2 * IOMMU for IPMMU/IPMMUI
3 * Copyright (C) 2012 Hideki EIRAKU
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; version 2 of the License.
10 #include <linux/dma-mapping.h>
12 #include <linux/iommu.h>
13 #include <linux/platform_device.h>
14 #include <linux/sizes.h>
15 #include <linux/slab.h>
16 #include <asm/dma-iommu.h>
17 #include "shmobile-ipmmu.h"
19 #define L1_SIZE CONFIG_SHMOBILE_IOMMU_L1SIZE
20 #define L1_LEN (L1_SIZE / 4)
21 #define L1_ALIGN L1_SIZE
23 #define L2_LEN (L2_SIZE / 4)
24 #define L2_ALIGN L2_SIZE
26 struct shmobile_iommu_domain_pgtable {
31 struct shmobile_iommu_archdata {
32 struct list_head attached_list;
33 struct dma_iommu_mapping *iommu_mapping;
34 spinlock_t attach_lock;
35 struct shmobile_iommu_domain *attached;
36 int num_attached_devices;
37 struct shmobile_ipmmu *ipmmu;
40 struct shmobile_iommu_domain {
41 struct shmobile_iommu_domain_pgtable l1, l2[L1_LEN];
43 spinlock_t attached_list_lock;
44 struct list_head attached_list;
45 struct iommu_domain domain;
48 static struct shmobile_iommu_archdata *ipmmu_archdata;
49 static struct kmem_cache *l1cache, *l2cache;
51 static struct shmobile_iommu_domain *to_sh_domain(struct iommu_domain *dom)
53 return container_of(dom, struct shmobile_iommu_domain, domain);
56 static int pgtable_alloc(struct shmobile_iommu_domain_pgtable *pgtable,
57 struct kmem_cache *cache, size_t size)
59 pgtable->pgtable = kmem_cache_zalloc(cache, GFP_ATOMIC);
60 if (!pgtable->pgtable)
62 pgtable->handle = dma_map_single(NULL, pgtable->pgtable, size,
67 static void pgtable_free(struct shmobile_iommu_domain_pgtable *pgtable,
68 struct kmem_cache *cache, size_t size)
70 dma_unmap_single(NULL, pgtable->handle, size, DMA_TO_DEVICE);
71 kmem_cache_free(cache, pgtable->pgtable);
74 static uint32_t pgtable_read(struct shmobile_iommu_domain_pgtable *pgtable,
77 return pgtable->pgtable[index];
80 static void pgtable_write(struct shmobile_iommu_domain_pgtable *pgtable,
81 unsigned int index, unsigned int count, uint32_t val)
85 for (i = 0; i < count; i++)
86 pgtable->pgtable[index + i] = val;
87 dma_sync_single_for_device(NULL, pgtable->handle + index * sizeof(val),
88 sizeof(val) * count, DMA_TO_DEVICE);
91 static struct iommu_domain *shmobile_iommu_domain_alloc(unsigned type)
93 struct shmobile_iommu_domain *sh_domain;
96 if (type != IOMMU_DOMAIN_UNMANAGED)
99 sh_domain = kzalloc(sizeof(*sh_domain), GFP_KERNEL);
102 ret = pgtable_alloc(&sh_domain->l1, l1cache, L1_SIZE);
107 for (i = 0; i < L1_LEN; i++)
108 sh_domain->l2[i].pgtable = NULL;
109 spin_lock_init(&sh_domain->map_lock);
110 spin_lock_init(&sh_domain->attached_list_lock);
111 INIT_LIST_HEAD(&sh_domain->attached_list);
112 return &sh_domain->domain;
115 static void shmobile_iommu_domain_free(struct iommu_domain *domain)
117 struct shmobile_iommu_domain *sh_domain = to_sh_domain(domain);
120 for (i = 0; i < L1_LEN; i++) {
121 if (sh_domain->l2[i].pgtable)
122 pgtable_free(&sh_domain->l2[i], l2cache, L2_SIZE);
124 pgtable_free(&sh_domain->l1, l1cache, L1_SIZE);
128 static int shmobile_iommu_attach_device(struct iommu_domain *domain,
131 struct shmobile_iommu_archdata *archdata = dev->archdata.iommu;
132 struct shmobile_iommu_domain *sh_domain = to_sh_domain(domain);
137 spin_lock(&sh_domain->attached_list_lock);
138 spin_lock(&archdata->attach_lock);
139 if (archdata->attached != sh_domain) {
140 if (archdata->attached)
142 ipmmu_tlb_set(archdata->ipmmu, sh_domain->l1.handle, L1_SIZE,
144 ipmmu_tlb_flush(archdata->ipmmu);
145 archdata->attached = sh_domain;
146 archdata->num_attached_devices = 0;
147 list_add(&archdata->attached_list, &sh_domain->attached_list);
149 archdata->num_attached_devices++;
152 spin_unlock(&archdata->attach_lock);
153 spin_unlock(&sh_domain->attached_list_lock);
157 static void shmobile_iommu_detach_device(struct iommu_domain *domain,
160 struct shmobile_iommu_archdata *archdata = dev->archdata.iommu;
161 struct shmobile_iommu_domain *sh_domain = to_sh_domain(domain);
165 spin_lock(&sh_domain->attached_list_lock);
166 spin_lock(&archdata->attach_lock);
167 archdata->num_attached_devices--;
168 if (!archdata->num_attached_devices) {
169 ipmmu_tlb_set(archdata->ipmmu, 0, 0, 0);
170 ipmmu_tlb_flush(archdata->ipmmu);
171 archdata->attached = NULL;
172 list_del(&archdata->attached_list);
174 spin_unlock(&archdata->attach_lock);
175 spin_unlock(&sh_domain->attached_list_lock);
178 static void domain_tlb_flush(struct shmobile_iommu_domain *sh_domain)
180 struct shmobile_iommu_archdata *archdata;
182 spin_lock(&sh_domain->attached_list_lock);
183 list_for_each_entry(archdata, &sh_domain->attached_list, attached_list)
184 ipmmu_tlb_flush(archdata->ipmmu);
185 spin_unlock(&sh_domain->attached_list_lock);
188 static int l2alloc(struct shmobile_iommu_domain *sh_domain,
189 unsigned int l1index)
193 if (!sh_domain->l2[l1index].pgtable) {
194 ret = pgtable_alloc(&sh_domain->l2[l1index], l2cache, L2_SIZE);
198 pgtable_write(&sh_domain->l1, l1index, 1,
199 sh_domain->l2[l1index].handle | 0x1);
203 static void l2realfree(struct shmobile_iommu_domain_pgtable *l2)
206 pgtable_free(l2, l2cache, L2_SIZE);
209 static void l2free(struct shmobile_iommu_domain *sh_domain,
210 unsigned int l1index,
211 struct shmobile_iommu_domain_pgtable *l2)
213 pgtable_write(&sh_domain->l1, l1index, 1, 0);
214 if (sh_domain->l2[l1index].pgtable) {
215 *l2 = sh_domain->l2[l1index];
216 sh_domain->l2[l1index].pgtable = NULL;
220 static int shmobile_iommu_map(struct iommu_domain *domain, unsigned long iova,
221 phys_addr_t paddr, size_t size, int prot)
223 struct shmobile_iommu_domain_pgtable l2 = { .pgtable = NULL };
224 struct shmobile_iommu_domain *sh_domain = to_sh_domain(domain);
225 unsigned int l1index, l2index;
228 l1index = iova >> 20;
231 l2index = (iova >> 12) & 0xff;
232 spin_lock(&sh_domain->map_lock);
233 ret = l2alloc(sh_domain, l1index);
235 pgtable_write(&sh_domain->l2[l1index], l2index, 1,
237 spin_unlock(&sh_domain->map_lock);
240 l2index = (iova >> 12) & 0xf0;
241 spin_lock(&sh_domain->map_lock);
242 ret = l2alloc(sh_domain, l1index);
244 pgtable_write(&sh_domain->l2[l1index], l2index, 0x10,
246 spin_unlock(&sh_domain->map_lock);
249 spin_lock(&sh_domain->map_lock);
250 l2free(sh_domain, l1index, &l2);
251 pgtable_write(&sh_domain->l1, l1index, 1, paddr | 0xc02);
252 spin_unlock(&sh_domain->map_lock);
259 domain_tlb_flush(sh_domain);
264 static size_t shmobile_iommu_unmap(struct iommu_domain *domain,
265 unsigned long iova, size_t size)
267 struct shmobile_iommu_domain_pgtable l2 = { .pgtable = NULL };
268 struct shmobile_iommu_domain *sh_domain = to_sh_domain(domain);
269 unsigned int l1index, l2index;
270 uint32_t l2entry = 0;
273 l1index = iova >> 20;
274 if (!(iova & 0xfffff) && size >= SZ_1M) {
275 spin_lock(&sh_domain->map_lock);
276 l2free(sh_domain, l1index, &l2);
277 spin_unlock(&sh_domain->map_lock);
281 l2index = (iova >> 12) & 0xff;
282 spin_lock(&sh_domain->map_lock);
283 if (sh_domain->l2[l1index].pgtable)
284 l2entry = pgtable_read(&sh_domain->l2[l1index], l2index);
285 switch (l2entry & 3) {
289 pgtable_write(&sh_domain->l2[l1index], l2index, 0x10, 0);
293 pgtable_write(&sh_domain->l2[l1index], l2index, 1, 0);
297 spin_unlock(&sh_domain->map_lock);
300 domain_tlb_flush(sh_domain);
305 static phys_addr_t shmobile_iommu_iova_to_phys(struct iommu_domain *domain,
308 struct shmobile_iommu_domain *sh_domain = to_sh_domain(domain);
309 uint32_t l1entry = 0, l2entry = 0;
310 unsigned int l1index, l2index;
312 l1index = iova >> 20;
313 l2index = (iova >> 12) & 0xff;
314 spin_lock(&sh_domain->map_lock);
315 if (sh_domain->l2[l1index].pgtable)
316 l2entry = pgtable_read(&sh_domain->l2[l1index], l2index);
318 l1entry = pgtable_read(&sh_domain->l1, l1index);
319 spin_unlock(&sh_domain->map_lock);
320 switch (l2entry & 3) {
322 return (l2entry & ~0xffff) | (iova & 0xffff);
324 return (l2entry & ~0xfff) | (iova & 0xfff);
326 if ((l1entry & 3) == 2)
327 return (l1entry & ~0xfffff) | (iova & 0xfffff);
332 static int find_dev_name(struct shmobile_ipmmu *ipmmu, const char *dev_name)
334 unsigned int i, n = ipmmu->num_dev_names;
336 for (i = 0; i < n; i++) {
337 if (strcmp(ipmmu->dev_names[i], dev_name) == 0)
343 static int shmobile_iommu_add_device(struct device *dev)
345 struct shmobile_iommu_archdata *archdata = ipmmu_archdata;
346 struct dma_iommu_mapping *mapping;
348 if (!find_dev_name(archdata->ipmmu, dev_name(dev)))
350 mapping = archdata->iommu_mapping;
352 mapping = arm_iommu_create_mapping(&platform_bus_type, 0,
355 return PTR_ERR(mapping);
356 archdata->iommu_mapping = mapping;
358 dev->archdata.iommu = archdata;
359 if (arm_iommu_attach_device(dev, mapping))
360 pr_err("arm_iommu_attach_device failed\n");
364 static const struct iommu_ops shmobile_iommu_ops = {
365 .domain_alloc = shmobile_iommu_domain_alloc,
366 .domain_free = shmobile_iommu_domain_free,
367 .attach_dev = shmobile_iommu_attach_device,
368 .detach_dev = shmobile_iommu_detach_device,
369 .map = shmobile_iommu_map,
370 .unmap = shmobile_iommu_unmap,
371 .map_sg = default_iommu_map_sg,
372 .iova_to_phys = shmobile_iommu_iova_to_phys,
373 .add_device = shmobile_iommu_add_device,
374 .pgsize_bitmap = SZ_1M | SZ_64K | SZ_4K,
377 int ipmmu_iommu_init(struct shmobile_ipmmu *ipmmu)
379 static struct shmobile_iommu_archdata *archdata;
381 l1cache = kmem_cache_create("shmobile-iommu-pgtable1", L1_SIZE,
382 L1_ALIGN, SLAB_HWCACHE_ALIGN, NULL);
385 l2cache = kmem_cache_create("shmobile-iommu-pgtable2", L2_SIZE,
386 L2_ALIGN, SLAB_HWCACHE_ALIGN, NULL);
388 kmem_cache_destroy(l1cache);
391 archdata = kzalloc(sizeof(*archdata), GFP_KERNEL);
393 kmem_cache_destroy(l1cache);
394 kmem_cache_destroy(l2cache);
397 spin_lock_init(&archdata->attach_lock);
398 archdata->ipmmu = ipmmu;
399 ipmmu_archdata = archdata;
400 bus_set_iommu(&platform_bus_type, &shmobile_iommu_ops);