University
54 * Pittsburgh PA 15213-3890
55 * any improvements or extensions that they make and grant Carnegie Mellon
56 * the rights to redistribute these changes.
57 *
58 * Utah $Hdr: pmap.c 1.28 92/06/23$
59 * Author: Mike Hibler, Bob Wheeler, University of Utah CSS, 10/90
60 */
61
62 /*
63 * Manages physical address maps for powerpc.
64 *
65 * In addition to hardware address maps, this
66 * module is called upon to provide software-use-only
67 * maps which may or may not be stored in the same
68 * form as hardware maps. These pseudo-maps are
69 * used to store intermediate results from copy
70 * operations to and from address spaces.
71 *
72 * Since the information managed by this module is
73 * also stored by the logical address mapping module,
74 * this module may throw away valid virtual-to-physical
75 * mappings at almost any time. However, invalidations
76 * of virtual-to-physical mappings must be done as
77 * requested.
78 *
79 * In order to cope with hardware architectures which
80 * make virtual-to-physical map invalidates expensive,
81 * this module may delay invalidate or reduced protection
82 * operations until such time as they are actually
83 * necessary. This module is given full information to
84 * when physical maps must be made correct.
85 *
86 */
87
88 #include <zone_debug.h>
89 #include <debug.h>
90 #include <mach_kgdb.h>
91 #include <mach_vm_debug.h>
92 #include <db_machine_commands.h>
93
94 #include <kern/thread.h>
95 #include <kern/simple_lock.h>
96 #include <mach/vm_attributes.h>
97 #include <mach/vm_param.h>
98 #include <vm/vm_kern.h>
99 #include <kern/spl.h>
100
101 #include <kern/misc_protos.h>
102 #include <ppc/misc_protos.h>
103 #include <ppc/proc_reg.h>
104
105 #include <vm/pmap.h>
106 #include <vm/vm_map.h>
107 #include <vm/vm_page.h>
108
109 #include <ppc/pmap.h>
110 #include <ppc/mem.h>
111 #include <ppc/mappings.h>
112
113 #include <ppc/new_screen.h>
114 #include <ppc/Firmware.h>
115 #include <ppc/savearea.h>
116 #include <ppc/cpu_internal.h>
117 #include <ppc/exception.h>
118 #include <ppc/low_trace.h>
119 #include <ppc/lowglobals.h>
120 #include <ddb/db_output.h>
121 #include <machine/cpu_capabilities.h>
122
123 #include <vm/vm_protos.h> /* must be last */
124
125
126 extern unsigned int avail_remaining;
127 unsigned int debugbackpocket; /* (TEST/DEBUG) */
128
129 vm_offset_t first_free_virt;
130 int current_free_region; /* Used in pmap_next_page */
131
132 pmapTransTab *pmapTrans; /* Point to the hash to pmap translations */
133 struct phys_entry *phys_table;
134
135 /* forward */
136 static void pmap_map_physical(void);
137 static void pmap_map_iohole(addr64_t paddr, addr64_t size);
138 void pmap_activate(pmap_t pmap, thread_t th, int which_cpu);
139 void pmap_deactivate(pmap_t pmap, thread_t th, int which_cpu);
140
141 extern void hw_hash_init(void);
142
143 /* NOTE: kernel_pmap_store must be in V=R storage and aligned!!!!!!!!!!!!!! */
144
145 extern struct pmap kernel_pmap_store;
146 pmap_t kernel_pmap; /* Pointer to kernel pmap and anchor for in-use pmaps */
147 addr64_t kernel_pmap_phys; /* Pointer to kernel pmap and anchor for in-use pmaps, physical address */
148 pmap_t cursor_pmap; /* Pointer to last pmap allocated or previous if removed from in-use list */
149 pmap_t sharedPmap; /* Pointer to common pmap for 64-bit address spaces */
150 struct zone *pmap_zone; /* zone of pmap structures */
151 boolean_t pmap_initialized = FALSE;
152
153 int ppc_max_pmaps; /* Maximum number of concurrent address spaces allowed. This is machine dependent */
154 addr64_t vm_max_address; /* Maximum effective address supported */
155 addr64_t vm_max_physical; /* Maximum physical address supported */
156
157 /*
158 * Physical-to-virtual translations are handled by inverted page table
159 * structures, phys_tables. Multiple mappings of a single page are handled
160 * by linking the affected mapping structures. We initialise one region
161 * for phys_tables of the physical memory we know about, but more may be
162 * added as it is discovered (eg. by drivers).
163 */
164
165 /*
166 * free pmap list. caches the first free_pmap_max pmaps that are freed up
167 */
168 int free_pmap_max = 32;
169 int free_pmap_count;
170 pmap_t free_pmap_list;
171 decl_simple_lock_data(,free_pmap_lock)
172
173 /*
174 * Function to get index into phys_table for a given physical address
175 */
176
177 struct phys_entry *pmap_find_physentry(ppnum_t pa)
178 {
179 int i;
180 unsigned int entry;
181
182 for (i = pmap_mem_regions_count - 1; i >= 0; i--) {
183 if (pa < pmap_mem_regions[i].mrStart) continue; /* See if we fit in this region */
184 if (pa > pmap_mem_regions[i].mrEnd) continue; /* Check the end too */
185
186 entry = (unsigned int)pmap_mem_regions[i].mrPhysTab + ((pa - pmap_mem_regions[i].mrStart) * sizeof(phys_entry_t));
187 return (struct phys_entry *)entry;
188 }
189 // kprintf("DEBUG - pmap_find_physentry: page 0x%08X not found\n", pa);
190 return 0;
191 }
192
193 /*
194 * kern_return_t
195 * pmap_add_physical_memory(vm_offset_t spa, vm_offset_t epa,
196 * boolean_t available, unsigned int attr)
197 *
198 * THIS IS NOT SUPPORTED
199 */
200 kern_return_t
201 pmap_add_physical_memory(
202 __unused vm_offset_t spa,
203 __unused vm_offset_t epa,
204 __unused boolean_t available,
205 __unused unsigned int attr)
206 {
207
208 panic("Forget it! You can't map no more memory, you greedy puke!\n");
209 return KERN_SUCCESS;
210 }
211
212 /*
213 * pmap_map(va, spa, epa, prot)
214 * is called during boot to map memory in the kernel's address map.
215 * A virtual address range starting at "va" is mapped to the physical
216 * address range "spa" to "epa" with machine independent protection
217 * "prot".
218 *
219 * "va", "spa", and "epa" are byte addresses and must be on machine
220 * independent page boundaries.
221 *
222 * Pages with a contiguous virtual address range, the same protection, and attributes.
223 * therefore, we map it with a single block.
224 *
225 * Note that this call will only map into 32-bit space
226 *
227 */
228
229 vm_offset_t
230 pmap_map(
231 vm_offset_t va,
232 vm_offset_t spa,
233 vm_offset_t epa,
234 vm_prot_t prot)
235 {
236
237 addr64_t colladr;
238
239 if (spa == epa) return(va);
240
241 assert(epa > spa);
242
243 colladr = mapping_make(kernel_pmap, (addr64_t)va, (ppnum_t)(spa >> 12), (mmFlgBlock | mmFlgPerm), (epa - spa) >> 12, prot & VM_PROT_ALL);
244
245 if(colladr) { /* Was something already mapped in the range? */
246 panic("pmap_map: attempt to map previously mapped range - va = %08X, pa = %08X, epa = %08X, collision = %016llX\n",
247 va, spa, epa, colladr);
248 }
249 return(va);
250 }
251
252 /*
253 * pmap_map_physical()
254 * Maps physical memory into the kernel's address map beginning at lgPMWvaddr, the
255 * physical memory window.
256 *
257 */
258 void
259 pmap_map_physical()
260 {
261 unsigned region;
262 uint64_t msize, size;
263 addr64_t paddr, vaddr, colladdr;
264
265 /* Iterate over physical memory regions, block mapping each into the kernel's address map */
266 for (region = 0; region < (unsigned)pmap_mem_regions_count; region++) {
267 paddr = ((addr64_t)pmap_mem_regions[region].mrStart << 12); /* Get starting physical address */
268 size = (((addr64_t)pmap_mem_regions[region].mrEnd + 1) << 12) - paddr;
269
270 vaddr = paddr + lowGlo.lgPMWvaddr; /* Get starting virtual address */
271
272 while (size > 0) {
273
274 msize = ((size > 0x0000020000000000ULL) ? 0x0000020000000000ULL : size); /* Get size, but no more than 2TBs */
275
276 colladdr = mapping_make(kernel_pmap, vaddr, (paddr >> 12),
277 (mmFlgBlock | mmFlgPerm), (msize >> 12),
278 (VM_PROT_READ | VM_PROT_WRITE));
279 if (colladdr) {
280 panic ("pmap_map_physical: mapping failure - va = %016llX, pa = %08X, size = %08X, collision = %016llX\n",
281 vaddr, (paddr >> 12), (msize >> 12), colladdr);
282 }
283
284 vaddr = vaddr + (uint64_t)msize; /* Point to the next virtual addr */
285 paddr = paddr + (uint64_t)msize; /* Point to the next physical addr */
286 size -= msize;
287 }
288 }
289 }
290
291 /*
292 * pmap_map_iohole(addr64_t paddr, addr64_t size)
293 * Maps an I/O hole into the kernel's address map at its proper offset in
294 * the physical memory window.
295 *
296 */
297 void
298 pmap_map_iohole(addr64_t paddr, addr64_t size)
299 {
300
301 addr64_t vaddr, colladdr, msize;
302 uint32_t psize;
303
304 vaddr = paddr + lowGlo.lgPMWvaddr; /* Get starting virtual address */
305
306 while (size > 0) {
307
308 msize = ((size > 0x0000020000000000ULL) ? 0x0000020000000000ULL : size); /* Get size, but no more than 2TBs */
309
310 colladdr = mapping_make(kernel_pmap, vaddr, (paddr >> 12),
311 (mmFlgBlock | mmFlgPerm | mmFlgGuarded | mmFlgCInhib), (msize >> 12),
312 (VM_PROT_READ | VM_PROT_WRITE));
313 if (colladdr) {
314 panic ("pmap_map_iohole: mapping failed - va = %016llX, pa = %08X, size = %08X, collision = %016llX\n",
315 vaddr, (paddr >> 12), (msize >> 12), colladdr);
316 }
317
318 vaddr = vaddr + (uint64_t)msize; /* Point to the next virtual addr */
319 paddr = paddr + (uint64_t)msize; /* Point to the next physical addr */
320 size -= msize;
321 }
322 }
323
324 /*
325 * Bootstrap the system enough to run with virtual memory.
326 * Map the kernel's code and data, and allocate the system page table.
327 * Called with mapping done by BATs. Page_size must already be set.
328 *
329 * Parameters:
330 * msize: Total memory present
331 * first_avail: First virtual address available
332 * kmapsize: Size of kernel text and data
333 */
334 void
335 pmap_bootstrap(uint64_t msize, vm_offset_t *first_avail, unsigned int kmapsize)
336 {
337 vm_offset_t addr;
338 vm_size_t size;
339 unsigned int i, num, mapsize, vmpagesz, vmmapsz, nbits;
340 signed bank;
341 uint64_t tmemsize;
342 uint_t htslop;
343 vm_offset_t first_used_addr, PCAsize;
344 struct phys_entry *phys_entry;
345
346 *first_avail = round_page(*first_avail); /* Make sure we start out on a page boundary */
347 vm_last_addr = VM_MAX_KERNEL_ADDRESS; /* Set the highest address know to VM */
348
349 /*
350 * Initialize kernel pmap
351 */
352 kernel_pmap = &kernel_pmap_store;
353 kernel_pmap_phys = (addr64_t)&kernel_pmap_store;
354 cursor_pmap = &kernel_pmap_store;
355
356 kernel_pmap->pmap_link.next = (queue_t)kernel_pmap; /* Set up anchor forward */
357 kernel_pmap->pmap_link.prev = (queue_t)kernel_pmap; /* Set up anchor reverse */
358 kernel_pmap->ref_count = 1;
359 kernel_pmap->pmapFlags = pmapKeyDef; /* Set the default keys */
360 kernel_pmap->pmapCCtl = pmapCCtlVal; /* Initialize cache control */
361 kernel_pmap->space = PPC_SID_KERNEL;
362 kernel_pmap->pmapvr = 0; /* Virtual = Real */
363
364 /*
365 * IBM's recommended hash table size is one PTEG for every 2 physical pages.
366 * However, we have found that OSX rarely uses more than 4 PTEs in a PTEG
367 * with this size table. Therefore, by default we allocate a hash table
368 * one half IBM's recommended size, ie one PTEG per 4 pages. The "ht_shift" boot-arg
369 * can be used to override the default hash table size.
370 * We will allocate the hash table in physical RAM, outside of kernel virtual memory,
371 * at the top of the highest bank that will contain it.
372 * Note that "bank" doesn't refer to a physical memory slot here, it is a range of
373 * physically contiguous memory.
374 *
375 * The PCA will go there as well, immediately before the hash table.
376 */
377
378 nbits = cntlzw(((msize << 1) - 1) >> 32); /* Get first bit in upper half */
379 if (nbits == 32) /* If upper half was empty, find bit in bottom half */
380 nbits = nbits + cntlzw((uint_t)((msize << 1) - 1));
381 tmemsize = 0x8000000000000000ULL >> nbits; /* Get memory size rounded up to power of 2 */
382
383 /* Calculate hash table size: First, make sure we don't overflow 32-bit arithmetic. */
384 if (tmemsize > 0x0000002000000000ULL)
385 tmemsize = 0x0000002000000000ULL;
386
387 /* Second, calculate IBM recommended hash table size, ie one PTEG per 2 physical pages */
388 hash_table_size = (uint_t)(tmemsize >> 13) * PerProcTable[0].ppe_vaddr->pf.pfPTEG;
389
390 /* Third, cut this in half to produce the OSX default, ie one PTEG per 4 physical pages */
391 hash_table_size >>= 1;
392
393 /* Fourth, adjust default size per "ht_shift" boot arg */
394 if (hash_table_shift >= 0) /* if positive, make size bigger */
395 hash_table_size <<= hash_table_shift;
396 else /* if "ht_shift" is negative, make smaller */
397 hash_table_size >>= (-hash_table_shift);
398
399 /* Fifth, make sure we are at least minimum size */
400 if (hash_table_size < (256 * 1024))
401 hash_table_size = (256 * 1024);
402
403 while(1) { /* Try to fit hash table in PCA into contiguous memory */
404
405 if(hash_table_size < (256 * 1024)) { /* Have we dropped too short? This should never, ever happen */
406 panic("pmap_bootstrap: Can't find space for hash table\n"); /* This will never print, system isn't up far enough... */
407 }
408
409 PCAsize = (hash_table_size / PerProcTable[0].ppe_vaddr->pf.pfPTEG) * sizeof(PCA_t); /* Get total size of PCA table */
410 PCAsize = round_page(PCAsize); /* Make sure it is at least a page long */
411
412 for(bank = pmap_mem_regions_count - 1; bank >= 0; bank--) { /* Search backwards through banks */
413
414 hash_table_base = ((addr64_t)pmap_mem_regions[bank].mrEnd << 12) - hash_table_size + PAGE_SIZE; /* Get tenative address */
415
416 htslop = hash_table_base & (hash_table_size - 1); /* Get the extra that we will round down when we align */
417 hash_table_base = hash_table_base & -(addr64_t)hash_table_size; /* Round down to correct boundary */
418
419 if((hash_table_base - round_page(PCAsize)) >= ((addr64_t)pmap_mem_regions[bank].mrStart << 12)) break; /* Leave if we fit */
420 }
421
422 if(bank >= 0) break; /* We are done if we found a suitable bank */
423
424 hash_table_size = hash_table_size >> 1; /* Try the next size down */
425 }
426
427 if(htslop) { /* If there was slop (i.e., wasted pages for alignment) add a new region */
428 for(i = pmap_mem_regions_count - 1; i >= (unsigned)bank; i--) { /* Copy from end to our bank, including our bank */
429 pmap_mem_regions[i + 1].mrStart = pmap_mem_regions[i].mrStart; /* Set the start of the bank */
430 pmap_mem_regions[i + 1].mrAStart = pmap_mem_regions[i].mrAStart; /* Set the start of allocatable area */
431 pmap_mem_regions[i + 1].mrEnd = pmap_mem_regions[i].mrEnd; /* Set the end address of bank */
432 pmap_mem_regions[i + 1].mrAEnd = pmap_mem_regions[i].mrAEnd; /* Set the end address of allocatable area */
433 }
434
435 pmap_mem_regions[i + 1].mrStart = (hash_table_base + hash_table_size) >> 12; /* Set the start of the next bank to the start of the slop area */
436 pmap_mem_regions[i + 1].mrAStart = (hash_table_base + hash_table_size) >> 12; /* Set the start of allocatable area to the start of the slop area */
437 pmap_mem_regions[i].mrEnd = (hash_table_base + hash_table_size - 4096) >> 12; /* Set the end of our bank to the end of the hash table */
438
439 }
440
441 pmap_mem_regions[bank].mrAEnd = (hash_table_base - PCAsize - 4096) >> 12; /* Set the maximum allocatable in this bank */
442
443 hw_hash_init(); /* Initiaize the hash table and PCA */
444 hw_setup_trans(); /* Set up hardware registers needed for translation */
445
446 /*
447 * The hash table is now all initialized and so is the PCA. Go on to do the rest of it.
448 * This allocation is from the bottom up.
449 */
450
451 num = atop_64(msize); /* Get number of pages in all of memory */
452
453 /* Figure out how much we need to allocate */
454
455 size = (vm_size_t) (
456 (InitialSaveBloks * PAGE_SIZE) + /* Allow space for the initial context saveareas */
457 (BackPocketSaveBloks * PAGE_SIZE) + /* For backpocket saveareas */
458 trcWork.traceSize + /* Size of trace table */
459 ((((1 << maxAdrSpb) * sizeof(pmapTransTab)) + 4095) & -4096) + /* Size of pmap translate table */
460 (((num * sizeof(struct phys_entry)) + 4095) & -4096) /* For the physical entries */
461 );
462
463 mapsize = size = round_page(size); /* Get size of area to map that we just calculated */
464 mapsize = mapsize + kmapsize; /* Account for the kernel text size */
465
466 vmpagesz = round_page(num * sizeof(struct vm_page)); /* Allow for all vm_pages needed to map physical mem */
467 vmmapsz = round_page((num / 8) * sizeof(struct vm_map_entry)); /* Allow for vm_maps */
468
469 mapsize = mapsize + vmpagesz + vmmapsz; /* Add the VM system estimates into the grand total */
470
471 mapsize = mapsize + (4 * 1024 * 1024); /* Allow for 4 meg of extra mappings */
472 mapsize = ((mapsize / PAGE_SIZE) + MAPPERBLOK - 1) / MAPPERBLOK; /* Get number of blocks of mappings we need */
473 mapsize = mapsize + ((mapsize + MAPPERBLOK - 1) / MAPPERBLOK); /* Account for the mappings themselves */
474
475 size = size + (mapsize * PAGE_SIZE); /* Get the true size we need */
476
477 /* hash table must be aligned to its size */
478
479 addr = *first_avail; /* Set the address to start allocations */
480 first_used_addr = addr; /* Remember where we started */
481
482 bzero((char *)addr, size); /* Clear everything that we are allocating */
483
484 savearea_init(addr); /* Initialize the savearea chains and data */
485
486 addr = (vm_offset_t)((unsigned int)addr + ((InitialSaveBloks + BackPocketSaveBloks) * PAGE_SIZE)); /* Point past saveareas */
487
488 trcWork.traceCurr = (unsigned int)addr; /* Set first trace slot to use */
489 trcWork.traceStart = (unsigned int)addr; /* Set start of trace table */
490 trcWork.traceEnd = (unsigned int)addr + trcWork.traceSize; /* Set end of trace table */
491
492 addr = (vm_offset_t)trcWork.traceEnd; /* Set next allocatable location */
493
494 pmapTrans = (pmapTransTab *)addr; /* Point to the pmap to hash translation table */
495
496 pmapTrans[PPC_SID_KERNEL].pmapPAddr = (addr64_t)((uintptr_t)kernel_pmap); /* Initialize the kernel pmap in the translate table */
497 pmapTrans[PPC_SID_KERNEL].pmapVAddr = CAST_DOWN(unsigned int, kernel_pmap); /* Initialize the kernel pmap in the translate table */
498
499 addr += ((((1 << maxAdrSpb) * sizeof(pmapTransTab)) + 4095) & -4096); /* Point past pmap translate table */
500
501 /* NOTE: the phys_table must be within the first 2GB of physical RAM. This makes sure we only need to do 32-bit arithmetic */
502
503 phys_entry = (struct phys_entry *) addr; /* Get pointer to physical table */
504
505 for (bank = 0; bank < pmap_mem_regions_count; bank++) { /* Set pointer and initialize all banks of ram */
506
507 pmap_mem_regions[bank].mrPhysTab = phys_entry; /* Set pointer to the physical table for this bank */
508
509 phys_entry = phys_entry + (pmap_mem_regions[bank].mrEnd - pmap_mem_regions[bank].mrStart + 1); /* Point to the next */
510 }
511
512 addr += (((num * sizeof(struct phys_entry)) + 4095) & -4096); /* Step on past the physical entries */
513
514 /*
515 * Remaining space is for mapping entries. Tell the initializer routine that
516 * the mapping system can't release this block because it's permanently assigned
517 */
518
519 mapping_init(); /* Initialize the mapping tables */
520
521 for(i = addr; i < first_used_addr + size; i += PAGE_SIZE) { /* Add initial mapping blocks */
522 mapping_free_init(i, 1, 0); /* Pass block address and say that this one is not releasable */
523 }
524 mapCtl.mapcmin = MAPPERBLOK; /* Make sure we only adjust one at a time */
525
526 /* Map V=R the page tables */
527 pmap_map(first_used_addr, first_used_addr,
528 round_page(first_used_addr + size), VM_PROT_READ | VM_PROT_WRITE);
529
530 *first_avail = round_page(first_used_addr + size); /* Set next available page */
531 first_free_virt = *first_avail; /* Ditto */
532
533 /* For 64-bit machines, block map physical memory and the I/O hole into kernel space */
534 if(BootProcInfo.pf.Available & pf64Bit) { /* Are we on a 64-bit machine? */
535 lowGlo.lgPMWvaddr = PHYS_MEM_WINDOW_VADDR; /* Initialize the physical memory window's virtual address */
536
537 pmap_map_physical(); /* Block map physical memory into the window */
538
539 pmap_map_iohole(IO_MEM_WINDOW_VADDR, IO_MEM_WINDOW_SIZE);
540 /* Block map the I/O hole */
541 }
542
543 /* All the rest of memory is free - add it to the free
544 * regions so that it can be allocated by pmap_steal
545 */
546
547 pmap_mem_regions[0].mrAStart = (*first_avail >> 12); /* Set up the free area to start allocations (always in the first bank) */
548
549 current_free_region = 0; /* Set that we will start allocating in bank 0 */
550 avail_remaining = 0; /* Clear free page count */
551 for(bank = 0; bank < pmap_mem_regions_count; bank++) { /* Total up all of the pages in the system that are available */
552 avail_remaining += (pmap_mem_regions[bank].mrAEnd - pmap_mem_regions[bank].mrAStart) + 1; /* Add in allocatable pages in this bank */
553 }
554
555
556 }
557
558 /*
559 * pmap_init(spa, epa)
560 * finishes the initialization of the pmap module.
561 * This procedure is called from vm_mem_init() in vm/vm_init.c
562 * to initialize any remaining data structures that the pmap module
563 * needs to map virtual memory (VM is already ON).
564 *
565 * Note that the pmap needs to be sized and aligned to
566 * a power of two. This is because it is used both in virtual and
567 * real so it can't span a page boundary.
568 */
569
570 void
571 pmap_init(void)
572 {
573
574 pmap_zone = zinit(pmapSize, 400 * pmapSize, 4096, "pmap");
575 #if ZONE_DEBUG
576 zone_debug_disable(pmap_zone); /* Can't debug this one 'cause it messes with size and alignment */
577 #endif /* ZONE_DEBUG */
578
579 pmap_initialized = TRUE;
580
581 /*
582 * Initialize list of freed up pmaps
583 */
584 free_pmap_list = 0; /* Set that there are no free pmaps */
585 free_pmap_count = 0;
586 simple_lock_init(&free_pmap_lock, 0);
587
588 }
589
590 unsigned int pmap_free_pages(void)
591 {
592 return avail_remaining;
593 }
594
595 /*
596 * This function allocates physical pages.
597 */
598
599 /* Non-optimal, but only used for virtual memory startup.
600 * Allocate memory from a table of free physical addresses
601 * If there are no more free entries, too bad.
602 */
603
604 boolean_t pmap_next_page(ppnum_t *addrp)
605 {
606 int i;
607
608 if(current_free_region >= pmap_mem_regions_count) return FALSE; /* Return failure if we have used everything... */
609
610 for(i = current_free_region; i < pmap_mem_regions_count; i++) { /* Find the next bank with free pages */
611 if(pmap_mem_regions[i].mrAStart <= pmap_mem_regions[i].mrAEnd) break; /* Found one */
612 }
613
614 current_free_region = i; /* Set our current bank */
615 if(i >= pmap_mem_regions_count) return FALSE; /* Couldn't find a free page */
616
617 *addrp = pmap_mem_regions[i].mrAStart; /* Allocate the page */
618 pmap_mem_regions[i].mrAStart = pmap_mem_regions[i].mrAStart + 1; /* Set the next one to go */
619 avail_remaining--; /* Drop free count */
620
621 return TRUE;
622 }
623
624 void pmap_virtual_space(
625 vm_offset_t *startp,
626 vm_offset_t *endp)
627 {
628 *startp = round_page(first_free_virt);
629 *endp = vm_last_addr;
630 }
631
632 /*
633 * pmap_create
634 *
635 * Create and return a physical map.
636 *
637 * If the size specified for the map is zero, the map is an actual physical
638 * map, and may be referenced by the hardware.
639 *
640 * A pmap is either in the free list or in the in-use list. The only use
641 * of the in-use list (aside from debugging) is to handle the VSID wrap situation.
642 * Whenever a new pmap is allocated (i.e., not recovered from the free list). The
643 * in-use list is matched until a hole in the VSID sequence is found. (Note
644 * that the in-use pmaps are queued in VSID sequence order.) This is all done
645 * while free_pmap_lock is held.
646 *
647 * If the size specified is non-zero, the map will be used in software
648 * only, and is bounded by that size.
649 */
650 pmap_t
651 pmap_create(vm_map_size_t size)
652 {
653 pmap_t pmap, ckpmap, fore;
654 int s;
655 unsigned int currSID;
656 addr64_t physpmap;
657
658 /*
659 * A software use-only map doesn't even need a pmap structure.
660 */
661 if (size)
662 return(PMAP_NULL);
663
664 /*
665 * If there is a pmap in the pmap free list, reuse it.
666 * Note that we use free_pmap_list for all chaining of pmaps, both to
667 * the free list and the in use chain (anchored from kernel_pmap).
668 */
669 s = splhigh();
670 simple_lock(&free_pmap_lock);
671
672 if(free_pmap_list) { /* Any free? */
673 pmap = free_pmap_list; /* Yes, allocate it */
674 free_pmap_list = (pmap_t)pmap->freepmap; /* Dequeue this one (we chain free ones through freepmap) */
675 free_pmap_count--;
676 }
677 else {
678 simple_unlock(&free_pmap_lock); /* Unlock just in case */
679 splx(s);
680
681 pmap = (pmap_t) zalloc(pmap_zone); /* Get one */
682 if (pmap == PMAP_NULL) return(PMAP_NULL); /* Handle out-of-memory condition */
683
684 bzero((char *)pmap, pmapSize); /* Clean up the pmap */
685
686 s = splhigh();
687 simple_lock(&free_pmap_lock); /* Lock it back up */
688
689 ckpmap = cursor_pmap; /* Get starting point for free ID search */
690 currSID = ckpmap->spaceNum; /* Get the actual space ID number */
691
692 while(1) { /* Keep trying until something happens */
693
694 currSID = (currSID + 1) & (maxAdrSp - 1); /* Get the next in the sequence */
695 if(((currSID * incrVSID) & (maxAdrSp - 1)) == invalSpace) continue; /* Skip the space we have reserved */
696 ckpmap = (pmap_t)ckpmap->pmap_link.next; /* On to the next in-use pmap */
697
698 if(ckpmap->spaceNum != currSID) break; /* If we are out of sequence, this is free */
699
700 if(ckpmap == cursor_pmap) { /* See if we have 2^20 already allocated */
701 panic("pmap_create: Maximum number (%d) active address spaces reached\n", maxAdrSp); /* Die pig dog */
702 }
703 }
704
705 pmap->space = (currSID * incrVSID) & (maxAdrSp - 1); /* Calculate the actual VSID */
706 pmap->spaceNum = currSID; /* Set the space ID number */
707 /*
708 * Now we link into the chain just before the out of sequence guy.
709 */
710
711 fore = (pmap_t)ckpmap->pmap_link.prev; /* Get the current's previous */
712 pmap->pmap_link.next = (queue_t)ckpmap; /* My next points to the current */
713 fore->pmap_link.next = (queue_t)pmap; /* Current's previous's next points to me */
714 pmap->pmap_link.prev = (queue_t)fore; /* My prev points to what the current pointed to */
715 ckpmap->pmap_link.prev = (queue_t)pmap; /* Current's prev points to me */
716
717 physpmap = ((addr64_t)pmap_find_phys(kernel_pmap, (addr64_t)((uintptr_t)pmap)) << 12) | (addr64_t)((unsigned int)pmap & 0xFFF); /* Get the physical address of the pmap */
718
719 pmap->pmapvr = (addr64_t)((uintptr_t)pmap) ^ physpmap; /* Make V to R translation mask */
720
721 pmapTrans[pmap->space].pmapPAddr = physpmap; /* Set translate table physical to point to us */
722 pmapTrans[pmap->space].pmapVAddr = CAST_DOWN(unsigned int, pmap); /* Set translate table virtual to point to us */
723 }
724
725 pmap->pmapVmmExt = 0; /* Clear VMM extension block vaddr */
726 pmap->pmapVmmExtPhys = 0; /* and the paddr, too */
727 pmap->pmapFlags = pmapKeyDef; /* Set default key */
728 pmap->pmapCCtl = pmapCCtlVal; /* Initialize cache control */
729 pmap->ref_count = 1;
730 pmap->stats.resident_count = 0;
731 pmap->stats.wired_count = 0;
732 pmap->pmapSCSubTag = 0x0000000000000000ULL; /* Make sure this is clean an tidy */
733 simple_unlock(&free_pmap_lock);
734
735 splx(s);
736 return(pmap);
737 }
738
739 /*
740 * pmap_destroy
741 *
742 * Gives up a reference to the specified pmap. When the reference count
743 * reaches zero the pmap structure is added to the pmap free list.
744 *
745 * Should only be called if the map contains no valid mappings.
746 */
747 void
748 pmap_destroy(pmap_t pmap)
749 {
750 int ref_count;
751 spl_t s;
752 pmap_t fore, aft;
753
754 if (pmap == PMAP_NULL)
755 return;
756
757 ref_count=hw_atomic_sub(&pmap->ref_count, 1); /* Back off the count */
758 if(ref_count>0) return; /* Still more users, leave now... */
759
760 if(ref_count < 0) /* Did we go too far? */
761 panic("pmap_destroy(): ref_count < 0");
762
763 if (!(pmap->pmapFlags & pmapVMgsaa)) { /* Don't try this for a shadow assist guest */
764 pmap_unmap_sharedpage(pmap); /* Remove any mapping of page -1 */
765 }
766
767 #ifdef notdef
768 if(pmap->stats.resident_count != 0)
769 panic("PMAP_DESTROY: pmap not empty");
770 #else
771 if(pmap->stats.resident_count != 0) {
772 pmap_remove(pmap, 0, 0xFFFFFFFFFFFFF000ULL);
773 }
774 #endif
775
776 /*
777 * Add the pmap to the pmap free list.
778 */
779
780 s = splhigh();
781 /*
782 * Add the pmap to the pmap free list.
783 */
784 simple_lock(&free_pmap_lock);
785
786 if (free_pmap_count <= free_pmap_max) { /* Do we have enough spares? */
787
788 pmap->freepmap = free_pmap_list; /* Queue in front */
789 free_pmap_list = pmap;
790 free_pmap_count++;
791 simple_unlock(&free_pmap_lock);
792
793 } else {
794 if(cursor_pmap == pmap) cursor_pmap = (pmap_t)pmap->pmap_link.prev; /* If we are releasing the cursor, back up */
795 fore = (pmap_t)pmap->pmap_link.prev;
796 aft = (pmap_t)pmap->pmap_link.next;
797 fore->pmap_link.next = pmap->pmap_link.next; /* My previous's next is my next */
798 aft->pmap_link.prev = pmap->pmap_link.prev; /* My next's previous is my previous */
799 simple_unlock(&free_pmap_lock);
800 pmapTrans[pmap->space].pmapPAddr = -1; /* Invalidate the translate table physical */
801 pmapTrans[pmap->space].pmapVAddr = -1; /* Invalidate the translate table virtual */
802 zfree(pmap_zone, pmap);
803 }
804 splx(s);
805 }
806
807 /*
808 * pmap_reference(pmap)
809 * gains a reference to the specified pmap.
810 */
811 void
812 pmap_reference(pmap_t pmap)
813 {
814 if (pmap != PMAP_NULL) hw_atomic_add(&pmap->ref_count, 1); /* Bump the count */
815 }
816
817 /*
818 * pmap_remove_some_phys
819 *
820 * Removes mappings of the associated page from the specified pmap
821 *
822 */
823 void pmap_remove_some_phys(
824 pmap_t pmap,
825 vm_offset_t pa)
826 {
827 register struct phys_entry *pp;
828 register struct mapping *mp;
829 unsigned int pindex;
830
831 if (pmap == PMAP_NULL) { /* This should never be called with a null pmap */
832 panic("pmap_remove_some_phys: null pmap\n");
833 }
834
835 pp = mapping_phys_lookup(pa, &pindex); /* Get physical entry */
836 if (pp == 0) return; /* Leave if not in physical RAM */
837
838 do { /* Keep going until we toss all pages from this pmap */
839 if (pmap->pmapFlags & pmapVMhost) {
840 mp = hw_purge_phys(pp); /* Toss a map */
841 switch ((unsigned int)mp & mapRetCode) {
842 case mapRtOK:
843 mapping_free(mp); /* Return mapping to free inventory */
844 break;
845 case mapRtGuest:
846 break; /* Don't try to return a guest mapping */
847 case mapRtEmpty:
848 break; /* Physent chain empty, we're done */
849 case mapRtNotFnd:
850 break; /* Mapping disappeared on us, retry */
851 default:
852 panic("pmap_remove_some_phys: hw_purge_phys failed - pp = %08X, pmap = %08X, code = %08X\n",
853 pp, pmap, mp); /* Handle failure with our usual lack of tact */
854 }
855 } else {
856 mp = hw_purge_space(pp, pmap); /* Toss a map */
857 switch ((unsigned int)mp & mapRetCode) {
858 case mapRtOK:
859 mapping_free(mp); /* Return mapping to free inventory */
860 break;
861 case mapRtEmpty:
862 break; /* Physent chain empty, we're done */
863 case mapRtNotFnd:
864 break; /* Mapping disappeared on us, retry */
865 default:
866 panic("pmap_remove_some_phys: hw_purge_phys failed - pp = %08X, pmap = %08X, code = %08X\n",
867 pp, pmap, mp); /* Handle failure with our usual lack of tact */
868 }
869 }
870 } while (mapRtEmpty != ((unsigned int)mp & mapRetCode));
871
872 #if DEBUG
873 if ((pmap->pmapFlags & pmapVMhost) && !pmap_verify_free(pa))
874 panic("pmap_remove_some_phys: cruft left behind - pa = %08X, pmap = %08X\n", pa, pmap);
875 #endif
876
877 return; /* Leave... */
878 }
879
880 /*
881 * pmap_remove(pmap, s, e)
882 * unmaps all virtual addresses v in the virtual address
883 * range determined by [s, e) and pmap.
884 * s and e must be on machine independent page boundaries and
885 * s must be less than or equal to e.
886 *
887 * Note that pmap_remove does not remove any mappings in nested pmaps. We just
888 * skip those segments.
889 */
890 void
891 pmap_remove(
892 pmap_t pmap,
893 addr64_t sva,
894 addr64_t eva)
895 {
896 addr64_t va, endva;
897
898 if (pmap == PMAP_NULL) return; /* Leave if software pmap */
899
900
901 /* It is just possible that eva might have wrapped around to zero,
902 * and sometimes we get asked to liberate something of size zero
903 * even though it's dumb (eg. after zero length read_overwrites)
904 */
905 assert(eva >= sva);
906
907 /* If these are not page aligned the loop might not terminate */
908 assert((sva == trunc_page_64(sva)) && (eva == trunc_page_64(eva)));
909
910 va = sva & -4096LL; /* Round start down to a page */
911 endva = eva & -4096LL; /* Round end down to a page */
912
913 while(1) { /* Go until we finish the range */
914 va = mapping_remove(pmap, va); /* Remove the mapping and see what's next */
915 va = va & -4096LL; /* Make sure the "not found" indication is clear */
916 if((va == 0) || (va >= endva)) break; /* End loop if we finish range or run off the end */
917 }
918
919 }
920
921 /*
922 * Routine:
923 * pmap_page_protect
924 *
925 * Function:
926 * Lower the permission for all mappings to a given page.
927 */
928 void
929 pmap_page_protect(
930 ppnum_t pa,
931 vm_prot_t prot)
932 {
933 register struct phys_entry *pp;
934 boolean_t remove;
935 unsigned int pindex;
936 mapping_t *mp;
937
938
939 switch (prot) {
940 case VM_PROT_READ:
941 case VM_PROT_READ|VM_PROT_EXECUTE:
942 remove = FALSE;
943 break;
944 case VM_PROT_ALL:
945 return;
946 default:
947 remove = TRUE;
948 break;
949 }
950
951
952 pp = mapping_phys_lookup(pa, &pindex); /* Get physical entry */
953 if (pp == 0) return; /* Leave if not in physical RAM */
954
955 if (remove) { /* If the protection was set to none, we'll remove all mappings */
956
957 do { /* Keep going until we toss all pages from this physical page */
958 mp = hw_purge_phys(pp); /* Toss a map */
959 switch ((unsigned int)mp & mapRetCode) {
960 case mapRtOK:
961 mapping_free(mp); /* Return mapping to free inventory */
962 break;
963 case mapRtGuest:
964 break; /* Don't try to return a guest mapping */
965 case mapRtNotFnd:
966 break; /* Mapping disappeared on us, retry */
967 case mapRtEmpty:
968 break; /* Physent chain empty, we're done */
969 default: panic("pmap_page_protect: hw_purge_phys failed - pp = %08X, code = %08X\n",
970 pp, mp); /* Handle failure with our usual lack of tact */
971 }
972 } while (mapRtEmpty != ((unsigned int)mp & mapRetCode));
973
974 #if DEBUG
975 if (!pmap_verify_free(pa))
976 panic("pmap_page_protect: cruft left behind - pa = %08X\n", pa);
977 #endif
978
979 return; /* Leave... */
980 }
981
982 /* When we get here, it means that we are to change the protection for a
983 * physical page.
984 */
985
986 mapping_protect_phys(pa, prot & VM_PROT_ALL); /* Change protection of all mappings to page. */
987
988 }
989
990 /*
991 * Routine:
992 * pmap_disconnect
993 *
994 * Function:
995 * Disconnect all mappings for this page and return reference and change status
996 * in generic format.
997 *
998 */
999 unsigned int pmap_disconnect(
1000 ppnum_t pa)
1001 {
1002 register struct phys_entry *pp;
1003 unsigned int pindex;
1004 mapping_t *mp;
1005
1006 pp = mapping_phys_lookup(pa, &pindex); /* Get physical entry */
1007 if (pp == 0) return (0); /* Return null ref and chg if not in physical RAM */
1008 do { /* Iterate until all mappings are dead and gone */
1009 mp = hw_purge_phys(pp); /* Disconnect a mapping */
1010 if (!mp) break; /* All mappings are gone, leave the loop */
1011 switch ((unsigned int)mp & mapRetCode) {
1012 case mapRtOK:
1013 mapping_free(mp); /* Return mapping to free inventory */
1014 break;
1015 case mapRtGuest:
1016 break; /* Don't try to return a guest mapping */
1017 case mapRtNotFnd:
1018 break; /* Mapping disappeared on us, retry */
1019 case mapRtEmpty:
1020 break; /* Physent chain empty, we're done */
1021 default: panic("hw_purge_phys: hw_purge_phys failed - pp = %08X, code = %08X\n",
1022 pp, mp); /* Handle failure with our usual lack of tact */
1023 }
1024 } while (mapRtEmpty != ((unsigned int)mp & mapRetCode));
1025
1026 #if DEBUG
1027 if (!pmap_verify_free(pa))
1028 panic("pmap_disconnect: cruft left behind - pa = %08X\n", pa);
1029 #endif
1030
1031 return (mapping_tst_refmod(pa)); /* Return page ref and chg in generic format */
1032 }
1033
1034 /*
1035 * pmap_protect(pmap, s, e, prot)
1036 * changes the protection on all virtual addresses v in the
1037 * virtual address range determined by [s, e] and pmap to prot.
1038 * s and e must be on machine independent page boundaries and
1039 * s must be less than or equal to e.
1040 *
1041 * Note that any requests to change the protection of a nested pmap are
1042 * ignored. Those changes MUST be done by calling this with the correct pmap.
1043 */
1044 void pmap_protect(
1045 pmap_t pmap,
1046 vm_map_offset_t sva,
1047 vm_map_offset_t eva,
1048 vm_prot_t prot)
1049 {
1050
1051 addr64_t va, endva;
1052
1053 if (pmap == PMAP_NULL) return; /* Do nothing if no pmap */
1054
1055 if (prot == VM_PROT_NONE) { /* Should we kill the address range?? */
1056 pmap_remove(pmap, (addr64_t)sva, (addr64_t)eva); /* Yeah, dump 'em */
1057 return; /* Leave... */
1058 }
1059
1060 va = sva & -4096LL; /* Round start down to a page */
1061 endva = eva & -4096LL; /* Round end down to a page */
1062
1063 while(1) { /* Go until we finish the range */
1064 mapping_protect(pmap, va, prot & VM_PROT_ALL, &va); /* Change the protection and see what's next */
1065 if((va == 0) || (va >= endva)) break; /* End loop if we finish range or run off the end */
1066 }
1067
1068 }
1069
1070
1071
1072 /*
1073 * pmap_enter
1074 *
1075 * Create a translation for the virtual address (virt) to the physical
1076 * address (phys) in the pmap with the protection requested. If the
1077 * translation is wired then we can not allow a full page fault, i.e.,
1078 * the mapping control block is not eligible to be stolen in a low memory
1079 * condition.
1080 *
1081 * NB: This is the only routine which MAY NOT lazy-evaluate
1082 * or lose information. That is, this routine must actually
1083 * insert this page into the given map NOW.
1084 */
1085 void
1086 pmap_enter(pmap_t pmap, vm_map_offset_t va, ppnum_t pa, vm_prot_t prot,
1087 unsigned int flags, __unused boolean_t wired)
1088 {
1089 unsigned int mflags;
1090 addr64_t colva;
1091
1092 if (pmap == PMAP_NULL) return; /* Leave if software pmap */
1093
1094 mflags = 0; /* Make sure this is initialized to nothing special */
1095 if(!(flags & VM_WIMG_USE_DEFAULT)) { /* Are they supplying the attributes? */
1096 mflags = mmFlgUseAttr | (flags & VM_MEM_GUARDED) | ((flags & VM_MEM_NOT_CACHEABLE) >> 1); /* Convert to our mapping_make flags */
1097 }
1098
1099 /*
1100 * It is possible to hang here if another processor is remapping any pages we collide with and are removing
1101 */
1102
1103 while(1) { /* Keep trying the enter until it goes in */
1104
1105 colva = mapping_make(pmap, va, pa, mflags, 1, prot & VM_PROT_ALL); /* Enter the mapping into the pmap */
1106
1107 if(!colva) break; /* If there were no collisions, we are done... */
1108
1109 mapping_remove(pmap, colva); /* Remove the mapping that collided */
1110 }
1111 }
1112
1113 /*
1114 * Enters translations for odd-sized V=F blocks.
1115 *
1116 * The higher level VM map should be locked to insure that we don't have a
1117 * double diddle here.
1118 *
1119 * We panic if we get a block that overlaps with another. We do not merge adjacent
1120 * blocks because removing any address within a block removes the entire block and if
1121 * would really mess things up if we trashed too much.
1122 *
1123 * Once a block is mapped, it is unmutable, that is, protection, catch mode, etc. can
1124 * not be changed. The block must be unmapped and then remapped with the new stuff.
1125 * We also do not keep track of reference or change flags.
1126 *
1127 * Any block that is larger than 256MB must be a multiple of 32MB. We panic if it is not.
1128 *
1129 * Note that pmap_map_block_rc is the same but doesn't panic if collision.
1130 *
1131 */
1132
1133 void pmap_map_block(pmap_t pmap, addr64_t va, ppnum_t pa, uint32_t size, vm_prot_t prot, int attr, unsigned int flags) { /* Map an autogenned block */
1134
1135 unsigned int mflags;
1136 addr64_t colva;
1137
1138
1139 if (pmap == PMAP_NULL) { /* Did they give us a pmap? */
1140 panic("pmap_map_block: null pmap\n"); /* No, like that's dumb... */
1141 }
1142
1143 // kprintf("pmap_map_block: (%08X) va = %016llX, pa = %08X, size = %08X, prot = %08X, attr = %08X, flags = %08X\n", /* (BRINGUP) */
1144 // current_thread(), va, pa, size, prot, attr, flags); /* (BRINGUP) */
1145
1146 mflags = mmFlgBlock | mmFlgUseAttr | (attr & VM_MEM_GUARDED) | ((attr & VM_MEM_NOT_CACHEABLE) >> 1); /* Convert to our mapping_make flags */
1147 if(flags) mflags |= mmFlgPerm; /* Mark permanent if requested */
1148
1149 colva = mapping_make(pmap, va, pa, mflags, size, prot); /* Enter the mapping into the pmap */
1150
1151 if(colva) { /* If there was a collision, panic */
1152 panic("pmap_map_block: mapping error %d, pmap = %08X, va = %016llX\n", (uint32_t)(colva & mapRetCode), pmap, va);
1153 }
1154
1155 return; /* Return */
1156 }
1157
1158 int pmap_map_block_rc(pmap_t pmap, addr64_t va, ppnum_t pa, uint32_t size, vm_prot_t prot, int attr, unsigned int flags) { /* Map an autogenned block */
1159
1160 unsigned int mflags;
1161 addr64_t colva;
1162
1163
1164 if (pmap == PMAP_NULL) { /* Did they give us a pmap? */
1165 panic("pmap_map_block_rc: null pmap\n"); /* No, like that's dumb... */
1166 }
1167
1168 mflags = mmFlgBlock | mmFlgUseAttr | (attr & VM_MEM_GUARDED) | ((attr & VM_MEM_NOT_CACHEABLE) >> 1); /* Convert to our mapping_make flags */
1169 if(flags) mflags |= mmFlgPerm; /* Mark permanent if requested */
1170
1171 colva = mapping_make(pmap, va, pa, mflags, size, prot); /* Enter the mapping into the pmap */
1172
1173 if(colva) return 0; /* If there was a collision, fail */
1174
1175 return 1; /* Return true of we worked */
1176 }
1177
1178 /*
1179 * pmap_extract(pmap, va)
1180 * returns the physical address corrsponding to the
1181 * virtual address specified by pmap and va if the
1182 * virtual address is mapped and 0 if it is not.
1183 * Note: we assume nothing is ever mapped to phys 0.
1184 *
1185 * NOTE: This call always will fail for physical addresses greater than 0xFFFFF000.
1186 */
1187 vm_offset_t pmap_extract(pmap_t pmap, vm_map_offset_t va) {
1188
1189 spl_t spl;
1190 register struct mapping *mp;
1191 register vm_offset_t pa;
1192 addr64_t nextva;
1193 ppnum_t ppoffset;
1194 unsigned int gva;
1195
1196 #ifdef BOGUSCOMPAT
1197 panic("pmap_extract: THIS CALL IS BOGUS. NEVER USE IT EVER. So there...\n"); /* Don't use this */
1198 #else
1199
1200 gva = (unsigned int)va; /* Make sure we don't have a sign */
1201
1202 spl = splhigh(); /* We can't allow any loss of control here */
1203
1204 mp = mapping_find(pmap, (addr64_t)gva, &nextva,1); /* Find the mapping for this address */
1205
1206 if(!mp) { /* Is the page mapped? */
1207 splx(spl); /* Enable interrupts */
1208 return 0; /* Pass back 0 if not found */
1209 }
1210
1211 ppoffset = (ppnum_t)(((gva & -4096LL) - (mp->mpVAddr & -4096LL)) >> 12); /* Get offset from va to base va */
1212
1213
1214 pa = mp->mpPAddr + ppoffset; /* Remember ppage because mapping may vanish after drop call */
1215
1216 mapping_drop_busy(mp); /* We have everything we need from the mapping */
1217 splx(spl); /* Restore 'rupts */
1218
1219 if(pa > maxPPage32) return 0; /* Force large addresses to fail */
1220
1221 pa = (pa << 12) | (va & 0xFFF); /* Convert physical page number to address */
1222
1223 #endif
1224 return pa; /* Return physical address or 0 */
1225 }
1226
1227 /*
1228 * ppnum_t pmap_find_phys(pmap, addr64_t va)
1229 * returns the physical page corrsponding to the
1230 * virtual address specified by pmap and va if the
1231 * virtual address is mapped and 0 if it is not.
1232 * Note: we assume nothing is ever mapped to phys 0.
1233 *
1234 */
1235 ppnum_t pmap_find_phys(pmap_t pmap, addr64_t va) {
1236
1237 spl_t spl;
1238 register struct mapping *mp;
1239 ppnum_t pa, ppoffset;
1240 addr64_t nextva;
1241
1242 spl = splhigh(); /* We can't allow any loss of control here */
1243
1244 mp = mapping_find(pmap, va, &nextva, 1); /* Find the mapping for this address */
1245
1246 if(!mp) { /* Is the page mapped? */
1247 splx(spl); /* Enable interrupts */
1248 return 0; /* Pass back 0 if not found */
1249 }
1250
1251
1252 ppoffset = (ppnum_t)(((va & -4096LL) - (mp->mpVAddr & -4096LL)) >> 12); /* Get offset from va to base va */
1253
1254 pa = mp->mpPAddr + ppoffset; /* Get the actual physical address */
1255
1256 mapping_drop_busy(mp); /* We have everything we need from the mapping */
1257
1258 splx(spl); /* Restore 'rupts */
1259 return pa; /* Return physical address or 0 */
1260 }
1261
1262
1263 /*
1264 * pmap_attributes:
1265 *
1266 * Set/Get special memory attributes; not implemented.
1267 *
1268 * Note: 'VAL_GET_INFO' is used to return info about a page.
1269 * If less than 1 page is specified, return the physical page
1270 * mapping and a count of the number of mappings to that page.
1271 * If more than one page is specified, return the number
1272 * of resident pages and the number of shared (more than
1273 * one mapping) pages in the range;
1274 *
1275 *
1276 */
1277 kern_return_t
1278 pmap_attribute(
1279 __unused pmap_t pmap,
1280 __unused vm_map_offset_t address,
1281 __unused vm_map_size_t size,
1282 __unused vm_machine_attribute_t attribute,
1283 __unused vm_machine_attribute_val_t* value)
1284 {
1285
1286 return KERN_INVALID_ARGUMENT;
1287
1288 }
1289
1290 /*
1291 * pmap_attribute_cache_sync(vm_offset_t pa)
1292 *
1293 * Invalidates all of the instruction cache on a physical page and
1294 * pushes any dirty data from the data cache for the same physical page
1295 */
1296
1297 kern_return_t pmap_attribute_cache_sync(ppnum_t pp, vm_size_t size,
1298 __unused vm_machine_attribute_t attribute,
1299 __unused vm_machine_attribute_val_t* value) {
1300
1301 spl_t s;
1302 unsigned int i, npages;
1303
1304 npages = round_page(size) >> 12; /* Get the number of pages to do */
1305
1306 for(i = 0; i < npages; i++) { /* Do all requested pages */
1307 s = splhigh(); /* No interruptions here */
1308 sync_ppage(pp + i); /* Go flush data cache and invalidate icache */
1309 splx(s); /* Allow interruptions */
1310 }
1311
1312 return KERN_SUCCESS;
1313 }
1314
1315 /*
1316 * pmap_sync_page_data_phys(ppnum_t pa)
1317 *
1318 * Invalidates all of the instruction cache on a physical page and
1319 * pushes any dirty data from the data cache for the same physical page
1320 */
1321
1322 void pmap_sync_page_data_phys(ppnum_t pa) {
1323
1324 spl_t s;
1325
1326 s = splhigh(); /* No interruptions here */
1327 sync_ppage(pa); /* Sync up dem caches */
1328 splx(s); /* Allow interruptions */
1329 return;
1330 }
1331
1332 void
1333 pmap_sync_page_attributes_phys(ppnum_t pa)
1334 {
1335 pmap_sync_page_data_phys(pa);
1336 }
1337
1338 /*
1339 * pmap_collect
1340 *
1341 * Garbage collects the physical map system for pages that are no longer used.
1342 * It isn't implemented or needed or wanted.
1343 */
1344 void
1345 pmap_collect(__unused pmap_t pmap)
1346 {
1347 return;
1348 }
1349
1350 /*
1351 * Routine: pmap_activate
1352 * Function:
1353 * Binds the given physical map to the given
1354 * processor, and returns a hardware map description.
1355 * It isn't implemented or needed or wanted.
1356 */
1357 void
1358 pmap_activate(
1359 __unused pmap_t pmap,
1360 __unused thread_t th,
1361 __unused int which_cpu)
1362 {
1363 return;
1364 }
1365 /*
1366 * pmap_deactivate:
1367 * It isn't implemented or needed or wanted.
1368 */
1369 void
1370 pmap_deactivate(
1371 __unused pmap_t pmap,
1372 __unused thread_t th,
1373 __unused int which_cpu)
1374 {
1375 return;
1376 }
1377
1378
1379 /*
1380 * pmap_pageable(pmap, s, e, pageable)
1381 * Make the specified pages (by pmap, offset)
1382 * pageable (or not) as requested.
1383 *
1384 * A page which is not pageable may not take
1385 * a fault; therefore, its page table entry
1386 * must remain valid for the duration.
1387 *
1388 * This routine is merely advisory; pmap_enter()
1389 * will specify that these pages are to be wired
1390 * down (or not) as appropriate.
1391 *
1392 * (called from vm/vm_fault.c).
1393 */
1394 void
1395 pmap_pageable(
1396 __unused pmap_t pmap,
1397 __unused vm_map_offset_t start,
1398 __unused vm_map_offset_t end,
1399 __unused boolean_t pageable)
1400 {
1401
1402 return; /* This is not used... */
1403
1404 }
1405 /*
1406 * Routine: pmap_change_wiring
1407 * NOT USED ANYMORE.
1408 */
1409 void
1410 pmap_change_wiring(
1411 __unused pmap_t pmap,
1412 __unused vm_map_offset_t va,
1413 __unused boolean_t wired)
1414 {
1415 return; /* This is not used... */
1416 }
1417
1418 /*
1419 * pmap_modify_pages(pmap, s, e)
1420 * sets the modified bit on all virtual addresses v in the
1421 * virtual address range determined by [s, e] and pmap,
1422 * s and e must be on machine independent page boundaries and
1423 * s must be less than or equal to e.
1424 *
1425 * Note that this function will not descend nested pmaps.
1426 */
1427 void
1428 pmap_modify_pages(
1429 pmap_t pmap,
1430 vm_map_offset_t sva,
1431 vm_map_offset_t eva)
1432 {
1433 spl_t spl;
1434 mapping_t *mp;
1435 ppnum_t pa;
1436 addr64_t va, endva;
1437 unsigned int savetype;
1438
1439 if (pmap == PMAP_NULL) return; /* If no pmap, can't do it... */
1440
1441 va = sva & -4096; /* Round to page */
1442 endva = eva & -4096; /* Round to page */
1443
1444 while (va < endva) { /* Walk through all pages */
1445
1446 spl = splhigh(); /* We can't allow any loss of control here */
1447
1448 mp = mapping_find(pmap, (addr64_t)va, &va, 0); /* Find the mapping for this address */
1449
1450 if(!mp) { /* Is the page mapped? */
1451 splx(spl); /* Page not mapped, restore interruptions */
1452 if((va == 0) || (va >= endva)) break; /* We are done if there are no more or we hit the end... */
1453 continue; /* We are not done and there is more to check... */
1454 }
1455
1456 savetype = mp->mpFlags & mpType; /* Remember the type */
1457 pa = mp->mpPAddr; /* Remember ppage because mapping may vanish after drop call */
1458
1459 mapping_drop_busy(mp); /* We have everything we need from the mapping */
1460
1461 splx(spl); /* Restore 'rupts */
1462
1463 if(savetype != mpNormal) continue; /* Can't mess around with these guys... */
1464
1465 mapping_set_mod(pa); /* Set the modfied bit for this page */
1466
1467 if(va == 0) break; /* We hit the end of the pmap, might as well leave now... */
1468 }
1469 return; /* Leave... */
1470 }
1471
1472 /*
1473 * pmap_clear_modify(phys)
1474 * clears the hardware modified ("dirty") bit for one
1475 * machine independant page starting at the given
1476 * physical address. phys must be aligned on a machine
1477 * independant page boundary.
1478 */
1479 void
1480 pmap_clear_modify(ppnum_t pa)
1481 {
1482
1483 mapping_clr_mod(pa); /* Clear all change bits for physical page */
1484
1485 }
1486
1487 /*
1488 * pmap_is_modified(phys)
1489 * returns TRUE if the given physical page has been modified
1490 * since the last call to pmap_clear_modify().
1491 */
1492 boolean_t
1493 pmap_is_modified(register ppnum_t pa)
1494 {
1495 return mapping_tst_mod(pa); /* Check for modified */
1496
1497 }
1498
1499 /*
1500 * pmap_clear_reference(phys)
1501 * clears the hardware referenced bit in the given machine
1502 * independant physical page.
1503 *
1504 */
1505 void
1506 pmap_clear_reference(ppnum_t pa)
1507 {
1508 mapping_clr_ref(pa); /* Check for modified */
1509 }
1510
1511 /*
1512 * pmap_is_referenced(phys)
1513 * returns TRUE if the given physical page has been referenced
1514 * since the last call to pmap_clear_reference().
1515 */
1516 boolean_t
1517 pmap_is_referenced(ppnum_t pa)
1518 {
1519 return mapping_tst_ref(pa); /* Check for referenced */
1520 }
1521
1522 /*
1523 * pmap_get_refmod(phys)
1524 * returns the referenced and modified bits of the specified
1525 * physical page.
1526 */
1527 unsigned int
1528 pmap_get_refmod(ppnum_t pa)
1529 {
1530 return (mapping_tst_refmod(pa));
1531 }
1532
1533 /*
1534 * pmap_clear_refmod(phys, mask)
1535 * clears the referenced and modified bits as specified by the mask
1536 * of the specified physical page.
1537 */
1538 void
1539 pmap_clear_refmod(ppnum_t pa, unsigned int mask)
1540 {
1541 mapping_clr_refmod(pa, mask);
1542 }
1543
1544 /*
1545 * pmap_eligible_for_execute(ppnum_t pa)
1546 * return true if physical address is eligible to contain executable code;
1547 * otherwise, return false
1548 */
1549 boolean_t
1550 pmap_eligible_for_execute(ppnum_t pa)
1551 {
1552 phys_entry_t *physent;
1553 unsigned int pindex;
1554
1555 physent = mapping_phys_lookup(pa, &pindex); /* Get physical entry */
1556
1557 if((!physent) || (physent->ppLink & ppG))
1558 return 0; /* If there is no physical entry or marked guarded,
1559 the entry is not eligible for execute */
1560
1561 return 1; /* Otherwise, entry is eligible for execute */
1562 }
1563
1564 #if MACH_VM_DEBUG
1565 int
1566 pmap_list_resident_pages(
1567 __unused pmap_t pmap,
1568 __unused vm_offset_t *listp,
1569 __unused int space)
1570 {
1571 return 0;
1572 }
1573 #endif /* MACH_VM_DEBUG */
1574
1575 /*
1576 * Locking:
1577 * spl: VM
1578 */
1579 void
1580 pmap_copy_part_page(
1581 vm_offset_t src,
1582 vm_offset_t src_offset,
1583 vm_offset_t dst,
1584 vm_offset_t dst_offset,
1585 vm_size_t len)
1586 {
1587 addr64_t fsrc, fdst;
1588
1589 assert(((dst <<12) & PAGE_MASK+dst_offset+len) <= PAGE_SIZE);
1590 assert(((src <<12) & PAGE_MASK+src_offset+len) <= PAGE_SIZE);
1591
1592 fsrc = ((addr64_t)src << 12) + src_offset;
1593 fdst = ((addr64_t)dst << 12) + dst_offset;
1594
1595 phys_copy(fsrc, fdst, len); /* Copy the stuff physically */
1596 }
1597
1598 void
1599 pmap_zero_part_page(
1600 __unused vm_offset_t p,
1601 __unused vm_offset_t offset,
1602 __unused vm_size_t len)
1603 {
1604 panic("pmap_zero_part_page");
1605 }
1606
1607 boolean_t pmap_verify_free(ppnum_t pa) {
1608
1609 struct phys_entry *pp;
1610 unsigned int pindex;
1611
1612 pp = mapping_phys_lookup(pa, &pindex); /* Get physical entry */
1613 if (pp == 0) return FALSE; /* If there isn't one, show no mapping... */
1614
1615 if(pp->ppLink & ~(ppLock | ppFlags)) return FALSE; /* We have at least one mapping */
1616 return TRUE; /* No mappings */
1617 }
1618
1619
1620 /* Determine if we need to switch space and set up for it if so */
1621
1622 void pmap_switch(pmap_t map)
1623 {
1624 hw_blow_seg(lowGlo.lgUMWvaddr); /* Blow off the first segment */
1625 hw_blow_seg(lowGlo.lgUMWvaddr + 0x10000000ULL); /* Blow off the second segment */
1626
1627 /* when changing to kernel space, don't bother
1628 * doing anything, the kernel is mapped from here already.
1629 */
1630 if (map->space == PPC_SID_KERNEL) { /* Are we switching into kernel space? */
1631 return; /* If so, we don't do anything... */
1632 }
1633
1634 hw_set_user_space(map); /* Indicate if we need to load the SRs or not */
1635 return; /* Bye, bye, butterfly... */
1636 }
1637
1638 /*
1639 * kern_return_t pmap_nest(grand, subord, vstart, size)
1640 *
1641 * grand = the pmap that we will nest subord into
1642 * subord = the pmap that goes into the grand
1643 * vstart = start of range in pmap to be inserted
1644 * nstart = start of range in pmap nested pmap
1645 * size = Size of nest area (up to 2TB)
1646 *
1647 * Inserts a pmap into another. This is used to implement shared segments.
1648 * On the current PPC processors, this is limited to segment (256MB) aligned
1649 * segment sized ranges.
1650 *
1651 * We actually kinda allow recursive nests. The gating factor is that we do not allow
1652 * nesting on top of something that is already mapped, i.e., the range must be empty.
1653 *
1654 * Note that we depend upon higher level VM locks to insure that things don't change while
1655 * we are doing this. For example, VM should not be doing any pmap enters while it is nesting
1656 * or do 2 nests at once.
1657 */
1658
1659 kern_return_t pmap_nest(pmap_t grand, pmap_t subord, addr64_t vstart, addr64_t nstart, uint64_t size) {
1660
1661 addr64_t vend, colladdr;
1662 unsigned int msize;
1663 int nlists;
1664 mapping_t *mp;
1665
1666 if(size & 0x0FFFFFFFULL) return KERN_INVALID_VALUE; /* We can only do this for multiples of 256MB */
1667 if((size >> 25) > 65536) return KERN_INVALID_VALUE; /* Max size we can nest is 2TB */
1668 if(vstart & 0x0FFFFFFFULL) return KERN_INVALID_VALUE; /* We can only do this aligned to 256MB */
1669 if(nstart & 0x0FFFFFFFULL) return KERN_INVALID_VALUE; /* We can only do this aligned to 256MB */
1670
1671 if(size == 0) { /* Is the size valid? */
1672 panic("pmap_nest: size is invalid - %016llX\n", size);
1673 }
1674
1675 msize = (size >> 25) - 1; /* Change size to blocks of 32MB */
1676
1677 nlists = mapSetLists(grand); /* Set number of lists this will be on */
1678
1679 mp = mapping_alloc(nlists); /* Get a spare mapping block */
1680
1681 mp->mpFlags = 0x01000000 | mpNest | mpPerm | mpBSu | nlists; /* Make this a permanent nested pmap with a 32MB basic size unit */
1682 /* Set the flags. Make sure busy count is 1 */
1683 mp->mpSpace = subord->space; /* Set the address space/pmap lookup ID */
1684 mp->u.mpBSize = msize; /* Set the size */
1685 mp->mpPte = 0; /* Set the PTE invalid */
1686 mp->mpPAddr = 0; /* Set the physical page number */
1687 mp->mpVAddr = vstart; /* Set the address */
1688 mp->mpNestReloc = nstart - vstart; /* Set grand to nested vaddr relocation value */
1689
1690 colladdr = hw_add_map(grand, mp); /* Go add the mapping to the pmap */
1691
1692 if(colladdr) { /* Did it collide? */
1693 vend = vstart + size - 4096; /* Point to the last page we would cover in nest */
1694 panic("pmap_nest: attempt to nest into a non-empty range - pmap = %08X, start = %016llX, end = %016llX\n",
1695 grand, vstart, vend);
1696 }
1697
1698 return KERN_SUCCESS;
1699 }
1700
1701 /*
1702 * kern_return_t pmap_unnest(grand, vaddr)
1703 *
1704 * grand = the pmap that we will nest subord into
1705 * vaddr = start of range in pmap to be unnested
1706 *
1707 * Removes a pmap from another. This is used to implement shared segments.
1708 * On the current PPC processors, this is limited to segment (256MB) aligned
1709 * segment sized ranges.
1710 */
1711
1712 kern_return_t pmap_unnest(pmap_t grand, addr64_t vaddr) {
1713
1714 unsigned int tstamp, i, mycpu;
1715 addr64_t nextva;
1716 spl_t s;
1717 mapping_t *mp;
1718
1719 s = splhigh(); /* Make sure interruptions are disabled */
1720
1721 mp = mapping_find(grand, vaddr, &nextva, 0); /* Find the nested map */
1722
1723 if(((unsigned int)mp & mapRetCode) != mapRtOK) { /* See if it was even nested */
1724 panic("pmap_unnest: Attempt to unnest an unnested segment - va = %016llX\n", vaddr);
1725 }
1726
1727 if((mp->mpFlags & mpType) != mpNest) { /* Did we find something other than a nest? */
1728 panic("pmap_unnest: Attempt to unnest something that is not a nest - va = %016llX\n", vaddr);
1729 }
1730
1731 if(mp->mpVAddr != vaddr) { /* Make sure the address is the same */
1732 panic("pmap_unnest: Attempt to unnest something that is not at start of nest - va = %016llX\n", vaddr);
1733 }
1734
1735 (void)hw_atomic_and(&mp->mpFlags, ~mpPerm); /* Show that this mapping is now removable */
1736
1737 mapping_drop_busy(mp); /* Go ahead and release the mapping now */
1738
1739 splx(s); /* Restore 'rupts */
1740
1741 (void)mapping_remove(grand, vaddr); /* Toss the nested pmap mapping */
1742
1743 invalidateSegs(grand); /* Invalidate the pmap segment cache */
1744
1745 /*
1746 * Note that the following will force the segment registers to be reloaded
1747 * on all processors (if they are using the pmap we just changed) before returning.
1748 *
1749 * This is needed. The reason is that until the segment register is
1750 * reloaded, another thread in the same task on a different processor will
1751 * be able to access memory that it isn't allowed to anymore. That can happen
1752 * because access to the subordinate pmap is being removed, but the pmap is still
1753 * valid.
1754 *
1755 * Note that we only kick the other processor if we see that it was using the pmap while we
1756 * were changing it.
1757 */
1758
1759
1760 for(i=0; i < real_ncpus; i++) { /* Cycle through processors */
1761 disable_preemption();
1762 mycpu = cpu_number(); /* Who am I? Am I just a dream? */
1763 if((unsigned int)grand == PerProcTable[i].ppe_vaddr->ppUserPmapVirt) { /* Is this guy using the changed pmap? */
1764
1765 PerProcTable[i].ppe_vaddr->ppInvSeg = 1; /* Show that we need to invalidate the segments */
1766
1767 if(i != mycpu) {
1768
1769 tstamp = PerProcTable[i].ppe_vaddr->ruptStamp[1]; /* Save the processor's last interrupt time stamp */
1770 if(cpu_signal(i, SIGPcpureq, CPRQsegload, 0) == KERN_SUCCESS) { /* Make sure we see the pmap change */
1771 if(!hw_cpu_wcng(&PerProcTable[i].ppe_vaddr->ruptStamp[1], tstamp, LockTimeOut)) { /* Wait for the other processors to enter debug */
1772 panic("pmap_unnest: Other processor (%d) did not see interruption request\n", i);
1773 }
1774 }
1775 }
1776 }
1777 enable_preemption();
1778 }
1779
1780 return KERN_SUCCESS; /* Bye, bye, butterfly... */
1781 }
1782
1783
1784 /*
1785 * void MapUserMemoryWindowInit(void)
1786 *
1787 * Initialize anything we need to in order to map user address space slices into
1788 * the kernel. Primarily used for copy in/out.
1789 *
1790 * Currently we only support one 512MB slot for this purpose. There are two special
1791 * mappings defined for the purpose: the special pmap nest, and linkage mapping.
1792 *
1793 * The special pmap nest (which is allocated in this function) is used as a place holder
1794 * in the kernel's pmap search list. It is 512MB long and covers the address range
1795 * starting at lgUMWvaddr. It points to no actual memory and when the fault handler
1796 * hits in it, it knows to look in the per_proc and start using the linkage
1797 * mapping contained therin.
1798 *
1799 * The linkage mapping is used to glue the user address space slice into the
1800 * kernel. It contains the relocation information used to transform the faulting
1801 * kernel address into the user address space. It also provides the link to the
1802 * user's pmap. This is pointed to by the per_proc and is switched in and out
1803 * whenever there is a context switch.
1804 *
1805 */
1806
1807 void MapUserMemoryWindowInit(void) {
1808
1809 addr64_t colladdr;
1810 int nlists;
1811 mapping_t *mp;
1812
1813 nlists = mapSetLists(kernel_pmap); /* Set number of lists this will be on */
1814
1815 mp = mapping_alloc(nlists); /* Get a spare mapping block */
1816
1817 mp->mpFlags = 0x01000000 | mpLinkage | mpPerm | mpBSu | nlists; /* Make this a permanent nested pmap with a 32MB basic size unit */
1818 /* Set the flags. Make sure busy count is 1 */
1819 mp->mpSpace = kernel_pmap->space; /* Set the address space/pmap lookup ID */
1820 mp->u.mpBSize = 15; /* Set the size to 2 segments in 32MB chunks - 1 */
1821 mp->mpPte = 0; /* Means nothing */
1822 mp->mpPAddr = 0; /* Means nothing */
1823 mp->mpVAddr = lowGlo.lgUMWvaddr; /* Set the address range we cover */
1824 mp->mpNestReloc = 0; /* Means nothing */
1825
1826 colladdr = hw_add_map(kernel_pmap, mp); /* Go add the mapping to the pmap */
1827
1828 if(colladdr) { /* Did it collide? */
1829 panic("MapUserMemoryWindowInit: MapUserMemoryWindow range already mapped\n");
1830 }
1831
1832 return;
1833 }
1834
1835 /*
1836 * addr64_t MapUserMemoryWindow(vm_map_t map, vm_offset_t va, size)
1837 *
1838 * map = the vm_map that we are mapping into the kernel
1839 * va = start of the address range we are mapping
1840 * Note that we do not test validty, we chose to trust our fellows...
1841 *
1842 * Maps a 512M slice of a user address space into a predefined kernel range
1843 * on a per-thread basis. We map only the first 256M segment, allowing the
1844 * second 256M segment to fault in as needed. This allows our clients to access
1845 * an arbitrarily aligned operand up to 256M in size.
1846 *
1847 * In the future, the restriction of a predefined range may be loosened.
1848 *
1849 * Builds the proper linkage map to map the user range
1850 * We will round this down to the previous segment boundary and calculate
1851 * the relocation to the kernel slot
1852 *
1853 * We always make a segment table entry here if we need to. This is mainly because of
1854 * copyin/out and if we don't, there will be multiple segment faults for
1855 * each system call. I have seen upwards of 30000 per second.
1856 *
1857 * We do check, however, to see if the slice is already mapped and if so,
1858 * we just exit. This is done for performance reasons. It was found that
1859 * there was a considerable boost in copyin/out performance if we did not
1860 * invalidate the segment at ReleaseUserAddressSpace time, so we dumped the
1861 * restriction that you had to bracket MapUserMemoryWindow. Further, there
1862 * is a yet further boost if you didn't need to map it each time. The theory
1863 * behind this is that many times copies are to or from the same segment and
1864 * done multiple times within the same system call. To take advantage of that,
1865 * we check umwSpace and umwRelo to see if we've already got it.
1866 *
1867 * We also need to half-invalidate the slice when we context switch or go
1868 * back to user state. A half-invalidate does not clear the actual mapping,
1869 * but it does force the MapUserMemoryWindow function to reload the segment
1870 * register/SLBE. If this is not done, we can end up some pretty severe
1871 * performance penalties. If we map a slice, and the cached space/relocation is
1872 * the same, we won't reload the segment registers. Howver, since we ran someone else,
1873 * our SR is cleared and we will take a fault. This is reasonable if we block
1874 * while copying (e.g., we took a page fault), but it is not reasonable when we
1875 * just start. For this reason, we half-invalidate to make sure that the SR is
1876 * explicitly reloaded.
1877 *
1878 * Note that we do not go to the trouble of making a pmap segment cache
1879 * entry for these guys because they are very short term -- 99.99% of the time
1880 * they will be unmapped before the next context switch.
1881 *
1882 */
1883
1884 addr64_t MapUserMemoryWindow(
1885 vm_map_t map,
1886 addr64_t va) {
1887
1888 addr64_t baddrs, reladd;
1889 thread_t thread;
1890 mapping_t *mp;
1891
1892 baddrs = va & 0xFFFFFFFFF0000000ULL; /* Isolate the segment */
1893 thread = current_thread(); /* Remember our activation */
1894
1895 reladd = baddrs - lowGlo.lgUMWvaddr; /* Get the relocation from user to kernel */
1896
1897 if((thread->machine.umwSpace == map->pmap->space) && (thread->machine.umwRelo == reladd)) { /* Already mapped? */
1898 return ((va & 0x0FFFFFFFULL) | lowGlo.lgUMWvaddr); /* Pass back the kernel address we are to use */
1899 }
1900
1901 disable_preemption(); /* Don't move... */
1902
1903 mp = (mapping_t *)&(getPerProc()->ppUMWmp); /* Make up for C */
1904 thread->machine.umwRelo = reladd; /* Relocation from user to kernel */
1905 mp->mpNestReloc = reladd; /* Relocation from user to kernel */
1906
1907 thread->machine.umwSpace = map->pmap->space; /* Set the address space/pmap lookup ID */
1908 mp->mpSpace = map->pmap->space; /* Set the address space/pmap lookup ID */
1909
1910 /*
1911 * Here we make an assumption that we are going to be using the base pmap's address space.
1912 * If we are wrong, and that would be very, very, very rare, the fault handler will fix us up.
1913 */
1914
1915 hw_map_seg(map->pmap, lowGlo.lgUMWvaddr, baddrs); /* Make the entry for the first segment */
1916
1917 enable_preemption(); /* Let's move */
1918 return ((va & 0x0FFFFFFFULL) | lowGlo.lgUMWvaddr); /* Pass back the kernel address we are to use */
1919 }
1920
1921
1922 /*
1923 * kern_return_t pmap_boot_map(size)
1924 *
1925 * size = size of virtual address range to be mapped
1926 *
1927 * This function is used to assign a range of virtual addresses before VM in
1928 * initialized. It starts at VM_MAX_KERNEL_ADDRESS and works downward.
1929 * The variable vm_last_addr contains the current highest possible VM
1930 * assignable address. It is a panic to attempt to call this after VM has
1931 * started up. The only problem is, is that we may not have the serial or
1932 * framebuffer mapped, so we'll never know we died.........
1933 */
1934
1935 vm_offset_t pmap_boot_map(vm_size_t size) {
1936
1937 if(kernel_map != VM_MAP_NULL) { /* Has VM already started? */
1938 panic("pmap_boot_map: VM started\n");
1939 }
1940
1941 size = round_page(size); /* Make sure this is in pages */
1942 vm_last_addr = vm_last_addr - size; /* Allocate the memory */
1943 return (vm_last_addr + 1); /* Return the vaddr we just allocated */
1944
1945 }
1946
1947
1948 /*
1949 * void pmap_init_sharedpage(void);
1950 *
1951 * Hack map for the 64-bit commpage
1952 */
1953
1954 void pmap_init_sharedpage(vm_offset_t cpg){
1955
1956 addr64_t cva, cpoff;
1957 ppnum_t cpphys;
1958
1959 sharedPmap = pmap_create(0); /* Get a pmap to hold the common segment */
1960 if(!sharedPmap) { /* Check for errors */
1961 panic("pmap_init_sharedpage: couldn't make sharedPmap\n");
1962 }
1963
1964 for(cpoff = 0; cpoff < _COMM_PAGE_AREA_USED; cpoff += 4096) { /* Step along now */
1965
1966 cpphys = pmap_find_phys(kernel_pmap, (addr64_t)cpg + cpoff);
1967 if(!cpphys) {
1968 panic("pmap_init_sharedpage: compage %08X not mapped in kernel\n", cpg + cpoff);
1969 }
1970
1971 cva = mapping_make(sharedPmap, (addr64_t)((uint32_t)_COMM_PAGE_BASE_ADDRESS) + cpoff,
1972 cpphys, mmFlgPerm, 1, VM_PROT_READ); /* Map the page read only */
1973 if(cva) { /* Check for errors */
1974 panic("pmap_init_sharedpage: couldn't map commpage page - cva = %016llX\n", cva);
1975 }
1976
1977 }
1978
1979 return;
1980 }
1981
1982
1983 /*
1984 * void pmap_map_sharedpage(pmap_t pmap);
1985 *
1986 * Maps the last segment in a 64-bit address space
1987 *
1988 *
1989 */
1990
1991 void pmap_map_sharedpage(task_t task, pmap_t pmap){
1992
1993 kern_return_t ret;
1994
1995 if(task_has_64BitAddr(task) || _cpu_capabilities & k64Bit) { /* Should we map the 64-bit page -1? */
1996 ret = pmap_nest(pmap, sharedPmap, 0xFFFFFFFFF0000000ULL, 0x00000000F0000000ULL,
1997 0x0000000010000000ULL); /* Nest the highest possible segment to map comm page */
1998 if(ret != KERN_SUCCESS) { /* Did it work? */
1999 panic("pmap_map_sharedpage: couldn't nest shared page - ret = %08X\n", ret);
2000 }
2001 }
2002
2003 return;
2004 }
2005
2006
2007 /*
2008 * void pmap_unmap_sharedpage(pmap_t pmap);
2009 *
2010 * Unmaps the last segment in a 64-bit address space
2011 *
2012 */
2013
2014 void pmap_unmap_sharedpage(pmap_t pmap){
2015
2016 kern_return_t ret;
2017 mapping_t *mp;
2018 boolean_t inter;
2019 int gotnest;
2020 addr64_t nextva;
2021
2022 if(BootProcInfo.pf.Available & pf64Bit) { /* Are we on a 64-bit machine? */
2023
2024 inter = ml_set_interrupts_enabled(FALSE); /* Disable interruptions for now */
2025 mp = hw_find_map(pmap, 0xFFFFFFFFF0000000ULL, &nextva); /* Find the mapping for this address */
2026 if((unsigned int)mp == mapRtBadLk) { /* Did we lock up ok? */
2027 panic("pmap_unmap_sharedpage: mapping lock failure - rc = %08X, pmap = %08X\n", mp, pmap); /* Die... */
2028 }
2029
2030 gotnest = 0; /* Assume nothing here */
2031 if(mp) {
2032 gotnest = ((mp->mpFlags & mpType) == mpNest);
2033 /* Remember if we have a nest here */
2034 mapping_drop_busy(mp); /* We have everything we need from the mapping */
2035 }
2036 ml_set_interrupts_enabled(inter); /* Put interrupts back to what they were */
2037
2038 if(!gotnest) return; /* Leave if there isn't any nesting here */
2039
2040 ret = pmap_unnest(pmap, 0xFFFFFFFFF0000000ULL); /* Unnest the max 64-bit page */
2041
2042 if(ret != KERN_SUCCESS) { /* Did it work? */
2043 panic("pmap_unmap_sharedpage: couldn't unnest shared page - ret = %08X\n", ret);
2044 }
2045 }
2046
2047 return;
2048 }
2049
2050
2051 /* temporary workaround */
2052 boolean_t
2053 coredumpok(
2054 __unused vm_map_t map,
2055 __unused vm_offset_t va)
2056 {
2057 return TRUE;
2058 }