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>
30 #include <asm/tlbflush.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);
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).
98 * WARNING: This will be called in real-mode on HV KVM and virtual
101 long kvmppc_tce_validate(struct kvmppc_spapr_tce_table *stt, unsigned long tce)
103 unsigned long gpa = tce & ~(TCE_PCI_READ | TCE_PCI_WRITE);
104 enum dma_data_direction dir = iommu_tce_direction(tce);
106 /* Allow userspace to poison TCE table */
110 if (iommu_tce_check_gpa(stt->page_shift, gpa))
115 EXPORT_SYMBOL_GPL(kvmppc_tce_validate);
117 /* Note on the use of page_address() in real mode,
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.
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.
133 * WARNING: This will be called in real-mode on HV KVM and virtual
136 static u64 *kvmppc_page_address(struct page *page)
138 #if defined(HASHED_PAGE_VIRTUAL) || defined(WANT_PAGE_VIRTUAL)
139 #error TODO: fix to avoid page_address() here
141 return (u64 *) page_address(page);
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.
150 * WARNING: This will be called in real-mode on HV KVM and virtual
153 void kvmppc_tce_put(struct kvmppc_spapr_tce_table *stt,
154 unsigned long idx, unsigned long tce)
160 page = stt->pages[idx / TCES_PER_PAGE];
161 tbl = kvmppc_page_address(page);
163 tbl[idx % TCES_PER_PAGE] = tce;
165 EXPORT_SYMBOL_GPL(kvmppc_tce_put);
167 long kvmppc_gpa_to_ua(struct kvm *kvm, unsigned long gpa,
168 unsigned long *ua, unsigned long **prmap)
170 unsigned long gfn = gpa >> PAGE_SHIFT;
171 struct kvm_memory_slot *memslot;
173 memslot = search_memslots(kvm_memslots(kvm), gfn);
177 *ua = __gfn_to_hva_memslot(memslot, gfn) |
178 (gpa & ~(PAGE_MASK | TCE_PCI_READ | TCE_PCI_WRITE));
180 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
182 *prmap = &memslot->arch.rmap[gfn - memslot->base_gfn];
187 EXPORT_SYMBOL_GPL(kvmppc_gpa_to_ua);
189 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
190 static void kvmppc_rm_clear_tce(struct iommu_table *tbl, unsigned long entry)
192 unsigned long hpa = 0;
193 enum dma_data_direction dir = DMA_NONE;
195 iommu_tce_xchg_rm(tbl, entry, &hpa, &dir);
198 static long kvmppc_rm_tce_iommu_mapped_dec(struct kvm *kvm,
199 struct iommu_table *tbl, unsigned long entry)
201 struct mm_iommu_table_group_mem_t *mem = NULL;
202 const unsigned long pgsize = 1ULL << tbl->it_page_shift;
203 unsigned long *pua = IOMMU_TABLE_USERSPACE_ENTRY(tbl, entry);
206 /* it_userspace allocation might be delayed */
209 pua = (void *) vmalloc_to_phys(pua);
210 if (WARN_ON_ONCE_RM(!pua))
213 mem = mm_iommu_lookup_rm(kvm->mm, *pua, pgsize);
217 mm_iommu_mapped_dec(mem);
224 static long kvmppc_rm_tce_iommu_do_unmap(struct kvm *kvm,
225 struct iommu_table *tbl, unsigned long entry)
227 enum dma_data_direction dir = DMA_NONE;
228 unsigned long hpa = 0;
231 if (iommu_tce_xchg_rm(tbl, entry, &hpa, &dir))
233 * real mode xchg can fail if struct page crosses
241 ret = kvmppc_rm_tce_iommu_mapped_dec(kvm, tbl, entry);
243 iommu_tce_xchg_rm(tbl, entry, &hpa, &dir);
248 static long kvmppc_rm_tce_iommu_unmap(struct kvm *kvm,
249 struct kvmppc_spapr_tce_table *stt, struct iommu_table *tbl,
252 unsigned long i, ret = H_SUCCESS;
253 unsigned long subpages = 1ULL << (stt->page_shift - tbl->it_page_shift);
254 unsigned long io_entry = entry * subpages;
256 for (i = 0; i < subpages; ++i) {
257 ret = kvmppc_rm_tce_iommu_do_unmap(kvm, tbl, io_entry + i);
258 if (ret != H_SUCCESS)
265 static long kvmppc_rm_tce_iommu_do_map(struct kvm *kvm, struct iommu_table *tbl,
266 unsigned long entry, unsigned long ua,
267 enum dma_data_direction dir)
270 unsigned long hpa = 0;
271 unsigned long *pua = IOMMU_TABLE_USERSPACE_ENTRY(tbl, entry);
272 struct mm_iommu_table_group_mem_t *mem;
275 /* it_userspace allocation might be delayed */
278 mem = mm_iommu_lookup_rm(kvm->mm, ua, 1ULL << tbl->it_page_shift);
282 if (WARN_ON_ONCE_RM(mm_iommu_ua_to_hpa_rm(mem, ua, &hpa)))
285 pua = (void *) vmalloc_to_phys(pua);
286 if (WARN_ON_ONCE_RM(!pua))
289 if (WARN_ON_ONCE_RM(mm_iommu_mapped_inc(mem)))
292 ret = iommu_tce_xchg_rm(tbl, entry, &hpa, &dir);
294 mm_iommu_mapped_dec(mem);
296 * real mode xchg can fail if struct page crosses
303 kvmppc_rm_tce_iommu_mapped_dec(kvm, tbl, entry);
310 static long kvmppc_rm_tce_iommu_map(struct kvm *kvm,
311 struct kvmppc_spapr_tce_table *stt, struct iommu_table *tbl,
312 unsigned long entry, unsigned long ua,
313 enum dma_data_direction dir)
315 unsigned long i, pgoff, ret = H_SUCCESS;
316 unsigned long subpages = 1ULL << (stt->page_shift - tbl->it_page_shift);
317 unsigned long io_entry = entry * subpages;
319 for (i = 0, pgoff = 0; i < subpages;
320 ++i, pgoff += IOMMU_PAGE_SIZE(tbl)) {
322 ret = kvmppc_rm_tce_iommu_do_map(kvm, tbl,
323 io_entry + i, ua + pgoff, dir);
324 if (ret != H_SUCCESS)
331 long kvmppc_rm_h_put_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
332 unsigned long ioba, unsigned long tce)
334 struct kvmppc_spapr_tce_table *stt;
336 struct kvmppc_spapr_tce_iommu_table *stit;
337 unsigned long entry, ua = 0;
338 enum dma_data_direction dir;
340 /* udbg_printf("H_PUT_TCE(): liobn=0x%lx ioba=0x%lx, tce=0x%lx\n", */
341 /* liobn, ioba, tce); */
343 /* For radix, we might be in virtual mode, so punt */
344 if (kvm_is_radix(vcpu->kvm))
347 stt = kvmppc_find_table(vcpu->kvm, liobn);
351 ret = kvmppc_ioba_validate(stt, ioba, 1);
352 if (ret != H_SUCCESS)
355 ret = kvmppc_tce_validate(stt, tce);
356 if (ret != H_SUCCESS)
359 dir = iommu_tce_direction(tce);
360 if ((dir != DMA_NONE) && kvmppc_gpa_to_ua(vcpu->kvm,
361 tce & ~(TCE_PCI_READ | TCE_PCI_WRITE), &ua, NULL))
364 entry = ioba >> stt->page_shift;
366 list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
368 ret = kvmppc_rm_tce_iommu_unmap(vcpu->kvm, stt,
371 ret = kvmppc_rm_tce_iommu_map(vcpu->kvm, stt,
372 stit->tbl, entry, ua, dir);
374 if (ret == H_SUCCESS)
377 if (ret == H_TOO_HARD)
381 kvmppc_rm_clear_tce(stit->tbl, entry);
384 kvmppc_tce_put(stt, entry, tce);
389 static long kvmppc_rm_ua_to_hpa(struct kvm_vcpu *vcpu,
390 unsigned long ua, unsigned long *phpa)
396 * Called in real mode with MSR_EE = 0. We are safe here.
397 * It is ok to do the lookup with arch.pgdir here, because
398 * we are doing this on secondary cpus and current task there
399 * is not the hypervisor. Also this is safe against THP in the
400 * host, because an IPI to primary thread will wait for the secondary
401 * to exit which will agains result in the below page table walk
404 ptep = __find_linux_pte(vcpu->arch.pgdir, ua, NULL, &shift);
405 if (!ptep || !pte_present(*ptep))
412 /* Avoid handling anything potentially complicated in realmode */
413 if (shift > PAGE_SHIFT)
419 *phpa = (pte_pfn(pte) << PAGE_SHIFT) | (ua & ((1ULL << shift) - 1)) |
425 long kvmppc_rm_h_put_tce_indirect(struct kvm_vcpu *vcpu,
426 unsigned long liobn, unsigned long ioba,
427 unsigned long tce_list, unsigned long npages)
429 struct kvmppc_spapr_tce_table *stt;
430 long i, ret = H_SUCCESS;
431 unsigned long tces, entry, ua = 0;
432 unsigned long *rmap = NULL;
434 struct kvmppc_spapr_tce_iommu_table *stit;
436 /* For radix, we might be in virtual mode, so punt */
437 if (kvm_is_radix(vcpu->kvm))
440 stt = kvmppc_find_table(vcpu->kvm, liobn);
444 entry = ioba >> stt->page_shift;
446 * The spec says that the maximum size of the list is 512 TCEs
447 * so the whole table addressed resides in 4K page
452 if (tce_list & (SZ_4K - 1))
455 ret = kvmppc_ioba_validate(stt, ioba, npages);
456 if (ret != H_SUCCESS)
459 if (mm_iommu_preregistered(vcpu->kvm->mm)) {
461 * We get here if guest memory was pre-registered which
462 * is normally VFIO case and gpa->hpa translation does not
465 struct mm_iommu_table_group_mem_t *mem;
467 if (kvmppc_gpa_to_ua(vcpu->kvm, tce_list, &ua, NULL))
470 mem = mm_iommu_lookup_rm(vcpu->kvm->mm, ua, IOMMU_PAGE_SIZE_4K);
472 prereg = mm_iommu_ua_to_hpa_rm(mem, ua, &tces) == 0;
477 * This is usually a case of a guest with emulated devices only
478 * when TCE list is not in preregistered memory.
479 * We do not require memory to be preregistered in this case
480 * so lock rmap and do __find_linux_pte_or_hugepte().
482 if (kvmppc_gpa_to_ua(vcpu->kvm, tce_list, &ua, &rmap))
485 rmap = (void *) vmalloc_to_phys(rmap);
486 if (WARN_ON_ONCE_RM(!rmap))
490 * Synchronize with the MMU notifier callbacks in
491 * book3s_64_mmu_hv.c (kvm_unmap_hva_range_hv etc.).
492 * While we have the rmap lock, code running on other CPUs
493 * cannot finish unmapping the host real page that backs
494 * this guest real page, so we are OK to access the host
498 if (kvmppc_rm_ua_to_hpa(vcpu, ua, &tces)) {
504 for (i = 0; i < npages; ++i) {
505 unsigned long tce = be64_to_cpu(((u64 *)tces)[i]);
507 ret = kvmppc_tce_validate(stt, tce);
508 if (ret != H_SUCCESS)
512 if (kvmppc_gpa_to_ua(vcpu->kvm,
513 tce & ~(TCE_PCI_READ | TCE_PCI_WRITE),
517 list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
518 ret = kvmppc_rm_tce_iommu_map(vcpu->kvm, stt,
519 stit->tbl, entry + i, ua,
520 iommu_tce_direction(tce));
522 if (ret == H_SUCCESS)
525 if (ret == H_TOO_HARD)
529 kvmppc_rm_clear_tce(stit->tbl, entry);
532 kvmppc_tce_put(stt, entry + i, tce);
542 long kvmppc_rm_h_stuff_tce(struct kvm_vcpu *vcpu,
543 unsigned long liobn, unsigned long ioba,
544 unsigned long tce_value, unsigned long npages)
546 struct kvmppc_spapr_tce_table *stt;
548 struct kvmppc_spapr_tce_iommu_table *stit;
550 /* For radix, we might be in virtual mode, so punt */
551 if (kvm_is_radix(vcpu->kvm))
554 stt = kvmppc_find_table(vcpu->kvm, liobn);
558 ret = kvmppc_ioba_validate(stt, ioba, npages);
559 if (ret != H_SUCCESS)
562 /* Check permission bits only to allow userspace poison TCE for debug */
563 if (tce_value & (TCE_PCI_WRITE | TCE_PCI_READ))
566 list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
567 unsigned long entry = ioba >> stt->page_shift;
569 for (i = 0; i < npages; ++i) {
570 ret = kvmppc_rm_tce_iommu_unmap(vcpu->kvm, stt,
571 stit->tbl, entry + i);
573 if (ret == H_SUCCESS)
576 if (ret == H_TOO_HARD)
580 kvmppc_rm_clear_tce(stit->tbl, entry);
584 for (i = 0; i < npages; ++i, ioba += (1ULL << stt->page_shift))
585 kvmppc_tce_put(stt, ioba >> stt->page_shift, tce_value);
590 /* This can be called in either virtual mode or real mode */
591 long kvmppc_h_get_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
594 struct kvmppc_spapr_tce_table *stt;
600 stt = kvmppc_find_table(vcpu->kvm, liobn);
604 ret = kvmppc_ioba_validate(stt, ioba, 1);
605 if (ret != H_SUCCESS)
608 idx = (ioba >> stt->page_shift) - stt->offset;
609 page = stt->pages[idx / TCES_PER_PAGE];
610 tbl = (u64 *)page_address(page);
612 vcpu->arch.regs.gpr[4] = tbl[idx % TCES_PER_PAGE];
616 EXPORT_SYMBOL_GPL(kvmppc_h_get_tce);
618 #endif /* KVM_BOOK3S_HV_POSSIBLE */