2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (C) 2020 Justin Hibbits
5 * Copyright (C) 2007-2009 Semihalf, Rafal Jaworowski <raj@semihalf.com>
6 * Copyright (C) 2006 Semihalf, Marian Balakowicz <m8@semihalf.com>
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
19 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
20 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN
21 * NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
22 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
23 * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
24 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
25 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
26 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
27 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29 * Some hw specific parts of this pmap were derived or influenced
30 * by NetBSD's ibm4xx pmap module. More generic code is shared with
31 * a few other pmap modules from the FreeBSD tree.
37 * Kernel and user threads run within one common virtual address space
41 * Virtual address space layout:
42 * -----------------------------
43 * 0x0000_0000 - 0x7fff_ffff : user process
44 * 0x8000_0000 - 0xbfff_ffff : pmap_mapdev()-ed area (PCI/PCIE etc.)
45 * 0xc000_0000 - 0xffff_efff : KVA
48 #include <sys/cdefs.h>
49 __FBSDID("$FreeBSD$");
52 #include "opt_kstack_pages.h"
54 #include <sys/param.h>
56 #include <sys/malloc.h>
60 #include <sys/queue.h>
61 #include <sys/systm.h>
62 #include <sys/kernel.h>
63 #include <sys/kerneldump.h>
64 #include <sys/linker.h>
65 #include <sys/msgbuf.h>
67 #include <sys/mutex.h>
68 #include <sys/rwlock.h>
69 #include <sys/sched.h>
71 #include <sys/vmmeter.h>
74 #include <vm/vm_page.h>
75 #include <vm/vm_kern.h>
76 #include <vm/vm_pageout.h>
77 #include <vm/vm_extern.h>
78 #include <vm/vm_object.h>
79 #include <vm/vm_param.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_pager.h>
82 #include <vm/vm_phys.h>
83 #include <vm/vm_pagequeue.h>
86 #include <machine/_inttypes.h>
87 #include <machine/cpu.h>
88 #include <machine/pcb.h>
89 #include <machine/platform.h>
91 #include <machine/tlb.h>
92 #include <machine/spr.h>
93 #include <machine/md_var.h>
94 #include <machine/mmuvar.h>
95 #include <machine/pmap.h>
96 #include <machine/pte.h>
100 #define PRI0ptrX "08x"
102 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
103 static vm_offset_t zero_page_va;
104 static struct mtx zero_page_mutex;
106 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
107 static vm_offset_t copy_page_src_va;
108 static vm_offset_t copy_page_dst_va;
109 static struct mtx copy_page_mutex;
111 static vm_offset_t kernel_ptbl_root;
112 static unsigned int kernel_ptbls; /* Number of KVA ptbls. */
114 /**************************************************************************/
116 /**************************************************************************/
118 #define VM_MAPDEV_BASE ((vm_offset_t)VM_MAXUSER_ADDRESS + PAGE_SIZE)
120 static void tid_flush(tlbtid_t tid);
121 static unsigned long ilog2(unsigned long);
123 /**************************************************************************/
124 /* Page table management */
125 /**************************************************************************/
127 #define PMAP_ROOT_SIZE (sizeof(pte_t**) * PDIR_NENTRIES)
128 static void ptbl_init(void);
129 static struct ptbl_buf *ptbl_buf_alloc(void);
130 static void ptbl_buf_free(struct ptbl_buf *);
131 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
133 static pte_t *ptbl_alloc(pmap_t, unsigned int, boolean_t);
134 static void ptbl_free(pmap_t, unsigned int);
135 static void ptbl_hold(pmap_t, unsigned int);
136 static int ptbl_unhold(pmap_t, unsigned int);
138 static vm_paddr_t pte_vatopa(pmap_t, vm_offset_t);
139 static int pte_enter(pmap_t, vm_page_t, vm_offset_t, uint32_t, boolean_t);
140 static int pte_remove(pmap_t, vm_offset_t, uint8_t);
141 static pte_t *pte_find(pmap_t, vm_offset_t);
144 TAILQ_ENTRY(ptbl_buf) link; /* list link */
145 vm_offset_t kva; /* va of mapping */
148 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
149 #define PTBL_BUFS (128 * 16)
151 /* ptbl free list and a lock used for access synchronization. */
152 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
153 static struct mtx ptbl_buf_freelist_lock;
155 /* Base address of kva space allocated fot ptbl bufs. */
156 static vm_offset_t ptbl_buf_pool_vabase;
158 /* Pointer to ptbl_buf structures. */
159 static struct ptbl_buf *ptbl_bufs;
161 /**************************************************************************/
162 /* Page table related */
163 /**************************************************************************/
165 /* Initialize pool of kva ptbl buffers. */
171 CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
172 (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
173 CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
174 __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
176 mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
177 TAILQ_INIT(&ptbl_buf_freelist);
179 for (i = 0; i < PTBL_BUFS; i++) {
181 ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
182 TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
186 /* Get a ptbl_buf from the freelist. */
187 static struct ptbl_buf *
190 struct ptbl_buf *buf;
192 mtx_lock(&ptbl_buf_freelist_lock);
193 buf = TAILQ_FIRST(&ptbl_buf_freelist);
195 TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
196 mtx_unlock(&ptbl_buf_freelist_lock);
198 CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
203 /* Return ptbl buff to free pool. */
205 ptbl_buf_free(struct ptbl_buf *buf)
208 CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
210 mtx_lock(&ptbl_buf_freelist_lock);
211 TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
212 mtx_unlock(&ptbl_buf_freelist_lock);
216 * Search the list of allocated ptbl bufs and find on list of allocated ptbls
219 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
221 struct ptbl_buf *pbuf;
223 CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
225 PMAP_LOCK_ASSERT(pmap, MA_OWNED);
227 TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
228 if (pbuf->kva == (vm_offset_t)ptbl) {
229 /* Remove from pmap ptbl buf list. */
230 TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
232 /* Free corresponding ptbl buf. */
238 /* Allocate page table. */
240 ptbl_alloc(pmap_t pmap, unsigned int pdir_idx, boolean_t nosleep)
242 vm_page_t mtbl[PTBL_PAGES];
244 struct ptbl_buf *pbuf;
249 CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
250 (pmap == kernel_pmap), pdir_idx);
252 KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
253 ("ptbl_alloc: invalid pdir_idx"));
254 KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
255 ("pte_alloc: valid ptbl entry exists!"));
257 pbuf = ptbl_buf_alloc();
259 panic("pte_alloc: couldn't alloc kernel virtual memory");
261 ptbl = (pte_t *)pbuf->kva;
263 CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
265 for (i = 0; i < PTBL_PAGES; i++) {
266 pidx = (PTBL_PAGES * pdir_idx) + i;
267 while ((m = vm_page_alloc_noobj(VM_ALLOC_WIRED)) == NULL) {
269 ptbl_free_pmap_ptbl(pmap, ptbl);
270 for (j = 0; j < i; j++)
271 vm_page_free(mtbl[j]);
276 rw_wunlock(&pvh_global_lock);
278 rw_wlock(&pvh_global_lock);
285 /* Map allocated pages into kernel_pmap. */
286 mmu_booke_qenter((vm_offset_t)ptbl, mtbl, PTBL_PAGES);
288 /* Zero whole ptbl. */
289 bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
291 /* Add pbuf to the pmap ptbl bufs list. */
292 TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
297 /* Free ptbl pages and invalidate pdir entry. */
299 ptbl_free(pmap_t pmap, unsigned int pdir_idx)
307 CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
308 (pmap == kernel_pmap), pdir_idx);
310 KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
311 ("ptbl_free: invalid pdir_idx"));
313 ptbl = pmap->pm_pdir[pdir_idx];
315 CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
317 KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
320 * Invalidate the pdir entry as soon as possible, so that other CPUs
321 * don't attempt to look up the page tables we are releasing.
323 mtx_lock_spin(&tlbivax_mutex);
326 pmap->pm_pdir[pdir_idx] = NULL;
329 mtx_unlock_spin(&tlbivax_mutex);
331 for (i = 0; i < PTBL_PAGES; i++) {
332 va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
333 pa = pte_vatopa(kernel_pmap, va);
334 m = PHYS_TO_VM_PAGE(pa);
335 vm_page_free_zero(m);
337 mmu_booke_kremove(va);
340 ptbl_free_pmap_ptbl(pmap, ptbl);
344 * Decrement ptbl pages hold count and attempt to free ptbl pages.
345 * Called when removing pte entry from ptbl.
347 * Return 1 if ptbl pages were freed.
350 ptbl_unhold(pmap_t pmap, unsigned int pdir_idx)
357 CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
358 (pmap == kernel_pmap), pdir_idx);
360 KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
361 ("ptbl_unhold: invalid pdir_idx"));
362 KASSERT((pmap != kernel_pmap),
363 ("ptbl_unhold: unholding kernel ptbl!"));
365 ptbl = pmap->pm_pdir[pdir_idx];
367 //debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
368 KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
369 ("ptbl_unhold: non kva ptbl"));
371 /* decrement hold count */
372 for (i = 0; i < PTBL_PAGES; i++) {
373 pa = pte_vatopa(kernel_pmap,
374 (vm_offset_t)ptbl + (i * PAGE_SIZE));
375 m = PHYS_TO_VM_PAGE(pa);
380 * Free ptbl pages if there are no pte etries in this ptbl.
381 * ref_count has the same value for all ptbl pages, so check the last
384 if (m->ref_count == 0) {
385 ptbl_free(pmap, pdir_idx);
387 //debugf("ptbl_unhold: e (freed ptbl)\n");
395 * Increment hold count for ptbl pages. This routine is used when a new pte
396 * entry is being inserted into the ptbl.
399 ptbl_hold(pmap_t pmap, unsigned int pdir_idx)
406 CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
409 KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
410 ("ptbl_hold: invalid pdir_idx"));
411 KASSERT((pmap != kernel_pmap),
412 ("ptbl_hold: holding kernel ptbl!"));
414 ptbl = pmap->pm_pdir[pdir_idx];
416 KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
418 for (i = 0; i < PTBL_PAGES; i++) {
419 pa = pte_vatopa(kernel_pmap,
420 (vm_offset_t)ptbl + (i * PAGE_SIZE));
421 m = PHYS_TO_VM_PAGE(pa);
427 * Clean pte entry, try to free page table page if requested.
429 * Return 1 if ptbl pages were freed, otherwise return 0.
432 pte_remove(pmap_t pmap, vm_offset_t va, uint8_t flags)
434 unsigned int pdir_idx = PDIR_IDX(va);
435 unsigned int ptbl_idx = PTBL_IDX(va);
440 //int su = (pmap == kernel_pmap);
441 //debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
442 // su, (u_int32_t)pmap, va, flags);
444 ptbl = pmap->pm_pdir[pdir_idx];
445 KASSERT(ptbl, ("pte_remove: null ptbl"));
447 pte = &ptbl[ptbl_idx];
449 if (pte == NULL || !PTE_ISVALID(pte))
452 if (PTE_ISWIRED(pte))
453 pmap->pm_stats.wired_count--;
455 /* Get vm_page_t for mapped pte. */
456 m = PHYS_TO_VM_PAGE(PTE_PA(pte));
458 /* Handle managed entry. */
459 if (PTE_ISMANAGED(pte)) {
460 if (PTE_ISMODIFIED(pte))
463 if (PTE_ISREFERENCED(pte))
464 vm_page_aflag_set(m, PGA_REFERENCED);
466 pv_remove(pmap, va, m);
467 } else if (pmap == kernel_pmap && m && m->md.pv_tracked) {
469 * Always pv_insert()/pv_remove() on MPC85XX, in case DPAA is
470 * used. This is needed by the NCSW support code for fast
471 * VA<->PA translation.
473 pv_remove(pmap, va, m);
474 if (TAILQ_EMPTY(&m->md.pv_list))
475 m->md.pv_tracked = false;
478 mtx_lock_spin(&tlbivax_mutex);
481 tlb0_flush_entry(va);
485 mtx_unlock_spin(&tlbivax_mutex);
487 pmap->pm_stats.resident_count--;
489 if (flags & PTBL_UNHOLD) {
490 //debugf("pte_remove: e (unhold)\n");
491 return (ptbl_unhold(pmap, pdir_idx));
494 //debugf("pte_remove: e\n");
499 * Insert PTE for a given page and virtual address.
502 pte_enter(pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
505 unsigned int pdir_idx = PDIR_IDX(va);
506 unsigned int ptbl_idx = PTBL_IDX(va);
507 pte_t *ptbl, *pte, pte_tmp;
509 CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
510 pmap == kernel_pmap, pmap, va);
512 /* Get the page table pointer. */
513 ptbl = pmap->pm_pdir[pdir_idx];
516 /* Allocate page table pages. */
517 ptbl = ptbl_alloc(pmap, pdir_idx, nosleep);
519 KASSERT(nosleep, ("nosleep and NULL ptbl"));
522 pmap->pm_pdir[pdir_idx] = ptbl;
523 pte = &ptbl[ptbl_idx];
526 * Check if there is valid mapping for requested
527 * va, if there is, remove it.
529 pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
530 if (PTE_ISVALID(pte)) {
531 pte_remove(pmap, va, PTBL_HOLD);
534 * pte is not used, increment hold count
537 if (pmap != kernel_pmap)
538 ptbl_hold(pmap, pdir_idx);
543 * Insert pv_entry into pv_list for mapped page if part of managed
546 if ((m->oflags & VPO_UNMANAGED) == 0) {
547 flags |= PTE_MANAGED;
549 /* Create and insert pv entry. */
550 pv_insert(pmap, va, m);
553 pmap->pm_stats.resident_count++;
555 pte_tmp = PTE_RPN_FROM_PA(VM_PAGE_TO_PHYS(m));
556 pte_tmp |= (PTE_VALID | flags | PTE_PS_4KB); /* 4KB pages only */
558 mtx_lock_spin(&tlbivax_mutex);
561 tlb0_flush_entry(va);
565 mtx_unlock_spin(&tlbivax_mutex);
569 /* Return the pa for the given pmap/va. */
571 pte_vatopa(pmap_t pmap, vm_offset_t va)
576 pte = pte_find(pmap, va);
577 if ((pte != NULL) && PTE_ISVALID(pte))
578 pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
582 /* Get a pointer to a PTE in a page table. */
584 pte_find(pmap_t pmap, vm_offset_t va)
586 unsigned int pdir_idx = PDIR_IDX(va);
587 unsigned int ptbl_idx = PTBL_IDX(va);
589 KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
591 if (pmap->pm_pdir[pdir_idx])
592 return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
597 /* Get a pointer to a PTE in a page table, or the next closest (greater) one. */
598 static __inline pte_t *
599 pte_find_next(pmap_t pmap, vm_offset_t *pva)
606 KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
611 pdir = pmap->pm_pdir;
612 for (; i < PDIR_NENTRIES; i++, j = 0) {
615 for (; j < PTBL_NENTRIES; j++) {
617 if (!PTE_ISVALID(pte))
619 *pva = PDIR_SIZE * i + PAGE_SIZE * j;
626 /* Set up kernel page tables. */
628 kernel_pte_alloc(vm_offset_t data_end, vm_offset_t addr)
632 vm_offset_t pdir_start;
635 kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
636 kernel_pmap->pm_pdir = (pte_t **)kernel_ptbl_root;
638 pdir_start = kernel_ptbl_root + PDIR_NENTRIES * sizeof(pte_t);
640 /* Initialize kernel pdir */
641 for (i = 0; i < kernel_ptbls; i++) {
642 kernel_pmap->pm_pdir[kptbl_min + i] =
643 (pte_t *)(pdir_start + (i * PAGE_SIZE * PTBL_PAGES));
647 * Fill in PTEs covering kernel code and data. They are not required
648 * for address translation, as this area is covered by static TLB1
649 * entries, but for pte_vatopa() to work correctly with kernel area
652 for (va = addr; va < data_end; va += PAGE_SIZE) {
653 pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
655 *pte = PTE_RPN_FROM_PA(kernload + (va - kernstart));
656 *pte |= PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
657 PTE_VALID | PTE_PS_4KB;
662 mmu_booke_alloc_kernel_pgtables(vm_offset_t data_end)
664 /* Allocate space for ptbl_bufs. */
665 ptbl_bufs = (struct ptbl_buf *)data_end;
666 data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
667 debugf(" ptbl_bufs at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
668 (uintptr_t)ptbl_bufs, data_end);
670 data_end = round_page(data_end);
672 kernel_ptbl_root = data_end;
673 data_end += PDIR_NENTRIES * sizeof(pte_t*);
675 /* Allocate PTE tables for kernel KVA. */
676 kernel_ptbls = howmany(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS,
678 data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
679 debugf(" kernel ptbls: %d\n", kernel_ptbls);
680 debugf(" kernel pdir at %#jx end = %#jx\n",
681 (uintmax_t)kernel_ptbl_root, (uintmax_t)data_end);
687 * Initialize a preallocated and zeroed pmap structure,
688 * such as one in a vmspace structure.
691 mmu_booke_pinit(pmap_t pmap)
695 CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
696 curthread->td_proc->p_pid, curthread->td_proc->p_comm);
698 KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
700 for (i = 0; i < MAXCPU; i++)
701 pmap->pm_tid[i] = TID_NONE;
702 CPU_ZERO(&kernel_pmap->pm_active);
703 bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
704 pmap->pm_pdir = uma_zalloc(ptbl_root_zone, M_WAITOK);
705 bzero(pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
706 TAILQ_INIT(&pmap->pm_ptbl_list);
712 * Release any resources held by the given physical map.
713 * Called when a pmap initialized by mmu_booke_pinit is being released.
714 * Should only be called if the map contains no valid mappings.
717 mmu_booke_release(pmap_t pmap)
720 KASSERT(pmap->pm_stats.resident_count == 0,
721 ("pmap_release: pmap resident count %ld != 0",
722 pmap->pm_stats.resident_count));
723 uma_zfree(ptbl_root_zone, pmap->pm_pdir);
727 mmu_booke_sync_icache(pmap_t pm, vm_offset_t va, vm_size_t sz)
737 rw_wlock(&pvh_global_lock);
738 pmap = PCPU_GET(curpmap);
739 active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
742 pte = pte_find(pm, va);
743 valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
747 sync_sz = PAGE_SIZE - (va & PAGE_MASK);
748 sync_sz = min(sync_sz, sz);
752 * Create a mapping in the active pmap.
754 * XXX: We use the zero page here, because
755 * it isn't likely to be in use.
756 * If we ever decide to support
757 * security.bsd.map_at_zero on Book-E, change
758 * this to some other address that isn't
762 m = PHYS_TO_VM_PAGE(pa);
764 pte_enter(pmap, m, addr,
765 PTE_SR | PTE_VALID, FALSE);
766 __syncicache((void *)(addr + (va & PAGE_MASK)),
768 pte_remove(pmap, addr, PTBL_UNHOLD);
771 __syncicache((void *)va, sync_sz);
776 rw_wunlock(&pvh_global_lock);
780 * mmu_booke_zero_page_area zeros the specified hardware page by
781 * mapping it into virtual memory and using bzero to clear
784 * off and size must reside within a single page.
787 mmu_booke_zero_page_area(vm_page_t m, int off, int size)
791 /* XXX KASSERT off and size are within a single page? */
793 mtx_lock(&zero_page_mutex);
796 mmu_booke_kenter(va, VM_PAGE_TO_PHYS(m));
797 bzero((caddr_t)va + off, size);
798 mmu_booke_kremove(va);
800 mtx_unlock(&zero_page_mutex);
804 * mmu_booke_zero_page zeros the specified hardware page.
807 mmu_booke_zero_page(vm_page_t m)
812 mtx_lock(&zero_page_mutex);
814 mmu_booke_kenter(va, VM_PAGE_TO_PHYS(m));
816 for (off = 0; off < PAGE_SIZE; off += cacheline_size)
817 __asm __volatile("dcbz 0,%0" :: "r"(va + off));
819 mmu_booke_kremove(va);
821 mtx_unlock(&zero_page_mutex);
825 * mmu_booke_copy_page copies the specified (machine independent) page by
826 * mapping the page into virtual memory and using memcopy to copy the page,
827 * one machine dependent page at a time.
830 mmu_booke_copy_page(vm_page_t sm, vm_page_t dm)
832 vm_offset_t sva, dva;
834 sva = copy_page_src_va;
835 dva = copy_page_dst_va;
837 mtx_lock(©_page_mutex);
838 mmu_booke_kenter(sva, VM_PAGE_TO_PHYS(sm));
839 mmu_booke_kenter(dva, VM_PAGE_TO_PHYS(dm));
841 memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
843 mmu_booke_kremove(dva);
844 mmu_booke_kremove(sva);
845 mtx_unlock(©_page_mutex);
849 mmu_booke_copy_pages(vm_page_t *ma, vm_offset_t a_offset,
850 vm_page_t *mb, vm_offset_t b_offset, int xfersize)
853 vm_offset_t a_pg_offset, b_pg_offset;
856 mtx_lock(©_page_mutex);
857 while (xfersize > 0) {
858 a_pg_offset = a_offset & PAGE_MASK;
859 cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
860 mmu_booke_kenter(copy_page_src_va,
861 VM_PAGE_TO_PHYS(ma[a_offset >> PAGE_SHIFT]));
862 a_cp = (char *)copy_page_src_va + a_pg_offset;
863 b_pg_offset = b_offset & PAGE_MASK;
864 cnt = min(cnt, PAGE_SIZE - b_pg_offset);
865 mmu_booke_kenter(copy_page_dst_va,
866 VM_PAGE_TO_PHYS(mb[b_offset >> PAGE_SHIFT]));
867 b_cp = (char *)copy_page_dst_va + b_pg_offset;
868 bcopy(a_cp, b_cp, cnt);
869 mmu_booke_kremove(copy_page_dst_va);
870 mmu_booke_kremove(copy_page_src_va);
875 mtx_unlock(©_page_mutex);
879 mmu_booke_quick_enter_page(vm_page_t m)
886 paddr = VM_PAGE_TO_PHYS(m);
888 flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
889 flags |= tlb_calc_wimg(paddr, pmap_page_get_memattr(m)) << PTE_MAS2_SHIFT;
893 qaddr = PCPU_GET(qmap_addr);
895 pte = pte_find(kernel_pmap, qaddr);
897 KASSERT(*pte == 0, ("mmu_booke_quick_enter_page: PTE busy"));
900 * XXX: tlbivax is broadcast to other cores, but qaddr should
901 * not be present in other TLBs. Is there a better instruction
902 * sequence to use? Or just forget it & use mmu_booke_kenter()...
904 __asm __volatile("tlbivax 0, %0" :: "r"(qaddr & MAS2_EPN_MASK));
905 __asm __volatile("isync; msync");
907 *pte = PTE_RPN_FROM_PA(paddr) | flags;
909 /* Flush the real memory from the instruction cache. */
910 if ((flags & (PTE_I | PTE_G)) == 0)
911 __syncicache((void *)qaddr, PAGE_SIZE);
917 mmu_booke_quick_remove_page(vm_offset_t addr)
921 pte = pte_find(kernel_pmap, addr);
923 KASSERT(PCPU_GET(qmap_addr) == addr,
924 ("mmu_booke_quick_remove_page: invalid address"));
926 ("mmu_booke_quick_remove_page: PTE not in use"));
932 /**************************************************************************/
934 /**************************************************************************/
937 * Return the largest uint value log such that 2^log <= num.
940 ilog2(unsigned long num)
944 __asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
949 * Invalidate all TLB0 entries which match the given TID. Note this is
950 * dedicated for cases when invalidations should NOT be propagated to other
954 tid_flush(tlbtid_t tid)
957 uint32_t mas0, mas1, mas2;
960 /* Don't evict kernel translations */
961 if (tid == TID_KERNEL)
965 __asm __volatile("wrteei 0");
968 * Newer (e500mc and later) have tlbilx, which doesn't broadcast, so use
969 * it for PID invalidation.
971 switch ((mfpvr() >> 16) & 0xffff) {
975 mtspr(SPR_MAS6, tid << MAS6_SPID0_SHIFT);
977 __asm __volatile("isync; .long 0x7c200024; isync; msync");
978 __asm __volatile("wrtee %0" :: "r"(msr));
982 for (way = 0; way < TLB0_WAYS; way++)
983 for (entry = 0; entry < TLB0_ENTRIES_PER_WAY; entry++) {
984 mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
985 mtspr(SPR_MAS0, mas0);
987 mas2 = entry << MAS2_TLB0_ENTRY_IDX_SHIFT;
988 mtspr(SPR_MAS2, mas2);
990 __asm __volatile("isync; tlbre");
992 mas1 = mfspr(SPR_MAS1);
994 if (!(mas1 & MAS1_VALID))
996 if (((mas1 & MAS1_TID_MASK) >> MAS1_TID_SHIFT) != tid)
999 mtspr(SPR_MAS1, mas1);
1000 __asm __volatile("isync; tlbwe; isync; msync");
1002 __asm __volatile("wrtee %0" :: "r"(msr));