xref: /openbmc/linux/arch/powerpc/kvm/e500_mmu_host.c (revision 77d84ff8)
1 /*
2  * Copyright (C) 2008-2013 Freescale Semiconductor, Inc. All rights reserved.
3  *
4  * Author: Yu Liu, yu.liu@freescale.com
5  *         Scott Wood, scottwood@freescale.com
6  *         Ashish Kalra, ashish.kalra@freescale.com
7  *         Varun Sethi, varun.sethi@freescale.com
8  *         Alexander Graf, agraf@suse.de
9  *
10  * Description:
11  * This file is based on arch/powerpc/kvm/44x_tlb.c,
12  * by Hollis Blanchard <hollisb@us.ibm.com>.
13  *
14  * This program is free software; you can redistribute it and/or modify
15  * it under the terms of the GNU General Public License, version 2, as
16  * published by the Free Software Foundation.
17  */
18 
19 #include <linux/kernel.h>
20 #include <linux/types.h>
21 #include <linux/slab.h>
22 #include <linux/string.h>
23 #include <linux/kvm.h>
24 #include <linux/kvm_host.h>
25 #include <linux/highmem.h>
26 #include <linux/log2.h>
27 #include <linux/uaccess.h>
28 #include <linux/sched.h>
29 #include <linux/rwsem.h>
30 #include <linux/vmalloc.h>
31 #include <linux/hugetlb.h>
32 #include <asm/kvm_ppc.h>
33 
34 #include "e500.h"
35 #include "timing.h"
36 #include "e500_mmu_host.h"
37 
38 #include "trace_booke.h"
39 
40 #define to_htlb1_esel(esel) (host_tlb_params[1].entries - (esel) - 1)
41 
42 static struct kvmppc_e500_tlb_params host_tlb_params[E500_TLB_NUM];
43 
44 static inline unsigned int tlb1_max_shadow_size(void)
45 {
46 	/* reserve one entry for magic page */
47 	return host_tlb_params[1].entries - tlbcam_index - 1;
48 }
49 
50 static inline u32 e500_shadow_mas3_attrib(u32 mas3, int usermode)
51 {
52 	/* Mask off reserved bits. */
53 	mas3 &= MAS3_ATTRIB_MASK;
54 
55 #ifndef CONFIG_KVM_BOOKE_HV
56 	if (!usermode) {
57 		/* Guest is in supervisor mode,
58 		 * so we need to translate guest
59 		 * supervisor permissions into user permissions. */
60 		mas3 &= ~E500_TLB_USER_PERM_MASK;
61 		mas3 |= (mas3 & E500_TLB_SUPER_PERM_MASK) << 1;
62 	}
63 	mas3 |= E500_TLB_SUPER_PERM_MASK;
64 #endif
65 	return mas3;
66 }
67 
68 static inline u32 e500_shadow_mas2_attrib(u32 mas2, int usermode)
69 {
70 #ifdef CONFIG_SMP
71 	return (mas2 & MAS2_ATTRIB_MASK) | MAS2_M;
72 #else
73 	return mas2 & MAS2_ATTRIB_MASK;
74 #endif
75 }
76 
77 /*
78  * writing shadow tlb entry to host TLB
79  */
80 static inline void __write_host_tlbe(struct kvm_book3e_206_tlb_entry *stlbe,
81 				     uint32_t mas0)
82 {
83 	unsigned long flags;
84 
85 	local_irq_save(flags);
86 	mtspr(SPRN_MAS0, mas0);
87 	mtspr(SPRN_MAS1, stlbe->mas1);
88 	mtspr(SPRN_MAS2, (unsigned long)stlbe->mas2);
89 	mtspr(SPRN_MAS3, (u32)stlbe->mas7_3);
90 	mtspr(SPRN_MAS7, (u32)(stlbe->mas7_3 >> 32));
91 #ifdef CONFIG_KVM_BOOKE_HV
92 	mtspr(SPRN_MAS8, stlbe->mas8);
93 #endif
94 	asm volatile("isync; tlbwe" : : : "memory");
95 
96 #ifdef CONFIG_KVM_BOOKE_HV
97 	/* Must clear mas8 for other host tlbwe's */
98 	mtspr(SPRN_MAS8, 0);
99 	isync();
100 #endif
101 	local_irq_restore(flags);
102 
103 	trace_kvm_booke206_stlb_write(mas0, stlbe->mas8, stlbe->mas1,
104 	                              stlbe->mas2, stlbe->mas7_3);
105 }
106 
107 /*
108  * Acquire a mas0 with victim hint, as if we just took a TLB miss.
109  *
110  * We don't care about the address we're searching for, other than that it's
111  * in the right set and is not present in the TLB.  Using a zero PID and a
112  * userspace address means we don't have to set and then restore MAS5, or
113  * calculate a proper MAS6 value.
114  */
115 static u32 get_host_mas0(unsigned long eaddr)
116 {
117 	unsigned long flags;
118 	u32 mas0;
119 
120 	local_irq_save(flags);
121 	mtspr(SPRN_MAS6, 0);
122 	asm volatile("tlbsx 0, %0" : : "b" (eaddr & ~CONFIG_PAGE_OFFSET));
123 	mas0 = mfspr(SPRN_MAS0);
124 	local_irq_restore(flags);
125 
126 	return mas0;
127 }
128 
129 /* sesel is for tlb1 only */
130 static inline void write_host_tlbe(struct kvmppc_vcpu_e500 *vcpu_e500,
131 		int tlbsel, int sesel, struct kvm_book3e_206_tlb_entry *stlbe)
132 {
133 	u32 mas0;
134 
135 	if (tlbsel == 0) {
136 		mas0 = get_host_mas0(stlbe->mas2);
137 		__write_host_tlbe(stlbe, mas0);
138 	} else {
139 		__write_host_tlbe(stlbe,
140 				  MAS0_TLBSEL(1) |
141 				  MAS0_ESEL(to_htlb1_esel(sesel)));
142 	}
143 }
144 
145 /* sesel is for tlb1 only */
146 static void write_stlbe(struct kvmppc_vcpu_e500 *vcpu_e500,
147 			struct kvm_book3e_206_tlb_entry *gtlbe,
148 			struct kvm_book3e_206_tlb_entry *stlbe,
149 			int stlbsel, int sesel)
150 {
151 	int stid;
152 
153 	preempt_disable();
154 	stid = kvmppc_e500_get_tlb_stid(&vcpu_e500->vcpu, gtlbe);
155 
156 	stlbe->mas1 |= MAS1_TID(stid);
157 	write_host_tlbe(vcpu_e500, stlbsel, sesel, stlbe);
158 	preempt_enable();
159 }
160 
161 #ifdef CONFIG_KVM_E500V2
162 /* XXX should be a hook in the gva2hpa translation */
163 void kvmppc_map_magic(struct kvm_vcpu *vcpu)
164 {
165 	struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
166 	struct kvm_book3e_206_tlb_entry magic;
167 	ulong shared_page = ((ulong)vcpu->arch.shared) & PAGE_MASK;
168 	unsigned int stid;
169 	pfn_t pfn;
170 
171 	pfn = (pfn_t)virt_to_phys((void *)shared_page) >> PAGE_SHIFT;
172 	get_page(pfn_to_page(pfn));
173 
174 	preempt_disable();
175 	stid = kvmppc_e500_get_sid(vcpu_e500, 0, 0, 0, 0);
176 
177 	magic.mas1 = MAS1_VALID | MAS1_TS | MAS1_TID(stid) |
178 		     MAS1_TSIZE(BOOK3E_PAGESZ_4K);
179 	magic.mas2 = vcpu->arch.magic_page_ea | MAS2_M;
180 	magic.mas7_3 = ((u64)pfn << PAGE_SHIFT) |
181 		       MAS3_SW | MAS3_SR | MAS3_UW | MAS3_UR;
182 	magic.mas8 = 0;
183 
184 	__write_host_tlbe(&magic, MAS0_TLBSEL(1) | MAS0_ESEL(tlbcam_index));
185 	preempt_enable();
186 }
187 #endif
188 
189 void inval_gtlbe_on_host(struct kvmppc_vcpu_e500 *vcpu_e500, int tlbsel,
190 			 int esel)
191 {
192 	struct kvm_book3e_206_tlb_entry *gtlbe =
193 		get_entry(vcpu_e500, tlbsel, esel);
194 	struct tlbe_ref *ref = &vcpu_e500->gtlb_priv[tlbsel][esel].ref;
195 
196 	/* Don't bother with unmapped entries */
197 	if (!(ref->flags & E500_TLB_VALID)) {
198 		WARN(ref->flags & (E500_TLB_BITMAP | E500_TLB_TLB0),
199 		     "%s: flags %x\n", __func__, ref->flags);
200 		WARN_ON(tlbsel == 1 && vcpu_e500->g2h_tlb1_map[esel]);
201 	}
202 
203 	if (tlbsel == 1 && ref->flags & E500_TLB_BITMAP) {
204 		u64 tmp = vcpu_e500->g2h_tlb1_map[esel];
205 		int hw_tlb_indx;
206 		unsigned long flags;
207 
208 		local_irq_save(flags);
209 		while (tmp) {
210 			hw_tlb_indx = __ilog2_u64(tmp & -tmp);
211 			mtspr(SPRN_MAS0,
212 			      MAS0_TLBSEL(1) |
213 			      MAS0_ESEL(to_htlb1_esel(hw_tlb_indx)));
214 			mtspr(SPRN_MAS1, 0);
215 			asm volatile("tlbwe");
216 			vcpu_e500->h2g_tlb1_rmap[hw_tlb_indx] = 0;
217 			tmp &= tmp - 1;
218 		}
219 		mb();
220 		vcpu_e500->g2h_tlb1_map[esel] = 0;
221 		ref->flags &= ~(E500_TLB_BITMAP | E500_TLB_VALID);
222 		local_irq_restore(flags);
223 	}
224 
225 	if (tlbsel == 1 && ref->flags & E500_TLB_TLB0) {
226 		/*
227 		 * TLB1 entry is backed by 4k pages. This should happen
228 		 * rarely and is not worth optimizing. Invalidate everything.
229 		 */
230 		kvmppc_e500_tlbil_all(vcpu_e500);
231 		ref->flags &= ~(E500_TLB_TLB0 | E500_TLB_VALID);
232 	}
233 
234 	/* Already invalidated in between */
235 	if (!(ref->flags & E500_TLB_VALID))
236 		return;
237 
238 	/* Guest tlbe is backed by at most one host tlbe per shadow pid. */
239 	kvmppc_e500_tlbil_one(vcpu_e500, gtlbe);
240 
241 	/* Mark the TLB as not backed by the host anymore */
242 	ref->flags &= ~E500_TLB_VALID;
243 }
244 
245 static inline int tlbe_is_writable(struct kvm_book3e_206_tlb_entry *tlbe)
246 {
247 	return tlbe->mas7_3 & (MAS3_SW|MAS3_UW);
248 }
249 
250 static inline void kvmppc_e500_ref_setup(struct tlbe_ref *ref,
251 					 struct kvm_book3e_206_tlb_entry *gtlbe,
252 					 pfn_t pfn)
253 {
254 	ref->pfn = pfn;
255 	ref->flags |= E500_TLB_VALID;
256 
257 	/* Mark the page accessed */
258 	kvm_set_pfn_accessed(pfn);
259 
260 	if (tlbe_is_writable(gtlbe))
261 		kvm_set_pfn_dirty(pfn);
262 }
263 
264 static inline void kvmppc_e500_ref_release(struct tlbe_ref *ref)
265 {
266 	if (ref->flags & E500_TLB_VALID) {
267 		/* FIXME: don't log bogus pfn for TLB1 */
268 		trace_kvm_booke206_ref_release(ref->pfn, ref->flags);
269 		ref->flags = 0;
270 	}
271 }
272 
273 static void clear_tlb1_bitmap(struct kvmppc_vcpu_e500 *vcpu_e500)
274 {
275 	if (vcpu_e500->g2h_tlb1_map)
276 		memset(vcpu_e500->g2h_tlb1_map, 0,
277 		       sizeof(u64) * vcpu_e500->gtlb_params[1].entries);
278 	if (vcpu_e500->h2g_tlb1_rmap)
279 		memset(vcpu_e500->h2g_tlb1_rmap, 0,
280 		       sizeof(unsigned int) * host_tlb_params[1].entries);
281 }
282 
283 static void clear_tlb_privs(struct kvmppc_vcpu_e500 *vcpu_e500)
284 {
285 	int tlbsel;
286 	int i;
287 
288 	for (tlbsel = 0; tlbsel <= 1; tlbsel++) {
289 		for (i = 0; i < vcpu_e500->gtlb_params[tlbsel].entries; i++) {
290 			struct tlbe_ref *ref =
291 				&vcpu_e500->gtlb_priv[tlbsel][i].ref;
292 			kvmppc_e500_ref_release(ref);
293 		}
294 	}
295 }
296 
297 void kvmppc_core_flush_tlb(struct kvm_vcpu *vcpu)
298 {
299 	struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
300 	kvmppc_e500_tlbil_all(vcpu_e500);
301 	clear_tlb_privs(vcpu_e500);
302 	clear_tlb1_bitmap(vcpu_e500);
303 }
304 
305 /* TID must be supplied by the caller */
306 static void kvmppc_e500_setup_stlbe(
307 	struct kvm_vcpu *vcpu,
308 	struct kvm_book3e_206_tlb_entry *gtlbe,
309 	int tsize, struct tlbe_ref *ref, u64 gvaddr,
310 	struct kvm_book3e_206_tlb_entry *stlbe)
311 {
312 	pfn_t pfn = ref->pfn;
313 	u32 pr = vcpu->arch.shared->msr & MSR_PR;
314 
315 	BUG_ON(!(ref->flags & E500_TLB_VALID));
316 
317 	/* Force IPROT=0 for all guest mappings. */
318 	stlbe->mas1 = MAS1_TSIZE(tsize) | get_tlb_sts(gtlbe) | MAS1_VALID;
319 	stlbe->mas2 = (gvaddr & MAS2_EPN) |
320 		      e500_shadow_mas2_attrib(gtlbe->mas2, pr);
321 	stlbe->mas7_3 = ((u64)pfn << PAGE_SHIFT) |
322 			e500_shadow_mas3_attrib(gtlbe->mas7_3, pr);
323 
324 #ifdef CONFIG_KVM_BOOKE_HV
325 	stlbe->mas8 = MAS8_TGS | vcpu->kvm->arch.lpid;
326 #endif
327 }
328 
329 static inline int kvmppc_e500_shadow_map(struct kvmppc_vcpu_e500 *vcpu_e500,
330 	u64 gvaddr, gfn_t gfn, struct kvm_book3e_206_tlb_entry *gtlbe,
331 	int tlbsel, struct kvm_book3e_206_tlb_entry *stlbe,
332 	struct tlbe_ref *ref)
333 {
334 	struct kvm_memory_slot *slot;
335 	unsigned long pfn = 0; /* silence GCC warning */
336 	unsigned long hva;
337 	int pfnmap = 0;
338 	int tsize = BOOK3E_PAGESZ_4K;
339 	int ret = 0;
340 	unsigned long mmu_seq;
341 	struct kvm *kvm = vcpu_e500->vcpu.kvm;
342 
343 	/* used to check for invalidations in progress */
344 	mmu_seq = kvm->mmu_notifier_seq;
345 	smp_rmb();
346 
347 	/*
348 	 * Translate guest physical to true physical, acquiring
349 	 * a page reference if it is normal, non-reserved memory.
350 	 *
351 	 * gfn_to_memslot() must succeed because otherwise we wouldn't
352 	 * have gotten this far.  Eventually we should just pass the slot
353 	 * pointer through from the first lookup.
354 	 */
355 	slot = gfn_to_memslot(vcpu_e500->vcpu.kvm, gfn);
356 	hva = gfn_to_hva_memslot(slot, gfn);
357 
358 	if (tlbsel == 1) {
359 		struct vm_area_struct *vma;
360 		down_read(&current->mm->mmap_sem);
361 
362 		vma = find_vma(current->mm, hva);
363 		if (vma && hva >= vma->vm_start &&
364 		    (vma->vm_flags & VM_PFNMAP)) {
365 			/*
366 			 * This VMA is a physically contiguous region (e.g.
367 			 * /dev/mem) that bypasses normal Linux page
368 			 * management.  Find the overlap between the
369 			 * vma and the memslot.
370 			 */
371 
372 			unsigned long start, end;
373 			unsigned long slot_start, slot_end;
374 
375 			pfnmap = 1;
376 
377 			start = vma->vm_pgoff;
378 			end = start +
379 			      ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT);
380 
381 			pfn = start + ((hva - vma->vm_start) >> PAGE_SHIFT);
382 
383 			slot_start = pfn - (gfn - slot->base_gfn);
384 			slot_end = slot_start + slot->npages;
385 
386 			if (start < slot_start)
387 				start = slot_start;
388 			if (end > slot_end)
389 				end = slot_end;
390 
391 			tsize = (gtlbe->mas1 & MAS1_TSIZE_MASK) >>
392 				MAS1_TSIZE_SHIFT;
393 
394 			/*
395 			 * e500 doesn't implement the lowest tsize bit,
396 			 * or 1K pages.
397 			 */
398 			tsize = max(BOOK3E_PAGESZ_4K, tsize & ~1);
399 
400 			/*
401 			 * Now find the largest tsize (up to what the guest
402 			 * requested) that will cover gfn, stay within the
403 			 * range, and for which gfn and pfn are mutually
404 			 * aligned.
405 			 */
406 
407 			for (; tsize > BOOK3E_PAGESZ_4K; tsize -= 2) {
408 				unsigned long gfn_start, gfn_end, tsize_pages;
409 				tsize_pages = 1 << (tsize - 2);
410 
411 				gfn_start = gfn & ~(tsize_pages - 1);
412 				gfn_end = gfn_start + tsize_pages;
413 
414 				if (gfn_start + pfn - gfn < start)
415 					continue;
416 				if (gfn_end + pfn - gfn > end)
417 					continue;
418 				if ((gfn & (tsize_pages - 1)) !=
419 				    (pfn & (tsize_pages - 1)))
420 					continue;
421 
422 				gvaddr &= ~((tsize_pages << PAGE_SHIFT) - 1);
423 				pfn &= ~(tsize_pages - 1);
424 				break;
425 			}
426 		} else if (vma && hva >= vma->vm_start &&
427 			   (vma->vm_flags & VM_HUGETLB)) {
428 			unsigned long psize = vma_kernel_pagesize(vma);
429 
430 			tsize = (gtlbe->mas1 & MAS1_TSIZE_MASK) >>
431 				MAS1_TSIZE_SHIFT;
432 
433 			/*
434 			 * Take the largest page size that satisfies both host
435 			 * and guest mapping
436 			 */
437 			tsize = min(__ilog2(psize) - 10, tsize);
438 
439 			/*
440 			 * e500 doesn't implement the lowest tsize bit,
441 			 * or 1K pages.
442 			 */
443 			tsize = max(BOOK3E_PAGESZ_4K, tsize & ~1);
444 		}
445 
446 		up_read(&current->mm->mmap_sem);
447 	}
448 
449 	if (likely(!pfnmap)) {
450 		unsigned long tsize_pages = 1 << (tsize + 10 - PAGE_SHIFT);
451 		pfn = gfn_to_pfn_memslot(slot, gfn);
452 		if (is_error_noslot_pfn(pfn)) {
453 			printk(KERN_ERR "Couldn't get real page for gfn %lx!\n",
454 					(long)gfn);
455 			return -EINVAL;
456 		}
457 
458 		/* Align guest and physical address to page map boundaries */
459 		pfn &= ~(tsize_pages - 1);
460 		gvaddr &= ~((tsize_pages << PAGE_SHIFT) - 1);
461 	}
462 
463 	spin_lock(&kvm->mmu_lock);
464 	if (mmu_notifier_retry(kvm, mmu_seq)) {
465 		ret = -EAGAIN;
466 		goto out;
467 	}
468 
469 	kvmppc_e500_ref_setup(ref, gtlbe, pfn);
470 
471 	kvmppc_e500_setup_stlbe(&vcpu_e500->vcpu, gtlbe, tsize,
472 				ref, gvaddr, stlbe);
473 
474 	/* Clear i-cache for new pages */
475 	kvmppc_mmu_flush_icache(pfn);
476 
477 out:
478 	spin_unlock(&kvm->mmu_lock);
479 
480 	/* Drop refcount on page, so that mmu notifiers can clear it */
481 	kvm_release_pfn_clean(pfn);
482 
483 	return ret;
484 }
485 
486 /* XXX only map the one-one case, for now use TLB0 */
487 static int kvmppc_e500_tlb0_map(struct kvmppc_vcpu_e500 *vcpu_e500, int esel,
488 				struct kvm_book3e_206_tlb_entry *stlbe)
489 {
490 	struct kvm_book3e_206_tlb_entry *gtlbe;
491 	struct tlbe_ref *ref;
492 	int stlbsel = 0;
493 	int sesel = 0;
494 	int r;
495 
496 	gtlbe = get_entry(vcpu_e500, 0, esel);
497 	ref = &vcpu_e500->gtlb_priv[0][esel].ref;
498 
499 	r = kvmppc_e500_shadow_map(vcpu_e500, get_tlb_eaddr(gtlbe),
500 			get_tlb_raddr(gtlbe) >> PAGE_SHIFT,
501 			gtlbe, 0, stlbe, ref);
502 	if (r)
503 		return r;
504 
505 	write_stlbe(vcpu_e500, gtlbe, stlbe, stlbsel, sesel);
506 
507 	return 0;
508 }
509 
510 static int kvmppc_e500_tlb1_map_tlb1(struct kvmppc_vcpu_e500 *vcpu_e500,
511 				     struct tlbe_ref *ref,
512 				     int esel)
513 {
514 	unsigned int sesel = vcpu_e500->host_tlb1_nv++;
515 
516 	if (unlikely(vcpu_e500->host_tlb1_nv >= tlb1_max_shadow_size()))
517 		vcpu_e500->host_tlb1_nv = 0;
518 
519 	if (vcpu_e500->h2g_tlb1_rmap[sesel]) {
520 		unsigned int idx = vcpu_e500->h2g_tlb1_rmap[sesel] - 1;
521 		vcpu_e500->g2h_tlb1_map[idx] &= ~(1ULL << sesel);
522 	}
523 
524 	vcpu_e500->gtlb_priv[1][esel].ref.flags |= E500_TLB_BITMAP;
525 	vcpu_e500->g2h_tlb1_map[esel] |= (u64)1 << sesel;
526 	vcpu_e500->h2g_tlb1_rmap[sesel] = esel + 1;
527 	WARN_ON(!(ref->flags & E500_TLB_VALID));
528 
529 	return sesel;
530 }
531 
532 /* Caller must ensure that the specified guest TLB entry is safe to insert into
533  * the shadow TLB. */
534 /* For both one-one and one-to-many */
535 static int kvmppc_e500_tlb1_map(struct kvmppc_vcpu_e500 *vcpu_e500,
536 		u64 gvaddr, gfn_t gfn, struct kvm_book3e_206_tlb_entry *gtlbe,
537 		struct kvm_book3e_206_tlb_entry *stlbe, int esel)
538 {
539 	struct tlbe_ref *ref = &vcpu_e500->gtlb_priv[1][esel].ref;
540 	int sesel;
541 	int r;
542 
543 	r = kvmppc_e500_shadow_map(vcpu_e500, gvaddr, gfn, gtlbe, 1, stlbe,
544 				   ref);
545 	if (r)
546 		return r;
547 
548 	/* Use TLB0 when we can only map a page with 4k */
549 	if (get_tlb_tsize(stlbe) == BOOK3E_PAGESZ_4K) {
550 		vcpu_e500->gtlb_priv[1][esel].ref.flags |= E500_TLB_TLB0;
551 		write_stlbe(vcpu_e500, gtlbe, stlbe, 0, 0);
552 		return 0;
553 	}
554 
555 	/* Otherwise map into TLB1 */
556 	sesel = kvmppc_e500_tlb1_map_tlb1(vcpu_e500, ref, esel);
557 	write_stlbe(vcpu_e500, gtlbe, stlbe, 1, sesel);
558 
559 	return 0;
560 }
561 
562 void kvmppc_mmu_map(struct kvm_vcpu *vcpu, u64 eaddr, gpa_t gpaddr,
563 		    unsigned int index)
564 {
565 	struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
566 	struct tlbe_priv *priv;
567 	struct kvm_book3e_206_tlb_entry *gtlbe, stlbe;
568 	int tlbsel = tlbsel_of(index);
569 	int esel = esel_of(index);
570 
571 	gtlbe = get_entry(vcpu_e500, tlbsel, esel);
572 
573 	switch (tlbsel) {
574 	case 0:
575 		priv = &vcpu_e500->gtlb_priv[tlbsel][esel];
576 
577 		/* Triggers after clear_tlb_privs or on initial mapping */
578 		if (!(priv->ref.flags & E500_TLB_VALID)) {
579 			kvmppc_e500_tlb0_map(vcpu_e500, esel, &stlbe);
580 		} else {
581 			kvmppc_e500_setup_stlbe(vcpu, gtlbe, BOOK3E_PAGESZ_4K,
582 						&priv->ref, eaddr, &stlbe);
583 			write_stlbe(vcpu_e500, gtlbe, &stlbe, 0, 0);
584 		}
585 		break;
586 
587 	case 1: {
588 		gfn_t gfn = gpaddr >> PAGE_SHIFT;
589 		kvmppc_e500_tlb1_map(vcpu_e500, eaddr, gfn, gtlbe, &stlbe,
590 				     esel);
591 		break;
592 	}
593 
594 	default:
595 		BUG();
596 		break;
597 	}
598 }
599 
600 /************* MMU Notifiers *************/
601 
602 int kvm_unmap_hva(struct kvm *kvm, unsigned long hva)
603 {
604 	trace_kvm_unmap_hva(hva);
605 
606 	/*
607 	 * Flush all shadow tlb entries everywhere. This is slow, but
608 	 * we are 100% sure that we catch the to be unmapped page
609 	 */
610 	kvm_flush_remote_tlbs(kvm);
611 
612 	return 0;
613 }
614 
615 int kvm_unmap_hva_range(struct kvm *kvm, unsigned long start, unsigned long end)
616 {
617 	/* kvm_unmap_hva flushes everything anyways */
618 	kvm_unmap_hva(kvm, start);
619 
620 	return 0;
621 }
622 
623 int kvm_age_hva(struct kvm *kvm, unsigned long hva)
624 {
625 	/* XXX could be more clever ;) */
626 	return 0;
627 }
628 
629 int kvm_test_age_hva(struct kvm *kvm, unsigned long hva)
630 {
631 	/* XXX could be more clever ;) */
632 	return 0;
633 }
634 
635 void kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte)
636 {
637 	/* The page will get remapped properly on its next fault */
638 	kvm_unmap_hva(kvm, hva);
639 }
640 
641 /*****************************************/
642 
643 int e500_mmu_host_init(struct kvmppc_vcpu_e500 *vcpu_e500)
644 {
645 	host_tlb_params[0].entries = mfspr(SPRN_TLB0CFG) & TLBnCFG_N_ENTRY;
646 	host_tlb_params[1].entries = mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY;
647 
648 	/*
649 	 * This should never happen on real e500 hardware, but is
650 	 * architecturally possible -- e.g. in some weird nested
651 	 * virtualization case.
652 	 */
653 	if (host_tlb_params[0].entries == 0 ||
654 	    host_tlb_params[1].entries == 0) {
655 		pr_err("%s: need to know host tlb size\n", __func__);
656 		return -ENODEV;
657 	}
658 
659 	host_tlb_params[0].ways = (mfspr(SPRN_TLB0CFG) & TLBnCFG_ASSOC) >>
660 				  TLBnCFG_ASSOC_SHIFT;
661 	host_tlb_params[1].ways = host_tlb_params[1].entries;
662 
663 	if (!is_power_of_2(host_tlb_params[0].entries) ||
664 	    !is_power_of_2(host_tlb_params[0].ways) ||
665 	    host_tlb_params[0].entries < host_tlb_params[0].ways ||
666 	    host_tlb_params[0].ways == 0) {
667 		pr_err("%s: bad tlb0 host config: %u entries %u ways\n",
668 		       __func__, host_tlb_params[0].entries,
669 		       host_tlb_params[0].ways);
670 		return -ENODEV;
671 	}
672 
673 	host_tlb_params[0].sets =
674 		host_tlb_params[0].entries / host_tlb_params[0].ways;
675 	host_tlb_params[1].sets = 1;
676 
677 	vcpu_e500->h2g_tlb1_rmap = kzalloc(sizeof(unsigned int) *
678 					   host_tlb_params[1].entries,
679 					   GFP_KERNEL);
680 	if (!vcpu_e500->h2g_tlb1_rmap)
681 		return -EINVAL;
682 
683 	return 0;
684 }
685 
686 void e500_mmu_host_uninit(struct kvmppc_vcpu_e500 *vcpu_e500)
687 {
688 	kfree(vcpu_e500->h2g_tlb1_rmap);
689 }
690