2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD AND 4-Clause-BSD
4 * Copyright (c) 2001 The NetBSD Foundation, Inc.
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Matt Thomas <matt@3am-software.com> of Allegro Networks, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 * POSSIBILITY OF SUCH DAMAGE.
32 * Copyright (C) 1995, 1996 Wolfgang Solfrank.
33 * Copyright (C) 1995, 1996 TooLs GmbH.
34 * All rights reserved.
36 * Redistribution and use in source and binary forms, with or without
37 * modification, are permitted provided that the following conditions
39 * 1. Redistributions of source code must retain the above copyright
40 * notice, this list of conditions and the following disclaimer.
41 * 2. Redistributions in binary form must reproduce the above copyright
42 * notice, this list of conditions and the following disclaimer in the
43 * documentation and/or other materials provided with the distribution.
44 * 3. All advertising materials mentioning features or use of this software
45 * must display the following acknowledgement:
46 * This product includes software developed by TooLs GmbH.
47 * 4. The name of TooLs GmbH may not be used to endorse or promote products
48 * derived from this software without specific prior written permission.
50 * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
51 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
52 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
53 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
54 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
55 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
56 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
57 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
58 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
59 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
61 * $NetBSD: pmap.c,v 1.28 2000/03/26 20:42:36 kleink Exp $
64 * Copyright (C) 2001 Benno Rice.
65 * All rights reserved.
67 * Redistribution and use in source and binary forms, with or without
68 * modification, are permitted provided that the following conditions
70 * 1. Redistributions of source code must retain the above copyright
71 * notice, this list of conditions and the following disclaimer.
72 * 2. Redistributions in binary form must reproduce the above copyright
73 * notice, this list of conditions and the following disclaimer in the
74 * documentation and/or other materials provided with the distribution.
76 * THIS SOFTWARE IS PROVIDED BY Benno Rice ``AS IS'' AND ANY EXPRESS OR
77 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
78 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
79 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
80 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
81 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
82 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
83 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
84 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
85 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
88 #include <sys/cdefs.h>
89 __FBSDID("$FreeBSD$");
92 * Native 64-bit page table operations for running without a hypervisor.
95 #include <sys/param.h>
96 #include <sys/kernel.h>
99 #include <sys/mutex.h>
100 #include <sys/proc.h>
101 #include <sys/sched.h>
102 #include <sys/sysctl.h>
103 #include <sys/systm.h>
104 #include <sys/rwlock.h>
105 #include <sys/endian.h>
110 #include <vm/vm_param.h>
111 #include <vm/vm_kern.h>
112 #include <vm/vm_page.h>
113 #include <vm/vm_map.h>
114 #include <vm/vm_object.h>
115 #include <vm/vm_extern.h>
116 #include <vm/vm_pageout.h>
118 #include <machine/cpu.h>
119 #include <machine/hid.h>
120 #include <machine/md_var.h>
121 #include <machine/mmuvar.h>
123 #include "mmu_oea64.h"
125 #include "moea64_if.h"
127 #define PTESYNC() __asm __volatile("ptesync");
128 #define TLBSYNC() __asm __volatile("tlbsync; ptesync");
129 #define SYNC() __asm __volatile("sync");
130 #define EIEIO() __asm __volatile("eieio");
132 #define VSID_HASH_MASK 0x0000007fffffffffULL
134 /* POWER9 only permits a 64k partition table size. */
135 #define PART_SIZE 0x10000
137 static int moea64_crop_tlbie;
140 TLBIE(uint64_t vpn) {
141 #ifndef __powerpc64__
142 register_t vpn_hi, vpn_lo;
144 register_t scratch, intr;
147 static volatile u_int tlbie_lock = 0;
149 vpn <<= ADDR_PIDX_SHFT;
151 /* Hobo spinlock: we need stronger guarantees than mutexes provide */
152 while (!atomic_cmpset_int(&tlbie_lock, 0, 1));
153 isync(); /* Flush instruction queue once lock acquired */
155 if (moea64_crop_tlbie)
156 vpn &= ~(0xffffULL << 48);
159 __asm __volatile("tlbie %0" :: "r"(vpn) : "memory");
160 __asm __volatile("eieio; tlbsync; ptesync" ::: "memory");
162 vpn_hi = (uint32_t)(vpn >> 32);
163 vpn_lo = (uint32_t)vpn;
165 intr = intr_disable();
180 : "=r"(msr), "=r"(scratch) : "r"(vpn_hi), "r"(vpn_lo), "r"(32), "r"(1)
185 /* No barriers or special ops -- taken care of by ptesync above */
189 #define DISABLE_TRANS(msr) msr = mfmsr(); mtmsr(msr & ~PSL_DR)
190 #define ENABLE_TRANS(msr) mtmsr(msr)
195 static volatile struct lpte *moea64_pteg_table;
196 static struct rwlock moea64_eviction_lock;
201 static int moea64_pte_insert_native(mmu_t, struct pvo_entry *);
202 static int64_t moea64_pte_synch_native(mmu_t, struct pvo_entry *);
203 static int64_t moea64_pte_clear_native(mmu_t, struct pvo_entry *, uint64_t);
204 static int64_t moea64_pte_replace_native(mmu_t, struct pvo_entry *, int);
205 static int64_t moea64_pte_unset_native(mmu_t mmu, struct pvo_entry *);
210 static void moea64_bootstrap_native(mmu_t mmup,
211 vm_offset_t kernelstart, vm_offset_t kernelend);
212 static void moea64_cpu_bootstrap_native(mmu_t, int ap);
213 static void tlbia(void);
215 static mmu_method_t moea64_native_methods[] = {
216 /* Internal interfaces */
217 MMUMETHOD(mmu_bootstrap, moea64_bootstrap_native),
218 MMUMETHOD(mmu_cpu_bootstrap, moea64_cpu_bootstrap_native),
220 MMUMETHOD(moea64_pte_synch, moea64_pte_synch_native),
221 MMUMETHOD(moea64_pte_clear, moea64_pte_clear_native),
222 MMUMETHOD(moea64_pte_unset, moea64_pte_unset_native),
223 MMUMETHOD(moea64_pte_replace, moea64_pte_replace_native),
224 MMUMETHOD(moea64_pte_insert, moea64_pte_insert_native),
229 MMU_DEF_INHERIT(oea64_mmu_native, MMU_TYPE_G5, moea64_native_methods,
233 moea64_pte_synch_native(mmu_t mmu, struct pvo_entry *pvo)
235 volatile struct lpte *pt = moea64_pteg_table + pvo->pvo_pte.slot;
236 struct lpte properpt;
239 PMAP_LOCK_ASSERT(pvo->pvo_pmap, MA_OWNED);
241 moea64_pte_from_pvo(pvo, &properpt);
243 rw_rlock(&moea64_eviction_lock);
244 if ((be64toh(pt->pte_hi) & LPTE_AVPN_MASK) !=
245 (properpt.pte_hi & LPTE_AVPN_MASK)) {
247 rw_runlock(&moea64_eviction_lock);
252 ptelo = be64toh(pt->pte_lo);
254 rw_runlock(&moea64_eviction_lock);
256 return (ptelo & (LPTE_REF | LPTE_CHG));
260 moea64_pte_clear_native(mmu_t mmu, struct pvo_entry *pvo, uint64_t ptebit)
262 volatile struct lpte *pt = moea64_pteg_table + pvo->pvo_pte.slot;
263 struct lpte properpt;
266 PMAP_LOCK_ASSERT(pvo->pvo_pmap, MA_OWNED);
268 moea64_pte_from_pvo(pvo, &properpt);
270 rw_rlock(&moea64_eviction_lock);
271 if ((be64toh(pt->pte_hi) & LPTE_AVPN_MASK) !=
272 (properpt.pte_hi & LPTE_AVPN_MASK)) {
274 rw_runlock(&moea64_eviction_lock);
278 if (ptebit == LPTE_REF) {
279 /* See "Resetting the Reference Bit" in arch manual */
281 /* 2-step here safe: precision is not guaranteed */
282 ptelo = be64toh(pt->pte_lo);
284 /* One-byte store to avoid touching the C bit */
285 ((volatile uint8_t *)(&pt->pte_lo))[6] =
286 #if BYTE_ORDER == BIG_ENDIAN
287 ((uint8_t *)(&properpt.pte_lo))[6];
289 ((uint8_t *)(&properpt.pte_lo))[1];
291 rw_runlock(&moea64_eviction_lock);
297 rw_runlock(&moea64_eviction_lock);
298 ptelo = moea64_pte_unset_native(mmu, pvo);
299 moea64_pte_insert_native(mmu, pvo);
302 return (ptelo & (LPTE_REF | LPTE_CHG));
306 moea64_pte_unset_native(mmu_t mmu, struct pvo_entry *pvo)
308 volatile struct lpte *pt = moea64_pteg_table + pvo->pvo_pte.slot;
309 struct lpte properpt;
312 moea64_pte_from_pvo(pvo, &properpt);
314 rw_rlock(&moea64_eviction_lock);
315 if ((be64toh(pt->pte_hi & LPTE_AVPN_MASK)) !=
316 (properpt.pte_hi & LPTE_AVPN_MASK)) {
318 moea64_pte_overflow--;
319 rw_runlock(&moea64_eviction_lock);
324 * Invalidate the pte, briefly locking it to collect RC bits. No
325 * atomics needed since this is protected against eviction by the lock.
329 pt->pte_hi = be64toh((pt->pte_hi & ~LPTE_VALID) | LPTE_LOCKED);
332 ptelo = be64toh(pt->pte_lo);
333 *((volatile int32_t *)(&pt->pte_hi) + 1) = 0; /* Release lock */
335 rw_runlock(&moea64_eviction_lock);
337 /* Keep statistics */
340 return (ptelo & (LPTE_CHG | LPTE_REF));
344 moea64_pte_replace_native(mmu_t mmu, struct pvo_entry *pvo, int flags)
346 volatile struct lpte *pt = moea64_pteg_table + pvo->pvo_pte.slot;
347 struct lpte properpt;
351 /* Just some software bits changing. */
352 moea64_pte_from_pvo(pvo, &properpt);
354 rw_rlock(&moea64_eviction_lock);
355 if ((be64toh(pt->pte_hi) & LPTE_AVPN_MASK) !=
356 (properpt.pte_hi & LPTE_AVPN_MASK)) {
357 rw_runlock(&moea64_eviction_lock);
360 pt->pte_hi = htobe64(properpt.pte_hi);
361 ptelo = be64toh(pt->pte_lo);
362 rw_runlock(&moea64_eviction_lock);
364 /* Otherwise, need reinsertion and deletion */
365 ptelo = moea64_pte_unset_native(mmu, pvo);
366 moea64_pte_insert_native(mmu, pvo);
373 moea64_cpu_bootstrap_native(mmu_t mmup, int ap)
377 struct slb *slb = PCPU_GET(aim.slb);
382 * Initialize segment registers and MMU
385 mtmsr(mfmsr() & ~PSL_DR & ~PSL_IR);
388 * Install kernel SLB entries
392 __asm __volatile ("slbia");
393 __asm __volatile ("slbmfee %0,%1; slbie %0;" : "=r"(seg0) :
396 for (i = 0; i < n_slbs; i++) {
397 if (!(slb[i].slbe & SLBE_VALID))
400 __asm __volatile ("slbmte %0, %1" ::
401 "r"(slb[i].slbv), "r"(slb[i].slbe));
404 for (i = 0; i < 16; i++)
405 mtsrin(i << ADDR_SR_SHFT, kernel_pmap->pm_sr[i]);
412 __asm __volatile ("ptesync; mtsdr1 %0; isync"
413 :: "r"(((uintptr_t)moea64_pteg_table & ~DMAP_BASE_ADDRESS)
414 | (uintptr_t)(flsl(moea64_pteg_mask >> 11))));
419 moea64_bootstrap_native(mmu_t mmup, vm_offset_t kernelstart,
420 vm_offset_t kernelend)
427 moea64_early_bootstrap(mmup, kernelstart, kernelend);
429 switch (mfpvr() >> 16) {
436 moea64_crop_tlbie = true;
439 * Allocate PTEG table.
442 size = moea64_pteg_count * sizeof(struct lpteg);
443 CTR2(KTR_PMAP, "moea64_bootstrap: %lu PTEGs, %lu bytes",
444 moea64_pteg_count, size);
445 rw_init(&moea64_eviction_lock, "pte eviction");
448 * We now need to allocate memory. This memory, to be allocated,
449 * has to reside in a page table. The page table we are about to
450 * allocate. We don't have BAT. So drop to data real mode for a minute
451 * as a measure of last resort. We do this a couple times.
454 * PTEG table must be aligned on a 256k boundary, but can be placed
455 * anywhere with that alignment on POWER ISA 3+ systems. On earlier
456 * systems, offset addition is done by the CPU with bitwise OR rather
457 * than addition, so the table must also be aligned on a boundary of
458 * its own size. Pick the larger of the two, which works on all
461 moea64_pteg_table = (struct lpte *)moea64_bootstrap_alloc(size,
462 MAX(256*1024, size));
465 (struct lpte *)PHYS_TO_DMAP((vm_offset_t)moea64_pteg_table);
467 bzero(__DEVOLATILE(void *, moea64_pteg_table), moea64_pteg_count *
468 sizeof(struct lpteg));
471 CTR1(KTR_PMAP, "moea64_bootstrap: PTEG table at %p", moea64_pteg_table);
473 moea64_mid_bootstrap(mmup, kernelstart, kernelend);
476 * Add a mapping for the page table itself if there is no direct map.
478 if (!hw_direct_map) {
479 size = moea64_pteg_count * sizeof(struct lpteg);
480 off = (vm_offset_t)(moea64_pteg_table);
482 for (pa = off; pa < off + size; pa += PAGE_SIZE)
487 /* Bring up virtual memory */
488 moea64_late_bootstrap(mmup, kernelstart, kernelend);
495 #ifndef __powerpc64__
496 register_t msr, scratch;
499 i = 0xc00; /* IS = 11 */
500 switch (mfpvr() >> 16) {
509 i = 0; /* IS not supported */
515 for (; i < 0x200000; i += 0x00001000) {
517 __asm __volatile("tlbiel %0" :: "r"(i));
530 : "=r"(msr), "=r"(scratch) : "r"(i), "r"(1));
539 atomic_pte_lock(volatile struct lpte *pte, uint64_t bitmask, uint64_t *oldhi)
545 * Note: in principle, if just the locked bit were set here, we
546 * could avoid needing the eviction lock. However, eviction occurs
547 * so rarely that it isn't worth bothering about in practice.
551 "1:\tlwarx %1, 0, %3\n\t" /* load old value */
552 "and. %0,%1,%4\n\t" /* check if any bits set */
553 "bne 2f\n\t" /* exit if any set */
554 "stwcx. %5, 0, %3\n\t" /* attempt to store */
555 "bne- 1b\n\t" /* spin if failed */
556 "li %0, 1\n\t" /* success - retval = 1 */
557 "b 3f\n\t" /* we've succeeded */
559 "stwcx. %1, 0, %3\n\t" /* clear reservation (74xx) */
560 "li %0, 0\n\t" /* failure - retval = 0 */
562 : "=&r" (ret), "=&r"(oldhihalf), "=m" (pte->pte_hi)
563 : "r" ((volatile char *)&pte->pte_hi + 4),
564 "r" ((uint32_t)bitmask), "r" ((uint32_t)LPTE_LOCKED),
566 : "cr0", "cr1", "cr2", "memory");
568 *oldhi = (pte->pte_hi & 0xffffffff00000000ULL) | oldhihalf;
574 moea64_insert_to_pteg_native(struct lpte *pvo_pt, uintptr_t slotbase,
577 volatile struct lpte *pt;
578 uint64_t oldptehi, va;
582 /* Start at a random slot */
584 for (j = 0; j < 8; j++) {
585 k = slotbase + (i + j) % 8;
586 pt = &moea64_pteg_table[k];
587 /* Invalidate and seize lock only if no bits in mask set */
588 if (atomic_pte_lock(pt, mask, &oldptehi)) /* Lock obtained */
595 if (oldptehi & LPTE_VALID) {
596 KASSERT(!(oldptehi & LPTE_WIRED), ("Unmapped wired entry"));
598 * Need to invalidate old entry completely: see
599 * "Modifying a Page Table Entry". Need to reconstruct
600 * the virtual address for the outgoing entry to do that.
602 if (oldptehi & LPTE_BIG)
603 va = oldptehi >> moea64_large_page_shift;
605 va = oldptehi >> ADDR_PIDX_SHFT;
606 if (oldptehi & LPTE_HID)
607 va = (((k >> 3) ^ moea64_pteg_mask) ^ va) &
610 va = ((k >> 3) ^ va) & VSID_HASH_MASK;
611 va |= (oldptehi & LPTE_AVPN_MASK) <<
612 (ADDR_API_SHFT64 - ADDR_PIDX_SHFT);
616 moea64_pte_overflow++;
620 * Update the PTE as per "Adding a Page Table Entry". Lock is released
621 * by setting the high doubleworld.
623 pt->pte_lo = htobe64(pvo_pt->pte_lo);
625 pt->pte_hi = htobe64(pvo_pt->pte_hi);
628 /* Keep statistics */
635 moea64_pte_insert_native(mmu_t mmu, struct pvo_entry *pvo)
637 struct lpte insertpt;
641 moea64_pte_from_pvo(pvo, &insertpt);
643 /* Make sure further insertion is locked out during evictions */
644 rw_rlock(&moea64_eviction_lock);
647 * First try primary hash.
649 pvo->pvo_pte.slot &= ~7ULL; /* Base slot address */
650 slot = moea64_insert_to_pteg_native(&insertpt, pvo->pvo_pte.slot,
651 LPTE_VALID | LPTE_WIRED | LPTE_LOCKED);
653 rw_runlock(&moea64_eviction_lock);
654 pvo->pvo_pte.slot = slot;
659 * Now try secondary hash.
661 pvo->pvo_vaddr ^= PVO_HID;
662 insertpt.pte_hi ^= LPTE_HID;
663 pvo->pvo_pte.slot ^= (moea64_pteg_mask << 3);
664 slot = moea64_insert_to_pteg_native(&insertpt, pvo->pvo_pte.slot,
665 LPTE_VALID | LPTE_WIRED | LPTE_LOCKED);
667 rw_runlock(&moea64_eviction_lock);
668 pvo->pvo_pte.slot = slot;
673 * Out of luck. Find a PTE to sacrifice.
676 /* Lock out all insertions for a bit */
677 if (!rw_try_upgrade(&moea64_eviction_lock)) {
678 rw_runlock(&moea64_eviction_lock);
679 rw_wlock(&moea64_eviction_lock);
682 slot = moea64_insert_to_pteg_native(&insertpt, pvo->pvo_pte.slot,
683 LPTE_WIRED | LPTE_LOCKED);
685 rw_wunlock(&moea64_eviction_lock);
686 pvo->pvo_pte.slot = slot;
690 /* Try other hash table. Now we're getting desperate... */
691 pvo->pvo_vaddr ^= PVO_HID;
692 insertpt.pte_hi ^= LPTE_HID;
693 pvo->pvo_pte.slot ^= (moea64_pteg_mask << 3);
694 slot = moea64_insert_to_pteg_native(&insertpt, pvo->pvo_pte.slot,
695 LPTE_WIRED | LPTE_LOCKED);
697 rw_wunlock(&moea64_eviction_lock);
698 pvo->pvo_pte.slot = slot;
702 /* No freeable slots in either PTEG? We're hosed. */
703 rw_wunlock(&moea64_eviction_lock);
704 panic("moea64_pte_insert: overflow");