xref: /freebsd/sys/x86/iommu/intel_dmar.h (revision 7b8dd078ea3c6b836f514f47587672fd062279c8)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2013-2015 The FreeBSD Foundation
5  *
6  * This software was developed by Konstantin Belousov <kib@FreeBSD.org>
7  * under sponsorship from the FreeBSD Foundation.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  */
30 
31 #ifndef __X86_IOMMU_INTEL_DMAR_H
32 #define	__X86_IOMMU_INTEL_DMAR_H
33 
34 #include <dev/iommu/iommu.h>
35 
36 struct dmar_unit;
37 
38 /*
39  * Locking annotations:
40  * (u) - Protected by iommu unit lock
41  * (d) - Protected by domain lock
42  * (c) - Immutable after initialization
43  */
44 
45 /*
46  * The domain abstraction.  Most non-constant members of the domain
47  * are protected by owning dmar unit lock, not by the domain lock.
48  * Most important, the dmar lock protects the contexts list.
49  *
50  * The domain lock protects the address map for the domain, and list
51  * of unload entries delayed.
52  *
53  * Page tables pages and pages content is protected by the vm object
54  * lock pgtbl_obj, which contains the page tables pages.
55  */
56 struct dmar_domain {
57 	struct iommu_domain iodom;
58 	int domain;			/* (c) DID, written in context entry */
59 	int mgaw;			/* (c) Real max address width */
60 	int agaw;			/* (c) Adjusted guest address width */
61 	int pglvl;			/* (c) The pagelevel */
62 	int awlvl;			/* (c) The pagelevel as the bitmask,
63 					   to set in context entry */
64 	u_int ctx_cnt;			/* (u) Number of contexts owned */
65 	u_int refs;			/* (u) Refs, including ctx */
66 	struct dmar_unit *dmar;		/* (c) */
67 	LIST_ENTRY(dmar_domain) link;	/* (u) Member in the dmar list */
68 	vm_object_t pgtbl_obj;		/* (c) Page table pages */
69 	u_int batch_no;
70 };
71 
72 struct dmar_ctx {
73 	struct iommu_ctx context;
74 	uint64_t last_fault_rec[2];	/* Last fault reported */
75 };
76 
77 #define	DMAR_DOMAIN_PGLOCK(dom)		VM_OBJECT_WLOCK((dom)->pgtbl_obj)
78 #define	DMAR_DOMAIN_PGTRYLOCK(dom)	VM_OBJECT_TRYWLOCK((dom)->pgtbl_obj)
79 #define	DMAR_DOMAIN_PGUNLOCK(dom)	VM_OBJECT_WUNLOCK((dom)->pgtbl_obj)
80 #define	DMAR_DOMAIN_ASSERT_PGLOCKED(dom) \
81 	VM_OBJECT_ASSERT_WLOCKED((dom)->pgtbl_obj)
82 
83 #define	DMAR_DOMAIN_LOCK(dom)	mtx_lock(&(dom)->iodom.lock)
84 #define	DMAR_DOMAIN_UNLOCK(dom)	mtx_unlock(&(dom)->iodom.lock)
85 #define	DMAR_DOMAIN_ASSERT_LOCKED(dom) mtx_assert(&(dom)->iodom.lock, MA_OWNED)
86 
87 #define	DMAR2IOMMU(dmar)	(&((dmar)->iommu))
88 #define	IOMMU2DMAR(dmar)	\
89 	__containerof((dmar), struct dmar_unit, iommu)
90 
91 #define	DOM2IODOM(domain)	(&((domain)->iodom))
92 #define	IODOM2DOM(domain)	\
93 	__containerof((domain), struct dmar_domain, iodom)
94 
95 #define	CTX2IOCTX(ctx)		(&((ctx)->context))
96 #define	IOCTX2CTX(ctx)		\
97 	__containerof((ctx), struct dmar_ctx, context)
98 
99 #define	CTX2DOM(ctx)		IODOM2DOM((ctx)->context.domain)
100 #define	CTX2DMAR(ctx)		(CTX2DOM(ctx)->dmar)
101 #define	DOM2DMAR(domain)	((domain)->dmar)
102 
103 #define	DMAR_INTR_FAULT		0
104 #define	DMAR_INTR_QI		1
105 #define	DMAR_INTR_TOTAL		2
106 
107 struct dmar_unit {
108 	struct iommu_unit iommu;
109 	struct x86_unit_common x86c;
110 	uint16_t segment;
111 	uint64_t base;
112 	int memdomain;
113 
114 	/* Resources */
115 	int reg_rid;
116 	struct resource *regs;
117 
118 	/* Hardware registers cache */
119 	uint32_t hw_ver;
120 	uint64_t hw_cap;
121 	uint64_t hw_ecap;
122 	uint32_t hw_gcmd;
123 
124 	/* Data for being a dmar */
125 	LIST_HEAD(, dmar_domain) domains;
126 	struct unrhdr *domids;
127 	vm_object_t ctx_obj;
128 	u_int barrier_flags;
129 
130 	/* Fault handler data */
131 	struct mtx fault_lock;
132 	uint64_t *fault_log;
133 	int fault_log_head;
134 	int fault_log_tail;
135 	int fault_log_size;
136 	struct task fault_task;
137 	struct taskqueue *fault_taskqueue;
138 
139 	/* QI */
140 	int qi_enabled;
141 
142 	/* IR */
143 	int ir_enabled;
144 	vm_paddr_t irt_phys;
145 	dmar_irte_t *irt;
146 	u_int irte_cnt;
147 	vmem_t *irtids;
148 };
149 
150 #define	DMAR_LOCK(dmar)		mtx_lock(&DMAR2IOMMU(dmar)->lock)
151 #define	DMAR_UNLOCK(dmar)	mtx_unlock(&DMAR2IOMMU(dmar)->lock)
152 #define	DMAR_ASSERT_LOCKED(dmar) mtx_assert(&DMAR2IOMMU(dmar)->lock, MA_OWNED)
153 
154 #define	DMAR_FAULT_LOCK(dmar)	mtx_lock_spin(&(dmar)->fault_lock)
155 #define	DMAR_FAULT_UNLOCK(dmar)	mtx_unlock_spin(&(dmar)->fault_lock)
156 #define	DMAR_FAULT_ASSERT_LOCKED(dmar) mtx_assert(&(dmar)->fault_lock, MA_OWNED)
157 
158 #define	DMAR_IS_COHERENT(dmar)	(((dmar)->hw_ecap & DMAR_ECAP_C) != 0)
159 #define	DMAR_HAS_QI(dmar)	(((dmar)->hw_ecap & DMAR_ECAP_QI) != 0)
160 #define	DMAR_X2APIC(dmar) \
161 	(x2apic_mode && ((dmar)->hw_ecap & DMAR_ECAP_EIM) != 0)
162 
163 /* Barrier ids */
164 #define	DMAR_BARRIER_RMRR	0
165 #define	DMAR_BARRIER_USEQ	1
166 
167 SYSCTL_DECL(_hw_iommu_dmar);
168 
169 struct dmar_unit *dmar_find(device_t dev, bool verbose);
170 struct dmar_unit *dmar_find_hpet(device_t dev, uint16_t *rid);
171 struct dmar_unit *dmar_find_ioapic(u_int apic_id, uint16_t *rid);
172 
173 u_int dmar_nd2mask(u_int nd);
174 bool dmar_pglvl_supported(struct dmar_unit *unit, int pglvl);
175 int domain_set_agaw(struct dmar_domain *domain, int mgaw);
176 int dmar_maxaddr2mgaw(struct dmar_unit *unit, iommu_gaddr_t maxaddr,
177     bool allow_less);
178 int domain_is_sp_lvl(struct dmar_domain *domain, int lvl);
179 iommu_gaddr_t domain_page_size(struct dmar_domain *domain, int lvl);
180 int calc_am(struct dmar_unit *unit, iommu_gaddr_t base, iommu_gaddr_t size,
181     iommu_gaddr_t *isizep);
182 int dmar_load_root_entry_ptr(struct dmar_unit *unit);
183 int dmar_inv_ctx_glob(struct dmar_unit *unit);
184 int dmar_inv_iotlb_glob(struct dmar_unit *unit);
185 int dmar_flush_write_bufs(struct dmar_unit *unit);
186 void dmar_flush_pte_to_ram(struct dmar_unit *unit, iommu_pte_t *dst);
187 void dmar_flush_ctx_to_ram(struct dmar_unit *unit, dmar_ctx_entry_t *dst);
188 void dmar_flush_root_to_ram(struct dmar_unit *unit, dmar_root_entry_t *dst);
189 int dmar_disable_protected_regions(struct dmar_unit *unit);
190 int dmar_enable_translation(struct dmar_unit *unit);
191 int dmar_disable_translation(struct dmar_unit *unit);
192 int dmar_load_irt_ptr(struct dmar_unit *unit);
193 int dmar_enable_ir(struct dmar_unit *unit);
194 int dmar_disable_ir(struct dmar_unit *unit);
195 bool dmar_barrier_enter(struct dmar_unit *dmar, u_int barrier_id);
196 void dmar_barrier_exit(struct dmar_unit *dmar, u_int barrier_id);
197 uint64_t dmar_get_timeout(void);
198 void dmar_update_timeout(uint64_t newval);
199 
200 int dmar_fault_intr(void *arg);
201 void dmar_enable_fault_intr(struct iommu_unit *unit);
202 void dmar_disable_fault_intr(struct iommu_unit *unit);
203 int dmar_init_fault_log(struct dmar_unit *unit);
204 void dmar_fini_fault_log(struct dmar_unit *unit);
205 
206 int dmar_qi_intr(void *arg);
207 void dmar_enable_qi_intr(struct iommu_unit *unit);
208 void dmar_disable_qi_intr(struct iommu_unit *unit);
209 int dmar_init_qi(struct dmar_unit *unit);
210 void dmar_fini_qi(struct dmar_unit *unit);
211 void dmar_qi_invalidate_locked(struct dmar_domain *domain,
212     struct iommu_map_entry *entry, bool emit_wait);
213 void dmar_qi_invalidate_sync(struct dmar_domain *domain, iommu_gaddr_t start,
214     iommu_gaddr_t size, bool cansleep);
215 void dmar_qi_invalidate_ctx_glob_locked(struct dmar_unit *unit);
216 void dmar_qi_invalidate_iotlb_glob_locked(struct dmar_unit *unit);
217 void dmar_qi_invalidate_iec_glob(struct dmar_unit *unit);
218 void dmar_qi_invalidate_iec(struct dmar_unit *unit, u_int start, u_int cnt);
219 
220 vm_object_t dmar_get_idmap_pgtbl(struct dmar_domain *domain,
221     iommu_gaddr_t maxaddr);
222 void dmar_put_idmap_pgtbl(vm_object_t obj);
223 void dmar_flush_iotlb_sync(struct dmar_domain *domain, iommu_gaddr_t base,
224     iommu_gaddr_t size);
225 int dmar_domain_alloc_pgtbl(struct dmar_domain *domain);
226 void dmar_domain_free_pgtbl(struct dmar_domain *domain);
227 extern const struct iommu_domain_map_ops dmar_domain_map_ops;
228 
229 int dmar_dev_depth(device_t child);
230 void dmar_dev_path(device_t child, int *busno, void *path1, int depth);
231 
232 struct dmar_ctx *dmar_get_ctx_for_dev(struct dmar_unit *dmar, device_t dev,
233     uint16_t rid, bool id_mapped, bool rmrr_init);
234 struct dmar_ctx *dmar_get_ctx_for_devpath(struct dmar_unit *dmar, uint16_t rid,
235     int dev_domain, int dev_busno, const void *dev_path, int dev_path_len,
236     bool id_mapped, bool rmrr_init);
237 int dmar_move_ctx_to_domain(struct dmar_domain *domain, struct dmar_ctx *ctx);
238 void dmar_free_ctx_locked_method(struct iommu_unit *dmar,
239     struct iommu_ctx *ctx);
240 void dmar_free_ctx_method(struct iommu_ctx *ctx);
241 struct dmar_ctx *dmar_find_ctx_locked(struct dmar_unit *dmar, uint16_t rid);
242 struct iommu_ctx *dmar_get_ctx(struct iommu_unit *iommu, device_t dev,
243     uint16_t rid, bool id_mapped, bool rmrr_init);
244 void dmar_domain_unload_entry(struct iommu_map_entry *entry, bool free,
245     bool cansleep);
246 void dmar_domain_unload(struct iommu_domain *iodom,
247     struct iommu_map_entries_tailq *entries, bool cansleep);
248 
249 void dmar_dev_parse_rmrr(struct dmar_domain *domain, int dev_domain,
250     int dev_busno, const void *dev_path, int dev_path_len,
251     struct iommu_map_entries_tailq *rmrr_entries);
252 int dmar_instantiate_rmrr_ctxs(struct iommu_unit *dmar);
253 
254 void dmar_quirks_post_ident(struct dmar_unit *dmar);
255 void dmar_quirks_pre_use(struct iommu_unit *dmar);
256 
257 int dmar_init_irt(struct dmar_unit *unit);
258 void dmar_fini_irt(struct dmar_unit *unit);
259 int dmar_alloc_msi_intr(device_t src, u_int *cookies, u_int count);
260 int dmar_map_msi_intr(device_t src, u_int cpu, u_int vector, u_int cookie,
261     uint64_t *addr, uint32_t *data);
262 int dmar_unmap_msi_intr(device_t src, u_int cookie);
263 int dmar_map_ioapic_intr(u_int ioapic_id, u_int cpu, u_int vector, bool edge,
264     bool activehi, int irq, u_int *cookie, uint32_t *hi, uint32_t *lo);
265 int dmar_unmap_ioapic_intr(u_int ioapic_id, u_int *cookie);
266 
267 extern int haw;
268 extern int dmar_rmrr_enable;
269 
270 static inline uint32_t
271 dmar_read4(const struct dmar_unit *unit, int reg)
272 {
273 
274 	return (bus_read_4(unit->regs, reg));
275 }
276 
277 static inline uint64_t
278 dmar_read8(const struct dmar_unit *unit, int reg)
279 {
280 #ifdef __i386__
281 	uint32_t high, low;
282 
283 	low = bus_read_4(unit->regs, reg);
284 	high = bus_read_4(unit->regs, reg + 4);
285 	return (low | ((uint64_t)high << 32));
286 #else
287 	return (bus_read_8(unit->regs, reg));
288 #endif
289 }
290 
291 static inline void
292 dmar_write4(const struct dmar_unit *unit, int reg, uint32_t val)
293 {
294 
295 	KASSERT(reg != DMAR_GCMD_REG || (val & DMAR_GCMD_TE) ==
296 	    (unit->hw_gcmd & DMAR_GCMD_TE),
297 	    ("dmar%d clearing TE 0x%08x 0x%08x", unit->iommu.unit,
298 	    unit->hw_gcmd, val));
299 	bus_write_4(unit->regs, reg, val);
300 }
301 
302 static inline void
303 dmar_write8(const struct dmar_unit *unit, int reg, uint64_t val)
304 {
305 
306 	KASSERT(reg != DMAR_GCMD_REG, ("8byte GCMD write"));
307 #ifdef __i386__
308 	uint32_t high, low;
309 
310 	low = val;
311 	high = val >> 32;
312 	bus_write_4(unit->regs, reg, low);
313 	bus_write_4(unit->regs, reg + 4, high);
314 #else
315 	bus_write_8(unit->regs, reg, val);
316 #endif
317 }
318 
319 /*
320  * dmar_pte_store and dmar_pte_clear ensure that on i386, 32bit writes
321  * are issued in the correct order.  For store, the lower word,
322  * containing the P or R and W bits, is set only after the high word
323  * is written.  For clear, the P bit is cleared first, then the high
324  * word is cleared.
325  *
326  * dmar_pte_update updates the pte.  For amd64, the update is atomic.
327  * For i386, it first disables the entry by clearing the word
328  * containing the P bit, and then defer to dmar_pte_store.  The locked
329  * cmpxchg8b is probably available on any machine having DMAR support,
330  * but interrupt translation table may be mapped uncached.
331  */
332 static inline void
333 dmar_pte_store1(volatile uint64_t *dst, uint64_t val)
334 {
335 #ifdef __i386__
336 	volatile uint32_t *p;
337 	uint32_t hi, lo;
338 
339 	hi = val >> 32;
340 	lo = val;
341 	p = (volatile uint32_t *)dst;
342 	*(p + 1) = hi;
343 	*p = lo;
344 #else
345 	*dst = val;
346 #endif
347 }
348 
349 static inline void
350 dmar_pte_store(volatile uint64_t *dst, uint64_t val)
351 {
352 
353 	KASSERT(*dst == 0, ("used pte %p oldval %jx newval %jx",
354 	    dst, (uintmax_t)*dst, (uintmax_t)val));
355 	dmar_pte_store1(dst, val);
356 }
357 
358 static inline void
359 dmar_pte_update(volatile uint64_t *dst, uint64_t val)
360 {
361 
362 #ifdef __i386__
363 	volatile uint32_t *p;
364 
365 	p = (volatile uint32_t *)dst;
366 	*p = 0;
367 #endif
368 	dmar_pte_store1(dst, val);
369 }
370 
371 static inline void
372 dmar_pte_clear(volatile uint64_t *dst)
373 {
374 #ifdef __i386__
375 	volatile uint32_t *p;
376 
377 	p = (volatile uint32_t *)dst;
378 	*p = 0;
379 	*(p + 1) = 0;
380 #else
381 	*dst = 0;
382 #endif
383 }
384 
385 extern struct timespec dmar_hw_timeout;
386 
387 #define	DMAR_WAIT_UNTIL(cond)					\
388 {								\
389 	struct timespec last, curr;				\
390 	bool forever;						\
391 								\
392 	if (dmar_hw_timeout.tv_sec == 0 &&			\
393 	    dmar_hw_timeout.tv_nsec == 0) {			\
394 		forever = true;					\
395 	} else {						\
396 		forever = false;				\
397 		nanouptime(&curr);				\
398 		timespecadd(&curr, &dmar_hw_timeout, &last);	\
399 	}							\
400 	for (;;) {						\
401 		if (cond) {					\
402 			error = 0;				\
403 			break;					\
404 		}						\
405 		nanouptime(&curr);				\
406 		if (!forever && timespeccmp(&last, &curr, <)) {	\
407 			error = ETIMEDOUT;			\
408 			break;					\
409 		}						\
410 		cpu_spinwait();					\
411 	}							\
412 }
413 
414 #ifdef INVARIANTS
415 #define	TD_PREP_PINNED_ASSERT						\
416 	int old_td_pinned;						\
417 	old_td_pinned = curthread->td_pinned
418 #define	TD_PINNED_ASSERT						\
419 	KASSERT(curthread->td_pinned == old_td_pinned,			\
420 	    ("pin count leak: %d %d %s:%d", curthread->td_pinned,	\
421 	    old_td_pinned, __FILE__, __LINE__))
422 #else
423 #define	TD_PREP_PINNED_ASSERT
424 #define	TD_PINNED_ASSERT
425 #endif
426 
427 #endif
428