xref: /freebsd/sys/x86/iommu/intel_intrmap.c (revision 59c8e88e72633afbc47a4ace0d2170d00d51f7dc)
1 /*-
2  * Copyright (c) 2015 The FreeBSD Foundation
3  *
4  * This software was developed by Konstantin Belousov <kib@FreeBSD.org>
5  * under sponsorship from the FreeBSD Foundation.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/param.h>
30 #include <sys/systm.h>
31 #include <sys/bus.h>
32 #include <sys/kernel.h>
33 #include <sys/lock.h>
34 #include <sys/malloc.h>
35 #include <sys/memdesc.h>
36 #include <sys/mutex.h>
37 #include <sys/rman.h>
38 #include <sys/rwlock.h>
39 #include <sys/sysctl.h>
40 #include <sys/taskqueue.h>
41 #include <sys/tree.h>
42 #include <sys/vmem.h>
43 #include <vm/vm.h>
44 #include <vm/vm_extern.h>
45 #include <vm/vm_kern.h>
46 #include <vm/vm_object.h>
47 #include <vm/vm_page.h>
48 #include <dev/pci/pcireg.h>
49 #include <dev/pci/pcivar.h>
50 #include <machine/bus.h>
51 #include <machine/intr_machdep.h>
52 #include <x86/include/apicreg.h>
53 #include <x86/include/apicvar.h>
54 #include <x86/include/busdma_impl.h>
55 #include <dev/iommu/busdma_iommu.h>
56 #include <x86/iommu/intel_reg.h>
57 #include <x86/iommu/intel_dmar.h>
58 #include <x86/iommu/iommu_intrmap.h>
59 
60 static struct dmar_unit *dmar_ir_find(device_t src, uint16_t *rid,
61     int *is_dmar);
62 static void dmar_ir_program_irte(struct dmar_unit *unit, u_int idx,
63     uint64_t low, uint16_t rid);
64 static int dmar_ir_free_irte(struct dmar_unit *unit, u_int cookie);
65 
66 int
67 iommu_alloc_msi_intr(device_t src, u_int *cookies, u_int count)
68 {
69 	struct dmar_unit *unit;
70 	vmem_addr_t vmem_res;
71 	u_int idx, i;
72 	int error;
73 
74 	unit = dmar_ir_find(src, NULL, NULL);
75 	if (unit == NULL || !unit->ir_enabled) {
76 		for (i = 0; i < count; i++)
77 			cookies[i] = -1;
78 		return (EOPNOTSUPP);
79 	}
80 
81 	error = vmem_alloc(unit->irtids, count, M_FIRSTFIT | M_NOWAIT,
82 	    &vmem_res);
83 	if (error != 0) {
84 		KASSERT(error != EOPNOTSUPP,
85 		    ("impossible EOPNOTSUPP from vmem"));
86 		return (error);
87 	}
88 	idx = vmem_res;
89 	for (i = 0; i < count; i++)
90 		cookies[i] = idx + i;
91 	return (0);
92 }
93 
94 int
95 iommu_map_msi_intr(device_t src, u_int cpu, u_int vector, u_int cookie,
96     uint64_t *addr, uint32_t *data)
97 {
98 	struct dmar_unit *unit;
99 	uint64_t low;
100 	uint16_t rid;
101 	int is_dmar;
102 
103 	unit = dmar_ir_find(src, &rid, &is_dmar);
104 	if (is_dmar) {
105 		KASSERT(unit == NULL, ("DMAR cannot translate itself"));
106 
107 		/*
108 		 * See VT-d specification, 5.1.6 Remapping Hardware -
109 		 * Interrupt Programming.
110 		 */
111 		*data = vector;
112 		*addr = MSI_INTEL_ADDR_BASE | ((cpu & 0xff) << 12);
113 		if (x2apic_mode)
114 			*addr |= ((uint64_t)cpu & 0xffffff00) << 32;
115 		else
116 			KASSERT(cpu <= 0xff, ("cpu id too big %d", cpu));
117 		return (0);
118 	}
119 	if (unit == NULL || !unit->ir_enabled || cookie == -1)
120 		return (EOPNOTSUPP);
121 
122 	low = (DMAR_X2APIC(unit) ? DMAR_IRTE1_DST_x2APIC(cpu) :
123 	    DMAR_IRTE1_DST_xAPIC(cpu)) | DMAR_IRTE1_V(vector) |
124 	    DMAR_IRTE1_DLM_FM | DMAR_IRTE1_TM_EDGE | DMAR_IRTE1_RH_DIRECT |
125 	    DMAR_IRTE1_DM_PHYSICAL | DMAR_IRTE1_P;
126 	dmar_ir_program_irte(unit, cookie, low, rid);
127 
128 	if (addr != NULL) {
129 		/*
130 		 * See VT-d specification, 5.1.5.2 MSI and MSI-X
131 		 * Register Programming.
132 		 */
133 		*addr = MSI_INTEL_ADDR_BASE | ((cookie & 0x7fff) << 5) |
134 		    ((cookie & 0x8000) << 2) | 0x18;
135 		*data = 0;
136 	}
137 	return (0);
138 }
139 
140 int
141 iommu_unmap_msi_intr(device_t src, u_int cookie)
142 {
143 	struct dmar_unit *unit;
144 
145 	if (cookie == -1)
146 		return (0);
147 	unit = dmar_ir_find(src, NULL, NULL);
148 	return (dmar_ir_free_irte(unit, cookie));
149 }
150 
151 int
152 iommu_map_ioapic_intr(u_int ioapic_id, u_int cpu, u_int vector, bool edge,
153     bool activehi, int irq, u_int *cookie, uint32_t *hi, uint32_t *lo)
154 {
155 	struct dmar_unit *unit;
156 	vmem_addr_t vmem_res;
157 	uint64_t low, iorte;
158 	u_int idx;
159 	int error;
160 	uint16_t rid;
161 
162 	unit = dmar_find_ioapic(ioapic_id, &rid);
163 	if (unit == NULL || !unit->ir_enabled) {
164 		*cookie = -1;
165 		return (EOPNOTSUPP);
166 	}
167 
168 	error = vmem_alloc(unit->irtids, 1, M_FIRSTFIT | M_NOWAIT, &vmem_res);
169 	if (error != 0) {
170 		KASSERT(error != EOPNOTSUPP,
171 		    ("impossible EOPNOTSUPP from vmem"));
172 		return (error);
173 	}
174 	idx = vmem_res;
175 	low = 0;
176 	switch (irq) {
177 	case IRQ_EXTINT:
178 		low |= DMAR_IRTE1_DLM_ExtINT;
179 		break;
180 	case IRQ_NMI:
181 		low |= DMAR_IRTE1_DLM_NMI;
182 		break;
183 	case IRQ_SMI:
184 		low |= DMAR_IRTE1_DLM_SMI;
185 		break;
186 	default:
187 		KASSERT(vector != 0, ("No vector for IRQ %u", irq));
188 		low |= DMAR_IRTE1_DLM_FM | DMAR_IRTE1_V(vector);
189 		break;
190 	}
191 	low |= (DMAR_X2APIC(unit) ? DMAR_IRTE1_DST_x2APIC(cpu) :
192 	    DMAR_IRTE1_DST_xAPIC(cpu)) |
193 	    (edge ? DMAR_IRTE1_TM_EDGE : DMAR_IRTE1_TM_LEVEL) |
194 	    DMAR_IRTE1_RH_DIRECT | DMAR_IRTE1_DM_PHYSICAL | DMAR_IRTE1_P;
195 	dmar_ir_program_irte(unit, idx, low, rid);
196 
197 	if (hi != NULL) {
198 		/*
199 		 * See VT-d specification, 5.1.5.1 I/OxAPIC
200 		 * Programming.
201 		 */
202 		iorte = (1ULL << 48) | ((uint64_t)(idx & 0x7fff) << 49) |
203 		    ((idx & 0x8000) != 0 ? (1 << 11) : 0) |
204 		    (edge ? IOART_TRGREDG : IOART_TRGRLVL) |
205 		    (activehi ? IOART_INTAHI : IOART_INTALO) |
206 		    IOART_DELFIXED | vector;
207 		*hi = iorte >> 32;
208 		*lo = iorte;
209 	}
210 	*cookie = idx;
211 	return (0);
212 }
213 
214 int
215 iommu_unmap_ioapic_intr(u_int ioapic_id, u_int *cookie)
216 {
217 	struct dmar_unit *unit;
218 	u_int idx;
219 
220 	idx = *cookie;
221 	if (idx == -1)
222 		return (0);
223 	*cookie = -1;
224 	unit = dmar_find_ioapic(ioapic_id, NULL);
225 	KASSERT(unit != NULL && unit->ir_enabled,
226 	    ("unmap: cookie %d unit %p", idx, unit));
227 	return (dmar_ir_free_irte(unit, idx));
228 }
229 
230 static struct dmar_unit *
231 dmar_ir_find(device_t src, uint16_t *rid, int *is_dmar)
232 {
233 	devclass_t src_class;
234 	struct dmar_unit *unit;
235 
236 	/*
237 	 * We need to determine if the interrupt source generates FSB
238 	 * interrupts.  If yes, it is either DMAR, in which case
239 	 * interrupts are not remapped.  Or it is HPET, and interrupts
240 	 * are remapped.  For HPET, source id is reported by HPET
241 	 * record in DMAR ACPI table.
242 	 */
243 	if (is_dmar != NULL)
244 		*is_dmar = FALSE;
245 	src_class = device_get_devclass(src);
246 	if (src_class == devclass_find("dmar")) {
247 		unit = NULL;
248 		if (is_dmar != NULL)
249 			*is_dmar = TRUE;
250 	} else if (src_class == devclass_find("hpet")) {
251 		unit = dmar_find_hpet(src, rid);
252 	} else {
253 		unit = dmar_find(src, bootverbose);
254 		if (unit != NULL && rid != NULL)
255 			iommu_get_requester(src, rid);
256 	}
257 	return (unit);
258 }
259 
260 static void
261 dmar_ir_program_irte(struct dmar_unit *unit, u_int idx, uint64_t low,
262     uint16_t rid)
263 {
264 	dmar_irte_t *irte;
265 	uint64_t high;
266 
267 	KASSERT(idx < unit->irte_cnt,
268 	    ("bad cookie %d %d", idx, unit->irte_cnt));
269 	irte = &(unit->irt[idx]);
270 	high = DMAR_IRTE2_SVT_RID | DMAR_IRTE2_SQ_RID |
271 	    DMAR_IRTE2_SID_RID(rid);
272 	if (bootverbose) {
273 		device_printf(unit->dev,
274 		    "programming irte[%d] rid %#x high %#jx low %#jx\n",
275 		    idx, rid, (uintmax_t)high, (uintmax_t)low);
276 	}
277 	DMAR_LOCK(unit);
278 	if ((irte->irte1 & DMAR_IRTE1_P) != 0) {
279 		/*
280 		 * The rte is already valid.  Assume that the request
281 		 * is to remap the interrupt for balancing.  Only low
282 		 * word of rte needs to be changed.  Assert that the
283 		 * high word contains expected value.
284 		 */
285 		KASSERT(irte->irte2 == high,
286 		    ("irte2 mismatch, %jx %jx", (uintmax_t)irte->irte2,
287 		    (uintmax_t)high));
288 		dmar_pte_update(&irte->irte1, low);
289 	} else {
290 		dmar_pte_store(&irte->irte2, high);
291 		dmar_pte_store(&irte->irte1, low);
292 	}
293 	dmar_qi_invalidate_iec(unit, idx, 1);
294 	DMAR_UNLOCK(unit);
295 
296 }
297 
298 static int
299 dmar_ir_free_irte(struct dmar_unit *unit, u_int cookie)
300 {
301 	dmar_irte_t *irte;
302 
303 	KASSERT(unit != NULL && unit->ir_enabled,
304 	    ("unmap: cookie %d unit %p", cookie, unit));
305 	KASSERT(cookie < unit->irte_cnt,
306 	    ("bad cookie %u %u", cookie, unit->irte_cnt));
307 	irte = &(unit->irt[cookie]);
308 	dmar_pte_clear(&irte->irte1);
309 	dmar_pte_clear(&irte->irte2);
310 	DMAR_LOCK(unit);
311 	dmar_qi_invalidate_iec(unit, cookie, 1);
312 	DMAR_UNLOCK(unit);
313 	vmem_free(unit->irtids, cookie, 1);
314 	return (0);
315 }
316 
317 static u_int
318 clp2(u_int v)
319 {
320 
321 	return (powerof2(v) ? v : 1 << fls(v));
322 }
323 
324 int
325 dmar_init_irt(struct dmar_unit *unit)
326 {
327 
328 	if ((unit->hw_ecap & DMAR_ECAP_IR) == 0)
329 		return (0);
330 	unit->ir_enabled = 1;
331 	TUNABLE_INT_FETCH("hw.dmar.ir", &unit->ir_enabled);
332 	if (!unit->ir_enabled)
333 		return (0);
334 	if (!unit->qi_enabled) {
335 		unit->ir_enabled = 0;
336 		if (bootverbose)
337 			device_printf(unit->dev,
338 	     "QI disabled, disabling interrupt remapping\n");
339 		return (0);
340 	}
341 	unit->irte_cnt = clp2(num_io_irqs);
342 	unit->irt = kmem_alloc_contig(unit->irte_cnt * sizeof(dmar_irte_t),
343 	    M_ZERO | M_WAITOK, 0, dmar_high, PAGE_SIZE, 0,
344 	    DMAR_IS_COHERENT(unit) ?
345 	    VM_MEMATTR_DEFAULT : VM_MEMATTR_UNCACHEABLE);
346 	if (unit->irt == NULL)
347 		return (ENOMEM);
348 	unit->irt_phys = pmap_kextract((vm_offset_t)unit->irt);
349 	unit->irtids = vmem_create("dmarirt", 0, unit->irte_cnt, 1, 0,
350 	    M_FIRSTFIT | M_NOWAIT);
351 	DMAR_LOCK(unit);
352 	dmar_load_irt_ptr(unit);
353 	dmar_qi_invalidate_iec_glob(unit);
354 	DMAR_UNLOCK(unit);
355 
356 	/*
357 	 * Initialize mappings for already configured interrupt pins.
358 	 * Required, because otherwise the interrupts fault without
359 	 * irtes.
360 	 */
361 	intr_reprogram();
362 
363 	DMAR_LOCK(unit);
364 	dmar_enable_ir(unit);
365 	DMAR_UNLOCK(unit);
366 	return (0);
367 }
368 
369 void
370 dmar_fini_irt(struct dmar_unit *unit)
371 {
372 
373 	unit->ir_enabled = 0;
374 	if (unit->irt != NULL) {
375 		dmar_disable_ir(unit);
376 		dmar_qi_invalidate_iec_glob(unit);
377 		vmem_destroy(unit->irtids);
378 		kmem_free(unit->irt, unit->irte_cnt * sizeof(dmar_irte_t));
379 	}
380 }
381