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