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.
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.
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.
15 * Copyright 2010 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
16 * Copyright 2011 David Gibson, IBM Corporation <dwg@au1.ibm.com>
17 * Copyright 2016 Alexey Kardashevskiy, IBM Corporation <aik@au1.ibm.com>
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 #include <linux/stringify.h>
31 #include <asm/kvm_ppc.h>
32 #include <asm/kvm_book3s.h>
33 #include <asm/book3s/64/mmu-hash.h>
34 #include <asm/mmu_context.h>
35 #include <asm/hvcall.h>
36 #include <asm/synch.h>
37 #include <asm/ppc-opcode.h>
38 #include <asm/kvm_host.h>
40 #include <asm/iommu.h>
42 #include <asm/pte-walk.h>
46 #define WARN_ON_ONCE_RM(condition) ({ \
47 static bool __section(.data.unlikely) __warned; \
48 int __ret_warn_once = !!(condition); \
50 if (unlikely(__ret_warn_once && !__warned)) { \
52 pr_err("WARN_ON_ONCE_RM: (%s) at %s:%u\n", \
53 __stringify(condition), \
54 __func__, __LINE__); \
57 unlikely(__ret_warn_once); \
62 #define WARN_ON_ONCE_RM(condition) ({ \
63 int __ret_warn_on = !!(condition); \
64 unlikely(__ret_warn_on); \
69 #define TCES_PER_PAGE (PAGE_SIZE / sizeof(u64))
72 * Finds a TCE table descriptor by LIOBN.
74 * WARNING: This will be called in real or virtual mode on HV KVM and virtual
77 struct kvmppc_spapr_tce_table *kvmppc_find_table(struct kvm *kvm,
80 struct kvmppc_spapr_tce_table *stt;
82 list_for_each_entry_lockless(stt, &kvm->arch.spapr_tce_tables, list)
83 if (stt->liobn == liobn)
88 EXPORT_SYMBOL_GPL(kvmppc_find_table);
90 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
92 * Validates TCE address.
93 * At the moment flags and page mask are validated.
94 * As the host kernel does not access those addresses (just puts them
95 * to the table and user space is supposed to process them), we can skip
96 * checking other things (such as TCE is a guest RAM address or the page
97 * was actually allocated).
99 static long kvmppc_rm_tce_validate(struct kvmppc_spapr_tce_table *stt,
102 unsigned long gpa = tce & ~(TCE_PCI_READ | TCE_PCI_WRITE);
103 enum dma_data_direction dir = iommu_tce_direction(tce);
104 struct kvmppc_spapr_tce_iommu_table *stit;
105 unsigned long ua = 0;
107 /* Allow userspace to poison TCE table */
111 if (iommu_tce_check_gpa(stt->page_shift, gpa))
114 if (kvmppc_gpa_to_ua(stt->kvm, tce & ~(TCE_PCI_READ | TCE_PCI_WRITE),
118 list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
119 unsigned long hpa = 0;
120 struct mm_iommu_table_group_mem_t *mem;
121 long shift = stit->tbl->it_page_shift;
123 mem = mm_iommu_lookup_rm(stt->kvm->mm, ua, 1ULL << shift);
127 if (mm_iommu_ua_to_hpa_rm(mem, ua, shift, &hpa))
133 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
135 /* Note on the use of page_address() in real mode,
137 * It is safe to use page_address() in real mode on ppc64 because
138 * page_address() is always defined as lowmem_page_address()
139 * which returns __va(PFN_PHYS(page_to_pfn(page))) which is arithmetic
140 * operation and does not access page struct.
142 * Theoretically page_address() could be defined different
143 * but either WANT_PAGE_VIRTUAL or HASHED_PAGE_VIRTUAL
144 * would have to be enabled.
145 * WANT_PAGE_VIRTUAL is never enabled on ppc32/ppc64,
146 * HASHED_PAGE_VIRTUAL could be enabled for ppc32 only and only
147 * if CONFIG_HIGHMEM is defined. As CONFIG_SPARSEMEM_VMEMMAP
148 * is not expected to be enabled on ppc32, page_address()
149 * is safe for ppc32 as well.
151 * WARNING: This will be called in real-mode on HV KVM and virtual
154 static u64 *kvmppc_page_address(struct page *page)
156 #if defined(HASHED_PAGE_VIRTUAL) || defined(WANT_PAGE_VIRTUAL)
157 #error TODO: fix to avoid page_address() here
159 return (u64 *) page_address(page);
163 * Handles TCE requests for emulated devices.
164 * Puts guest TCE values to the table and expects user space to convert them.
165 * Called in both real and virtual modes.
166 * Cannot fail so kvmppc_tce_validate must be called before it.
168 * WARNING: This will be called in real-mode on HV KVM and virtual
171 void kvmppc_tce_put(struct kvmppc_spapr_tce_table *stt,
172 unsigned long idx, unsigned long tce)
178 page = stt->pages[idx / TCES_PER_PAGE];
179 tbl = kvmppc_page_address(page);
181 tbl[idx % TCES_PER_PAGE] = tce;
183 EXPORT_SYMBOL_GPL(kvmppc_tce_put);
185 long kvmppc_gpa_to_ua(struct kvm *kvm, unsigned long gpa,
186 unsigned long *ua, unsigned long **prmap)
188 unsigned long gfn = gpa >> PAGE_SHIFT;
189 struct kvm_memory_slot *memslot;
191 memslot = search_memslots(kvm_memslots(kvm), gfn);
195 *ua = __gfn_to_hva_memslot(memslot, gfn) |
196 (gpa & ~(PAGE_MASK | TCE_PCI_READ | TCE_PCI_WRITE));
198 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
200 *prmap = &memslot->arch.rmap[gfn - memslot->base_gfn];
205 EXPORT_SYMBOL_GPL(kvmppc_gpa_to_ua);
207 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
208 static long iommu_tce_xchg_rm(struct mm_struct *mm, struct iommu_table *tbl,
209 unsigned long entry, unsigned long *hpa,
210 enum dma_data_direction *direction)
214 ret = tbl->it_ops->exchange_rm(tbl, entry, hpa, direction);
216 if (!ret && ((*direction == DMA_FROM_DEVICE) ||
217 (*direction == DMA_BIDIRECTIONAL))) {
218 __be64 *pua = IOMMU_TABLE_USERSPACE_ENTRY_RM(tbl, entry);
220 * kvmppc_rm_tce_iommu_do_map() updates the UA cache after
221 * calling this so we still get here a valid UA.
224 mm_iommu_ua_mark_dirty_rm(mm, be64_to_cpu(*pua));
230 static void kvmppc_rm_clear_tce(struct kvm *kvm, struct iommu_table *tbl,
233 unsigned long hpa = 0;
234 enum dma_data_direction dir = DMA_NONE;
236 iommu_tce_xchg_rm(kvm->mm, tbl, entry, &hpa, &dir);
239 static long kvmppc_rm_tce_iommu_mapped_dec(struct kvm *kvm,
240 struct iommu_table *tbl, unsigned long entry)
242 struct mm_iommu_table_group_mem_t *mem = NULL;
243 const unsigned long pgsize = 1ULL << tbl->it_page_shift;
244 __be64 *pua = IOMMU_TABLE_USERSPACE_ENTRY_RM(tbl, entry);
247 /* it_userspace allocation might be delayed */
250 mem = mm_iommu_lookup_rm(kvm->mm, be64_to_cpu(*pua), pgsize);
254 mm_iommu_mapped_dec(mem);
256 *pua = cpu_to_be64(0);
261 static long kvmppc_rm_tce_iommu_do_unmap(struct kvm *kvm,
262 struct iommu_table *tbl, unsigned long entry)
264 enum dma_data_direction dir = DMA_NONE;
265 unsigned long hpa = 0;
268 if (iommu_tce_xchg_rm(kvm->mm, tbl, entry, &hpa, &dir))
270 * real mode xchg can fail if struct page crosses
278 ret = kvmppc_rm_tce_iommu_mapped_dec(kvm, tbl, entry);
280 iommu_tce_xchg_rm(kvm->mm, tbl, entry, &hpa, &dir);
285 static long kvmppc_rm_tce_iommu_unmap(struct kvm *kvm,
286 struct kvmppc_spapr_tce_table *stt, struct iommu_table *tbl,
289 unsigned long i, ret = H_SUCCESS;
290 unsigned long subpages = 1ULL << (stt->page_shift - tbl->it_page_shift);
291 unsigned long io_entry = entry * subpages;
293 for (i = 0; i < subpages; ++i) {
294 ret = kvmppc_rm_tce_iommu_do_unmap(kvm, tbl, io_entry + i);
295 if (ret != H_SUCCESS)
302 static long kvmppc_rm_tce_iommu_do_map(struct kvm *kvm, struct iommu_table *tbl,
303 unsigned long entry, unsigned long ua,
304 enum dma_data_direction dir)
307 unsigned long hpa = 0;
308 __be64 *pua = IOMMU_TABLE_USERSPACE_ENTRY_RM(tbl, entry);
309 struct mm_iommu_table_group_mem_t *mem;
312 /* it_userspace allocation might be delayed */
315 mem = mm_iommu_lookup_rm(kvm->mm, ua, 1ULL << tbl->it_page_shift);
319 if (WARN_ON_ONCE_RM(mm_iommu_ua_to_hpa_rm(mem, ua, tbl->it_page_shift,
323 if (WARN_ON_ONCE_RM(mm_iommu_mapped_inc(mem)))
326 ret = iommu_tce_xchg_rm(kvm->mm, tbl, entry, &hpa, &dir);
328 mm_iommu_mapped_dec(mem);
330 * real mode xchg can fail if struct page crosses
337 kvmppc_rm_tce_iommu_mapped_dec(kvm, tbl, entry);
339 *pua = cpu_to_be64(ua);
344 static long kvmppc_rm_tce_iommu_map(struct kvm *kvm,
345 struct kvmppc_spapr_tce_table *stt, struct iommu_table *tbl,
346 unsigned long entry, unsigned long ua,
347 enum dma_data_direction dir)
349 unsigned long i, pgoff, ret = H_SUCCESS;
350 unsigned long subpages = 1ULL << (stt->page_shift - tbl->it_page_shift);
351 unsigned long io_entry = entry * subpages;
353 for (i = 0, pgoff = 0; i < subpages;
354 ++i, pgoff += IOMMU_PAGE_SIZE(tbl)) {
356 ret = kvmppc_rm_tce_iommu_do_map(kvm, tbl,
357 io_entry + i, ua + pgoff, dir);
358 if (ret != H_SUCCESS)
365 long kvmppc_rm_h_put_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
366 unsigned long ioba, unsigned long tce)
368 struct kvmppc_spapr_tce_table *stt;
370 struct kvmppc_spapr_tce_iommu_table *stit;
371 unsigned long entry, ua = 0;
372 enum dma_data_direction dir;
374 /* udbg_printf("H_PUT_TCE(): liobn=0x%lx ioba=0x%lx, tce=0x%lx\n", */
375 /* liobn, ioba, tce); */
377 /* For radix, we might be in virtual mode, so punt */
378 if (kvm_is_radix(vcpu->kvm))
381 stt = kvmppc_find_table(vcpu->kvm, liobn);
385 ret = kvmppc_ioba_validate(stt, ioba, 1);
386 if (ret != H_SUCCESS)
389 ret = kvmppc_rm_tce_validate(stt, tce);
390 if (ret != H_SUCCESS)
393 dir = iommu_tce_direction(tce);
394 if ((dir != DMA_NONE) && kvmppc_gpa_to_ua(vcpu->kvm,
395 tce & ~(TCE_PCI_READ | TCE_PCI_WRITE), &ua, NULL))
398 entry = ioba >> stt->page_shift;
400 list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
402 ret = kvmppc_rm_tce_iommu_unmap(vcpu->kvm, stt,
405 ret = kvmppc_rm_tce_iommu_map(vcpu->kvm, stt,
406 stit->tbl, entry, ua, dir);
408 if (ret == H_SUCCESS)
411 if (ret == H_TOO_HARD)
415 kvmppc_rm_clear_tce(vcpu->kvm, stit->tbl, entry);
418 kvmppc_tce_put(stt, entry, tce);
423 static long kvmppc_rm_ua_to_hpa(struct kvm_vcpu *vcpu,
424 unsigned long ua, unsigned long *phpa)
430 * Called in real mode with MSR_EE = 0. We are safe here.
431 * It is ok to do the lookup with arch.pgdir here, because
432 * we are doing this on secondary cpus and current task there
433 * is not the hypervisor. Also this is safe against THP in the
434 * host, because an IPI to primary thread will wait for the secondary
435 * to exit which will agains result in the below page table walk
438 ptep = __find_linux_pte(vcpu->arch.pgdir, ua, NULL, &shift);
439 if (!ptep || !pte_present(*ptep))
446 /* Avoid handling anything potentially complicated in realmode */
447 if (shift > PAGE_SHIFT)
453 *phpa = (pte_pfn(pte) << PAGE_SHIFT) | (ua & ((1ULL << shift) - 1)) |
459 long kvmppc_rm_h_put_tce_indirect(struct kvm_vcpu *vcpu,
460 unsigned long liobn, unsigned long ioba,
461 unsigned long tce_list, unsigned long npages)
463 struct kvmppc_spapr_tce_table *stt;
464 long i, ret = H_SUCCESS;
465 unsigned long tces, entry, ua = 0;
466 unsigned long *rmap = NULL;
468 struct kvmppc_spapr_tce_iommu_table *stit;
470 /* For radix, we might be in virtual mode, so punt */
471 if (kvm_is_radix(vcpu->kvm))
474 stt = kvmppc_find_table(vcpu->kvm, liobn);
478 entry = ioba >> stt->page_shift;
480 * The spec says that the maximum size of the list is 512 TCEs
481 * so the whole table addressed resides in 4K page
486 if (tce_list & (SZ_4K - 1))
489 ret = kvmppc_ioba_validate(stt, ioba, npages);
490 if (ret != H_SUCCESS)
493 if (mm_iommu_preregistered(vcpu->kvm->mm)) {
495 * We get here if guest memory was pre-registered which
496 * is normally VFIO case and gpa->hpa translation does not
499 struct mm_iommu_table_group_mem_t *mem;
501 if (kvmppc_gpa_to_ua(vcpu->kvm, tce_list, &ua, NULL))
504 mem = mm_iommu_lookup_rm(vcpu->kvm->mm, ua, IOMMU_PAGE_SIZE_4K);
506 prereg = mm_iommu_ua_to_hpa_rm(mem, ua,
507 IOMMU_PAGE_SHIFT_4K, &tces) == 0;
512 * This is usually a case of a guest with emulated devices only
513 * when TCE list is not in preregistered memory.
514 * We do not require memory to be preregistered in this case
515 * so lock rmap and do __find_linux_pte_or_hugepte().
517 if (kvmppc_gpa_to_ua(vcpu->kvm, tce_list, &ua, &rmap))
520 rmap = (void *) vmalloc_to_phys(rmap);
521 if (WARN_ON_ONCE_RM(!rmap))
525 * Synchronize with the MMU notifier callbacks in
526 * book3s_64_mmu_hv.c (kvm_unmap_hva_range_hv etc.).
527 * While we have the rmap lock, code running on other CPUs
528 * cannot finish unmapping the host real page that backs
529 * this guest real page, so we are OK to access the host
533 if (kvmppc_rm_ua_to_hpa(vcpu, ua, &tces)) {
539 for (i = 0; i < npages; ++i) {
540 unsigned long tce = be64_to_cpu(((u64 *)tces)[i]);
542 ret = kvmppc_rm_tce_validate(stt, tce);
543 if (ret != H_SUCCESS)
547 for (i = 0; i < npages; ++i) {
548 unsigned long tce = be64_to_cpu(((u64 *)tces)[i]);
551 if (kvmppc_gpa_to_ua(vcpu->kvm,
552 tce & ~(TCE_PCI_READ | TCE_PCI_WRITE),
556 list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
557 ret = kvmppc_rm_tce_iommu_map(vcpu->kvm, stt,
558 stit->tbl, entry + i, ua,
559 iommu_tce_direction(tce));
561 if (ret == H_SUCCESS)
564 if (ret == H_TOO_HARD)
568 kvmppc_rm_clear_tce(vcpu->kvm, stit->tbl, entry);
571 kvmppc_tce_put(stt, entry + i, tce);
581 long kvmppc_rm_h_stuff_tce(struct kvm_vcpu *vcpu,
582 unsigned long liobn, unsigned long ioba,
583 unsigned long tce_value, unsigned long npages)
585 struct kvmppc_spapr_tce_table *stt;
587 struct kvmppc_spapr_tce_iommu_table *stit;
589 /* For radix, we might be in virtual mode, so punt */
590 if (kvm_is_radix(vcpu->kvm))
593 stt = kvmppc_find_table(vcpu->kvm, liobn);
597 ret = kvmppc_ioba_validate(stt, ioba, npages);
598 if (ret != H_SUCCESS)
601 /* Check permission bits only to allow userspace poison TCE for debug */
602 if (tce_value & (TCE_PCI_WRITE | TCE_PCI_READ))
605 list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
606 unsigned long entry = ioba >> stt->page_shift;
608 for (i = 0; i < npages; ++i) {
609 ret = kvmppc_rm_tce_iommu_unmap(vcpu->kvm, stt,
610 stit->tbl, entry + i);
612 if (ret == H_SUCCESS)
615 if (ret == H_TOO_HARD)
619 kvmppc_rm_clear_tce(vcpu->kvm, stit->tbl, entry);
623 for (i = 0; i < npages; ++i, ioba += (1ULL << stt->page_shift))
624 kvmppc_tce_put(stt, ioba >> stt->page_shift, tce_value);
629 /* This can be called in either virtual mode or real mode */
630 long kvmppc_h_get_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
633 struct kvmppc_spapr_tce_table *stt;
639 stt = kvmppc_find_table(vcpu->kvm, liobn);
643 ret = kvmppc_ioba_validate(stt, ioba, 1);
644 if (ret != H_SUCCESS)
647 idx = (ioba >> stt->page_shift) - stt->offset;
648 page = stt->pages[idx / TCES_PER_PAGE];
649 tbl = (u64 *)page_address(page);
651 vcpu->arch.regs.gpr[4] = tbl[idx % TCES_PER_PAGE];
655 EXPORT_SYMBOL_GPL(kvmppc_h_get_tce);
657 #endif /* KVM_BOOK3S_HV_POSSIBLE */