Commit | Line | Data |
---|---|---|
54738c09 DG |
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 | * Copyright 2011 David Gibson, IBM Corporation <dwg@au1.ibm.com> | |
d3695aa4 | 17 | * Copyright 2016 Alexey Kardashevskiy, IBM Corporation <aik@au1.ibm.com> |
54738c09 DG |
18 | */ |
19 | ||
20 | #include <linux/types.h> | |
21 | #include <linux/string.h> | |
22 | #include <linux/kvm.h> | |
23 | #include <linux/kvm_host.h> | |
24 | #include <linux/highmem.h> | |
25 | #include <linux/gfp.h> | |
26 | #include <linux/slab.h> | |
27 | #include <linux/hugetlb.h> | |
28 | #include <linux/list.h> | |
29 | ||
30 | #include <asm/tlbflush.h> | |
31 | #include <asm/kvm_ppc.h> | |
32 | #include <asm/kvm_book3s.h> | |
f64e8084 | 33 | #include <asm/book3s/64/mmu-hash.h> |
d3695aa4 | 34 | #include <asm/mmu_context.h> |
54738c09 DG |
35 | #include <asm/hvcall.h> |
36 | #include <asm/synch.h> | |
37 | #include <asm/ppc-opcode.h> | |
38 | #include <asm/kvm_host.h> | |
39 | #include <asm/udbg.h> | |
fcbb2ce6 | 40 | #include <asm/iommu.h> |
5ee7af18 | 41 | #include <asm/tce.h> |
d3695aa4 | 42 | #include <asm/iommu.h> |
54738c09 DG |
43 | |
44 | #define TCES_PER_PAGE (PAGE_SIZE / sizeof(u64)) | |
45 | ||
fcbb2ce6 AK |
46 | /* |
47 | * Finds a TCE table descriptor by LIOBN. | |
48 | * | |
49 | * WARNING: This will be called in real or virtual mode on HV KVM and virtual | |
50 | * mode on PR KVM | |
51 | */ | |
d3695aa4 | 52 | struct kvmppc_spapr_tce_table *kvmppc_find_table(struct kvm_vcpu *vcpu, |
fcbb2ce6 AK |
53 | unsigned long liobn) |
54 | { | |
55 | struct kvm *kvm = vcpu->kvm; | |
56 | struct kvmppc_spapr_tce_table *stt; | |
57 | ||
366baf28 | 58 | list_for_each_entry_lockless(stt, &kvm->arch.spapr_tce_tables, list) |
fcbb2ce6 AK |
59 | if (stt->liobn == liobn) |
60 | return stt; | |
61 | ||
62 | return NULL; | |
63 | } | |
d3695aa4 | 64 | EXPORT_SYMBOL_GPL(kvmppc_find_table); |
fcbb2ce6 AK |
65 | |
66 | /* | |
67 | * Validates IO address. | |
68 | * | |
69 | * WARNING: This will be called in real-mode on HV KVM and virtual | |
70 | * mode on PR KVM | |
71 | */ | |
5ee7af18 | 72 | long kvmppc_ioba_validate(struct kvmppc_spapr_tce_table *stt, |
fcbb2ce6 AK |
73 | unsigned long ioba, unsigned long npages) |
74 | { | |
fe26e527 AK |
75 | unsigned long mask = (1ULL << stt->page_shift) - 1; |
76 | unsigned long idx = ioba >> stt->page_shift; | |
fcbb2ce6 | 77 | |
14f853f1 AK |
78 | if ((ioba & mask) || (idx < stt->offset) || |
79 | (idx - stt->offset + npages > stt->size) || | |
80 | (idx + npages < idx)) | |
fcbb2ce6 AK |
81 | return H_PARAMETER; |
82 | ||
83 | return H_SUCCESS; | |
84 | } | |
5ee7af18 AK |
85 | EXPORT_SYMBOL_GPL(kvmppc_ioba_validate); |
86 | ||
87 | /* | |
88 | * Validates TCE address. | |
89 | * At the moment flags and page mask are validated. | |
90 | * As the host kernel does not access those addresses (just puts them | |
91 | * to the table and user space is supposed to process them), we can skip | |
92 | * checking other things (such as TCE is a guest RAM address or the page | |
93 | * was actually allocated). | |
94 | * | |
95 | * WARNING: This will be called in real-mode on HV KVM and virtual | |
96 | * mode on PR KVM | |
97 | */ | |
98 | long kvmppc_tce_validate(struct kvmppc_spapr_tce_table *stt, unsigned long tce) | |
99 | { | |
fe26e527 AK |
100 | unsigned long page_mask = ~((1ULL << stt->page_shift) - 1); |
101 | unsigned long mask = ~(page_mask | TCE_PCI_WRITE | TCE_PCI_READ); | |
5ee7af18 AK |
102 | |
103 | if (tce & mask) | |
104 | return H_PARAMETER; | |
105 | ||
106 | return H_SUCCESS; | |
107 | } | |
108 | EXPORT_SYMBOL_GPL(kvmppc_tce_validate); | |
109 | ||
110 | /* Note on the use of page_address() in real mode, | |
111 | * | |
112 | * It is safe to use page_address() in real mode on ppc64 because | |
113 | * page_address() is always defined as lowmem_page_address() | |
114 | * which returns __va(PFN_PHYS(page_to_pfn(page))) which is arithmetic | |
115 | * operation and does not access page struct. | |
116 | * | |
117 | * Theoretically page_address() could be defined different | |
118 | * but either WANT_PAGE_VIRTUAL or HASHED_PAGE_VIRTUAL | |
119 | * would have to be enabled. | |
120 | * WANT_PAGE_VIRTUAL is never enabled on ppc32/ppc64, | |
121 | * HASHED_PAGE_VIRTUAL could be enabled for ppc32 only and only | |
122 | * if CONFIG_HIGHMEM is defined. As CONFIG_SPARSEMEM_VMEMMAP | |
123 | * is not expected to be enabled on ppc32, page_address() | |
124 | * is safe for ppc32 as well. | |
125 | * | |
126 | * WARNING: This will be called in real-mode on HV KVM and virtual | |
127 | * mode on PR KVM | |
128 | */ | |
129 | static u64 *kvmppc_page_address(struct page *page) | |
130 | { | |
131 | #if defined(HASHED_PAGE_VIRTUAL) || defined(WANT_PAGE_VIRTUAL) | |
132 | #error TODO: fix to avoid page_address() here | |
133 | #endif | |
134 | return (u64 *) page_address(page); | |
135 | } | |
136 | ||
137 | /* | |
138 | * Handles TCE requests for emulated devices. | |
139 | * Puts guest TCE values to the table and expects user space to convert them. | |
140 | * Called in both real and virtual modes. | |
141 | * Cannot fail so kvmppc_tce_validate must be called before it. | |
142 | * | |
143 | * WARNING: This will be called in real-mode on HV KVM and virtual | |
144 | * mode on PR KVM | |
145 | */ | |
146 | void kvmppc_tce_put(struct kvmppc_spapr_tce_table *stt, | |
147 | unsigned long idx, unsigned long tce) | |
148 | { | |
149 | struct page *page; | |
150 | u64 *tbl; | |
151 | ||
14f853f1 | 152 | idx -= stt->offset; |
5ee7af18 AK |
153 | page = stt->pages[idx / TCES_PER_PAGE]; |
154 | tbl = kvmppc_page_address(page); | |
155 | ||
156 | tbl[idx % TCES_PER_PAGE] = tce; | |
157 | } | |
158 | EXPORT_SYMBOL_GPL(kvmppc_tce_put); | |
fcbb2ce6 | 159 | |
d3695aa4 AK |
160 | long kvmppc_gpa_to_ua(struct kvm *kvm, unsigned long gpa, |
161 | unsigned long *ua, unsigned long **prmap) | |
162 | { | |
163 | unsigned long gfn = gpa >> PAGE_SHIFT; | |
164 | struct kvm_memory_slot *memslot; | |
165 | ||
166 | memslot = search_memslots(kvm_memslots(kvm), gfn); | |
167 | if (!memslot) | |
168 | return -EINVAL; | |
169 | ||
170 | *ua = __gfn_to_hva_memslot(memslot, gfn) | | |
171 | (gpa & ~(PAGE_MASK | TCE_PCI_READ | TCE_PCI_WRITE)); | |
172 | ||
173 | #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE | |
174 | if (prmap) | |
175 | *prmap = &memslot->arch.rmap[gfn - memslot->base_gfn]; | |
176 | #endif | |
177 | ||
178 | return 0; | |
179 | } | |
180 | EXPORT_SYMBOL_GPL(kvmppc_gpa_to_ua); | |
181 | ||
182 | #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE | |
31217db7 AK |
183 | long kvmppc_rm_h_put_tce(struct kvm_vcpu *vcpu, unsigned long liobn, |
184 | unsigned long ioba, unsigned long tce) | |
54738c09 | 185 | { |
fcbb2ce6 AK |
186 | struct kvmppc_spapr_tce_table *stt = kvmppc_find_table(vcpu, liobn); |
187 | long ret; | |
54738c09 DG |
188 | |
189 | /* udbg_printf("H_PUT_TCE(): liobn=0x%lx ioba=0x%lx, tce=0x%lx\n", */ | |
190 | /* liobn, ioba, tce); */ | |
191 | ||
fcbb2ce6 AK |
192 | if (!stt) |
193 | return H_TOO_HARD; | |
194 | ||
195 | ret = kvmppc_ioba_validate(stt, ioba, 1); | |
196 | if (ret != H_SUCCESS) | |
197 | return ret; | |
198 | ||
5ee7af18 AK |
199 | ret = kvmppc_tce_validate(stt, tce); |
200 | if (ret != H_SUCCESS) | |
201 | return ret; | |
fcbb2ce6 | 202 | |
fe26e527 | 203 | kvmppc_tce_put(stt, ioba >> stt->page_shift, tce); |
fcbb2ce6 AK |
204 | |
205 | return H_SUCCESS; | |
54738c09 | 206 | } |
69e9fbb2 | 207 | |
d3695aa4 AK |
208 | static long kvmppc_rm_ua_to_hpa(struct kvm_vcpu *vcpu, |
209 | unsigned long ua, unsigned long *phpa) | |
210 | { | |
211 | pte_t *ptep, pte; | |
212 | unsigned shift = 0; | |
213 | ||
214 | ptep = __find_linux_pte_or_hugepte(vcpu->arch.pgdir, ua, NULL, &shift); | |
215 | if (!ptep || !pte_present(*ptep)) | |
216 | return -ENXIO; | |
217 | pte = *ptep; | |
218 | ||
219 | if (!shift) | |
220 | shift = PAGE_SHIFT; | |
221 | ||
222 | /* Avoid handling anything potentially complicated in realmode */ | |
223 | if (shift > PAGE_SHIFT) | |
224 | return -EAGAIN; | |
225 | ||
226 | if (!pte_young(pte)) | |
227 | return -EAGAIN; | |
228 | ||
229 | *phpa = (pte_pfn(pte) << PAGE_SHIFT) | (ua & ((1ULL << shift) - 1)) | | |
230 | (ua & ~PAGE_MASK); | |
231 | ||
232 | return 0; | |
233 | } | |
234 | ||
235 | long kvmppc_rm_h_put_tce_indirect(struct kvm_vcpu *vcpu, | |
236 | unsigned long liobn, unsigned long ioba, | |
237 | unsigned long tce_list, unsigned long npages) | |
238 | { | |
239 | struct kvmppc_spapr_tce_table *stt; | |
240 | long i, ret = H_SUCCESS; | |
241 | unsigned long tces, entry, ua = 0; | |
242 | unsigned long *rmap = NULL; | |
243 | ||
244 | stt = kvmppc_find_table(vcpu, liobn); | |
245 | if (!stt) | |
246 | return H_TOO_HARD; | |
247 | ||
fe26e527 | 248 | entry = ioba >> stt->page_shift; |
d3695aa4 AK |
249 | /* |
250 | * The spec says that the maximum size of the list is 512 TCEs | |
251 | * so the whole table addressed resides in 4K page | |
252 | */ | |
253 | if (npages > 512) | |
254 | return H_PARAMETER; | |
255 | ||
256 | if (tce_list & (SZ_4K - 1)) | |
257 | return H_PARAMETER; | |
258 | ||
259 | ret = kvmppc_ioba_validate(stt, ioba, npages); | |
260 | if (ret != H_SUCCESS) | |
261 | return ret; | |
262 | ||
263 | if (kvmppc_gpa_to_ua(vcpu->kvm, tce_list, &ua, &rmap)) | |
264 | return H_TOO_HARD; | |
265 | ||
266 | rmap = (void *) vmalloc_to_phys(rmap); | |
267 | ||
268 | /* | |
269 | * Synchronize with the MMU notifier callbacks in | |
270 | * book3s_64_mmu_hv.c (kvm_unmap_hva_hv etc.). | |
271 | * While we have the rmap lock, code running on other CPUs | |
272 | * cannot finish unmapping the host real page that backs | |
273 | * this guest real page, so we are OK to access the host | |
274 | * real page. | |
275 | */ | |
276 | lock_rmap(rmap); | |
277 | if (kvmppc_rm_ua_to_hpa(vcpu, ua, &tces)) { | |
278 | ret = H_TOO_HARD; | |
279 | goto unlock_exit; | |
280 | } | |
281 | ||
282 | for (i = 0; i < npages; ++i) { | |
283 | unsigned long tce = be64_to_cpu(((u64 *)tces)[i]); | |
284 | ||
285 | ret = kvmppc_tce_validate(stt, tce); | |
286 | if (ret != H_SUCCESS) | |
287 | goto unlock_exit; | |
288 | ||
289 | kvmppc_tce_put(stt, entry + i, tce); | |
290 | } | |
291 | ||
292 | unlock_exit: | |
293 | unlock_rmap(rmap); | |
294 | ||
295 | return ret; | |
296 | } | |
297 | ||
31217db7 | 298 | long kvmppc_rm_h_stuff_tce(struct kvm_vcpu *vcpu, |
d3695aa4 AK |
299 | unsigned long liobn, unsigned long ioba, |
300 | unsigned long tce_value, unsigned long npages) | |
301 | { | |
302 | struct kvmppc_spapr_tce_table *stt; | |
303 | long i, ret; | |
304 | ||
305 | stt = kvmppc_find_table(vcpu, liobn); | |
306 | if (!stt) | |
307 | return H_TOO_HARD; | |
308 | ||
309 | ret = kvmppc_ioba_validate(stt, ioba, npages); | |
310 | if (ret != H_SUCCESS) | |
311 | return ret; | |
312 | ||
313 | /* Check permission bits only to allow userspace poison TCE for debug */ | |
314 | if (tce_value & (TCE_PCI_WRITE | TCE_PCI_READ)) | |
315 | return H_PARAMETER; | |
316 | ||
fe26e527 AK |
317 | for (i = 0; i < npages; ++i, ioba += (1ULL << stt->page_shift)) |
318 | kvmppc_tce_put(stt, ioba >> stt->page_shift, tce_value); | |
d3695aa4 AK |
319 | |
320 | return H_SUCCESS; | |
321 | } | |
d3695aa4 | 322 | |
69e9fbb2 LD |
323 | long kvmppc_h_get_tce(struct kvm_vcpu *vcpu, unsigned long liobn, |
324 | unsigned long ioba) | |
325 | { | |
fcbb2ce6 AK |
326 | struct kvmppc_spapr_tce_table *stt = kvmppc_find_table(vcpu, liobn); |
327 | long ret; | |
328 | unsigned long idx; | |
329 | struct page *page; | |
330 | u64 *tbl; | |
69e9fbb2 | 331 | |
fcbb2ce6 AK |
332 | if (!stt) |
333 | return H_TOO_HARD; | |
69e9fbb2 | 334 | |
fcbb2ce6 AK |
335 | ret = kvmppc_ioba_validate(stt, ioba, 1); |
336 | if (ret != H_SUCCESS) | |
337 | return ret; | |
69e9fbb2 | 338 | |
14f853f1 | 339 | idx = (ioba >> stt->page_shift) - stt->offset; |
fcbb2ce6 AK |
340 | page = stt->pages[idx / TCES_PER_PAGE]; |
341 | tbl = (u64 *)page_address(page); | |
69e9fbb2 | 342 | |
fcbb2ce6 | 343 | vcpu->arch.gpr[4] = tbl[idx % TCES_PER_PAGE]; |
69e9fbb2 | 344 | |
fcbb2ce6 | 345 | return H_SUCCESS; |
69e9fbb2 LD |
346 | } |
347 | EXPORT_SYMBOL_GPL(kvmppc_h_get_tce); | |
d3695aa4 AK |
348 | |
349 | #endif /* KVM_BOOK3S_HV_POSSIBLE */ |