~hc/RK356X_SDK_RELEASE.git

..	..	@@ -1,7 +1,5 @@
	1	+// SPDX-License-Identifier: GPL-2.0-only
1	2	/*
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	3	*
6	4	* Copyright 2016 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
7	5	*/
..	..	@@ -10,14 +8,19 @@
10	8	#include <linux/string.h>
11	9	#include <linux/kvm.h>
12	10	#include <linux/kvm_host.h>
	11	+#include <linux/anon_inodes.h>
	12	+#include <linux/file.h>
	13	+#include <linux/debugfs.h>
	14	+#include <linux/pgtable.h>
13	15
14	16	#include <asm/kvm_ppc.h>
15	17	#include <asm/kvm_book3s.h>
16	18	#include <asm/page.h>
17	19	#include <asm/mmu.h>
18		-#include <asm/pgtable.h>
19	20	#include <asm/pgalloc.h>
20	21	#include <asm/pte-walk.h>
	22	+#include <asm/ultravisor.h>
	23	+#include <asm/kvm_book3s_uvmem.h>
21	24
22	25	/*
23	26	* Supported radix tree geometry.
..	..	@@ -26,18 +29,236 @@
26	29	*/
27	30	static int p9_supported_radix_bits[4] = { 5, 9, 9, 13 };
28	31
	32	+unsigned long __kvmhv_copy_tofrom_guest_radix(int lpid, int pid,
	33	+ gva_t eaddr, void to, void from,
	34	+ unsigned long n)
	35	+{
	36	+ int old_pid, old_lpid;
	37	+ unsigned long quadrant, ret = n;
	38	+ bool is_load = !!to;
	39	+
	40	+ /* Can't access quadrants 1 or 2 in non-HV mode, call the HV to do it */
	41	+ if (kvmhv_on_pseries())
	42	+ return plpar_hcall_norets(H_COPY_TOFROM_GUEST, lpid, pid, eaddr,
	43	+ (to != NULL) ? __pa(to): 0,
	44	+ (from != NULL) ? __pa(from): 0, n);
	45	+
	46	+ quadrant = 1;
	47	+ if (!pid)
	48	+ quadrant = 2;
	49	+ if (is_load)
	50	+ from = (void *) (eaddr \| (quadrant << 62));
	51	+ else
	52	+ to = (void *) (eaddr \| (quadrant << 62));
	53	+
	54	+ preempt_disable();
	55	+
	56	+ /* switch the lpid first to avoid running host with unallocated pid */
	57	+ old_lpid = mfspr(SPRN_LPID);
	58	+ if (old_lpid != lpid)
	59	+ mtspr(SPRN_LPID, lpid);
	60	+ if (quadrant == 1) {
	61	+ old_pid = mfspr(SPRN_PID);
	62	+ if (old_pid != pid)
	63	+ mtspr(SPRN_PID, pid);
	64	+ }
	65	+ isync();
	66	+
	67	+ pagefault_disable();
	68	+ if (is_load)
	69	+ ret = __copy_from_user_inatomic(to, (const void __user *)from, n);
	70	+ else
	71	+ ret = __copy_to_user_inatomic((void __user *)to, from, n);
	72	+ pagefault_enable();
	73	+
	74	+ /* switch the pid first to avoid running host with unallocated pid */
	75	+ if (quadrant == 1 && pid != old_pid)
	76	+ mtspr(SPRN_PID, old_pid);
	77	+ if (lpid != old_lpid)
	78	+ mtspr(SPRN_LPID, old_lpid);
	79	+ isync();
	80	+
	81	+ preempt_enable();
	82	+
	83	+ return ret;
	84	+}
	85	+EXPORT_SYMBOL_GPL(__kvmhv_copy_tofrom_guest_radix);
	86	+
	87	+static long kvmhv_copy_tofrom_guest_radix(struct kvm_vcpu *vcpu, gva_t eaddr,
	88	+ void to, void from, unsigned long n)
	89	+{
	90	+ int lpid = vcpu->kvm->arch.lpid;
	91	+ int pid = vcpu->arch.pid;
	92	+
	93	+ /* This would cause a data segment intr so don't allow the access */
	94	+ if (eaddr & (0x3FFUL << 52))
	95	+ return -EINVAL;
	96	+
	97	+ /* Should we be using the nested lpid */
	98	+ if (vcpu->arch.nested)
	99	+ lpid = vcpu->arch.nested->shadow_lpid;
	100	+
	101	+ /* If accessing quadrant 3 then pid is expected to be 0 */
	102	+ if (((eaddr >> 62) & 0x3) == 0x3)
	103	+ pid = 0;
	104	+
	105	+ eaddr &= ~(0xFFFUL << 52);
	106	+
	107	+ return __kvmhv_copy_tofrom_guest_radix(lpid, pid, eaddr, to, from, n);
	108	+}
	109	+
	110	+long kvmhv_copy_from_guest_radix(struct kvm_vcpu vcpu, gva_t eaddr, void to,
	111	+ unsigned long n)
	112	+{
	113	+ long ret;
	114	+
	115	+ ret = kvmhv_copy_tofrom_guest_radix(vcpu, eaddr, to, NULL, n);
	116	+ if (ret > 0)
	117	+ memset(to + (n - ret), 0, ret);
	118	+
	119	+ return ret;
	120	+}
	121	+EXPORT_SYMBOL_GPL(kvmhv_copy_from_guest_radix);
	122	+
	123	+long kvmhv_copy_to_guest_radix(struct kvm_vcpu vcpu, gva_t eaddr, void from,
	124	+ unsigned long n)
	125	+{
	126	+ return kvmhv_copy_tofrom_guest_radix(vcpu, eaddr, NULL, from, n);
	127	+}
	128	+EXPORT_SYMBOL_GPL(kvmhv_copy_to_guest_radix);
	129	+
	130	+int kvmppc_mmu_walk_radix_tree(struct kvm_vcpu *vcpu, gva_t eaddr,
	131	+ struct kvmppc_pte *gpte, u64 root,
	132	+ u64 *pte_ret_p)
	133	+{
	134	+ struct kvm *kvm = vcpu->kvm;
	135	+ int ret, level, ps;
	136	+ unsigned long rts, bits, offset, index;
	137	+ u64 pte, base, gpa;
	138	+ __be64 rpte;
	139	+
	140	+ rts = ((root & RTS1_MASK) >> (RTS1_SHIFT - 3)) \|
	141	+ ((root & RTS2_MASK) >> RTS2_SHIFT);
	142	+ bits = root & RPDS_MASK;
	143	+ base = root & RPDB_MASK;
	144	+
	145	+ offset = rts + 31;
	146	+
	147	+ /* Current implementations only support 52-bit space */
	148	+ if (offset != 52)
	149	+ return -EINVAL;
	150	+
	151	+ /* Walk each level of the radix tree */
	152	+ for (level = 3; level >= 0; --level) {
	153	+ u64 addr;
	154	+ /* Check a valid size */
	155	+ if (level && bits != p9_supported_radix_bits[level])
	156	+ return -EINVAL;
	157	+ if (level == 0 && !(bits == 5 \|\| bits == 9))
	158	+ return -EINVAL;
	159	+ offset -= bits;
	160	+ index = (eaddr >> offset) & ((1UL << bits) - 1);
	161	+ /* Check that low bits of page table base are zero */
	162	+ if (base & ((1UL << (bits + 3)) - 1))
	163	+ return -EINVAL;
	164	+ /* Read the entry from guest memory */
	165	+ addr = base + (index * sizeof(rpte));
	166	+ vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
	167	+ ret = kvm_read_guest(kvm, addr, &rpte, sizeof(rpte));
	168	+ srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
	169	+ if (ret) {
	170	+ if (pte_ret_p)
	171	+ *pte_ret_p = addr;
	172	+ return ret;
	173	+ }
	174	+ pte = __be64_to_cpu(rpte);
	175	+ if (!(pte & _PAGE_PRESENT))
	176	+ return -ENOENT;
	177	+ /* Check if a leaf entry */
	178	+ if (pte & _PAGE_PTE)
	179	+ break;
	180	+ /* Get ready to walk the next level */
	181	+ base = pte & RPDB_MASK;
	182	+ bits = pte & RPDS_MASK;
	183	+ }
	184	+
	185	+ /* Need a leaf at lowest level; 512GB pages not supported */
	186	+ if (level < 0 \|\| level == 3)
	187	+ return -EINVAL;
	188	+
	189	+ /* We found a valid leaf PTE */
	190	+ /* Offset is now log base 2 of the page size */
	191	+ gpa = pte & 0x01fffffffffff000ul;
	192	+ if (gpa & ((1ul << offset) - 1))
	193	+ return -EINVAL;
	194	+ gpa \|= eaddr & ((1ul << offset) - 1);
	195	+ for (ps = MMU_PAGE_4K; ps < MMU_PAGE_COUNT; ++ps)
	196	+ if (offset == mmu_psize_defs[ps].shift)
	197	+ break;
	198	+ gpte->page_size = ps;
	199	+ gpte->page_shift = offset;
	200	+
	201	+ gpte->eaddr = eaddr;
	202	+ gpte->raddr = gpa;
	203	+
	204	+ /* Work out permissions */
	205	+ gpte->may_read = !!(pte & _PAGE_READ);
	206	+ gpte->may_write = !!(pte & _PAGE_WRITE);
	207	+ gpte->may_execute = !!(pte & _PAGE_EXEC);
	208	+
	209	+ gpte->rc = pte & (_PAGE_ACCESSED \| _PAGE_DIRTY);
	210	+
	211	+ if (pte_ret_p)
	212	+ *pte_ret_p = pte;
	213	+
	214	+ return 0;
	215	+}
	216	+
	217	+/*
	218	+ * Used to walk a partition or process table radix tree in guest memory
	219	+ * Note: We exploit the fact that a partition table and a process
	220	+ * table have the same layout, a partition-scoped page table and a
	221	+ * process-scoped page table have the same layout, and the 2nd
	222	+ * doubleword of a partition table entry has the same layout as
	223	+ * the PTCR register.
	224	+ */
	225	+int kvmppc_mmu_radix_translate_table(struct kvm_vcpu *vcpu, gva_t eaddr,
	226	+ struct kvmppc_pte *gpte, u64 table,
	227	+ int table_index, u64 *pte_ret_p)
	228	+{
	229	+ struct kvm *kvm = vcpu->kvm;
	230	+ int ret;
	231	+ unsigned long size, ptbl, root;
	232	+ struct prtb_entry entry;
	233	+
	234	+ if ((table & PRTS_MASK) > 24)
	235	+ return -EINVAL;
	236	+ size = 1ul << ((table & PRTS_MASK) + 12);
	237	+
	238	+ /* Is the table big enough to contain this entry? */
	239	+ if ((table_index * sizeof(entry)) >= size)
	240	+ return -EINVAL;
	241	+
	242	+ /* Read the table to find the root of the radix tree */
	243	+ ptbl = (table & PRTB_MASK) + (table_index * sizeof(entry));
	244	+ vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
	245	+ ret = kvm_read_guest(kvm, ptbl, &entry, sizeof(entry));
	246	+ srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
	247	+ if (ret)
	248	+ return ret;
	249	+
	250	+ /* Root is stored in the first double word */
	251	+ root = be64_to_cpu(entry.prtb0);
	252	+
	253	+ return kvmppc_mmu_walk_radix_tree(vcpu, eaddr, gpte, root, pte_ret_p);
	254	+}
	255	+
29	256	int kvmppc_mmu_radix_xlate(struct kvm_vcpu *vcpu, gva_t eaddr,
30	257	struct kvmppc_pte *gpte, bool data, bool iswrite)
31	258	{
32		- struct kvm *kvm = vcpu->kvm;
33	259	u32 pid;
34		- int ret, level, ps;
35		- __be64 prte, rpte;
36		- unsigned long ptbl;
37		- unsigned long root, pte, index;
38		- unsigned long rts, bits, offset;
39		- unsigned long gpa;
40		- unsigned long proc_tbl_size;
	260	+ u64 pte;
	261	+ int ret;
41	262
42	263	/* Work out effective PID */
43	264	switch (eaddr >> 62) {
..	..	@@ -50,71 +271,13 @@
50	271	default:
51	272	return -EINVAL;
52	273	}
53		- proc_tbl_size = 1 << ((kvm->arch.process_table & PRTS_MASK) + 12);
54		- if (pid * 16 >= proc_tbl_size)
55		- return -EINVAL;
56	274
57		- /* Read partition table to find root of tree for effective PID */
58		- ptbl = (kvm->arch.process_table & PRTB_MASK) + (pid * 16);
59		- ret = kvm_read_guest(kvm, ptbl, &prte, sizeof(prte));
	275	+ ret = kvmppc_mmu_radix_translate_table(vcpu, eaddr, gpte,
	276	+ vcpu->kvm->arch.process_table, pid, &pte);
60	277	if (ret)
61	278	return ret;
62	279
63		- root = be64_to_cpu(prte);
64		- rts = ((root & RTS1_MASK) >> (RTS1_SHIFT - 3)) \|
65		- ((root & RTS2_MASK) >> RTS2_SHIFT);
66		- bits = root & RPDS_MASK;
67		- root = root & RPDB_MASK;
68		-
69		- offset = rts + 31;
70		-
71		- /* current implementations only support 52-bit space */
72		- if (offset != 52)
73		- return -EINVAL;
74		-
75		- for (level = 3; level >= 0; --level) {
76		- if (level && bits != p9_supported_radix_bits[level])
77		- return -EINVAL;
78		- if (level == 0 && !(bits == 5 \|\| bits == 9))
79		- return -EINVAL;
80		- offset -= bits;
81		- index = (eaddr >> offset) & ((1UL << bits) - 1);
82		- /* check that low bits of page table base are zero */
83		- if (root & ((1UL << (bits + 3)) - 1))
84		- return -EINVAL;
85		- ret = kvm_read_guest(kvm, root + index * 8,
86		- &rpte, sizeof(rpte));
87		- if (ret)
88		- return ret;
89		- pte = __be64_to_cpu(rpte);
90		- if (!(pte & _PAGE_PRESENT))
91		- return -ENOENT;
92		- if (pte & _PAGE_PTE)
93		- break;
94		- bits = pte & 0x1f;
95		- root = pte & 0x0fffffffffffff00ul;
96		- }
97		- /* need a leaf at lowest level; 512GB pages not supported */
98		- if (level < 0 \|\| level == 3)
99		- return -EINVAL;
100		-
101		- /* offset is now log base 2 of the page size */
102		- gpa = pte & 0x01fffffffffff000ul;
103		- if (gpa & ((1ul << offset) - 1))
104		- return -EINVAL;
105		- gpa += eaddr & ((1ul << offset) - 1);
106		- for (ps = MMU_PAGE_4K; ps < MMU_PAGE_COUNT; ++ps)
107		- if (offset == mmu_psize_defs[ps].shift)
108		- break;
109		- gpte->page_size = ps;
110		-
111		- gpte->eaddr = eaddr;
112		- gpte->raddr = gpa;
113		-
114		- /* Work out permissions */
115		- gpte->may_read = !!(pte & _PAGE_READ);
116		- gpte->may_write = !!(pte & _PAGE_WRITE);
117		- gpte->may_execute = !!(pte & _PAGE_EXEC);
	280	+ /* Check privilege (applies only to process scoped translations) */
118	281	if (kvmppc_get_msr(vcpu) & MSR_PR) {
119	282	if (pte & _PAGE_PRIVILEGED) {
120	283	gpte->may_read = 0;
..	..	@@ -136,21 +299,47 @@
136	299	return 0;
137	300	}
138	301
139		-static void kvmppc_radix_tlbie_page(struct kvm *kvm, unsigned long addr,
140		- unsigned int pshift)
	302	+void kvmppc_radix_tlbie_page(struct kvm *kvm, unsigned long addr,
	303	+ unsigned int pshift, unsigned int lpid)
141	304	{
142	305	unsigned long psize = PAGE_SIZE;
	306	+ int psi;
	307	+ long rc;
	308	+ unsigned long rb;
143	309
144	310	if (pshift)
145	311	psize = 1UL << pshift;
	312	+ else
	313	+ pshift = PAGE_SHIFT;
146	314
147	315	addr &= ~(psize - 1);
148		- radix__flush_tlb_lpid_page(kvm->arch.lpid, addr, psize);
	316	+
	317	+ if (!kvmhv_on_pseries()) {
	318	+ radix__flush_tlb_lpid_page(lpid, addr, psize);
	319	+ return;
	320	+ }
	321	+
	322	+ psi = shift_to_mmu_psize(pshift);
	323	+ rb = addr \| (mmu_get_ap(psi) << PPC_BITLSHIFT(58));
	324	+ rc = plpar_hcall_norets(H_TLB_INVALIDATE, H_TLBIE_P1_ENC(0, 0, 1),
	325	+ lpid, rb);
	326	+ if (rc)
	327	+ pr_err("KVM: TLB page invalidation hcall failed, rc=%ld\n", rc);
149	328	}
150	329
151		-static void kvmppc_radix_flush_pwc(struct kvm *kvm)
	330	+static void kvmppc_radix_flush_pwc(struct kvm *kvm, unsigned int lpid)
152	331	{
153		- radix__flush_pwc_lpid(kvm->arch.lpid);
	332	+ long rc;
	333	+
	334	+ if (!kvmhv_on_pseries()) {
	335	+ radix__flush_pwc_lpid(lpid);
	336	+ return;
	337	+ }
	338	+
	339	+ rc = plpar_hcall_norets(H_TLB_INVALIDATE, H_TLBIE_P1_ENC(1, 0, 1),
	340	+ lpid, TLBIEL_INVAL_SET_LPID);
	341	+ if (rc)
	342	+ pr_err("KVM: TLB PWC invalidation hcall failed, rc=%ld\n", rc);
154	343	}
155	344
156	345	static unsigned long kvmppc_radix_update_pte(struct kvm kvm, pte_t ptep,
..	..	@@ -160,7 +349,7 @@
160	349	return __radix_pte_update(ptep, clr, set);
161	350	}
162	351
163		-void kvmppc_radix_set_pte_at(struct kvm *kvm, unsigned long addr,
	352	+static void kvmppc_radix_set_pte_at(struct kvm *kvm, unsigned long addr,
164	353	pte_t *ptep, pte_t pte)
165	354	{
166	355	radix__set_pte_at(kvm->mm, addr, ptep, pte, 0);
..	..	@@ -185,12 +374,6 @@
185	374	kmem_cache_free(kvm_pte_cache, ptep);
186	375	}
187	376
188		-/* Like pmd_huge() and pmd_large(), but works regardless of config options */
189		-static inline int pmd_is_leaf(pmd_t pmd)
190		-{
191		- return !!(pmd_val(pmd) & _PAGE_PTE);
192		-}
193		-
194	377	static pmd_t *kvmppc_pmd_alloc(void)
195	378	{
196	379	pmd_t *pmd;
..	..	@@ -207,23 +390,44 @@
207	390	kmem_cache_free(kvm_pmd_cache, pmdp);
208	391	}
209	392
210		-static void kvmppc_unmap_pte(struct kvm kvm, pte_t pte,
211		- unsigned long gpa, unsigned int shift)
	393	+/* Called with kvm->mmu_lock held */
	394	+void kvmppc_unmap_pte(struct kvm kvm, pte_t pte, unsigned long gpa,
	395	+ unsigned int shift,
	396	+ const struct kvm_memory_slot *memslot,
	397	+ unsigned int lpid)
212	398
213	399	{
214		- unsigned long page_size = 1ul << shift;
215	400	unsigned long old;
	401	+ unsigned long gfn = gpa >> PAGE_SHIFT;
	402	+ unsigned long page_size = PAGE_SIZE;
	403	+ unsigned long hpa;
216	404
217	405	old = kvmppc_radix_update_pte(kvm, pte, ~0UL, 0, gpa, shift);
218		- kvmppc_radix_tlbie_page(kvm, gpa, shift);
219		- if (old & _PAGE_DIRTY) {
220		- unsigned long gfn = gpa >> PAGE_SHIFT;
221		- struct kvm_memory_slot *memslot;
	406	+ kvmppc_radix_tlbie_page(kvm, gpa, shift, lpid);
222	407
	408	+ /* The following only applies to L1 entries */
	409	+ if (lpid != kvm->arch.lpid)
	410	+ return;
	411	+
	412	+ if (!memslot) {
223	413	memslot = gfn_to_memslot(kvm, gfn);
224		- if (memslot && memslot->dirty_bitmap)
225		- kvmppc_update_dirty_map(memslot, gfn, page_size);
	414	+ if (!memslot)
	415	+ return;
226	416	}
	417	+ if (shift) { /* 1GB or 2MB page */
	418	+ page_size = 1ul << shift;
	419	+ if (shift == PMD_SHIFT)
	420	+ kvm->stat.num_2M_pages--;
	421	+ else if (shift == PUD_SHIFT)
	422	+ kvm->stat.num_1G_pages--;
	423	+ }
	424	+
	425	+ gpa &= ~(page_size - 1);
	426	+ hpa = old & PTE_RPN_MASK;
	427	+ kvmhv_remove_nest_rmap_range(kvm, memslot, gpa, hpa, page_size);
	428	+
	429	+ if ((old & _PAGE_DIRTY) && memslot->dirty_bitmap)
	430	+ kvmppc_update_dirty_map(memslot, gfn, page_size);
227	431	}
228	432
229	433	/*
..	..	@@ -232,14 +436,19 @@
232	436	* Callers are responsible for flushing the PWC.
233	437	*
234	438	* When page tables are being unmapped/freed as part of page fault path
235		- * (full == false), ptes are not expected. There is code to unmap them
236		- * and emit a warning if encountered, but there may already be data
237		- * corruption due to the unexpected mappings.
	439	+ * (full == false), valid ptes are generally not expected; however, there
	440	+ * is one situation where they arise, which is when dirty page logging is
	441	+ * turned off for a memslot while the VM is running. The new memslot
	442	+ * becomes visible to page faults before the memslot commit function
	443	+ * gets to flush the memslot, which can lead to a 2MB page mapping being
	444	+ * installed for a guest physical address where there are already 64kB
	445	+ * (or 4kB) mappings (of sub-pages of the same 2MB page).
238	446	*/
239		-static void kvmppc_unmap_free_pte(struct kvm kvm, pte_t pte, bool full)
	447	+static void kvmppc_unmap_free_pte(struct kvm kvm, pte_t pte, bool full,
	448	+ unsigned int lpid)
240	449	{
241	450	if (full) {
242		- memset(pte, 0, sizeof(long) << PTE_INDEX_SIZE);
	451	+ memset(pte, 0, sizeof(long) << RADIX_PTE_INDEX_SIZE);
243	452	} else {
244	453	pte_t *p = pte;
245	454	unsigned long it;
..	..	@@ -247,17 +456,17 @@
247	456	for (it = 0; it < PTRS_PER_PTE; ++it, ++p) {
248	457	if (pte_val(*p) == 0)
249	458	continue;
250		- WARN_ON_ONCE(1);
251	459	kvmppc_unmap_pte(kvm, p,
252	460	pte_pfn(*p) << PAGE_SHIFT,
253		- PAGE_SHIFT);
	461	+ PAGE_SHIFT, NULL, lpid);
254	462	}
255	463	}
256	464
257	465	kvmppc_pte_free(pte);
258	466	}
259	467
260		-static void kvmppc_unmap_free_pmd(struct kvm kvm, pmd_t pmd, bool full)
	468	+static void kvmppc_unmap_free_pmd(struct kvm kvm, pmd_t pmd, bool full,
	469	+ unsigned int lpid)
261	470	{
262	471	unsigned long im;
263	472	pmd_t *p = pmd;
..	..	@@ -272,20 +481,21 @@
272	481	WARN_ON_ONCE(1);
273	482	kvmppc_unmap_pte(kvm, (pte_t *)p,
274	483	pte_pfn((pte_t )p) << PAGE_SHIFT,
275		- PMD_SHIFT);
	484	+ PMD_SHIFT, NULL, lpid);
276	485	}
277	486	} else {
278	487	pte_t *pte;
279	488
280	489	pte = pte_offset_map(p, 0);
281		- kvmppc_unmap_free_pte(kvm, pte, full);
	490	+ kvmppc_unmap_free_pte(kvm, pte, full, lpid);
282	491	pmd_clear(p);
283	492	}
284	493	}
285	494	kvmppc_pmd_free(pmd);
286	495	}
287	496
288		-static void kvmppc_unmap_free_pud(struct kvm kvm, pud_t pud)
	497	+static void kvmppc_unmap_free_pud(struct kvm kvm, pud_t pud,
	498	+ unsigned int lpid)
289	499	{
290	500	unsigned long iu;
291	501	pud_t *p = pud;
..	..	@@ -293,42 +503,47 @@
293	503	for (iu = 0; iu < PTRS_PER_PUD; ++iu, ++p) {
294	504	if (!pud_present(*p))
295	505	continue;
296		- if (pud_huge(*p)) {
	506	+ if (pud_is_leaf(*p)) {
297	507	pud_clear(p);
298	508	} else {
299	509	pmd_t *pmd;
300	510
301	511	pmd = pmd_offset(p, 0);
302		- kvmppc_unmap_free_pmd(kvm, pmd, true);
	512	+ kvmppc_unmap_free_pmd(kvm, pmd, true, lpid);
303	513	pud_clear(p);
304	514	}
305	515	}
306	516	pud_free(kvm->mm, pud);
307	517	}
308	518
309		-void kvmppc_free_radix(struct kvm *kvm)
	519	+void kvmppc_free_pgtable_radix(struct kvm kvm, pgd_t pgd, unsigned int lpid)
310	520	{
311	521	unsigned long ig;
312		- pgd_t *pgd;
313	522
314		- if (!kvm->arch.pgtable)
315		- return;
316		- pgd = kvm->arch.pgtable;
317	523	for (ig = 0; ig < PTRS_PER_PGD; ++ig, ++pgd) {
	524	+ p4d_t *p4d = p4d_offset(pgd, 0);
318	525	pud_t *pud;
319	526
320		- if (!pgd_present(*pgd))
	527	+ if (!p4d_present(*p4d))
321	528	continue;
322		- pud = pud_offset(pgd, 0);
323		- kvmppc_unmap_free_pud(kvm, pud);
324		- pgd_clear(pgd);
	529	+ pud = pud_offset(p4d, 0);
	530	+ kvmppc_unmap_free_pud(kvm, pud, lpid);
	531	+ p4d_clear(p4d);
325	532	}
326		- pgd_free(kvm->mm, kvm->arch.pgtable);
327		- kvm->arch.pgtable = NULL;
	533	+}
	534	+
	535	+void kvmppc_free_radix(struct kvm *kvm)
	536	+{
	537	+ if (kvm->arch.pgtable) {
	538	+ kvmppc_free_pgtable_radix(kvm, kvm->arch.pgtable,
	539	+ kvm->arch.lpid);
	540	+ pgd_free(kvm->mm, kvm->arch.pgtable);
	541	+ kvm->arch.pgtable = NULL;
	542	+ }
328	543	}
329	544
330	545	static void kvmppc_unmap_free_pmd_entry_table(struct kvm kvm, pmd_t pmd,
331		- unsigned long gpa)
	546	+ unsigned long gpa, unsigned int lpid)
332	547	{
333	548	pte_t *pte = pte_offset_kernel(pmd, 0);
334	549
..	..	@@ -338,13 +553,13 @@
338	553	* flushing the PWC again.
339	554	*/
340	555	pmd_clear(pmd);
341		- kvmppc_radix_flush_pwc(kvm);
	556	+ kvmppc_radix_flush_pwc(kvm, lpid);
342	557
343		- kvmppc_unmap_free_pte(kvm, pte, false);
	558	+ kvmppc_unmap_free_pte(kvm, pte, false, lpid);
344	559	}
345	560
346	561	static void kvmppc_unmap_free_pud_entry_table(struct kvm kvm, pud_t pud,
347		- unsigned long gpa)
	562	+ unsigned long gpa, unsigned int lpid)
348	563	{
349	564	pmd_t *pmd = pmd_offset(pud, 0);
350	565
..	..	@@ -354,9 +569,9 @@
354	569	* so can be freed without flushing the PWC again.
355	570	*/
356	571	pud_clear(pud);
357		- kvmppc_radix_flush_pwc(kvm);
	572	+ kvmppc_radix_flush_pwc(kvm, lpid);
358	573
359		- kvmppc_unmap_free_pmd(kvm, pmd, false);
	574	+ kvmppc_unmap_free_pmd(kvm, pmd, false, lpid);
360	575	}
361	576
362	577	/*
..	..	@@ -368,25 +583,30 @@
368	583	*/
369	584	#define PTE_BITS_MUST_MATCH (~(_PAGE_WRITE \| _PAGE_DIRTY \| _PAGE_ACCESSED))
370	585
371		-static int kvmppc_create_pte(struct kvm *kvm, pte_t pte, unsigned long gpa,
372		- unsigned int level, unsigned long mmu_seq)
	586	+int kvmppc_create_pte(struct kvm kvm, pgd_t pgtable, pte_t pte,
	587	+ unsigned long gpa, unsigned int level,
	588	+ unsigned long mmu_seq, unsigned int lpid,
	589	+ unsigned long rmapp, struct rmap_nested *n_rmap)
373	590	{
374	591	pgd_t *pgd;
	592	+ p4d_t *p4d;
375	593	pud_t pud, new_pud = NULL;
376	594	pmd_t pmd, new_pmd = NULL;
377	595	pte_t ptep, new_ptep = NULL;
378	596	int ret;
379	597
380	598	/* Traverse the guest's 2nd-level tree, allocate new levels needed */
381		- pgd = kvm->arch.pgtable + pgd_index(gpa);
	599	+ pgd = pgtable + pgd_index(gpa);
	600	+ p4d = p4d_offset(pgd, gpa);
	601	+
382	602	pud = NULL;
383		- if (pgd_present(*pgd))
384		- pud = pud_offset(pgd, gpa);
	603	+ if (p4d_present(*p4d))
	604	+ pud = pud_offset(p4d, gpa);
385	605	else
386	606	new_pud = pud_alloc_one(kvm->mm, gpa);
387	607
388	608	pmd = NULL;
389		- if (pud && pud_present(pud) && !pud_huge(pud))
	609	+ if (pud && pud_present(pud) && !pud_is_leaf(pud))
390	610	pmd = pmd_offset(pud, gpa);
391	611	else if (level <= 1)
392	612	new_pmd = kvmppc_pmd_alloc();
..	..	@@ -402,14 +622,14 @@
402	622
403	623	/* Now traverse again under the lock and change the tree */
404	624	ret = -ENOMEM;
405		- if (pgd_none(*pgd)) {
	625	+ if (p4d_none(*p4d)) {
406	626	if (!new_pud)
407	627	goto out_unlock;
408		- pgd_populate(kvm->mm, pgd, new_pud);
	628	+ p4d_populate(kvm->mm, p4d, new_pud);
409	629	new_pud = NULL;
410	630	}
411		- pud = pud_offset(pgd, gpa);
412		- if (pud_huge(*pud)) {
	631	+ pud = pud_offset(p4d, gpa);
	632	+ if (pud_is_leaf(*pud)) {
413	633	unsigned long hgpa = gpa & PUD_MASK;
414	634
415	635	/* Check if we raced and someone else has set the same thing */
..	..	@@ -435,7 +655,8 @@
435	655	goto out_unlock;
436	656	}
437	657	/* Valid 1GB page here already, remove it */
438		- kvmppc_unmap_pte(kvm, (pte_t *)pud, hgpa, PUD_SHIFT);
	658	+ kvmppc_unmap_pte(kvm, (pte_t *)pud, hgpa, PUD_SHIFT, NULL,
	659	+ lpid);
439	660	}
440	661	if (level == 2) {
441	662	if (!pud_none(*pud)) {
..	..	@@ -444,9 +665,11 @@
444	665	* install a large page, so remove and free the page
445	666	* table page.
446	667	*/
447		- kvmppc_unmap_free_pud_entry_table(kvm, pud, gpa);
	668	+ kvmppc_unmap_free_pud_entry_table(kvm, pud, gpa, lpid);
448	669	}
449	670	kvmppc_radix_set_pte_at(kvm, gpa, (pte_t *)pud, pte);
	671	+ if (rmapp && n_rmap)
	672	+ kvmhv_insert_nest_rmap(kvm, rmapp, n_rmap);
450	673	ret = 0;
451	674	goto out_unlock;
452	675	}
..	..	@@ -470,7 +693,7 @@
470	693	WARN_ON_ONCE((pmd_val(*pmd) ^ pte_val(pte)) &
471	694	PTE_BITS_MUST_MATCH);
472	695	kvmppc_radix_update_pte(kvm, pmdp_ptep(pmd),
473		- 0, pte_val(pte), lgpa, PMD_SHIFT);
	696	+ 0, pte_val(pte), lgpa, PMD_SHIFT);
474	697	ret = 0;
475	698	goto out_unlock;
476	699	}
..	..	@@ -484,7 +707,8 @@
484	707	goto out_unlock;
485	708	}
486	709	/* Valid 2MB page here already, remove it */
487		- kvmppc_unmap_pte(kvm, pmdp_ptep(pmd), lgpa, PMD_SHIFT);
	710	+ kvmppc_unmap_pte(kvm, pmdp_ptep(pmd), lgpa, PMD_SHIFT, NULL,
	711	+ lpid);
488	712	}
489	713	if (level == 1) {
490	714	if (!pmd_none(*pmd)) {
..	..	@@ -493,9 +717,11 @@
493	717	* install a large page, so remove and free the page
494	718	* table page.
495	719	*/
496		- kvmppc_unmap_free_pmd_entry_table(kvm, pmd, gpa);
	720	+ kvmppc_unmap_free_pmd_entry_table(kvm, pmd, gpa, lpid);
497	721	}
498	722	kvmppc_radix_set_pte_at(kvm, gpa, pmdp_ptep(pmd), pte);
	723	+ if (rmapp && n_rmap)
	724	+ kvmhv_insert_nest_rmap(kvm, rmapp, n_rmap);
499	725	ret = 0;
500	726	goto out_unlock;
501	727	}
..	..	@@ -520,6 +746,8 @@
520	746	goto out_unlock;
521	747	}
522	748	kvmppc_radix_set_pte_at(kvm, gpa, ptep, pte);
	749	+ if (rmapp && n_rmap)
	750	+ kvmhv_insert_nest_rmap(kvm, rmapp, n_rmap);
523	751	ret = 0;
524	752
525	753	out_unlock:
..	..	@@ -533,95 +761,50 @@
533	761	return ret;
534	762	}
535	763
536		-int kvmppc_book3s_radix_page_fault(struct kvm_run run, struct kvm_vcpu vcpu,
537		- unsigned long ea, unsigned long dsisr)
	764	+bool kvmppc_hv_handle_set_rc(struct kvm *kvm, bool nested, bool writing,
	765	+ unsigned long gpa, unsigned int lpid)
	766	+{
	767	+ unsigned long pgflags;
	768	+ unsigned int shift;
	769	+ pte_t *ptep;
	770	+
	771	+ /*
	772	+ * Need to set an R or C bit in the 2nd-level tables;
	773	+ * since we are just helping out the hardware here,
	774	+ * it is sufficient to do what the hardware does.
	775	+ */
	776	+ pgflags = _PAGE_ACCESSED;
	777	+ if (writing)
	778	+ pgflags \|= _PAGE_DIRTY;
	779	+
	780	+ if (nested)
	781	+ ptep = find_kvm_nested_guest_pte(kvm, lpid, gpa, &shift);
	782	+ else
	783	+ ptep = find_kvm_secondary_pte(kvm, gpa, &shift);
	784	+
	785	+ if (ptep && pte_present(ptep) && (!writing \|\| pte_write(ptep))) {
	786	+ kvmppc_radix_update_pte(kvm, ptep, 0, pgflags, gpa, shift);
	787	+ return true;
	788	+ }
	789	+ return false;
	790	+}
	791	+
	792	+int kvmppc_book3s_instantiate_page(struct kvm_vcpu *vcpu,
	793	+ unsigned long gpa,
	794	+ struct kvm_memory_slot *memslot,
	795	+ bool writing, bool kvm_ro,
	796	+ pte_t inserted_pte, unsigned int levelp)
538	797	{
539	798	struct kvm *kvm = vcpu->kvm;
540		- unsigned long mmu_seq;
541		- unsigned long gpa, gfn, hva;
542		- struct kvm_memory_slot *memslot;
543	799	struct page *page = NULL;
544		- long ret;
545		- bool writing;
	800	+ unsigned long mmu_seq;
	801	+ unsigned long hva, gfn = gpa >> PAGE_SHIFT;
546	802	bool upgrade_write = false;
547	803	bool *upgrade_p = &upgrade_write;
548	804	pte_t pte, *ptep;
549		- unsigned long pgflags;
550	805	unsigned int shift, level;
551		-
552		- /* Check for unusual errors */
553		- if (dsisr & DSISR_UNSUPP_MMU) {
554		- pr_err("KVM: Got unsupported MMU fault\n");
555		- return -EFAULT;
556		- }
557		- if (dsisr & DSISR_BADACCESS) {
558		- /* Reflect to the guest as DSI */
559		- pr_err("KVM: Got radix HV page fault with DSISR=%lx\n", dsisr);
560		- kvmppc_core_queue_data_storage(vcpu, ea, dsisr);
561		- return RESUME_GUEST;
562		- }
563		-
564		- /* Translate the logical address and get the page */
565		- gpa = vcpu->arch.fault_gpa & ~0xfffUL;
566		- gpa &= ~0xF000000000000000ul;
567		- gfn = gpa >> PAGE_SHIFT;
568		- if (!(dsisr & DSISR_PRTABLE_FAULT))
569		- gpa \|= ea & 0xfff;
570		- memslot = gfn_to_memslot(kvm, gfn);
571		-
572		- /* No memslot means it's an emulated MMIO region */
573		- if (!memslot \|\| (memslot->flags & KVM_MEMSLOT_INVALID)) {
574		- if (dsisr & (DSISR_PRTABLE_FAULT \| DSISR_BADACCESS \|
575		- DSISR_SET_RC)) {
576		- /*
577		- * Bad address in guest page table tree, or other
578		- * unusual error - reflect it to the guest as DSI.
579		- */
580		- kvmppc_core_queue_data_storage(vcpu, ea, dsisr);
581		- return RESUME_GUEST;
582		- }
583		- return kvmppc_hv_emulate_mmio(run, vcpu, gpa, ea,
584		- dsisr & DSISR_ISSTORE);
585		- }
586		-
587		- writing = (dsisr & DSISR_ISSTORE) != 0;
588		- if (memslot->flags & KVM_MEM_READONLY) {
589		- if (writing) {
590		- /* give the guest a DSI */
591		- dsisr = DSISR_ISSTORE \| DSISR_PROTFAULT;
592		- kvmppc_core_queue_data_storage(vcpu, ea, dsisr);
593		- return RESUME_GUEST;
594		- }
595		- upgrade_p = NULL;
596		- }
597		-
598		- if (dsisr & DSISR_SET_RC) {
599		- /*
600		- * Need to set an R or C bit in the 2nd-level tables;
601		- * since we are just helping out the hardware here,
602		- * it is sufficient to do what the hardware does.
603		- */
604		- pgflags = _PAGE_ACCESSED;
605		- if (writing)
606		- pgflags \|= _PAGE_DIRTY;
607		- /*
608		- * We are walking the secondary page table here. We can do this
609		- * without disabling irq.
610		- */
611		- spin_lock(&kvm->mmu_lock);
612		- ptep = __find_linux_pte(kvm->arch.pgtable,
613		- gpa, NULL, &shift);
614		- if (ptep && pte_present(*ptep) &&
615		- (!writing \|\| pte_write(*ptep))) {
616		- kvmppc_radix_update_pte(kvm, ptep, 0, pgflags,
617		- gpa, shift);
618		- dsisr &= ~DSISR_SET_RC;
619		- }
620		- spin_unlock(&kvm->mmu_lock);
621		- if (!(dsisr & (DSISR_BAD_FAULT_64S \| DSISR_NOHPTE \|
622		- DSISR_PROTFAULT \| DSISR_SET_RC)))
623		- return RESUME_GUEST;
624		- }
	806	+ int ret;
	807	+ bool large_enable;
625	808
626	809	/* used to check for invalidations in progress */
627	810	mmu_seq = kvm->mmu_notifier_seq;
..	..	@@ -634,7 +817,7 @@
634	817	* is that the page is writable.
635	818	*/
636	819	hva = gfn_to_hva_memslot(memslot, gfn);
637		- if (upgrade_p && __get_user_pages_fast(hva, 1, 1, &page) == 1) {
	820	+ if (!kvm_ro && get_user_page_fast_only(hva, FOLL_WRITE, &page)) {
638	821	upgrade_write = true;
639	822	} else {
640	823	unsigned long pfn;
..	..	@@ -656,27 +839,31 @@
656	839	* Read the PTE from the process' radix tree and use that
657	840	* so we get the shift and attribute bits.
658	841	*/
659		- local_irq_disable();
660		- ptep = __find_linux_pte(vcpu->arch.pgdir, hva, NULL, &shift);
	842	+ spin_lock(&kvm->mmu_lock);
	843	+ ptep = find_kvm_host_pte(kvm, mmu_seq, hva, &shift);
	844	+ pte = __pte(0);
	845	+ if (ptep)
	846	+ pte = READ_ONCE(*ptep);
	847	+ spin_unlock(&kvm->mmu_lock);
661	848	/*
662	849	* If the PTE disappeared temporarily due to a THP
663	850	* collapse, just return and let the guest try again.
664	851	*/
665		- if (!ptep) {
666		- local_irq_enable();
	852	+ if (!pte_present(pte)) {
667	853	if (page)
668	854	put_page(page);
669	855	return RESUME_GUEST;
670	856	}
671		- pte = *ptep;
672		- local_irq_enable();
	857	+
	858	+ /* If we're logging dirty pages, always map single pages */
	859	+ large_enable = !(memslot->flags & KVM_MEM_LOG_DIRTY_PAGES);
673	860
674	861	/* Get pte level from shift/size */
675		- if (shift == PUD_SHIFT &&
	862	+ if (large_enable && shift == PUD_SHIFT &&
676	863	(gpa & (PUD_SIZE - PAGE_SIZE)) ==
677	864	(hva & (PUD_SIZE - PAGE_SIZE))) {
678	865	level = 2;
679		- } else if (shift == PMD_SHIFT &&
	866	+ } else if (large_enable && shift == PMD_SHIFT &&
680	867	(gpa & (PMD_SIZE - PAGE_SIZE)) ==
681	868	(hva & (PMD_SIZE - PAGE_SIZE))) {
682	869	level = 1;
..	..	@@ -702,7 +889,12 @@
702	889	}
703	890
704	891	/* Allocate space in the tree and write the PTE */
705		- ret = kvmppc_create_pte(kvm, pte, gpa, level, mmu_seq);
	892	+ ret = kvmppc_create_pte(kvm, kvm->arch.pgtable, pte, gpa, level,
	893	+ mmu_seq, kvm->arch.lpid, NULL, NULL);
	894	+ if (inserted_pte)
	895	+ *inserted_pte = pte;
	896	+ if (levelp)
	897	+ *levelp = level;
706	898
707	899	if (page) {
708	900	if (!ret && (pte_val(pte) & _PAGE_WRITE))
..	..	@@ -710,36 +902,119 @@
710	902	put_page(page);
711	903	}
712	904
	905	+ /* Increment number of large pages if we (successfully) inserted one */
	906	+ if (!ret) {
	907	+ if (level == 1)
	908	+ kvm->stat.num_2M_pages++;
	909	+ else if (level == 2)
	910	+ kvm->stat.num_1G_pages++;
	911	+ }
	912	+
	913	+ return ret;
	914	+}
	915	+
	916	+int kvmppc_book3s_radix_page_fault(struct kvm_vcpu *vcpu,
	917	+ unsigned long ea, unsigned long dsisr)
	918	+{
	919	+ struct kvm *kvm = vcpu->kvm;
	920	+ unsigned long gpa, gfn;
	921	+ struct kvm_memory_slot *memslot;
	922	+ long ret;
	923	+ bool writing = !!(dsisr & DSISR_ISSTORE);
	924	+ bool kvm_ro = false;
	925	+
	926	+ /* Check for unusual errors */
	927	+ if (dsisr & DSISR_UNSUPP_MMU) {
	928	+ pr_err("KVM: Got unsupported MMU fault\n");
	929	+ return -EFAULT;
	930	+ }
	931	+ if (dsisr & DSISR_BADACCESS) {
	932	+ /* Reflect to the guest as DSI */
	933	+ pr_err("KVM: Got radix HV page fault with DSISR=%lx\n", dsisr);
	934	+ kvmppc_core_queue_data_storage(vcpu, ea, dsisr);
	935	+ return RESUME_GUEST;
	936	+ }
	937	+
	938	+ /* Translate the logical address */
	939	+ gpa = vcpu->arch.fault_gpa & ~0xfffUL;
	940	+ gpa &= ~0xF000000000000000ul;
	941	+ gfn = gpa >> PAGE_SHIFT;
	942	+ if (!(dsisr & DSISR_PRTABLE_FAULT))
	943	+ gpa \|= ea & 0xfff;
	944	+
	945	+ if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE)
	946	+ return kvmppc_send_page_to_uv(kvm, gfn);
	947	+
	948	+ /* Get the corresponding memslot */
	949	+ memslot = gfn_to_memslot(kvm, gfn);
	950	+
	951	+ /* No memslot means it's an emulated MMIO region */
	952	+ if (!memslot \|\| (memslot->flags & KVM_MEMSLOT_INVALID)) {
	953	+ if (dsisr & (DSISR_PRTABLE_FAULT \| DSISR_BADACCESS \|
	954	+ DSISR_SET_RC)) {
	955	+ /*
	956	+ * Bad address in guest page table tree, or other
	957	+ * unusual error - reflect it to the guest as DSI.
	958	+ */
	959	+ kvmppc_core_queue_data_storage(vcpu, ea, dsisr);
	960	+ return RESUME_GUEST;
	961	+ }
	962	+ return kvmppc_hv_emulate_mmio(vcpu, gpa, ea, writing);
	963	+ }
	964	+
	965	+ if (memslot->flags & KVM_MEM_READONLY) {
	966	+ if (writing) {
	967	+ /* give the guest a DSI */
	968	+ kvmppc_core_queue_data_storage(vcpu, ea, DSISR_ISSTORE \|
	969	+ DSISR_PROTFAULT);
	970	+ return RESUME_GUEST;
	971	+ }
	972	+ kvm_ro = true;
	973	+ }
	974	+
	975	+ /* Failed to set the reference/change bits */
	976	+ if (dsisr & DSISR_SET_RC) {
	977	+ spin_lock(&kvm->mmu_lock);
	978	+ if (kvmppc_hv_handle_set_rc(kvm, false, writing,
	979	+ gpa, kvm->arch.lpid))
	980	+ dsisr &= ~DSISR_SET_RC;
	981	+ spin_unlock(&kvm->mmu_lock);
	982	+
	983	+ if (!(dsisr & (DSISR_BAD_FAULT_64S \| DSISR_NOHPTE \|
	984	+ DSISR_PROTFAULT \| DSISR_SET_RC)))
	985	+ return RESUME_GUEST;
	986	+ }
	987	+
	988	+ /* Try to insert a pte */
	989	+ ret = kvmppc_book3s_instantiate_page(vcpu, gpa, memslot, writing,
	990	+ kvm_ro, NULL, NULL);
	991	+
713	992	if (ret == 0 \|\| ret == -EAGAIN)
714	993	ret = RESUME_GUEST;
715	994	return ret;
716	995	}
717	996
718		-/* Called with kvm->lock held */
	997	+/* Called with kvm->mmu_lock held */
719	998	int kvm_unmap_radix(struct kvm kvm, struct kvm_memory_slot memslot,
720	999	unsigned long gfn)
721	1000	{
722	1001	pte_t *ptep;
723	1002	unsigned long gpa = gfn << PAGE_SHIFT;
724	1003	unsigned int shift;
725		- unsigned long old;
726	1004
727		- ptep = __find_linux_pte(kvm->arch.pgtable, gpa, NULL, &shift);
728		- if (ptep && pte_present(*ptep)) {
729		- old = kvmppc_radix_update_pte(kvm, ptep, ~0UL, 0,
730		- gpa, shift);
731		- kvmppc_radix_tlbie_page(kvm, gpa, shift);
732		- if ((old & _PAGE_DIRTY) && memslot->dirty_bitmap) {
733		- unsigned long psize = PAGE_SIZE;
734		- if (shift)
735		- psize = 1ul << shift;
736		- kvmppc_update_dirty_map(memslot, gfn, psize);
737		- }
	1005	+ if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE) {
	1006	+ uv_page_inval(kvm->arch.lpid, gpa, PAGE_SHIFT);
	1007	+ return 0;
738	1008	}
739		- return 0;
	1009	+
	1010	+ ptep = find_kvm_secondary_pte(kvm, gpa, &shift);
	1011	+ if (ptep && pte_present(*ptep))
	1012	+ kvmppc_unmap_pte(kvm, ptep, gpa, shift, memslot,
	1013	+ kvm->arch.lpid);
	1014	+ return 0;
740	1015	}
741	1016
742		-/* Called with kvm->lock held */
	1017	+/* Called with kvm->mmu_lock held */
743	1018	int kvm_age_radix(struct kvm kvm, struct kvm_memory_slot memslot,
744	1019	unsigned long gfn)
745	1020	{
..	..	@@ -747,18 +1022,27 @@
747	1022	unsigned long gpa = gfn << PAGE_SHIFT;
748	1023	unsigned int shift;
749	1024	int ref = 0;
	1025	+ unsigned long old, *rmapp;
750	1026
751		- ptep = __find_linux_pte(kvm->arch.pgtable, gpa, NULL, &shift);
	1027	+ if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE)
	1028	+ return ref;
	1029	+
	1030	+ ptep = find_kvm_secondary_pte(kvm, gpa, &shift);
752	1031	if (ptep && pte_present(ptep) && pte_young(ptep)) {
753		- kvmppc_radix_update_pte(kvm, ptep, _PAGE_ACCESSED, 0,
754		- gpa, shift);
	1032	+ old = kvmppc_radix_update_pte(kvm, ptep, _PAGE_ACCESSED, 0,
	1033	+ gpa, shift);
755	1034	/* XXX need to flush tlb here? */
	1035	+ /* Also clear bit in ptes in shadow pgtable for nested guests */
	1036	+ rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
	1037	+ kvmhv_update_nest_rmap_rc_list(kvm, rmapp, _PAGE_ACCESSED, 0,
	1038	+ old & PTE_RPN_MASK,
	1039	+ 1UL << shift);
756	1040	ref = 1;
757	1041	}
758	1042	return ref;
759	1043	}
760	1044
761		-/* Called with kvm->lock held */
	1045	+/* Called with kvm->mmu_lock held */
762	1046	int kvm_test_age_radix(struct kvm kvm, struct kvm_memory_slot memslot,
763	1047	unsigned long gfn)
764	1048	{
..	..	@@ -767,7 +1051,10 @@
767	1051	unsigned int shift;
768	1052	int ref = 0;
769	1053
770		- ptep = __find_linux_pte(kvm->arch.pgtable, gpa, NULL, &shift);
	1054	+ if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE)
	1055	+ return ref;
	1056	+
	1057	+ ptep = find_kvm_secondary_pte(kvm, gpa, &shift);
771	1058	if (ptep && pte_present(ptep) && pte_young(ptep))
772	1059	ref = 1;
773	1060	return ref;
..	..	@@ -779,18 +1066,52 @@
779	1066	{
780	1067	unsigned long gfn = memslot->base_gfn + pagenum;
781	1068	unsigned long gpa = gfn << PAGE_SHIFT;
782		- pte_t *ptep;
	1069	+ pte_t *ptep, pte;
783	1070	unsigned int shift;
784	1071	int ret = 0;
	1072	+ unsigned long old, *rmapp;
785	1073
786		- ptep = __find_linux_pte(kvm->arch.pgtable, gpa, NULL, &shift);
787		- if (ptep && pte_present(ptep) && pte_dirty(ptep)) {
	1074	+ if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE)
	1075	+ return ret;
	1076	+
	1077	+ /*
	1078	+ * For performance reasons we don't hold kvm->mmu_lock while walking the
	1079	+ * partition scoped table.
	1080	+ */
	1081	+ ptep = find_kvm_secondary_pte_unlocked(kvm, gpa, &shift);
	1082	+ if (!ptep)
	1083	+ return 0;
	1084	+
	1085	+ pte = READ_ONCE(*ptep);
	1086	+ if (pte_present(pte) && pte_dirty(pte)) {
	1087	+ spin_lock(&kvm->mmu_lock);
	1088	+ /*
	1089	+ * Recheck the pte again
	1090	+ */
	1091	+ if (pte_val(pte) != pte_val(*ptep)) {
	1092	+ /*
	1093	+ * We have KVM_MEM_LOG_DIRTY_PAGES enabled. Hence we can
	1094	+ * only find PAGE_SIZE pte entries here. We can continue
	1095	+ * to use the pte addr returned by above page table
	1096	+ * walk.
	1097	+ */
	1098	+ if (!pte_present(ptep) \|\| !pte_dirty(ptep)) {
	1099	+ spin_unlock(&kvm->mmu_lock);
	1100	+ return 0;
	1101	+ }
	1102	+ }
	1103	+
788	1104	ret = 1;
789		- if (shift)
790		- ret = 1 << (shift - PAGE_SHIFT);
791		- kvmppc_radix_update_pte(kvm, ptep, _PAGE_DIRTY, 0,
792		- gpa, shift);
793		- kvmppc_radix_tlbie_page(kvm, gpa, shift);
	1105	+ VM_BUG_ON(shift);
	1106	+ old = kvmppc_radix_update_pte(kvm, ptep, _PAGE_DIRTY, 0,
	1107	+ gpa, shift);
	1108	+ kvmppc_radix_tlbie_page(kvm, gpa, shift, kvm->arch.lpid);
	1109	+ /* Also clear bit in ptes in shadow pgtable for nested guests */
	1110	+ rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
	1111	+ kvmhv_update_nest_rmap_rc_list(kvm, rmapp, _PAGE_DIRTY, 0,
	1112	+ old & PTE_RPN_MASK,
	1113	+ 1UL << shift);
	1114	+ spin_unlock(&kvm->mmu_lock);
794	1115	}
795	1116	return ret;
796	1117	}
..	..	@@ -818,6 +1139,37 @@
818	1139	}
819	1140	}
820	1141	return 0;
	1142	+}
	1143	+
	1144	+void kvmppc_radix_flush_memslot(struct kvm *kvm,
	1145	+ const struct kvm_memory_slot *memslot)
	1146	+{
	1147	+ unsigned long n;
	1148	+ pte_t *ptep;
	1149	+ unsigned long gpa;
	1150	+ unsigned int shift;
	1151	+
	1152	+ if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_START)
	1153	+ kvmppc_uvmem_drop_pages(memslot, kvm, true);
	1154	+
	1155	+ if (kvm->arch.secure_guest & KVMPPC_SECURE_INIT_DONE)
	1156	+ return;
	1157	+
	1158	+ gpa = memslot->base_gfn << PAGE_SHIFT;
	1159	+ spin_lock(&kvm->mmu_lock);
	1160	+ for (n = memslot->npages; n; --n) {
	1161	+ ptep = find_kvm_secondary_pte(kvm, gpa, &shift);
	1162	+ if (ptep && pte_present(*ptep))
	1163	+ kvmppc_unmap_pte(kvm, ptep, gpa, shift, memslot,
	1164	+ kvm->arch.lpid);
	1165	+ gpa += PAGE_SIZE;
	1166	+ }
	1167	+ /*
	1168	+ * Increase the mmu notifier sequence number to prevent any page
	1169	+ * fault that read the memslot earlier from writing a PTE.
	1170	+ */
	1171	+ kvm->mmu_notifier_seq++;
	1172	+ spin_unlock(&kvm->mmu_lock);
821	1173	}
822	1174
823	1175	static void add_rmmu_ap_encoding(struct kvm_ppc_rmmu_info *info,
..	..	@@ -875,6 +1227,216 @@
875	1227	memset(addr, 0, RADIX_PMD_TABLE_SIZE);
876	1228	}
877	1229
	1230	+struct debugfs_radix_state {
	1231	+ struct kvm *kvm;
	1232	+ struct mutex mutex;
	1233	+ unsigned long gpa;
	1234	+ int lpid;
	1235	+ int chars_left;
	1236	+ int buf_index;
	1237	+ char buf[128];
	1238	+ u8 hdr;
	1239	+};
	1240	+
	1241	+static int debugfs_radix_open(struct inode inode, struct file file)
	1242	+{
	1243	+ struct kvm *kvm = inode->i_private;
	1244	+ struct debugfs_radix_state *p;
	1245	+
	1246	+ p = kzalloc(sizeof(*p), GFP_KERNEL);
	1247	+ if (!p)
	1248	+ return -ENOMEM;
	1249	+
	1250	+ kvm_get_kvm(kvm);
	1251	+ p->kvm = kvm;
	1252	+ mutex_init(&p->mutex);
	1253	+ file->private_data = p;
	1254	+
	1255	+ return nonseekable_open(inode, file);
	1256	+}
	1257	+
	1258	+static int debugfs_radix_release(struct inode inode, struct file file)
	1259	+{
	1260	+ struct debugfs_radix_state *p = file->private_data;
	1261	+
	1262	+ kvm_put_kvm(p->kvm);
	1263	+ kfree(p);
	1264	+ return 0;
	1265	+}
	1266	+
	1267	+static ssize_t debugfs_radix_read(struct file file, char __user buf,
	1268	+ size_t len, loff_t *ppos)
	1269	+{
	1270	+ struct debugfs_radix_state *p = file->private_data;
	1271	+ ssize_t ret, r;
	1272	+ unsigned long n;
	1273	+ struct kvm *kvm;
	1274	+ unsigned long gpa;
	1275	+ pgd_t *pgt;
	1276	+ struct kvm_nested_guest *nested;
	1277	+ pgd_t *pgdp;
	1278	+ p4d_t p4d, *p4dp;
	1279	+ pud_t pud, *pudp;
	1280	+ pmd_t pmd, *pmdp;
	1281	+ pte_t *ptep;
	1282	+ int shift;
	1283	+ unsigned long pte;
	1284	+
	1285	+ kvm = p->kvm;
	1286	+ if (!kvm_is_radix(kvm))
	1287	+ return 0;
	1288	+
	1289	+ ret = mutex_lock_interruptible(&p->mutex);
	1290	+ if (ret)
	1291	+ return ret;
	1292	+
	1293	+ if (p->chars_left) {
	1294	+ n = p->chars_left;
	1295	+ if (n > len)
	1296	+ n = len;
	1297	+ r = copy_to_user(buf, p->buf + p->buf_index, n);
	1298	+ n -= r;
	1299	+ p->chars_left -= n;
	1300	+ p->buf_index += n;
	1301	+ buf += n;
	1302	+ len -= n;
	1303	+ ret = n;
	1304	+ if (r) {
	1305	+ if (!n)
	1306	+ ret = -EFAULT;
	1307	+ goto out;
	1308	+ }
	1309	+ }
	1310	+
	1311	+ gpa = p->gpa;
	1312	+ nested = NULL;
	1313	+ pgt = NULL;
	1314	+ while (len != 0 && p->lpid >= 0) {
	1315	+ if (gpa >= RADIX_PGTABLE_RANGE) {
	1316	+ gpa = 0;
	1317	+ pgt = NULL;
	1318	+ if (nested) {
	1319	+ kvmhv_put_nested(nested);
	1320	+ nested = NULL;
	1321	+ }
	1322	+ p->lpid = kvmhv_nested_next_lpid(kvm, p->lpid);
	1323	+ p->hdr = 0;
	1324	+ if (p->lpid < 0)
	1325	+ break;
	1326	+ }
	1327	+ if (!pgt) {
	1328	+ if (p->lpid == 0) {
	1329	+ pgt = kvm->arch.pgtable;
	1330	+ } else {
	1331	+ nested = kvmhv_get_nested(kvm, p->lpid, false);
	1332	+ if (!nested) {
	1333	+ gpa = RADIX_PGTABLE_RANGE;
	1334	+ continue;
	1335	+ }
	1336	+ pgt = nested->shadow_pgtable;
	1337	+ }
	1338	+ }
	1339	+ n = 0;
	1340	+ if (!p->hdr) {
	1341	+ if (p->lpid > 0)
	1342	+ n = scnprintf(p->buf, sizeof(p->buf),
	1343	+ "\nNested LPID %d: ", p->lpid);
	1344	+ n += scnprintf(p->buf + n, sizeof(p->buf) - n,
	1345	+ "pgdir: %lx\n", (unsigned long)pgt);
	1346	+ p->hdr = 1;
	1347	+ goto copy;
	1348	+ }
	1349	+
	1350	+ pgdp = pgt + pgd_index(gpa);
	1351	+ p4dp = p4d_offset(pgdp, gpa);
	1352	+ p4d = READ_ONCE(*p4dp);
	1353	+ if (!(p4d_val(p4d) & _PAGE_PRESENT)) {
	1354	+ gpa = (gpa & P4D_MASK) + P4D_SIZE;
	1355	+ continue;
	1356	+ }
	1357	+
	1358	+ pudp = pud_offset(&p4d, gpa);
	1359	+ pud = READ_ONCE(*pudp);
	1360	+ if (!(pud_val(pud) & _PAGE_PRESENT)) {
	1361	+ gpa = (gpa & PUD_MASK) + PUD_SIZE;
	1362	+ continue;
	1363	+ }
	1364	+ if (pud_val(pud) & _PAGE_PTE) {
	1365	+ pte = pud_val(pud);
	1366	+ shift = PUD_SHIFT;
	1367	+ goto leaf;
	1368	+ }
	1369	+
	1370	+ pmdp = pmd_offset(&pud, gpa);
	1371	+ pmd = READ_ONCE(*pmdp);
	1372	+ if (!(pmd_val(pmd) & _PAGE_PRESENT)) {
	1373	+ gpa = (gpa & PMD_MASK) + PMD_SIZE;
	1374	+ continue;
	1375	+ }
	1376	+ if (pmd_val(pmd) & _PAGE_PTE) {
	1377	+ pte = pmd_val(pmd);
	1378	+ shift = PMD_SHIFT;
	1379	+ goto leaf;
	1380	+ }
	1381	+
	1382	+ ptep = pte_offset_kernel(&pmd, gpa);
	1383	+ pte = pte_val(READ_ONCE(*ptep));
	1384	+ if (!(pte & _PAGE_PRESENT)) {
	1385	+ gpa += PAGE_SIZE;
	1386	+ continue;
	1387	+ }
	1388	+ shift = PAGE_SHIFT;
	1389	+ leaf:
	1390	+ n = scnprintf(p->buf, sizeof(p->buf),
	1391	+ " %lx: %lx %d\n", gpa, pte, shift);
	1392	+ gpa += 1ul << shift;
	1393	+ copy:
	1394	+ p->chars_left = n;
	1395	+ if (n > len)
	1396	+ n = len;
	1397	+ r = copy_to_user(buf, p->buf, n);
	1398	+ n -= r;
	1399	+ p->chars_left -= n;
	1400	+ p->buf_index = n;
	1401	+ buf += n;
	1402	+ len -= n;
	1403	+ ret += n;
	1404	+ if (r) {
	1405	+ if (!ret)
	1406	+ ret = -EFAULT;
	1407	+ break;
	1408	+ }
	1409	+ }
	1410	+ p->gpa = gpa;
	1411	+ if (nested)
	1412	+ kvmhv_put_nested(nested);
	1413	+
	1414	+ out:
	1415	+ mutex_unlock(&p->mutex);
	1416	+ return ret;
	1417	+}
	1418	+
	1419	+static ssize_t debugfs_radix_write(struct file file, const char __user buf,
	1420	+ size_t len, loff_t *ppos)
	1421	+{
	1422	+ return -EACCES;
	1423	+}
	1424	+
	1425	+static const struct file_operations debugfs_radix_fops = {
	1426	+ .owner = THIS_MODULE,
	1427	+ .open = debugfs_radix_open,
	1428	+ .release = debugfs_radix_release,
	1429	+ .read = debugfs_radix_read,
	1430	+ .write = debugfs_radix_write,
	1431	+ .llseek = generic_file_llseek,
	1432	+};
	1433	+
	1434	+void kvmhv_radix_debugfs_init(struct kvm *kvm)
	1435	+{
	1436	+ debugfs_create_file("radix", 0400, kvm->arch.debugfs_dir, kvm,
	1437	+ &debugfs_radix_fops);
	1438	+}
	1439	+
878	1440	int kvmppc_radix_init(void)
879	1441	{
880	1442	unsigned long size = sizeof(void *) << RADIX_PTE_INDEX_SIZE;