projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
iommu/vt-d: Enhance intel_irq_remapping driver to support DMAR unit hotplug
[deliverable/linux.git] / drivers / iommu / intel_irq_remapping.c
1 #include <linux/interrupt.h>
2 #include <linux/dmar.h>
3 #include <linux/spinlock.h>
4 #include <linux/slab.h>
5 #include <linux/jiffies.h>
6 #include <linux/hpet.h>
7 #include <linux/pci.h>
8 #include <linux/irq.h>
9 #include <linux/intel-iommu.h>
10 #include <linux/acpi.h>
11 #include <asm/io_apic.h>
12 #include <asm/smp.h>
13 #include <asm/cpu.h>
14 #include <asm/irq_remapping.h>
15 #include <asm/pci-direct.h>
16 #include <asm/msidef.h>
17
18 #include "irq_remapping.h"
19
20 struct ioapic_scope {
21 struct intel_iommu *iommu;
22 unsigned int id;
23 unsigned int bus; /* PCI bus number */
24 unsigned int devfn; /* PCI devfn number */
25 };
26
27 struct hpet_scope {
28 struct intel_iommu *iommu;
29 u8 id;
30 unsigned int bus;
31 unsigned int devfn;
32 };
33
34 #define IR_X2APIC_MODE(mode) (mode ? (1 << 11) : 0)
35 #define IRTE_DEST(dest) ((x2apic_mode) ? dest : dest << 8)
36
37 static struct ioapic_scope ir_ioapic[MAX_IO_APICS];
38 static struct hpet_scope ir_hpet[MAX_HPET_TBS];
39
40 /*
41 * Lock ordering:
42 * ->dmar_global_lock
43 * ->irq_2_ir_lock
44 * ->qi->q_lock
45 * ->iommu->register_lock
46 * Note:
47 * intel_irq_remap_ops.{supported,prepare,enable,disable,reenable} are called
48 * in single-threaded environment with interrupt disabled, so no need to tabke
49 * the dmar_global_lock.
50 */
51 static DEFINE_RAW_SPINLOCK(irq_2_ir_lock);
52
53 static int __init parse_ioapics_under_ir(void);
54
55 static struct irq_2_iommu *irq_2_iommu(unsigned int irq)
56 {
57 struct irq_cfg *cfg = irq_get_chip_data(irq);
58 return cfg ? &cfg->irq_2_iommu : NULL;
59 }
60
61 static int get_irte(int irq, struct irte *entry)
62 {
63 struct irq_2_iommu *irq_iommu = irq_2_iommu(irq);
64 unsigned long flags;
65 int index;
66
67 if (!entry || !irq_iommu)
68 return -1;
69
70 raw_spin_lock_irqsave(&irq_2_ir_lock, flags);
71
72 if (unlikely(!irq_iommu->iommu)) {
73 raw_spin_unlock_irqrestore(&irq_2_ir_lock, flags);
74 return -1;
75 }
76
77 index = irq_iommu->irte_index + irq_iommu->sub_handle;
78 *entry = *(irq_iommu->iommu->ir_table->base + index);
79
80 raw_spin_unlock_irqrestore(&irq_2_ir_lock, flags);
81 return 0;
82 }
83
84 static int alloc_irte(struct intel_iommu *iommu, int irq, u16 count)
85 {
86 struct ir_table *table = iommu->ir_table;
87 struct irq_2_iommu *irq_iommu = irq_2_iommu(irq);
88 struct irq_cfg *cfg = irq_get_chip_data(irq);
89 unsigned int mask = 0;
90 unsigned long flags;
91 int index;
92
93 if (!count || !irq_iommu)
94 return -1;
95
96 if (count > 1) {
97 count = __roundup_pow_of_two(count);
98 mask = ilog2(count);
99 }
100
101 if (mask > ecap_max_handle_mask(iommu->ecap)) {
102 printk(KERN_ERR
103 "Requested mask %x exceeds the max invalidation handle"
104 " mask value %Lx\n", mask,
105 ecap_max_handle_mask(iommu->ecap));
106 return -1;
107 }
108
109 raw_spin_lock_irqsave(&irq_2_ir_lock, flags);
110 index = bitmap_find_free_region(table->bitmap,
111 INTR_REMAP_TABLE_ENTRIES, mask);
112 if (index < 0) {
113 pr_warn("IR%d: can't allocate an IRTE\n", iommu->seq_id);
114 } else {
115 cfg->remapped = 1;
116 irq_iommu->iommu = iommu;
117 irq_iommu->irte_index = index;
118 irq_iommu->sub_handle = 0;
119 irq_iommu->irte_mask = mask;
120 }
121 raw_spin_unlock_irqrestore(&irq_2_ir_lock, flags);
122
123 return index;
124 }
125
126 static int qi_flush_iec(struct intel_iommu *iommu, int index, int mask)
127 {
128 struct qi_desc desc;
129
130 desc.low = QI_IEC_IIDEX(index) | QI_IEC_TYPE | QI_IEC_IM(mask)
131 | QI_IEC_SELECTIVE;
132 desc.high = 0;
133
134 return qi_submit_sync(&desc, iommu);
135 }
136
137 static int map_irq_to_irte_handle(int irq, u16 *sub_handle)
138 {
139 struct irq_2_iommu *irq_iommu = irq_2_iommu(irq);
140 unsigned long flags;
141 int index;
142
143 if (!irq_iommu)
144 return -1;
145
146 raw_spin_lock_irqsave(&irq_2_ir_lock, flags);
147 *sub_handle = irq_iommu->sub_handle;
148 index = irq_iommu->irte_index;
149 raw_spin_unlock_irqrestore(&irq_2_ir_lock, flags);
150 return index;
151 }
152
153 static int set_irte_irq(int irq, struct intel_iommu *iommu, u16 index, u16 subhandle)
154 {
155 struct irq_2_iommu *irq_iommu = irq_2_iommu(irq);
156 struct irq_cfg *cfg = irq_get_chip_data(irq);
157 unsigned long flags;
158
159 if (!irq_iommu)
160 return -1;
161
162 raw_spin_lock_irqsave(&irq_2_ir_lock, flags);
163
164 cfg->remapped = 1;
165 irq_iommu->iommu = iommu;
166 irq_iommu->irte_index = index;
167 irq_iommu->sub_handle = subhandle;
168 irq_iommu->irte_mask = 0;
169
170 raw_spin_unlock_irqrestore(&irq_2_ir_lock, flags);
171
172 return 0;
173 }
174
175 static int modify_irte(int irq, struct irte *irte_modified)
176 {
177 struct irq_2_iommu *irq_iommu = irq_2_iommu(irq);
178 struct intel_iommu *iommu;
179 unsigned long flags;
180 struct irte *irte;
181 int rc, index;
182
183 if (!irq_iommu)
184 return -1;
185
186 raw_spin_lock_irqsave(&irq_2_ir_lock, flags);
187
188 iommu = irq_iommu->iommu;
189
190 index = irq_iommu->irte_index + irq_iommu->sub_handle;
191 irte = &iommu->ir_table->base[index];
192
193 set_64bit(&irte->low, irte_modified->low);
194 set_64bit(&irte->high, irte_modified->high);
195 __iommu_flush_cache(iommu, irte, sizeof(*irte));
196
197 rc = qi_flush_iec(iommu, index, 0);
198 raw_spin_unlock_irqrestore(&irq_2_ir_lock, flags);
199
200 return rc;
201 }
202
203 static struct intel_iommu *map_hpet_to_ir(u8 hpet_id)
204 {
205 int i;
206
207 for (i = 0; i < MAX_HPET_TBS; i++)
208 if (ir_hpet[i].id == hpet_id && ir_hpet[i].iommu)
209 return ir_hpet[i].iommu;
210 return NULL;
211 }
212
213 static struct intel_iommu *map_ioapic_to_ir(int apic)
214 {
215 int i;
216
217 for (i = 0; i < MAX_IO_APICS; i++)
218 if (ir_ioapic[i].id == apic && ir_ioapic[i].iommu)
219 return ir_ioapic[i].iommu;
220 return NULL;
221 }
222
223 static struct intel_iommu *map_dev_to_ir(struct pci_dev *dev)
224 {
225 struct dmar_drhd_unit *drhd;
226
227 drhd = dmar_find_matched_drhd_unit(dev);
228 if (!drhd)
229 return NULL;
230
231 return drhd->iommu;
232 }
233
234 static int clear_entries(struct irq_2_iommu *irq_iommu)
235 {
236 struct irte *start, *entry, *end;
237 struct intel_iommu *iommu;
238 int index;
239
240 if (irq_iommu->sub_handle)
241 return 0;
242
243 iommu = irq_iommu->iommu;
244 index = irq_iommu->irte_index + irq_iommu->sub_handle;
245
246 start = iommu->ir_table->base + index;
247 end = start + (1 << irq_iommu->irte_mask);
248
249 for (entry = start; entry < end; entry++) {
250 set_64bit(&entry->low, 0);
251 set_64bit(&entry->high, 0);
252 }
253 bitmap_release_region(iommu->ir_table->bitmap, index,
254 irq_iommu->irte_mask);
255
256 return qi_flush_iec(iommu, index, irq_iommu->irte_mask);
257 }
258
259 static int free_irte(int irq)
260 {
261 struct irq_2_iommu *irq_iommu = irq_2_iommu(irq);
262 unsigned long flags;
263 int rc;
264
265 if (!irq_iommu)
266 return -1;
267
268 raw_spin_lock_irqsave(&irq_2_ir_lock, flags);
269
270 rc = clear_entries(irq_iommu);
271
272 irq_iommu->iommu = NULL;
273 irq_iommu->irte_index = 0;
274 irq_iommu->sub_handle = 0;
275 irq_iommu->irte_mask = 0;
276
277 raw_spin_unlock_irqrestore(&irq_2_ir_lock, flags);
278
279 return rc;
280 }
281
282 /*
283 * source validation type
284 */
285 #define SVT_NO_VERIFY 0x0 /* no verification is required */
286 #define SVT_VERIFY_SID_SQ 0x1 /* verify using SID and SQ fields */
287 #define SVT_VERIFY_BUS 0x2 /* verify bus of request-id */
288
289 /*
290 * source-id qualifier
291 */
292 #define SQ_ALL_16 0x0 /* verify all 16 bits of request-id */
293 #define SQ_13_IGNORE_1 0x1 /* verify most significant 13 bits, ignore
294 * the third least significant bit
295 */
296 #define SQ_13_IGNORE_2 0x2 /* verify most significant 13 bits, ignore
297 * the second and third least significant bits
298 */
299 #define SQ_13_IGNORE_3 0x3 /* verify most significant 13 bits, ignore
300 * the least three significant bits
301 */
302
303 /*
304 * set SVT, SQ and SID fields of irte to verify
305 * source ids of interrupt requests
306 */
307 static void set_irte_sid(struct irte *irte, unsigned int svt,
308 unsigned int sq, unsigned int sid)
309 {
310 if (disable_sourceid_checking)
311 svt = SVT_NO_VERIFY;
312 irte->svt = svt;
313 irte->sq = sq;
314 irte->sid = sid;
315 }
316
317 static int set_ioapic_sid(struct irte *irte, int apic)
318 {
319 int i;
320 u16 sid = 0;
321
322 if (!irte)
323 return -1;
324
325 down_read(&dmar_global_lock);
326 for (i = 0; i < MAX_IO_APICS; i++) {
327 if (ir_ioapic[i].iommu && ir_ioapic[i].id == apic) {
328 sid = (ir_ioapic[i].bus << 8) | ir_ioapic[i].devfn;
329 break;
330 }
331 }
332 up_read(&dmar_global_lock);
333
334 if (sid == 0) {
335 pr_warning("Failed to set source-id of IOAPIC (%d)\n", apic);
336 return -1;
337 }
338
339 set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_ALL_16, sid);
340
341 return 0;
342 }
343
344 static int set_hpet_sid(struct irte *irte, u8 id)
345 {
346 int i;
347 u16 sid = 0;
348
349 if (!irte)
350 return -1;
351
352 down_read(&dmar_global_lock);
353 for (i = 0; i < MAX_HPET_TBS; i++) {
354 if (ir_hpet[i].iommu && ir_hpet[i].id == id) {
355 sid = (ir_hpet[i].bus << 8) | ir_hpet[i].devfn;
356 break;
357 }
358 }
359 up_read(&dmar_global_lock);
360
361 if (sid == 0) {
362 pr_warning("Failed to set source-id of HPET block (%d)\n", id);
363 return -1;
364 }
365
366 /*
367 * Should really use SQ_ALL_16. Some platforms are broken.
368 * While we figure out the right quirks for these broken platforms, use
369 * SQ_13_IGNORE_3 for now.
370 */
371 set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_13_IGNORE_3, sid);
372
373 return 0;
374 }
375
376 struct set_msi_sid_data {
377 struct pci_dev *pdev;
378 u16 alias;
379 };
380
381 static int set_msi_sid_cb(struct pci_dev *pdev, u16 alias, void *opaque)
382 {
383 struct set_msi_sid_data *data = opaque;
384
385 data->pdev = pdev;
386 data->alias = alias;
387
388 return 0;
389 }
390
391 static int set_msi_sid(struct irte *irte, struct pci_dev *dev)
392 {
393 struct set_msi_sid_data data;
394
395 if (!irte || !dev)
396 return -1;
397
398 pci_for_each_dma_alias(dev, set_msi_sid_cb, &data);
399
400 /*
401 * DMA alias provides us with a PCI device and alias. The only case
402 * where the it will return an alias on a different bus than the
403 * device is the case of a PCIe-to-PCI bridge, where the alias is for
404 * the subordinate bus. In this case we can only verify the bus.
405 *
406 * If the alias device is on a different bus than our source device
407 * then we have a topology based alias, use it.
408 *
409 * Otherwise, the alias is for a device DMA quirk and we cannot
410 * assume that MSI uses the same requester ID. Therefore use the
411 * original device.
412 */
413 if (PCI_BUS_NUM(data.alias) != data.pdev->bus->number)
414 set_irte_sid(irte, SVT_VERIFY_BUS, SQ_ALL_16,
415 PCI_DEVID(PCI_BUS_NUM(data.alias),
416 dev->bus->number));
417 else if (data.pdev->bus->number != dev->bus->number)
418 set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_ALL_16, data.alias);
419 else
420 set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_ALL_16,
421 PCI_DEVID(dev->bus->number, dev->devfn));
422
423 return 0;
424 }
425
426 static void iommu_set_irq_remapping(struct intel_iommu *iommu, int mode)
427 {
428 u64 addr;
429 u32 sts;
430 unsigned long flags;
431
432 addr = virt_to_phys((void *)iommu->ir_table->base);
433
434 raw_spin_lock_irqsave(&iommu->register_lock, flags);
435
436 dmar_writeq(iommu->reg + DMAR_IRTA_REG,
437 (addr) | IR_X2APIC_MODE(mode) | INTR_REMAP_TABLE_REG_SIZE);
438
439 /* Set interrupt-remapping table pointer */
440 writel(iommu->gcmd | DMA_GCMD_SIRTP, iommu->reg + DMAR_GCMD_REG);
441
442 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
443 readl, (sts & DMA_GSTS_IRTPS), sts);
444 raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
445
446 /*
447 * global invalidation of interrupt entry cache before enabling
448 * interrupt-remapping.
449 */
450 qi_global_iec(iommu);
451
452 raw_spin_lock_irqsave(&iommu->register_lock, flags);
453
454 /* Enable interrupt-remapping */
455 iommu->gcmd |= DMA_GCMD_IRE;
456 iommu->gcmd &= ~DMA_GCMD_CFI; /* Block compatibility-format MSIs */
457 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
458
459 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
460 readl, (sts & DMA_GSTS_IRES), sts);
461
462 /*
463 * With CFI clear in the Global Command register, we should be
464 * protected from dangerous (i.e. compatibility) interrupts
465 * regardless of x2apic status. Check just to be sure.
466 */
467 if (sts & DMA_GSTS_CFIS)
468 WARN(1, KERN_WARNING
469 "Compatibility-format IRQs enabled despite intr remapping;\n"
470 "you are vulnerable to IRQ injection.\n");
471
472 raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
473 }
474
475 static int intel_setup_irq_remapping(struct intel_iommu *iommu)
476 {
477 struct ir_table *ir_table;
478 struct page *pages;
479 unsigned long *bitmap;
480
481 if (iommu->ir_table)
482 return 0;
483
484 ir_table = kzalloc(sizeof(struct ir_table), GFP_ATOMIC);
485 if (!ir_table)
486 return -ENOMEM;
487
488 pages = alloc_pages_node(iommu->node, GFP_ATOMIC | __GFP_ZERO,
489 INTR_REMAP_PAGE_ORDER);
490
491 if (!pages) {
492 pr_err("IR%d: failed to allocate pages of order %d\n",
493 iommu->seq_id, INTR_REMAP_PAGE_ORDER);
494 goto out_free_table;
495 }
496
497 bitmap = kcalloc(BITS_TO_LONGS(INTR_REMAP_TABLE_ENTRIES),
498 sizeof(long), GFP_ATOMIC);
499 if (bitmap == NULL) {
500 pr_err("IR%d: failed to allocate bitmap\n", iommu->seq_id);
501 goto out_free_pages;
502 }
503
504 ir_table->base = page_address(pages);
505 ir_table->bitmap = bitmap;
506 iommu->ir_table = ir_table;
507 return 0;
508
509 out_free_pages:
510 __free_pages(pages, INTR_REMAP_PAGE_ORDER);
511 out_free_table:
512 kfree(ir_table);
513 return -ENOMEM;
514 }
515
516 static void intel_teardown_irq_remapping(struct intel_iommu *iommu)
517 {
518 if (iommu && iommu->ir_table) {
519 free_pages((unsigned long)iommu->ir_table->base,
520 INTR_REMAP_PAGE_ORDER);
521 kfree(iommu->ir_table->bitmap);
522 kfree(iommu->ir_table);
523 iommu->ir_table = NULL;
524 }
525 }
526
527 /*
528 * Disable Interrupt Remapping.
529 */
530 static void iommu_disable_irq_remapping(struct intel_iommu *iommu)
531 {
532 unsigned long flags;
533 u32 sts;
534
535 if (!ecap_ir_support(iommu->ecap))
536 return;
537
538 /*
539 * global invalidation of interrupt entry cache before disabling
540 * interrupt-remapping.
541 */
542 qi_global_iec(iommu);
543
544 raw_spin_lock_irqsave(&iommu->register_lock, flags);
545
546 sts = dmar_readq(iommu->reg + DMAR_GSTS_REG);
547 if (!(sts & DMA_GSTS_IRES))
548 goto end;
549
550 iommu->gcmd &= ~DMA_GCMD_IRE;
551 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
552
553 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
554 readl, !(sts & DMA_GSTS_IRES), sts);
555
556 end:
557 raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
558 }
559
560 static int __init dmar_x2apic_optout(void)
561 {
562 struct acpi_table_dmar *dmar;
563 dmar = (struct acpi_table_dmar *)dmar_tbl;
564 if (!dmar || no_x2apic_optout)
565 return 0;
566 return dmar->flags & DMAR_X2APIC_OPT_OUT;
567 }
568
569 static int __init intel_irq_remapping_supported(void)
570 {
571 struct dmar_drhd_unit *drhd;
572 struct intel_iommu *iommu;
573
574 if (disable_irq_remap)
575 return 0;
576 if (irq_remap_broken) {
577 printk(KERN_WARNING
578 "This system BIOS has enabled interrupt remapping\n"
579 "on a chipset that contains an erratum making that\n"
580 "feature unstable. To maintain system stability\n"
581 "interrupt remapping is being disabled. Please\n"
582 "contact your BIOS vendor for an update\n");
583 add_taint(TAINT_FIRMWARE_WORKAROUND, LOCKDEP_STILL_OK);
584 disable_irq_remap = 1;
585 return 0;
586 }
587
588 if (!dmar_ir_support())
589 return 0;
590
591 for_each_iommu(iommu, drhd)
592 if (!ecap_ir_support(iommu->ecap))
593 return 0;
594
595 return 1;
596 }
597
598 static int __init intel_enable_irq_remapping(void)
599 {
600 struct dmar_drhd_unit *drhd;
601 struct intel_iommu *iommu;
602 bool x2apic_present;
603 int setup = 0;
604 int eim = 0;
605
606 x2apic_present = x2apic_supported();
607
608 if (parse_ioapics_under_ir() != 1) {
609 printk(KERN_INFO "Not enable interrupt remapping\n");
610 goto error;
611 }
612
613 if (x2apic_present) {
614 pr_info("Queued invalidation will be enabled to support x2apic and Intr-remapping.\n");
615
616 eim = !dmar_x2apic_optout();
617 if (!eim)
618 printk(KERN_WARNING
619 "Your BIOS is broken and requested that x2apic be disabled.\n"
620 "This will slightly decrease performance.\n"
621 "Use 'intremap=no_x2apic_optout' to override BIOS request.\n");
622 }
623
624 for_each_iommu(iommu, drhd) {
625 /*
626 * If the queued invalidation is already initialized,
627 * shouldn't disable it.
628 */
629 if (iommu->qi)
630 continue;
631
632 /*
633 * Clear previous faults.
634 */
635 dmar_fault(-1, iommu);
636
637 /*
638 * Disable intr remapping and queued invalidation, if already
639 * enabled prior to OS handover.
640 */
641 iommu_disable_irq_remapping(iommu);
642
643 dmar_disable_qi(iommu);
644 }
645
646 /*
647 * check for the Interrupt-remapping support
648 */
649 for_each_iommu(iommu, drhd) {
650 if (!ecap_ir_support(iommu->ecap))
651 continue;
652
653 if (eim && !ecap_eim_support(iommu->ecap)) {
654 printk(KERN_INFO "DRHD %Lx: EIM not supported by DRHD, "
655 " ecap %Lx\n", drhd->reg_base_addr, iommu->ecap);
656 goto error;
657 }
658 }
659
660 /*
661 * Enable queued invalidation for all the DRHD's.
662 */
663 for_each_iommu(iommu, drhd) {
664 int ret = dmar_enable_qi(iommu);
665
666 if (ret) {
667 printk(KERN_ERR "DRHD %Lx: failed to enable queued, "
668 " invalidation, ecap %Lx, ret %d\n",
669 drhd->reg_base_addr, iommu->ecap, ret);
670 goto error;
671 }
672 }
673
674 /*
675 * Setup Interrupt-remapping for all the DRHD's now.
676 */
677 for_each_iommu(iommu, drhd) {
678 if (!ecap_ir_support(iommu->ecap))
679 continue;
680
681 if (intel_setup_irq_remapping(iommu))
682 goto error;
683
684 iommu_set_irq_remapping(iommu, eim);
685 setup = 1;
686 }
687
688 if (!setup)
689 goto error;
690
691 irq_remapping_enabled = 1;
692
693 /*
694 * VT-d has a different layout for IO-APIC entries when
695 * interrupt remapping is enabled. So it needs a special routine
696 * to print IO-APIC entries for debugging purposes too.
697 */
698 x86_io_apic_ops.print_entries = intel_ir_io_apic_print_entries;
699
700 pr_info("Enabled IRQ remapping in %s mode\n", eim ? "x2apic" : "xapic");
701
702 return eim ? IRQ_REMAP_X2APIC_MODE : IRQ_REMAP_XAPIC_MODE;
703
704 error:
705 /*
706 * handle error condition gracefully here!
707 */
708
709 if (x2apic_present)
710 pr_warn("Failed to enable irq remapping. You are vulnerable to irq-injection attacks.\n");
711
712 return -1;
713 }
714
715 static int ir_parse_one_hpet_scope(struct acpi_dmar_device_scope *scope,
716 struct intel_iommu *iommu,
717 struct acpi_dmar_hardware_unit *drhd)
718 {
719 struct acpi_dmar_pci_path *path;
720 u8 bus;
721 int count, free = -1;
722
723 bus = scope->bus;
724 path = (struct acpi_dmar_pci_path *)(scope + 1);
725 count = (scope->length - sizeof(struct acpi_dmar_device_scope))
726 / sizeof(struct acpi_dmar_pci_path);
727
728 while (--count > 0) {
729 /*
730 * Access PCI directly due to the PCI
731 * subsystem isn't initialized yet.
732 */
733 bus = read_pci_config_byte(bus, path->device, path->function,
734 PCI_SECONDARY_BUS);
735 path++;
736 }
737
738 for (count = 0; count < MAX_HPET_TBS; count++) {
739 if (ir_hpet[count].iommu == iommu &&
740 ir_hpet[count].id == scope->enumeration_id)
741 return 0;
742 else if (ir_hpet[count].iommu == NULL && free == -1)
743 free = count;
744 }
745 if (free == -1) {
746 pr_warn("Exceeded Max HPET blocks\n");
747 return -ENOSPC;
748 }
749
750 ir_hpet[free].iommu = iommu;
751 ir_hpet[free].id = scope->enumeration_id;
752 ir_hpet[free].bus = bus;
753 ir_hpet[free].devfn = PCI_DEVFN(path->device, path->function);
754 pr_info("HPET id %d under DRHD base 0x%Lx\n",
755 scope->enumeration_id, drhd->address);
756
757 return 0;
758 }
759
760 static int ir_parse_one_ioapic_scope(struct acpi_dmar_device_scope *scope,
761 struct intel_iommu *iommu,
762 struct acpi_dmar_hardware_unit *drhd)
763 {
764 struct acpi_dmar_pci_path *path;
765 u8 bus;
766 int count, free = -1;
767
768 bus = scope->bus;
769 path = (struct acpi_dmar_pci_path *)(scope + 1);
770 count = (scope->length - sizeof(struct acpi_dmar_device_scope))
771 / sizeof(struct acpi_dmar_pci_path);
772
773 while (--count > 0) {
774 /*
775 * Access PCI directly due to the PCI
776 * subsystem isn't initialized yet.
777 */
778 bus = read_pci_config_byte(bus, path->device, path->function,
779 PCI_SECONDARY_BUS);
780 path++;
781 }
782
783 for (count = 0; count < MAX_IO_APICS; count++) {
784 if (ir_ioapic[count].iommu == iommu &&
785 ir_ioapic[count].id == scope->enumeration_id)
786 return 0;
787 else if (ir_ioapic[count].iommu == NULL && free == -1)
788 free = count;
789 }
790 if (free == -1) {
791 pr_warn("Exceeded Max IO APICS\n");
792 return -ENOSPC;
793 }
794
795 ir_ioapic[free].bus = bus;
796 ir_ioapic[free].devfn = PCI_DEVFN(path->device, path->function);
797 ir_ioapic[free].iommu = iommu;
798 ir_ioapic[free].id = scope->enumeration_id;
799 pr_info("IOAPIC id %d under DRHD base 0x%Lx IOMMU %d\n",
800 scope->enumeration_id, drhd->address, iommu->seq_id);
801
802 return 0;
803 }
804
805 static int ir_parse_ioapic_hpet_scope(struct acpi_dmar_header *header,
806 struct intel_iommu *iommu)
807 {
808 int ret = 0;
809 struct acpi_dmar_hardware_unit *drhd;
810 struct acpi_dmar_device_scope *scope;
811 void *start, *end;
812
813 drhd = (struct acpi_dmar_hardware_unit *)header;
814 start = (void *)(drhd + 1);
815 end = ((void *)drhd) + header->length;
816
817 while (start < end && ret == 0) {
818 scope = start;
819 if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_IOAPIC)
820 ret = ir_parse_one_ioapic_scope(scope, iommu, drhd);
821 else if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_HPET)
822 ret = ir_parse_one_hpet_scope(scope, iommu, drhd);
823 start += scope->length;
824 }
825
826 return ret;
827 }
828
829 static void ir_remove_ioapic_hpet_scope(struct intel_iommu *iommu)
830 {
831 int i;
832
833 for (i = 0; i < MAX_HPET_TBS; i++)
834 if (ir_hpet[i].iommu == iommu)
835 ir_hpet[i].iommu = NULL;
836
837 for (i = 0; i < MAX_IO_APICS; i++)
838 if (ir_ioapic[i].iommu == iommu)
839 ir_ioapic[i].iommu = NULL;
840 }
841
842 /*
843 * Finds the assocaition between IOAPIC's and its Interrupt-remapping
844 * hardware unit.
845 */
846 static int __init parse_ioapics_under_ir(void)
847 {
848 struct dmar_drhd_unit *drhd;
849 struct intel_iommu *iommu;
850 int ir_supported = 0;
851 int ioapic_idx;
852
853 for_each_iommu(iommu, drhd)
854 if (ecap_ir_support(iommu->ecap)) {
855 if (ir_parse_ioapic_hpet_scope(drhd->hdr, iommu))
856 return -1;
857
858 ir_supported = 1;
859 }
860
861 if (!ir_supported)
862 return 0;
863
864 for (ioapic_idx = 0; ioapic_idx < nr_ioapics; ioapic_idx++) {
865 int ioapic_id = mpc_ioapic_id(ioapic_idx);
866 if (!map_ioapic_to_ir(ioapic_id)) {
867 pr_err(FW_BUG "ioapic %d has no mapping iommu, "
868 "interrupt remapping will be disabled\n",
869 ioapic_id);
870 return -1;
871 }
872 }
873
874 return 1;
875 }
876
877 static int __init ir_dev_scope_init(void)
878 {
879 int ret;
880
881 if (!irq_remapping_enabled)
882 return 0;
883
884 down_write(&dmar_global_lock);
885 ret = dmar_dev_scope_init();
886 up_write(&dmar_global_lock);
887
888 return ret;
889 }
890 rootfs_initcall(ir_dev_scope_init);
891
892 static void disable_irq_remapping(void)
893 {
894 struct dmar_drhd_unit *drhd;
895 struct intel_iommu *iommu = NULL;
896
897 /*
898 * Disable Interrupt-remapping for all the DRHD's now.
899 */
900 for_each_iommu(iommu, drhd) {
901 if (!ecap_ir_support(iommu->ecap))
902 continue;
903
904 iommu_disable_irq_remapping(iommu);
905 }
906 }
907
908 static int reenable_irq_remapping(int eim)
909 {
910 struct dmar_drhd_unit *drhd;
911 int setup = 0;
912 struct intel_iommu *iommu = NULL;
913
914 for_each_iommu(iommu, drhd)
915 if (iommu->qi)
916 dmar_reenable_qi(iommu);
917
918 /*
919 * Setup Interrupt-remapping for all the DRHD's now.
920 */
921 for_each_iommu(iommu, drhd) {
922 if (!ecap_ir_support(iommu->ecap))
923 continue;
924
925 /* Set up interrupt remapping for iommu.*/
926 iommu_set_irq_remapping(iommu, eim);
927 setup = 1;
928 }
929
930 if (!setup)
931 goto error;
932
933 return 0;
934
935 error:
936 /*
937 * handle error condition gracefully here!
938 */
939 return -1;
940 }
941
942 static void prepare_irte(struct irte *irte, int vector,
943 unsigned int dest)
944 {
945 memset(irte, 0, sizeof(*irte));
946
947 irte->present = 1;
948 irte->dst_mode = apic->irq_dest_mode;
949 /*
950 * Trigger mode in the IRTE will always be edge, and for IO-APIC, the
951 * actual level or edge trigger will be setup in the IO-APIC
952 * RTE. This will help simplify level triggered irq migration.
953 * For more details, see the comments (in io_apic.c) explainig IO-APIC
954 * irq migration in the presence of interrupt-remapping.
955 */
956 irte->trigger_mode = 0;
957 irte->dlvry_mode = apic->irq_delivery_mode;
958 irte->vector = vector;
959 irte->dest_id = IRTE_DEST(dest);
960 irte->redir_hint = 1;
961 }
962
963 static int intel_setup_ioapic_entry(int irq,
964 struct IO_APIC_route_entry *route_entry,
965 unsigned int destination, int vector,
966 struct io_apic_irq_attr *attr)
967 {
968 int ioapic_id = mpc_ioapic_id(attr->ioapic);
969 struct intel_iommu *iommu;
970 struct IR_IO_APIC_route_entry *entry;
971 struct irte irte;
972 int index;
973
974 down_read(&dmar_global_lock);
975 iommu = map_ioapic_to_ir(ioapic_id);
976 if (!iommu) {
977 pr_warn("No mapping iommu for ioapic %d\n", ioapic_id);
978 index = -ENODEV;
979 } else {
980 index = alloc_irte(iommu, irq, 1);
981 if (index < 0) {
982 pr_warn("Failed to allocate IRTE for ioapic %d\n",
983 ioapic_id);
984 index = -ENOMEM;
985 }
986 }
987 up_read(&dmar_global_lock);
988 if (index < 0)
989 return index;
990
991 prepare_irte(&irte, vector, destination);
992
993 /* Set source-id of interrupt request */
994 set_ioapic_sid(&irte, ioapic_id);
995
996 modify_irte(irq, &irte);
997
998 apic_printk(APIC_VERBOSE, KERN_DEBUG "IOAPIC[%d]: "
999 "Set IRTE entry (P:%d FPD:%d Dst_Mode:%d "
1000 "Redir_hint:%d Trig_Mode:%d Dlvry_Mode:%X "
1001 "Avail:%X Vector:%02X Dest:%08X "
1002 "SID:%04X SQ:%X SVT:%X)\n",
1003 attr->ioapic, irte.present, irte.fpd, irte.dst_mode,
1004 irte.redir_hint, irte.trigger_mode, irte.dlvry_mode,
1005 irte.avail, irte.vector, irte.dest_id,
1006 irte.sid, irte.sq, irte.svt);
1007
1008 entry = (struct IR_IO_APIC_route_entry *)route_entry;
1009 memset(entry, 0, sizeof(*entry));
1010
1011 entry->index2 = (index >> 15) & 0x1;
1012 entry->zero = 0;
1013 entry->format = 1;
1014 entry->index = (index & 0x7fff);
1015 /*
1016 * IO-APIC RTE will be configured with virtual vector.
1017 * irq handler will do the explicit EOI to the io-apic.
1018 */
1019 entry->vector = attr->ioapic_pin;
1020 entry->mask = 0; /* enable IRQ */
1021 entry->trigger = attr->trigger;
1022 entry->polarity = attr->polarity;
1023
1024 /* Mask level triggered irqs.
1025 * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
1026 */
1027 if (attr->trigger)
1028 entry->mask = 1;
1029
1030 return 0;
1031 }
1032
1033 /*
1034 * Migrate the IO-APIC irq in the presence of intr-remapping.
1035 *
1036 * For both level and edge triggered, irq migration is a simple atomic
1037 * update(of vector and cpu destination) of IRTE and flush the hardware cache.
1038 *
1039 * For level triggered, we eliminate the io-apic RTE modification (with the
1040 * updated vector information), by using a virtual vector (io-apic pin number).
1041 * Real vector that is used for interrupting cpu will be coming from
1042 * the interrupt-remapping table entry.
1043 *
1044 * As the migration is a simple atomic update of IRTE, the same mechanism
1045 * is used to migrate MSI irq's in the presence of interrupt-remapping.
1046 */
1047 static int
1048 intel_ioapic_set_affinity(struct irq_data *data, const struct cpumask *mask,
1049 bool force)
1050 {
1051 struct irq_cfg *cfg = data->chip_data;
1052 unsigned int dest, irq = data->irq;
1053 struct irte irte;
1054 int err;
1055
1056 if (!config_enabled(CONFIG_SMP))
1057 return -EINVAL;
1058
1059 if (!cpumask_intersects(mask, cpu_online_mask))
1060 return -EINVAL;
1061
1062 if (get_irte(irq, &irte))
1063 return -EBUSY;
1064
1065 err = assign_irq_vector(irq, cfg, mask);
1066 if (err)
1067 return err;
1068
1069 err = apic->cpu_mask_to_apicid_and(cfg->domain, mask, &dest);
1070 if (err) {
1071 if (assign_irq_vector(irq, cfg, data->affinity))
1072 pr_err("Failed to recover vector for irq %d\n", irq);
1073 return err;
1074 }
1075
1076 irte.vector = cfg->vector;
1077 irte.dest_id = IRTE_DEST(dest);
1078
1079 /*
1080 * Atomically updates the IRTE with the new destination, vector
1081 * and flushes the interrupt entry cache.
1082 */
1083 modify_irte(irq, &irte);
1084
1085 /*
1086 * After this point, all the interrupts will start arriving
1087 * at the new destination. So, time to cleanup the previous
1088 * vector allocation.
1089 */
1090 if (cfg->move_in_progress)
1091 send_cleanup_vector(cfg);
1092
1093 cpumask_copy(data->affinity, mask);
1094 return 0;
1095 }
1096
1097 static void intel_compose_msi_msg(struct pci_dev *pdev,
1098 unsigned int irq, unsigned int dest,
1099 struct msi_msg *msg, u8 hpet_id)
1100 {
1101 struct irq_cfg *cfg;
1102 struct irte irte;
1103 u16 sub_handle = 0;
1104 int ir_index;
1105
1106 cfg = irq_get_chip_data(irq);
1107
1108 ir_index = map_irq_to_irte_handle(irq, &sub_handle);
1109 BUG_ON(ir_index == -1);
1110
1111 prepare_irte(&irte, cfg->vector, dest);
1112
1113 /* Set source-id of interrupt request */
1114 if (pdev)
1115 set_msi_sid(&irte, pdev);
1116 else
1117 set_hpet_sid(&irte, hpet_id);
1118
1119 modify_irte(irq, &irte);
1120
1121 msg->address_hi = MSI_ADDR_BASE_HI;
1122 msg->data = sub_handle;
1123 msg->address_lo = MSI_ADDR_BASE_LO | MSI_ADDR_IR_EXT_INT |
1124 MSI_ADDR_IR_SHV |
1125 MSI_ADDR_IR_INDEX1(ir_index) |
1126 MSI_ADDR_IR_INDEX2(ir_index);
1127 }
1128
1129 /*
1130 * Map the PCI dev to the corresponding remapping hardware unit
1131 * and allocate 'nvec' consecutive interrupt-remapping table entries
1132 * in it.
1133 */
1134 static int intel_msi_alloc_irq(struct pci_dev *dev, int irq, int nvec)
1135 {
1136 struct intel_iommu *iommu;
1137 int index;
1138
1139 down_read(&dmar_global_lock);
1140 iommu = map_dev_to_ir(dev);
1141 if (!iommu) {
1142 printk(KERN_ERR
1143 "Unable to map PCI %s to iommu\n", pci_name(dev));
1144 index = -ENOENT;
1145 } else {
1146 index = alloc_irte(iommu, irq, nvec);
1147 if (index < 0) {
1148 printk(KERN_ERR
1149 "Unable to allocate %d IRTE for PCI %s\n",
1150 nvec, pci_name(dev));
1151 index = -ENOSPC;
1152 }
1153 }
1154 up_read(&dmar_global_lock);
1155
1156 return index;
1157 }
1158
1159 static int intel_msi_setup_irq(struct pci_dev *pdev, unsigned int irq,
1160 int index, int sub_handle)
1161 {
1162 struct intel_iommu *iommu;
1163 int ret = -ENOENT;
1164
1165 down_read(&dmar_global_lock);
1166 iommu = map_dev_to_ir(pdev);
1167 if (iommu) {
1168 /*
1169 * setup the mapping between the irq and the IRTE
1170 * base index, the sub_handle pointing to the
1171 * appropriate interrupt remap table entry.
1172 */
1173 set_irte_irq(irq, iommu, index, sub_handle);
1174 ret = 0;
1175 }
1176 up_read(&dmar_global_lock);
1177
1178 return ret;
1179 }
1180
1181 static int intel_alloc_hpet_msi(unsigned int irq, unsigned int id)
1182 {
1183 int ret = -1;
1184 struct intel_iommu *iommu;
1185 int index;
1186
1187 down_read(&dmar_global_lock);
1188 iommu = map_hpet_to_ir(id);
1189 if (iommu) {
1190 index = alloc_irte(iommu, irq, 1);
1191 if (index >= 0)
1192 ret = 0;
1193 }
1194 up_read(&dmar_global_lock);
1195
1196 return ret;
1197 }
1198
1199 struct irq_remap_ops intel_irq_remap_ops = {
1200 .supported = intel_irq_remapping_supported,
1201 .prepare = dmar_table_init,
1202 .enable = intel_enable_irq_remapping,
1203 .disable = disable_irq_remapping,
1204 .reenable = reenable_irq_remapping,
1205 .enable_faulting = enable_drhd_fault_handling,
1206 .setup_ioapic_entry = intel_setup_ioapic_entry,
1207 .set_affinity = intel_ioapic_set_affinity,
1208 .free_irq = free_irte,
1209 .compose_msi_msg = intel_compose_msi_msg,
1210 .msi_alloc_irq = intel_msi_alloc_irq,
1211 .msi_setup_irq = intel_msi_setup_irq,
1212 .alloc_hpet_msi = intel_alloc_hpet_msi,
1213 };
1214
1215 /*
1216 * Support of Interrupt Remapping Unit Hotplug
1217 */
1218 static int dmar_ir_add(struct dmar_drhd_unit *dmaru, struct intel_iommu *iommu)
1219 {
1220 int ret;
1221 int eim = x2apic_enabled();
1222
1223 if (eim && !ecap_eim_support(iommu->ecap)) {
1224 pr_info("DRHD %Lx: EIM not supported by DRHD, ecap %Lx\n",
1225 iommu->reg_phys, iommu->ecap);
1226 return -ENODEV;
1227 }
1228
1229 if (ir_parse_ioapic_hpet_scope(dmaru->hdr, iommu)) {
1230 pr_warn("DRHD %Lx: failed to parse managed IOAPIC/HPET\n",
1231 iommu->reg_phys);
1232 return -ENODEV;
1233 }
1234
1235 /* TODO: check all IOAPICs are covered by IOMMU */
1236
1237 /* Setup Interrupt-remapping now. */
1238 ret = intel_setup_irq_remapping(iommu);
1239 if (ret) {
1240 pr_err("DRHD %Lx: failed to allocate resource\n",
1241 iommu->reg_phys);
1242 ir_remove_ioapic_hpet_scope(iommu);
1243 return ret;
1244 }
1245
1246 if (!iommu->qi) {
1247 /* Clear previous faults. */
1248 dmar_fault(-1, iommu);
1249 iommu_disable_irq_remapping(iommu);
1250 dmar_disable_qi(iommu);
1251 }
1252
1253 /* Enable queued invalidation */
1254 ret = dmar_enable_qi(iommu);
1255 if (!ret) {
1256 iommu_set_irq_remapping(iommu, eim);
1257 } else {
1258 pr_err("DRHD %Lx: failed to enable queued invalidation, ecap %Lx, ret %d\n",
1259 iommu->reg_phys, iommu->ecap, ret);
1260 intel_teardown_irq_remapping(iommu);
1261 ir_remove_ioapic_hpet_scope(iommu);
1262 }
1263
1264 return ret;
1265 }
1266
1267 int dmar_ir_hotplug(struct dmar_drhd_unit *dmaru, bool insert)
1268 {
1269 int ret = 0;
1270 struct intel_iommu *iommu = dmaru->iommu;
1271
1272 if (!irq_remapping_enabled)
1273 return 0;
1274 if (iommu == NULL)
1275 return -EINVAL;
1276 if (!ecap_ir_support(iommu->ecap))
1277 return 0;
1278
1279 if (insert) {
1280 if (!iommu->ir_table)
1281 ret = dmar_ir_add(dmaru, iommu);
1282 } else {
1283 if (iommu->ir_table) {
1284 if (!bitmap_empty(iommu->ir_table->bitmap,
1285 INTR_REMAP_TABLE_ENTRIES)) {
1286 ret = -EBUSY;
1287 } else {
1288 iommu_disable_irq_remapping(iommu);
1289 intel_teardown_irq_remapping(iommu);
1290 ir_remove_ioapic_hpet_scope(iommu);
1291 }
1292 }
1293 }
1294
1295 return ret;
1296 }
Linux kernel - Deliverables
This page took 0.121022 seconds and 5 git commands to generate.