1	// SPDX-License-Identifier: GPL-2.0-only
2	/*
3	*
4	* Copyright 2010 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
5	*/
6
7	#include <linux/types.h>
8	#include <linux/string.h>
9	#include <linux/kvm.h>
10	#include <linux/kvm_host.h>
11	#include <linux/highmem.h>
12	#include <linux/gfp.h>
13	#include <linux/slab.h>
14	#include <linux/hugetlb.h>
15	#include <linux/vmalloc.h>
16	#include <linux/srcu.h>
17	#include <linux/anon_inodes.h>
18	#include <linux/file.h>
19	#include <linux/debugfs.h>
20
21	#include <asm/kvm_ppc.h>
22	#include <asm/kvm_book3s.h>
23	#include <asm/book3s/64/mmu-hash.h>
24	#include <asm/hvcall.h>
25	#include <asm/synch.h>
26	#include <asm/ppc-opcode.h>
27	#include <asm/cputable.h>
28	#include <asm/pte-walk.h>
29
30	#include "book3s.h"
31	#include "book3s_hv.h"
32	#include "trace_hv.h"
33
34	//#define DEBUG_RESIZE_HPT 1
35
36	#ifdef DEBUG_RESIZE_HPT
37	#define resize_hpt_debug(resize, ...) \
38	do { \
39	printk(KERN_DEBUG "RESIZE HPT %p: ", resize); \
40	printk(__VA_ARGS__); \
41	} while (0)
42	#else
43	#define resize_hpt_debug(resize, ...) \
44	do { } while (0)
45	#endif
46
47	static long kvmppc_virtmode_do_h_enter(struct kvm *kvm, unsigned long flags,
48	long pte_index, unsigned long pteh,
49	unsigned long ptel, unsigned long *pte_idx_ret);
50
51	struct kvm_resize_hpt {
52	/* These fields read-only after init */
53	struct kvm *kvm;
54	struct work_struct work;
55	u32 order;
56
57	/* These fields protected by kvm->arch.mmu_setup_lock */
58
59	/* Possible values and their usage:
60	* <0 an error occurred during allocation,
61	* -EBUSY allocation is in the progress,
62	* 0 allocation made successfully.
63	*/
64	int error;
65
66	/* Private to the work thread, until error != -EBUSY,
67	* then protected by kvm->arch.mmu_setup_lock.
68	*/
69	struct kvm_hpt_info hpt;
70	};
71
72	int kvmppc_allocate_hpt(struct kvm_hpt_info *info, u32 order)
73	{
74	unsigned long hpt = 0;
75	int cma = 0;
76	struct page *page = NULL;
77	struct revmap_entry *rev;
78	unsigned long npte;
79
80	if ((order < PPC_MIN_HPT_ORDER) || (order > PPC_MAX_HPT_ORDER))
81	return -EINVAL;
82
83	page = kvm_alloc_hpt_cma(1ul << (order - PAGE_SHIFT));
84	if (page) {
85	hpt = (unsigned long)pfn_to_kaddr(page_to_pfn(page));
86	memset((void *)hpt, 0, (1ul << order));
87	cma = 1;
88	}
89
90	if (!hpt)
91	hpt = __get_free_pages(GFP_KERNEL|__GFP_ZERO|__GFP_RETRY_MAYFAIL
92	|__GFP_NOWARN, order: order - PAGE_SHIFT);
93
94	if (!hpt)
95	return -ENOMEM;
96
97	/* HPTEs are 2**4 bytes long */
98	npte = 1ul << (order - 4);
99
100	/* Allocate reverse map array */
101	rev = vmalloc(array_size(npte, sizeof(struct revmap_entry)));
102	if (!rev) {
103	if (cma)
104	kvm_free_hpt_cma(page, 1 << (order - PAGE_SHIFT));
105	else
106	free_pages(addr: hpt, order: order - PAGE_SHIFT);
107	return -ENOMEM;
108	}
109
110	info->order = order;
111	info->virt = hpt;
112	info->cma = cma;
113	info->rev = rev;
114
115	return 0;
116	}
117
118	void kvmppc_set_hpt(struct kvm *kvm, struct kvm_hpt_info *info)
119	{
120	atomic64_set(v: &kvm->arch.mmio_update, i: 0);
121	kvm->arch.hpt = *info;
122	kvm->arch.sdr1 = __pa(info->virt) | (info->order - 18);
123
124	pr_debug("KVM guest htab at %lx (order %ld), LPID %llx\n",
125	info->virt, (long)info->order, kvm->arch.lpid);
126	}
127
128	int kvmppc_alloc_reset_hpt(struct kvm *kvm, int order)
129	{
130	int err = -EBUSY;
131	struct kvm_hpt_info info;
132
133	mutex_lock(&kvm->arch.mmu_setup_lock);
134	if (kvm->arch.mmu_ready) {
135	kvm->arch.mmu_ready = 0;
136	/* order mmu_ready vs. vcpus_running */
137	smp_mb();
138	if (atomic_read(v: &kvm->arch.vcpus_running)) {
139	kvm->arch.mmu_ready = 1;
140	goto out;
141	}
142	}
143	if (kvm_is_radix(kvm)) {
144	err = kvmppc_switch_mmu_to_hpt(kvm);
145	if (err)
146	goto out;
147	}
148
149	if (kvm->arch.hpt.order == order) {
150	/* We already have a suitable HPT */
151
152	/* Set the entire HPT to 0, i.e. invalid HPTEs */
153	memset((void *)kvm->arch.hpt.virt, 0, 1ul << order);
154	/*
155	* Reset all the reverse-mapping chains for all memslots
156	*/
157	kvmppc_rmap_reset(kvm);
158	err = 0;
159	goto out;
160	}
161
162	if (kvm->arch.hpt.virt) {
163	kvmppc_free_hpt(&kvm->arch.hpt);
164	kvmppc_rmap_reset(kvm);
165	}
166
167	err = kvmppc_allocate_hpt(info: &info, order);
168	if (err < 0)
169	goto out;
170	kvmppc_set_hpt(kvm, info: &info);
171
172	out:
173	if (err == 0)
174	/* Ensure that each vcpu will flush its TLB on next entry. */
175	cpumask_setall(dstp: &kvm->arch.need_tlb_flush);
176
177	mutex_unlock(lock: &kvm->arch.mmu_setup_lock);
178	return err;
179	}
180
181	void kvmppc_free_hpt(struct kvm_hpt_info *info)
182	{
183	vfree(addr: info->rev);
184	info->rev = NULL;
185	if (info->cma)
186	kvm_free_hpt_cma(virt_to_page((void *)info->virt),
187	1 << (info->order - PAGE_SHIFT));
188	else if (info->virt)
189	free_pages(addr: info->virt, order: 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 nr_lpids;
261
262	if (!mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE))
263	return -EINVAL;
264
265	if (cpu_has_feature(CPU_FTR_HVMODE)) {
266	if (WARN_ON(mfspr(SPRN_LPID) != 0))
267	return -EINVAL;
268	nr_lpids = 1UL << mmu_lpid_bits;
269	} else {
270	nr_lpids = 1UL << KVM_MAX_NESTED_GUESTS_SHIFT;
271	}
272
273	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
274	/* POWER7 has 10-bit LPIDs, POWER8 has 12-bit LPIDs */
275	if (cpu_has_feature(CPU_FTR_ARCH_207S))
276	WARN_ON(nr_lpids != 1UL << 12);
277	else
278	WARN_ON(nr_lpids != 1UL << 10);
279
280	/*
281	* Reserve the last implemented LPID use in partition
282	* switching for POWER7 and POWER8.
283	*/
284	nr_lpids -= 1;
285	}
286
287	kvmppc_init_lpid(nr_lpids);
288
289	return 0;
290	}
291
292	static long kvmppc_virtmode_do_h_enter(struct kvm *kvm, unsigned long flags,
293	long pte_index, unsigned long pteh,
294	unsigned long ptel, unsigned long *pte_idx_ret)
295	{
296	long ret;
297
298	preempt_disable();
299	ret = kvmppc_do_h_enter(kvm, flags, pte_index, pteh, ptel,
300	kvm->mm->pgd, false, pte_idx_ret);
301	preempt_enable();
302	if (ret == H_TOO_HARD) {
303	/* this can't happen */
304	pr_err("KVM: Oops, kvmppc_h_enter returned too hard!\n");
305	ret = H_RESOURCE; /* or something */
306	}
307	return ret;
308
309	}
310
311	static struct kvmppc_slb *kvmppc_mmu_book3s_hv_find_slbe(struct kvm_vcpu *vcpu,
312	gva_t eaddr)
313	{
314	u64 mask;
315	int i;
316
317	for (i = 0; i < vcpu->arch.slb_nr; i++) {
318	if (!(vcpu->arch.slb[i].orige & SLB_ESID_V))
319	continue;
320
321	if (vcpu->arch.slb[i].origv & SLB_VSID_B_1T)
322	mask = ESID_MASK_1T;
323	else
324	mask = ESID_MASK;
325
326	if (((vcpu->arch.slb[i].orige ^ eaddr) & mask) == 0)
327	return &vcpu->arch.slb[i];
328	}
329	return NULL;
330	}
331
332	static unsigned long kvmppc_mmu_get_real_addr(unsigned long v, unsigned long r,
333	unsigned long ea)
334	{
335	unsigned long ra_mask;
336
337	ra_mask = kvmppc_actual_pgsz(v, r) - 1;
338	return (r & HPTE_R_RPN & ~ra_mask) | (ea & ra_mask);
339	}
340
341	static int kvmppc_mmu_book3s_64_hv_xlate(struct kvm_vcpu *vcpu, gva_t eaddr,
342	struct kvmppc_pte *gpte, bool data, bool iswrite)
343	{
344	struct kvm *kvm = vcpu->kvm;
345	struct kvmppc_slb *slbe;
346	unsigned long slb_v;
347	unsigned long pp, key;
348	unsigned long v, orig_v, gr;
349	__be64 *hptep;
350	long int index;
351	int virtmode = __kvmppc_get_msr_hv(vcpu) & (data ? MSR_DR : MSR_IR);
352
353	if (kvm_is_radix(vcpu->kvm))
354	return kvmppc_mmu_radix_xlate(vcpu, eaddr, gpte, data, iswrite);
355
356	/* Get SLB entry */
357	if (virtmode) {
358	slbe = kvmppc_mmu_book3s_hv_find_slbe(vcpu, eaddr);
359	if (!slbe)
360	return -EINVAL;
361	slb_v = slbe->origv;
362	} else {
363	/* real mode access */
364	slb_v = vcpu->kvm->arch.vrma_slb_v;
365	}
366
367	preempt_disable();
368	/* Find the HPTE in the hash table */
369	index = kvmppc_hv_find_lock_hpte(kvm, eaddr, slb_v,
370	HPTE_V_VALID | HPTE_V_ABSENT);
371	if (index < 0) {
372	preempt_enable();
373	return -ENOENT;
374	}
375	hptep = (__be64 *)(kvm->arch.hpt.virt + (index << 4));
376	v = orig_v = be64_to_cpu(hptep[0]) & ~HPTE_V_HVLOCK;
377	if (cpu_has_feature(CPU_FTR_ARCH_300))
378	v = hpte_new_to_old_v(v, be64_to_cpu(hptep[1]));
379	gr = kvm->arch.hpt.rev[index].guest_rpte;
380
381	unlock_hpte(hptep, orig_v);
382	preempt_enable();
383
384	gpte->eaddr = eaddr;
385	gpte->vpage = ((v & HPTE_V_AVPN) << 4) | ((eaddr >> 12) & 0xfff);
386
387	/* Get PP bits and key for permission check */
388	pp = gr & (HPTE_R_PP0 | HPTE_R_PP);
389	key = (__kvmppc_get_msr_hv(vcpu) & MSR_PR) ? SLB_VSID_KP : SLB_VSID_KS;
390	key &= slb_v;
391
392	/* Calculate permissions */
393	gpte->may_read = hpte_read_permission(pp, key);
394	gpte->may_write = hpte_write_permission(pp, key);
395	gpte->may_execute = gpte->may_read && !(gr & (HPTE_R_N | HPTE_R_G));
396
397	/* Storage key permission check for POWER7 */
398	if (data && virtmode) {
399	int amrfield = hpte_get_skey_perm(gr, vcpu->arch.amr);
400	if (amrfield & 1)
401	gpte->may_read = 0;
402	if (amrfield & 2)
403	gpte->may_write = 0;
404	}
405
406	/* Get the guest physical address */
407	gpte->raddr = kvmppc_mmu_get_real_addr(v, r: gr, ea: eaddr);
408	return 0;
409	}
410
411	/*
412	* Quick test for whether an instruction is a load or a store.
413	* If the instruction is a load or a store, then this will indicate
414	* which it is, at least on server processors. (Embedded processors
415	* have some external PID instructions that don't follow the rule
416	* embodied here.) If the instruction isn't a load or store, then
417	* this doesn't return anything useful.
418	*/
419	static int instruction_is_store(ppc_inst_t instr)
420	{
421	unsigned int mask;
422	unsigned int suffix;
423
424	mask = 0x10000000;
425	suffix = ppc_inst_val(instr);
426	if (ppc_inst_prefixed(instr))
427	suffix = ppc_inst_suffix(instr);
428	else if ((suffix & 0xfc000000) == 0x7c000000)
429	mask = 0x100; /* major opcode 31 */
430	return (suffix & mask) != 0;
431	}
432
433	int kvmppc_hv_emulate_mmio(struct kvm_vcpu *vcpu,
434	unsigned long gpa, gva_t ea, int is_store)
435	{
436	ppc_inst_t last_inst;
437	bool is_prefixed = !!(kvmppc_get_msr(vcpu) & SRR1_PREFIXED);
438
439	/*
440	* Fast path - check if the guest physical address corresponds to a
441	* device on the FAST_MMIO_BUS, if so we can avoid loading the
442	* instruction all together, then we can just handle it and return.
443	*/
444	if (is_store) {
445	int idx, ret;
446
447	idx = srcu_read_lock(ssp: &vcpu->kvm->srcu);
448	ret = kvm_io_bus_write(vcpu, bus_idx: KVM_FAST_MMIO_BUS, addr: (gpa_t) gpa, len: 0,
449	NULL);
450	srcu_read_unlock(ssp: &vcpu->kvm->srcu, idx);
451	if (!ret) {
452	kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + (is_prefixed ? 8 : 4));
453	return RESUME_GUEST;
454	}
455	}
456
457	/*
458	* If we fail, we just return to the guest and try executing it again.
459	*/
460	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
461	EMULATE_DONE)
462	return RESUME_GUEST;
463
464	/*
465	* WARNING: We do not know for sure whether the instruction we just
466	* read from memory is the same that caused the fault in the first
467	* place.
468	*
469	* If the fault is prefixed but the instruction is not or vice
470	* versa, try again so that we don't advance pc the wrong amount.
471	*/
472	if (ppc_inst_prefixed(last_inst) != is_prefixed)
473	return RESUME_GUEST;
474
475	/*
476	* If the instruction we read is neither an load or a store,
477	* then it can't access memory, so we don't need to worry about
478	* enforcing access permissions. So, assuming it is a load or
479	* store, we just check that its direction (load or store) is
480	* consistent with the original fault, since that's what we
481	* checked the access permissions against. If there is a mismatch
482	* we just return and retry the instruction.
483	*/
484
485	if (instruction_is_store(last_inst) != !!is_store)
486	return RESUME_GUEST;
487
488	/*
489	* Emulated accesses are emulated by looking at the hash for
490	* translation once, then performing the access later. The
491	* translation could be invalidated in the meantime in which
492	* point performing the subsequent memory access on the old
493	* physical address could possibly be a security hole for the
494	* guest (but not the host).
495	*
496	* This is less of an issue for MMIO stores since they aren't
497	* globally visible. It could be an issue for MMIO loads to
498	* a certain extent but we'll ignore it for now.
499	*/
500
501	vcpu->arch.paddr_accessed = gpa;
502	vcpu->arch.vaddr_accessed = ea;
503	return kvmppc_emulate_mmio(vcpu);
504	}
505
506	int kvmppc_book3s_hv_page_fault(struct kvm_vcpu *vcpu,
507	unsigned long ea, unsigned long dsisr)
508	{
509	struct kvm *kvm = vcpu->kvm;
510	unsigned long hpte[3], r;
511	unsigned long hnow_v, hnow_r;
512	__be64 *hptep;
513	unsigned long mmu_seq, psize, pte_size;
514	unsigned long gpa_base, gfn_base;
515	unsigned long gpa, gfn, hva, pfn, hpa;
516	struct kvm_memory_slot *memslot;
517	unsigned long *rmap;
518	struct revmap_entry *rev;
519	struct page *page;
520	long index, ret;
521	bool is_ci;
522	bool writing, write_ok;
523	unsigned int shift;
524	unsigned long rcbits;
525	long mmio_update;
526	pte_t pte, *ptep;
527
528	if (kvm_is_radix(kvm))
529	return kvmppc_book3s_radix_page_fault(vcpu, ea, dsisr);
530
531	/*
532	* Real-mode code has already searched the HPT and found the
533	* entry we're interested in. Lock the entry and check that
534	* it hasn't changed. If it has, just return and re-execute the
535	* instruction.
536	*/
537	if (ea != vcpu->arch.pgfault_addr)
538	return RESUME_GUEST;
539
540	if (vcpu->arch.pgfault_cache) {
541	mmio_update = atomic64_read(v: &kvm->arch.mmio_update);
542	if (mmio_update == vcpu->arch.pgfault_cache->mmio_update) {
543	r = vcpu->arch.pgfault_cache->rpte;
544	psize = kvmppc_actual_pgsz(vcpu->arch.pgfault_hpte[0],
545	r);
546	gpa_base = r & HPTE_R_RPN & ~(psize - 1);
547	gfn_base = gpa_base >> PAGE_SHIFT;
548	gpa = gpa_base | (ea & (psize - 1));
549	return kvmppc_hv_emulate_mmio(vcpu, gpa, ea,
550	dsisr & DSISR_ISSTORE);
551	}
552	}
553	index = vcpu->arch.pgfault_index;
554	hptep = (__be64 *)(kvm->arch.hpt.virt + (index << 4));
555	rev = &kvm->arch.hpt.rev[index];
556	preempt_disable();
557	while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
558	cpu_relax();
559	hpte[0] = be64_to_cpu(hptep[0]) & ~HPTE_V_HVLOCK;
560	hpte[1] = be64_to_cpu(hptep[1]);
561	hpte[2] = r = rev->guest_rpte;
562	unlock_hpte(hptep, hpte[0]);
563	preempt_enable();
564
565	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
566	hpte[0] = hpte_new_to_old_v(hpte[0], hpte[1]);
567	hpte[1] = hpte_new_to_old_r(hpte[1]);
568	}
569	if (hpte[0] != vcpu->arch.pgfault_hpte[0] ||
570	hpte[1] != vcpu->arch.pgfault_hpte[1])
571	return RESUME_GUEST;
572
573	/* Translate the logical address and get the page */
574	psize = kvmppc_actual_pgsz(hpte[0], r);
575	gpa_base = r & HPTE_R_RPN & ~(psize - 1);
576	gfn_base = gpa_base >> PAGE_SHIFT;
577	gpa = gpa_base | (ea & (psize - 1));
578	gfn = gpa >> PAGE_SHIFT;
579	memslot = gfn_to_memslot(kvm, gfn);
580
581	trace_kvm_page_fault_enter(vcpu, hptep: hpte, memslot, ea, dsisr);
582
583	/* No memslot means it's an emulated MMIO region */
584	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
585	return kvmppc_hv_emulate_mmio(vcpu, gpa, ea,
586	dsisr & DSISR_ISSTORE);
587
588	/*
589	* This should never happen, because of the slot_is_aligned()
590	* check in kvmppc_do_h_enter().
591	*/
592	if (gfn_base < memslot->base_gfn)
593	return -EFAULT;
594
595	/* used to check for invalidations in progress */
596	mmu_seq = kvm->mmu_invalidate_seq;
597	smp_rmb();
598
599	ret = -EFAULT;
600	page = NULL;
601	writing = (dsisr & DSISR_ISSTORE) != 0;
602	/* If writing != 0, then the HPTE must allow writing, if we get here */
603	write_ok = writing;
604	hva = gfn_to_hva_memslot(slot: memslot, gfn);
605
606	/*
607	* Do a fast check first, since __gfn_to_pfn_memslot doesn't
608	* do it with !atomic && !async, which is how we call it.
609	* We always ask for write permission since the common case
610	* is that the page is writable.
611	*/
612	if (get_user_page_fast_only(addr: hva, gup_flags: FOLL_WRITE, pagep: &page)) {
613	write_ok = true;
614	} else {
615	/* Call KVM generic code to do the slow-path check */
616	pfn = __gfn_to_pfn_memslot(slot: memslot, gfn, atomic: false, interruptible: false, NULL,
617	write_fault: writing, writable: &write_ok, NULL);
618	if (is_error_noslot_pfn(pfn))
619	return -EFAULT;
620	page = NULL;
621	if (pfn_valid(pfn)) {
622	page = pfn_to_page(pfn);
623	if (PageReserved(page))
624	page = NULL;
625	}
626	}
627
628	/*
629	* Read the PTE from the process' radix tree and use that
630	* so we get the shift and attribute bits.
631	*/
632	spin_lock(lock: &kvm->mmu_lock);
633	ptep = find_kvm_host_pte(kvm, mmu_seq, hva, &shift);
634	pte = __pte(val: 0);
635	if (ptep)
636	pte = READ_ONCE(*ptep);
637	spin_unlock(lock: &kvm->mmu_lock);
638	/*
639	* If the PTE disappeared temporarily due to a THP
640	* collapse, just return and let the guest try again.
641	*/
642	if (!pte_present(a: pte)) {
643	if (page)
644	put_page(page);
645	return RESUME_GUEST;
646	}
647	hpa = pte_pfn(pte) << PAGE_SHIFT;
648	pte_size = PAGE_SIZE;
649	if (shift)
650	pte_size = 1ul << shift;
651	is_ci = pte_ci(pte);
652
653	if (psize > pte_size)
654	goto out_put;
655	if (pte_size > psize)
656	hpa |= hva & (pte_size - psize);
657
658	/* Check WIMG vs. the actual page we're accessing */
659	if (!hpte_cache_flags_ok(r, is_ci)) {
660	if (is_ci)
661	goto out_put;
662	/*
663	* Allow guest to map emulated device memory as
664	* uncacheable, but actually make it cacheable.
665	*/
666	r = (r & ~(HPTE_R_W|HPTE_R_I|HPTE_R_G)) | HPTE_R_M;
667	}
668
669	/*
670	* Set the HPTE to point to hpa.
671	* Since the hpa is at PAGE_SIZE granularity, make sure we
672	* don't mask out lower-order bits if psize < PAGE_SIZE.
673	*/
674	if (psize < PAGE_SIZE)
675	psize = PAGE_SIZE;
676	r = (r & HPTE_R_KEY_HI) | (r & ~(HPTE_R_PP0 - psize)) | hpa;
677	if (hpte_is_writable(r) && !write_ok)
678	r = hpte_make_readonly(r);
679	ret = RESUME_GUEST;
680	preempt_disable();
681	while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
682	cpu_relax();
683	hnow_v = be64_to_cpu(hptep[0]);
684	hnow_r = be64_to_cpu(hptep[1]);
685	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
686	hnow_v = hpte_new_to_old_v(hnow_v, hnow_r);
687	hnow_r = hpte_new_to_old_r(hnow_r);
688	}
689
690	/*
691	* If the HPT is being resized, don't update the HPTE,
692	* instead let the guest retry after the resize operation is complete.
693	* The synchronization for mmu_ready test vs. set is provided
694	* by the HPTE lock.
695	*/
696	if (!kvm->arch.mmu_ready)
697	goto out_unlock;
698
699	if ((hnow_v & ~HPTE_V_HVLOCK) != hpte[0] || hnow_r != hpte[1] ||
700	rev->guest_rpte != hpte[2])
701	/* HPTE has been changed under us; let the guest retry */
702	goto out_unlock;
703	hpte[0] = (hpte[0] & ~HPTE_V_ABSENT) | HPTE_V_VALID;
704
705	/* Always put the HPTE in the rmap chain for the page base address */
706	rmap = &memslot->arch.rmap[gfn_base - memslot->base_gfn];
707	lock_rmap(rmap);
708
709	/* Check if we might have been invalidated; let the guest retry if so */
710	ret = RESUME_GUEST;
711	if (mmu_invalidate_retry(kvm: vcpu->kvm, mmu_seq)) {
712	unlock_rmap(rmap);
713	goto out_unlock;
714	}
715
716	/* Only set R/C in real HPTE if set in both *rmap and guest_rpte */
717	rcbits = *rmap >> KVMPPC_RMAP_RC_SHIFT;
718	r &= rcbits | ~(HPTE_R_R | HPTE_R_C);
719
720	if (be64_to_cpu(hptep[0]) & HPTE_V_VALID) {
721	/* HPTE was previously valid, so we need to invalidate it */
722	unlock_rmap(rmap);
723	hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
724	kvmppc_invalidate_hpte(kvm, hptep, index);
725	/* don't lose previous R and C bits */
726	r |= be64_to_cpu(hptep[1]) & (HPTE_R_R | HPTE_R_C);
727	} else {
728	kvmppc_add_revmap_chain(kvm, rev, rmap, index, 0);
729	}
730
731	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
732	r = hpte_old_to_new_r(hpte[0], r);
733	hpte[0] = hpte_old_to_new_v(hpte[0]);
734	}
735	hptep[1] = cpu_to_be64(r);
736	eieio();
737	__unlock_hpte(hptep, hpte[0]);
738	asm volatile("ptesync" : : : "memory");
739	preempt_enable();
740	if (page && hpte_is_writable(r))
741	set_page_dirty_lock(page);
742
743	out_put:
744	trace_kvm_page_fault_exit(vcpu, hptep: hpte, ret);
745
746	if (page)
747	put_page(page);
748	return ret;
749
750	out_unlock:
751	__unlock_hpte(hptep, be64_to_cpu(hptep[0]));
752	preempt_enable();
753	goto out_put;
754	}
755
756	void kvmppc_rmap_reset(struct kvm *kvm)
757	{
758	struct kvm_memslots *slots;
759	struct kvm_memory_slot *memslot;
760	int srcu_idx, bkt;
761
762	srcu_idx = srcu_read_lock(ssp: &kvm->srcu);
763	slots = kvm_memslots(kvm);
764	kvm_for_each_memslot(memslot, bkt, slots) {
765	/* Mutual exclusion with kvm_unmap_hva_range etc. */
766	spin_lock(lock: &kvm->mmu_lock);
767	/*
768	* This assumes it is acceptable to lose reference and
769	* change bits across a reset.
770	*/
771	memset(memslot->arch.rmap, 0,
772	memslot->npages * sizeof(*memslot->arch.rmap));
773	spin_unlock(lock: &kvm->mmu_lock);
774	}
775	srcu_read_unlock(ssp: &kvm->srcu, idx: srcu_idx);
776	}
777
778	/* Must be called with both HPTE and rmap locked */
779	static void kvmppc_unmap_hpte(struct kvm *kvm, unsigned long i,
780	struct kvm_memory_slot *memslot,
781	unsigned long *rmapp, unsigned long gfn)
782	{
783	__be64 *hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
784	struct revmap_entry *rev = kvm->arch.hpt.rev;
785	unsigned long j, h;
786	unsigned long ptel, psize, rcbits;
787
788	j = rev[i].forw;
789	if (j == i) {
790	/* chain is now empty */
791	*rmapp &= ~(KVMPPC_RMAP_PRESENT | KVMPPC_RMAP_INDEX);
792	} else {
793	/* remove i from chain */
794	h = rev[i].back;
795	rev[h].forw = j;
796	rev[j].back = h;
797	rev[i].forw = rev[i].back = i;
798	*rmapp = (*rmapp & ~KVMPPC_RMAP_INDEX) | j;
799	}
800
801	/* Now check and modify the HPTE */
802	ptel = rev[i].guest_rpte;
803	psize = kvmppc_actual_pgsz(be64_to_cpu(hptep[0]), ptel);
804	if ((be64_to_cpu(hptep[0]) & HPTE_V_VALID) &&
805	hpte_rpn(ptel, psize) == gfn) {
806	hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
807	kvmppc_invalidate_hpte(kvm, hptep, i);
808	hptep[1] &= ~cpu_to_be64(HPTE_R_KEY_HI | HPTE_R_KEY_LO);
809	/* Harvest R and C */
810	rcbits = be64_to_cpu(hptep[1]) & (HPTE_R_R | HPTE_R_C);
811	*rmapp |= rcbits << KVMPPC_RMAP_RC_SHIFT;
812	if ((rcbits & HPTE_R_C) && memslot->dirty_bitmap)
813	kvmppc_update_dirty_map(memslot, gfn, psize);
814	if (rcbits & ~rev[i].guest_rpte) {
815	rev[i].guest_rpte = ptel | rcbits;
816	note_hpte_modification(kvm, &rev[i]);
817	}
818	}
819	}
820
821	static void kvm_unmap_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot,
822	unsigned long gfn)
823	{
824	unsigned long i;
825	__be64 *hptep;
826	unsigned long *rmapp;
827
828	rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
829	for (;;) {
830	lock_rmap(rmapp);
831	if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
832	unlock_rmap(rmapp);
833	break;
834	}
835
836	/*
837	* To avoid an ABBA deadlock with the HPTE lock bit,
838	* we can't spin on the HPTE lock while holding the
839	* rmap chain lock.
840	*/
841	i = *rmapp & KVMPPC_RMAP_INDEX;
842	hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
843	if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
844	/* unlock rmap before spinning on the HPTE lock */
845	unlock_rmap(rmapp);
846	while (be64_to_cpu(hptep[0]) & HPTE_V_HVLOCK)
847	cpu_relax();
848	continue;
849	}
850
851	kvmppc_unmap_hpte(kvm, i, memslot, rmapp, gfn);
852	unlock_rmap(rmapp);
853	__unlock_hpte(hptep, be64_to_cpu(hptep[0]));
854	}
855	}
856
857	bool kvm_unmap_gfn_range_hv(struct kvm *kvm, struct kvm_gfn_range *range)
858	{
859	gfn_t gfn;
860
861	if (kvm_is_radix(kvm)) {
862	for (gfn = range->start; gfn < range->end; gfn++)
863	kvm_unmap_radix(kvm, range->slot, gfn);
864	} else {
865	for (gfn = range->start; gfn < range->end; gfn++)
866	kvm_unmap_rmapp(kvm, memslot: range->slot, gfn);
867	}
868
869	return false;
870	}
871
872	void kvmppc_core_flush_memslot_hv(struct kvm *kvm,
873	struct kvm_memory_slot *memslot)
874	{
875	unsigned long gfn;
876	unsigned long n;
877	unsigned long *rmapp;
878
879	gfn = memslot->base_gfn;
880	rmapp = memslot->arch.rmap;
881	if (kvm_is_radix(kvm)) {
882	kvmppc_radix_flush_memslot(kvm, memslot);
883	return;
884	}
885
886	for (n = memslot->npages; n; --n, ++gfn) {
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 bool 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	bool ret = false;
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 = true;
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 = true;
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	bool kvm_age_gfn_hv(struct kvm *kvm, struct kvm_gfn_range *range)
955	{
956	gfn_t gfn;
957	bool ret = false;
958
959	if (kvm_is_radix(kvm)) {
960	for (gfn = range->start; gfn < range->end; gfn++)
961	ret |= kvm_age_radix(kvm, range->slot, gfn);
962	} else {
963	for (gfn = range->start; gfn < range->end; gfn++)
964	ret |= kvm_age_rmapp(kvm, memslot: range->slot, gfn);
965	}
966
967	return ret;
968	}
969
970	static bool kvm_test_age_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot,
971	unsigned long gfn)
972	{
973	struct revmap_entry *rev = kvm->arch.hpt.rev;
974	unsigned long head, i, j;
975	unsigned long *hp;
976	bool ret = true;
977	unsigned long *rmapp;
978
979	rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
980	if (*rmapp & KVMPPC_RMAP_REFERENCED)
981	return true;
982
983	lock_rmap(rmapp);
984	if (*rmapp & KVMPPC_RMAP_REFERENCED)
985	goto out;
986
987	if (*rmapp & KVMPPC_RMAP_PRESENT) {
988	i = head = *rmapp & KVMPPC_RMAP_INDEX;
989	do {
990	hp = (unsigned long *)(kvm->arch.hpt.virt + (i << 4));
991	j = rev[i].forw;
992	if (be64_to_cpu(hp[1]) & HPTE_R_R)
993	goto out;
994	} while ((i = j) != head);
995	}
996	ret = false;
997
998	out:
999	unlock_rmap(rmapp);
1000	return ret;
1001	}
1002
1003	bool kvm_test_age_gfn_hv(struct kvm *kvm, struct kvm_gfn_range *range)
1004	{
1005	WARN_ON(range->start + 1 != range->end);
1006
1007	if (kvm_is_radix(kvm))
1008	return kvm_test_age_radix(kvm, range->slot, range->start);
1009	else
1010	return kvm_test_age_rmapp(kvm, memslot: range->slot, gfn: range->start);
1011	}
1012
1013	bool kvm_set_spte_gfn_hv(struct kvm *kvm, struct kvm_gfn_range *range)
1014	{
1015	WARN_ON(range->start + 1 != range->end);
1016
1017	if (kvm_is_radix(kvm))
1018	kvm_unmap_radix(kvm, range->slot, range->start);
1019	else
1020	kvm_unmap_rmapp(kvm, memslot: range->slot, gfn: range->start);
1021
1022	return false;
1023	}
1024
1025	static int vcpus_running(struct kvm *kvm)
1026	{
1027	return atomic_read(v: &kvm->arch.vcpus_running) != 0;
1028	}
1029
1030	/*
1031	* Returns the number of system pages that are dirty.
1032	* This can be more than 1 if we find a huge-page HPTE.
1033	*/
1034	static int kvm_test_clear_dirty_npages(struct kvm *kvm, unsigned long *rmapp)
1035	{
1036	struct revmap_entry *rev = kvm->arch.hpt.rev;
1037	unsigned long head, i, j;
1038	unsigned long n;
1039	unsigned long v, r;
1040	__be64 *hptep;
1041	int npages_dirty = 0;
1042
1043	retry:
1044	lock_rmap(rmapp);
1045	if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
1046	unlock_rmap(rmapp);
1047	return npages_dirty;
1048	}
1049
1050	i = head = *rmapp & KVMPPC_RMAP_INDEX;
1051	do {
1052	unsigned long hptep1;
1053	hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
1054	j = rev[i].forw;
1055
1056	/*
1057	* Checking the C (changed) bit here is racy since there
1058	* is no guarantee about when the hardware writes it back.
1059	* If the HPTE is not writable then it is stable since the
1060	* page can't be written to, and we would have done a tlbie
1061	* (which forces the hardware to complete any writeback)
1062	* when making the HPTE read-only.
1063	* If vcpus are running then this call is racy anyway
1064	* since the page could get dirtied subsequently, so we
1065	* expect there to be a further call which would pick up
1066	* any delayed C bit writeback.
1067	* Otherwise we need to do the tlbie even if C==0 in
1068	* order to pick up any delayed writeback of C.
1069	*/
1070	hptep1 = be64_to_cpu(hptep[1]);
1071	if (!(hptep1 & HPTE_R_C) &&
1072	(!hpte_is_writable(hptep1) || vcpus_running(kvm)))
1073	continue;
1074
1075	if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
1076	/* unlock rmap before spinning on the HPTE lock */
1077	unlock_rmap(rmapp);
1078	while (hptep[0] & cpu_to_be64(HPTE_V_HVLOCK))
1079	cpu_relax();
1080	goto retry;
1081	}
1082
1083	/* Now check and modify the HPTE */
1084	if (!(hptep[0] & cpu_to_be64(HPTE_V_VALID))) {
1085	__unlock_hpte(hptep, be64_to_cpu(hptep[0]));
1086	continue;
1087	}
1088
1089	/* need to make it temporarily absent so C is stable */
1090	hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
1091	kvmppc_invalidate_hpte(kvm, hptep, i);
1092	v = be64_to_cpu(hptep[0]);
1093	r = be64_to_cpu(hptep[1]);
1094	if (r & HPTE_R_C) {
1095	hptep[1] = cpu_to_be64(r & ~HPTE_R_C);
1096	if (!(rev[i].guest_rpte & HPTE_R_C)) {
1097	rev[i].guest_rpte |= HPTE_R_C;
1098	note_hpte_modification(kvm, &rev[i]);
1099	}
1100	n = kvmppc_actual_pgsz(v, r);
1101	n = (n + PAGE_SIZE - 1) >> PAGE_SHIFT;
1102	if (n > npages_dirty)
1103	npages_dirty = n;
1104	eieio();
1105	}
1106	v &= ~HPTE_V_ABSENT;
1107	v |= HPTE_V_VALID;
1108	__unlock_hpte(hptep, v);
1109	} while ((i = j) != head);
1110
1111	unlock_rmap(rmapp);
1112	return npages_dirty;
1113	}
1114
1115	void kvmppc_harvest_vpa_dirty(struct kvmppc_vpa *vpa,
1116	struct kvm_memory_slot *memslot,
1117	unsigned long *map)
1118	{
1119	unsigned long gfn;
1120
1121	if (!vpa->dirty || !vpa->pinned_addr)
1122	return;
1123	gfn = vpa->gpa >> PAGE_SHIFT;
1124	if (gfn < memslot->base_gfn ||
1125	gfn >= memslot->base_gfn + memslot->npages)
1126	return;
1127
1128	vpa->dirty = false;
1129	if (map)
1130	__set_bit_le(nr: gfn - memslot->base_gfn, addr: map);
1131	}
1132
1133	long kvmppc_hv_get_dirty_log_hpt(struct kvm *kvm,
1134	struct kvm_memory_slot *memslot, unsigned long *map)
1135	{
1136	unsigned long i;
1137	unsigned long *rmapp;
1138
1139	preempt_disable();
1140	rmapp = memslot->arch.rmap;
1141	for (i = 0; i < memslot->npages; ++i) {
1142	int npages = kvm_test_clear_dirty_npages(kvm, rmapp);
1143	/*
1144	* Note that if npages > 0 then i must be a multiple of npages,
1145	* since we always put huge-page HPTEs in the rmap chain
1146	* corresponding to their page base address.
1147	*/
1148	if (npages)
1149	set_dirty_bits(map, i, npages);
1150	++rmapp;
1151	}
1152	preempt_enable();
1153	return 0;
1154	}
1155
1156	void *kvmppc_pin_guest_page(struct kvm *kvm, unsigned long gpa,
1157	unsigned long *nb_ret)
1158	{
1159	struct kvm_memory_slot *memslot;
1160	unsigned long gfn = gpa >> PAGE_SHIFT;
1161	struct page *page, *pages[1];
1162	int npages;
1163	unsigned long hva, offset;
1164	int srcu_idx;
1165
1166	srcu_idx = srcu_read_lock(ssp: &kvm->srcu);
1167	memslot = gfn_to_memslot(kvm, gfn);
1168	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
1169	goto err;
1170	hva = gfn_to_hva_memslot(slot: memslot, gfn);
1171	npages = get_user_pages_fast(start: hva, nr_pages: 1, gup_flags: FOLL_WRITE, pages);
1172	if (npages < 1)
1173	goto err;
1174	page = pages[0];
1175	srcu_read_unlock(ssp: &kvm->srcu, idx: srcu_idx);
1176
1177	offset = gpa & (PAGE_SIZE - 1);
1178	if (nb_ret)
1179	*nb_ret = PAGE_SIZE - offset;
1180	return page_address(page) + offset;
1181
1182	err:
1183	srcu_read_unlock(ssp: &kvm->srcu, idx: srcu_idx);
1184	return NULL;
1185	}
1186
1187	void kvmppc_unpin_guest_page(struct kvm *kvm, void *va, unsigned long gpa,
1188	bool dirty)
1189	{
1190	struct page *page = virt_to_page(va);
1191	struct kvm_memory_slot *memslot;
1192	unsigned long gfn;
1193	int srcu_idx;
1194
1195	put_page(page);
1196
1197	if (!dirty)
1198	return;
1199
1200	/* We need to mark this page dirty in the memslot dirty_bitmap, if any */
1201	gfn = gpa >> PAGE_SHIFT;
1202	srcu_idx = srcu_read_lock(ssp: &kvm->srcu);
1203	memslot = gfn_to_memslot(kvm, gfn);
1204	if (memslot && memslot->dirty_bitmap)
1205	set_bit_le(nr: gfn - memslot->base_gfn, addr: memslot->dirty_bitmap);
1206	srcu_read_unlock(ssp: &kvm->srcu, idx: srcu_idx);
1207	}
1208
1209	/*
1210	* HPT resizing
1211	*/
1212	static int resize_hpt_allocate(struct kvm_resize_hpt *resize)
1213	{
1214	int rc;
1215
1216	rc = kvmppc_allocate_hpt(info: &resize->hpt, order: resize->order);
1217	if (rc < 0)
1218	return rc;
1219
1220	resize_hpt_debug(resize, "%s(): HPT @ 0x%lx\n", __func__,
1221	resize->hpt.virt);
1222
1223	return 0;
1224	}
1225
1226	static unsigned long resize_hpt_rehash_hpte(struct kvm_resize_hpt *resize,
1227	unsigned long idx)
1228	{
1229	struct kvm *kvm = resize->kvm;
1230	struct kvm_hpt_info *old = &kvm->arch.hpt;
1231	struct kvm_hpt_info *new = &resize->hpt;
1232	unsigned long old_hash_mask = (1ULL << (old->order - 7)) - 1;
1233	unsigned long new_hash_mask = (1ULL << (new->order - 7)) - 1;
1234	__be64 *hptep, *new_hptep;
1235	unsigned long vpte, rpte, guest_rpte;
1236	int ret;
1237	struct revmap_entry *rev;
1238	unsigned long apsize, avpn, pteg, hash;
1239	unsigned long new_idx, new_pteg, replace_vpte;
1240	int pshift;
1241
1242	hptep = (__be64 *)(old->virt + (idx << 4));
1243
1244	/* Guest is stopped, so new HPTEs can't be added or faulted
1245	* in, only unmapped or altered by host actions. So, it's
1246	* safe to check this before we take the HPTE lock */
1247	vpte = be64_to_cpu(hptep[0]);
1248	if (!(vpte & HPTE_V_VALID) && !(vpte & HPTE_V_ABSENT))
1249	return 0; /* nothing to do */
1250
1251	while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
1252	cpu_relax();
1253
1254	vpte = be64_to_cpu(hptep[0]);
1255
1256	ret = 0;
1257	if (!(vpte & HPTE_V_VALID) && !(vpte & HPTE_V_ABSENT))
1258	/* Nothing to do */
1259	goto out;
1260
1261	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1262	rpte = be64_to_cpu(hptep[1]);
1263	vpte = hpte_new_to_old_v(vpte, rpte);
1264	}
1265
1266	/* Unmap */
1267	rev = &old->rev[idx];
1268	guest_rpte = rev->guest_rpte;
1269
1270	ret = -EIO;
1271	apsize = kvmppc_actual_pgsz(vpte, guest_rpte);
1272	if (!apsize)
1273	goto out;
1274
1275	if (vpte & HPTE_V_VALID) {
1276	unsigned long gfn = hpte_rpn(guest_rpte, apsize);
1277	int srcu_idx = srcu_read_lock(ssp: &kvm->srcu);
1278	struct kvm_memory_slot *memslot =
1279	__gfn_to_memslot(slots: kvm_memslots(kvm), gfn);
1280
1281	if (memslot) {
1282	unsigned long *rmapp;
1283	rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
1284
1285	lock_rmap(rmapp);
1286	kvmppc_unmap_hpte(kvm, i: idx, memslot, rmapp, gfn);
1287	unlock_rmap(rmapp);
1288	}
1289
1290	srcu_read_unlock(ssp: &kvm->srcu, idx: srcu_idx);
1291	}
1292
1293	/* Reload PTE after unmap */
1294	vpte = be64_to_cpu(hptep[0]);
1295	BUG_ON(vpte & HPTE_V_VALID);
1296	BUG_ON(!(vpte & HPTE_V_ABSENT));
1297
1298	ret = 0;
1299	if (!(vpte & HPTE_V_BOLTED))
1300	goto out;
1301
1302	rpte = be64_to_cpu(hptep[1]);
1303
1304	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1305	vpte = hpte_new_to_old_v(vpte, rpte);
1306	rpte = hpte_new_to_old_r(rpte);
1307	}
1308
1309	pshift = kvmppc_hpte_base_page_shift(vpte, rpte);
1310	avpn = HPTE_V_AVPN_VAL(vpte) & ~(((1ul << pshift) - 1) >> 23);
1311	pteg = idx / HPTES_PER_GROUP;
1312	if (vpte & HPTE_V_SECONDARY)
1313	pteg = ~pteg;
1314
1315	if (!(vpte & HPTE_V_1TB_SEG)) {
1316	unsigned long offset, vsid;
1317
1318	/* We only have 28 - 23 bits of offset in avpn */
1319	offset = (avpn & 0x1f) << 23;
1320	vsid = avpn >> 5;
1321	/* We can find more bits from the pteg value */
1322	if (pshift < 23)
1323	offset |= ((vsid ^ pteg) & old_hash_mask) << pshift;
1324
1325	hash = vsid ^ (offset >> pshift);
1326	} else {
1327	unsigned long offset, vsid;
1328
1329	/* We only have 40 - 23 bits of seg_off in avpn */
1330	offset = (avpn & 0x1ffff) << 23;
1331	vsid = avpn >> 17;
1332	if (pshift < 23)
1333	offset |= ((vsid ^ (vsid << 25) ^ pteg) & old_hash_mask) << pshift;
1334
1335	hash = vsid ^ (vsid << 25) ^ (offset >> pshift);
1336	}
1337
1338	new_pteg = hash & new_hash_mask;
1339	if (vpte & HPTE_V_SECONDARY)
1340	new_pteg = ~hash & new_hash_mask;
1341
1342	new_idx = new_pteg * HPTES_PER_GROUP + (idx % HPTES_PER_GROUP);
1343	new_hptep = (__be64 *)(new->virt + (new_idx << 4));
1344
1345	replace_vpte = be64_to_cpu(new_hptep[0]);
1346	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1347	unsigned long replace_rpte = be64_to_cpu(new_hptep[1]);
1348	replace_vpte = hpte_new_to_old_v(replace_vpte, replace_rpte);
1349	}
1350
1351	if (replace_vpte & (HPTE_V_VALID | HPTE_V_ABSENT)) {
1352	BUG_ON(new->order >= old->order);
1353
1354	if (replace_vpte & HPTE_V_BOLTED) {
1355	if (vpte & HPTE_V_BOLTED)
1356	/* Bolted collision, nothing we can do */
1357	ret = -ENOSPC;
1358	/* Discard the new HPTE */
1359	goto out;
1360	}
1361
1362	/* Discard the previous HPTE */
1363	}
1364
1365	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1366	rpte = hpte_old_to_new_r(vpte, rpte);
1367	vpte = hpte_old_to_new_v(vpte);
1368	}
1369
1370	new_hptep[1] = cpu_to_be64(rpte);
1371	new->rev[new_idx].guest_rpte = guest_rpte;
1372	/* No need for a barrier, since new HPT isn't active */
1373	new_hptep[0] = cpu_to_be64(vpte);
1374	unlock_hpte(new_hptep, vpte);
1375
1376	out:
1377	unlock_hpte(hptep, vpte);
1378	return ret;
1379	}
1380
1381	static int resize_hpt_rehash(struct kvm_resize_hpt *resize)
1382	{
1383	struct kvm *kvm = resize->kvm;
1384	unsigned long i;
1385	int rc;
1386
1387	for (i = 0; i < kvmppc_hpt_npte(&kvm->arch.hpt); i++) {
1388	rc = resize_hpt_rehash_hpte(resize, idx: i);
1389	if (rc != 0)
1390	return rc;
1391	}
1392
1393	return 0;
1394	}
1395
1396	static void resize_hpt_pivot(struct kvm_resize_hpt *resize)
1397	{
1398	struct kvm *kvm = resize->kvm;
1399	struct kvm_hpt_info hpt_tmp;
1400
1401	/* Exchange the pending tables in the resize structure with
1402	* the active tables */
1403
1404	resize_hpt_debug(resize, "resize_hpt_pivot()\n");
1405
1406	spin_lock(lock: &kvm->mmu_lock);
1407	asm volatile("ptesync" : : : "memory");
1408
1409	hpt_tmp = kvm->arch.hpt;
1410	kvmppc_set_hpt(kvm, info: &resize->hpt);
1411	resize->hpt = hpt_tmp;
1412
1413	spin_unlock(lock: &kvm->mmu_lock);
1414
1415	synchronize_srcu_expedited(ssp: &kvm->srcu);
1416
1417	if (cpu_has_feature(CPU_FTR_ARCH_300))
1418	kvmppc_setup_partition_table(kvm);
1419
1420	resize_hpt_debug(resize, "resize_hpt_pivot() done\n");
1421	}
1422
1423	static void resize_hpt_release(struct kvm *kvm, struct kvm_resize_hpt *resize)
1424	{
1425	if (WARN_ON(!mutex_is_locked(&kvm->arch.mmu_setup_lock)))
1426	return;
1427
1428	if (!resize)
1429	return;
1430
1431	if (resize->error != -EBUSY) {
1432	if (resize->hpt.virt)
1433	kvmppc_free_hpt(&resize->hpt);
1434	kfree(objp: resize);
1435	}
1436
1437	if (kvm->arch.resize_hpt == resize)
1438	kvm->arch.resize_hpt = NULL;
1439	}
1440
1441	static void resize_hpt_prepare_work(struct work_struct *work)
1442	{
1443	struct kvm_resize_hpt *resize = container_of(work,
1444	struct kvm_resize_hpt,
1445	work);
1446	struct kvm *kvm = resize->kvm;
1447	int err = 0;
1448
1449	if (WARN_ON(resize->error != -EBUSY))
1450	return;
1451
1452	mutex_lock(&kvm->arch.mmu_setup_lock);
1453
1454	/* Request is still current? */
1455	if (kvm->arch.resize_hpt == resize) {
1456	/* We may request large allocations here:
1457	* do not sleep with kvm->arch.mmu_setup_lock held for a while.
1458	*/
1459	mutex_unlock(lock: &kvm->arch.mmu_setup_lock);
1460
1461	resize_hpt_debug(resize, "%s(): order = %d\n", __func__,
1462	resize->order);
1463
1464	err = resize_hpt_allocate(resize);
1465
1466	/* We have strict assumption about -EBUSY
1467	* when preparing for HPT resize.
1468	*/
1469	if (WARN_ON(err == -EBUSY))
1470	err = -EINPROGRESS;
1471
1472	mutex_lock(&kvm->arch.mmu_setup_lock);
1473	/* It is possible that kvm->arch.resize_hpt != resize
1474	* after we grab kvm->arch.mmu_setup_lock again.
1475	*/
1476	}
1477
1478	resize->error = err;
1479
1480	if (kvm->arch.resize_hpt != resize)
1481	resize_hpt_release(kvm, resize);
1482
1483	mutex_unlock(lock: &kvm->arch.mmu_setup_lock);
1484	}
1485
1486	int kvm_vm_ioctl_resize_hpt_prepare(struct kvm *kvm,
1487	struct kvm_ppc_resize_hpt *rhpt)
1488	{
1489	unsigned long flags = rhpt->flags;
1490	unsigned long shift = rhpt->shift;
1491	struct kvm_resize_hpt *resize;
1492	int ret;
1493
1494	if (flags != 0 || kvm_is_radix(kvm))
1495	return -EINVAL;
1496
1497	if (shift && ((shift < 18) || (shift > 46)))
1498	return -EINVAL;
1499
1500	mutex_lock(&kvm->arch.mmu_setup_lock);
1501
1502	resize = kvm->arch.resize_hpt;
1503
1504	if (resize) {
1505	if (resize->order == shift) {
1506	/* Suitable resize in progress? */
1507	ret = resize->error;
1508	if (ret == -EBUSY)
1509	ret = 100; /* estimated time in ms */
1510	else if (ret)
1511	resize_hpt_release(kvm, resize);
1512
1513	goto out;
1514	}
1515
1516	/* not suitable, cancel it */
1517	resize_hpt_release(kvm, resize);
1518	}
1519
1520	ret = 0;
1521	if (!shift)
1522	goto out; /* nothing to do */
1523
1524	/* start new resize */
1525
1526	resize = kzalloc(size: sizeof(*resize), GFP_KERNEL);
1527	if (!resize) {
1528	ret = -ENOMEM;
1529	goto out;
1530	}
1531
1532	resize->error = -EBUSY;
1533	resize->order = shift;
1534	resize->kvm = kvm;
1535	INIT_WORK(&resize->work, resize_hpt_prepare_work);
1536	kvm->arch.resize_hpt = resize;
1537
1538	schedule_work(work: &resize->work);
1539
1540	ret = 100; /* estimated time in ms */
1541
1542	out:
1543	mutex_unlock(lock: &kvm->arch.mmu_setup_lock);
1544	return ret;
1545	}
1546
1547	static void resize_hpt_boot_vcpu(void *opaque)
1548	{
1549	/* Nothing to do, just force a KVM exit */
1550	}
1551
1552	int kvm_vm_ioctl_resize_hpt_commit(struct kvm *kvm,
1553	struct kvm_ppc_resize_hpt *rhpt)
1554	{
1555	unsigned long flags = rhpt->flags;
1556	unsigned long shift = rhpt->shift;
1557	struct kvm_resize_hpt *resize;
1558	int ret;
1559
1560	if (flags != 0 || kvm_is_radix(kvm))
1561	return -EINVAL;
1562
1563	if (shift && ((shift < 18) || (shift > 46)))
1564	return -EINVAL;
1565
1566	mutex_lock(&kvm->arch.mmu_setup_lock);
1567
1568	resize = kvm->arch.resize_hpt;
1569
1570	/* This shouldn't be possible */
1571	ret = -EIO;
1572	if (WARN_ON(!kvm->arch.mmu_ready))
1573	goto out_no_hpt;
1574
1575	/* Stop VCPUs from running while we mess with the HPT */
1576	kvm->arch.mmu_ready = 0;
1577	smp_mb();
1578
1579	/* Boot all CPUs out of the guest so they re-read
1580	* mmu_ready */
1581	on_each_cpu(func: resize_hpt_boot_vcpu, NULL, wait: 1);
1582
1583	ret = -ENXIO;
1584	if (!resize || (resize->order != shift))
1585	goto out;
1586
1587	ret = resize->error;
1588	if (ret)
1589	goto out;
1590
1591	ret = resize_hpt_rehash(resize);
1592	if (ret)
1593	goto out;
1594
1595	resize_hpt_pivot(resize);
1596
1597	out:
1598	/* Let VCPUs run again */
1599	kvm->arch.mmu_ready = 1;
1600	smp_mb();
1601	out_no_hpt:
1602	resize_hpt_release(kvm, resize);
1603	mutex_unlock(lock: &kvm->arch.mmu_setup_lock);
1604	return ret;
1605	}
1606
1607	/*
1608	* Functions for reading and writing the hash table via reads and
1609	* writes on a file descriptor.
1610	*
1611	* Reads return the guest view of the hash table, which has to be
1612	* pieced together from the real hash table and the guest_rpte
1613	* values in the revmap array.
1614	*
1615	* On writes, each HPTE written is considered in turn, and if it
1616	* is valid, it is written to the HPT as if an H_ENTER with the
1617	* exact flag set was done. When the invalid count is non-zero
1618	* in the header written to the stream, the kernel will make
1619	* sure that that many HPTEs are invalid, and invalidate them
1620	* if not.
1621	*/
1622
1623	struct kvm_htab_ctx {
1624	unsigned long index;
1625	unsigned long flags;
1626	struct kvm *kvm;
1627	int first_pass;
1628	};
1629
1630	#define HPTE_SIZE (2 * sizeof(unsigned long))
1631
1632	/*
1633	* Returns 1 if this HPT entry has been modified or has pending
1634	* R/C bit changes.
1635	*/
1636	static int hpte_dirty(struct revmap_entry *revp, __be64 *hptp)
1637	{
1638	unsigned long rcbits_unset;
1639
1640	if (revp->guest_rpte & HPTE_GR_MODIFIED)
1641	return 1;
1642
1643	/* Also need to consider changes in reference and changed bits */
1644	rcbits_unset = ~revp->guest_rpte & (HPTE_R_R | HPTE_R_C);
1645	if ((be64_to_cpu(hptp[0]) & HPTE_V_VALID) &&
1646	(be64_to_cpu(hptp[1]) & rcbits_unset))
1647	return 1;
1648
1649	return 0;
1650	}
1651
1652	static long record_hpte(unsigned long flags, __be64 *hptp,
1653	unsigned long *hpte, struct revmap_entry *revp,
1654	int want_valid, int first_pass)
1655	{
1656	unsigned long v, r, hr;
1657	unsigned long rcbits_unset;
1658	int ok = 1;
1659	int valid, dirty;
1660
1661	/* Unmodified entries are uninteresting except on the first pass */
1662	dirty = hpte_dirty(revp, hptp);
1663	if (!first_pass && !dirty)
1664	return 0;
1665
1666	valid = 0;
1667	if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT)) {
1668	valid = 1;
1669	if ((flags & KVM_GET_HTAB_BOLTED_ONLY) &&
1670	!(be64_to_cpu(hptp[0]) & HPTE_V_BOLTED))
1671	valid = 0;
1672	}
1673	if (valid != want_valid)
1674	return 0;
1675
1676	v = r = 0;
1677	if (valid || dirty) {
1678	/* lock the HPTE so it's stable and read it */
1679	preempt_disable();
1680	while (!try_lock_hpte(hptp, HPTE_V_HVLOCK))
1681	cpu_relax();
1682	v = be64_to_cpu(hptp[0]);
1683	hr = be64_to_cpu(hptp[1]);
1684	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1685	v = hpte_new_to_old_v(v, hr);
1686	hr = hpte_new_to_old_r(hr);
1687	}
1688
1689	/* re-evaluate valid and dirty from synchronized HPTE value */
1690	valid = !!(v & HPTE_V_VALID);
1691	dirty = !!(revp->guest_rpte & HPTE_GR_MODIFIED);
1692
1693	/* Harvest R and C into guest view if necessary */
1694	rcbits_unset = ~revp->guest_rpte & (HPTE_R_R | HPTE_R_C);
1695	if (valid && (rcbits_unset & hr)) {
1696	revp->guest_rpte |= (hr &
1697	(HPTE_R_R | HPTE_R_C)) | HPTE_GR_MODIFIED;
1698	dirty = 1;
1699	}
1700
1701	if (v & HPTE_V_ABSENT) {
1702	v &= ~HPTE_V_ABSENT;
1703	v |= HPTE_V_VALID;
1704	valid = 1;
1705	}
1706	if ((flags & KVM_GET_HTAB_BOLTED_ONLY) && !(v & HPTE_V_BOLTED))
1707	valid = 0;
1708
1709	r = revp->guest_rpte;
1710	/* only clear modified if this is the right sort of entry */
1711	if (valid == want_valid && dirty) {
1712	r &= ~HPTE_GR_MODIFIED;
1713	revp->guest_rpte = r;
1714	}
1715	unlock_hpte(hptp, be64_to_cpu(hptp[0]));
1716	preempt_enable();
1717	if (!(valid == want_valid && (first_pass || dirty)))
1718	ok = 0;
1719	}
1720	hpte[0] = cpu_to_be64(v);
1721	hpte[1] = cpu_to_be64(r);
1722	return ok;
1723	}
1724
1725	static ssize_t kvm_htab_read(struct file *file, char __user *buf,
1726	size_t count, loff_t *ppos)
1727	{
1728	struct kvm_htab_ctx *ctx = file->private_data;
1729	struct kvm *kvm = ctx->kvm;
1730	struct kvm_get_htab_header hdr;
1731	__be64 *hptp;
1732	struct revmap_entry *revp;
1733	unsigned long i, nb, nw;
1734	unsigned long __user *lbuf;
1735	struct kvm_get_htab_header __user *hptr;
1736	unsigned long flags;
1737	int first_pass;
1738	unsigned long hpte[2];
1739
1740	if (!access_ok(buf, count))
1741	return -EFAULT;
1742	if (kvm_is_radix(kvm))
1743	return 0;
1744
1745	first_pass = ctx->first_pass;
1746	flags = ctx->flags;
1747
1748	i = ctx->index;
1749	hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE));
1750	revp = kvm->arch.hpt.rev + i;
1751	lbuf = (unsigned long __user *)buf;
1752
1753	nb = 0;
1754	while (nb + sizeof(hdr) + HPTE_SIZE < count) {
1755	/* Initialize header */
1756	hptr = (struct kvm_get_htab_header __user *)buf;
1757	hdr.n_valid = 0;
1758	hdr.n_invalid = 0;
1759	nw = nb;
1760	nb += sizeof(hdr);
1761	lbuf = (unsigned long __user *)(buf + sizeof(hdr));
1762
1763	/* Skip uninteresting entries, i.e. clean on not-first pass */
1764	if (!first_pass) {
1765	while (i < kvmppc_hpt_npte(&kvm->arch.hpt) &&
1766	!hpte_dirty(revp, hptp)) {
1767	++i;
1768	hptp += 2;
1769	++revp;
1770	}
1771	}
1772	hdr.index = i;
1773
1774	/* Grab a series of valid entries */
1775	while (i < kvmppc_hpt_npte(&kvm->arch.hpt) &&
1776	hdr.n_valid < 0xffff &&
1777	nb + HPTE_SIZE < count &&
1778	record_hpte(flags, hptp, hpte, revp, want_valid: 1, first_pass)) {
1779	/* valid entry, write it out */
1780	++hdr.n_valid;
1781	if (__put_user(hpte[0], lbuf) ||
1782	__put_user(hpte[1], lbuf + 1))
1783	return -EFAULT;
1784	nb += HPTE_SIZE;
1785	lbuf += 2;
1786	++i;
1787	hptp += 2;
1788	++revp;
1789	}
1790	/* Now skip invalid entries while we can */
1791	while (i < kvmppc_hpt_npte(&kvm->arch.hpt) &&
1792	hdr.n_invalid < 0xffff &&
1793	record_hpte(flags, hptp, hpte, revp, want_valid: 0, first_pass)) {
1794	/* found an invalid entry */
1795	++hdr.n_invalid;
1796	++i;
1797	hptp += 2;
1798	++revp;
1799	}
1800
1801	if (hdr.n_valid || hdr.n_invalid) {
1802	/* write back the header */
1803	if (__copy_to_user(to: hptr, from: &hdr, n: sizeof(hdr)))
1804	return -EFAULT;
1805	nw = nb;
1806	buf = (char __user *)lbuf;
1807	} else {
1808	nb = nw;
1809	}
1810
1811	/* Check if we've wrapped around the hash table */
1812	if (i >= kvmppc_hpt_npte(&kvm->arch.hpt)) {
1813	i = 0;
1814	ctx->first_pass = 0;
1815	break;
1816	}
1817	}
1818
1819	ctx->index = i;
1820
1821	return nb;
1822	}
1823
1824	static ssize_t kvm_htab_write(struct file *file, const char __user *buf,
1825	size_t count, loff_t *ppos)
1826	{
1827	struct kvm_htab_ctx *ctx = file->private_data;
1828	struct kvm *kvm = ctx->kvm;
1829	struct kvm_get_htab_header hdr;
1830	unsigned long i, j;
1831	unsigned long v, r;
1832	unsigned long __user *lbuf;
1833	__be64 *hptp;
1834	unsigned long tmp[2];
1835	ssize_t nb;
1836	long int err, ret;
1837	int mmu_ready;
1838	int pshift;
1839
1840	if (!access_ok(buf, count))
1841	return -EFAULT;
1842	if (kvm_is_radix(kvm))
1843	return -EINVAL;
1844
1845	/* lock out vcpus from running while we're doing this */
1846	mutex_lock(&kvm->arch.mmu_setup_lock);
1847	mmu_ready = kvm->arch.mmu_ready;
1848	if (mmu_ready) {
1849	kvm->arch.mmu_ready = 0; /* temporarily */
1850	/* order mmu_ready vs. vcpus_running */
1851	smp_mb();
1852	if (atomic_read(v: &kvm->arch.vcpus_running)) {
1853	kvm->arch.mmu_ready = 1;
1854	mutex_unlock(lock: &kvm->arch.mmu_setup_lock);
1855	return -EBUSY;
1856	}
1857	}
1858
1859	err = 0;
1860	for (nb = 0; nb + sizeof(hdr) <= count; ) {
1861	err = -EFAULT;
1862	if (__copy_from_user(to: &hdr, from: buf, n: sizeof(hdr)))
1863	break;
1864
1865	err = 0;
1866	if (nb + hdr.n_valid * HPTE_SIZE > count)
1867	break;
1868
1869	nb += sizeof(hdr);
1870	buf += sizeof(hdr);
1871
1872	err = -EINVAL;
1873	i = hdr.index;
1874	if (i >= kvmppc_hpt_npte(&kvm->arch.hpt) ||
1875	i + hdr.n_valid + hdr.n_invalid > kvmppc_hpt_npte(&kvm->arch.hpt))
1876	break;
1877
1878	hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE));
1879	lbuf = (unsigned long __user *)buf;
1880	for (j = 0; j < hdr.n_valid; ++j) {
1881	__be64 hpte_v;
1882	__be64 hpte_r;
1883
1884	err = -EFAULT;
1885	if (__get_user(hpte_v, lbuf) ||
1886	__get_user(hpte_r, lbuf + 1))
1887	goto out;
1888	v = be64_to_cpu(hpte_v);
1889	r = be64_to_cpu(hpte_r);
1890	err = -EINVAL;
1891	if (!(v & HPTE_V_VALID))
1892	goto out;
1893	pshift = kvmppc_hpte_base_page_shift(v, r);
1894	if (pshift <= 0)
1895	goto out;
1896	lbuf += 2;
1897	nb += HPTE_SIZE;
1898
1899	if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT))
1900	kvmppc_do_h_remove(kvm, 0, i, 0, tmp);
1901	err = -EIO;
1902	ret = kvmppc_virtmode_do_h_enter(kvm, H_EXACT, i, v, r,
1903	tmp);
1904	if (ret != H_SUCCESS) {
1905	pr_err("%s ret %ld i=%ld v=%lx r=%lx\n", __func__, ret, i, v, r);
1906	goto out;
1907	}
1908	if (!mmu_ready && is_vrma_hpte(v)) {
1909	unsigned long senc, lpcr;
1910
1911	senc = slb_pgsize_encoding(1ul << pshift);
1912	kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
1913	(VRMA_VSID << SLB_VSID_SHIFT_1T);
1914	if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
1915	lpcr = senc << (LPCR_VRMASD_SH - 4);
1916	kvmppc_update_lpcr(kvm, lpcr,
1917	LPCR_VRMASD);
1918	} else {
1919	kvmppc_setup_partition_table(kvm);
1920	}
1921	mmu_ready = 1;
1922	}
1923	++i;
1924	hptp += 2;
1925	}
1926
1927	for (j = 0; j < hdr.n_invalid; ++j) {
1928	if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT))
1929	kvmppc_do_h_remove(kvm, 0, i, 0, tmp);
1930	++i;
1931	hptp += 2;
1932	}
1933	err = 0;
1934	}
1935
1936	out:
1937	/* Order HPTE updates vs. mmu_ready */
1938	smp_wmb();
1939	kvm->arch.mmu_ready = mmu_ready;
1940	mutex_unlock(lock: &kvm->arch.mmu_setup_lock);
1941
1942	if (err)
1943	return err;
1944	return nb;
1945	}
1946
1947	static int kvm_htab_release(struct inode *inode, struct file *filp)
1948	{
1949	struct kvm_htab_ctx *ctx = filp->private_data;
1950
1951	filp->private_data = NULL;
1952	if (!(ctx->flags & KVM_GET_HTAB_WRITE))
1953	atomic_dec(v: &ctx->kvm->arch.hpte_mod_interest);
1954	kvm_put_kvm(kvm: ctx->kvm);
1955	kfree(objp: ctx);
1956	return 0;
1957	}
1958
1959	static const struct file_operations kvm_htab_fops = {
1960	.read = kvm_htab_read,
1961	.write = kvm_htab_write,
1962	.llseek = default_llseek,
1963	.release = kvm_htab_release,
1964	};
1965
1966	int kvm_vm_ioctl_get_htab_fd(struct kvm *kvm, struct kvm_get_htab_fd *ghf)
1967	{
1968	int ret;
1969	struct kvm_htab_ctx *ctx;
1970	int rwflag;
1971
1972	/* reject flags we don't recognize */
1973	if (ghf->flags & ~(KVM_GET_HTAB_BOLTED_ONLY | KVM_GET_HTAB_WRITE))
1974	return -EINVAL;
1975	ctx = kzalloc(size: sizeof(*ctx), GFP_KERNEL);
1976	if (!ctx)
1977	return -ENOMEM;
1978	kvm_get_kvm(kvm);
1979	ctx->kvm = kvm;
1980	ctx->index = ghf->start_index;
1981	ctx->flags = ghf->flags;
1982	ctx->first_pass = 1;
1983
1984	rwflag = (ghf->flags & KVM_GET_HTAB_WRITE) ? O_WRONLY : O_RDONLY;
1985	ret = anon_inode_getfd(name: "kvm-htab", fops: &kvm_htab_fops, priv: ctx, flags: rwflag | O_CLOEXEC);
1986	if (ret < 0) {
1987	kfree(objp: ctx);
1988	kvm_put_kvm_no_destroy(kvm);
1989	return ret;
1990	}
1991
1992	if (rwflag == O_RDONLY) {
1993	mutex_lock(&kvm->slots_lock);
1994	atomic_inc(v: &kvm->arch.hpte_mod_interest);
1995	/* make sure kvmppc_do_h_enter etc. see the increment */
1996	synchronize_srcu_expedited(ssp: &kvm->srcu);
1997	mutex_unlock(lock: &kvm->slots_lock);
1998	}
1999
2000	return ret;
2001	}
2002
2003	struct debugfs_htab_state {
2004	struct kvm *kvm;
2005	struct mutex mutex;
2006	unsigned long hpt_index;
2007	int chars_left;
2008	int buf_index;
2009	char buf[64];
2010	};
2011
2012	static int debugfs_htab_open(struct inode *inode, struct file *file)
2013	{
2014	struct kvm *kvm = inode->i_private;
2015	struct debugfs_htab_state *p;
2016
2017	p = kzalloc(size: sizeof(*p), GFP_KERNEL);
2018	if (!p)
2019	return -ENOMEM;
2020
2021	kvm_get_kvm(kvm);
2022	p->kvm = kvm;
2023	mutex_init(&p->mutex);
2024	file->private_data = p;
2025
2026	return nonseekable_open(inode, filp: file);
2027	}
2028
2029	static int debugfs_htab_release(struct inode *inode, struct file *file)
2030	{
2031	struct debugfs_htab_state *p = file->private_data;
2032
2033	kvm_put_kvm(kvm: p->kvm);
2034	kfree(objp: p);
2035	return 0;
2036	}
2037
2038	static ssize_t debugfs_htab_read(struct file *file, char __user *buf,
2039	size_t len, loff_t *ppos)
2040	{
2041	struct debugfs_htab_state *p = file->private_data;
2042	ssize_t ret, r;
2043	unsigned long i, n;
2044	unsigned long v, hr, gr;
2045	struct kvm *kvm;
2046	__be64 *hptp;
2047
2048	kvm = p->kvm;
2049	if (kvm_is_radix(kvm))
2050	return 0;
2051
2052	ret = mutex_lock_interruptible(&p->mutex);
2053	if (ret)
2054	return ret;
2055
2056	if (p->chars_left) {
2057	n = p->chars_left;
2058	if (n > len)
2059	n = len;
2060	r = copy_to_user(to: buf, from: p->buf + p->buf_index, n);
2061	n -= r;
2062	p->chars_left -= n;
2063	p->buf_index += n;
2064	buf += n;
2065	len -= n;
2066	ret = n;
2067	if (r) {
2068	if (!n)
2069	ret = -EFAULT;
2070	goto out;
2071	}
2072	}
2073
2074	i = p->hpt_index;
2075	hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE));
2076	for (; len != 0 && i < kvmppc_hpt_npte(&kvm->arch.hpt);
2077	++i, hptp += 2) {
2078	if (!(be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT)))
2079	continue;
2080
2081	/* lock the HPTE so it's stable and read it */
2082	preempt_disable();
2083	while (!try_lock_hpte(hptp, HPTE_V_HVLOCK))
2084	cpu_relax();
2085	v = be64_to_cpu(hptp[0]) & ~HPTE_V_HVLOCK;
2086	hr = be64_to_cpu(hptp[1]);
2087	gr = kvm->arch.hpt.rev[i].guest_rpte;
2088	unlock_hpte(hptp, v);
2089	preempt_enable();
2090
2091	if (!(v & (HPTE_V_VALID | HPTE_V_ABSENT)))
2092	continue;
2093
2094	n = scnprintf(buf: p->buf, size: sizeof(p->buf),
2095	fmt: "%6lx %.16lx %.16lx %.16lx\n",
2096	i, v, hr, gr);
2097	p->chars_left = n;
2098	if (n > len)
2099	n = len;
2100	r = copy_to_user(to: buf, from: p->buf, n);
2101	n -= r;
2102	p->chars_left -= n;
2103	p->buf_index = n;
2104	buf += n;
2105	len -= n;
2106	ret += n;
2107	if (r) {
2108	if (!ret)
2109	ret = -EFAULT;
2110	goto out;
2111	}
2112	}
2113	p->hpt_index = i;
2114
2115	out:
2116	mutex_unlock(lock: &p->mutex);
2117	return ret;
2118	}
2119
2120	static ssize_t debugfs_htab_write(struct file *file, const char __user *buf,
2121	size_t len, loff_t *ppos)
2122	{
2123	return -EACCES;
2124	}
2125
2126	static const struct file_operations debugfs_htab_fops = {
2127	.owner = THIS_MODULE,
2128	.open = debugfs_htab_open,
2129	.release = debugfs_htab_release,
2130	.read = debugfs_htab_read,
2131	.write = debugfs_htab_write,
2132	.llseek = generic_file_llseek,
2133	};
2134
2135	void kvmppc_mmu_debugfs_init(struct kvm *kvm)
2136	{
2137	debugfs_create_file(name: "htab", mode: 0400, parent: kvm->debugfs_dentry, data: kvm,
2138	fops: &debugfs_htab_fops);
2139	}
2140
2141	void kvmppc_mmu_book3s_hv_init(struct kvm_vcpu *vcpu)
2142	{
2143	struct kvmppc_mmu *mmu = &vcpu->arch.mmu;
2144
2145	vcpu->arch.slb_nr = 32; /* POWER7/POWER8 */
2146
2147	mmu->xlate = kvmppc_mmu_book3s_64_hv_xlate;
2148
2149	vcpu->arch.hflags |= BOOK3S_HFLAG_SLB;
2150	}
2151