xref: /linux/arch/powerpc/include/asm/kvm_book3s_64.h (revision 4949009eb8d40a441dcddcd96e101e77d31cf1b2)
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 SUSE Linux Products GmbH 2010
16  *
17  * Authors: Alexander Graf <agraf@suse.de>
18  */
19 
20 #ifndef __ASM_KVM_BOOK3S_64_H__
21 #define __ASM_KVM_BOOK3S_64_H__
22 
23 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
24 static inline struct kvmppc_book3s_shadow_vcpu *svcpu_get(struct kvm_vcpu *vcpu)
25 {
26 	preempt_disable();
27 	return &get_paca()->shadow_vcpu;
28 }
29 
30 static inline void svcpu_put(struct kvmppc_book3s_shadow_vcpu *svcpu)
31 {
32 	preempt_enable();
33 }
34 #endif
35 
36 #define SPAPR_TCE_SHIFT		12
37 
38 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
39 #define KVM_DEFAULT_HPT_ORDER	24	/* 16MB HPT by default */
40 #endif
41 
42 #define VRMA_VSID	0x1ffffffUL	/* 1TB VSID reserved for VRMA */
43 
44 /*
45  * We use a lock bit in HPTE dword 0 to synchronize updates and
46  * accesses to each HPTE, and another bit to indicate non-present
47  * HPTEs.
48  */
49 #define HPTE_V_HVLOCK	0x40UL
50 #define HPTE_V_ABSENT	0x20UL
51 
52 /*
53  * We use this bit in the guest_rpte field of the revmap entry
54  * to indicate a modified HPTE.
55  */
56 #define HPTE_GR_MODIFIED	(1ul << 62)
57 
58 /* These bits are reserved in the guest view of the HPTE */
59 #define HPTE_GR_RESERVED	HPTE_GR_MODIFIED
60 
61 static inline long try_lock_hpte(__be64 *hpte, unsigned long bits)
62 {
63 	unsigned long tmp, old;
64 	__be64 be_lockbit, be_bits;
65 
66 	/*
67 	 * We load/store in native endian, but the HTAB is in big endian. If
68 	 * we byte swap all data we apply on the PTE we're implicitly correct
69 	 * again.
70 	 */
71 	be_lockbit = cpu_to_be64(HPTE_V_HVLOCK);
72 	be_bits = cpu_to_be64(bits);
73 
74 	asm volatile("	ldarx	%0,0,%2\n"
75 		     "	and.	%1,%0,%3\n"
76 		     "	bne	2f\n"
77 		     "	or	%0,%0,%4\n"
78 		     "  stdcx.	%0,0,%2\n"
79 		     "	beq+	2f\n"
80 		     "	mr	%1,%3\n"
81 		     "2:	isync"
82 		     : "=&r" (tmp), "=&r" (old)
83 		     : "r" (hpte), "r" (be_bits), "r" (be_lockbit)
84 		     : "cc", "memory");
85 	return old == 0;
86 }
87 
88 static inline int __hpte_actual_psize(unsigned int lp, int psize)
89 {
90 	int i, shift;
91 	unsigned int mask;
92 
93 	/* start from 1 ignoring MMU_PAGE_4K */
94 	for (i = 1; i < MMU_PAGE_COUNT; i++) {
95 
96 		/* invalid penc */
97 		if (mmu_psize_defs[psize].penc[i] == -1)
98 			continue;
99 		/*
100 		 * encoding bits per actual page size
101 		 *        PTE LP     actual page size
102 		 *    rrrr rrrz		>=8KB
103 		 *    rrrr rrzz		>=16KB
104 		 *    rrrr rzzz		>=32KB
105 		 *    rrrr zzzz		>=64KB
106 		 * .......
107 		 */
108 		shift = mmu_psize_defs[i].shift - LP_SHIFT;
109 		if (shift > LP_BITS)
110 			shift = LP_BITS;
111 		mask = (1 << shift) - 1;
112 		if ((lp & mask) == mmu_psize_defs[psize].penc[i])
113 			return i;
114 	}
115 	return -1;
116 }
117 
118 static inline unsigned long compute_tlbie_rb(unsigned long v, unsigned long r,
119 					     unsigned long pte_index)
120 {
121 	int b_psize = MMU_PAGE_4K, a_psize = MMU_PAGE_4K;
122 	unsigned int penc;
123 	unsigned long rb = 0, va_low, sllp;
124 	unsigned int lp = (r >> LP_SHIFT) & ((1 << LP_BITS) - 1);
125 
126 	if (v & HPTE_V_LARGE) {
127 		for (b_psize = 0; b_psize < MMU_PAGE_COUNT; b_psize++) {
128 
129 			/* valid entries have a shift value */
130 			if (!mmu_psize_defs[b_psize].shift)
131 				continue;
132 
133 			a_psize = __hpte_actual_psize(lp, b_psize);
134 			if (a_psize != -1)
135 				break;
136 		}
137 	}
138 	/*
139 	 * Ignore the top 14 bits of va
140 	 * v have top two bits covering segment size, hence move
141 	 * by 16 bits, Also clear the lower HPTE_V_AVPN_SHIFT (7) bits.
142 	 * AVA field in v also have the lower 23 bits ignored.
143 	 * For base page size 4K we need 14 .. 65 bits (so need to
144 	 * collect extra 11 bits)
145 	 * For others we need 14..14+i
146 	 */
147 	/* This covers 14..54 bits of va*/
148 	rb = (v & ~0x7fUL) << 16;		/* AVA field */
149 
150 	rb |= (v >> HPTE_V_SSIZE_SHIFT) << 8;	/*  B field */
151 	/*
152 	 * AVA in v had cleared lower 23 bits. We need to derive
153 	 * that from pteg index
154 	 */
155 	va_low = pte_index >> 3;
156 	if (v & HPTE_V_SECONDARY)
157 		va_low = ~va_low;
158 	/*
159 	 * get the vpn bits from va_low using reverse of hashing.
160 	 * In v we have va with 23 bits dropped and then left shifted
161 	 * HPTE_V_AVPN_SHIFT (7) bits. Now to find vsid we need
162 	 * right shift it with (SID_SHIFT - (23 - 7))
163 	 */
164 	if (!(v & HPTE_V_1TB_SEG))
165 		va_low ^= v >> (SID_SHIFT - 16);
166 	else
167 		va_low ^= v >> (SID_SHIFT_1T - 16);
168 	va_low &= 0x7ff;
169 
170 	switch (b_psize) {
171 	case MMU_PAGE_4K:
172 		sllp = ((mmu_psize_defs[a_psize].sllp & SLB_VSID_L) >> 6) |
173 			((mmu_psize_defs[a_psize].sllp & SLB_VSID_LP) >> 4);
174 		rb |= sllp << 5;	/*  AP field */
175 		rb |= (va_low & 0x7ff) << 12;	/* remaining 11 bits of AVA */
176 		break;
177 	default:
178 	{
179 		int aval_shift;
180 		/*
181 		 * remaining bits of AVA/LP fields
182 		 * Also contain the rr bits of LP
183 		 */
184 		rb |= (va_low << mmu_psize_defs[b_psize].shift) & 0x7ff000;
185 		/*
186 		 * Now clear not needed LP bits based on actual psize
187 		 */
188 		rb &= ~((1ul << mmu_psize_defs[a_psize].shift) - 1);
189 		/*
190 		 * AVAL field 58..77 - base_page_shift bits of va
191 		 * we have space for 58..64 bits, Missing bits should
192 		 * be zero filled. +1 is to take care of L bit shift
193 		 */
194 		aval_shift = 64 - (77 - mmu_psize_defs[b_psize].shift) + 1;
195 		rb |= ((va_low << aval_shift) & 0xfe);
196 
197 		rb |= 1;		/* L field */
198 		penc = mmu_psize_defs[b_psize].penc[a_psize];
199 		rb |= penc << 12;	/* LP field */
200 		break;
201 	}
202 	}
203 	rb |= (v >> 54) & 0x300;		/* B field */
204 	return rb;
205 }
206 
207 static inline unsigned long __hpte_page_size(unsigned long h, unsigned long l,
208 					     bool is_base_size)
209 {
210 
211 	int size, a_psize;
212 	/* Look at the 8 bit LP value */
213 	unsigned int lp = (l >> LP_SHIFT) & ((1 << LP_BITS) - 1);
214 
215 	/* only handle 4k, 64k and 16M pages for now */
216 	if (!(h & HPTE_V_LARGE))
217 		return 1ul << 12;
218 	else {
219 		for (size = 0; size < MMU_PAGE_COUNT; size++) {
220 			/* valid entries have a shift value */
221 			if (!mmu_psize_defs[size].shift)
222 				continue;
223 
224 			a_psize = __hpte_actual_psize(lp, size);
225 			if (a_psize != -1) {
226 				if (is_base_size)
227 					return 1ul << mmu_psize_defs[size].shift;
228 				return 1ul << mmu_psize_defs[a_psize].shift;
229 			}
230 		}
231 
232 	}
233 	return 0;
234 }
235 
236 static inline unsigned long hpte_page_size(unsigned long h, unsigned long l)
237 {
238 	return __hpte_page_size(h, l, 0);
239 }
240 
241 static inline unsigned long hpte_base_page_size(unsigned long h, unsigned long l)
242 {
243 	return __hpte_page_size(h, l, 1);
244 }
245 
246 static inline unsigned long hpte_rpn(unsigned long ptel, unsigned long psize)
247 {
248 	return ((ptel & HPTE_R_RPN) & ~(psize - 1)) >> PAGE_SHIFT;
249 }
250 
251 static inline int hpte_is_writable(unsigned long ptel)
252 {
253 	unsigned long pp = ptel & (HPTE_R_PP0 | HPTE_R_PP);
254 
255 	return pp != PP_RXRX && pp != PP_RXXX;
256 }
257 
258 static inline unsigned long hpte_make_readonly(unsigned long ptel)
259 {
260 	if ((ptel & HPTE_R_PP0) || (ptel & HPTE_R_PP) == PP_RWXX)
261 		ptel = (ptel & ~HPTE_R_PP) | PP_RXXX;
262 	else
263 		ptel |= PP_RXRX;
264 	return ptel;
265 }
266 
267 static inline int hpte_cache_flags_ok(unsigned long ptel, unsigned long io_type)
268 {
269 	unsigned int wimg = ptel & HPTE_R_WIMG;
270 
271 	/* Handle SAO */
272 	if (wimg == (HPTE_R_W | HPTE_R_I | HPTE_R_M) &&
273 	    cpu_has_feature(CPU_FTR_ARCH_206))
274 		wimg = HPTE_R_M;
275 
276 	if (!io_type)
277 		return wimg == HPTE_R_M;
278 
279 	return (wimg & (HPTE_R_W | HPTE_R_I)) == io_type;
280 }
281 
282 /*
283  * If it's present and writable, atomically set dirty and referenced bits and
284  * return the PTE, otherwise return 0. If we find a transparent hugepage
285  * and if it is marked splitting we return 0;
286  */
287 static inline pte_t kvmppc_read_update_linux_pte(pte_t *ptep, int writing,
288 						 unsigned int hugepage)
289 {
290 	pte_t old_pte, new_pte = __pte(0);
291 
292 	while (1) {
293 		old_pte = pte_val(*ptep);
294 		/*
295 		 * wait until _PAGE_BUSY is clear then set it atomically
296 		 */
297 		if (unlikely(old_pte & _PAGE_BUSY)) {
298 			cpu_relax();
299 			continue;
300 		}
301 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
302 		/* If hugepage and is trans splitting return None */
303 		if (unlikely(hugepage &&
304 			     pmd_trans_splitting(pte_pmd(old_pte))))
305 			return __pte(0);
306 #endif
307 		/* If pte is not present return None */
308 		if (unlikely(!(old_pte & _PAGE_PRESENT)))
309 			return __pte(0);
310 
311 		new_pte = pte_mkyoung(old_pte);
312 		if (writing && pte_write(old_pte))
313 			new_pte = pte_mkdirty(new_pte);
314 
315 		if (old_pte == __cmpxchg_u64((unsigned long *)ptep, old_pte,
316 					     new_pte))
317 			break;
318 	}
319 	return new_pte;
320 }
321 
322 
323 /* Return HPTE cache control bits corresponding to Linux pte bits */
324 static inline unsigned long hpte_cache_bits(unsigned long pte_val)
325 {
326 #if _PAGE_NO_CACHE == HPTE_R_I && _PAGE_WRITETHRU == HPTE_R_W
327 	return pte_val & (HPTE_R_W | HPTE_R_I);
328 #else
329 	return ((pte_val & _PAGE_NO_CACHE) ? HPTE_R_I : 0) +
330 		((pte_val & _PAGE_WRITETHRU) ? HPTE_R_W : 0);
331 #endif
332 }
333 
334 static inline bool hpte_read_permission(unsigned long pp, unsigned long key)
335 {
336 	if (key)
337 		return PP_RWRX <= pp && pp <= PP_RXRX;
338 	return 1;
339 }
340 
341 static inline bool hpte_write_permission(unsigned long pp, unsigned long key)
342 {
343 	if (key)
344 		return pp == PP_RWRW;
345 	return pp <= PP_RWRW;
346 }
347 
348 static inline int hpte_get_skey_perm(unsigned long hpte_r, unsigned long amr)
349 {
350 	unsigned long skey;
351 
352 	skey = ((hpte_r & HPTE_R_KEY_HI) >> 57) |
353 		((hpte_r & HPTE_R_KEY_LO) >> 9);
354 	return (amr >> (62 - 2 * skey)) & 3;
355 }
356 
357 static inline void lock_rmap(unsigned long *rmap)
358 {
359 	do {
360 		while (test_bit(KVMPPC_RMAP_LOCK_BIT, rmap))
361 			cpu_relax();
362 	} while (test_and_set_bit_lock(KVMPPC_RMAP_LOCK_BIT, rmap));
363 }
364 
365 static inline void unlock_rmap(unsigned long *rmap)
366 {
367 	__clear_bit_unlock(KVMPPC_RMAP_LOCK_BIT, rmap);
368 }
369 
370 static inline bool slot_is_aligned(struct kvm_memory_slot *memslot,
371 				   unsigned long pagesize)
372 {
373 	unsigned long mask = (pagesize >> PAGE_SHIFT) - 1;
374 
375 	if (pagesize <= PAGE_SIZE)
376 		return 1;
377 	return !(memslot->base_gfn & mask) && !(memslot->npages & mask);
378 }
379 
380 /*
381  * This works for 4k, 64k and 16M pages on POWER7,
382  * and 4k and 16M pages on PPC970.
383  */
384 static inline unsigned long slb_pgsize_encoding(unsigned long psize)
385 {
386 	unsigned long senc = 0;
387 
388 	if (psize > 0x1000) {
389 		senc = SLB_VSID_L;
390 		if (psize == 0x10000)
391 			senc |= SLB_VSID_LP_01;
392 	}
393 	return senc;
394 }
395 
396 static inline int is_vrma_hpte(unsigned long hpte_v)
397 {
398 	return (hpte_v & ~0xffffffUL) ==
399 		(HPTE_V_1TB_SEG | (VRMA_VSID << (40 - 16)));
400 }
401 
402 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
403 /*
404  * Note modification of an HPTE; set the HPTE modified bit
405  * if anyone is interested.
406  */
407 static inline void note_hpte_modification(struct kvm *kvm,
408 					  struct revmap_entry *rev)
409 {
410 	if (atomic_read(&kvm->arch.hpte_mod_interest))
411 		rev->guest_rpte |= HPTE_GR_MODIFIED;
412 }
413 
414 /*
415  * Like kvm_memslots(), but for use in real mode when we can't do
416  * any RCU stuff (since the secondary threads are offline from the
417  * kernel's point of view), and we can't print anything.
418  * Thus we use rcu_dereference_raw() rather than rcu_dereference_check().
419  */
420 static inline struct kvm_memslots *kvm_memslots_raw(struct kvm *kvm)
421 {
422 	return rcu_dereference_raw_notrace(kvm->memslots);
423 }
424 
425 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
426 
427 #endif /* __ASM_KVM_BOOK3S_64_H__ */
428