xref: /openbmc/linux/arch/powerpc/kvm/book3s_64_vio.c (revision 4f6cce39)
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>
17  * Copyright 2016 Alexey Kardashevskiy, IBM Corporation <aik@au1.ibm.com>
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/sched/signal.h>
28 #include <linux/hugetlb.h>
29 #include <linux/list.h>
30 #include <linux/anon_inodes.h>
31 
32 #include <asm/tlbflush.h>
33 #include <asm/kvm_ppc.h>
34 #include <asm/kvm_book3s.h>
35 #include <asm/book3s/64/mmu-hash.h>
36 #include <asm/hvcall.h>
37 #include <asm/synch.h>
38 #include <asm/ppc-opcode.h>
39 #include <asm/kvm_host.h>
40 #include <asm/udbg.h>
41 #include <asm/iommu.h>
42 #include <asm/tce.h>
43 
44 static unsigned long kvmppc_tce_pages(unsigned long iommu_pages)
45 {
46 	return ALIGN(iommu_pages * sizeof(u64), PAGE_SIZE) / PAGE_SIZE;
47 }
48 
49 static unsigned long kvmppc_stt_pages(unsigned long tce_pages)
50 {
51 	unsigned long stt_bytes = sizeof(struct kvmppc_spapr_tce_table) +
52 			(tce_pages * sizeof(struct page *));
53 
54 	return tce_pages + ALIGN(stt_bytes, PAGE_SIZE) / PAGE_SIZE;
55 }
56 
57 static long kvmppc_account_memlimit(unsigned long stt_pages, bool inc)
58 {
59 	long ret = 0;
60 
61 	if (!current || !current->mm)
62 		return ret; /* process exited */
63 
64 	down_write(&current->mm->mmap_sem);
65 
66 	if (inc) {
67 		unsigned long locked, lock_limit;
68 
69 		locked = current->mm->locked_vm + stt_pages;
70 		lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
71 		if (locked > lock_limit && !capable(CAP_IPC_LOCK))
72 			ret = -ENOMEM;
73 		else
74 			current->mm->locked_vm += stt_pages;
75 	} else {
76 		if (WARN_ON_ONCE(stt_pages > current->mm->locked_vm))
77 			stt_pages = current->mm->locked_vm;
78 
79 		current->mm->locked_vm -= stt_pages;
80 	}
81 
82 	pr_debug("[%d] RLIMIT_MEMLOCK KVM %c%ld %ld/%ld%s\n", current->pid,
83 			inc ? '+' : '-',
84 			stt_pages << PAGE_SHIFT,
85 			current->mm->locked_vm << PAGE_SHIFT,
86 			rlimit(RLIMIT_MEMLOCK),
87 			ret ? " - exceeded" : "");
88 
89 	up_write(&current->mm->mmap_sem);
90 
91 	return ret;
92 }
93 
94 static void release_spapr_tce_table(struct rcu_head *head)
95 {
96 	struct kvmppc_spapr_tce_table *stt = container_of(head,
97 			struct kvmppc_spapr_tce_table, rcu);
98 	unsigned long i, npages = kvmppc_tce_pages(stt->size);
99 
100 	for (i = 0; i < npages; i++)
101 		__free_page(stt->pages[i]);
102 
103 	kfree(stt);
104 }
105 
106 static int kvm_spapr_tce_fault(struct vm_fault *vmf)
107 {
108 	struct kvmppc_spapr_tce_table *stt = vmf->vma->vm_file->private_data;
109 	struct page *page;
110 
111 	if (vmf->pgoff >= kvmppc_tce_pages(stt->size))
112 		return VM_FAULT_SIGBUS;
113 
114 	page = stt->pages[vmf->pgoff];
115 	get_page(page);
116 	vmf->page = page;
117 	return 0;
118 }
119 
120 static const struct vm_operations_struct kvm_spapr_tce_vm_ops = {
121 	.fault = kvm_spapr_tce_fault,
122 };
123 
124 static int kvm_spapr_tce_mmap(struct file *file, struct vm_area_struct *vma)
125 {
126 	vma->vm_ops = &kvm_spapr_tce_vm_ops;
127 	return 0;
128 }
129 
130 static int kvm_spapr_tce_release(struct inode *inode, struct file *filp)
131 {
132 	struct kvmppc_spapr_tce_table *stt = filp->private_data;
133 
134 	list_del_rcu(&stt->list);
135 
136 	kvm_put_kvm(stt->kvm);
137 
138 	kvmppc_account_memlimit(
139 		kvmppc_stt_pages(kvmppc_tce_pages(stt->size)), false);
140 	call_rcu(&stt->rcu, release_spapr_tce_table);
141 
142 	return 0;
143 }
144 
145 static const struct file_operations kvm_spapr_tce_fops = {
146 	.mmap           = kvm_spapr_tce_mmap,
147 	.release	= kvm_spapr_tce_release,
148 };
149 
150 long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
151 				   struct kvm_create_spapr_tce_64 *args)
152 {
153 	struct kvmppc_spapr_tce_table *stt = NULL;
154 	unsigned long npages, size;
155 	int ret = -ENOMEM;
156 	int i;
157 
158 	if (!args->size)
159 		return -EINVAL;
160 
161 	/* Check this LIOBN hasn't been previously allocated */
162 	list_for_each_entry(stt, &kvm->arch.spapr_tce_tables, list) {
163 		if (stt->liobn == args->liobn)
164 			return -EBUSY;
165 	}
166 
167 	size = args->size;
168 	npages = kvmppc_tce_pages(size);
169 	ret = kvmppc_account_memlimit(kvmppc_stt_pages(npages), true);
170 	if (ret) {
171 		stt = NULL;
172 		goto fail;
173 	}
174 
175 	ret = -ENOMEM;
176 	stt = kzalloc(sizeof(*stt) + npages * sizeof(struct page *),
177 		      GFP_KERNEL);
178 	if (!stt)
179 		goto fail;
180 
181 	stt->liobn = args->liobn;
182 	stt->page_shift = args->page_shift;
183 	stt->offset = args->offset;
184 	stt->size = size;
185 	stt->kvm = kvm;
186 
187 	for (i = 0; i < npages; i++) {
188 		stt->pages[i] = alloc_page(GFP_KERNEL | __GFP_ZERO);
189 		if (!stt->pages[i])
190 			goto fail;
191 	}
192 
193 	kvm_get_kvm(kvm);
194 
195 	mutex_lock(&kvm->lock);
196 	list_add_rcu(&stt->list, &kvm->arch.spapr_tce_tables);
197 
198 	mutex_unlock(&kvm->lock);
199 
200 	return anon_inode_getfd("kvm-spapr-tce", &kvm_spapr_tce_fops,
201 				stt, O_RDWR | O_CLOEXEC);
202 
203 fail:
204 	if (stt) {
205 		for (i = 0; i < npages; i++)
206 			if (stt->pages[i])
207 				__free_page(stt->pages[i]);
208 
209 		kfree(stt);
210 	}
211 	return ret;
212 }
213 
214 long kvmppc_h_put_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
215 		      unsigned long ioba, unsigned long tce)
216 {
217 	struct kvmppc_spapr_tce_table *stt = kvmppc_find_table(vcpu, liobn);
218 	long ret;
219 
220 	/* udbg_printf("H_PUT_TCE(): liobn=0x%lx ioba=0x%lx, tce=0x%lx\n", */
221 	/* 	    liobn, ioba, tce); */
222 
223 	if (!stt)
224 		return H_TOO_HARD;
225 
226 	ret = kvmppc_ioba_validate(stt, ioba, 1);
227 	if (ret != H_SUCCESS)
228 		return ret;
229 
230 	ret = kvmppc_tce_validate(stt, tce);
231 	if (ret != H_SUCCESS)
232 		return ret;
233 
234 	kvmppc_tce_put(stt, ioba >> stt->page_shift, tce);
235 
236 	return H_SUCCESS;
237 }
238 EXPORT_SYMBOL_GPL(kvmppc_h_put_tce);
239 
240 long kvmppc_h_put_tce_indirect(struct kvm_vcpu *vcpu,
241 		unsigned long liobn, unsigned long ioba,
242 		unsigned long tce_list, unsigned long npages)
243 {
244 	struct kvmppc_spapr_tce_table *stt;
245 	long i, ret = H_SUCCESS, idx;
246 	unsigned long entry, ua = 0;
247 	u64 __user *tces;
248 	u64 tce;
249 
250 	stt = kvmppc_find_table(vcpu, liobn);
251 	if (!stt)
252 		return H_TOO_HARD;
253 
254 	entry = ioba >> stt->page_shift;
255 	/*
256 	 * SPAPR spec says that the maximum size of the list is 512 TCEs
257 	 * so the whole table fits in 4K page
258 	 */
259 	if (npages > 512)
260 		return H_PARAMETER;
261 
262 	if (tce_list & (SZ_4K - 1))
263 		return H_PARAMETER;
264 
265 	ret = kvmppc_ioba_validate(stt, ioba, npages);
266 	if (ret != H_SUCCESS)
267 		return ret;
268 
269 	idx = srcu_read_lock(&vcpu->kvm->srcu);
270 	if (kvmppc_gpa_to_ua(vcpu->kvm, tce_list, &ua, NULL)) {
271 		ret = H_TOO_HARD;
272 		goto unlock_exit;
273 	}
274 	tces = (u64 __user *) ua;
275 
276 	for (i = 0; i < npages; ++i) {
277 		if (get_user(tce, tces + i)) {
278 			ret = H_TOO_HARD;
279 			goto unlock_exit;
280 		}
281 		tce = be64_to_cpu(tce);
282 
283 		ret = kvmppc_tce_validate(stt, tce);
284 		if (ret != H_SUCCESS)
285 			goto unlock_exit;
286 
287 		kvmppc_tce_put(stt, entry + i, tce);
288 	}
289 
290 unlock_exit:
291 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
292 
293 	return ret;
294 }
295 EXPORT_SYMBOL_GPL(kvmppc_h_put_tce_indirect);
296 
297 long kvmppc_h_stuff_tce(struct kvm_vcpu *vcpu,
298 		unsigned long liobn, unsigned long ioba,
299 		unsigned long tce_value, unsigned long npages)
300 {
301 	struct kvmppc_spapr_tce_table *stt;
302 	long i, ret;
303 
304 	stt = kvmppc_find_table(vcpu, liobn);
305 	if (!stt)
306 		return H_TOO_HARD;
307 
308 	ret = kvmppc_ioba_validate(stt, ioba, npages);
309 	if (ret != H_SUCCESS)
310 		return ret;
311 
312 	/* Check permission bits only to allow userspace poison TCE for debug */
313 	if (tce_value & (TCE_PCI_WRITE | TCE_PCI_READ))
314 		return H_PARAMETER;
315 
316 	for (i = 0; i < npages; ++i, ioba += (1ULL << stt->page_shift))
317 		kvmppc_tce_put(stt, ioba >> stt->page_shift, tce_value);
318 
319 	return H_SUCCESS;
320 }
321 EXPORT_SYMBOL_GPL(kvmppc_h_stuff_tce);
322