1 /*-
2 * Copyright (c) 2015 The FreeBSD Foundation
3 *
4 * This software was developed by Konstantin Belousov <[email protected]>
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/cdefs.h>
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/bus.h>
33 #include <sys/domainset.h>
34 #include <sys/kernel.h>
35 #include <sys/lock.h>
36 #include <sys/malloc.h>
37 #include <sys/memdesc.h>
38 #include <sys/mutex.h>
39 #include <sys/rman.h>
40 #include <sys/rwlock.h>
41 #include <sys/sysctl.h>
42 #include <sys/taskqueue.h>
43 #include <sys/tree.h>
44 #include <sys/vmem.h>
45 #include <vm/vm.h>
46 #include <vm/vm_extern.h>
47 #include <vm/vm_kern.h>
48 #include <vm/vm_object.h>
49 #include <vm/vm_page.h>
50 #include <dev/pci/pcireg.h>
51 #include <dev/pci/pcivar.h>
52 #include <machine/bus.h>
53 #include <machine/intr_machdep.h>
54 #include <x86/include/apicreg.h>
55 #include <x86/include/apicvar.h>
56 #include <x86/include/busdma_impl.h>
57 #include <dev/iommu/busdma_iommu.h>
58 #include <x86/iommu/intel_reg.h>
59 #include <x86/iommu/x86_iommu.h>
60 #include <x86/iommu/intel_dmar.h>
61 #include <x86/iommu/iommu_intrmap.h>
62
63 static struct dmar_unit *dmar_ir_find(device_t src, uint16_t *rid,
64 int *is_dmar);
65 static void dmar_ir_program_irte(struct dmar_unit *unit, u_int idx,
66 uint64_t low, uint16_t rid);
67 static int dmar_ir_free_irte(struct dmar_unit *unit, u_int cookie);
68
69 int
dmar_alloc_msi_intr(device_t src,u_int * cookies,u_int count)70 dmar_alloc_msi_intr(device_t src, u_int *cookies, u_int count)
71 {
72 struct dmar_unit *unit;
73 vmem_addr_t vmem_res;
74 u_int idx, i;
75 int error;
76
77 unit = dmar_ir_find(src, NULL, NULL);
78 if (unit == NULL || !unit->ir_enabled) {
79 for (i = 0; i < count; i++)
80 cookies[i] = -1;
81 return (EOPNOTSUPP);
82 }
83
84 error = vmem_alloc(unit->irtids, count, M_FIRSTFIT | M_NOWAIT,
85 &vmem_res);
86 if (error != 0) {
87 KASSERT(error != EOPNOTSUPP,
88 ("impossible EOPNOTSUPP from vmem"));
89 return (error);
90 }
91 idx = vmem_res;
92 for (i = 0; i < count; i++)
93 cookies[i] = idx + i;
94 return (0);
95 }
96
97 int
dmar_map_msi_intr(device_t src,u_int cpu,u_int vector,u_int cookie,uint64_t * addr,uint32_t * data)98 dmar_map_msi_intr(device_t src, u_int cpu, u_int vector, u_int cookie,
99 uint64_t *addr, uint32_t *data)
100 {
101 struct dmar_unit *unit;
102 uint64_t low;
103 uint16_t rid;
104 int is_dmar;
105
106 unit = dmar_ir_find(src, &rid, &is_dmar);
107 if (is_dmar) {
108 KASSERT(unit == NULL, ("DMAR cannot translate itself"));
109
110 /*
111 * See VT-d specification, 5.1.6 Remapping Hardware -
112 * Interrupt Programming.
113 */
114 *data = vector;
115 *addr = MSI_INTEL_ADDR_BASE | ((cpu & 0xff) << 12);
116 if (x2apic_mode)
117 *addr |= ((uint64_t)cpu & 0xffffff00) << 32;
118 else
119 KASSERT(cpu <= 0xff, ("cpu id too big %d", cpu));
120 return (0);
121 }
122 if (unit == NULL || !unit->ir_enabled || cookie == -1)
123 return (EOPNOTSUPP);
124
125 low = (DMAR_X2APIC(unit) ? DMAR_IRTE1_DST_x2APIC(cpu) :
126 DMAR_IRTE1_DST_xAPIC(cpu)) | DMAR_IRTE1_V(vector) |
127 DMAR_IRTE1_DLM_FM | DMAR_IRTE1_TM_EDGE | DMAR_IRTE1_RH_DIRECT |
128 DMAR_IRTE1_DM_PHYSICAL | DMAR_IRTE1_P;
129 dmar_ir_program_irte(unit, cookie, low, rid);
130
131 if (addr != NULL) {
132 /*
133 * See VT-d specification, 5.1.5.2 MSI and MSI-X
134 * Register Programming.
135 */
136 *addr = MSI_INTEL_ADDR_BASE | ((cookie & 0x7fff) << 5) |
137 ((cookie & 0x8000) << 2) | 0x18;
138 *data = 0;
139 }
140 return (0);
141 }
142
143 int
dmar_unmap_msi_intr(device_t src,u_int cookie)144 dmar_unmap_msi_intr(device_t src, u_int cookie)
145 {
146 struct dmar_unit *unit;
147
148 if (cookie == -1)
149 return (0);
150 unit = dmar_ir_find(src, NULL, NULL);
151 return (dmar_ir_free_irte(unit, cookie));
152 }
153
154 int
dmar_map_ioapic_intr(u_int ioapic_id,u_int cpu,u_int vector,bool edge,bool activehi,int irq,u_int * cookie,uint32_t * hi,uint32_t * lo)155 dmar_map_ioapic_intr(u_int ioapic_id, u_int cpu, u_int vector, bool edge,
156 bool activehi, int irq, u_int *cookie, uint32_t *hi, uint32_t *lo)
157 {
158 struct dmar_unit *unit;
159 vmem_addr_t vmem_res;
160 uint64_t low, iorte;
161 u_int idx;
162 int error;
163 uint16_t rid;
164
165 unit = dmar_find_ioapic(ioapic_id, &rid);
166 if (unit == NULL || !unit->ir_enabled) {
167 *cookie = -1;
168 return (EOPNOTSUPP);
169 }
170
171 error = vmem_alloc(unit->irtids, 1, M_FIRSTFIT | M_NOWAIT, &vmem_res);
172 if (error != 0) {
173 KASSERT(error != EOPNOTSUPP,
174 ("impossible EOPNOTSUPP from vmem"));
175 return (error);
176 }
177 idx = vmem_res;
178 low = 0;
179 switch (irq) {
180 case IRQ_EXTINT:
181 low |= DMAR_IRTE1_DLM_ExtINT;
182 break;
183 case IRQ_NMI:
184 low |= DMAR_IRTE1_DLM_NMI;
185 break;
186 case IRQ_SMI:
187 low |= DMAR_IRTE1_DLM_SMI;
188 break;
189 default:
190 KASSERT(vector != 0, ("No vector for IRQ %u", irq));
191 low |= DMAR_IRTE1_DLM_FM | DMAR_IRTE1_V(vector);
192 break;
193 }
194 low |= (DMAR_X2APIC(unit) ? DMAR_IRTE1_DST_x2APIC(cpu) :
195 DMAR_IRTE1_DST_xAPIC(cpu)) |
196 (edge ? DMAR_IRTE1_TM_EDGE : DMAR_IRTE1_TM_LEVEL) |
197 DMAR_IRTE1_RH_DIRECT | DMAR_IRTE1_DM_PHYSICAL | DMAR_IRTE1_P;
198 dmar_ir_program_irte(unit, idx, low, rid);
199
200 if (hi != NULL) {
201 /*
202 * See VT-d specification, 5.1.5.1 I/OxAPIC
203 * Programming.
204 */
205 iorte = (1ULL << 48) | ((uint64_t)(idx & 0x7fff) << 49) |
206 ((idx & 0x8000) != 0 ? (1 << 11) : 0) |
207 (edge ? IOART_TRGREDG : IOART_TRGRLVL) |
208 (activehi ? IOART_INTAHI : IOART_INTALO) |
209 IOART_DELFIXED | vector;
210 *hi = iorte >> 32;
211 *lo = iorte;
212 }
213 *cookie = idx;
214 return (0);
215 }
216
217 int
dmar_unmap_ioapic_intr(u_int ioapic_id,u_int * cookie)218 dmar_unmap_ioapic_intr(u_int ioapic_id, u_int *cookie)
219 {
220 struct dmar_unit *unit;
221 u_int idx;
222
223 idx = *cookie;
224 if (idx == -1)
225 return (0);
226 *cookie = -1;
227 unit = dmar_find_ioapic(ioapic_id, NULL);
228 KASSERT(unit != NULL && unit->ir_enabled,
229 ("unmap: cookie %d unit %p", idx, unit));
230 return (dmar_ir_free_irte(unit, idx));
231 }
232
233 static struct dmar_unit *
dmar_ir_find(device_t src,uint16_t * rid,int * is_dmar)234 dmar_ir_find(device_t src, uint16_t *rid, int *is_dmar)
235 {
236 devclass_t src_class;
237 struct dmar_unit *unit;
238
239 /*
240 * We need to determine if the interrupt source generates FSB
241 * interrupts. If yes, it is either DMAR, in which case
242 * interrupts are not remapped. Or it is HPET, and interrupts
243 * are remapped. For HPET, source id is reported by HPET
244 * record in DMAR ACPI table.
245 */
246 if (is_dmar != NULL)
247 *is_dmar = FALSE;
248 src_class = device_get_devclass(src);
249 if (src_class == devclass_find("dmar")) {
250 unit = NULL;
251 if (is_dmar != NULL)
252 *is_dmar = TRUE;
253 } else if (src_class == devclass_find("hpet")) {
254 unit = dmar_find_hpet(src, rid);
255 } else {
256 unit = dmar_find(src, bootverbose);
257 if (unit != NULL && rid != NULL)
258 iommu_get_requester(src, rid);
259 }
260 return (unit);
261 }
262
263 static void
dmar_ir_program_irte(struct dmar_unit * unit,u_int idx,uint64_t low,uint16_t rid)264 dmar_ir_program_irte(struct dmar_unit *unit, u_int idx, uint64_t low,
265 uint16_t rid)
266 {
267 dmar_irte_t *irte;
268 uint64_t high;
269
270 KASSERT(idx < unit->irte_cnt,
271 ("bad cookie %d %d", idx, unit->irte_cnt));
272 irte = &(unit->irt[idx]);
273 high = DMAR_IRTE2_SVT_RID | DMAR_IRTE2_SQ_RID |
274 DMAR_IRTE2_SID_RID(rid);
275 if (bootverbose) {
276 device_printf(unit->iommu.dev,
277 "programming irte[%d] rid %#x high %#jx low %#jx\n",
278 idx, rid, (uintmax_t)high, (uintmax_t)low);
279 }
280 DMAR_LOCK(unit);
281 if ((irte->irte1 & DMAR_IRTE1_P) != 0) {
282 /*
283 * The rte is already valid. Assume that the request
284 * is to remap the interrupt for balancing. Only low
285 * word of rte needs to be changed. Assert that the
286 * high word contains expected value.
287 */
288 KASSERT(irte->irte2 == high,
289 ("irte2 mismatch, %jx %jx", (uintmax_t)irte->irte2,
290 (uintmax_t)high));
291 dmar_pte_update(&irte->irte1, low);
292 } else {
293 dmar_pte_store(&irte->irte2, high);
294 dmar_pte_store(&irte->irte1, low);
295 }
296 dmar_qi_invalidate_iec(unit, idx, 1);
297 DMAR_UNLOCK(unit);
298
299 }
300
301 static int
dmar_ir_free_irte(struct dmar_unit * unit,u_int cookie)302 dmar_ir_free_irte(struct dmar_unit *unit, u_int cookie)
303 {
304 dmar_irte_t *irte;
305
306 KASSERT(unit != NULL && unit->ir_enabled,
307 ("unmap: cookie %d unit %p", cookie, unit));
308 KASSERT(cookie < unit->irte_cnt,
309 ("bad cookie %u %u", cookie, unit->irte_cnt));
310 irte = &(unit->irt[cookie]);
311 dmar_pte_clear(&irte->irte1);
312 dmar_pte_clear(&irte->irte2);
313 DMAR_LOCK(unit);
314 dmar_qi_invalidate_iec(unit, cookie, 1);
315 DMAR_UNLOCK(unit);
316 vmem_free(unit->irtids, cookie, 1);
317 return (0);
318 }
319
320 static u_int
clp2(u_int v)321 clp2(u_int v)
322 {
323
324 return (powerof2(v) ? v : 1 << fls(v));
325 }
326
327 int
dmar_init_irt(struct dmar_unit * unit)328 dmar_init_irt(struct dmar_unit *unit)
329 {
330 SYSCTL_ADD_INT(&unit->iommu.sysctl_ctx,
331 SYSCTL_CHILDREN(device_get_sysctl_tree(unit->iommu.dev)),
332 OID_AUTO, "ir", CTLFLAG_RD, &unit->ir_enabled, 0,
333 "Interrupt remapping ops enabled");
334 if ((unit->hw_ecap & DMAR_ECAP_IR) == 0)
335 return (0);
336 unit->ir_enabled = 1;
337 TUNABLE_INT_FETCH("hw.dmar.ir", &unit->ir_enabled);
338 TUNABLE_INT_FETCH("hw.iommu.ir", &unit->ir_enabled);
339 if (!unit->ir_enabled)
340 return (0);
341 if (!unit->qi_enabled) {
342 unit->ir_enabled = 0;
343 if (bootverbose)
344 device_printf(unit->iommu.dev,
345 "QI disabled, disabling interrupt remapping\n");
346 return (0);
347 }
348 unit->irte_cnt = clp2(num_io_irqs);
349 if (unit->memdomain == -1) {
350 unit->irt = kmem_alloc_contig(
351 unit->irte_cnt * sizeof(dmar_irte_t),
352 M_ZERO | M_WAITOK, 0, iommu_high, PAGE_SIZE, 0,
353 DMAR_IS_COHERENT(unit) ?
354 VM_MEMATTR_DEFAULT : VM_MEMATTR_UNCACHEABLE);
355 } else {
356 unit->irt = kmem_alloc_contig_domainset(
357 DOMAINSET_PREF(unit->memdomain),
358 unit->irte_cnt * sizeof(dmar_irte_t),
359 M_ZERO | M_WAITOK, 0, iommu_high, PAGE_SIZE, 0,
360 DMAR_IS_COHERENT(unit) ?
361 VM_MEMATTR_DEFAULT : VM_MEMATTR_UNCACHEABLE);
362 }
363 if (unit->irt == NULL)
364 return (ENOMEM);
365 unit->irt_phys = pmap_kextract((vm_offset_t)unit->irt);
366 unit->irtids = vmem_create("dmarirt", 0, unit->irte_cnt, 1, 0,
367 M_FIRSTFIT | M_NOWAIT);
368 DMAR_LOCK(unit);
369 dmar_load_irt_ptr(unit);
370 dmar_qi_invalidate_iec_glob(unit);
371 DMAR_UNLOCK(unit);
372
373 /*
374 * Initialize mappings for already configured interrupt pins.
375 * Required, because otherwise the interrupts fault without
376 * irtes.
377 */
378 intr_reprogram();
379
380 DMAR_LOCK(unit);
381 dmar_enable_ir(unit);
382 DMAR_UNLOCK(unit);
383 return (0);
384 }
385
386 void
dmar_fini_irt(struct dmar_unit * unit)387 dmar_fini_irt(struct dmar_unit *unit)
388 {
389
390 unit->ir_enabled = 0;
391 if (unit->irt != NULL) {
392 dmar_disable_ir(unit);
393 dmar_qi_invalidate_iec_glob(unit);
394 vmem_destroy(unit->irtids);
395 kmem_free(unit->irt, unit->irte_cnt * sizeof(dmar_irte_t));
396 }
397 }
398