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 
28 #include <asm/tlbflush.h>
29 #include <asm/kvm_ppc.h>
30 #include <asm/kvm_book3s.h>
31 #include <asm/mmu-hash64.h>
32 #include <asm/hvcall.h>
33 #include <asm/synch.h>
34 #include <asm/ppc-opcode.h>
35 #include <asm/cputable.h>
36 
37 /* POWER7 has 10-bit LPIDs, PPC970 has 6-bit LPIDs */
38 #define MAX_LPID_970	63
39 #define NR_LPIDS	(LPID_RSVD + 1)
40 unsigned long lpid_inuse[BITS_TO_LONGS(NR_LPIDS)];
41 
kvmppc_alloc_hpt(struct kvm * kvm)42 long kvmppc_alloc_hpt(struct kvm *kvm)
43 {
44 	unsigned long hpt;
45 	unsigned long lpid;
46 	struct revmap_entry *rev;
47 	struct kvmppc_linear_info *li;
48 
49 	/* Allocate guest's hashed page table */
50 	li = kvm_alloc_hpt();
51 	if (li) {
52 		/* using preallocated memory */
53 		hpt = (ulong)li->base_virt;
54 		kvm->arch.hpt_li = li;
55 	} else {
56 		/* using dynamic memory */
57 		hpt = __get_free_pages(GFP_KERNEL|__GFP_ZERO|__GFP_REPEAT|
58 				       __GFP_NOWARN, HPT_ORDER - PAGE_SHIFT);
59 	}
60 
61 	if (!hpt) {
62 		pr_err("kvm_alloc_hpt: Couldn't alloc HPT\n");
63 		return -ENOMEM;
64 	}
65 	kvm->arch.hpt_virt = hpt;
66 
67 	/* Allocate reverse map array */
68 	rev = vmalloc(sizeof(struct revmap_entry) * HPT_NPTE);
69 	if (!rev) {
70 		pr_err("kvmppc_alloc_hpt: Couldn't alloc reverse map array\n");
71 		goto out_freehpt;
72 	}
73 	kvm->arch.revmap = rev;
74 
75 	/* Allocate the guest's logical partition ID */
76 	do {
77 		lpid = find_first_zero_bit(lpid_inuse, NR_LPIDS);
78 		if (lpid >= NR_LPIDS) {
79 			pr_err("kvm_alloc_hpt: No LPIDs free\n");
80 			goto out_freeboth;
81 		}
82 	} while (test_and_set_bit(lpid, lpid_inuse));
83 
84 	kvm->arch.sdr1 = __pa(hpt) | (HPT_ORDER - 18);
85 	kvm->arch.lpid = lpid;
86 
87 	pr_info("KVM guest htab at %lx, LPID %lx\n", hpt, lpid);
88 	return 0;
89 
90  out_freeboth:
91 	vfree(rev);
92  out_freehpt:
93 	free_pages(hpt, HPT_ORDER - PAGE_SHIFT);
94 	return -ENOMEM;
95 }
96 
kvmppc_free_hpt(struct kvm * kvm)97 void kvmppc_free_hpt(struct kvm *kvm)
98 {
99 	clear_bit(kvm->arch.lpid, lpid_inuse);
100 	vfree(kvm->arch.revmap);
101 	if (kvm->arch.hpt_li)
102 		kvm_release_hpt(kvm->arch.hpt_li);
103 	else
104 		free_pages(kvm->arch.hpt_virt, HPT_ORDER - PAGE_SHIFT);
105 }
106 
107 /* Bits in first HPTE dword for pagesize 4k, 64k or 16M */
hpte0_pgsize_encoding(unsigned long pgsize)108 static inline unsigned long hpte0_pgsize_encoding(unsigned long pgsize)
109 {
110 	return (pgsize > 0x1000) ? HPTE_V_LARGE : 0;
111 }
112 
113 /* Bits in second HPTE dword for pagesize 4k, 64k or 16M */
hpte1_pgsize_encoding(unsigned long pgsize)114 static inline unsigned long hpte1_pgsize_encoding(unsigned long pgsize)
115 {
116 	return (pgsize == 0x10000) ? 0x1000 : 0;
117 }
118 
kvmppc_map_vrma(struct kvm_vcpu * vcpu,struct kvm_memory_slot * memslot,unsigned long porder)119 void kvmppc_map_vrma(struct kvm_vcpu *vcpu, struct kvm_memory_slot *memslot,
120 		     unsigned long porder)
121 {
122 	unsigned long i;
123 	unsigned long npages;
124 	unsigned long hp_v, hp_r;
125 	unsigned long addr, hash;
126 	unsigned long psize;
127 	unsigned long hp0, hp1;
128 	long ret;
129 
130 	psize = 1ul << porder;
131 	npages = memslot->npages >> (porder - PAGE_SHIFT);
132 
133 	/* VRMA can't be > 1TB */
134 	if (npages > 1ul << (40 - porder))
135 		npages = 1ul << (40 - porder);
136 	/* Can't use more than 1 HPTE per HPTEG */
137 	if (npages > HPT_NPTEG)
138 		npages = HPT_NPTEG;
139 
140 	hp0 = HPTE_V_1TB_SEG | (VRMA_VSID << (40 - 16)) |
141 		HPTE_V_BOLTED | hpte0_pgsize_encoding(psize);
142 	hp1 = hpte1_pgsize_encoding(psize) |
143 		HPTE_R_R | HPTE_R_C | HPTE_R_M | PP_RWXX;
144 
145 	for (i = 0; i < npages; ++i) {
146 		addr = i << porder;
147 		/* can't use hpt_hash since va > 64 bits */
148 		hash = (i ^ (VRMA_VSID ^ (VRMA_VSID << 25))) & HPT_HASH_MASK;
149 		/*
150 		 * We assume that the hash table is empty and no
151 		 * vcpus are using it at this stage.  Since we create
152 		 * at most one HPTE per HPTEG, we just assume entry 7
153 		 * is available and use it.
154 		 */
155 		hash = (hash << 3) + 7;
156 		hp_v = hp0 | ((addr >> 16) & ~0x7fUL);
157 		hp_r = hp1 | addr;
158 		ret = kvmppc_virtmode_h_enter(vcpu, H_EXACT, hash, hp_v, hp_r);
159 		if (ret != H_SUCCESS) {
160 			pr_err("KVM: map_vrma at %lx failed, ret=%ld\n",
161 			       addr, ret);
162 			break;
163 		}
164 	}
165 }
166 
kvmppc_mmu_hv_init(void)167 int kvmppc_mmu_hv_init(void)
168 {
169 	unsigned long host_lpid, rsvd_lpid;
170 
171 	if (!cpu_has_feature(CPU_FTR_HVMODE))
172 		return -EINVAL;
173 
174 	memset(lpid_inuse, 0, sizeof(lpid_inuse));
175 
176 	if (cpu_has_feature(CPU_FTR_ARCH_206)) {
177 		host_lpid = mfspr(SPRN_LPID);	/* POWER7 */
178 		rsvd_lpid = LPID_RSVD;
179 	} else {
180 		host_lpid = 0;			/* PPC970 */
181 		rsvd_lpid = MAX_LPID_970;
182 	}
183 
184 	set_bit(host_lpid, lpid_inuse);
185 	/* rsvd_lpid is reserved for use in partition switching */
186 	set_bit(rsvd_lpid, lpid_inuse);
187 
188 	return 0;
189 }
190 
kvmppc_mmu_destroy(struct kvm_vcpu * vcpu)191 void kvmppc_mmu_destroy(struct kvm_vcpu *vcpu)
192 {
193 }
194 
kvmppc_mmu_book3s_64_hv_reset_msr(struct kvm_vcpu * vcpu)195 static void kvmppc_mmu_book3s_64_hv_reset_msr(struct kvm_vcpu *vcpu)
196 {
197 	kvmppc_set_msr(vcpu, MSR_SF | MSR_ME);
198 }
199 
200 /*
201  * This is called to get a reference to a guest page if there isn't
202  * one already in the kvm->arch.slot_phys[][] arrays.
203  */
kvmppc_get_guest_page(struct kvm * kvm,unsigned long gfn,struct kvm_memory_slot * memslot,unsigned long psize)204 static long kvmppc_get_guest_page(struct kvm *kvm, unsigned long gfn,
205 				  struct kvm_memory_slot *memslot,
206 				  unsigned long psize)
207 {
208 	unsigned long start;
209 	long np, err;
210 	struct page *page, *hpage, *pages[1];
211 	unsigned long s, pgsize;
212 	unsigned long *physp;
213 	unsigned int is_io, got, pgorder;
214 	struct vm_area_struct *vma;
215 	unsigned long pfn, i, npages;
216 
217 	physp = kvm->arch.slot_phys[memslot->id];
218 	if (!physp)
219 		return -EINVAL;
220 	if (physp[gfn - memslot->base_gfn])
221 		return 0;
222 
223 	is_io = 0;
224 	got = 0;
225 	page = NULL;
226 	pgsize = psize;
227 	err = -EINVAL;
228 	start = gfn_to_hva_memslot(memslot, gfn);
229 
230 	/* Instantiate and get the page we want access to */
231 	np = get_user_pages_fast(start, 1, 1, pages);
232 	if (np != 1) {
233 		/* Look up the vma for the page */
234 		down_read(&current->mm->mmap_sem);
235 		vma = find_vma(current->mm, start);
236 		if (!vma || vma->vm_start > start ||
237 		    start + psize > vma->vm_end ||
238 		    !(vma->vm_flags & VM_PFNMAP))
239 			goto up_err;
240 		is_io = hpte_cache_bits(pgprot_val(vma->vm_page_prot));
241 		pfn = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
242 		/* check alignment of pfn vs. requested page size */
243 		if (psize > PAGE_SIZE && (pfn & ((psize >> PAGE_SHIFT) - 1)))
244 			goto up_err;
245 		up_read(&current->mm->mmap_sem);
246 
247 	} else {
248 		page = pages[0];
249 		got = KVMPPC_GOT_PAGE;
250 
251 		/* See if this is a large page */
252 		s = PAGE_SIZE;
253 		if (PageHuge(page)) {
254 			hpage = compound_head(page);
255 			s <<= compound_order(hpage);
256 			/* Get the whole large page if slot alignment is ok */
257 			if (s > psize && slot_is_aligned(memslot, s) &&
258 			    !(memslot->userspace_addr & (s - 1))) {
259 				start &= ~(s - 1);
260 				pgsize = s;
261 				get_page(hpage);
262 				put_page(page);
263 				page = hpage;
264 			}
265 		}
266 		if (s < psize)
267 			goto out;
268 		pfn = page_to_pfn(page);
269 	}
270 
271 	npages = pgsize >> PAGE_SHIFT;
272 	pgorder = __ilog2(npages);
273 	physp += (gfn - memslot->base_gfn) & ~(npages - 1);
274 	spin_lock(&kvm->arch.slot_phys_lock);
275 	for (i = 0; i < npages; ++i) {
276 		if (!physp[i]) {
277 			physp[i] = ((pfn + i) << PAGE_SHIFT) +
278 				got + is_io + pgorder;
279 			got = 0;
280 		}
281 	}
282 	spin_unlock(&kvm->arch.slot_phys_lock);
283 	err = 0;
284 
285  out:
286 	if (got)
287 		put_page(page);
288 	return err;
289 
290  up_err:
291 	up_read(&current->mm->mmap_sem);
292 	return err;
293 }
294 
295 /*
296  * We come here on a H_ENTER call from the guest when we are not
297  * using mmu notifiers and we don't have the requested page pinned
298  * already.
299  */
kvmppc_virtmode_h_enter(struct kvm_vcpu * vcpu,unsigned long flags,long pte_index,unsigned long pteh,unsigned long ptel)300 long kvmppc_virtmode_h_enter(struct kvm_vcpu *vcpu, unsigned long flags,
301 			long pte_index, unsigned long pteh, unsigned long ptel)
302 {
303 	struct kvm *kvm = vcpu->kvm;
304 	unsigned long psize, gpa, gfn;
305 	struct kvm_memory_slot *memslot;
306 	long ret;
307 
308 	if (kvm->arch.using_mmu_notifiers)
309 		goto do_insert;
310 
311 	psize = hpte_page_size(pteh, ptel);
312 	if (!psize)
313 		return H_PARAMETER;
314 
315 	pteh &= ~(HPTE_V_HVLOCK | HPTE_V_ABSENT | HPTE_V_VALID);
316 
317 	/* Find the memslot (if any) for this address */
318 	gpa = (ptel & HPTE_R_RPN) & ~(psize - 1);
319 	gfn = gpa >> PAGE_SHIFT;
320 	memslot = gfn_to_memslot(kvm, gfn);
321 	if (memslot && !(memslot->flags & KVM_MEMSLOT_INVALID)) {
322 		if (!slot_is_aligned(memslot, psize))
323 			return H_PARAMETER;
324 		if (kvmppc_get_guest_page(kvm, gfn, memslot, psize) < 0)
325 			return H_PARAMETER;
326 	}
327 
328  do_insert:
329 	/* Protect linux PTE lookup from page table destruction */
330 	rcu_read_lock_sched();	/* this disables preemption too */
331 	vcpu->arch.pgdir = current->mm->pgd;
332 	ret = kvmppc_h_enter(vcpu, flags, pte_index, pteh, ptel);
333 	rcu_read_unlock_sched();
334 	if (ret == H_TOO_HARD) {
335 		/* this can't happen */
336 		pr_err("KVM: Oops, kvmppc_h_enter returned too hard!\n");
337 		ret = H_RESOURCE;	/* or something */
338 	}
339 	return ret;
340 
341 }
342 
kvmppc_mmu_book3s_hv_find_slbe(struct kvm_vcpu * vcpu,gva_t eaddr)343 static struct kvmppc_slb *kvmppc_mmu_book3s_hv_find_slbe(struct kvm_vcpu *vcpu,
344 							 gva_t eaddr)
345 {
346 	u64 mask;
347 	int i;
348 
349 	for (i = 0; i < vcpu->arch.slb_nr; i++) {
350 		if (!(vcpu->arch.slb[i].orige & SLB_ESID_V))
351 			continue;
352 
353 		if (vcpu->arch.slb[i].origv & SLB_VSID_B_1T)
354 			mask = ESID_MASK_1T;
355 		else
356 			mask = ESID_MASK;
357 
358 		if (((vcpu->arch.slb[i].orige ^ eaddr) & mask) == 0)
359 			return &vcpu->arch.slb[i];
360 	}
361 	return NULL;
362 }
363 
kvmppc_mmu_get_real_addr(unsigned long v,unsigned long r,unsigned long ea)364 static unsigned long kvmppc_mmu_get_real_addr(unsigned long v, unsigned long r,
365 			unsigned long ea)
366 {
367 	unsigned long ra_mask;
368 
369 	ra_mask = hpte_page_size(v, r) - 1;
370 	return (r & HPTE_R_RPN & ~ra_mask) | (ea & ra_mask);
371 }
372 
kvmppc_mmu_book3s_64_hv_xlate(struct kvm_vcpu * vcpu,gva_t eaddr,struct kvmppc_pte * gpte,bool data)373 static int kvmppc_mmu_book3s_64_hv_xlate(struct kvm_vcpu *vcpu, gva_t eaddr,
374 			struct kvmppc_pte *gpte, bool data)
375 {
376 	struct kvm *kvm = vcpu->kvm;
377 	struct kvmppc_slb *slbe;
378 	unsigned long slb_v;
379 	unsigned long pp, key;
380 	unsigned long v, gr;
381 	unsigned long *hptep;
382 	int index;
383 	int virtmode = vcpu->arch.shregs.msr & (data ? MSR_DR : MSR_IR);
384 
385 	/* Get SLB entry */
386 	if (virtmode) {
387 		slbe = kvmppc_mmu_book3s_hv_find_slbe(vcpu, eaddr);
388 		if (!slbe)
389 			return -EINVAL;
390 		slb_v = slbe->origv;
391 	} else {
392 		/* real mode access */
393 		slb_v = vcpu->kvm->arch.vrma_slb_v;
394 	}
395 
396 	preempt_disable();
397 	/* Find the HPTE in the hash table */
398 	index = kvmppc_hv_find_lock_hpte(kvm, eaddr, slb_v,
399 					 HPTE_V_VALID | HPTE_V_ABSENT);
400 	if (index < 0) {
401 		preempt_enable();
402 		return -ENOENT;
403 	}
404 	hptep = (unsigned long *)(kvm->arch.hpt_virt + (index << 4));
405 	v = hptep[0] & ~HPTE_V_HVLOCK;
406 	gr = kvm->arch.revmap[index].guest_rpte;
407 
408 	/* Unlock the HPTE */
409 	asm volatile("lwsync" : : : "memory");
410 	hptep[0] = v;
411 	preempt_enable();
412 
413 	gpte->eaddr = eaddr;
414 	gpte->vpage = ((v & HPTE_V_AVPN) << 4) | ((eaddr >> 12) & 0xfff);
415 
416 	/* Get PP bits and key for permission check */
417 	pp = gr & (HPTE_R_PP0 | HPTE_R_PP);
418 	key = (vcpu->arch.shregs.msr & MSR_PR) ? SLB_VSID_KP : SLB_VSID_KS;
419 	key &= slb_v;
420 
421 	/* Calculate permissions */
422 	gpte->may_read = hpte_read_permission(pp, key);
423 	gpte->may_write = hpte_write_permission(pp, key);
424 	gpte->may_execute = gpte->may_read && !(gr & (HPTE_R_N | HPTE_R_G));
425 
426 	/* Storage key permission check for POWER7 */
427 	if (data && virtmode && cpu_has_feature(CPU_FTR_ARCH_206)) {
428 		int amrfield = hpte_get_skey_perm(gr, vcpu->arch.amr);
429 		if (amrfield & 1)
430 			gpte->may_read = 0;
431 		if (amrfield & 2)
432 			gpte->may_write = 0;
433 	}
434 
435 	/* Get the guest physical address */
436 	gpte->raddr = kvmppc_mmu_get_real_addr(v, gr, eaddr);
437 	return 0;
438 }
439 
440 /*
441  * Quick test for whether an instruction is a load or a store.
442  * If the instruction is a load or a store, then this will indicate
443  * which it is, at least on server processors.  (Embedded processors
444  * have some external PID instructions that don't follow the rule
445  * embodied here.)  If the instruction isn't a load or store, then
446  * this doesn't return anything useful.
447  */
instruction_is_store(unsigned int instr)448 static int instruction_is_store(unsigned int instr)
449 {
450 	unsigned int mask;
451 
452 	mask = 0x10000000;
453 	if ((instr & 0xfc000000) == 0x7c000000)
454 		mask = 0x100;		/* major opcode 31 */
455 	return (instr & mask) != 0;
456 }
457 
kvmppc_hv_emulate_mmio(struct kvm_run * run,struct kvm_vcpu * vcpu,unsigned long gpa,int is_store)458 static int kvmppc_hv_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu,
459 				  unsigned long gpa, int is_store)
460 {
461 	int ret;
462 	u32 last_inst;
463 	unsigned long srr0 = kvmppc_get_pc(vcpu);
464 
465 	/* We try to load the last instruction.  We don't let
466 	 * emulate_instruction do it as it doesn't check what
467 	 * kvmppc_ld returns.
468 	 * If we fail, we just return to the guest and try executing it again.
469 	 */
470 	if (vcpu->arch.last_inst == KVM_INST_FETCH_FAILED) {
471 		ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false);
472 		if (ret != EMULATE_DONE || last_inst == KVM_INST_FETCH_FAILED)
473 			return RESUME_GUEST;
474 		vcpu->arch.last_inst = last_inst;
475 	}
476 
477 	/*
478 	 * WARNING: We do not know for sure whether the instruction we just
479 	 * read from memory is the same that caused the fault in the first
480 	 * place.  If the instruction we read is neither an load or a store,
481 	 * then it can't access memory, so we don't need to worry about
482 	 * enforcing access permissions.  So, assuming it is a load or
483 	 * store, we just check that its direction (load or store) is
484 	 * consistent with the original fault, since that's what we
485 	 * checked the access permissions against.  If there is a mismatch
486 	 * we just return and retry the instruction.
487 	 */
488 
489 	if (instruction_is_store(vcpu->arch.last_inst) != !!is_store)
490 		return RESUME_GUEST;
491 
492 	/*
493 	 * Emulated accesses are emulated by looking at the hash for
494 	 * translation once, then performing the access later. The
495 	 * translation could be invalidated in the meantime in which
496 	 * point performing the subsequent memory access on the old
497 	 * physical address could possibly be a security hole for the
498 	 * guest (but not the host).
499 	 *
500 	 * This is less of an issue for MMIO stores since they aren't
501 	 * globally visible. It could be an issue for MMIO loads to
502 	 * a certain extent but we'll ignore it for now.
503 	 */
504 
505 	vcpu->arch.paddr_accessed = gpa;
506 	return kvmppc_emulate_mmio(run, vcpu);
507 }
508 
kvmppc_book3s_hv_page_fault(struct kvm_run * run,struct kvm_vcpu * vcpu,unsigned long ea,unsigned long dsisr)509 int kvmppc_book3s_hv_page_fault(struct kvm_run *run, struct kvm_vcpu *vcpu,
510 				unsigned long ea, unsigned long dsisr)
511 {
512 	struct kvm *kvm = vcpu->kvm;
513 	unsigned long *hptep, hpte[3], r;
514 	unsigned long mmu_seq, psize, pte_size;
515 	unsigned long gfn, hva, pfn;
516 	struct kvm_memory_slot *memslot;
517 	unsigned long *rmap;
518 	struct revmap_entry *rev;
519 	struct page *page, *pages[1];
520 	long index, ret, npages;
521 	unsigned long is_io;
522 	unsigned int writing, write_ok;
523 	struct vm_area_struct *vma;
524 	unsigned long rcbits;
525 
526 	/*
527 	 * Real-mode code has already searched the HPT and found the
528 	 * entry we're interested in.  Lock the entry and check that
529 	 * it hasn't changed.  If it has, just return and re-execute the
530 	 * instruction.
531 	 */
532 	if (ea != vcpu->arch.pgfault_addr)
533 		return RESUME_GUEST;
534 	index = vcpu->arch.pgfault_index;
535 	hptep = (unsigned long *)(kvm->arch.hpt_virt + (index << 4));
536 	rev = &kvm->arch.revmap[index];
537 	preempt_disable();
538 	while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
539 		cpu_relax();
540 	hpte[0] = hptep[0] & ~HPTE_V_HVLOCK;
541 	hpte[1] = hptep[1];
542 	hpte[2] = r = rev->guest_rpte;
543 	asm volatile("lwsync" : : : "memory");
544 	hptep[0] = hpte[0];
545 	preempt_enable();
546 
547 	if (hpte[0] != vcpu->arch.pgfault_hpte[0] ||
548 	    hpte[1] != vcpu->arch.pgfault_hpte[1])
549 		return RESUME_GUEST;
550 
551 	/* Translate the logical address and get the page */
552 	psize = hpte_page_size(hpte[0], r);
553 	gfn = hpte_rpn(r, psize);
554 	memslot = gfn_to_memslot(kvm, gfn);
555 
556 	/* No memslot means it's an emulated MMIO region */
557 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID)) {
558 		unsigned long gpa = (gfn << PAGE_SHIFT) | (ea & (psize - 1));
559 		return kvmppc_hv_emulate_mmio(run, vcpu, gpa,
560 					      dsisr & DSISR_ISSTORE);
561 	}
562 
563 	if (!kvm->arch.using_mmu_notifiers)
564 		return -EFAULT;		/* should never get here */
565 
566 	/* used to check for invalidations in progress */
567 	mmu_seq = kvm->mmu_notifier_seq;
568 	smp_rmb();
569 
570 	is_io = 0;
571 	pfn = 0;
572 	page = NULL;
573 	pte_size = PAGE_SIZE;
574 	writing = (dsisr & DSISR_ISSTORE) != 0;
575 	/* If writing != 0, then the HPTE must allow writing, if we get here */
576 	write_ok = writing;
577 	hva = gfn_to_hva_memslot(memslot, gfn);
578 	npages = get_user_pages_fast(hva, 1, writing, pages);
579 	if (npages < 1) {
580 		/* Check if it's an I/O mapping */
581 		down_read(&current->mm->mmap_sem);
582 		vma = find_vma(current->mm, hva);
583 		if (vma && vma->vm_start <= hva && hva + psize <= vma->vm_end &&
584 		    (vma->vm_flags & VM_PFNMAP)) {
585 			pfn = vma->vm_pgoff +
586 				((hva - vma->vm_start) >> PAGE_SHIFT);
587 			pte_size = psize;
588 			is_io = hpte_cache_bits(pgprot_val(vma->vm_page_prot));
589 			write_ok = vma->vm_flags & VM_WRITE;
590 		}
591 		up_read(&current->mm->mmap_sem);
592 		if (!pfn)
593 			return -EFAULT;
594 	} else {
595 		page = pages[0];
596 		if (PageHuge(page)) {
597 			page = compound_head(page);
598 			pte_size <<= compound_order(page);
599 		}
600 		/* if the guest wants write access, see if that is OK */
601 		if (!writing && hpte_is_writable(r)) {
602 			pte_t *ptep, pte;
603 
604 			/*
605 			 * We need to protect against page table destruction
606 			 * while looking up and updating the pte.
607 			 */
608 			rcu_read_lock_sched();
609 			ptep = find_linux_pte_or_hugepte(current->mm->pgd,
610 							 hva, NULL);
611 			if (ptep && pte_present(*ptep)) {
612 				pte = kvmppc_read_update_linux_pte(ptep, 1);
613 				if (pte_write(pte))
614 					write_ok = 1;
615 			}
616 			rcu_read_unlock_sched();
617 		}
618 		pfn = page_to_pfn(page);
619 	}
620 
621 	ret = -EFAULT;
622 	if (psize > pte_size)
623 		goto out_put;
624 
625 	/* Check WIMG vs. the actual page we're accessing */
626 	if (!hpte_cache_flags_ok(r, is_io)) {
627 		if (is_io)
628 			return -EFAULT;
629 		/*
630 		 * Allow guest to map emulated device memory as
631 		 * uncacheable, but actually make it cacheable.
632 		 */
633 		r = (r & ~(HPTE_R_W|HPTE_R_I|HPTE_R_G)) | HPTE_R_M;
634 	}
635 
636 	/* Set the HPTE to point to pfn */
637 	r = (r & ~(HPTE_R_PP0 - pte_size)) | (pfn << PAGE_SHIFT);
638 	if (hpte_is_writable(r) && !write_ok)
639 		r = hpte_make_readonly(r);
640 	ret = RESUME_GUEST;
641 	preempt_disable();
642 	while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
643 		cpu_relax();
644 	if ((hptep[0] & ~HPTE_V_HVLOCK) != hpte[0] || hptep[1] != hpte[1] ||
645 	    rev->guest_rpte != hpte[2])
646 		/* HPTE has been changed under us; let the guest retry */
647 		goto out_unlock;
648 	hpte[0] = (hpte[0] & ~HPTE_V_ABSENT) | HPTE_V_VALID;
649 
650 	rmap = &memslot->rmap[gfn - memslot->base_gfn];
651 	lock_rmap(rmap);
652 
653 	/* Check if we might have been invalidated; let the guest retry if so */
654 	ret = RESUME_GUEST;
655 	if (mmu_notifier_retry(vcpu, mmu_seq)) {
656 		unlock_rmap(rmap);
657 		goto out_unlock;
658 	}
659 
660 	/* Only set R/C in real HPTE if set in both *rmap and guest_rpte */
661 	rcbits = *rmap >> KVMPPC_RMAP_RC_SHIFT;
662 	r &= rcbits | ~(HPTE_R_R | HPTE_R_C);
663 
664 	if (hptep[0] & HPTE_V_VALID) {
665 		/* HPTE was previously valid, so we need to invalidate it */
666 		unlock_rmap(rmap);
667 		hptep[0] |= HPTE_V_ABSENT;
668 		kvmppc_invalidate_hpte(kvm, hptep, index);
669 		/* don't lose previous R and C bits */
670 		r |= hptep[1] & (HPTE_R_R | HPTE_R_C);
671 	} else {
672 		kvmppc_add_revmap_chain(kvm, rev, rmap, index, 0);
673 	}
674 
675 	hptep[1] = r;
676 	eieio();
677 	hptep[0] = hpte[0];
678 	asm volatile("ptesync" : : : "memory");
679 	preempt_enable();
680 	if (page && hpte_is_writable(r))
681 		SetPageDirty(page);
682 
683  out_put:
684 	if (page) {
685 		/*
686 		 * We drop pages[0] here, not page because page might
687 		 * have been set to the head page of a compound, but
688 		 * we have to drop the reference on the correct tail
689 		 * page to match the get inside gup()
690 		 */
691 		put_page(pages[0]);
692 	}
693 	return ret;
694 
695  out_unlock:
696 	hptep[0] &= ~HPTE_V_HVLOCK;
697 	preempt_enable();
698 	goto out_put;
699 }
700 
kvm_handle_hva(struct kvm * kvm,unsigned long hva,int (* handler)(struct kvm * kvm,unsigned long * rmapp,unsigned long gfn))701 static int kvm_handle_hva(struct kvm *kvm, unsigned long hva,
702 			  int (*handler)(struct kvm *kvm, unsigned long *rmapp,
703 					 unsigned long gfn))
704 {
705 	int ret;
706 	int retval = 0;
707 	struct kvm_memslots *slots;
708 	struct kvm_memory_slot *memslot;
709 
710 	slots = kvm_memslots(kvm);
711 	kvm_for_each_memslot(memslot, slots) {
712 		unsigned long start = memslot->userspace_addr;
713 		unsigned long end;
714 
715 		end = start + (memslot->npages << PAGE_SHIFT);
716 		if (hva >= start && hva < end) {
717 			gfn_t gfn_offset = (hva - start) >> PAGE_SHIFT;
718 
719 			ret = handler(kvm, &memslot->rmap[gfn_offset],
720 				      memslot->base_gfn + gfn_offset);
721 			retval |= ret;
722 		}
723 	}
724 
725 	return retval;
726 }
727 
kvm_unmap_rmapp(struct kvm * kvm,unsigned long * rmapp,unsigned long gfn)728 static int kvm_unmap_rmapp(struct kvm *kvm, unsigned long *rmapp,
729 			   unsigned long gfn)
730 {
731 	struct revmap_entry *rev = kvm->arch.revmap;
732 	unsigned long h, i, j;
733 	unsigned long *hptep;
734 	unsigned long ptel, psize, rcbits;
735 
736 	for (;;) {
737 		lock_rmap(rmapp);
738 		if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
739 			unlock_rmap(rmapp);
740 			break;
741 		}
742 
743 		/*
744 		 * To avoid an ABBA deadlock with the HPTE lock bit,
745 		 * we can't spin on the HPTE lock while holding the
746 		 * rmap chain lock.
747 		 */
748 		i = *rmapp & KVMPPC_RMAP_INDEX;
749 		hptep = (unsigned long *) (kvm->arch.hpt_virt + (i << 4));
750 		if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
751 			/* unlock rmap before spinning on the HPTE lock */
752 			unlock_rmap(rmapp);
753 			while (hptep[0] & HPTE_V_HVLOCK)
754 				cpu_relax();
755 			continue;
756 		}
757 		j = rev[i].forw;
758 		if (j == i) {
759 			/* chain is now empty */
760 			*rmapp &= ~(KVMPPC_RMAP_PRESENT | KVMPPC_RMAP_INDEX);
761 		} else {
762 			/* remove i from chain */
763 			h = rev[i].back;
764 			rev[h].forw = j;
765 			rev[j].back = h;
766 			rev[i].forw = rev[i].back = i;
767 			*rmapp = (*rmapp & ~KVMPPC_RMAP_INDEX) | j;
768 		}
769 
770 		/* Now check and modify the HPTE */
771 		ptel = rev[i].guest_rpte;
772 		psize = hpte_page_size(hptep[0], ptel);
773 		if ((hptep[0] & HPTE_V_VALID) &&
774 		    hpte_rpn(ptel, psize) == gfn) {
775 			hptep[0] |= HPTE_V_ABSENT;
776 			kvmppc_invalidate_hpte(kvm, hptep, i);
777 			/* Harvest R and C */
778 			rcbits = hptep[1] & (HPTE_R_R | HPTE_R_C);
779 			*rmapp |= rcbits << KVMPPC_RMAP_RC_SHIFT;
780 			rev[i].guest_rpte = ptel | rcbits;
781 		}
782 		unlock_rmap(rmapp);
783 		hptep[0] &= ~HPTE_V_HVLOCK;
784 	}
785 	return 0;
786 }
787 
kvm_unmap_hva(struct kvm * kvm,unsigned long hva)788 int kvm_unmap_hva(struct kvm *kvm, unsigned long hva)
789 {
790 	if (kvm->arch.using_mmu_notifiers)
791 		kvm_handle_hva(kvm, hva, kvm_unmap_rmapp);
792 	return 0;
793 }
794 
kvm_age_rmapp(struct kvm * kvm,unsigned long * rmapp,unsigned long gfn)795 static int kvm_age_rmapp(struct kvm *kvm, unsigned long *rmapp,
796 			 unsigned long gfn)
797 {
798 	struct revmap_entry *rev = kvm->arch.revmap;
799 	unsigned long head, i, j;
800 	unsigned long *hptep;
801 	int ret = 0;
802 
803  retry:
804 	lock_rmap(rmapp);
805 	if (*rmapp & KVMPPC_RMAP_REFERENCED) {
806 		*rmapp &= ~KVMPPC_RMAP_REFERENCED;
807 		ret = 1;
808 	}
809 	if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
810 		unlock_rmap(rmapp);
811 		return ret;
812 	}
813 
814 	i = head = *rmapp & KVMPPC_RMAP_INDEX;
815 	do {
816 		hptep = (unsigned long *) (kvm->arch.hpt_virt + (i << 4));
817 		j = rev[i].forw;
818 
819 		/* If this HPTE isn't referenced, ignore it */
820 		if (!(hptep[1] & HPTE_R_R))
821 			continue;
822 
823 		if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
824 			/* unlock rmap before spinning on the HPTE lock */
825 			unlock_rmap(rmapp);
826 			while (hptep[0] & HPTE_V_HVLOCK)
827 				cpu_relax();
828 			goto retry;
829 		}
830 
831 		/* Now check and modify the HPTE */
832 		if ((hptep[0] & HPTE_V_VALID) && (hptep[1] & HPTE_R_R)) {
833 			kvmppc_clear_ref_hpte(kvm, hptep, i);
834 			rev[i].guest_rpte |= HPTE_R_R;
835 			ret = 1;
836 		}
837 		hptep[0] &= ~HPTE_V_HVLOCK;
838 	} while ((i = j) != head);
839 
840 	unlock_rmap(rmapp);
841 	return ret;
842 }
843 
kvm_age_hva(struct kvm * kvm,unsigned long hva)844 int kvm_age_hva(struct kvm *kvm, unsigned long hva)
845 {
846 	if (!kvm->arch.using_mmu_notifiers)
847 		return 0;
848 	return kvm_handle_hva(kvm, hva, kvm_age_rmapp);
849 }
850 
kvm_test_age_rmapp(struct kvm * kvm,unsigned long * rmapp,unsigned long gfn)851 static int kvm_test_age_rmapp(struct kvm *kvm, unsigned long *rmapp,
852 			      unsigned long gfn)
853 {
854 	struct revmap_entry *rev = kvm->arch.revmap;
855 	unsigned long head, i, j;
856 	unsigned long *hp;
857 	int ret = 1;
858 
859 	if (*rmapp & KVMPPC_RMAP_REFERENCED)
860 		return 1;
861 
862 	lock_rmap(rmapp);
863 	if (*rmapp & KVMPPC_RMAP_REFERENCED)
864 		goto out;
865 
866 	if (*rmapp & KVMPPC_RMAP_PRESENT) {
867 		i = head = *rmapp & KVMPPC_RMAP_INDEX;
868 		do {
869 			hp = (unsigned long *)(kvm->arch.hpt_virt + (i << 4));
870 			j = rev[i].forw;
871 			if (hp[1] & HPTE_R_R)
872 				goto out;
873 		} while ((i = j) != head);
874 	}
875 	ret = 0;
876 
877  out:
878 	unlock_rmap(rmapp);
879 	return ret;
880 }
881 
kvm_test_age_hva(struct kvm * kvm,unsigned long hva)882 int kvm_test_age_hva(struct kvm *kvm, unsigned long hva)
883 {
884 	if (!kvm->arch.using_mmu_notifiers)
885 		return 0;
886 	return kvm_handle_hva(kvm, hva, kvm_test_age_rmapp);
887 }
888 
kvm_set_spte_hva(struct kvm * kvm,unsigned long hva,pte_t pte)889 void kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte)
890 {
891 	if (!kvm->arch.using_mmu_notifiers)
892 		return;
893 	kvm_handle_hva(kvm, hva, kvm_unmap_rmapp);
894 }
895 
kvm_test_clear_dirty(struct kvm * kvm,unsigned long * rmapp)896 static int kvm_test_clear_dirty(struct kvm *kvm, unsigned long *rmapp)
897 {
898 	struct revmap_entry *rev = kvm->arch.revmap;
899 	unsigned long head, i, j;
900 	unsigned long *hptep;
901 	int ret = 0;
902 
903  retry:
904 	lock_rmap(rmapp);
905 	if (*rmapp & KVMPPC_RMAP_CHANGED) {
906 		*rmapp &= ~KVMPPC_RMAP_CHANGED;
907 		ret = 1;
908 	}
909 	if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
910 		unlock_rmap(rmapp);
911 		return ret;
912 	}
913 
914 	i = head = *rmapp & KVMPPC_RMAP_INDEX;
915 	do {
916 		hptep = (unsigned long *) (kvm->arch.hpt_virt + (i << 4));
917 		j = rev[i].forw;
918 
919 		if (!(hptep[1] & HPTE_R_C))
920 			continue;
921 
922 		if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
923 			/* unlock rmap before spinning on the HPTE lock */
924 			unlock_rmap(rmapp);
925 			while (hptep[0] & HPTE_V_HVLOCK)
926 				cpu_relax();
927 			goto retry;
928 		}
929 
930 		/* Now check and modify the HPTE */
931 		if ((hptep[0] & HPTE_V_VALID) && (hptep[1] & HPTE_R_C)) {
932 			/* need to make it temporarily absent to clear C */
933 			hptep[0] |= HPTE_V_ABSENT;
934 			kvmppc_invalidate_hpte(kvm, hptep, i);
935 			hptep[1] &= ~HPTE_R_C;
936 			eieio();
937 			hptep[0] = (hptep[0] & ~HPTE_V_ABSENT) | HPTE_V_VALID;
938 			rev[i].guest_rpte |= HPTE_R_C;
939 			ret = 1;
940 		}
941 		hptep[0] &= ~HPTE_V_HVLOCK;
942 	} while ((i = j) != head);
943 
944 	unlock_rmap(rmapp);
945 	return ret;
946 }
947 
kvmppc_hv_get_dirty_log(struct kvm * kvm,struct kvm_memory_slot * memslot)948 long kvmppc_hv_get_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
949 {
950 	unsigned long i;
951 	unsigned long *rmapp, *map;
952 
953 	preempt_disable();
954 	rmapp = memslot->rmap;
955 	map = memslot->dirty_bitmap;
956 	for (i = 0; i < memslot->npages; ++i) {
957 		if (kvm_test_clear_dirty(kvm, rmapp))
958 			__set_bit_le(i, map);
959 		++rmapp;
960 	}
961 	preempt_enable();
962 	return 0;
963 }
964 
kvmppc_pin_guest_page(struct kvm * kvm,unsigned long gpa,unsigned long * nb_ret)965 void *kvmppc_pin_guest_page(struct kvm *kvm, unsigned long gpa,
966 			    unsigned long *nb_ret)
967 {
968 	struct kvm_memory_slot *memslot;
969 	unsigned long gfn = gpa >> PAGE_SHIFT;
970 	struct page *page, *pages[1];
971 	int npages;
972 	unsigned long hva, psize, offset;
973 	unsigned long pa;
974 	unsigned long *physp;
975 
976 	memslot = gfn_to_memslot(kvm, gfn);
977 	if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
978 		return NULL;
979 	if (!kvm->arch.using_mmu_notifiers) {
980 		physp = kvm->arch.slot_phys[memslot->id];
981 		if (!physp)
982 			return NULL;
983 		physp += gfn - memslot->base_gfn;
984 		pa = *physp;
985 		if (!pa) {
986 			if (kvmppc_get_guest_page(kvm, gfn, memslot,
987 						  PAGE_SIZE) < 0)
988 				return NULL;
989 			pa = *physp;
990 		}
991 		page = pfn_to_page(pa >> PAGE_SHIFT);
992 		get_page(page);
993 	} else {
994 		hva = gfn_to_hva_memslot(memslot, gfn);
995 		npages = get_user_pages_fast(hva, 1, 1, pages);
996 		if (npages < 1)
997 			return NULL;
998 		page = pages[0];
999 	}
1000 	psize = PAGE_SIZE;
1001 	if (PageHuge(page)) {
1002 		page = compound_head(page);
1003 		psize <<= compound_order(page);
1004 	}
1005 	offset = gpa & (psize - 1);
1006 	if (nb_ret)
1007 		*nb_ret = psize - offset;
1008 	return page_address(page) + offset;
1009 }
1010 
kvmppc_unpin_guest_page(struct kvm * kvm,void * va)1011 void kvmppc_unpin_guest_page(struct kvm *kvm, void *va)
1012 {
1013 	struct page *page = virt_to_page(va);
1014 
1015 	put_page(page);
1016 }
1017 
kvmppc_mmu_book3s_hv_init(struct kvm_vcpu * vcpu)1018 void kvmppc_mmu_book3s_hv_init(struct kvm_vcpu *vcpu)
1019 {
1020 	struct kvmppc_mmu *mmu = &vcpu->arch.mmu;
1021 
1022 	if (cpu_has_feature(CPU_FTR_ARCH_206))
1023 		vcpu->arch.slb_nr = 32;		/* POWER7 */
1024 	else
1025 		vcpu->arch.slb_nr = 64;
1026 
1027 	mmu->xlate = kvmppc_mmu_book3s_64_hv_xlate;
1028 	mmu->reset_msr = kvmppc_mmu_book3s_64_hv_reset_msr;
1029 
1030 	vcpu->arch.hflags |= BOOK3S_HFLAG_SLB;
1031 }
1032