xref: /freebsd/sys/x86/iommu/intel_fault.c (revision 6ef6ba9950260f42b47499d17874d00ca9290955)
1 /*-
2  * Copyright (c) 2013 The FreeBSD Foundation
3  * All rights reserved.
4  *
5  * This software was developed by Konstantin Belousov <kib@FreeBSD.org>
6  * under sponsorship from the FreeBSD Foundation.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32 
33 #include "opt_acpi.h"
34 
35 #include <sys/param.h>
36 #include <sys/bus.h>
37 #include <sys/kernel.h>
38 #include <sys/malloc.h>
39 #include <sys/memdesc.h>
40 #include <sys/module.h>
41 #include <sys/rman.h>
42 #include <sys/taskqueue.h>
43 #include <sys/tree.h>
44 #include <machine/bus.h>
45 #include <contrib/dev/acpica/include/acpi.h>
46 #include <contrib/dev/acpica/include/accommon.h>
47 #include <dev/acpica/acpivar.h>
48 #include <vm/vm.h>
49 #include <vm/vm_extern.h>
50 #include <vm/vm_kern.h>
51 #include <vm/vm_page.h>
52 #include <vm/vm_map.h>
53 #include <x86/include/busdma_impl.h>
54 #include <x86/iommu/intel_reg.h>
55 #include <x86/iommu/busdma_dmar.h>
56 #include <x86/iommu/intel_dmar.h>
57 
58 /*
59  * Fault interrupt handling for DMARs.  If advanced fault logging is
60  * not implemented by hardware, the code emulates it.  Fast interrupt
61  * handler flushes the fault registers into circular buffer at
62  * unit->fault_log, and schedules a task.
63  *
64  * The fast handler is used since faults usually come in bursts, and
65  * number of fault log registers is limited, e.g. down to one for 5400
66  * MCH.  We are trying to reduce the latency for clearing the fault
67  * register file.  The task is usually long-running, since printf() is
68  * slow, but this is not problematic because bursts are rare.
69  *
70  * For the same reason, each translation unit task is executed in its
71  * own thread.
72  *
73  * XXXKIB It seems there is no hardware available which implements
74  * advanced fault logging, so the code to handle AFL is not written.
75  */
76 
77 static int
78 dmar_fault_next(struct dmar_unit *unit, int faultp)
79 {
80 
81 	faultp += 2;
82 	if (faultp == unit->fault_log_size)
83 		faultp = 0;
84 	return (faultp);
85 }
86 
87 static void
88 dmar_fault_intr_clear(struct dmar_unit *unit, uint32_t fsts)
89 {
90 	uint32_t clear;
91 
92 	clear = 0;
93 	if ((fsts & DMAR_FSTS_ITE) != 0) {
94 		printf("DMAR%d: Invalidation timed out\n", unit->unit);
95 		clear |= DMAR_FSTS_ITE;
96 	}
97 	if ((fsts & DMAR_FSTS_ICE) != 0) {
98 		printf("DMAR%d: Invalidation completion error\n",
99 		    unit->unit);
100 		clear |= DMAR_FSTS_ICE;
101 	}
102 	if ((fsts & DMAR_FSTS_IQE) != 0) {
103 		printf("DMAR%d: Invalidation queue error\n",
104 		    unit->unit);
105 		clear |= DMAR_FSTS_IQE;
106 	}
107 	if ((fsts & DMAR_FSTS_APF) != 0) {
108 		printf("DMAR%d: Advanced pending fault\n", unit->unit);
109 		clear |= DMAR_FSTS_APF;
110 	}
111 	if ((fsts & DMAR_FSTS_AFO) != 0) {
112 		printf("DMAR%d: Advanced fault overflow\n", unit->unit);
113 		clear |= DMAR_FSTS_AFO;
114 	}
115 	if (clear != 0)
116 		dmar_write4(unit, DMAR_FSTS_REG, clear);
117 }
118 
119 int
120 dmar_fault_intr(void *arg)
121 {
122 	struct dmar_unit *unit;
123 	uint64_t fault_rec[2];
124 	uint32_t fsts;
125 	int fri, frir, faultp;
126 	bool enqueue;
127 
128 	unit = arg;
129 	enqueue = false;
130 	fsts = dmar_read4(unit, DMAR_FSTS_REG);
131 	dmar_fault_intr_clear(unit, fsts);
132 
133 	if ((fsts & DMAR_FSTS_PPF) == 0)
134 		goto done;
135 
136 	fri = DMAR_FSTS_FRI(fsts);
137 	for (;;) {
138 		frir = (DMAR_CAP_FRO(unit->hw_cap) + fri) * 16;
139 		fault_rec[1] = dmar_read8(unit, frir + 8);
140 		if ((fault_rec[1] & DMAR_FRCD2_F) == 0)
141 			break;
142 		fault_rec[0] = dmar_read8(unit, frir);
143 		dmar_write4(unit, frir + 12, DMAR_FRCD2_F32);
144 		DMAR_FAULT_LOCK(unit);
145 		faultp = unit->fault_log_head;
146 		if (dmar_fault_next(unit, faultp) == unit->fault_log_tail) {
147 			/* XXXKIB log overflow */
148 		} else {
149 			unit->fault_log[faultp] = fault_rec[0];
150 			unit->fault_log[faultp + 1] = fault_rec[1];
151 			unit->fault_log_head = dmar_fault_next(unit, faultp);
152 			enqueue = true;
153 		}
154 		DMAR_FAULT_UNLOCK(unit);
155 		fri += 1;
156 		if (fri >= DMAR_CAP_NFR(unit->hw_cap))
157 			fri = 0;
158 	}
159 
160 done:
161 	/*
162 	 * On SandyBridge, due to errata BJ124, IvyBridge errata
163 	 * BV100, and Haswell errata HSD40, "Spurious Intel VT-d
164 	 * Interrupts May Occur When the PFO Bit is Set".  Handle the
165 	 * cases by clearing overflow bit even if no fault is
166 	 * reported.
167 	 *
168 	 * On IvyBridge, errata BV30 states that clearing clear
169 	 * DMAR_FRCD2_F bit in the fault register causes spurious
170 	 * interrupt.  Do nothing.
171 	 *
172 	 */
173 	if ((fsts & DMAR_FSTS_PFO) != 0) {
174 		printf("DMAR%d: Fault Overflow\n", unit->unit);
175 		dmar_write4(unit, DMAR_FSTS_REG, DMAR_FSTS_PFO);
176 	}
177 
178 	if (enqueue) {
179 		taskqueue_enqueue_fast(unit->fault_taskqueue,
180 		    &unit->fault_task);
181 	}
182 	return (FILTER_HANDLED);
183 }
184 
185 static void
186 dmar_fault_task(void *arg, int pending __unused)
187 {
188 	struct dmar_unit *unit;
189 	struct dmar_ctx *ctx;
190 	uint64_t fault_rec[2];
191 	int sid, bus, slot, func, faultp;
192 
193 	unit = arg;
194 	DMAR_FAULT_LOCK(unit);
195 	for (;;) {
196 		faultp = unit->fault_log_tail;
197 		if (faultp == unit->fault_log_head)
198 			break;
199 
200 		fault_rec[0] = unit->fault_log[faultp];
201 		fault_rec[1] = unit->fault_log[faultp + 1];
202 		unit->fault_log_tail = dmar_fault_next(unit, faultp);
203 		DMAR_FAULT_UNLOCK(unit);
204 
205 		sid = DMAR_FRCD2_SID(fault_rec[1]);
206 		bus = (sid >> 8) & 0xf;
207 		slot = (sid >> 3) & 0x1f;
208 		func = sid & 0x7;
209 		printf("DMAR%d: ", unit->unit);
210 		DMAR_LOCK(unit);
211 		ctx = dmar_find_ctx_locked(unit, bus, slot, func);
212 		if (ctx == NULL) {
213 			printf("<unknown dev>:");
214 		} else {
215 			ctx->flags |= DMAR_CTX_FAULTED;
216 			ctx->last_fault_rec[0] = fault_rec[0];
217 			ctx->last_fault_rec[1] = fault_rec[1];
218 			device_print_prettyname(ctx->ctx_tag.owner);
219 		}
220 		DMAR_UNLOCK(unit);
221 		printf(
222 		    "pci%d:%d:%d fault acc %x adt 0x%x reason 0x%x addr %jx\n",
223 		    bus, slot, func, DMAR_FRCD2_T(fault_rec[1]),
224 		    DMAR_FRCD2_AT(fault_rec[1]), DMAR_FRCD2_FR(fault_rec[1]),
225 		    (uintmax_t)fault_rec[0]);
226 		DMAR_FAULT_LOCK(unit);
227 	}
228 	DMAR_FAULT_UNLOCK(unit);
229 }
230 
231 static void
232 dmar_clear_faults(struct dmar_unit *unit)
233 {
234 	uint32_t frec, frir, fsts;
235 	int i;
236 
237 	for (i = 0; i < DMAR_CAP_NFR(unit->hw_cap); i++) {
238 		frir = (DMAR_CAP_FRO(unit->hw_cap) + i) * 16;
239 		frec = dmar_read4(unit, frir + 12);
240 		if ((frec & DMAR_FRCD2_F32) == 0)
241 			continue;
242 		dmar_write4(unit, frir + 12, DMAR_FRCD2_F32);
243 	}
244 	fsts = dmar_read4(unit, DMAR_FSTS_REG);
245 	dmar_write4(unit, DMAR_FSTS_REG, fsts);
246 }
247 
248 int
249 dmar_init_fault_log(struct dmar_unit *unit)
250 {
251 
252 	mtx_init(&unit->fault_lock, "dmarflt", NULL, MTX_SPIN);
253 	unit->fault_log_size = 256; /* 128 fault log entries */
254 	TUNABLE_INT_FETCH("hw.dmar.fault_log_size", &unit->fault_log_size);
255 	if (unit->fault_log_size % 2 != 0)
256 		panic("hw.dmar_fault_log_size must be even");
257 	unit->fault_log = malloc(sizeof(uint64_t) * unit->fault_log_size,
258 	    M_DEVBUF, M_WAITOK | M_ZERO);
259 
260 	TASK_INIT(&unit->fault_task, 0, dmar_fault_task, unit);
261 	unit->fault_taskqueue = taskqueue_create_fast("dmar", M_WAITOK,
262 	    taskqueue_thread_enqueue, &unit->fault_taskqueue);
263 	taskqueue_start_threads(&unit->fault_taskqueue, 1, PI_AV,
264 	    "dmar%d fault taskq", unit->unit);
265 
266 	DMAR_LOCK(unit);
267 	dmar_disable_fault_intr(unit);
268 	dmar_clear_faults(unit);
269 	dmar_enable_fault_intr(unit);
270 	DMAR_UNLOCK(unit);
271 
272 	return (0);
273 }
274 
275 void
276 dmar_fini_fault_log(struct dmar_unit *unit)
277 {
278 
279 	DMAR_LOCK(unit);
280 	dmar_disable_fault_intr(unit);
281 	DMAR_UNLOCK(unit);
282 
283 	if (unit->fault_taskqueue == NULL)
284 		return;
285 
286 	taskqueue_drain(unit->fault_taskqueue, &unit->fault_task);
287 	taskqueue_free(unit->fault_taskqueue);
288 	unit->fault_taskqueue = NULL;
289 	mtx_destroy(&unit->fault_lock);
290 
291 	free(unit->fault_log, M_DEVBUF);
292 	unit->fault_log = NULL;
293 	unit->fault_log_head = unit->fault_log_tail = 0;
294 }
295 
296 void
297 dmar_enable_fault_intr(struct dmar_unit *unit)
298 {
299 	uint32_t fectl;
300 
301 	DMAR_ASSERT_LOCKED(unit);
302 	fectl = dmar_read4(unit, DMAR_FECTL_REG);
303 	fectl &= ~DMAR_FECTL_IM;
304 	dmar_write4(unit, DMAR_FECTL_REG, fectl);
305 }
306 
307 void
308 dmar_disable_fault_intr(struct dmar_unit *unit)
309 {
310 	uint32_t fectl;
311 
312 	DMAR_ASSERT_LOCKED(unit);
313 	fectl = dmar_read4(unit, DMAR_FECTL_REG);
314 	dmar_write4(unit, DMAR_FECTL_REG, fectl | DMAR_FECTL_IM);
315 }
316