1 /*
2  * This program is free software; you can redistribute it and/or modify
3  * it under the terms of the GNU General Public License, version 2, as
4  * published by the Free Software Foundation.
5  *
6  * This program is distributed in the hope that it will be useful,
7  * but WITHOUT ANY WARRANTY; without even the implied warranty of
8  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
9  * GNU General Public License for more details.
10  *
11  * You should have received a copy of the GNU General Public License
12  * along with this program; if not, write to the Free Software
13  * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
14  *
15  * Copyright 2010 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
16  */
17 
18 #include <linux/types.h>
19 #include <linux/string.h>
20 #include <linux/kvm.h>
21 #include <linux/kvm_host.h>
22 #include <linux/highmem.h>
23 #include <linux/gfp.h>
24 #include <linux/slab.h>
25 #include <linux/hugetlb.h>
26 #include <linux/vmalloc.h>
27 #include <linux/srcu.h>
28 #include <linux/anon_inodes.h>
29 #include <linux/file.h>
30 #include <linux/debugfs.h>
31 
32 #include <asm/tlbflush.h>
33 #include <asm/kvm_ppc.h>
34 #include <asm/kvm_book3s.h>
35 #include <asm/book3s/64/mmu-hash.h>
36 #include <asm/hvcall.h>
37 #include <asm/synch.h>
38 #include <asm/ppc-opcode.h>
39 #include <asm/cputable.h>
40 
41 #include "trace_hv.h"
42 
43 //#define DEBUG_RESIZE_HPT	1
44 
45 #ifdef DEBUG_RESIZE_HPT
46 #define resize_hpt_debug(resize, ...)				\
47 	do {							\
48 		printk(KERN_DEBUG "RESIZE HPT %p: ", resize);	\
49 		printk(__VA_ARGS__);				\
50 	} while (0)
51 #else
52 #define resize_hpt_debug(resize, ...)				\
53 	do { } while (0)
54 #endif
55 
56 static long kvmppc_virtmode_do_h_enter(struct kvm *kvm, unsigned long flags,
57 				long pte_index, unsigned long pteh,
58 				unsigned long ptel, unsigned long *pte_idx_ret);
59 
60 struct kvm_resize_hpt {
61 	/* These fields read-only after init */
62 	struct kvm *kvm;
63 	struct work_struct work;
64 	u32 order;
65 
66 	/* These fields protected by kvm->lock */
67 	int error;
68 	bool prepare_done;
69 
70 	/* Private to the work thread, until prepare_done is true,
71 	 * then protected by kvm->resize_hpt_sem */
72 	struct kvm_hpt_info hpt;
73 };
74 
75 static void kvmppc_rmap_reset(struct kvm *kvm);
76 
77 int kvmppc_allocate_hpt(struct kvm_hpt_info *info, u32 order)
78 {
79 	unsigned long hpt = 0;
80 	int cma = 0;
81 	struct page *page = NULL;
82 	struct revmap_entry *rev;
83 	unsigned long npte;
84 
85 	if ((order < PPC_MIN_HPT_ORDER) || (order > PPC_MAX_HPT_ORDER))
86 		return -EINVAL;
87 
88 	page = kvm_alloc_hpt_cma(1ul << (order - PAGE_SHIFT));
89 	if (page) {
90 		hpt = (unsigned long)pfn_to_kaddr(page_to_pfn(page));
91 		memset((void *)hpt, 0, (1ul << order));
92 		cma = 1;
93 	}
94 
95 	if (!hpt)
96 		hpt = __get_free_pages(GFP_KERNEL|__GFP_ZERO|__GFP_RETRY_MAYFAIL
97 				       |__GFP_NOWARN, order - PAGE_SHIFT);
98 
99 	if (!hpt)
100 		return -ENOMEM;
101 
102 	/* HPTEs are 2**4 bytes long */
103 	npte = 1ul << (order - 4);
104 
105 	/* Allocate reverse map array */
106 	rev = vmalloc(sizeof(struct revmap_entry) * npte);
107 	if (!rev) {
108 		pr_err("kvmppc_allocate_hpt: Couldn't alloc reverse map array\n");
109 		if (cma)
110 			kvm_free_hpt_cma(page, 1 << (order - PAGE_SHIFT));
111 		else
112 			free_pages(hpt, order - PAGE_SHIFT);
113 		return -ENOMEM;
114 	}
115 
116 	info->order = order;
117 	info->virt = hpt;
118 	info->cma = cma;
119 	info->rev = rev;
120 
121 	return 0;
122 }
123 
124 void kvmppc_set_hpt(struct kvm *kvm, struct kvm_hpt_info *info)
125 {
126 	atomic64_set(&kvm->arch.mmio_update, 0);
127 	kvm->arch.hpt = *info;
128 	kvm->arch.sdr1 = __pa(info->virt) | (info->order - 18);
129 
130 	pr_debug("KVM guest htab at %lx (order %ld), LPID %x\n",
131 		 info->virt, (long)info->order, kvm->arch.lpid);
132 }
133 
134 long kvmppc_alloc_reset_hpt(struct kvm *kvm, int order)
135 {
136 	long err = -EBUSY;
137 	struct kvm_hpt_info info;
138 
139 	if (kvm_is_radix(kvm))
140 		return -EINVAL;
141 
142 	mutex_lock(&kvm->lock);
143 	if (kvm->arch.hpte_setup_done) {
144 		kvm->arch.hpte_setup_done = 0;
145 		/* order hpte_setup_done vs. vcpus_running */
146 		smp_mb();
147 		if (atomic_read(&kvm->arch.vcpus_running)) {
148 			kvm->arch.hpte_setup_done = 1;
149 			goto out;
150 		}
151 	}
152 	if (kvm->arch.hpt.order == order) {
153 		/* We already have a suitable HPT */
154 
155 		/* Set the entire HPT to 0, i.e. invalid HPTEs */
156 		memset((void *)kvm->arch.hpt.virt, 0, 1ul << order);
157 		/*
158 		 * Reset all the reverse-mapping chains for all memslots
159 		 */
160 		kvmppc_rmap_reset(kvm);
161 		/* Ensure that each vcpu will flush its TLB on next entry. */
162 		cpumask_setall(&kvm->arch.need_tlb_flush);
163 		err = 0;
164 		goto out;
165 	}
166 
167 	if (kvm->arch.hpt.virt) {
168 		kvmppc_free_hpt(&kvm->arch.hpt);
169 		kvmppc_rmap_reset(kvm);
170 	}
171 
172 	err = kvmppc_allocate_hpt(&info, order);
173 	if (err < 0)
174 		goto out;
175 	kvmppc_set_hpt(kvm, &info);
176 
177 out:
178 	mutex_unlock(&kvm->lock);
179 	return err;
180 }
181 
182 void kvmppc_free_hpt(struct kvm_hpt_info *info)
183 {
184 	vfree(info->rev);
185 	if (info->cma)
186 		kvm_free_hpt_cma(virt_to_page(info->virt),
187 				 1 << (info->order - PAGE_SHIFT));
188 	else if (info->virt)
189 		free_pages(info->virt, info->order - PAGE_SHIFT);
190 	info->virt = 0;
191 	info->order = 0;
192 }
193 
194 /* Bits in first HPTE dword for pagesize 4k, 64k or 16M */
195 static inline unsigned long hpte0_pgsize_encoding(unsigned long pgsize)
196 {
197 	return (pgsize > 0x1000) ? HPTE_V_LARGE : 0;
198 }
199 
200 /* Bits in second HPTE dword for pagesize 4k, 64k or 16M */
201 static inline unsigned long hpte1_pgsize_encoding(unsigned long pgsize)
202 {
203 	return (pgsize == 0x10000) ? 0x1000 : 0;
204 }
205 
206 void kvmppc_map_vrma(struct kvm_vcpu *vcpu, struct kvm_memory_slot *memslot,
207 		     unsigned long porder)
208 {
209 	unsigned long i;
210 	unsigned long npages;
211 	unsigned long hp_v, hp_r;
212 	unsigned long addr, hash;
213 	unsigned long psize;
214 	unsigned long hp0, hp1;
215 	unsigned long idx_ret;
216 	long ret;
217 	struct kvm *kvm = vcpu->kvm;
218 
219 	psize = 1ul << porder;
220 	npages = memslot->npages >> (porder - PAGE_SHIFT);
221 
222 	/* VRMA can't be > 1TB */
223 	if (npages > 1ul << (40 - porder))
224 		npages = 1ul << (40 - porder);
225 	/* Can't use more than 1 HPTE per HPTEG */
226 	if (npages > kvmppc_hpt_mask(&kvm->arch.hpt) + 1)
227 		npages = kvmppc_hpt_mask(&kvm->arch.hpt) + 1;
228 
229 	hp0 = HPTE_V_1TB_SEG | (VRMA_VSID << (40 - 16)) |
230 		HPTE_V_BOLTED | hpte0_pgsize_encoding(psize);
231 	hp1 = hpte1_pgsize_encoding(psize) |
232 		HPTE_R_R | HPTE_R_C | HPTE_R_M | PP_RWXX;
233 
234 	for (i = 0; i < npages; ++i) {
235 		addr = i << porder;
236 		/* can't use hpt_hash since va > 64 bits */
237 		hash = (i ^ (VRMA_VSID ^ (VRMA_VSID << 25)))
238 			& kvmppc_hpt_mask(&kvm->arch.hpt);
239 		/*
240 		 * We assume that the hash table is empty and no
241 		 * vcpus are using it at this stage.  Since we create
242 		 * at most one HPTE per HPTEG, we just assume entry 7
243 		 * is available and use it.
244 		 */
245 		hash = (hash << 3) + 7;
246 		hp_v = hp0 | ((addr >> 16) & ~0x7fUL);
247 		hp_r = hp1 | addr;
248 		ret = kvmppc_virtmode_do_h_enter(kvm, H_EXACT, hash, hp_v, hp_r,
249 						 &idx_ret);
250 		if (ret != H_SUCCESS) {
251 			pr_err("KVM: map_vrma at %lx failed, ret=%ld\n",
252 			       addr, ret);
253 			break;
254 		}
255 	}
256 }
257 
258 int kvmppc_mmu_hv_init(void)
259 {
260 	unsigned long host_lpid, rsvd_lpid;
261 
262 	if (!cpu_has_feature(CPU_FTR_HVMODE))
263 		return -EINVAL;
264 
265 	/* POWER7 has 10-bit LPIDs (12-bit in POWER8) */
266 	host_lpid = mfspr(SPRN_LPID);
267 	rsvd_lpid = LPID_RSVD;
268 
269 	kvmppc_init_lpid(rsvd_lpid + 1);
270 
271 	kvmppc_claim_lpid(host_lpid);
272 	/* rsvd_lpid is reserved for use in partition switching */
273 	kvmppc_claim_lpid(rsvd_lpid);
274 
275 	return 0;
276 }
277 
278 static void kvmppc_mmu_book3s_64_hv_reset_msr(struct kvm_vcpu *vcpu)
279 {
280 	unsigned long msr = vcpu->arch.intr_msr;
281 
282 	/* If transactional, change to suspend mode on IRQ delivery */
283 	if (MSR_TM_TRANSACTIONAL(vcpu->arch.shregs.msr))
284 		msr |= MSR_TS_S;
285 	else
286 		msr |= vcpu->arch.shregs.msr & MSR_TS_MASK;
287 	kvmppc_set_msr(vcpu, msr);
288 }
289 
290 static long kvmppc_virtmode_do_h_enter(struct kvm *kvm, unsigned long flags,
291 				long pte_index, unsigned long pteh,
292 				unsigned long ptel, unsigned long *pte_idx_ret)
293 {
294 	long ret;
295 
296 	/* Protect linux PTE lookup from page table destruction */
297 	rcu_read_lock_sched();	/* this disables preemption too */
298 	ret = kvmppc_do_h_enter(kvm, flags, pte_index, pteh, ptel,
299 				current->mm->pgd, false, pte_idx_ret);
300 	rcu_read_unlock_sched();
301 	if (ret == H_TOO_HARD) {
302 		/* this can't happen */
303 		pr_err("KVM: Oops, kvmppc_h_enter returned too hard!\n");
304 		ret = H_RESOURCE;	/* or something */
305 	}
306 	return ret;
307 
308 }
309 
310 static struct kvmppc_slb *kvmppc_mmu_book3s_hv_find_slbe(struct kvm_vcpu *vcpu,
311 							 gva_t eaddr)
312 {
313 	u64 mask;
314 	int i;
315 
316 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
317 		if (!(vcpu->arch.slb[i].orige & SLB_ESID_V))
318 			continue;
319 
320 		if (vcpu->arch.slb[i].origv & SLB_VSID_B_1T)
321 			mask = ESID_MASK_1T;
322 		else
323 			mask = ESID_MASK;
324 
325 		if (((vcpu->arch.slb[i].orige ^ eaddr) & mask) == 0)
326 			return &vcpu->arch.slb[i];
327 	}
328 	return NULL;
329 }
330 
331 static unsigned long kvmppc_mmu_get_real_addr(unsigned long v, unsigned long r,
332 			unsigned long ea)
333 {
334 	unsigned long ra_mask;
335 
336 	ra_mask = hpte_page_size(v, r) - 1;
337 	return (r & HPTE_R_RPN & ~ra_mask) | (ea & ra_mask);
338 }
339 
340 static int kvmppc_mmu_book3s_64_hv_xlate(struct kvm_vcpu *vcpu, gva_t eaddr,
341 			struct kvmppc_pte *gpte, bool data, bool iswrite)
342 {
343 	struct kvm *kvm = vcpu->kvm;
344 	struct kvmppc_slb *slbe;
345 	unsigned long slb_v;
346 	unsigned long pp, key;
347 	unsigned long v, orig_v, gr;
348 	__be64 *hptep;
349 	int index;
350 	int virtmode = vcpu->arch.shregs.msr & (data ? MSR_DR : MSR_IR);
351 
352 	/* Get SLB entry */
353 	if (virtmode) {
354 		slbe = kvmppc_mmu_book3s_hv_find_slbe(vcpu, eaddr);
355 		if (!slbe)
356 			return -EINVAL;
357 		slb_v = slbe->origv;
358 	} else {
359 		/* real mode access */
360 		slb_v = vcpu->kvm->arch.vrma_slb_v;
361 	}
362 
363 	preempt_disable();
364 	/* Find the HPTE in the hash table */
365 	index = kvmppc_hv_find_lock_hpte(kvm, eaddr, slb_v,
366 					 HPTE_V_VALID | HPTE_V_ABSENT);
367 	if (index < 0) {
368 		preempt_enable();
369 		return -ENOENT;
370 	}
371 	hptep = (__be64 *)(kvm->arch.hpt.virt + (index << 4));
372 	v = orig_v = be64_to_cpu(hptep[0]) & ~HPTE_V_HVLOCK;
373 	if (cpu_has_feature(CPU_FTR_ARCH_300))
374 		v = hpte_new_to_old_v(v, be64_to_cpu(hptep[1]));
375 	gr = kvm->arch.hpt.rev[index].guest_rpte;
376 
377 	unlock_hpte(hptep, orig_v);
378 	preempt_enable();
379 
380 	gpte->eaddr = eaddr;
381 	gpte->vpage = ((v & HPTE_V_AVPN) << 4) | ((eaddr >> 12) & 0xfff);
382 
383 	/* Get PP bits and key for permission check */
384 	pp = gr & (HPTE_R_PP0 | HPTE_R_PP);
385 	key = (vcpu->arch.shregs.msr & MSR_PR) ? SLB_VSID_KP : SLB_VSID_KS;
386 	key &= slb_v;
387 
388 	/* Calculate permissions */
389 	gpte->may_read = hpte_read_permission(pp, key);
390 	gpte->may_write = hpte_write_permission(pp, key);
391 	gpte->may_execute = gpte->may_read && !(gr & (HPTE_R_N | HPTE_R_G));
392 
393 	/* Storage key permission check for POWER7 */
394 	if (data && virtmode) {
395 		int amrfield = hpte_get_skey_perm(gr, vcpu->arch.amr);
396 		if (amrfield & 1)
397 			gpte->may_read = 0;
398 		if (amrfield & 2)
399 			gpte->may_write = 0;
400 	}
401 
402 	/* Get the guest physical address */
403 	gpte->raddr = kvmppc_mmu_get_real_addr(v, gr, eaddr);
404 	return 0;
405 }
406 
407 /*
408  * Quick test for whether an instruction is a load or a store.
409  * If the instruction is a load or a store, then this will indicate
410  * which it is, at least on server processors.  (Embedded processors
411  * have some external PID instructions that don't follow the rule
412  * embodied here.)  If the instruction isn't a load or store, then
413  * this doesn't return anything useful.
414  */
415 static int instruction_is_store(unsigned int instr)
416 {
417 	unsigned int mask;
418 
419 	mask = 0x10000000;
420 	if ((instr & 0xfc000000) == 0x7c000000)
421 		mask = 0x100;		/* major opcode 31 */
422 	return (instr & mask) != 0;
423 }
424 
425 int kvmppc_hv_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu,
426 			   unsigned long gpa, gva_t ea, int is_store)
427 {
428 	u32 last_inst;
429 
430 	/*
431 	 * If we fail, we just return to the guest and try executing it again.
432 	 */
433 	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
434 		EMULATE_DONE)
435 		return RESUME_GUEST;
436 
437 	/*
438 	 * WARNING: We do not know for sure whether the instruction we just
439 	 * read from memory is the same that caused the fault in the first
440 	 * place.  If the instruction we read is neither an load or a store,
441 	 * then it can't access memory, so we don't need to worry about
442 	 * enforcing access permissions.  So, assuming it is a load or
443 	 * store, we just check that its direction (load or store) is
444 	 * consistent with the original fault, since that's what we
445 	 * checked the access permissions against.  If there is a mismatch
446 	 * we just return and retry the instruction.
447 	 */
448 
449 	if (instruction_is_store(last_inst) != !!is_store)
450 		return RESUME_GUEST;
451 
452 	/*
453 	 * Emulated accesses are emulated by looking at the hash for
454 	 * translation once, then performing the access later. The
455 	 * translation could be invalidated in the meantime in which
456 	 * point performing the subsequent memory access on the old
457 	 * physical address could possibly be a security hole for the
458 	 * guest (but not the host).
459 	 *
460 	 * This is less of an issue for MMIO stores since they aren't
461 	 * globally visible. It could be an issue for MMIO loads to
462 	 * a certain extent but we'll ignore it for now.
463 	 */
464 
465 	vcpu->arch.paddr_accessed = gpa;
466 	vcpu->arch.vaddr_accessed = ea;
467 	return kvmppc_emulate_mmio(run, vcpu);
468 }
469 
470 int kvmppc_book3s_hv_page_fault(struct kvm_run *run, struct kvm_vcpu *vcpu,
471 				unsigned long ea, unsigned long dsisr)
472 {
473 	struct kvm *kvm = vcpu->kvm;
474 	unsigned long hpte[3], r;
475 	unsigned long hnow_v, hnow_r;
476 	__be64 *hptep;
477 	unsigned long mmu_seq, psize, pte_size;
478 	unsigned long gpa_base, gfn_base;
479 	unsigned long gpa, gfn, hva, pfn;
480 	struct kvm_memory_slot *memslot;
481 	unsigned long *rmap;
482 	struct revmap_entry *rev;
483 	struct page *page, *pages[1];
484 	long index, ret, npages;
485 	bool is_ci;
486 	unsigned int writing, write_ok;
487 	struct vm_area_struct *vma;
488 	unsigned long rcbits;
489 	long mmio_update;
490 
491 	if (kvm_is_radix(kvm))
492 		return kvmppc_book3s_radix_page_fault(run, vcpu, ea, dsisr);
493 
494 	/*
495 	 * Real-mode code has already searched the HPT and found the
496 	 * entry we're interested in.  Lock the entry and check that
497 	 * it hasn't changed.  If it has, just return and re-execute the
498 	 * instruction.
499 	 */
500 	if (ea != vcpu->arch.pgfault_addr)
501 		return RESUME_GUEST;
502 
503 	if (vcpu->arch.pgfault_cache) {
504 		mmio_update = atomic64_read(&kvm->arch.mmio_update);
505 		if (mmio_update == vcpu->arch.pgfault_cache->mmio_update) {
506 			r = vcpu->arch.pgfault_cache->rpte;
507 			psize = hpte_page_size(vcpu->arch.pgfault_hpte[0], r);
508 			gpa_base = r & HPTE_R_RPN & ~(psize - 1);
509 			gfn_base = gpa_base >> PAGE_SHIFT;
510 			gpa = gpa_base | (ea & (psize - 1));
511 			return kvmppc_hv_emulate_mmio(run, vcpu, gpa, ea,
512 						dsisr & DSISR_ISSTORE);
513 		}
514 	}
515 	index = vcpu->arch.pgfault_index;
516 	hptep = (__be64 *)(kvm->arch.hpt.virt + (index << 4));
517 	rev = &kvm->arch.hpt.rev[index];
518 	preempt_disable();
519 	while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
520 		cpu_relax();
521 	hpte[0] = be64_to_cpu(hptep[0]) & ~HPTE_V_HVLOCK;
522 	hpte[1] = be64_to_cpu(hptep[1]);
523 	hpte[2] = r = rev->guest_rpte;
524 	unlock_hpte(hptep, hpte[0]);
525 	preempt_enable();
526 
527 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
528 		hpte[0] = hpte_new_to_old_v(hpte[0], hpte[1]);
529 		hpte[1] = hpte_new_to_old_r(hpte[1]);
530 	}
531 	if (hpte[0] != vcpu->arch.pgfault_hpte[0] ||
532 	    hpte[1] != vcpu->arch.pgfault_hpte[1])
533 		return RESUME_GUEST;
534 
535 	/* Translate the logical address and get the page */
536 	psize = hpte_page_size(hpte[0], r);
537 	gpa_base = r & HPTE_R_RPN & ~(psize - 1);
538 	gfn_base = gpa_base >> PAGE_SHIFT;
539 	gpa = gpa_base | (ea & (psize - 1));
540 	gfn = gpa >> PAGE_SHIFT;
541 	memslot = gfn_to_memslot(kvm, gfn);
542 
543 	trace_kvm_page_fault_enter(vcpu, hpte, memslot, ea, dsisr);
544 
545 	/* No memslot means it's an emulated MMIO region */
546 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
547 		return kvmppc_hv_emulate_mmio(run, vcpu, gpa, ea,
548 					      dsisr & DSISR_ISSTORE);
549 
550 	/*
551 	 * This should never happen, because of the slot_is_aligned()
552 	 * check in kvmppc_do_h_enter().
553 	 */
554 	if (gfn_base < memslot->base_gfn)
555 		return -EFAULT;
556 
557 	/* used to check for invalidations in progress */
558 	mmu_seq = kvm->mmu_notifier_seq;
559 	smp_rmb();
560 
561 	ret = -EFAULT;
562 	is_ci = false;
563 	pfn = 0;
564 	page = NULL;
565 	pte_size = PAGE_SIZE;
566 	writing = (dsisr & DSISR_ISSTORE) != 0;
567 	/* If writing != 0, then the HPTE must allow writing, if we get here */
568 	write_ok = writing;
569 	hva = gfn_to_hva_memslot(memslot, gfn);
570 	npages = get_user_pages_fast(hva, 1, writing, pages);
571 	if (npages < 1) {
572 		/* Check if it's an I/O mapping */
573 		down_read(&current->mm->mmap_sem);
574 		vma = find_vma(current->mm, hva);
575 		if (vma && vma->vm_start <= hva && hva + psize <= vma->vm_end &&
576 		    (vma->vm_flags & VM_PFNMAP)) {
577 			pfn = vma->vm_pgoff +
578 				((hva - vma->vm_start) >> PAGE_SHIFT);
579 			pte_size = psize;
580 			is_ci = pte_ci(__pte((pgprot_val(vma->vm_page_prot))));
581 			write_ok = vma->vm_flags & VM_WRITE;
582 		}
583 		up_read(&current->mm->mmap_sem);
584 		if (!pfn)
585 			goto out_put;
586 	} else {
587 		page = pages[0];
588 		pfn = page_to_pfn(page);
589 		if (PageHuge(page)) {
590 			page = compound_head(page);
591 			pte_size <<= compound_order(page);
592 		}
593 		/* if the guest wants write access, see if that is OK */
594 		if (!writing && hpte_is_writable(r)) {
595 			pte_t *ptep, pte;
596 			unsigned long flags;
597 			/*
598 			 * We need to protect against page table destruction
599 			 * hugepage split and collapse.
600 			 */
601 			local_irq_save(flags);
602 			ptep = find_linux_pte_or_hugepte(current->mm->pgd,
603 							 hva, NULL, NULL);
604 			if (ptep) {
605 				pte = kvmppc_read_update_linux_pte(ptep, 1);
606 				if (__pte_write(pte))
607 					write_ok = 1;
608 			}
609 			local_irq_restore(flags);
610 		}
611 	}
612 
613 	if (psize > pte_size)
614 		goto out_put;
615 
616 	/* Check WIMG vs. the actual page we're accessing */
617 	if (!hpte_cache_flags_ok(r, is_ci)) {
618 		if (is_ci)
619 			goto out_put;
620 		/*
621 		 * Allow guest to map emulated device memory as
622 		 * uncacheable, but actually make it cacheable.
623 		 */
624 		r = (r & ~(HPTE_R_W|HPTE_R_I|HPTE_R_G)) | HPTE_R_M;
625 	}
626 
627 	/*
628 	 * Set the HPTE to point to pfn.
629 	 * Since the pfn is at PAGE_SIZE granularity, make sure we
630 	 * don't mask out lower-order bits if psize < PAGE_SIZE.
631 	 */
632 	if (psize < PAGE_SIZE)
633 		psize = PAGE_SIZE;
634 	r = (r & HPTE_R_KEY_HI) | (r & ~(HPTE_R_PP0 - psize)) |
635 					((pfn << PAGE_SHIFT) & ~(psize - 1));
636 	if (hpte_is_writable(r) && !write_ok)
637 		r = hpte_make_readonly(r);
638 	ret = RESUME_GUEST;
639 	preempt_disable();
640 	while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
641 		cpu_relax();
642 	hnow_v = be64_to_cpu(hptep[0]);
643 	hnow_r = be64_to_cpu(hptep[1]);
644 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
645 		hnow_v = hpte_new_to_old_v(hnow_v, hnow_r);
646 		hnow_r = hpte_new_to_old_r(hnow_r);
647 	}
648 	if ((hnow_v & ~HPTE_V_HVLOCK) != hpte[0] || hnow_r != hpte[1] ||
649 	    rev->guest_rpte != hpte[2])
650 		/* HPTE has been changed under us; let the guest retry */
651 		goto out_unlock;
652 	hpte[0] = (hpte[0] & ~HPTE_V_ABSENT) | HPTE_V_VALID;
653 
654 	/* Always put the HPTE in the rmap chain for the page base address */
655 	rmap = &memslot->arch.rmap[gfn_base - memslot->base_gfn];
656 	lock_rmap(rmap);
657 
658 	/* Check if we might have been invalidated; let the guest retry if so */
659 	ret = RESUME_GUEST;
660 	if (mmu_notifier_retry(vcpu->kvm, mmu_seq)) {
661 		unlock_rmap(rmap);
662 		goto out_unlock;
663 	}
664 
665 	/* Only set R/C in real HPTE if set in both *rmap and guest_rpte */
666 	rcbits = *rmap >> KVMPPC_RMAP_RC_SHIFT;
667 	r &= rcbits | ~(HPTE_R_R | HPTE_R_C);
668 
669 	if (be64_to_cpu(hptep[0]) & HPTE_V_VALID) {
670 		/* HPTE was previously valid, so we need to invalidate it */
671 		unlock_rmap(rmap);
672 		hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
673 		kvmppc_invalidate_hpte(kvm, hptep, index);
674 		/* don't lose previous R and C bits */
675 		r |= be64_to_cpu(hptep[1]) & (HPTE_R_R | HPTE_R_C);
676 	} else {
677 		kvmppc_add_revmap_chain(kvm, rev, rmap, index, 0);
678 	}
679 
680 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
681 		r = hpte_old_to_new_r(hpte[0], r);
682 		hpte[0] = hpte_old_to_new_v(hpte[0]);
683 	}
684 	hptep[1] = cpu_to_be64(r);
685 	eieio();
686 	__unlock_hpte(hptep, hpte[0]);
687 	asm volatile("ptesync" : : : "memory");
688 	preempt_enable();
689 	if (page && hpte_is_writable(r))
690 		SetPageDirty(page);
691 
692  out_put:
693 	trace_kvm_page_fault_exit(vcpu, hpte, ret);
694 
695 	if (page) {
696 		/*
697 		 * We drop pages[0] here, not page because page might
698 		 * have been set to the head page of a compound, but
699 		 * we have to drop the reference on the correct tail
700 		 * page to match the get inside gup()
701 		 */
702 		put_page(pages[0]);
703 	}
704 	return ret;
705 
706  out_unlock:
707 	__unlock_hpte(hptep, be64_to_cpu(hptep[0]));
708 	preempt_enable();
709 	goto out_put;
710 }
711 
712 static void kvmppc_rmap_reset(struct kvm *kvm)
713 {
714 	struct kvm_memslots *slots;
715 	struct kvm_memory_slot *memslot;
716 	int srcu_idx;
717 
718 	srcu_idx = srcu_read_lock(&kvm->srcu);
719 	slots = kvm_memslots(kvm);
720 	kvm_for_each_memslot(memslot, slots) {
721 		/*
722 		 * This assumes it is acceptable to lose reference and
723 		 * change bits across a reset.
724 		 */
725 		memset(memslot->arch.rmap, 0,
726 		       memslot->npages * sizeof(*memslot->arch.rmap));
727 	}
728 	srcu_read_unlock(&kvm->srcu, srcu_idx);
729 }
730 
731 typedef int (*hva_handler_fn)(struct kvm *kvm, struct kvm_memory_slot *memslot,
732 			      unsigned long gfn);
733 
734 static int kvm_handle_hva_range(struct kvm *kvm,
735 				unsigned long start,
736 				unsigned long end,
737 				hva_handler_fn handler)
738 {
739 	int ret;
740 	int retval = 0;
741 	struct kvm_memslots *slots;
742 	struct kvm_memory_slot *memslot;
743 
744 	slots = kvm_memslots(kvm);
745 	kvm_for_each_memslot(memslot, slots) {
746 		unsigned long hva_start, hva_end;
747 		gfn_t gfn, gfn_end;
748 
749 		hva_start = max(start, memslot->userspace_addr);
750 		hva_end = min(end, memslot->userspace_addr +
751 					(memslot->npages << PAGE_SHIFT));
752 		if (hva_start >= hva_end)
753 			continue;
754 		/*
755 		 * {gfn(page) | page intersects with [hva_start, hva_end)} =
756 		 * {gfn, gfn+1, ..., gfn_end-1}.
757 		 */
758 		gfn = hva_to_gfn_memslot(hva_start, memslot);
759 		gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot);
760 
761 		for (; gfn < gfn_end; ++gfn) {
762 			ret = handler(kvm, memslot, gfn);
763 			retval |= ret;
764 		}
765 	}
766 
767 	return retval;
768 }
769 
770 static int kvm_handle_hva(struct kvm *kvm, unsigned long hva,
771 			  hva_handler_fn handler)
772 {
773 	return kvm_handle_hva_range(kvm, hva, hva + 1, handler);
774 }
775 
776 /* Must be called with both HPTE and rmap locked */
777 static void kvmppc_unmap_hpte(struct kvm *kvm, unsigned long i,
778 			      unsigned long *rmapp, unsigned long gfn)
779 {
780 	__be64 *hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
781 	struct revmap_entry *rev = kvm->arch.hpt.rev;
782 	unsigned long j, h;
783 	unsigned long ptel, psize, rcbits;
784 
785 	j = rev[i].forw;
786 	if (j == i) {
787 		/* chain is now empty */
788 		*rmapp &= ~(KVMPPC_RMAP_PRESENT | KVMPPC_RMAP_INDEX);
789 	} else {
790 		/* remove i from chain */
791 		h = rev[i].back;
792 		rev[h].forw = j;
793 		rev[j].back = h;
794 		rev[i].forw = rev[i].back = i;
795 		*rmapp = (*rmapp & ~KVMPPC_RMAP_INDEX) | j;
796 	}
797 
798 	/* Now check and modify the HPTE */
799 	ptel = rev[i].guest_rpte;
800 	psize = hpte_page_size(be64_to_cpu(hptep[0]), ptel);
801 	if ((be64_to_cpu(hptep[0]) & HPTE_V_VALID) &&
802 	    hpte_rpn(ptel, psize) == gfn) {
803 		hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
804 		kvmppc_invalidate_hpte(kvm, hptep, i);
805 		hptep[1] &= ~cpu_to_be64(HPTE_R_KEY_HI | HPTE_R_KEY_LO);
806 		/* Harvest R and C */
807 		rcbits = be64_to_cpu(hptep[1]) & (HPTE_R_R | HPTE_R_C);
808 		*rmapp |= rcbits << KVMPPC_RMAP_RC_SHIFT;
809 		if (rcbits & HPTE_R_C)
810 			kvmppc_update_rmap_change(rmapp, psize);
811 		if (rcbits & ~rev[i].guest_rpte) {
812 			rev[i].guest_rpte = ptel | rcbits;
813 			note_hpte_modification(kvm, &rev[i]);
814 		}
815 	}
816 }
817 
818 static int kvm_unmap_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot,
819 			   unsigned long gfn)
820 {
821 	unsigned long i;
822 	__be64 *hptep;
823 	unsigned long *rmapp;
824 
825 	rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
826 	for (;;) {
827 		lock_rmap(rmapp);
828 		if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
829 			unlock_rmap(rmapp);
830 			break;
831 		}
832 
833 		/*
834 		 * To avoid an ABBA deadlock with the HPTE lock bit,
835 		 * we can't spin on the HPTE lock while holding the
836 		 * rmap chain lock.
837 		 */
838 		i = *rmapp & KVMPPC_RMAP_INDEX;
839 		hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
840 		if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
841 			/* unlock rmap before spinning on the HPTE lock */
842 			unlock_rmap(rmapp);
843 			while (be64_to_cpu(hptep[0]) & HPTE_V_HVLOCK)
844 				cpu_relax();
845 			continue;
846 		}
847 
848 		kvmppc_unmap_hpte(kvm, i, rmapp, gfn);
849 		unlock_rmap(rmapp);
850 		__unlock_hpte(hptep, be64_to_cpu(hptep[0]));
851 	}
852 	return 0;
853 }
854 
855 int kvm_unmap_hva_hv(struct kvm *kvm, unsigned long hva)
856 {
857 	hva_handler_fn handler;
858 
859 	handler = kvm_is_radix(kvm) ? kvm_unmap_radix : kvm_unmap_rmapp;
860 	kvm_handle_hva(kvm, hva, handler);
861 	return 0;
862 }
863 
864 int kvm_unmap_hva_range_hv(struct kvm *kvm, unsigned long start, unsigned long end)
865 {
866 	hva_handler_fn handler;
867 
868 	handler = kvm_is_radix(kvm) ? kvm_unmap_radix : kvm_unmap_rmapp;
869 	kvm_handle_hva_range(kvm, start, end, handler);
870 	return 0;
871 }
872 
873 void kvmppc_core_flush_memslot_hv(struct kvm *kvm,
874 				  struct kvm_memory_slot *memslot)
875 {
876 	unsigned long gfn;
877 	unsigned long n;
878 	unsigned long *rmapp;
879 
880 	gfn = memslot->base_gfn;
881 	rmapp = memslot->arch.rmap;
882 	for (n = memslot->npages; n; --n, ++gfn) {
883 		if (kvm_is_radix(kvm)) {
884 			kvm_unmap_radix(kvm, memslot, gfn);
885 			continue;
886 		}
887 		/*
888 		 * Testing the present bit without locking is OK because
889 		 * the memslot has been marked invalid already, and hence
890 		 * no new HPTEs referencing this page can be created,
891 		 * thus the present bit can't go from 0 to 1.
892 		 */
893 		if (*rmapp & KVMPPC_RMAP_PRESENT)
894 			kvm_unmap_rmapp(kvm, memslot, gfn);
895 		++rmapp;
896 	}
897 }
898 
899 static int kvm_age_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot,
900 			 unsigned long gfn)
901 {
902 	struct revmap_entry *rev = kvm->arch.hpt.rev;
903 	unsigned long head, i, j;
904 	__be64 *hptep;
905 	int ret = 0;
906 	unsigned long *rmapp;
907 
908 	rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
909  retry:
910 	lock_rmap(rmapp);
911 	if (*rmapp & KVMPPC_RMAP_REFERENCED) {
912 		*rmapp &= ~KVMPPC_RMAP_REFERENCED;
913 		ret = 1;
914 	}
915 	if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
916 		unlock_rmap(rmapp);
917 		return ret;
918 	}
919 
920 	i = head = *rmapp & KVMPPC_RMAP_INDEX;
921 	do {
922 		hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
923 		j = rev[i].forw;
924 
925 		/* If this HPTE isn't referenced, ignore it */
926 		if (!(be64_to_cpu(hptep[1]) & HPTE_R_R))
927 			continue;
928 
929 		if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
930 			/* unlock rmap before spinning on the HPTE lock */
931 			unlock_rmap(rmapp);
932 			while (be64_to_cpu(hptep[0]) & HPTE_V_HVLOCK)
933 				cpu_relax();
934 			goto retry;
935 		}
936 
937 		/* Now check and modify the HPTE */
938 		if ((be64_to_cpu(hptep[0]) & HPTE_V_VALID) &&
939 		    (be64_to_cpu(hptep[1]) & HPTE_R_R)) {
940 			kvmppc_clear_ref_hpte(kvm, hptep, i);
941 			if (!(rev[i].guest_rpte & HPTE_R_R)) {
942 				rev[i].guest_rpte |= HPTE_R_R;
943 				note_hpte_modification(kvm, &rev[i]);
944 			}
945 			ret = 1;
946 		}
947 		__unlock_hpte(hptep, be64_to_cpu(hptep[0]));
948 	} while ((i = j) != head);
949 
950 	unlock_rmap(rmapp);
951 	return ret;
952 }
953 
954 int kvm_age_hva_hv(struct kvm *kvm, unsigned long start, unsigned long end)
955 {
956 	hva_handler_fn handler;
957 
958 	handler = kvm_is_radix(kvm) ? kvm_age_radix : kvm_age_rmapp;
959 	return kvm_handle_hva_range(kvm, start, end, handler);
960 }
961 
962 static int kvm_test_age_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot,
963 			      unsigned long gfn)
964 {
965 	struct revmap_entry *rev = kvm->arch.hpt.rev;
966 	unsigned long head, i, j;
967 	unsigned long *hp;
968 	int ret = 1;
969 	unsigned long *rmapp;
970 
971 	rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
972 	if (*rmapp & KVMPPC_RMAP_REFERENCED)
973 		return 1;
974 
975 	lock_rmap(rmapp);
976 	if (*rmapp & KVMPPC_RMAP_REFERENCED)
977 		goto out;
978 
979 	if (*rmapp & KVMPPC_RMAP_PRESENT) {
980 		i = head = *rmapp & KVMPPC_RMAP_INDEX;
981 		do {
982 			hp = (unsigned long *)(kvm->arch.hpt.virt + (i << 4));
983 			j = rev[i].forw;
984 			if (be64_to_cpu(hp[1]) & HPTE_R_R)
985 				goto out;
986 		} while ((i = j) != head);
987 	}
988 	ret = 0;
989 
990  out:
991 	unlock_rmap(rmapp);
992 	return ret;
993 }
994 
995 int kvm_test_age_hva_hv(struct kvm *kvm, unsigned long hva)
996 {
997 	hva_handler_fn handler;
998 
999 	handler = kvm_is_radix(kvm) ? kvm_test_age_radix : kvm_test_age_rmapp;
1000 	return kvm_handle_hva(kvm, hva, handler);
1001 }
1002 
1003 void kvm_set_spte_hva_hv(struct kvm *kvm, unsigned long hva, pte_t pte)
1004 {
1005 	hva_handler_fn handler;
1006 
1007 	handler = kvm_is_radix(kvm) ? kvm_unmap_radix : kvm_unmap_rmapp;
1008 	kvm_handle_hva(kvm, hva, handler);
1009 }
1010 
1011 static int vcpus_running(struct kvm *kvm)
1012 {
1013 	return atomic_read(&kvm->arch.vcpus_running) != 0;
1014 }
1015 
1016 /*
1017  * Returns the number of system pages that are dirty.
1018  * This can be more than 1 if we find a huge-page HPTE.
1019  */
1020 static int kvm_test_clear_dirty_npages(struct kvm *kvm, unsigned long *rmapp)
1021 {
1022 	struct revmap_entry *rev = kvm->arch.hpt.rev;
1023 	unsigned long head, i, j;
1024 	unsigned long n;
1025 	unsigned long v, r;
1026 	__be64 *hptep;
1027 	int npages_dirty = 0;
1028 
1029  retry:
1030 	lock_rmap(rmapp);
1031 	if (*rmapp & KVMPPC_RMAP_CHANGED) {
1032 		long change_order = (*rmapp & KVMPPC_RMAP_CHG_ORDER)
1033 			>> KVMPPC_RMAP_CHG_SHIFT;
1034 		*rmapp &= ~(KVMPPC_RMAP_CHANGED | KVMPPC_RMAP_CHG_ORDER);
1035 		npages_dirty = 1;
1036 		if (change_order > PAGE_SHIFT)
1037 			npages_dirty = 1ul << (change_order - PAGE_SHIFT);
1038 	}
1039 	if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
1040 		unlock_rmap(rmapp);
1041 		return npages_dirty;
1042 	}
1043 
1044 	i = head = *rmapp & KVMPPC_RMAP_INDEX;
1045 	do {
1046 		unsigned long hptep1;
1047 		hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
1048 		j = rev[i].forw;
1049 
1050 		/*
1051 		 * Checking the C (changed) bit here is racy since there
1052 		 * is no guarantee about when the hardware writes it back.
1053 		 * If the HPTE is not writable then it is stable since the
1054 		 * page can't be written to, and we would have done a tlbie
1055 		 * (which forces the hardware to complete any writeback)
1056 		 * when making the HPTE read-only.
1057 		 * If vcpus are running then this call is racy anyway
1058 		 * since the page could get dirtied subsequently, so we
1059 		 * expect there to be a further call which would pick up
1060 		 * any delayed C bit writeback.
1061 		 * Otherwise we need to do the tlbie even if C==0 in
1062 		 * order to pick up any delayed writeback of C.
1063 		 */
1064 		hptep1 = be64_to_cpu(hptep[1]);
1065 		if (!(hptep1 & HPTE_R_C) &&
1066 		    (!hpte_is_writable(hptep1) || vcpus_running(kvm)))
1067 			continue;
1068 
1069 		if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
1070 			/* unlock rmap before spinning on the HPTE lock */
1071 			unlock_rmap(rmapp);
1072 			while (hptep[0] & cpu_to_be64(HPTE_V_HVLOCK))
1073 				cpu_relax();
1074 			goto retry;
1075 		}
1076 
1077 		/* Now check and modify the HPTE */
1078 		if (!(hptep[0] & cpu_to_be64(HPTE_V_VALID))) {
1079 			__unlock_hpte(hptep, be64_to_cpu(hptep[0]));
1080 			continue;
1081 		}
1082 
1083 		/* need to make it temporarily absent so C is stable */
1084 		hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
1085 		kvmppc_invalidate_hpte(kvm, hptep, i);
1086 		v = be64_to_cpu(hptep[0]);
1087 		r = be64_to_cpu(hptep[1]);
1088 		if (r & HPTE_R_C) {
1089 			hptep[1] = cpu_to_be64(r & ~HPTE_R_C);
1090 			if (!(rev[i].guest_rpte & HPTE_R_C)) {
1091 				rev[i].guest_rpte |= HPTE_R_C;
1092 				note_hpte_modification(kvm, &rev[i]);
1093 			}
1094 			n = hpte_page_size(v, r);
1095 			n = (n + PAGE_SIZE - 1) >> PAGE_SHIFT;
1096 			if (n > npages_dirty)
1097 				npages_dirty = n;
1098 			eieio();
1099 		}
1100 		v &= ~HPTE_V_ABSENT;
1101 		v |= HPTE_V_VALID;
1102 		__unlock_hpte(hptep, v);
1103 	} while ((i = j) != head);
1104 
1105 	unlock_rmap(rmapp);
1106 	return npages_dirty;
1107 }
1108 
1109 void kvmppc_harvest_vpa_dirty(struct kvmppc_vpa *vpa,
1110 			      struct kvm_memory_slot *memslot,
1111 			      unsigned long *map)
1112 {
1113 	unsigned long gfn;
1114 
1115 	if (!vpa->dirty || !vpa->pinned_addr)
1116 		return;
1117 	gfn = vpa->gpa >> PAGE_SHIFT;
1118 	if (gfn < memslot->base_gfn ||
1119 	    gfn >= memslot->base_gfn + memslot->npages)
1120 		return;
1121 
1122 	vpa->dirty = false;
1123 	if (map)
1124 		__set_bit_le(gfn - memslot->base_gfn, map);
1125 }
1126 
1127 long kvmppc_hv_get_dirty_log_hpt(struct kvm *kvm,
1128 			struct kvm_memory_slot *memslot, unsigned long *map)
1129 {
1130 	unsigned long i, j;
1131 	unsigned long *rmapp;
1132 
1133 	preempt_disable();
1134 	rmapp = memslot->arch.rmap;
1135 	for (i = 0; i < memslot->npages; ++i) {
1136 		int npages = kvm_test_clear_dirty_npages(kvm, rmapp);
1137 		/*
1138 		 * Note that if npages > 0 then i must be a multiple of npages,
1139 		 * since we always put huge-page HPTEs in the rmap chain
1140 		 * corresponding to their page base address.
1141 		 */
1142 		if (npages && map)
1143 			for (j = i; npages; ++j, --npages)
1144 				__set_bit_le(j, map);
1145 		++rmapp;
1146 	}
1147 	preempt_enable();
1148 	return 0;
1149 }
1150 
1151 void *kvmppc_pin_guest_page(struct kvm *kvm, unsigned long gpa,
1152 			    unsigned long *nb_ret)
1153 {
1154 	struct kvm_memory_slot *memslot;
1155 	unsigned long gfn = gpa >> PAGE_SHIFT;
1156 	struct page *page, *pages[1];
1157 	int npages;
1158 	unsigned long hva, offset;
1159 	int srcu_idx;
1160 
1161 	srcu_idx = srcu_read_lock(&kvm->srcu);
1162 	memslot = gfn_to_memslot(kvm, gfn);
1163 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
1164 		goto err;
1165 	hva = gfn_to_hva_memslot(memslot, gfn);
1166 	npages = get_user_pages_fast(hva, 1, 1, pages);
1167 	if (npages < 1)
1168 		goto err;
1169 	page = pages[0];
1170 	srcu_read_unlock(&kvm->srcu, srcu_idx);
1171 
1172 	offset = gpa & (PAGE_SIZE - 1);
1173 	if (nb_ret)
1174 		*nb_ret = PAGE_SIZE - offset;
1175 	return page_address(page) + offset;
1176 
1177  err:
1178 	srcu_read_unlock(&kvm->srcu, srcu_idx);
1179 	return NULL;
1180 }
1181 
1182 void kvmppc_unpin_guest_page(struct kvm *kvm, void *va, unsigned long gpa,
1183 			     bool dirty)
1184 {
1185 	struct page *page = virt_to_page(va);
1186 	struct kvm_memory_slot *memslot;
1187 	unsigned long gfn;
1188 	unsigned long *rmap;
1189 	int srcu_idx;
1190 
1191 	put_page(page);
1192 
1193 	if (!dirty)
1194 		return;
1195 
1196 	/* We need to mark this page dirty in the rmap chain */
1197 	gfn = gpa >> PAGE_SHIFT;
1198 	srcu_idx = srcu_read_lock(&kvm->srcu);
1199 	memslot = gfn_to_memslot(kvm, gfn);
1200 	if (memslot) {
1201 		if (!kvm_is_radix(kvm)) {
1202 			rmap = &memslot->arch.rmap[gfn - memslot->base_gfn];
1203 			lock_rmap(rmap);
1204 			*rmap |= KVMPPC_RMAP_CHANGED;
1205 			unlock_rmap(rmap);
1206 		} else if (memslot->dirty_bitmap) {
1207 			mark_page_dirty(kvm, gfn);
1208 		}
1209 	}
1210 	srcu_read_unlock(&kvm->srcu, srcu_idx);
1211 }
1212 
1213 /*
1214  * HPT resizing
1215  */
1216 static int resize_hpt_allocate(struct kvm_resize_hpt *resize)
1217 {
1218 	int rc;
1219 
1220 	rc = kvmppc_allocate_hpt(&resize->hpt, resize->order);
1221 	if (rc < 0)
1222 		return rc;
1223 
1224 	resize_hpt_debug(resize, "resize_hpt_allocate(): HPT @ 0x%lx\n",
1225 			 resize->hpt.virt);
1226 
1227 	return 0;
1228 }
1229 
1230 static unsigned long resize_hpt_rehash_hpte(struct kvm_resize_hpt *resize,
1231 					    unsigned long idx)
1232 {
1233 	struct kvm *kvm = resize->kvm;
1234 	struct kvm_hpt_info *old = &kvm->arch.hpt;
1235 	struct kvm_hpt_info *new = &resize->hpt;
1236 	unsigned long old_hash_mask = (1ULL << (old->order - 7)) - 1;
1237 	unsigned long new_hash_mask = (1ULL << (new->order - 7)) - 1;
1238 	__be64 *hptep, *new_hptep;
1239 	unsigned long vpte, rpte, guest_rpte;
1240 	int ret;
1241 	struct revmap_entry *rev;
1242 	unsigned long apsize, psize, avpn, pteg, hash;
1243 	unsigned long new_idx, new_pteg, replace_vpte;
1244 
1245 	hptep = (__be64 *)(old->virt + (idx << 4));
1246 
1247 	/* Guest is stopped, so new HPTEs can't be added or faulted
1248 	 * in, only unmapped or altered by host actions.  So, it's
1249 	 * safe to check this before we take the HPTE lock */
1250 	vpte = be64_to_cpu(hptep[0]);
1251 	if (!(vpte & HPTE_V_VALID) && !(vpte & HPTE_V_ABSENT))
1252 		return 0; /* nothing to do */
1253 
1254 	while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
1255 		cpu_relax();
1256 
1257 	vpte = be64_to_cpu(hptep[0]);
1258 
1259 	ret = 0;
1260 	if (!(vpte & HPTE_V_VALID) && !(vpte & HPTE_V_ABSENT))
1261 		/* Nothing to do */
1262 		goto out;
1263 
1264 	/* Unmap */
1265 	rev = &old->rev[idx];
1266 	guest_rpte = rev->guest_rpte;
1267 
1268 	ret = -EIO;
1269 	apsize = hpte_page_size(vpte, guest_rpte);
1270 	if (!apsize)
1271 		goto out;
1272 
1273 	if (vpte & HPTE_V_VALID) {
1274 		unsigned long gfn = hpte_rpn(guest_rpte, apsize);
1275 		int srcu_idx = srcu_read_lock(&kvm->srcu);
1276 		struct kvm_memory_slot *memslot =
1277 			__gfn_to_memslot(kvm_memslots(kvm), gfn);
1278 
1279 		if (memslot) {
1280 			unsigned long *rmapp;
1281 			rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
1282 
1283 			lock_rmap(rmapp);
1284 			kvmppc_unmap_hpte(kvm, idx, rmapp, gfn);
1285 			unlock_rmap(rmapp);
1286 		}
1287 
1288 		srcu_read_unlock(&kvm->srcu, srcu_idx);
1289 	}
1290 
1291 	/* Reload PTE after unmap */
1292 	vpte = be64_to_cpu(hptep[0]);
1293 
1294 	BUG_ON(vpte & HPTE_V_VALID);
1295 	BUG_ON(!(vpte & HPTE_V_ABSENT));
1296 
1297 	ret = 0;
1298 	if (!(vpte & HPTE_V_BOLTED))
1299 		goto out;
1300 
1301 	rpte = be64_to_cpu(hptep[1]);
1302 	psize = hpte_base_page_size(vpte, rpte);
1303 	avpn = HPTE_V_AVPN_VAL(vpte) & ~((psize - 1) >> 23);
1304 	pteg = idx / HPTES_PER_GROUP;
1305 	if (vpte & HPTE_V_SECONDARY)
1306 		pteg = ~pteg;
1307 
1308 	if (!(vpte & HPTE_V_1TB_SEG)) {
1309 		unsigned long offset, vsid;
1310 
1311 		/* We only have 28 - 23 bits of offset in avpn */
1312 		offset = (avpn & 0x1f) << 23;
1313 		vsid = avpn >> 5;
1314 		/* We can find more bits from the pteg value */
1315 		if (psize < (1ULL << 23))
1316 			offset |= ((vsid ^ pteg) & old_hash_mask) * psize;
1317 
1318 		hash = vsid ^ (offset / psize);
1319 	} else {
1320 		unsigned long offset, vsid;
1321 
1322 		/* We only have 40 - 23 bits of seg_off in avpn */
1323 		offset = (avpn & 0x1ffff) << 23;
1324 		vsid = avpn >> 17;
1325 		if (psize < (1ULL << 23))
1326 			offset |= ((vsid ^ (vsid << 25) ^ pteg) & old_hash_mask) * psize;
1327 
1328 		hash = vsid ^ (vsid << 25) ^ (offset / psize);
1329 	}
1330 
1331 	new_pteg = hash & new_hash_mask;
1332 	if (vpte & HPTE_V_SECONDARY) {
1333 		BUG_ON(~pteg != (hash & old_hash_mask));
1334 		new_pteg = ~new_pteg;
1335 	} else {
1336 		BUG_ON(pteg != (hash & old_hash_mask));
1337 	}
1338 
1339 	new_idx = new_pteg * HPTES_PER_GROUP + (idx % HPTES_PER_GROUP);
1340 	new_hptep = (__be64 *)(new->virt + (new_idx << 4));
1341 
1342 	replace_vpte = be64_to_cpu(new_hptep[0]);
1343 
1344 	if (replace_vpte & (HPTE_V_VALID | HPTE_V_ABSENT)) {
1345 		BUG_ON(new->order >= old->order);
1346 
1347 		if (replace_vpte & HPTE_V_BOLTED) {
1348 			if (vpte & HPTE_V_BOLTED)
1349 				/* Bolted collision, nothing we can do */
1350 				ret = -ENOSPC;
1351 			/* Discard the new HPTE */
1352 			goto out;
1353 		}
1354 
1355 		/* Discard the previous HPTE */
1356 	}
1357 
1358 	new_hptep[1] = cpu_to_be64(rpte);
1359 	new->rev[new_idx].guest_rpte = guest_rpte;
1360 	/* No need for a barrier, since new HPT isn't active */
1361 	new_hptep[0] = cpu_to_be64(vpte);
1362 	unlock_hpte(new_hptep, vpte);
1363 
1364 out:
1365 	unlock_hpte(hptep, vpte);
1366 	return ret;
1367 }
1368 
1369 static int resize_hpt_rehash(struct kvm_resize_hpt *resize)
1370 {
1371 	struct kvm *kvm = resize->kvm;
1372 	unsigned  long i;
1373 	int rc;
1374 
1375 	/*
1376 	 * resize_hpt_rehash_hpte() doesn't handle the new-format HPTEs
1377 	 * that POWER9 uses, and could well hit a BUG_ON on POWER9.
1378 	 */
1379 	if (cpu_has_feature(CPU_FTR_ARCH_300))
1380 		return -EIO;
1381 	for (i = 0; i < kvmppc_hpt_npte(&kvm->arch.hpt); i++) {
1382 		rc = resize_hpt_rehash_hpte(resize, i);
1383 		if (rc != 0)
1384 			return rc;
1385 	}
1386 
1387 	return 0;
1388 }
1389 
1390 static void resize_hpt_pivot(struct kvm_resize_hpt *resize)
1391 {
1392 	struct kvm *kvm = resize->kvm;
1393 	struct kvm_hpt_info hpt_tmp;
1394 
1395 	/* Exchange the pending tables in the resize structure with
1396 	 * the active tables */
1397 
1398 	resize_hpt_debug(resize, "resize_hpt_pivot()\n");
1399 
1400 	spin_lock(&kvm->mmu_lock);
1401 	asm volatile("ptesync" : : : "memory");
1402 
1403 	hpt_tmp = kvm->arch.hpt;
1404 	kvmppc_set_hpt(kvm, &resize->hpt);
1405 	resize->hpt = hpt_tmp;
1406 
1407 	spin_unlock(&kvm->mmu_lock);
1408 
1409 	synchronize_srcu_expedited(&kvm->srcu);
1410 
1411 	resize_hpt_debug(resize, "resize_hpt_pivot() done\n");
1412 }
1413 
1414 static void resize_hpt_release(struct kvm *kvm, struct kvm_resize_hpt *resize)
1415 {
1416 	BUG_ON(kvm->arch.resize_hpt != resize);
1417 
1418 	if (!resize)
1419 		return;
1420 
1421 	if (resize->hpt.virt)
1422 		kvmppc_free_hpt(&resize->hpt);
1423 
1424 	kvm->arch.resize_hpt = NULL;
1425 	kfree(resize);
1426 }
1427 
1428 static void resize_hpt_prepare_work(struct work_struct *work)
1429 {
1430 	struct kvm_resize_hpt *resize = container_of(work,
1431 						     struct kvm_resize_hpt,
1432 						     work);
1433 	struct kvm *kvm = resize->kvm;
1434 	int err;
1435 
1436 	resize_hpt_debug(resize, "resize_hpt_prepare_work(): order = %d\n",
1437 			 resize->order);
1438 
1439 	err = resize_hpt_allocate(resize);
1440 
1441 	mutex_lock(&kvm->lock);
1442 
1443 	resize->error = err;
1444 	resize->prepare_done = true;
1445 
1446 	mutex_unlock(&kvm->lock);
1447 }
1448 
1449 long kvm_vm_ioctl_resize_hpt_prepare(struct kvm *kvm,
1450 				     struct kvm_ppc_resize_hpt *rhpt)
1451 {
1452 	unsigned long flags = rhpt->flags;
1453 	unsigned long shift = rhpt->shift;
1454 	struct kvm_resize_hpt *resize;
1455 	int ret;
1456 
1457 	if (flags != 0)
1458 		return -EINVAL;
1459 
1460 	if (shift && ((shift < 18) || (shift > 46)))
1461 		return -EINVAL;
1462 
1463 	mutex_lock(&kvm->lock);
1464 
1465 	resize = kvm->arch.resize_hpt;
1466 
1467 	if (resize) {
1468 		if (resize->order == shift) {
1469 			/* Suitable resize in progress */
1470 			if (resize->prepare_done) {
1471 				ret = resize->error;
1472 				if (ret != 0)
1473 					resize_hpt_release(kvm, resize);
1474 			} else {
1475 				ret = 100; /* estimated time in ms */
1476 			}
1477 
1478 			goto out;
1479 		}
1480 
1481 		/* not suitable, cancel it */
1482 		resize_hpt_release(kvm, resize);
1483 	}
1484 
1485 	ret = 0;
1486 	if (!shift)
1487 		goto out; /* nothing to do */
1488 
1489 	/* start new resize */
1490 
1491 	resize = kzalloc(sizeof(*resize), GFP_KERNEL);
1492 	if (!resize) {
1493 		ret = -ENOMEM;
1494 		goto out;
1495 	}
1496 	resize->order = shift;
1497 	resize->kvm = kvm;
1498 	INIT_WORK(&resize->work, resize_hpt_prepare_work);
1499 	kvm->arch.resize_hpt = resize;
1500 
1501 	schedule_work(&resize->work);
1502 
1503 	ret = 100; /* estimated time in ms */
1504 
1505 out:
1506 	mutex_unlock(&kvm->lock);
1507 	return ret;
1508 }
1509 
1510 static void resize_hpt_boot_vcpu(void *opaque)
1511 {
1512 	/* Nothing to do, just force a KVM exit */
1513 }
1514 
1515 long kvm_vm_ioctl_resize_hpt_commit(struct kvm *kvm,
1516 				    struct kvm_ppc_resize_hpt *rhpt)
1517 {
1518 	unsigned long flags = rhpt->flags;
1519 	unsigned long shift = rhpt->shift;
1520 	struct kvm_resize_hpt *resize;
1521 	long ret;
1522 
1523 	if (flags != 0)
1524 		return -EINVAL;
1525 
1526 	if (shift && ((shift < 18) || (shift > 46)))
1527 		return -EINVAL;
1528 
1529 	mutex_lock(&kvm->lock);
1530 
1531 	resize = kvm->arch.resize_hpt;
1532 
1533 	/* This shouldn't be possible */
1534 	ret = -EIO;
1535 	if (WARN_ON(!kvm->arch.hpte_setup_done))
1536 		goto out_no_hpt;
1537 
1538 	/* Stop VCPUs from running while we mess with the HPT */
1539 	kvm->arch.hpte_setup_done = 0;
1540 	smp_mb();
1541 
1542 	/* Boot all CPUs out of the guest so they re-read
1543 	 * hpte_setup_done */
1544 	on_each_cpu(resize_hpt_boot_vcpu, NULL, 1);
1545 
1546 	ret = -ENXIO;
1547 	if (!resize || (resize->order != shift))
1548 		goto out;
1549 
1550 	ret = -EBUSY;
1551 	if (!resize->prepare_done)
1552 		goto out;
1553 
1554 	ret = resize->error;
1555 	if (ret != 0)
1556 		goto out;
1557 
1558 	ret = resize_hpt_rehash(resize);
1559 	if (ret != 0)
1560 		goto out;
1561 
1562 	resize_hpt_pivot(resize);
1563 
1564 out:
1565 	/* Let VCPUs run again */
1566 	kvm->arch.hpte_setup_done = 1;
1567 	smp_mb();
1568 out_no_hpt:
1569 	resize_hpt_release(kvm, resize);
1570 	mutex_unlock(&kvm->lock);
1571 	return ret;
1572 }
1573 
1574 /*
1575  * Functions for reading and writing the hash table via reads and
1576  * writes on a file descriptor.
1577  *
1578  * Reads return the guest view of the hash table, which has to be
1579  * pieced together from the real hash table and the guest_rpte
1580  * values in the revmap array.
1581  *
1582  * On writes, each HPTE written is considered in turn, and if it
1583  * is valid, it is written to the HPT as if an H_ENTER with the
1584  * exact flag set was done.  When the invalid count is non-zero
1585  * in the header written to the stream, the kernel will make
1586  * sure that that many HPTEs are invalid, and invalidate them
1587  * if not.
1588  */
1589 
1590 struct kvm_htab_ctx {
1591 	unsigned long	index;
1592 	unsigned long	flags;
1593 	struct kvm	*kvm;
1594 	int		first_pass;
1595 };
1596 
1597 #define HPTE_SIZE	(2 * sizeof(unsigned long))
1598 
1599 /*
1600  * Returns 1 if this HPT entry has been modified or has pending
1601  * R/C bit changes.
1602  */
1603 static int hpte_dirty(struct revmap_entry *revp, __be64 *hptp)
1604 {
1605 	unsigned long rcbits_unset;
1606 
1607 	if (revp->guest_rpte & HPTE_GR_MODIFIED)
1608 		return 1;
1609 
1610 	/* Also need to consider changes in reference and changed bits */
1611 	rcbits_unset = ~revp->guest_rpte & (HPTE_R_R | HPTE_R_C);
1612 	if ((be64_to_cpu(hptp[0]) & HPTE_V_VALID) &&
1613 	    (be64_to_cpu(hptp[1]) & rcbits_unset))
1614 		return 1;
1615 
1616 	return 0;
1617 }
1618 
1619 static long record_hpte(unsigned long flags, __be64 *hptp,
1620 			unsigned long *hpte, struct revmap_entry *revp,
1621 			int want_valid, int first_pass)
1622 {
1623 	unsigned long v, r, hr;
1624 	unsigned long rcbits_unset;
1625 	int ok = 1;
1626 	int valid, dirty;
1627 
1628 	/* Unmodified entries are uninteresting except on the first pass */
1629 	dirty = hpte_dirty(revp, hptp);
1630 	if (!first_pass && !dirty)
1631 		return 0;
1632 
1633 	valid = 0;
1634 	if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT)) {
1635 		valid = 1;
1636 		if ((flags & KVM_GET_HTAB_BOLTED_ONLY) &&
1637 		    !(be64_to_cpu(hptp[0]) & HPTE_V_BOLTED))
1638 			valid = 0;
1639 	}
1640 	if (valid != want_valid)
1641 		return 0;
1642 
1643 	v = r = 0;
1644 	if (valid || dirty) {
1645 		/* lock the HPTE so it's stable and read it */
1646 		preempt_disable();
1647 		while (!try_lock_hpte(hptp, HPTE_V_HVLOCK))
1648 			cpu_relax();
1649 		v = be64_to_cpu(hptp[0]);
1650 		hr = be64_to_cpu(hptp[1]);
1651 		if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1652 			v = hpte_new_to_old_v(v, hr);
1653 			hr = hpte_new_to_old_r(hr);
1654 		}
1655 
1656 		/* re-evaluate valid and dirty from synchronized HPTE value */
1657 		valid = !!(v & HPTE_V_VALID);
1658 		dirty = !!(revp->guest_rpte & HPTE_GR_MODIFIED);
1659 
1660 		/* Harvest R and C into guest view if necessary */
1661 		rcbits_unset = ~revp->guest_rpte & (HPTE_R_R | HPTE_R_C);
1662 		if (valid && (rcbits_unset & hr)) {
1663 			revp->guest_rpte |= (hr &
1664 				(HPTE_R_R | HPTE_R_C)) | HPTE_GR_MODIFIED;
1665 			dirty = 1;
1666 		}
1667 
1668 		if (v & HPTE_V_ABSENT) {
1669 			v &= ~HPTE_V_ABSENT;
1670 			v |= HPTE_V_VALID;
1671 			valid = 1;
1672 		}
1673 		if ((flags & KVM_GET_HTAB_BOLTED_ONLY) && !(v & HPTE_V_BOLTED))
1674 			valid = 0;
1675 
1676 		r = revp->guest_rpte;
1677 		/* only clear modified if this is the right sort of entry */
1678 		if (valid == want_valid && dirty) {
1679 			r &= ~HPTE_GR_MODIFIED;
1680 			revp->guest_rpte = r;
1681 		}
1682 		unlock_hpte(hptp, be64_to_cpu(hptp[0]));
1683 		preempt_enable();
1684 		if (!(valid == want_valid && (first_pass || dirty)))
1685 			ok = 0;
1686 	}
1687 	hpte[0] = cpu_to_be64(v);
1688 	hpte[1] = cpu_to_be64(r);
1689 	return ok;
1690 }
1691 
1692 static ssize_t kvm_htab_read(struct file *file, char __user *buf,
1693 			     size_t count, loff_t *ppos)
1694 {
1695 	struct kvm_htab_ctx *ctx = file->private_data;
1696 	struct kvm *kvm = ctx->kvm;
1697 	struct kvm_get_htab_header hdr;
1698 	__be64 *hptp;
1699 	struct revmap_entry *revp;
1700 	unsigned long i, nb, nw;
1701 	unsigned long __user *lbuf;
1702 	struct kvm_get_htab_header __user *hptr;
1703 	unsigned long flags;
1704 	int first_pass;
1705 	unsigned long hpte[2];
1706 
1707 	if (!access_ok(VERIFY_WRITE, buf, count))
1708 		return -EFAULT;
1709 
1710 	first_pass = ctx->first_pass;
1711 	flags = ctx->flags;
1712 
1713 	i = ctx->index;
1714 	hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE));
1715 	revp = kvm->arch.hpt.rev + i;
1716 	lbuf = (unsigned long __user *)buf;
1717 
1718 	nb = 0;
1719 	while (nb + sizeof(hdr) + HPTE_SIZE < count) {
1720 		/* Initialize header */
1721 		hptr = (struct kvm_get_htab_header __user *)buf;
1722 		hdr.n_valid = 0;
1723 		hdr.n_invalid = 0;
1724 		nw = nb;
1725 		nb += sizeof(hdr);
1726 		lbuf = (unsigned long __user *)(buf + sizeof(hdr));
1727 
1728 		/* Skip uninteresting entries, i.e. clean on not-first pass */
1729 		if (!first_pass) {
1730 			while (i < kvmppc_hpt_npte(&kvm->arch.hpt) &&
1731 			       !hpte_dirty(revp, hptp)) {
1732 				++i;
1733 				hptp += 2;
1734 				++revp;
1735 			}
1736 		}
1737 		hdr.index = i;
1738 
1739 		/* Grab a series of valid entries */
1740 		while (i < kvmppc_hpt_npte(&kvm->arch.hpt) &&
1741 		       hdr.n_valid < 0xffff &&
1742 		       nb + HPTE_SIZE < count &&
1743 		       record_hpte(flags, hptp, hpte, revp, 1, first_pass)) {
1744 			/* valid entry, write it out */
1745 			++hdr.n_valid;
1746 			if (__put_user(hpte[0], lbuf) ||
1747 			    __put_user(hpte[1], lbuf + 1))
1748 				return -EFAULT;
1749 			nb += HPTE_SIZE;
1750 			lbuf += 2;
1751 			++i;
1752 			hptp += 2;
1753 			++revp;
1754 		}
1755 		/* Now skip invalid entries while we can */
1756 		while (i < kvmppc_hpt_npte(&kvm->arch.hpt) &&
1757 		       hdr.n_invalid < 0xffff &&
1758 		       record_hpte(flags, hptp, hpte, revp, 0, first_pass)) {
1759 			/* found an invalid entry */
1760 			++hdr.n_invalid;
1761 			++i;
1762 			hptp += 2;
1763 			++revp;
1764 		}
1765 
1766 		if (hdr.n_valid || hdr.n_invalid) {
1767 			/* write back the header */
1768 			if (__copy_to_user(hptr, &hdr, sizeof(hdr)))
1769 				return -EFAULT;
1770 			nw = nb;
1771 			buf = (char __user *)lbuf;
1772 		} else {
1773 			nb = nw;
1774 		}
1775 
1776 		/* Check if we've wrapped around the hash table */
1777 		if (i >= kvmppc_hpt_npte(&kvm->arch.hpt)) {
1778 			i = 0;
1779 			ctx->first_pass = 0;
1780 			break;
1781 		}
1782 	}
1783 
1784 	ctx->index = i;
1785 
1786 	return nb;
1787 }
1788 
1789 static ssize_t kvm_htab_write(struct file *file, const char __user *buf,
1790 			      size_t count, loff_t *ppos)
1791 {
1792 	struct kvm_htab_ctx *ctx = file->private_data;
1793 	struct kvm *kvm = ctx->kvm;
1794 	struct kvm_get_htab_header hdr;
1795 	unsigned long i, j;
1796 	unsigned long v, r;
1797 	unsigned long __user *lbuf;
1798 	__be64 *hptp;
1799 	unsigned long tmp[2];
1800 	ssize_t nb;
1801 	long int err, ret;
1802 	int hpte_setup;
1803 
1804 	if (!access_ok(VERIFY_READ, buf, count))
1805 		return -EFAULT;
1806 
1807 	/* lock out vcpus from running while we're doing this */
1808 	mutex_lock(&kvm->lock);
1809 	hpte_setup = kvm->arch.hpte_setup_done;
1810 	if (hpte_setup) {
1811 		kvm->arch.hpte_setup_done = 0;	/* temporarily */
1812 		/* order hpte_setup_done vs. vcpus_running */
1813 		smp_mb();
1814 		if (atomic_read(&kvm->arch.vcpus_running)) {
1815 			kvm->arch.hpte_setup_done = 1;
1816 			mutex_unlock(&kvm->lock);
1817 			return -EBUSY;
1818 		}
1819 	}
1820 
1821 	err = 0;
1822 	for (nb = 0; nb + sizeof(hdr) <= count; ) {
1823 		err = -EFAULT;
1824 		if (__copy_from_user(&hdr, buf, sizeof(hdr)))
1825 			break;
1826 
1827 		err = 0;
1828 		if (nb + hdr.n_valid * HPTE_SIZE > count)
1829 			break;
1830 
1831 		nb += sizeof(hdr);
1832 		buf += sizeof(hdr);
1833 
1834 		err = -EINVAL;
1835 		i = hdr.index;
1836 		if (i >= kvmppc_hpt_npte(&kvm->arch.hpt) ||
1837 		    i + hdr.n_valid + hdr.n_invalid > kvmppc_hpt_npte(&kvm->arch.hpt))
1838 			break;
1839 
1840 		hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE));
1841 		lbuf = (unsigned long __user *)buf;
1842 		for (j = 0; j < hdr.n_valid; ++j) {
1843 			__be64 hpte_v;
1844 			__be64 hpte_r;
1845 
1846 			err = -EFAULT;
1847 			if (__get_user(hpte_v, lbuf) ||
1848 			    __get_user(hpte_r, lbuf + 1))
1849 				goto out;
1850 			v = be64_to_cpu(hpte_v);
1851 			r = be64_to_cpu(hpte_r);
1852 			err = -EINVAL;
1853 			if (!(v & HPTE_V_VALID))
1854 				goto out;
1855 			lbuf += 2;
1856 			nb += HPTE_SIZE;
1857 
1858 			if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT))
1859 				kvmppc_do_h_remove(kvm, 0, i, 0, tmp);
1860 			err = -EIO;
1861 			ret = kvmppc_virtmode_do_h_enter(kvm, H_EXACT, i, v, r,
1862 							 tmp);
1863 			if (ret != H_SUCCESS) {
1864 				pr_err("kvm_htab_write ret %ld i=%ld v=%lx "
1865 				       "r=%lx\n", ret, i, v, r);
1866 				goto out;
1867 			}
1868 			if (!hpte_setup && is_vrma_hpte(v)) {
1869 				unsigned long psize = hpte_base_page_size(v, r);
1870 				unsigned long senc = slb_pgsize_encoding(psize);
1871 				unsigned long lpcr;
1872 
1873 				kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
1874 					(VRMA_VSID << SLB_VSID_SHIFT_1T);
1875 				lpcr = senc << (LPCR_VRMASD_SH - 4);
1876 				kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
1877 				hpte_setup = 1;
1878 			}
1879 			++i;
1880 			hptp += 2;
1881 		}
1882 
1883 		for (j = 0; j < hdr.n_invalid; ++j) {
1884 			if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT))
1885 				kvmppc_do_h_remove(kvm, 0, i, 0, tmp);
1886 			++i;
1887 			hptp += 2;
1888 		}
1889 		err = 0;
1890 	}
1891 
1892  out:
1893 	/* Order HPTE updates vs. hpte_setup_done */
1894 	smp_wmb();
1895 	kvm->arch.hpte_setup_done = hpte_setup;
1896 	mutex_unlock(&kvm->lock);
1897 
1898 	if (err)
1899 		return err;
1900 	return nb;
1901 }
1902 
1903 static int kvm_htab_release(struct inode *inode, struct file *filp)
1904 {
1905 	struct kvm_htab_ctx *ctx = filp->private_data;
1906 
1907 	filp->private_data = NULL;
1908 	if (!(ctx->flags & KVM_GET_HTAB_WRITE))
1909 		atomic_dec(&ctx->kvm->arch.hpte_mod_interest);
1910 	kvm_put_kvm(ctx->kvm);
1911 	kfree(ctx);
1912 	return 0;
1913 }
1914 
1915 static const struct file_operations kvm_htab_fops = {
1916 	.read		= kvm_htab_read,
1917 	.write		= kvm_htab_write,
1918 	.llseek		= default_llseek,
1919 	.release	= kvm_htab_release,
1920 };
1921 
1922 int kvm_vm_ioctl_get_htab_fd(struct kvm *kvm, struct kvm_get_htab_fd *ghf)
1923 {
1924 	int ret;
1925 	struct kvm_htab_ctx *ctx;
1926 	int rwflag;
1927 
1928 	/* reject flags we don't recognize */
1929 	if (ghf->flags & ~(KVM_GET_HTAB_BOLTED_ONLY | KVM_GET_HTAB_WRITE))
1930 		return -EINVAL;
1931 	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1932 	if (!ctx)
1933 		return -ENOMEM;
1934 	kvm_get_kvm(kvm);
1935 	ctx->kvm = kvm;
1936 	ctx->index = ghf->start_index;
1937 	ctx->flags = ghf->flags;
1938 	ctx->first_pass = 1;
1939 
1940 	rwflag = (ghf->flags & KVM_GET_HTAB_WRITE) ? O_WRONLY : O_RDONLY;
1941 	ret = anon_inode_getfd("kvm-htab", &kvm_htab_fops, ctx, rwflag | O_CLOEXEC);
1942 	if (ret < 0) {
1943 		kvm_put_kvm(kvm);
1944 		return ret;
1945 	}
1946 
1947 	if (rwflag == O_RDONLY) {
1948 		mutex_lock(&kvm->slots_lock);
1949 		atomic_inc(&kvm->arch.hpte_mod_interest);
1950 		/* make sure kvmppc_do_h_enter etc. see the increment */
1951 		synchronize_srcu_expedited(&kvm->srcu);
1952 		mutex_unlock(&kvm->slots_lock);
1953 	}
1954 
1955 	return ret;
1956 }
1957 
1958 struct debugfs_htab_state {
1959 	struct kvm	*kvm;
1960 	struct mutex	mutex;
1961 	unsigned long	hpt_index;
1962 	int		chars_left;
1963 	int		buf_index;
1964 	char		buf[64];
1965 };
1966 
1967 static int debugfs_htab_open(struct inode *inode, struct file *file)
1968 {
1969 	struct kvm *kvm = inode->i_private;
1970 	struct debugfs_htab_state *p;
1971 
1972 	p = kzalloc(sizeof(*p), GFP_KERNEL);
1973 	if (!p)
1974 		return -ENOMEM;
1975 
1976 	kvm_get_kvm(kvm);
1977 	p->kvm = kvm;
1978 	mutex_init(&p->mutex);
1979 	file->private_data = p;
1980 
1981 	return nonseekable_open(inode, file);
1982 }
1983 
1984 static int debugfs_htab_release(struct inode *inode, struct file *file)
1985 {
1986 	struct debugfs_htab_state *p = file->private_data;
1987 
1988 	kvm_put_kvm(p->kvm);
1989 	kfree(p);
1990 	return 0;
1991 }
1992 
1993 static ssize_t debugfs_htab_read(struct file *file, char __user *buf,
1994 				 size_t len, loff_t *ppos)
1995 {
1996 	struct debugfs_htab_state *p = file->private_data;
1997 	ssize_t ret, r;
1998 	unsigned long i, n;
1999 	unsigned long v, hr, gr;
2000 	struct kvm *kvm;
2001 	__be64 *hptp;
2002 
2003 	ret = mutex_lock_interruptible(&p->mutex);
2004 	if (ret)
2005 		return ret;
2006 
2007 	if (p->chars_left) {
2008 		n = p->chars_left;
2009 		if (n > len)
2010 			n = len;
2011 		r = copy_to_user(buf, p->buf + p->buf_index, n);
2012 		n -= r;
2013 		p->chars_left -= n;
2014 		p->buf_index += n;
2015 		buf += n;
2016 		len -= n;
2017 		ret = n;
2018 		if (r) {
2019 			if (!n)
2020 				ret = -EFAULT;
2021 			goto out;
2022 		}
2023 	}
2024 
2025 	kvm = p->kvm;
2026 	i = p->hpt_index;
2027 	hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE));
2028 	for (; len != 0 && i < kvmppc_hpt_npte(&kvm->arch.hpt);
2029 	     ++i, hptp += 2) {
2030 		if (!(be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT)))
2031 			continue;
2032 
2033 		/* lock the HPTE so it's stable and read it */
2034 		preempt_disable();
2035 		while (!try_lock_hpte(hptp, HPTE_V_HVLOCK))
2036 			cpu_relax();
2037 		v = be64_to_cpu(hptp[0]) & ~HPTE_V_HVLOCK;
2038 		hr = be64_to_cpu(hptp[1]);
2039 		gr = kvm->arch.hpt.rev[i].guest_rpte;
2040 		unlock_hpte(hptp, v);
2041 		preempt_enable();
2042 
2043 		if (!(v & (HPTE_V_VALID | HPTE_V_ABSENT)))
2044 			continue;
2045 
2046 		n = scnprintf(p->buf, sizeof(p->buf),
2047 			      "%6lx %.16lx %.16lx %.16lx\n",
2048 			      i, v, hr, gr);
2049 		p->chars_left = n;
2050 		if (n > len)
2051 			n = len;
2052 		r = copy_to_user(buf, p->buf, n);
2053 		n -= r;
2054 		p->chars_left -= n;
2055 		p->buf_index = n;
2056 		buf += n;
2057 		len -= n;
2058 		ret += n;
2059 		if (r) {
2060 			if (!ret)
2061 				ret = -EFAULT;
2062 			goto out;
2063 		}
2064 	}
2065 	p->hpt_index = i;
2066 
2067  out:
2068 	mutex_unlock(&p->mutex);
2069 	return ret;
2070 }
2071 
2072 static ssize_t debugfs_htab_write(struct file *file, const char __user *buf,
2073 			   size_t len, loff_t *ppos)
2074 {
2075 	return -EACCES;
2076 }
2077 
2078 static const struct file_operations debugfs_htab_fops = {
2079 	.owner	 = THIS_MODULE,
2080 	.open	 = debugfs_htab_open,
2081 	.release = debugfs_htab_release,
2082 	.read	 = debugfs_htab_read,
2083 	.write	 = debugfs_htab_write,
2084 	.llseek	 = generic_file_llseek,
2085 };
2086 
2087 void kvmppc_mmu_debugfs_init(struct kvm *kvm)
2088 {
2089 	kvm->arch.htab_dentry = debugfs_create_file("htab", 0400,
2090 						    kvm->arch.debugfs_dir, kvm,
2091 						    &debugfs_htab_fops);
2092 }
2093 
2094 void kvmppc_mmu_book3s_hv_init(struct kvm_vcpu *vcpu)
2095 {
2096 	struct kvmppc_mmu *mmu = &vcpu->arch.mmu;
2097 
2098 	vcpu->arch.slb_nr = 32;		/* POWER7/POWER8 */
2099 
2100 	if (kvm_is_radix(vcpu->kvm))
2101 		mmu->xlate = kvmppc_mmu_radix_xlate;
2102 	else
2103 		mmu->xlate = kvmppc_mmu_book3s_64_hv_xlate;
2104 	mmu->reset_msr = kvmppc_mmu_book3s_64_hv_reset_msr;
2105 
2106 	vcpu->arch.hflags |= BOOK3S_HFLAG_SLB;
2107 }
2108