1// SPDX-License-Identifier: GPL-2.0
2/*
3 * IOMMU API for s390 PCI devices
4 *
5 * Copyright IBM Corp. 2015
6 * Author(s): Gerald Schaefer <gerald.schaefer@de.ibm.com>
7 */
8
9#include <linux/pci.h>
10#include <linux/iommu.h>
11#include <linux/iommu-helper.h>
12#include <linux/sizes.h>
13#include <asm/pci_dma.h>
14
15/*
16 * Physically contiguous memory regions can be mapped with 4 KiB alignment,
17 * we allow all page sizes that are an order of 4KiB (no special large page
18 * support so far).
19 */
20#define S390_IOMMU_PGSIZES	(~0xFFFUL)
21
22static const struct iommu_ops s390_iommu_ops;
23
24struct s390_domain {
25	struct iommu_domain	domain;
26	struct list_head	devices;
27	unsigned long		*dma_table;
28	spinlock_t		dma_table_lock;
29	spinlock_t		list_lock;
30};
31
32struct s390_domain_device {
33	struct list_head	list;
34	struct zpci_dev		*zdev;
35};
36
37static struct s390_domain *to_s390_domain(struct iommu_domain *dom)
38{
39	return container_of(dom, struct s390_domain, domain);
40}
41
42static bool s390_iommu_capable(enum iommu_cap cap)
43{
44	switch (cap) {
45	case IOMMU_CAP_CACHE_COHERENCY:
46		return true;
47	case IOMMU_CAP_INTR_REMAP:
48		return true;
49	default:
50		return false;
51	}
52}
53
54static struct iommu_domain *s390_domain_alloc(unsigned domain_type)
55{
56	struct s390_domain *s390_domain;
57
58	if (domain_type != IOMMU_DOMAIN_UNMANAGED)
59		return NULL;
60
61	s390_domain = kzalloc(sizeof(*s390_domain), GFP_KERNEL);
62	if (!s390_domain)
63		return NULL;
64
65	s390_domain->dma_table = dma_alloc_cpu_table();
66	if (!s390_domain->dma_table) {
67		kfree(s390_domain);
68		return NULL;
69	}
70
71	spin_lock_init(&s390_domain->dma_table_lock);
72	spin_lock_init(&s390_domain->list_lock);
73	INIT_LIST_HEAD(&s390_domain->devices);
74
75	return &s390_domain->domain;
76}
77
78static void s390_domain_free(struct iommu_domain *domain)
79{
80	struct s390_domain *s390_domain = to_s390_domain(domain);
81
82	dma_cleanup_tables(s390_domain->dma_table);
83	kfree(s390_domain);
84}
85
86static int s390_iommu_attach_device(struct iommu_domain *domain,
87				    struct device *dev)
88{
89	struct s390_domain *s390_domain = to_s390_domain(domain);
90	struct zpci_dev *zdev = to_zpci_dev(dev);
91	struct s390_domain_device *domain_device;
92	unsigned long flags;
93	int rc;
94
95	if (!zdev)
96		return -ENODEV;
97
98	domain_device = kzalloc(sizeof(*domain_device), GFP_KERNEL);
99	if (!domain_device)
100		return -ENOMEM;
101
102	if (zdev->dma_table)
103		zpci_dma_exit_device(zdev);
104
105	zdev->dma_table = s390_domain->dma_table;
106	rc = zpci_register_ioat(zdev, 0, zdev->start_dma, zdev->end_dma,
107				(u64) zdev->dma_table);
108	if (rc)
109		goto out_restore;
110
111	spin_lock_irqsave(&s390_domain->list_lock, flags);
112	/* First device defines the DMA range limits */
113	if (list_empty(&s390_domain->devices)) {
114		domain->geometry.aperture_start = zdev->start_dma;
115		domain->geometry.aperture_end = zdev->end_dma;
116		domain->geometry.force_aperture = true;
117	/* Allow only devices with identical DMA range limits */
118	} else if (domain->geometry.aperture_start != zdev->start_dma ||
119		   domain->geometry.aperture_end != zdev->end_dma) {
120		rc = -EINVAL;
121		spin_unlock_irqrestore(&s390_domain->list_lock, flags);
122		goto out_restore;
123	}
124	domain_device->zdev = zdev;
125	zdev->s390_domain = s390_domain;
126	list_add(&domain_device->list, &s390_domain->devices);
127	spin_unlock_irqrestore(&s390_domain->list_lock, flags);
128
129	return 0;
130
131out_restore:
132	zpci_dma_init_device(zdev);
133	kfree(domain_device);
134
135	return rc;
136}
137
138static void s390_iommu_detach_device(struct iommu_domain *domain,
139				     struct device *dev)
140{
141	struct s390_domain *s390_domain = to_s390_domain(domain);
142	struct zpci_dev *zdev = to_zpci_dev(dev);
143	struct s390_domain_device *domain_device, *tmp;
144	unsigned long flags;
145	int found = 0;
146
147	if (!zdev)
148		return;
149
150	spin_lock_irqsave(&s390_domain->list_lock, flags);
151	list_for_each_entry_safe(domain_device, tmp, &s390_domain->devices,
152				 list) {
153		if (domain_device->zdev == zdev) {
154			list_del(&domain_device->list);
155			kfree(domain_device);
156			found = 1;
157			break;
158		}
159	}
160	spin_unlock_irqrestore(&s390_domain->list_lock, flags);
161
162	if (found) {
163		zdev->s390_domain = NULL;
164		zpci_unregister_ioat(zdev, 0);
165		zpci_dma_init_device(zdev);
166	}
167}
168
169static struct iommu_device *s390_iommu_probe_device(struct device *dev)
170{
171	struct zpci_dev *zdev = to_zpci_dev(dev);
172
173	return &zdev->iommu_dev;
174}
175
176static void s390_iommu_release_device(struct device *dev)
177{
178	struct zpci_dev *zdev = to_zpci_dev(dev);
179	struct iommu_domain *domain;
180
181	/*
182	 * This is a workaround for a scenario where the IOMMU API common code
183	 * "forgets" to call the detach_dev callback: After binding a device
184	 * to vfio-pci and completing the VFIO_SET_IOMMU ioctl (which triggers
185	 * the attach_dev), removing the device via
186	 * "echo 1 > /sys/bus/pci/devices/.../remove" won't trigger detach_dev,
187	 * only release_device will be called via the BUS_NOTIFY_REMOVED_DEVICE
188	 * notifier.
189	 *
190	 * So let's call detach_dev from here if it hasn't been called before.
191	 */
192	if (zdev && zdev->s390_domain) {
193		domain = iommu_get_domain_for_dev(dev);
194		if (domain)
195			s390_iommu_detach_device(domain, dev);
196	}
197}
198
199static int s390_iommu_update_trans(struct s390_domain *s390_domain,
200				   unsigned long pa, dma_addr_t dma_addr,
201				   size_t size, int flags)
202{
203	struct s390_domain_device *domain_device;
204	u8 *page_addr = (u8 *) (pa & PAGE_MASK);
205	dma_addr_t start_dma_addr = dma_addr;
206	unsigned long irq_flags, nr_pages, i;
207	unsigned long *entry;
208	int rc = 0;
209
210	if (dma_addr < s390_domain->domain.geometry.aperture_start ||
211	    dma_addr + size > s390_domain->domain.geometry.aperture_end)
212		return -EINVAL;
213
214	nr_pages = PAGE_ALIGN(size) >> PAGE_SHIFT;
215	if (!nr_pages)
216		return 0;
217
218	spin_lock_irqsave(&s390_domain->dma_table_lock, irq_flags);
219	for (i = 0; i < nr_pages; i++) {
220		entry = dma_walk_cpu_trans(s390_domain->dma_table, dma_addr);
221		if (!entry) {
222			rc = -ENOMEM;
223			goto undo_cpu_trans;
224		}
225		dma_update_cpu_trans(entry, page_addr, flags);
226		page_addr += PAGE_SIZE;
227		dma_addr += PAGE_SIZE;
228	}
229
230	spin_lock(&s390_domain->list_lock);
231	list_for_each_entry(domain_device, &s390_domain->devices, list) {
232		rc = zpci_refresh_trans((u64) domain_device->zdev->fh << 32,
233					start_dma_addr, nr_pages * PAGE_SIZE);
234		if (rc)
235			break;
236	}
237	spin_unlock(&s390_domain->list_lock);
238
239undo_cpu_trans:
240	if (rc && ((flags & ZPCI_PTE_VALID_MASK) == ZPCI_PTE_VALID)) {
241		flags = ZPCI_PTE_INVALID;
242		while (i-- > 0) {
243			page_addr -= PAGE_SIZE;
244			dma_addr -= PAGE_SIZE;
245			entry = dma_walk_cpu_trans(s390_domain->dma_table,
246						   dma_addr);
247			if (!entry)
248				break;
249			dma_update_cpu_trans(entry, page_addr, flags);
250		}
251	}
252	spin_unlock_irqrestore(&s390_domain->dma_table_lock, irq_flags);
253
254	return rc;
255}
256
257static int s390_iommu_map(struct iommu_domain *domain, unsigned long iova,
258			  phys_addr_t paddr, size_t size, int prot, gfp_t gfp)
259{
260	struct s390_domain *s390_domain = to_s390_domain(domain);
261	int flags = ZPCI_PTE_VALID, rc = 0;
262
263	if (!(prot & IOMMU_READ))
264		return -EINVAL;
265
266	if (!(prot & IOMMU_WRITE))
267		flags |= ZPCI_TABLE_PROTECTED;
268
269	rc = s390_iommu_update_trans(s390_domain, (unsigned long) paddr, iova,
270				     size, flags);
271
272	return rc;
273}
274
275static phys_addr_t s390_iommu_iova_to_phys(struct iommu_domain *domain,
276					   dma_addr_t iova)
277{
278	struct s390_domain *s390_domain = to_s390_domain(domain);
279	unsigned long *sto, *pto, *rto, flags;
280	unsigned int rtx, sx, px;
281	phys_addr_t phys = 0;
282
283	if (iova < domain->geometry.aperture_start ||
284	    iova > domain->geometry.aperture_end)
285		return 0;
286
287	rtx = calc_rtx(iova);
288	sx = calc_sx(iova);
289	px = calc_px(iova);
290	rto = s390_domain->dma_table;
291
292	spin_lock_irqsave(&s390_domain->dma_table_lock, flags);
293	if (rto && reg_entry_isvalid(rto[rtx])) {
294		sto = get_rt_sto(rto[rtx]);
295		if (sto && reg_entry_isvalid(sto[sx])) {
296			pto = get_st_pto(sto[sx]);
297			if (pto && pt_entry_isvalid(pto[px]))
298				phys = pto[px] & ZPCI_PTE_ADDR_MASK;
299		}
300	}
301	spin_unlock_irqrestore(&s390_domain->dma_table_lock, flags);
302
303	return phys;
304}
305
306static size_t s390_iommu_unmap(struct iommu_domain *domain,
307			       unsigned long iova, size_t size,
308			       struct iommu_iotlb_gather *gather)
309{
310	struct s390_domain *s390_domain = to_s390_domain(domain);
311	int flags = ZPCI_PTE_INVALID;
312	phys_addr_t paddr;
313	int rc;
314
315	paddr = s390_iommu_iova_to_phys(domain, iova);
316	if (!paddr)
317		return 0;
318
319	rc = s390_iommu_update_trans(s390_domain, (unsigned long) paddr, iova,
320				     size, flags);
321	if (rc)
322		return 0;
323
324	return size;
325}
326
327int zpci_init_iommu(struct zpci_dev *zdev)
328{
329	int rc = 0;
330
331	rc = iommu_device_sysfs_add(&zdev->iommu_dev, NULL, NULL,
332				    "s390-iommu.%08x", zdev->fid);
333	if (rc)
334		goto out_err;
335
336	iommu_device_set_ops(&zdev->iommu_dev, &s390_iommu_ops);
337
338	rc = iommu_device_register(&zdev->iommu_dev);
339	if (rc)
340		goto out_sysfs;
341
342	return 0;
343
344out_sysfs:
345	iommu_device_sysfs_remove(&zdev->iommu_dev);
346
347out_err:
348	return rc;
349}
350
351void zpci_destroy_iommu(struct zpci_dev *zdev)
352{
353	iommu_device_unregister(&zdev->iommu_dev);
354	iommu_device_sysfs_remove(&zdev->iommu_dev);
355}
356
357static const struct iommu_ops s390_iommu_ops = {
358	.capable = s390_iommu_capable,
359	.domain_alloc = s390_domain_alloc,
360	.domain_free = s390_domain_free,
361	.attach_dev = s390_iommu_attach_device,
362	.detach_dev = s390_iommu_detach_device,
363	.map = s390_iommu_map,
364	.unmap = s390_iommu_unmap,
365	.iova_to_phys = s390_iommu_iova_to_phys,
366	.probe_device = s390_iommu_probe_device,
367	.release_device = s390_iommu_release_device,
368	.device_group = generic_device_group,
369	.pgsize_bitmap = S390_IOMMU_PGSIZES,
370};
371
372static int __init s390_iommu_init(void)
373{
374	return bus_set_iommu(&pci_bus_type, &s390_iommu_ops);
375}
376subsys_initcall(s390_iommu_init);
377