1 /*
2  * Copyright (C) 2007-2010 Advanced Micro Devices, Inc.
3  * Author: Joerg Roedel <joerg.roedel@amd.com>
4  *         Leo Duran <leo.duran@amd.com>
5  *
6  * This program is free software; you can redistribute it and/or modify it
7  * under the terms of the GNU General Public License version 2 as published
8  * by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License
16  * along with this program; if not, write to the Free Software
17  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
18  */
19 
20 #ifndef _ASM_X86_AMD_IOMMU_TYPES_H
21 #define _ASM_X86_AMD_IOMMU_TYPES_H
22 
23 #include <linux/types.h>
24 #include <linux/mutex.h>
25 #include <linux/list.h>
26 #include <linux/spinlock.h>
27 
28 /*
29  * Maximum number of IOMMUs supported
30  */
31 #define MAX_IOMMUS	32
32 
33 /*
34  * some size calculation constants
35  */
36 #define DEV_TABLE_ENTRY_SIZE		32
37 #define ALIAS_TABLE_ENTRY_SIZE		2
38 #define RLOOKUP_TABLE_ENTRY_SIZE	(sizeof(void *))
39 
40 /* Length of the MMIO region for the AMD IOMMU */
41 #define MMIO_REGION_LENGTH       0x4000
42 
43 /* Capability offsets used by the driver */
44 #define MMIO_CAP_HDR_OFFSET	0x00
45 #define MMIO_RANGE_OFFSET	0x0c
46 #define MMIO_MISC_OFFSET	0x10
47 
48 /* Masks, shifts and macros to parse the device range capability */
49 #define MMIO_RANGE_LD_MASK	0xff000000
50 #define MMIO_RANGE_FD_MASK	0x00ff0000
51 #define MMIO_RANGE_BUS_MASK	0x0000ff00
52 #define MMIO_RANGE_LD_SHIFT	24
53 #define MMIO_RANGE_FD_SHIFT	16
54 #define MMIO_RANGE_BUS_SHIFT	8
55 #define MMIO_GET_LD(x)  (((x) & MMIO_RANGE_LD_MASK) >> MMIO_RANGE_LD_SHIFT)
56 #define MMIO_GET_FD(x)  (((x) & MMIO_RANGE_FD_MASK) >> MMIO_RANGE_FD_SHIFT)
57 #define MMIO_GET_BUS(x) (((x) & MMIO_RANGE_BUS_MASK) >> MMIO_RANGE_BUS_SHIFT)
58 #define MMIO_MSI_NUM(x)	((x) & 0x1f)
59 
60 /* Flag masks for the AMD IOMMU exclusion range */
61 #define MMIO_EXCL_ENABLE_MASK 0x01ULL
62 #define MMIO_EXCL_ALLOW_MASK  0x02ULL
63 
64 /* Used offsets into the MMIO space */
65 #define MMIO_DEV_TABLE_OFFSET   0x0000
66 #define MMIO_CMD_BUF_OFFSET     0x0008
67 #define MMIO_EVT_BUF_OFFSET     0x0010
68 #define MMIO_CONTROL_OFFSET     0x0018
69 #define MMIO_EXCL_BASE_OFFSET   0x0020
70 #define MMIO_EXCL_LIMIT_OFFSET  0x0028
71 #define MMIO_CMD_HEAD_OFFSET	0x2000
72 #define MMIO_CMD_TAIL_OFFSET	0x2008
73 #define MMIO_EVT_HEAD_OFFSET	0x2010
74 #define MMIO_EVT_TAIL_OFFSET	0x2018
75 #define MMIO_STATUS_OFFSET	0x2020
76 
77 /* MMIO status bits */
78 #define MMIO_STATUS_COM_WAIT_INT_MASK	0x04
79 
80 /* event logging constants */
81 #define EVENT_ENTRY_SIZE	0x10
82 #define EVENT_TYPE_SHIFT	28
83 #define EVENT_TYPE_MASK		0xf
84 #define EVENT_TYPE_ILL_DEV	0x1
85 #define EVENT_TYPE_IO_FAULT	0x2
86 #define EVENT_TYPE_DEV_TAB_ERR	0x3
87 #define EVENT_TYPE_PAGE_TAB_ERR	0x4
88 #define EVENT_TYPE_ILL_CMD	0x5
89 #define EVENT_TYPE_CMD_HARD_ERR	0x6
90 #define EVENT_TYPE_IOTLB_INV_TO	0x7
91 #define EVENT_TYPE_INV_DEV_REQ	0x8
92 #define EVENT_DEVID_MASK	0xffff
93 #define EVENT_DEVID_SHIFT	0
94 #define EVENT_DOMID_MASK	0xffff
95 #define EVENT_DOMID_SHIFT	0
96 #define EVENT_FLAGS_MASK	0xfff
97 #define EVENT_FLAGS_SHIFT	0x10
98 
99 /* feature control bits */
100 #define CONTROL_IOMMU_EN        0x00ULL
101 #define CONTROL_HT_TUN_EN       0x01ULL
102 #define CONTROL_EVT_LOG_EN      0x02ULL
103 #define CONTROL_EVT_INT_EN      0x03ULL
104 #define CONTROL_COMWAIT_EN      0x04ULL
105 #define CONTROL_PASSPW_EN       0x08ULL
106 #define CONTROL_RESPASSPW_EN    0x09ULL
107 #define CONTROL_COHERENT_EN     0x0aULL
108 #define CONTROL_ISOC_EN         0x0bULL
109 #define CONTROL_CMDBUF_EN       0x0cULL
110 #define CONTROL_PPFLOG_EN       0x0dULL
111 #define CONTROL_PPFINT_EN       0x0eULL
112 
113 /* command specific defines */
114 #define CMD_COMPL_WAIT          0x01
115 #define CMD_INV_DEV_ENTRY       0x02
116 #define CMD_INV_IOMMU_PAGES     0x03
117 
118 #define CMD_COMPL_WAIT_STORE_MASK	0x01
119 #define CMD_COMPL_WAIT_INT_MASK		0x02
120 #define CMD_INV_IOMMU_PAGES_SIZE_MASK	0x01
121 #define CMD_INV_IOMMU_PAGES_PDE_MASK	0x02
122 
123 #define CMD_INV_IOMMU_ALL_PAGES_ADDRESS	0x7fffffffffffffffULL
124 
125 /* macros and definitions for device table entries */
126 #define DEV_ENTRY_VALID         0x00
127 #define DEV_ENTRY_TRANSLATION   0x01
128 #define DEV_ENTRY_IR            0x3d
129 #define DEV_ENTRY_IW            0x3e
130 #define DEV_ENTRY_NO_PAGE_FAULT	0x62
131 #define DEV_ENTRY_EX            0x67
132 #define DEV_ENTRY_SYSMGT1       0x68
133 #define DEV_ENTRY_SYSMGT2       0x69
134 #define DEV_ENTRY_INIT_PASS     0xb8
135 #define DEV_ENTRY_EINT_PASS     0xb9
136 #define DEV_ENTRY_NMI_PASS      0xba
137 #define DEV_ENTRY_LINT0_PASS    0xbe
138 #define DEV_ENTRY_LINT1_PASS    0xbf
139 #define DEV_ENTRY_MODE_MASK	0x07
140 #define DEV_ENTRY_MODE_SHIFT	0x09
141 
142 /* constants to configure the command buffer */
143 #define CMD_BUFFER_SIZE    8192
144 #define CMD_BUFFER_UNINITIALIZED 1
145 #define CMD_BUFFER_ENTRIES 512
146 #define MMIO_CMD_SIZE_SHIFT 56
147 #define MMIO_CMD_SIZE_512 (0x9ULL << MMIO_CMD_SIZE_SHIFT)
148 
149 /* constants for event buffer handling */
150 #define EVT_BUFFER_SIZE		8192 /* 512 entries */
151 #define EVT_LEN_MASK		(0x9ULL << 56)
152 
153 #define PAGE_MODE_NONE    0x00
154 #define PAGE_MODE_1_LEVEL 0x01
155 #define PAGE_MODE_2_LEVEL 0x02
156 #define PAGE_MODE_3_LEVEL 0x03
157 #define PAGE_MODE_4_LEVEL 0x04
158 #define PAGE_MODE_5_LEVEL 0x05
159 #define PAGE_MODE_6_LEVEL 0x06
160 
161 #define PM_LEVEL_SHIFT(x)	(12 + ((x) * 9))
162 #define PM_LEVEL_SIZE(x)	(((x) < 6) ? \
163 				  ((1ULL << PM_LEVEL_SHIFT((x))) - 1): \
164 				   (0xffffffffffffffffULL))
165 #define PM_LEVEL_INDEX(x, a)	(((a) >> PM_LEVEL_SHIFT((x))) & 0x1ffULL)
166 #define PM_LEVEL_ENC(x)		(((x) << 9) & 0xe00ULL)
167 #define PM_LEVEL_PDE(x, a)	((a) | PM_LEVEL_ENC((x)) | \
168 				 IOMMU_PTE_P | IOMMU_PTE_IR | IOMMU_PTE_IW)
169 #define PM_PTE_LEVEL(pte)	(((pte) >> 9) & 0x7ULL)
170 
171 #define PM_MAP_4k		0
172 #define PM_ADDR_MASK		0x000ffffffffff000ULL
173 #define PM_MAP_MASK(lvl)	(PM_ADDR_MASK & \
174 				(~((1ULL << (12 + ((lvl) * 9))) - 1)))
175 #define PM_ALIGNED(lvl, addr)	((PM_MAP_MASK(lvl) & (addr)) == (addr))
176 
177 /*
178  * Returns the page table level to use for a given page size
179  * Pagesize is expected to be a power-of-two
180  */
181 #define PAGE_SIZE_LEVEL(pagesize) \
182 		((__ffs(pagesize) - 12) / 9)
183 /*
184  * Returns the number of ptes to use for a given page size
185  * Pagesize is expected to be a power-of-two
186  */
187 #define PAGE_SIZE_PTE_COUNT(pagesize) \
188 		(1ULL << ((__ffs(pagesize) - 12) % 9))
189 
190 /*
191  * Aligns a given io-virtual address to a given page size
192  * Pagesize is expected to be a power-of-two
193  */
194 #define PAGE_SIZE_ALIGN(address, pagesize) \
195 		((address) & ~((pagesize) - 1))
196 /*
197  * Creates an IOMMU PTE for an address an a given pagesize
198  * The PTE has no permission bits set
199  * Pagesize is expected to be a power-of-two larger than 4096
200  */
201 #define PAGE_SIZE_PTE(address, pagesize)		\
202 		(((address) | ((pagesize) - 1)) &	\
203 		 (~(pagesize >> 1)) & PM_ADDR_MASK)
204 
205 /*
206  * Takes a PTE value with mode=0x07 and returns the page size it maps
207  */
208 #define PTE_PAGE_SIZE(pte) \
209 	(1ULL << (1 + ffz(((pte) | 0xfffULL))))
210 
211 #define IOMMU_PTE_P  (1ULL << 0)
212 #define IOMMU_PTE_TV (1ULL << 1)
213 #define IOMMU_PTE_U  (1ULL << 59)
214 #define IOMMU_PTE_FC (1ULL << 60)
215 #define IOMMU_PTE_IR (1ULL << 61)
216 #define IOMMU_PTE_IW (1ULL << 62)
217 
218 #define IOMMU_PAGE_MASK (((1ULL << 52) - 1) & ~0xfffULL)
219 #define IOMMU_PTE_PRESENT(pte) ((pte) & IOMMU_PTE_P)
220 #define IOMMU_PTE_PAGE(pte) (phys_to_virt((pte) & IOMMU_PAGE_MASK))
221 #define IOMMU_PTE_MODE(pte) (((pte) >> 9) & 0x07)
222 
223 #define IOMMU_PROT_MASK 0x03
224 #define IOMMU_PROT_IR 0x01
225 #define IOMMU_PROT_IW 0x02
226 
227 /* IOMMU capabilities */
228 #define IOMMU_CAP_IOTLB   24
229 #define IOMMU_CAP_NPCACHE 26
230 
231 #define MAX_DOMAIN_ID 65536
232 
233 /* FIXME: move this macro to <linux/pci.h> */
234 #define PCI_BUS(x) (((x) >> 8) & 0xff)
235 
236 /* Protection domain flags */
237 #define PD_DMA_OPS_MASK		(1UL << 0) /* domain used for dma_ops */
238 #define PD_DEFAULT_MASK		(1UL << 1) /* domain is a default dma_ops
239 					      domain for an IOMMU */
240 #define PD_PASSTHROUGH_MASK	(1UL << 2) /* domain has no page
241 					      translation */
242 
243 extern bool amd_iommu_dump;
244 #define DUMP_printk(format, arg...)					\
245 	do {								\
246 		if (amd_iommu_dump)						\
247 			printk(KERN_INFO "AMD-Vi: " format, ## arg);	\
248 	} while(0);
249 
250 /* global flag if IOMMUs cache non-present entries */
251 extern bool amd_iommu_np_cache;
252 
253 /*
254  * Make iterating over all IOMMUs easier
255  */
256 #define for_each_iommu(iommu) \
257 	list_for_each_entry((iommu), &amd_iommu_list, list)
258 #define for_each_iommu_safe(iommu, next) \
259 	list_for_each_entry_safe((iommu), (next), &amd_iommu_list, list)
260 
261 #define APERTURE_RANGE_SHIFT	27	/* 128 MB */
262 #define APERTURE_RANGE_SIZE	(1ULL << APERTURE_RANGE_SHIFT)
263 #define APERTURE_RANGE_PAGES	(APERTURE_RANGE_SIZE >> PAGE_SHIFT)
264 #define APERTURE_MAX_RANGES	32	/* allows 4GB of DMA address space */
265 #define APERTURE_RANGE_INDEX(a)	((a) >> APERTURE_RANGE_SHIFT)
266 #define APERTURE_PAGE_INDEX(a)	(((a) >> 21) & 0x3fULL)
267 
268 /*
269  * This structure contains generic data for  IOMMU protection domains
270  * independent of their use.
271  */
272 struct protection_domain {
273 	struct list_head list;  /* for list of all protection domains */
274 	struct list_head dev_list; /* List of all devices in this domain */
275 	spinlock_t lock;	/* mostly used to lock the page table*/
276 	struct mutex api_lock;	/* protect page tables in the iommu-api path */
277 	u16 id;			/* the domain id written to the device table */
278 	int mode;		/* paging mode (0-6 levels) */
279 	u64 *pt_root;		/* page table root pointer */
280 	unsigned long flags;	/* flags to find out type of domain */
281 	bool updated;		/* complete domain flush required */
282 	unsigned dev_cnt;	/* devices assigned to this domain */
283 	unsigned dev_iommu[MAX_IOMMUS]; /* per-IOMMU reference count */
284 	void *priv;		/* private data */
285 
286 };
287 
288 /*
289  * This struct contains device specific data for the IOMMU
290  */
291 struct iommu_dev_data {
292 	struct list_head list;		  /* For domain->dev_list */
293 	struct device *dev;		  /* Device this data belong to */
294 	struct device *alias;		  /* The Alias Device */
295 	struct protection_domain *domain; /* Domain the device is bound to */
296 	atomic_t bind;			  /* Domain attach reverent count */
297 };
298 
299 /*
300  * For dynamic growth the aperture size is split into ranges of 128MB of
301  * DMA address space each. This struct represents one such range.
302  */
303 struct aperture_range {
304 
305 	/* address allocation bitmap */
306 	unsigned long *bitmap;
307 
308 	/*
309 	 * Array of PTE pages for the aperture. In this array we save all the
310 	 * leaf pages of the domain page table used for the aperture. This way
311 	 * we don't need to walk the page table to find a specific PTE. We can
312 	 * just calculate its address in constant time.
313 	 */
314 	u64 *pte_pages[64];
315 
316 	unsigned long offset;
317 };
318 
319 /*
320  * Data container for a dma_ops specific protection domain
321  */
322 struct dma_ops_domain {
323 	struct list_head list;
324 
325 	/* generic protection domain information */
326 	struct protection_domain domain;
327 
328 	/* size of the aperture for the mappings */
329 	unsigned long aperture_size;
330 
331 	/* address we start to search for free addresses */
332 	unsigned long next_address;
333 
334 	/* address space relevant data */
335 	struct aperture_range *aperture[APERTURE_MAX_RANGES];
336 
337 	/* This will be set to true when TLB needs to be flushed */
338 	bool need_flush;
339 
340 	/*
341 	 * if this is a preallocated domain, keep the device for which it was
342 	 * preallocated in this variable
343 	 */
344 	u16 target_dev;
345 };
346 
347 /*
348  * Structure where we save information about one hardware AMD IOMMU in the
349  * system.
350  */
351 struct amd_iommu {
352 	struct list_head list;
353 
354 	/* Index within the IOMMU array */
355 	int index;
356 
357 	/* locks the accesses to the hardware */
358 	spinlock_t lock;
359 
360 	/* Pointer to PCI device of this IOMMU */
361 	struct pci_dev *dev;
362 
363 	/* physical address of MMIO space */
364 	u64 mmio_phys;
365 	/* virtual address of MMIO space */
366 	u8 *mmio_base;
367 
368 	/* capabilities of that IOMMU read from ACPI */
369 	u32 cap;
370 
371 	/* flags read from acpi table */
372 	u8 acpi_flags;
373 
374 	/*
375 	 * Capability pointer. There could be more than one IOMMU per PCI
376 	 * device function if there are more than one AMD IOMMU capability
377 	 * pointers.
378 	 */
379 	u16 cap_ptr;
380 
381 	/* pci domain of this IOMMU */
382 	u16 pci_seg;
383 
384 	/* first device this IOMMU handles. read from PCI */
385 	u16 first_device;
386 	/* last device this IOMMU handles. read from PCI */
387 	u16 last_device;
388 
389 	/* start of exclusion range of that IOMMU */
390 	u64 exclusion_start;
391 	/* length of exclusion range of that IOMMU */
392 	u64 exclusion_length;
393 
394 	/* command buffer virtual address */
395 	u8 *cmd_buf;
396 	/* size of command buffer */
397 	u32 cmd_buf_size;
398 
399 	/* size of event buffer */
400 	u32 evt_buf_size;
401 	/* event buffer virtual address */
402 	u8 *evt_buf;
403 	/* MSI number for event interrupt */
404 	u16 evt_msi_num;
405 
406 	/* true if interrupts for this IOMMU are already enabled */
407 	bool int_enabled;
408 
409 	/* if one, we need to send a completion wait command */
410 	bool need_sync;
411 
412 	/* becomes true if a command buffer reset is running */
413 	bool reset_in_progress;
414 
415 	/* default dma_ops domain for that IOMMU */
416 	struct dma_ops_domain *default_dom;
417 
418 	/*
419 	 * We can't rely on the BIOS to restore all values on reinit, so we
420 	 * need to stash them
421 	 */
422 
423 	/* The iommu BAR */
424 	u32 stored_addr_lo;
425 	u32 stored_addr_hi;
426 
427 	/*
428 	 * Each iommu has 6 l1s, each of which is documented as having 0x12
429 	 * registers
430 	 */
431 	u32 stored_l1[6][0x12];
432 
433 	/* The l2 indirect registers */
434 	u32 stored_l2[0x83];
435 };
436 
437 /*
438  * List with all IOMMUs in the system. This list is not locked because it is
439  * only written and read at driver initialization or suspend time
440  */
441 extern struct list_head amd_iommu_list;
442 
443 /*
444  * Array with pointers to each IOMMU struct
445  * The indices are referenced in the protection domains
446  */
447 extern struct amd_iommu *amd_iommus[MAX_IOMMUS];
448 
449 /* Number of IOMMUs present in the system */
450 extern int amd_iommus_present;
451 
452 /*
453  * Declarations for the global list of all protection domains
454  */
455 extern spinlock_t amd_iommu_pd_lock;
456 extern struct list_head amd_iommu_pd_list;
457 
458 /*
459  * Structure defining one entry in the device table
460  */
461 struct dev_table_entry {
462 	u32 data[8];
463 };
464 
465 /*
466  * One entry for unity mappings parsed out of the ACPI table.
467  */
468 struct unity_map_entry {
469 	struct list_head list;
470 
471 	/* starting device id this entry is used for (including) */
472 	u16 devid_start;
473 	/* end device id this entry is used for (including) */
474 	u16 devid_end;
475 
476 	/* start address to unity map (including) */
477 	u64 address_start;
478 	/* end address to unity map (including) */
479 	u64 address_end;
480 
481 	/* required protection */
482 	int prot;
483 };
484 
485 /*
486  * List of all unity mappings. It is not locked because as runtime it is only
487  * read. It is created at ACPI table parsing time.
488  */
489 extern struct list_head amd_iommu_unity_map;
490 
491 /*
492  * Data structures for device handling
493  */
494 
495 /*
496  * Device table used by hardware. Read and write accesses by software are
497  * locked with the amd_iommu_pd_table lock.
498  */
499 extern struct dev_table_entry *amd_iommu_dev_table;
500 
501 /*
502  * Alias table to find requestor ids to device ids. Not locked because only
503  * read on runtime.
504  */
505 extern u16 *amd_iommu_alias_table;
506 
507 /*
508  * Reverse lookup table to find the IOMMU which translates a specific device.
509  */
510 extern struct amd_iommu **amd_iommu_rlookup_table;
511 
512 /* size of the dma_ops aperture as power of 2 */
513 extern unsigned amd_iommu_aperture_order;
514 
515 /* largest PCI device id we expect translation requests for */
516 extern u16 amd_iommu_last_bdf;
517 
518 /* allocation bitmap for domain ids */
519 extern unsigned long *amd_iommu_pd_alloc_bitmap;
520 
521 /*
522  * If true, the addresses will be flushed on unmap time, not when
523  * they are reused
524  */
525 extern bool amd_iommu_unmap_flush;
526 
527 /* takes bus and device/function and returns the device id
528  * FIXME: should that be in generic PCI code? */
calc_devid(u8 bus,u8 devfn)529 static inline u16 calc_devid(u8 bus, u8 devfn)
530 {
531 	return (((u16)bus) << 8) | devfn;
532 }
533 
534 #ifdef CONFIG_AMD_IOMMU_STATS
535 
536 struct __iommu_counter {
537 	char *name;
538 	struct dentry *dent;
539 	u64 value;
540 };
541 
542 #define DECLARE_STATS_COUNTER(nm) \
543 	static struct __iommu_counter nm = {	\
544 		.name = #nm,			\
545 	}
546 
547 #define INC_STATS_COUNTER(name)		name.value += 1
548 #define ADD_STATS_COUNTER(name, x)	name.value += (x)
549 #define SUB_STATS_COUNTER(name, x)	name.value -= (x)
550 
551 #else /* CONFIG_AMD_IOMMU_STATS */
552 
553 #define DECLARE_STATS_COUNTER(name)
554 #define INC_STATS_COUNTER(name)
555 #define ADD_STATS_COUNTER(name, x)
556 #define SUB_STATS_COUNTER(name, x)
557 
558 #endif /* CONFIG_AMD_IOMMU_STATS */
559 
560 #endif /* _ASM_X86_AMD_IOMMU_TYPES_H */
561