xref: /freebsd/sys/x86/iommu/intel_fault.c (revision 7afb8adff33d47f10a11368ff54bb2eec5b30165)
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 <sys/vmem.h>
45 #include <machine/bus.h>
46 #include <contrib/dev/acpica/include/acpi.h>
47 #include <contrib/dev/acpica/include/accommon.h>
48 #include <dev/acpica/acpivar.h>
49 #include <dev/pci/pcireg.h>
50 #include <dev/pci/pcivar.h>
51 #include <vm/vm.h>
52 #include <vm/vm_extern.h>
53 #include <vm/vm_kern.h>
54 #include <vm/vm_page.h>
55 #include <vm/vm_map.h>
56 #include <x86/include/busdma_impl.h>
57 #include <x86/iommu/intel_reg.h>
58 #include <x86/iommu/busdma_dmar.h>
59 #include <x86/iommu/intel_dmar.h>
60 
61 /*
62  * Fault interrupt handling for DMARs.  If advanced fault logging is
63  * not implemented by hardware, the code emulates it.  Fast interrupt
64  * handler flushes the fault registers into circular buffer at
65  * unit->fault_log, and schedules a task.
66  *
67  * The fast handler is used since faults usually come in bursts, and
68  * number of fault log registers is limited, e.g. down to one for 5400
69  * MCH.  We are trying to reduce the latency for clearing the fault
70  * register file.  The task is usually long-running, since printf() is
71  * slow, but this is not problematic because bursts are rare.
72  *
73  * For the same reason, each translation unit task is executed in its
74  * own thread.
75  *
76  * XXXKIB It seems there is no hardware available which implements
77  * advanced fault logging, so the code to handle AFL is not written.
78  */
79 
80 static int
81 dmar_fault_next(struct dmar_unit *unit, int faultp)
82 {
83 
84 	faultp += 2;
85 	if (faultp == unit->fault_log_size)
86 		faultp = 0;
87 	return (faultp);
88 }
89 
90 static void
91 dmar_fault_intr_clear(struct dmar_unit *unit, uint32_t fsts)
92 {
93 	uint32_t clear;
94 
95 	clear = 0;
96 	if ((fsts & DMAR_FSTS_ITE) != 0) {
97 		printf("DMAR%d: Invalidation timed out\n", unit->unit);
98 		clear |= DMAR_FSTS_ITE;
99 	}
100 	if ((fsts & DMAR_FSTS_ICE) != 0) {
101 		printf("DMAR%d: Invalidation completion error\n",
102 		    unit->unit);
103 		clear |= DMAR_FSTS_ICE;
104 	}
105 	if ((fsts & DMAR_FSTS_IQE) != 0) {
106 		printf("DMAR%d: Invalidation queue error\n",
107 		    unit->unit);
108 		clear |= DMAR_FSTS_IQE;
109 	}
110 	if ((fsts & DMAR_FSTS_APF) != 0) {
111 		printf("DMAR%d: Advanced pending fault\n", unit->unit);
112 		clear |= DMAR_FSTS_APF;
113 	}
114 	if ((fsts & DMAR_FSTS_AFO) != 0) {
115 		printf("DMAR%d: Advanced fault overflow\n", unit->unit);
116 		clear |= DMAR_FSTS_AFO;
117 	}
118 	if (clear != 0)
119 		dmar_write4(unit, DMAR_FSTS_REG, clear);
120 }
121 
122 int
123 dmar_fault_intr(void *arg)
124 {
125 	struct dmar_unit *unit;
126 	uint64_t fault_rec[2];
127 	uint32_t fsts;
128 	int fri, frir, faultp;
129 	bool enqueue;
130 
131 	unit = arg;
132 	enqueue = false;
133 	fsts = dmar_read4(unit, DMAR_FSTS_REG);
134 	dmar_fault_intr_clear(unit, fsts);
135 
136 	if ((fsts & DMAR_FSTS_PPF) == 0)
137 		goto done;
138 
139 	fri = DMAR_FSTS_FRI(fsts);
140 	for (;;) {
141 		frir = (DMAR_CAP_FRO(unit->hw_cap) + fri) * 16;
142 		fault_rec[1] = dmar_read8(unit, frir + 8);
143 		if ((fault_rec[1] & DMAR_FRCD2_F) == 0)
144 			break;
145 		fault_rec[0] = dmar_read8(unit, frir);
146 		dmar_write4(unit, frir + 12, DMAR_FRCD2_F32);
147 		DMAR_FAULT_LOCK(unit);
148 		faultp = unit->fault_log_head;
149 		if (dmar_fault_next(unit, faultp) == unit->fault_log_tail) {
150 			/* XXXKIB log overflow */
151 		} else {
152 			unit->fault_log[faultp] = fault_rec[0];
153 			unit->fault_log[faultp + 1] = fault_rec[1];
154 			unit->fault_log_head = dmar_fault_next(unit, faultp);
155 			enqueue = true;
156 		}
157 		DMAR_FAULT_UNLOCK(unit);
158 		fri += 1;
159 		if (fri >= DMAR_CAP_NFR(unit->hw_cap))
160 			fri = 0;
161 	}
162 
163 done:
164 	/*
165 	 * On SandyBridge, due to errata BJ124, IvyBridge errata
166 	 * BV100, and Haswell errata HSD40, "Spurious Intel VT-d
167 	 * Interrupts May Occur When the PFO Bit is Set".  Handle the
168 	 * cases by clearing overflow bit even if no fault is
169 	 * reported.
170 	 *
171 	 * On IvyBridge, errata BV30 states that clearing clear
172 	 * DMAR_FRCD2_F bit in the fault register causes spurious
173 	 * interrupt.  Do nothing.
174 	 *
175 	 */
176 	if ((fsts & DMAR_FSTS_PFO) != 0) {
177 		printf("DMAR%d: Fault Overflow\n", unit->unit);
178 		dmar_write4(unit, DMAR_FSTS_REG, DMAR_FSTS_PFO);
179 	}
180 
181 	if (enqueue) {
182 		taskqueue_enqueue_fast(unit->fault_taskqueue,
183 		    &unit->fault_task);
184 	}
185 	return (FILTER_HANDLED);
186 }
187 
188 static void
189 dmar_fault_task(void *arg, int pending __unused)
190 {
191 	struct dmar_unit *unit;
192 	struct dmar_ctx *ctx;
193 	uint64_t fault_rec[2];
194 	int sid, bus, slot, func, faultp;
195 
196 	unit = arg;
197 	DMAR_FAULT_LOCK(unit);
198 	for (;;) {
199 		faultp = unit->fault_log_tail;
200 		if (faultp == unit->fault_log_head)
201 			break;
202 
203 		fault_rec[0] = unit->fault_log[faultp];
204 		fault_rec[1] = unit->fault_log[faultp + 1];
205 		unit->fault_log_tail = dmar_fault_next(unit, faultp);
206 		DMAR_FAULT_UNLOCK(unit);
207 
208 		sid = DMAR_FRCD2_SID(fault_rec[1]);
209 		printf("DMAR%d: ", unit->unit);
210 		DMAR_LOCK(unit);
211 		ctx = dmar_find_ctx_locked(unit, sid);
212 		if (ctx == NULL) {
213 			printf("<unknown dev>:");
214 
215 			/*
216 			 * Note that the slot and function will not be correct
217 			 * if ARI is in use, but without a ctx entry we have
218 			 * no way of knowing whether ARI is in use or not.
219 			 */
220 			bus = PCI_RID2BUS(sid);
221 			slot = PCI_RID2SLOT(sid);
222 			func = PCI_RID2FUNC(sid);
223 		} else {
224 			ctx->flags |= DMAR_CTX_FAULTED;
225 			ctx->last_fault_rec[0] = fault_rec[0];
226 			ctx->last_fault_rec[1] = fault_rec[1];
227 			device_print_prettyname(ctx->ctx_tag.owner);
228 			bus = pci_get_bus(ctx->ctx_tag.owner);
229 			slot = pci_get_slot(ctx->ctx_tag.owner);
230 			func = pci_get_function(ctx->ctx_tag.owner);
231 		}
232 		DMAR_UNLOCK(unit);
233 		printf(
234 		    "pci%d:%d:%d sid %x fault acc %x adt 0x%x reason 0x%x "
235 		    "addr %jx\n",
236 		    bus, slot, func, sid, DMAR_FRCD2_T(fault_rec[1]),
237 		    DMAR_FRCD2_AT(fault_rec[1]), DMAR_FRCD2_FR(fault_rec[1]),
238 		    (uintmax_t)fault_rec[0]);
239 		DMAR_FAULT_LOCK(unit);
240 	}
241 	DMAR_FAULT_UNLOCK(unit);
242 }
243 
244 static void
245 dmar_clear_faults(struct dmar_unit *unit)
246 {
247 	uint32_t frec, frir, fsts;
248 	int i;
249 
250 	for (i = 0; i < DMAR_CAP_NFR(unit->hw_cap); i++) {
251 		frir = (DMAR_CAP_FRO(unit->hw_cap) + i) * 16;
252 		frec = dmar_read4(unit, frir + 12);
253 		if ((frec & DMAR_FRCD2_F32) == 0)
254 			continue;
255 		dmar_write4(unit, frir + 12, DMAR_FRCD2_F32);
256 	}
257 	fsts = dmar_read4(unit, DMAR_FSTS_REG);
258 	dmar_write4(unit, DMAR_FSTS_REG, fsts);
259 }
260 
261 int
262 dmar_init_fault_log(struct dmar_unit *unit)
263 {
264 
265 	mtx_init(&unit->fault_lock, "dmarflt", NULL, MTX_SPIN);
266 	unit->fault_log_size = 256; /* 128 fault log entries */
267 	TUNABLE_INT_FETCH("hw.dmar.fault_log_size", &unit->fault_log_size);
268 	if (unit->fault_log_size % 2 != 0)
269 		panic("hw.dmar_fault_log_size must be even");
270 	unit->fault_log = malloc(sizeof(uint64_t) * unit->fault_log_size,
271 	    M_DEVBUF, M_WAITOK | M_ZERO);
272 
273 	TASK_INIT(&unit->fault_task, 0, dmar_fault_task, unit);
274 	unit->fault_taskqueue = taskqueue_create_fast("dmar", M_WAITOK,
275 	    taskqueue_thread_enqueue, &unit->fault_taskqueue);
276 	taskqueue_start_threads(&unit->fault_taskqueue, 1, PI_AV,
277 	    "dmar%d fault taskq", unit->unit);
278 
279 	DMAR_LOCK(unit);
280 	dmar_disable_fault_intr(unit);
281 	dmar_clear_faults(unit);
282 	dmar_enable_fault_intr(unit);
283 	DMAR_UNLOCK(unit);
284 
285 	return (0);
286 }
287 
288 void
289 dmar_fini_fault_log(struct dmar_unit *unit)
290 {
291 
292 	DMAR_LOCK(unit);
293 	dmar_disable_fault_intr(unit);
294 	DMAR_UNLOCK(unit);
295 
296 	if (unit->fault_taskqueue == NULL)
297 		return;
298 
299 	taskqueue_drain(unit->fault_taskqueue, &unit->fault_task);
300 	taskqueue_free(unit->fault_taskqueue);
301 	unit->fault_taskqueue = NULL;
302 	mtx_destroy(&unit->fault_lock);
303 
304 	free(unit->fault_log, M_DEVBUF);
305 	unit->fault_log = NULL;
306 	unit->fault_log_head = unit->fault_log_tail = 0;
307 }
308 
309 void
310 dmar_enable_fault_intr(struct dmar_unit *unit)
311 {
312 	uint32_t fectl;
313 
314 	DMAR_ASSERT_LOCKED(unit);
315 	fectl = dmar_read4(unit, DMAR_FECTL_REG);
316 	fectl &= ~DMAR_FECTL_IM;
317 	dmar_write4(unit, DMAR_FECTL_REG, fectl);
318 }
319 
320 void
321 dmar_disable_fault_intr(struct dmar_unit *unit)
322 {
323 	uint32_t fectl;
324 
325 	DMAR_ASSERT_LOCKED(unit);
326 	fectl = dmar_read4(unit, DMAR_FECTL_REG);
327 	dmar_write4(unit, DMAR_FECTL_REG, fectl | DMAR_FECTL_IM);
328 }
329