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> | |
5c35a02c | 29 | #include <linux/stringify.h> |
54738c09 | 30 | |
54738c09 DG |
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> |
94171b19 | 42 | #include <asm/pte-walk.h> |
54738c09 | 43 | |
121f80ba AK |
44 | #ifdef CONFIG_BUG |
45 | ||
46 | #define WARN_ON_ONCE_RM(condition) ({ \ | |
47 | static bool __section(.data.unlikely) __warned; \ | |
48 | int __ret_warn_once = !!(condition); \ | |
49 | \ | |
50 | if (unlikely(__ret_warn_once && !__warned)) { \ | |
51 | __warned = true; \ | |
52 | pr_err("WARN_ON_ONCE_RM: (%s) at %s:%u\n", \ | |
53 | __stringify(condition), \ | |
54 | __func__, __LINE__); \ | |
55 | dump_stack(); \ | |
56 | } \ | |
57 | unlikely(__ret_warn_once); \ | |
58 | }) | |
59 | ||
60 | #else | |
61 | ||
62 | #define WARN_ON_ONCE_RM(condition) ({ \ | |
63 | int __ret_warn_on = !!(condition); \ | |
64 | unlikely(__ret_warn_on); \ | |
65 | }) | |
66 | ||
67 | #endif | |
68 | ||
54738c09 DG |
69 | #define TCES_PER_PAGE (PAGE_SIZE / sizeof(u64)) |
70 | ||
fcbb2ce6 AK |
71 | /* |
72 | * Finds a TCE table descriptor by LIOBN. | |
73 | * | |
74 | * WARNING: This will be called in real or virtual mode on HV KVM and virtual | |
75 | * mode on PR KVM | |
76 | */ | |
503bfcbe | 77 | struct kvmppc_spapr_tce_table *kvmppc_find_table(struct kvm *kvm, |
fcbb2ce6 AK |
78 | unsigned long liobn) |
79 | { | |
fcbb2ce6 AK |
80 | struct kvmppc_spapr_tce_table *stt; |
81 | ||
366baf28 | 82 | list_for_each_entry_lockless(stt, &kvm->arch.spapr_tce_tables, list) |
fcbb2ce6 AK |
83 | if (stt->liobn == liobn) |
84 | return stt; | |
85 | ||
86 | return NULL; | |
87 | } | |
d3695aa4 | 88 | EXPORT_SYMBOL_GPL(kvmppc_find_table); |
fcbb2ce6 | 89 | |
5ee7af18 AK |
90 | /* |
91 | * Validates TCE address. | |
92 | * At the moment flags and page mask are validated. | |
93 | * As the host kernel does not access those addresses (just puts them | |
94 | * to the table and user space is supposed to process them), we can skip | |
95 | * checking other things (such as TCE is a guest RAM address or the page | |
96 | * was actually allocated). | |
97 | * | |
98 | * WARNING: This will be called in real-mode on HV KVM and virtual | |
99 | * mode on PR KVM | |
100 | */ | |
101 | long kvmppc_tce_validate(struct kvmppc_spapr_tce_table *stt, unsigned long tce) | |
102 | { | |
b1af23d8 AK |
103 | unsigned long gpa = tce & ~(TCE_PCI_READ | TCE_PCI_WRITE); |
104 | enum dma_data_direction dir = iommu_tce_direction(tce); | |
105 | ||
106 | /* Allow userspace to poison TCE table */ | |
107 | if (dir == DMA_NONE) | |
108 | return H_SUCCESS; | |
5ee7af18 | 109 | |
b1af23d8 | 110 | if (iommu_tce_check_gpa(stt->page_shift, gpa)) |
5ee7af18 AK |
111 | return H_PARAMETER; |
112 | ||
113 | return H_SUCCESS; | |
114 | } | |
115 | EXPORT_SYMBOL_GPL(kvmppc_tce_validate); | |
116 | ||
117 | /* Note on the use of page_address() in real mode, | |
118 | * | |
119 | * It is safe to use page_address() in real mode on ppc64 because | |
120 | * page_address() is always defined as lowmem_page_address() | |
121 | * which returns __va(PFN_PHYS(page_to_pfn(page))) which is arithmetic | |
122 | * operation and does not access page struct. | |
123 | * | |
124 | * Theoretically page_address() could be defined different | |
125 | * but either WANT_PAGE_VIRTUAL or HASHED_PAGE_VIRTUAL | |
126 | * would have to be enabled. | |
127 | * WANT_PAGE_VIRTUAL is never enabled on ppc32/ppc64, | |
128 | * HASHED_PAGE_VIRTUAL could be enabled for ppc32 only and only | |
129 | * if CONFIG_HIGHMEM is defined. As CONFIG_SPARSEMEM_VMEMMAP | |
130 | * is not expected to be enabled on ppc32, page_address() | |
131 | * is safe for ppc32 as well. | |
132 | * | |
133 | * WARNING: This will be called in real-mode on HV KVM and virtual | |
134 | * mode on PR KVM | |
135 | */ | |
136 | static u64 *kvmppc_page_address(struct page *page) | |
137 | { | |
138 | #if defined(HASHED_PAGE_VIRTUAL) || defined(WANT_PAGE_VIRTUAL) | |
139 | #error TODO: fix to avoid page_address() here | |
140 | #endif | |
141 | return (u64 *) page_address(page); | |
142 | } | |
143 | ||
144 | /* | |
145 | * Handles TCE requests for emulated devices. | |
146 | * Puts guest TCE values to the table and expects user space to convert them. | |
147 | * Called in both real and virtual modes. | |
148 | * Cannot fail so kvmppc_tce_validate must be called before it. | |
149 | * | |
150 | * WARNING: This will be called in real-mode on HV KVM and virtual | |
151 | * mode on PR KVM | |
152 | */ | |
153 | void kvmppc_tce_put(struct kvmppc_spapr_tce_table *stt, | |
154 | unsigned long idx, unsigned long tce) | |
155 | { | |
156 | struct page *page; | |
157 | u64 *tbl; | |
158 | ||
14f853f1 | 159 | idx -= stt->offset; |
5ee7af18 AK |
160 | page = stt->pages[idx / TCES_PER_PAGE]; |
161 | tbl = kvmppc_page_address(page); | |
162 | ||
163 | tbl[idx % TCES_PER_PAGE] = tce; | |
164 | } | |
165 | EXPORT_SYMBOL_GPL(kvmppc_tce_put); | |
fcbb2ce6 | 166 | |
d3695aa4 AK |
167 | long kvmppc_gpa_to_ua(struct kvm *kvm, unsigned long gpa, |
168 | unsigned long *ua, unsigned long **prmap) | |
169 | { | |
170 | unsigned long gfn = gpa >> PAGE_SHIFT; | |
171 | struct kvm_memory_slot *memslot; | |
172 | ||
173 | memslot = search_memslots(kvm_memslots(kvm), gfn); | |
174 | if (!memslot) | |
175 | return -EINVAL; | |
176 | ||
177 | *ua = __gfn_to_hva_memslot(memslot, gfn) | | |
178 | (gpa & ~(PAGE_MASK | TCE_PCI_READ | TCE_PCI_WRITE)); | |
179 | ||
180 | #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE | |
181 | if (prmap) | |
182 | *prmap = &memslot->arch.rmap[gfn - memslot->base_gfn]; | |
183 | #endif | |
184 | ||
185 | return 0; | |
186 | } | |
187 | EXPORT_SYMBOL_GPL(kvmppc_gpa_to_ua); | |
188 | ||
189 | #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE | |
425333bf AK |
190 | static long iommu_tce_xchg_rm(struct mm_struct *mm, struct iommu_table *tbl, |
191 | unsigned long entry, unsigned long *hpa, | |
192 | enum dma_data_direction *direction) | |
193 | { | |
194 | long ret; | |
195 | ||
196 | ret = tbl->it_ops->exchange_rm(tbl, entry, hpa, direction); | |
197 | ||
198 | if (!ret && ((*direction == DMA_FROM_DEVICE) || | |
199 | (*direction == DMA_BIDIRECTIONAL))) { | |
200 | __be64 *pua = IOMMU_TABLE_USERSPACE_ENTRY_RM(tbl, entry); | |
201 | /* | |
202 | * kvmppc_rm_tce_iommu_do_map() updates the UA cache after | |
203 | * calling this so we still get here a valid UA. | |
204 | */ | |
205 | if (pua && *pua) | |
206 | mm_iommu_ua_mark_dirty_rm(mm, be64_to_cpu(*pua)); | |
207 | } | |
208 | ||
209 | return ret; | |
210 | } | |
211 | ||
212 | static void kvmppc_rm_clear_tce(struct kvm *kvm, struct iommu_table *tbl, | |
213 | unsigned long entry) | |
121f80ba AK |
214 | { |
215 | unsigned long hpa = 0; | |
216 | enum dma_data_direction dir = DMA_NONE; | |
217 | ||
425333bf | 218 | iommu_tce_xchg_rm(kvm->mm, tbl, entry, &hpa, &dir); |
121f80ba AK |
219 | } |
220 | ||
221 | static long kvmppc_rm_tce_iommu_mapped_dec(struct kvm *kvm, | |
222 | struct iommu_table *tbl, unsigned long entry) | |
223 | { | |
224 | struct mm_iommu_table_group_mem_t *mem = NULL; | |
225 | const unsigned long pgsize = 1ULL << tbl->it_page_shift; | |
a68bd126 | 226 | __be64 *pua = IOMMU_TABLE_USERSPACE_ENTRY_RM(tbl, entry); |
121f80ba AK |
227 | |
228 | if (!pua) | |
229 | /* it_userspace allocation might be delayed */ | |
230 | return H_TOO_HARD; | |
231 | ||
00a5c58d | 232 | mem = mm_iommu_lookup_rm(kvm->mm, be64_to_cpu(*pua), pgsize); |
121f80ba AK |
233 | if (!mem) |
234 | return H_TOO_HARD; | |
235 | ||
236 | mm_iommu_mapped_dec(mem); | |
237 | ||
00a5c58d | 238 | *pua = cpu_to_be64(0); |
121f80ba AK |
239 | |
240 | return H_SUCCESS; | |
241 | } | |
242 | ||
ca1fc489 | 243 | static long kvmppc_rm_tce_iommu_do_unmap(struct kvm *kvm, |
121f80ba AK |
244 | struct iommu_table *tbl, unsigned long entry) |
245 | { | |
246 | enum dma_data_direction dir = DMA_NONE; | |
247 | unsigned long hpa = 0; | |
248 | long ret; | |
249 | ||
425333bf | 250 | if (iommu_tce_xchg_rm(kvm->mm, tbl, entry, &hpa, &dir)) |
121f80ba AK |
251 | /* |
252 | * real mode xchg can fail if struct page crosses | |
253 | * a page boundary | |
254 | */ | |
255 | return H_TOO_HARD; | |
256 | ||
257 | if (dir == DMA_NONE) | |
258 | return H_SUCCESS; | |
259 | ||
260 | ret = kvmppc_rm_tce_iommu_mapped_dec(kvm, tbl, entry); | |
261 | if (ret) | |
425333bf | 262 | iommu_tce_xchg_rm(kvm->mm, tbl, entry, &hpa, &dir); |
121f80ba AK |
263 | |
264 | return ret; | |
265 | } | |
266 | ||
ca1fc489 AK |
267 | static long kvmppc_rm_tce_iommu_unmap(struct kvm *kvm, |
268 | struct kvmppc_spapr_tce_table *stt, struct iommu_table *tbl, | |
269 | unsigned long entry) | |
270 | { | |
271 | unsigned long i, ret = H_SUCCESS; | |
272 | unsigned long subpages = 1ULL << (stt->page_shift - tbl->it_page_shift); | |
273 | unsigned long io_entry = entry * subpages; | |
274 | ||
275 | for (i = 0; i < subpages; ++i) { | |
276 | ret = kvmppc_rm_tce_iommu_do_unmap(kvm, tbl, io_entry + i); | |
277 | if (ret != H_SUCCESS) | |
278 | break; | |
279 | } | |
280 | ||
281 | return ret; | |
282 | } | |
283 | ||
284 | static long kvmppc_rm_tce_iommu_do_map(struct kvm *kvm, struct iommu_table *tbl, | |
121f80ba AK |
285 | unsigned long entry, unsigned long ua, |
286 | enum dma_data_direction dir) | |
287 | { | |
288 | long ret; | |
289 | unsigned long hpa = 0; | |
a68bd126 | 290 | __be64 *pua = IOMMU_TABLE_USERSPACE_ENTRY_RM(tbl, entry); |
121f80ba AK |
291 | struct mm_iommu_table_group_mem_t *mem; |
292 | ||
293 | if (!pua) | |
294 | /* it_userspace allocation might be delayed */ | |
295 | return H_TOO_HARD; | |
296 | ||
297 | mem = mm_iommu_lookup_rm(kvm->mm, ua, 1ULL << tbl->it_page_shift); | |
298 | if (!mem) | |
299 | return H_TOO_HARD; | |
300 | ||
76fa4975 AK |
301 | if (WARN_ON_ONCE_RM(mm_iommu_ua_to_hpa_rm(mem, ua, tbl->it_page_shift, |
302 | &hpa))) | |
121f80ba AK |
303 | return H_HARDWARE; |
304 | ||
121f80ba AK |
305 | if (WARN_ON_ONCE_RM(mm_iommu_mapped_inc(mem))) |
306 | return H_CLOSED; | |
307 | ||
425333bf | 308 | ret = iommu_tce_xchg_rm(kvm->mm, tbl, entry, &hpa, &dir); |
121f80ba AK |
309 | if (ret) { |
310 | mm_iommu_mapped_dec(mem); | |
311 | /* | |
312 | * real mode xchg can fail if struct page crosses | |
313 | * a page boundary | |
314 | */ | |
315 | return H_TOO_HARD; | |
316 | } | |
317 | ||
318 | if (dir != DMA_NONE) | |
319 | kvmppc_rm_tce_iommu_mapped_dec(kvm, tbl, entry); | |
320 | ||
00a5c58d | 321 | *pua = cpu_to_be64(ua); |
121f80ba AK |
322 | |
323 | return 0; | |
324 | } | |
325 | ||
ca1fc489 AK |
326 | static long kvmppc_rm_tce_iommu_map(struct kvm *kvm, |
327 | struct kvmppc_spapr_tce_table *stt, struct iommu_table *tbl, | |
328 | unsigned long entry, unsigned long ua, | |
329 | enum dma_data_direction dir) | |
330 | { | |
331 | unsigned long i, pgoff, ret = H_SUCCESS; | |
332 | unsigned long subpages = 1ULL << (stt->page_shift - tbl->it_page_shift); | |
333 | unsigned long io_entry = entry * subpages; | |
334 | ||
335 | for (i = 0, pgoff = 0; i < subpages; | |
336 | ++i, pgoff += IOMMU_PAGE_SIZE(tbl)) { | |
337 | ||
338 | ret = kvmppc_rm_tce_iommu_do_map(kvm, tbl, | |
339 | io_entry + i, ua + pgoff, dir); | |
340 | if (ret != H_SUCCESS) | |
341 | break; | |
342 | } | |
343 | ||
344 | return ret; | |
345 | } | |
346 | ||
31217db7 AK |
347 | long kvmppc_rm_h_put_tce(struct kvm_vcpu *vcpu, unsigned long liobn, |
348 | unsigned long ioba, unsigned long tce) | |
54738c09 | 349 | { |
503bfcbe | 350 | struct kvmppc_spapr_tce_table *stt; |
fcbb2ce6 | 351 | long ret; |
121f80ba AK |
352 | struct kvmppc_spapr_tce_iommu_table *stit; |
353 | unsigned long entry, ua = 0; | |
354 | enum dma_data_direction dir; | |
54738c09 DG |
355 | |
356 | /* udbg_printf("H_PUT_TCE(): liobn=0x%lx ioba=0x%lx, tce=0x%lx\n", */ | |
357 | /* liobn, ioba, tce); */ | |
358 | ||
acde2572 PM |
359 | /* For radix, we might be in virtual mode, so punt */ |
360 | if (kvm_is_radix(vcpu->kvm)) | |
361 | return H_TOO_HARD; | |
362 | ||
503bfcbe | 363 | stt = kvmppc_find_table(vcpu->kvm, liobn); |
fcbb2ce6 AK |
364 | if (!stt) |
365 | return H_TOO_HARD; | |
366 | ||
367 | ret = kvmppc_ioba_validate(stt, ioba, 1); | |
368 | if (ret != H_SUCCESS) | |
369 | return ret; | |
370 | ||
5ee7af18 AK |
371 | ret = kvmppc_tce_validate(stt, tce); |
372 | if (ret != H_SUCCESS) | |
373 | return ret; | |
fcbb2ce6 | 374 | |
121f80ba AK |
375 | dir = iommu_tce_direction(tce); |
376 | if ((dir != DMA_NONE) && kvmppc_gpa_to_ua(vcpu->kvm, | |
377 | tce & ~(TCE_PCI_READ | TCE_PCI_WRITE), &ua, NULL)) | |
378 | return H_PARAMETER; | |
379 | ||
380 | entry = ioba >> stt->page_shift; | |
381 | ||
382 | list_for_each_entry_lockless(stit, &stt->iommu_tables, next) { | |
383 | if (dir == DMA_NONE) | |
ca1fc489 | 384 | ret = kvmppc_rm_tce_iommu_unmap(vcpu->kvm, stt, |
121f80ba AK |
385 | stit->tbl, entry); |
386 | else | |
ca1fc489 | 387 | ret = kvmppc_rm_tce_iommu_map(vcpu->kvm, stt, |
121f80ba AK |
388 | stit->tbl, entry, ua, dir); |
389 | ||
390 | if (ret == H_SUCCESS) | |
391 | continue; | |
392 | ||
393 | if (ret == H_TOO_HARD) | |
394 | return ret; | |
395 | ||
396 | WARN_ON_ONCE_RM(1); | |
425333bf | 397 | kvmppc_rm_clear_tce(vcpu->kvm, stit->tbl, entry); |
121f80ba AK |
398 | } |
399 | ||
400 | kvmppc_tce_put(stt, entry, tce); | |
fcbb2ce6 AK |
401 | |
402 | return H_SUCCESS; | |
54738c09 | 403 | } |
69e9fbb2 | 404 | |
d3695aa4 AK |
405 | static long kvmppc_rm_ua_to_hpa(struct kvm_vcpu *vcpu, |
406 | unsigned long ua, unsigned long *phpa) | |
407 | { | |
408 | pte_t *ptep, pte; | |
409 | unsigned shift = 0; | |
410 | ||
94171b19 AK |
411 | /* |
412 | * Called in real mode with MSR_EE = 0. We are safe here. | |
413 | * It is ok to do the lookup with arch.pgdir here, because | |
414 | * we are doing this on secondary cpus and current task there | |
415 | * is not the hypervisor. Also this is safe against THP in the | |
416 | * host, because an IPI to primary thread will wait for the secondary | |
417 | * to exit which will agains result in the below page table walk | |
418 | * to finish. | |
419 | */ | |
420 | ptep = __find_linux_pte(vcpu->arch.pgdir, ua, NULL, &shift); | |
d3695aa4 AK |
421 | if (!ptep || !pte_present(*ptep)) |
422 | return -ENXIO; | |
423 | pte = *ptep; | |
424 | ||
425 | if (!shift) | |
426 | shift = PAGE_SHIFT; | |
427 | ||
428 | /* Avoid handling anything potentially complicated in realmode */ | |
429 | if (shift > PAGE_SHIFT) | |
430 | return -EAGAIN; | |
431 | ||
432 | if (!pte_young(pte)) | |
433 | return -EAGAIN; | |
434 | ||
435 | *phpa = (pte_pfn(pte) << PAGE_SHIFT) | (ua & ((1ULL << shift) - 1)) | | |
436 | (ua & ~PAGE_MASK); | |
437 | ||
438 | return 0; | |
439 | } | |
440 | ||
441 | long kvmppc_rm_h_put_tce_indirect(struct kvm_vcpu *vcpu, | |
442 | unsigned long liobn, unsigned long ioba, | |
443 | unsigned long tce_list, unsigned long npages) | |
444 | { | |
445 | struct kvmppc_spapr_tce_table *stt; | |
446 | long i, ret = H_SUCCESS; | |
447 | unsigned long tces, entry, ua = 0; | |
448 | unsigned long *rmap = NULL; | |
da6f59e1 | 449 | bool prereg = false; |
121f80ba | 450 | struct kvmppc_spapr_tce_iommu_table *stit; |
d3695aa4 | 451 | |
acde2572 PM |
452 | /* For radix, we might be in virtual mode, so punt */ |
453 | if (kvm_is_radix(vcpu->kvm)) | |
454 | return H_TOO_HARD; | |
455 | ||
503bfcbe | 456 | stt = kvmppc_find_table(vcpu->kvm, liobn); |
d3695aa4 AK |
457 | if (!stt) |
458 | return H_TOO_HARD; | |
459 | ||
fe26e527 | 460 | entry = ioba >> stt->page_shift; |
d3695aa4 AK |
461 | /* |
462 | * The spec says that the maximum size of the list is 512 TCEs | |
463 | * so the whole table addressed resides in 4K page | |
464 | */ | |
465 | if (npages > 512) | |
466 | return H_PARAMETER; | |
467 | ||
468 | if (tce_list & (SZ_4K - 1)) | |
469 | return H_PARAMETER; | |
470 | ||
471 | ret = kvmppc_ioba_validate(stt, ioba, npages); | |
472 | if (ret != H_SUCCESS) | |
473 | return ret; | |
474 | ||
da6f59e1 AK |
475 | if (mm_iommu_preregistered(vcpu->kvm->mm)) { |
476 | /* | |
477 | * We get here if guest memory was pre-registered which | |
478 | * is normally VFIO case and gpa->hpa translation does not | |
479 | * depend on hpt. | |
480 | */ | |
481 | struct mm_iommu_table_group_mem_t *mem; | |
d3695aa4 | 482 | |
da6f59e1 AK |
483 | if (kvmppc_gpa_to_ua(vcpu->kvm, tce_list, &ua, NULL)) |
484 | return H_TOO_HARD; | |
d3695aa4 | 485 | |
da6f59e1 AK |
486 | mem = mm_iommu_lookup_rm(vcpu->kvm->mm, ua, IOMMU_PAGE_SIZE_4K); |
487 | if (mem) | |
76fa4975 AK |
488 | prereg = mm_iommu_ua_to_hpa_rm(mem, ua, |
489 | IOMMU_PAGE_SHIFT_4K, &tces) == 0; | |
da6f59e1 AK |
490 | } |
491 | ||
492 | if (!prereg) { | |
493 | /* | |
494 | * This is usually a case of a guest with emulated devices only | |
495 | * when TCE list is not in preregistered memory. | |
496 | * We do not require memory to be preregistered in this case | |
497 | * so lock rmap and do __find_linux_pte_or_hugepte(). | |
498 | */ | |
499 | if (kvmppc_gpa_to_ua(vcpu->kvm, tce_list, &ua, &rmap)) | |
500 | return H_TOO_HARD; | |
501 | ||
502 | rmap = (void *) vmalloc_to_phys(rmap); | |
121f80ba AK |
503 | if (WARN_ON_ONCE_RM(!rmap)) |
504 | return H_HARDWARE; | |
da6f59e1 AK |
505 | |
506 | /* | |
507 | * Synchronize with the MMU notifier callbacks in | |
39c983ea | 508 | * book3s_64_mmu_hv.c (kvm_unmap_hva_range_hv etc.). |
da6f59e1 AK |
509 | * While we have the rmap lock, code running on other CPUs |
510 | * cannot finish unmapping the host real page that backs | |
511 | * this guest real page, so we are OK to access the host | |
512 | * real page. | |
513 | */ | |
514 | lock_rmap(rmap); | |
515 | if (kvmppc_rm_ua_to_hpa(vcpu, ua, &tces)) { | |
516 | ret = H_TOO_HARD; | |
517 | goto unlock_exit; | |
518 | } | |
d3695aa4 AK |
519 | } |
520 | ||
521 | for (i = 0; i < npages; ++i) { | |
522 | unsigned long tce = be64_to_cpu(((u64 *)tces)[i]); | |
523 | ||
524 | ret = kvmppc_tce_validate(stt, tce); | |
525 | if (ret != H_SUCCESS) | |
526 | goto unlock_exit; | |
527 | ||
121f80ba AK |
528 | ua = 0; |
529 | if (kvmppc_gpa_to_ua(vcpu->kvm, | |
530 | tce & ~(TCE_PCI_READ | TCE_PCI_WRITE), | |
531 | &ua, NULL)) | |
532 | return H_PARAMETER; | |
533 | ||
534 | list_for_each_entry_lockless(stit, &stt->iommu_tables, next) { | |
ca1fc489 | 535 | ret = kvmppc_rm_tce_iommu_map(vcpu->kvm, stt, |
121f80ba AK |
536 | stit->tbl, entry + i, ua, |
537 | iommu_tce_direction(tce)); | |
538 | ||
539 | if (ret == H_SUCCESS) | |
540 | continue; | |
541 | ||
542 | if (ret == H_TOO_HARD) | |
543 | goto unlock_exit; | |
544 | ||
545 | WARN_ON_ONCE_RM(1); | |
425333bf | 546 | kvmppc_rm_clear_tce(vcpu->kvm, stit->tbl, entry); |
121f80ba AK |
547 | } |
548 | ||
d3695aa4 AK |
549 | kvmppc_tce_put(stt, entry + i, tce); |
550 | } | |
551 | ||
552 | unlock_exit: | |
da6f59e1 AK |
553 | if (rmap) |
554 | unlock_rmap(rmap); | |
d3695aa4 AK |
555 | |
556 | return ret; | |
557 | } | |
558 | ||
31217db7 | 559 | long kvmppc_rm_h_stuff_tce(struct kvm_vcpu *vcpu, |
d3695aa4 AK |
560 | unsigned long liobn, unsigned long ioba, |
561 | unsigned long tce_value, unsigned long npages) | |
562 | { | |
563 | struct kvmppc_spapr_tce_table *stt; | |
564 | long i, ret; | |
121f80ba | 565 | struct kvmppc_spapr_tce_iommu_table *stit; |
d3695aa4 | 566 | |
acde2572 PM |
567 | /* For radix, we might be in virtual mode, so punt */ |
568 | if (kvm_is_radix(vcpu->kvm)) | |
569 | return H_TOO_HARD; | |
570 | ||
503bfcbe | 571 | stt = kvmppc_find_table(vcpu->kvm, liobn); |
d3695aa4 AK |
572 | if (!stt) |
573 | return H_TOO_HARD; | |
574 | ||
575 | ret = kvmppc_ioba_validate(stt, ioba, npages); | |
576 | if (ret != H_SUCCESS) | |
577 | return ret; | |
578 | ||
579 | /* Check permission bits only to allow userspace poison TCE for debug */ | |
580 | if (tce_value & (TCE_PCI_WRITE | TCE_PCI_READ)) | |
581 | return H_PARAMETER; | |
582 | ||
121f80ba | 583 | list_for_each_entry_lockless(stit, &stt->iommu_tables, next) { |
c6b61661 | 584 | unsigned long entry = ioba >> stt->page_shift; |
121f80ba AK |
585 | |
586 | for (i = 0; i < npages; ++i) { | |
ca1fc489 | 587 | ret = kvmppc_rm_tce_iommu_unmap(vcpu->kvm, stt, |
121f80ba AK |
588 | stit->tbl, entry + i); |
589 | ||
590 | if (ret == H_SUCCESS) | |
591 | continue; | |
592 | ||
593 | if (ret == H_TOO_HARD) | |
594 | return ret; | |
595 | ||
596 | WARN_ON_ONCE_RM(1); | |
425333bf | 597 | kvmppc_rm_clear_tce(vcpu->kvm, stit->tbl, entry); |
121f80ba AK |
598 | } |
599 | } | |
600 | ||
fe26e527 AK |
601 | for (i = 0; i < npages; ++i, ioba += (1ULL << stt->page_shift)) |
602 | kvmppc_tce_put(stt, ioba >> stt->page_shift, tce_value); | |
d3695aa4 AK |
603 | |
604 | return H_SUCCESS; | |
605 | } | |
d3695aa4 | 606 | |
acde2572 | 607 | /* This can be called in either virtual mode or real mode */ |
69e9fbb2 LD |
608 | long kvmppc_h_get_tce(struct kvm_vcpu *vcpu, unsigned long liobn, |
609 | unsigned long ioba) | |
610 | { | |
503bfcbe | 611 | struct kvmppc_spapr_tce_table *stt; |
fcbb2ce6 AK |
612 | long ret; |
613 | unsigned long idx; | |
614 | struct page *page; | |
615 | u64 *tbl; | |
69e9fbb2 | 616 | |
503bfcbe | 617 | stt = kvmppc_find_table(vcpu->kvm, liobn); |
fcbb2ce6 AK |
618 | if (!stt) |
619 | return H_TOO_HARD; | |
69e9fbb2 | 620 | |
fcbb2ce6 AK |
621 | ret = kvmppc_ioba_validate(stt, ioba, 1); |
622 | if (ret != H_SUCCESS) | |
623 | return ret; | |
69e9fbb2 | 624 | |
14f853f1 | 625 | idx = (ioba >> stt->page_shift) - stt->offset; |
fcbb2ce6 AK |
626 | page = stt->pages[idx / TCES_PER_PAGE]; |
627 | tbl = (u64 *)page_address(page); | |
69e9fbb2 | 628 | |
1143a706 | 629 | vcpu->arch.regs.gpr[4] = tbl[idx % TCES_PER_PAGE]; |
69e9fbb2 | 630 | |
fcbb2ce6 | 631 | return H_SUCCESS; |
69e9fbb2 LD |
632 | } |
633 | EXPORT_SYMBOL_GPL(kvmppc_h_get_tce); | |
d3695aa4 AK |
634 | |
635 | #endif /* KVM_BOOK3S_HV_POSSIBLE */ |