aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/pci
diff options
context:
space:
mode:
authorBenjamin Herrenschmidt <benh@kernel.crashing.org>2008-10-14 20:31:54 -0400
committerBenjamin Herrenschmidt <benh@kernel.crashing.org>2008-10-14 20:31:54 -0400
commit6dc6472581f693b5fc95aebedf67b4960fb85cf0 (patch)
tree06a5a9a08519950575505273eabced331ed51405 /drivers/pci
parentee673eaa72d8d185012b1027a05e25aba18c267f (diff)
parent8acd3a60bcca17c6d89c73cee3ad6057eb83ba1e (diff)
Merge commit 'origin'
Manual fixup of conflicts on: arch/powerpc/include/asm/dcr-regs.h drivers/net/ibm_newemac/core.h
Diffstat (limited to 'drivers/pci')
-rw-r--r--drivers/pci/Makefile2
-rw-r--r--drivers/pci/dma_remapping.h157
-rw-r--r--drivers/pci/dmar.c397
-rw-r--r--drivers/pci/hotplug/fakephp.c6
-rw-r--r--drivers/pci/hotplug/pciehp_hpc.c2
-rw-r--r--drivers/pci/intel-iommu.c164
-rw-r--r--drivers/pci/intel-iommu.h233
-rw-r--r--drivers/pci/intr_remapping.c471
-rw-r--r--drivers/pci/intr_remapping.h8
-rw-r--r--drivers/pci/pci-sysfs.c19
-rw-r--r--drivers/pci/pcie/aspm.c2
-rw-r--r--drivers/pci/probe.c18
-rw-r--r--drivers/pci/search.c6
-rw-r--r--drivers/pci/setup-bus.c6
14 files changed, 1187 insertions, 304 deletions
diff --git a/drivers/pci/Makefile b/drivers/pci/Makefile
index 7d63f8ced24b..4b47f4ece5b7 100644
--- a/drivers/pci/Makefile
+++ b/drivers/pci/Makefile
@@ -26,6 +26,8 @@ obj-$(CONFIG_HT_IRQ) += htirq.o
26# Build Intel IOMMU support 26# Build Intel IOMMU support
27obj-$(CONFIG_DMAR) += dmar.o iova.o intel-iommu.o 27obj-$(CONFIG_DMAR) += dmar.o iova.o intel-iommu.o
28 28
29obj-$(CONFIG_INTR_REMAP) += dmar.o intr_remapping.o
30
29# 31#
30# Some architectures use the generic PCI setup functions 32# Some architectures use the generic PCI setup functions
31# 33#
diff --git a/drivers/pci/dma_remapping.h b/drivers/pci/dma_remapping.h
new file mode 100644
index 000000000000..bff5c65f81dc
--- /dev/null
+++ b/drivers/pci/dma_remapping.h
@@ -0,0 +1,157 @@
1#ifndef _DMA_REMAPPING_H
2#define _DMA_REMAPPING_H
3
4/*
5 * We need a fixed PAGE_SIZE of 4K irrespective of
6 * arch PAGE_SIZE for IOMMU page tables.
7 */
8#define PAGE_SHIFT_4K (12)
9#define PAGE_SIZE_4K (1UL << PAGE_SHIFT_4K)
10#define PAGE_MASK_4K (((u64)-1) << PAGE_SHIFT_4K)
11#define PAGE_ALIGN_4K(addr) (((addr) + PAGE_SIZE_4K - 1) & PAGE_MASK_4K)
12
13#define IOVA_PFN(addr) ((addr) >> PAGE_SHIFT_4K)
14#define DMA_32BIT_PFN IOVA_PFN(DMA_32BIT_MASK)
15#define DMA_64BIT_PFN IOVA_PFN(DMA_64BIT_MASK)
16
17
18/*
19 * 0: Present
20 * 1-11: Reserved
21 * 12-63: Context Ptr (12 - (haw-1))
22 * 64-127: Reserved
23 */
24struct root_entry {
25 u64 val;
26 u64 rsvd1;
27};
28#define ROOT_ENTRY_NR (PAGE_SIZE_4K/sizeof(struct root_entry))
29static inline bool root_present(struct root_entry *root)
30{
31 return (root->val & 1);
32}
33static inline void set_root_present(struct root_entry *root)
34{
35 root->val |= 1;
36}
37static inline void set_root_value(struct root_entry *root, unsigned long value)
38{
39 root->val |= value & PAGE_MASK_4K;
40}
41
42struct context_entry;
43static inline struct context_entry *
44get_context_addr_from_root(struct root_entry *root)
45{
46 return (struct context_entry *)
47 (root_present(root)?phys_to_virt(
48 root->val & PAGE_MASK_4K):
49 NULL);
50}
51
52/*
53 * low 64 bits:
54 * 0: present
55 * 1: fault processing disable
56 * 2-3: translation type
57 * 12-63: address space root
58 * high 64 bits:
59 * 0-2: address width
60 * 3-6: aval
61 * 8-23: domain id
62 */
63struct context_entry {
64 u64 lo;
65 u64 hi;
66};
67#define context_present(c) ((c).lo & 1)
68#define context_fault_disable(c) (((c).lo >> 1) & 1)
69#define context_translation_type(c) (((c).lo >> 2) & 3)
70#define context_address_root(c) ((c).lo & PAGE_MASK_4K)
71#define context_address_width(c) ((c).hi & 7)
72#define context_domain_id(c) (((c).hi >> 8) & ((1 << 16) - 1))
73
74#define context_set_present(c) do {(c).lo |= 1;} while (0)
75#define context_set_fault_enable(c) \
76 do {(c).lo &= (((u64)-1) << 2) | 1;} while (0)
77#define context_set_translation_type(c, val) \
78 do { \
79 (c).lo &= (((u64)-1) << 4) | 3; \
80 (c).lo |= ((val) & 3) << 2; \
81 } while (0)
82#define CONTEXT_TT_MULTI_LEVEL 0
83#define context_set_address_root(c, val) \
84 do {(c).lo |= (val) & PAGE_MASK_4K;} while (0)
85#define context_set_address_width(c, val) do {(c).hi |= (val) & 7;} while (0)
86#define context_set_domain_id(c, val) \
87 do {(c).hi |= ((val) & ((1 << 16) - 1)) << 8;} while (0)
88#define context_clear_entry(c) do {(c).lo = 0; (c).hi = 0;} while (0)
89
90/*
91 * 0: readable
92 * 1: writable
93 * 2-6: reserved
94 * 7: super page
95 * 8-11: available
96 * 12-63: Host physcial address
97 */
98struct dma_pte {
99 u64 val;
100};
101#define dma_clear_pte(p) do {(p).val = 0;} while (0)
102
103#define DMA_PTE_READ (1)
104#define DMA_PTE_WRITE (2)
105
106#define dma_set_pte_readable(p) do {(p).val |= DMA_PTE_READ;} while (0)
107#define dma_set_pte_writable(p) do {(p).val |= DMA_PTE_WRITE;} while (0)
108#define dma_set_pte_prot(p, prot) \
109 do {(p).val = ((p).val & ~3) | ((prot) & 3); } while (0)
110#define dma_pte_addr(p) ((p).val & PAGE_MASK_4K)
111#define dma_set_pte_addr(p, addr) do {\
112 (p).val |= ((addr) & PAGE_MASK_4K); } while (0)
113#define dma_pte_present(p) (((p).val & 3) != 0)
114
115struct intel_iommu;
116
117struct dmar_domain {
118 int id; /* domain id */
119 struct intel_iommu *iommu; /* back pointer to owning iommu */
120
121 struct list_head devices; /* all devices' list */
122 struct iova_domain iovad; /* iova's that belong to this domain */
123
124 struct dma_pte *pgd; /* virtual address */
125 spinlock_t mapping_lock; /* page table lock */
126 int gaw; /* max guest address width */
127
128 /* adjusted guest address width, 0 is level 2 30-bit */
129 int agaw;
130
131#define DOMAIN_FLAG_MULTIPLE_DEVICES 1
132 int flags;
133};
134
135/* PCI domain-device relationship */
136struct device_domain_info {
137 struct list_head link; /* link to domain siblings */
138 struct list_head global; /* link to global list */
139 u8 bus; /* PCI bus numer */
140 u8 devfn; /* PCI devfn number */
141 struct pci_dev *dev; /* it's NULL for PCIE-to-PCI bridge */
142 struct dmar_domain *domain; /* pointer to domain */
143};
144
145extern int init_dmars(void);
146extern void free_dmar_iommu(struct intel_iommu *iommu);
147
148extern int dmar_disabled;
149
150#ifndef CONFIG_DMAR_GFX_WA
151static inline void iommu_prepare_gfx_mapping(void)
152{
153 return;
154}
155#endif /* !CONFIG_DMAR_GFX_WA */
156
157#endif
diff --git a/drivers/pci/dmar.c b/drivers/pci/dmar.c
index 8bf86ae2333f..bd2c01674f5e 100644
--- a/drivers/pci/dmar.c
+++ b/drivers/pci/dmar.c
@@ -19,13 +19,16 @@
19 * Author: Shaohua Li <shaohua.li@intel.com> 19 * Author: Shaohua Li <shaohua.li@intel.com>
20 * Author: Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com> 20 * Author: Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
21 * 21 *
22 * This file implements early detection/parsing of DMA Remapping Devices 22 * This file implements early detection/parsing of Remapping Devices
23 * reported to OS through BIOS via DMA remapping reporting (DMAR) ACPI 23 * reported to OS through BIOS via DMA remapping reporting (DMAR) ACPI
24 * tables. 24 * tables.
25 *
26 * These routines are used by both DMA-remapping and Interrupt-remapping
25 */ 27 */
26 28
27#include <linux/pci.h> 29#include <linux/pci.h>
28#include <linux/dmar.h> 30#include <linux/dmar.h>
31#include <linux/timer.h>
29#include "iova.h" 32#include "iova.h"
30#include "intel-iommu.h" 33#include "intel-iommu.h"
31 34
@@ -37,7 +40,6 @@
37 * these units are not supported by the architecture. 40 * these units are not supported by the architecture.
38 */ 41 */
39LIST_HEAD(dmar_drhd_units); 42LIST_HEAD(dmar_drhd_units);
40LIST_HEAD(dmar_rmrr_units);
41 43
42static struct acpi_table_header * __initdata dmar_tbl; 44static struct acpi_table_header * __initdata dmar_tbl;
43 45
@@ -53,11 +55,6 @@ static void __init dmar_register_drhd_unit(struct dmar_drhd_unit *drhd)
53 list_add(&drhd->list, &dmar_drhd_units); 55 list_add(&drhd->list, &dmar_drhd_units);
54} 56}
55 57
56static void __init dmar_register_rmrr_unit(struct dmar_rmrr_unit *rmrr)
57{
58 list_add(&rmrr->list, &dmar_rmrr_units);
59}
60
61static int __init dmar_parse_one_dev_scope(struct acpi_dmar_device_scope *scope, 58static int __init dmar_parse_one_dev_scope(struct acpi_dmar_device_scope *scope,
62 struct pci_dev **dev, u16 segment) 59 struct pci_dev **dev, u16 segment)
63{ 60{
@@ -172,19 +169,37 @@ dmar_parse_one_drhd(struct acpi_dmar_header *header)
172 struct acpi_dmar_hardware_unit *drhd; 169 struct acpi_dmar_hardware_unit *drhd;
173 struct dmar_drhd_unit *dmaru; 170 struct dmar_drhd_unit *dmaru;
174 int ret = 0; 171 int ret = 0;
175 static int include_all;
176 172
177 dmaru = kzalloc(sizeof(*dmaru), GFP_KERNEL); 173 dmaru = kzalloc(sizeof(*dmaru), GFP_KERNEL);
178 if (!dmaru) 174 if (!dmaru)
179 return -ENOMEM; 175 return -ENOMEM;
180 176
177 dmaru->hdr = header;
181 drhd = (struct acpi_dmar_hardware_unit *)header; 178 drhd = (struct acpi_dmar_hardware_unit *)header;
182 dmaru->reg_base_addr = drhd->address; 179 dmaru->reg_base_addr = drhd->address;
183 dmaru->include_all = drhd->flags & 0x1; /* BIT0: INCLUDE_ALL */ 180 dmaru->include_all = drhd->flags & 0x1; /* BIT0: INCLUDE_ALL */
184 181
182 ret = alloc_iommu(dmaru);
183 if (ret) {
184 kfree(dmaru);
185 return ret;
186 }
187 dmar_register_drhd_unit(dmaru);
188 return 0;
189}
190
191static int __init
192dmar_parse_dev(struct dmar_drhd_unit *dmaru)
193{
194 struct acpi_dmar_hardware_unit *drhd;
195 static int include_all;
196 int ret;
197
198 drhd = (struct acpi_dmar_hardware_unit *) dmaru->hdr;
199
185 if (!dmaru->include_all) 200 if (!dmaru->include_all)
186 ret = dmar_parse_dev_scope((void *)(drhd + 1), 201 ret = dmar_parse_dev_scope((void *)(drhd + 1),
187 ((void *)drhd) + header->length, 202 ((void *)drhd) + drhd->header.length,
188 &dmaru->devices_cnt, &dmaru->devices, 203 &dmaru->devices_cnt, &dmaru->devices,
189 drhd->segment); 204 drhd->segment);
190 else { 205 else {
@@ -197,37 +212,59 @@ dmar_parse_one_drhd(struct acpi_dmar_header *header)
197 include_all = 1; 212 include_all = 1;
198 } 213 }
199 214
200 if (ret || (dmaru->devices_cnt == 0 && !dmaru->include_all)) 215 if (ret || (dmaru->devices_cnt == 0 && !dmaru->include_all)) {
216 list_del(&dmaru->list);
201 kfree(dmaru); 217 kfree(dmaru);
202 else 218 }
203 dmar_register_drhd_unit(dmaru);
204 return ret; 219 return ret;
205} 220}
206 221
222#ifdef CONFIG_DMAR
223LIST_HEAD(dmar_rmrr_units);
224
225static void __init dmar_register_rmrr_unit(struct dmar_rmrr_unit *rmrr)
226{
227 list_add(&rmrr->list, &dmar_rmrr_units);
228}
229
230
207static int __init 231static int __init
208dmar_parse_one_rmrr(struct acpi_dmar_header *header) 232dmar_parse_one_rmrr(struct acpi_dmar_header *header)
209{ 233{
210 struct acpi_dmar_reserved_memory *rmrr; 234 struct acpi_dmar_reserved_memory *rmrr;
211 struct dmar_rmrr_unit *rmrru; 235 struct dmar_rmrr_unit *rmrru;
212 int ret = 0;
213 236
214 rmrru = kzalloc(sizeof(*rmrru), GFP_KERNEL); 237 rmrru = kzalloc(sizeof(*rmrru), GFP_KERNEL);
215 if (!rmrru) 238 if (!rmrru)
216 return -ENOMEM; 239 return -ENOMEM;
217 240
241 rmrru->hdr = header;
218 rmrr = (struct acpi_dmar_reserved_memory *)header; 242 rmrr = (struct acpi_dmar_reserved_memory *)header;
219 rmrru->base_address = rmrr->base_address; 243 rmrru->base_address = rmrr->base_address;
220 rmrru->end_address = rmrr->end_address; 244 rmrru->end_address = rmrr->end_address;
245
246 dmar_register_rmrr_unit(rmrru);
247 return 0;
248}
249
250static int __init
251rmrr_parse_dev(struct dmar_rmrr_unit *rmrru)
252{
253 struct acpi_dmar_reserved_memory *rmrr;
254 int ret;
255
256 rmrr = (struct acpi_dmar_reserved_memory *) rmrru->hdr;
221 ret = dmar_parse_dev_scope((void *)(rmrr + 1), 257 ret = dmar_parse_dev_scope((void *)(rmrr + 1),
222 ((void *)rmrr) + header->length, 258 ((void *)rmrr) + rmrr->header.length,
223 &rmrru->devices_cnt, &rmrru->devices, rmrr->segment); 259 &rmrru->devices_cnt, &rmrru->devices, rmrr->segment);
224 260
225 if (ret || (rmrru->devices_cnt == 0)) 261 if (ret || (rmrru->devices_cnt == 0)) {
262 list_del(&rmrru->list);
226 kfree(rmrru); 263 kfree(rmrru);
227 else 264 }
228 dmar_register_rmrr_unit(rmrru);
229 return ret; 265 return ret;
230} 266}
267#endif
231 268
232static void __init 269static void __init
233dmar_table_print_dmar_entry(struct acpi_dmar_header *header) 270dmar_table_print_dmar_entry(struct acpi_dmar_header *header)
@@ -252,6 +289,7 @@ dmar_table_print_dmar_entry(struct acpi_dmar_header *header)
252 } 289 }
253} 290}
254 291
292
255/** 293/**
256 * parse_dmar_table - parses the DMA reporting table 294 * parse_dmar_table - parses the DMA reporting table
257 */ 295 */
@@ -284,7 +322,9 @@ parse_dmar_table(void)
284 ret = dmar_parse_one_drhd(entry_header); 322 ret = dmar_parse_one_drhd(entry_header);
285 break; 323 break;
286 case ACPI_DMAR_TYPE_RESERVED_MEMORY: 324 case ACPI_DMAR_TYPE_RESERVED_MEMORY:
325#ifdef CONFIG_DMAR
287 ret = dmar_parse_one_rmrr(entry_header); 326 ret = dmar_parse_one_rmrr(entry_header);
327#endif
288 break; 328 break;
289 default: 329 default:
290 printk(KERN_WARNING PREFIX 330 printk(KERN_WARNING PREFIX
@@ -300,15 +340,77 @@ parse_dmar_table(void)
300 return ret; 340 return ret;
301} 341}
302 342
343int dmar_pci_device_match(struct pci_dev *devices[], int cnt,
344 struct pci_dev *dev)
345{
346 int index;
347
348 while (dev) {
349 for (index = 0; index < cnt; index++)
350 if (dev == devices[index])
351 return 1;
303 352
304int __init dmar_table_init(void) 353 /* Check our parent */
354 dev = dev->bus->self;
355 }
356
357 return 0;
358}
359
360struct dmar_drhd_unit *
361dmar_find_matched_drhd_unit(struct pci_dev *dev)
305{ 362{
363 struct dmar_drhd_unit *drhd = NULL;
364
365 list_for_each_entry(drhd, &dmar_drhd_units, list) {
366 if (drhd->include_all || dmar_pci_device_match(drhd->devices,
367 drhd->devices_cnt, dev))
368 return drhd;
369 }
370
371 return NULL;
372}
373
374int __init dmar_dev_scope_init(void)
375{
376 struct dmar_drhd_unit *drhd;
377 int ret = -ENODEV;
378
379 for_each_drhd_unit(drhd) {
380 ret = dmar_parse_dev(drhd);
381 if (ret)
382 return ret;
383 }
384
385#ifdef CONFIG_DMAR
386 {
387 struct dmar_rmrr_unit *rmrr;
388 for_each_rmrr_units(rmrr) {
389 ret = rmrr_parse_dev(rmrr);
390 if (ret)
391 return ret;
392 }
393 }
394#endif
395
396 return ret;
397}
306 398
399
400int __init dmar_table_init(void)
401{
402 static int dmar_table_initialized;
307 int ret; 403 int ret;
308 404
405 if (dmar_table_initialized)
406 return 0;
407
408 dmar_table_initialized = 1;
409
309 ret = parse_dmar_table(); 410 ret = parse_dmar_table();
310 if (ret) { 411 if (ret) {
311 printk(KERN_INFO PREFIX "parse DMAR table failure.\n"); 412 if (ret != -ENODEV)
413 printk(KERN_INFO PREFIX "parse DMAR table failure.\n");
312 return ret; 414 return ret;
313 } 415 }
314 416
@@ -317,9 +419,14 @@ int __init dmar_table_init(void)
317 return -ENODEV; 419 return -ENODEV;
318 } 420 }
319 421
422#ifdef CONFIG_DMAR
320 if (list_empty(&dmar_rmrr_units)) 423 if (list_empty(&dmar_rmrr_units))
321 printk(KERN_INFO PREFIX "No RMRR found\n"); 424 printk(KERN_INFO PREFIX "No RMRR found\n");
425#endif
322 426
427#ifdef CONFIG_INTR_REMAP
428 parse_ioapics_under_ir();
429#endif
323 return 0; 430 return 0;
324} 431}
325 432
@@ -341,3 +448,255 @@ int __init early_dmar_detect(void)
341 448
342 return (ACPI_SUCCESS(status) ? 1 : 0); 449 return (ACPI_SUCCESS(status) ? 1 : 0);
343} 450}
451
452void __init detect_intel_iommu(void)
453{
454 int ret;
455
456 ret = early_dmar_detect();
457
458#ifdef CONFIG_DMAR
459 {
460 struct acpi_table_dmar *dmar;
461 /*
462 * for now we will disable dma-remapping when interrupt
463 * remapping is enabled.
464 * When support for queued invalidation for IOTLB invalidation
465 * is added, we will not need this any more.
466 */
467 dmar = (struct acpi_table_dmar *) dmar_tbl;
468 if (ret && cpu_has_x2apic && dmar->flags & 0x1) {
469 printk(KERN_INFO
470 "Queued invalidation will be enabled to support "
471 "x2apic and Intr-remapping.\n");
472 printk(KERN_INFO
473 "Disabling IOMMU detection, because of missing "
474 "queued invalidation support for IOTLB "
475 "invalidation\n");
476 printk(KERN_INFO
477 "Use \"nox2apic\", if you want to use Intel "
478 " IOMMU for DMA-remapping and don't care about "
479 " x2apic support\n");
480
481 dmar_disabled = 1;
482 return;
483 }
484
485 if (ret && !no_iommu && !iommu_detected && !swiotlb &&
486 !dmar_disabled)
487 iommu_detected = 1;
488 }
489#endif
490}
491
492
493int alloc_iommu(struct dmar_drhd_unit *drhd)
494{
495 struct intel_iommu *iommu;
496 int map_size;
497 u32 ver;
498 static int iommu_allocated = 0;
499
500 iommu = kzalloc(sizeof(*iommu), GFP_KERNEL);
501 if (!iommu)
502 return -ENOMEM;
503
504 iommu->seq_id = iommu_allocated++;
505
506 iommu->reg = ioremap(drhd->reg_base_addr, PAGE_SIZE_4K);
507 if (!iommu->reg) {
508 printk(KERN_ERR "IOMMU: can't map the region\n");
509 goto error;
510 }
511 iommu->cap = dmar_readq(iommu->reg + DMAR_CAP_REG);
512 iommu->ecap = dmar_readq(iommu->reg + DMAR_ECAP_REG);
513
514 /* the registers might be more than one page */
515 map_size = max_t(int, ecap_max_iotlb_offset(iommu->ecap),
516 cap_max_fault_reg_offset(iommu->cap));
517 map_size = PAGE_ALIGN_4K(map_size);
518 if (map_size > PAGE_SIZE_4K) {
519 iounmap(iommu->reg);
520 iommu->reg = ioremap(drhd->reg_base_addr, map_size);
521 if (!iommu->reg) {
522 printk(KERN_ERR "IOMMU: can't map the region\n");
523 goto error;
524 }
525 }
526
527 ver = readl(iommu->reg + DMAR_VER_REG);
528 pr_debug("IOMMU %llx: ver %d:%d cap %llx ecap %llx\n",
529 drhd->reg_base_addr, DMAR_VER_MAJOR(ver), DMAR_VER_MINOR(ver),
530 iommu->cap, iommu->ecap);
531
532 spin_lock_init(&iommu->register_lock);
533
534 drhd->iommu = iommu;
535 return 0;
536error:
537 kfree(iommu);
538 return -1;
539}
540
541void free_iommu(struct intel_iommu *iommu)
542{
543 if (!iommu)
544 return;
545
546#ifdef CONFIG_DMAR
547 free_dmar_iommu(iommu);
548#endif
549
550 if (iommu->reg)
551 iounmap(iommu->reg);
552 kfree(iommu);
553}
554
555/*
556 * Reclaim all the submitted descriptors which have completed its work.
557 */
558static inline void reclaim_free_desc(struct q_inval *qi)
559{
560 while (qi->desc_status[qi->free_tail] == QI_DONE) {
561 qi->desc_status[qi->free_tail] = QI_FREE;
562 qi->free_tail = (qi->free_tail + 1) % QI_LENGTH;
563 qi->free_cnt++;
564 }
565}
566
567/*
568 * Submit the queued invalidation descriptor to the remapping
569 * hardware unit and wait for its completion.
570 */
571void qi_submit_sync(struct qi_desc *desc, struct intel_iommu *iommu)
572{
573 struct q_inval *qi = iommu->qi;
574 struct qi_desc *hw, wait_desc;
575 int wait_index, index;
576 unsigned long flags;
577
578 if (!qi)
579 return;
580
581 hw = qi->desc;
582
583 spin_lock(&qi->q_lock);
584 while (qi->free_cnt < 3) {
585 spin_unlock(&qi->q_lock);
586 cpu_relax();
587 spin_lock(&qi->q_lock);
588 }
589
590 index = qi->free_head;
591 wait_index = (index + 1) % QI_LENGTH;
592
593 qi->desc_status[index] = qi->desc_status[wait_index] = QI_IN_USE;
594
595 hw[index] = *desc;
596
597 wait_desc.low = QI_IWD_STATUS_DATA(2) | QI_IWD_STATUS_WRITE | QI_IWD_TYPE;
598 wait_desc.high = virt_to_phys(&qi->desc_status[wait_index]);
599
600 hw[wait_index] = wait_desc;
601
602 __iommu_flush_cache(iommu, &hw[index], sizeof(struct qi_desc));
603 __iommu_flush_cache(iommu, &hw[wait_index], sizeof(struct qi_desc));
604
605 qi->free_head = (qi->free_head + 2) % QI_LENGTH;
606 qi->free_cnt -= 2;
607
608 spin_lock_irqsave(&iommu->register_lock, flags);
609 /*
610 * update the HW tail register indicating the presence of
611 * new descriptors.
612 */
613 writel(qi->free_head << 4, iommu->reg + DMAR_IQT_REG);
614 spin_unlock_irqrestore(&iommu->register_lock, flags);
615
616 while (qi->desc_status[wait_index] != QI_DONE) {
617 spin_unlock(&qi->q_lock);
618 cpu_relax();
619 spin_lock(&qi->q_lock);
620 }
621
622 qi->desc_status[index] = QI_DONE;
623
624 reclaim_free_desc(qi);
625 spin_unlock(&qi->q_lock);
626}
627
628/*
629 * Flush the global interrupt entry cache.
630 */
631void qi_global_iec(struct intel_iommu *iommu)
632{
633 struct qi_desc desc;
634
635 desc.low = QI_IEC_TYPE;
636 desc.high = 0;
637
638 qi_submit_sync(&desc, iommu);
639}
640
641/*
642 * Enable Queued Invalidation interface. This is a must to support
643 * interrupt-remapping. Also used by DMA-remapping, which replaces
644 * register based IOTLB invalidation.
645 */
646int dmar_enable_qi(struct intel_iommu *iommu)
647{
648 u32 cmd, sts;
649 unsigned long flags;
650 struct q_inval *qi;
651
652 if (!ecap_qis(iommu->ecap))
653 return -ENOENT;
654
655 /*
656 * queued invalidation is already setup and enabled.
657 */
658 if (iommu->qi)
659 return 0;
660
661 iommu->qi = kmalloc(sizeof(*qi), GFP_KERNEL);
662 if (!iommu->qi)
663 return -ENOMEM;
664
665 qi = iommu->qi;
666
667 qi->desc = (void *)(get_zeroed_page(GFP_KERNEL));
668 if (!qi->desc) {
669 kfree(qi);
670 iommu->qi = 0;
671 return -ENOMEM;
672 }
673
674 qi->desc_status = kmalloc(QI_LENGTH * sizeof(int), GFP_KERNEL);
675 if (!qi->desc_status) {
676 free_page((unsigned long) qi->desc);
677 kfree(qi);
678 iommu->qi = 0;
679 return -ENOMEM;
680 }
681
682 qi->free_head = qi->free_tail = 0;
683 qi->free_cnt = QI_LENGTH;
684
685 spin_lock_init(&qi->q_lock);
686
687 spin_lock_irqsave(&iommu->register_lock, flags);
688 /* write zero to the tail reg */
689 writel(0, iommu->reg + DMAR_IQT_REG);
690
691 dmar_writeq(iommu->reg + DMAR_IQA_REG, virt_to_phys(qi->desc));
692
693 cmd = iommu->gcmd | DMA_GCMD_QIE;
694 iommu->gcmd |= DMA_GCMD_QIE;
695 writel(cmd, iommu->reg + DMAR_GCMD_REG);
696
697 /* Make sure hardware complete it */
698 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl, (sts & DMA_GSTS_QIES), sts);
699 spin_unlock_irqrestore(&iommu->register_lock, flags);
700
701 return 0;
702}
diff --git a/drivers/pci/hotplug/fakephp.c b/drivers/pci/hotplug/fakephp.c
index 40337a06c18a..146ca9cd1567 100644
--- a/drivers/pci/hotplug/fakephp.c
+++ b/drivers/pci/hotplug/fakephp.c
@@ -320,15 +320,15 @@ static int disable_slot(struct hotplug_slot *slot)
320 return -ENODEV; 320 return -ENODEV;
321 } 321 }
322 322
323 /* remove the device from the pci core */
324 pci_remove_bus_device(dev);
325
323 /* queue work item to blow away this sysfs entry and other 326 /* queue work item to blow away this sysfs entry and other
324 * parts. 327 * parts.
325 */ 328 */
326 INIT_WORK(&dslot->remove_work, remove_slot_worker); 329 INIT_WORK(&dslot->remove_work, remove_slot_worker);
327 queue_work(dummyphp_wq, &dslot->remove_work); 330 queue_work(dummyphp_wq, &dslot->remove_work);
328 331
329 /* blow away this sysfs entry and other parts. */
330 remove_slot(dslot);
331
332 pci_dev_put(dev); 332 pci_dev_put(dev);
333 } 333 }
334 return 0; 334 return 0;
diff --git a/drivers/pci/hotplug/pciehp_hpc.c b/drivers/pci/hotplug/pciehp_hpc.c
index ab31f5ba665d..9d934ddee956 100644
--- a/drivers/pci/hotplug/pciehp_hpc.c
+++ b/drivers/pci/hotplug/pciehp_hpc.c
@@ -258,7 +258,7 @@ static int pcie_poll_cmd(struct controller *ctrl)
258 return 1; 258 return 1;
259 } 259 }
260 } 260 }
261 while (timeout > 1000) { 261 while (timeout > 0) {
262 msleep(10); 262 msleep(10);
263 timeout -= 10; 263 timeout -= 10;
264 if (!pciehp_readw(ctrl, SLOTSTATUS, &slot_status)) { 264 if (!pciehp_readw(ctrl, SLOTSTATUS, &slot_status)) {
diff --git a/drivers/pci/intel-iommu.c b/drivers/pci/intel-iommu.c
index 8d0e60ac849c..389fdd6f4a9f 100644
--- a/drivers/pci/intel-iommu.c
+++ b/drivers/pci/intel-iommu.c
@@ -49,8 +49,6 @@
49 49
50#define DEFAULT_DOMAIN_ADDRESS_WIDTH 48 50#define DEFAULT_DOMAIN_ADDRESS_WIDTH 48
51 51
52#define DMAR_OPERATION_TIMEOUT ((cycles_t) tsc_khz*10*1000) /* 10sec */
53
54#define DOMAIN_MAX_ADDR(gaw) ((((u64)1) << gaw) - 1) 52#define DOMAIN_MAX_ADDR(gaw) ((((u64)1) << gaw) - 1)
55 53
56 54
@@ -58,8 +56,6 @@ static void flush_unmaps_timeout(unsigned long data);
58 56
59DEFINE_TIMER(unmap_timer, flush_unmaps_timeout, 0, 0); 57DEFINE_TIMER(unmap_timer, flush_unmaps_timeout, 0, 0);
60 58
61static struct intel_iommu *g_iommus;
62
63#define HIGH_WATER_MARK 250 59#define HIGH_WATER_MARK 250
64struct deferred_flush_tables { 60struct deferred_flush_tables {
65 int next; 61 int next;
@@ -80,7 +76,7 @@ static long list_size;
80 76
81static void domain_remove_dev_info(struct dmar_domain *domain); 77static void domain_remove_dev_info(struct dmar_domain *domain);
82 78
83static int dmar_disabled; 79int dmar_disabled;
84static int __initdata dmar_map_gfx = 1; 80static int __initdata dmar_map_gfx = 1;
85static int dmar_forcedac; 81static int dmar_forcedac;
86static int intel_iommu_strict; 82static int intel_iommu_strict;
@@ -185,13 +181,6 @@ void free_iova_mem(struct iova *iova)
185 kmem_cache_free(iommu_iova_cache, iova); 181 kmem_cache_free(iommu_iova_cache, iova);
186} 182}
187 183
188static inline void __iommu_flush_cache(
189 struct intel_iommu *iommu, void *addr, int size)
190{
191 if (!ecap_coherent(iommu->ecap))
192 clflush_cache_range(addr, size);
193}
194
195/* Gets context entry for a given bus and devfn */ 184/* Gets context entry for a given bus and devfn */
196static struct context_entry * device_to_context_entry(struct intel_iommu *iommu, 185static struct context_entry * device_to_context_entry(struct intel_iommu *iommu,
197 u8 bus, u8 devfn) 186 u8 bus, u8 devfn)
@@ -488,19 +477,6 @@ static int iommu_alloc_root_entry(struct intel_iommu *iommu)
488 return 0; 477 return 0;
489} 478}
490 479
491#define IOMMU_WAIT_OP(iommu, offset, op, cond, sts) \
492{\
493 cycles_t start_time = get_cycles();\
494 while (1) {\
495 sts = op (iommu->reg + offset);\
496 if (cond)\
497 break;\
498 if (DMAR_OPERATION_TIMEOUT < (get_cycles() - start_time))\
499 panic("DMAR hardware is malfunctioning\n");\
500 cpu_relax();\
501 }\
502}
503
504static void iommu_set_root_entry(struct intel_iommu *iommu) 480static void iommu_set_root_entry(struct intel_iommu *iommu)
505{ 481{
506 void *addr; 482 void *addr;
@@ -990,6 +966,8 @@ static int iommu_init_domains(struct intel_iommu *iommu)
990 return -ENOMEM; 966 return -ENOMEM;
991 } 967 }
992 968
969 spin_lock_init(&iommu->lock);
970
993 /* 971 /*
994 * if Caching mode is set, then invalid translations are tagged 972 * if Caching mode is set, then invalid translations are tagged
995 * with domainid 0. Hence we need to pre-allocate it. 973 * with domainid 0. Hence we need to pre-allocate it.
@@ -998,62 +976,15 @@ static int iommu_init_domains(struct intel_iommu *iommu)
998 set_bit(0, iommu->domain_ids); 976 set_bit(0, iommu->domain_ids);
999 return 0; 977 return 0;
1000} 978}
1001static struct intel_iommu *alloc_iommu(struct intel_iommu *iommu,
1002 struct dmar_drhd_unit *drhd)
1003{
1004 int ret;
1005 int map_size;
1006 u32 ver;
1007
1008 iommu->reg = ioremap(drhd->reg_base_addr, PAGE_SIZE_4K);
1009 if (!iommu->reg) {
1010 printk(KERN_ERR "IOMMU: can't map the region\n");
1011 goto error;
1012 }
1013 iommu->cap = dmar_readq(iommu->reg + DMAR_CAP_REG);
1014 iommu->ecap = dmar_readq(iommu->reg + DMAR_ECAP_REG);
1015
1016 /* the registers might be more than one page */
1017 map_size = max_t(int, ecap_max_iotlb_offset(iommu->ecap),
1018 cap_max_fault_reg_offset(iommu->cap));
1019 map_size = PAGE_ALIGN_4K(map_size);
1020 if (map_size > PAGE_SIZE_4K) {
1021 iounmap(iommu->reg);
1022 iommu->reg = ioremap(drhd->reg_base_addr, map_size);
1023 if (!iommu->reg) {
1024 printk(KERN_ERR "IOMMU: can't map the region\n");
1025 goto error;
1026 }
1027 }
1028
1029 ver = readl(iommu->reg + DMAR_VER_REG);
1030 pr_debug("IOMMU %llx: ver %d:%d cap %llx ecap %llx\n",
1031 drhd->reg_base_addr, DMAR_VER_MAJOR(ver), DMAR_VER_MINOR(ver),
1032 iommu->cap, iommu->ecap);
1033 ret = iommu_init_domains(iommu);
1034 if (ret)
1035 goto error_unmap;
1036 spin_lock_init(&iommu->lock);
1037 spin_lock_init(&iommu->register_lock);
1038 979
1039 drhd->iommu = iommu;
1040 return iommu;
1041error_unmap:
1042 iounmap(iommu->reg);
1043error:
1044 kfree(iommu);
1045 return NULL;
1046}
1047 980
1048static void domain_exit(struct dmar_domain *domain); 981static void domain_exit(struct dmar_domain *domain);
1049static void free_iommu(struct intel_iommu *iommu) 982
983void free_dmar_iommu(struct intel_iommu *iommu)
1050{ 984{
1051 struct dmar_domain *domain; 985 struct dmar_domain *domain;
1052 int i; 986 int i;
1053 987
1054 if (!iommu)
1055 return;
1056
1057 i = find_first_bit(iommu->domain_ids, cap_ndoms(iommu->cap)); 988 i = find_first_bit(iommu->domain_ids, cap_ndoms(iommu->cap));
1058 for (; i < cap_ndoms(iommu->cap); ) { 989 for (; i < cap_ndoms(iommu->cap); ) {
1059 domain = iommu->domains[i]; 990 domain = iommu->domains[i];
@@ -1078,10 +1009,6 @@ static void free_iommu(struct intel_iommu *iommu)
1078 1009
1079 /* free context mapping */ 1010 /* free context mapping */
1080 free_context_table(iommu); 1011 free_context_table(iommu);
1081
1082 if (iommu->reg)
1083 iounmap(iommu->reg);
1084 kfree(iommu);
1085} 1012}
1086 1013
1087static struct dmar_domain * iommu_alloc_domain(struct intel_iommu *iommu) 1014static struct dmar_domain * iommu_alloc_domain(struct intel_iommu *iommu)
@@ -1426,37 +1353,6 @@ find_domain(struct pci_dev *pdev)
1426 return NULL; 1353 return NULL;
1427} 1354}
1428 1355
1429static int dmar_pci_device_match(struct pci_dev *devices[], int cnt,
1430 struct pci_dev *dev)
1431{
1432 int index;
1433
1434 while (dev) {
1435 for (index = 0; index < cnt; index++)
1436 if (dev == devices[index])
1437 return 1;
1438
1439 /* Check our parent */
1440 dev = dev->bus->self;
1441 }
1442
1443 return 0;
1444}
1445
1446static struct dmar_drhd_unit *
1447dmar_find_matched_drhd_unit(struct pci_dev *dev)
1448{
1449 struct dmar_drhd_unit *drhd = NULL;
1450
1451 list_for_each_entry(drhd, &dmar_drhd_units, list) {
1452 if (drhd->include_all || dmar_pci_device_match(drhd->devices,
1453 drhd->devices_cnt, dev))
1454 return drhd;
1455 }
1456
1457 return NULL;
1458}
1459
1460/* domain is initialized */ 1356/* domain is initialized */
1461static struct dmar_domain *get_domain_for_dev(struct pci_dev *pdev, int gaw) 1357static struct dmar_domain *get_domain_for_dev(struct pci_dev *pdev, int gaw)
1462{ 1358{
@@ -1729,8 +1625,6 @@ int __init init_dmars(void)
1729 * endfor 1625 * endfor
1730 */ 1626 */
1731 for_each_drhd_unit(drhd) { 1627 for_each_drhd_unit(drhd) {
1732 if (drhd->ignored)
1733 continue;
1734 g_num_of_iommus++; 1628 g_num_of_iommus++;
1735 /* 1629 /*
1736 * lock not needed as this is only incremented in the single 1630 * lock not needed as this is only incremented in the single
@@ -1739,12 +1633,6 @@ int __init init_dmars(void)
1739 */ 1633 */
1740 } 1634 }
1741 1635
1742 g_iommus = kzalloc(g_num_of_iommus * sizeof(*iommu), GFP_KERNEL);
1743 if (!g_iommus) {
1744 ret = -ENOMEM;
1745 goto error;
1746 }
1747
1748 deferred_flush = kzalloc(g_num_of_iommus * 1636 deferred_flush = kzalloc(g_num_of_iommus *
1749 sizeof(struct deferred_flush_tables), GFP_KERNEL); 1637 sizeof(struct deferred_flush_tables), GFP_KERNEL);
1750 if (!deferred_flush) { 1638 if (!deferred_flush) {
@@ -1752,16 +1640,15 @@ int __init init_dmars(void)
1752 goto error; 1640 goto error;
1753 } 1641 }
1754 1642
1755 i = 0;
1756 for_each_drhd_unit(drhd) { 1643 for_each_drhd_unit(drhd) {
1757 if (drhd->ignored) 1644 if (drhd->ignored)
1758 continue; 1645 continue;
1759 iommu = alloc_iommu(&g_iommus[i], drhd); 1646
1760 i++; 1647 iommu = drhd->iommu;
1761 if (!iommu) { 1648
1762 ret = -ENOMEM; 1649 ret = iommu_init_domains(iommu);
1650 if (ret)
1763 goto error; 1651 goto error;
1764 }
1765 1652
1766 /* 1653 /*
1767 * TBD: 1654 * TBD:
@@ -1845,7 +1732,6 @@ error:
1845 iommu = drhd->iommu; 1732 iommu = drhd->iommu;
1846 free_iommu(iommu); 1733 free_iommu(iommu);
1847 } 1734 }
1848 kfree(g_iommus);
1849 return ret; 1735 return ret;
1850} 1736}
1851 1737
@@ -2002,7 +1888,10 @@ static void flush_unmaps(void)
2002 /* just flush them all */ 1888 /* just flush them all */
2003 for (i = 0; i < g_num_of_iommus; i++) { 1889 for (i = 0; i < g_num_of_iommus; i++) {
2004 if (deferred_flush[i].next) { 1890 if (deferred_flush[i].next) {
2005 iommu_flush_iotlb_global(&g_iommus[i], 0); 1891 struct intel_iommu *iommu =
1892 deferred_flush[i].domain[0]->iommu;
1893
1894 iommu_flush_iotlb_global(iommu, 0);
2006 for (j = 0; j < deferred_flush[i].next; j++) { 1895 for (j = 0; j < deferred_flush[i].next; j++) {
2007 __free_iova(&deferred_flush[i].domain[j]->iovad, 1896 __free_iova(&deferred_flush[i].domain[j]->iovad,
2008 deferred_flush[i].iova[j]); 1897 deferred_flush[i].iova[j]);
@@ -2032,7 +1921,8 @@ static void add_unmap(struct dmar_domain *dom, struct iova *iova)
2032 if (list_size == HIGH_WATER_MARK) 1921 if (list_size == HIGH_WATER_MARK)
2033 flush_unmaps(); 1922 flush_unmaps();
2034 1923
2035 iommu_id = dom->iommu - g_iommus; 1924 iommu_id = dom->iommu->seq_id;
1925
2036 next = deferred_flush[iommu_id].next; 1926 next = deferred_flush[iommu_id].next;
2037 deferred_flush[iommu_id].domain[next] = dom; 1927 deferred_flush[iommu_id].domain[next] = dom;
2038 deferred_flush[iommu_id].iova[next] = iova; 1928 deferred_flush[iommu_id].iova[next] = iova;
@@ -2348,15 +2238,6 @@ static void __init iommu_exit_mempool(void)
2348 2238
2349} 2239}
2350 2240
2351void __init detect_intel_iommu(void)
2352{
2353 if (swiotlb || no_iommu || iommu_detected || dmar_disabled)
2354 return;
2355 if (early_dmar_detect()) {
2356 iommu_detected = 1;
2357 }
2358}
2359
2360static void __init init_no_remapping_devices(void) 2241static void __init init_no_remapping_devices(void)
2361{ 2242{
2362 struct dmar_drhd_unit *drhd; 2243 struct dmar_drhd_unit *drhd;
@@ -2403,12 +2284,19 @@ int __init intel_iommu_init(void)
2403{ 2284{
2404 int ret = 0; 2285 int ret = 0;
2405 2286
2406 if (no_iommu || swiotlb || dmar_disabled)
2407 return -ENODEV;
2408
2409 if (dmar_table_init()) 2287 if (dmar_table_init())
2410 return -ENODEV; 2288 return -ENODEV;
2411 2289
2290 if (dmar_dev_scope_init())
2291 return -ENODEV;
2292
2293 /*
2294 * Check the need for DMA-remapping initialization now.
2295 * Above initialization will also be used by Interrupt-remapping.
2296 */
2297 if (no_iommu || swiotlb || dmar_disabled)
2298 return -ENODEV;
2299
2412 iommu_init_mempool(); 2300 iommu_init_mempool();
2413 dmar_init_reserved_ranges(); 2301 dmar_init_reserved_ranges();
2414 2302
diff --git a/drivers/pci/intel-iommu.h b/drivers/pci/intel-iommu.h
index afc0ad96122e..2142c01e0143 100644
--- a/drivers/pci/intel-iommu.h
+++ b/drivers/pci/intel-iommu.h
@@ -27,19 +27,8 @@
27#include <linux/sysdev.h> 27#include <linux/sysdev.h>
28#include "iova.h" 28#include "iova.h"
29#include <linux/io.h> 29#include <linux/io.h>
30 30#include <asm/cacheflush.h>
31/* 31#include "dma_remapping.h"
32 * We need a fixed PAGE_SIZE of 4K irrespective of
33 * arch PAGE_SIZE for IOMMU page tables.
34 */
35#define PAGE_SHIFT_4K (12)
36#define PAGE_SIZE_4K (1UL << PAGE_SHIFT_4K)
37#define PAGE_MASK_4K (((u64)-1) << PAGE_SHIFT_4K)
38#define PAGE_ALIGN_4K(addr) (((addr) + PAGE_SIZE_4K - 1) & PAGE_MASK_4K)
39
40#define IOVA_PFN(addr) ((addr) >> PAGE_SHIFT_4K)
41#define DMA_32BIT_PFN IOVA_PFN(DMA_32BIT_MASK)
42#define DMA_64BIT_PFN IOVA_PFN(DMA_64BIT_MASK)
43 32
44/* 33/*
45 * Intel IOMMU register specification per version 1.0 public spec. 34 * Intel IOMMU register specification per version 1.0 public spec.
@@ -63,6 +52,11 @@
63#define DMAR_PLMLIMIT_REG 0x6c /* PMRR low limit */ 52#define DMAR_PLMLIMIT_REG 0x6c /* PMRR low limit */
64#define DMAR_PHMBASE_REG 0x70 /* pmrr high base addr */ 53#define DMAR_PHMBASE_REG 0x70 /* pmrr high base addr */
65#define DMAR_PHMLIMIT_REG 0x78 /* pmrr high limit */ 54#define DMAR_PHMLIMIT_REG 0x78 /* pmrr high limit */
55#define DMAR_IQH_REG 0x80 /* Invalidation queue head register */
56#define DMAR_IQT_REG 0x88 /* Invalidation queue tail register */
57#define DMAR_IQA_REG 0x90 /* Invalidation queue addr register */
58#define DMAR_ICS_REG 0x98 /* Invalidation complete status register */
59#define DMAR_IRTA_REG 0xb8 /* Interrupt remapping table addr register */
66 60
67#define OFFSET_STRIDE (9) 61#define OFFSET_STRIDE (9)
68/* 62/*
@@ -126,6 +120,10 @@ static inline void dmar_writeq(void __iomem *addr, u64 val)
126#define ecap_max_iotlb_offset(e) \ 120#define ecap_max_iotlb_offset(e) \
127 (ecap_iotlb_offset(e) + ecap_niotlb_iunits(e) * 16) 121 (ecap_iotlb_offset(e) + ecap_niotlb_iunits(e) * 16)
128#define ecap_coherent(e) ((e) & 0x1) 122#define ecap_coherent(e) ((e) & 0x1)
123#define ecap_qis(e) ((e) & 0x2)
124#define ecap_eim_support(e) ((e >> 4) & 0x1)
125#define ecap_ir_support(e) ((e >> 3) & 0x1)
126#define ecap_max_handle_mask(e) ((e >> 20) & 0xf)
129 127
130 128
131/* IOTLB_REG */ 129/* IOTLB_REG */
@@ -141,6 +139,17 @@ static inline void dmar_writeq(void __iomem *addr, u64 val)
141#define DMA_TLB_IH_NONLEAF (((u64)1) << 6) 139#define DMA_TLB_IH_NONLEAF (((u64)1) << 6)
142#define DMA_TLB_MAX_SIZE (0x3f) 140#define DMA_TLB_MAX_SIZE (0x3f)
143 141
142/* INVALID_DESC */
143#define DMA_ID_TLB_GLOBAL_FLUSH (((u64)1) << 3)
144#define DMA_ID_TLB_DSI_FLUSH (((u64)2) << 3)
145#define DMA_ID_TLB_PSI_FLUSH (((u64)3) << 3)
146#define DMA_ID_TLB_READ_DRAIN (((u64)1) << 7)
147#define DMA_ID_TLB_WRITE_DRAIN (((u64)1) << 6)
148#define DMA_ID_TLB_DID(id) (((u64)((id & 0xffff) << 16)))
149#define DMA_ID_TLB_IH_NONLEAF (((u64)1) << 6)
150#define DMA_ID_TLB_ADDR(addr) (addr)
151#define DMA_ID_TLB_ADDR_MASK(mask) (mask)
152
144/* PMEN_REG */ 153/* PMEN_REG */
145#define DMA_PMEN_EPM (((u32)1)<<31) 154#define DMA_PMEN_EPM (((u32)1)<<31)
146#define DMA_PMEN_PRS (((u32)1)<<0) 155#define DMA_PMEN_PRS (((u32)1)<<0)
@@ -151,6 +160,9 @@ static inline void dmar_writeq(void __iomem *addr, u64 val)
151#define DMA_GCMD_SFL (((u32)1) << 29) 160#define DMA_GCMD_SFL (((u32)1) << 29)
152#define DMA_GCMD_EAFL (((u32)1) << 28) 161#define DMA_GCMD_EAFL (((u32)1) << 28)
153#define DMA_GCMD_WBF (((u32)1) << 27) 162#define DMA_GCMD_WBF (((u32)1) << 27)
163#define DMA_GCMD_QIE (((u32)1) << 26)
164#define DMA_GCMD_SIRTP (((u32)1) << 24)
165#define DMA_GCMD_IRE (((u32) 1) << 25)
154 166
155/* GSTS_REG */ 167/* GSTS_REG */
156#define DMA_GSTS_TES (((u32)1) << 31) 168#define DMA_GSTS_TES (((u32)1) << 31)
@@ -158,6 +170,9 @@ static inline void dmar_writeq(void __iomem *addr, u64 val)
158#define DMA_GSTS_FLS (((u32)1) << 29) 170#define DMA_GSTS_FLS (((u32)1) << 29)
159#define DMA_GSTS_AFLS (((u32)1) << 28) 171#define DMA_GSTS_AFLS (((u32)1) << 28)
160#define DMA_GSTS_WBFS (((u32)1) << 27) 172#define DMA_GSTS_WBFS (((u32)1) << 27)
173#define DMA_GSTS_QIES (((u32)1) << 26)
174#define DMA_GSTS_IRTPS (((u32)1) << 24)
175#define DMA_GSTS_IRES (((u32)1) << 25)
161 176
162/* CCMD_REG */ 177/* CCMD_REG */
163#define DMA_CCMD_ICC (((u64)1) << 63) 178#define DMA_CCMD_ICC (((u64)1) << 63)
@@ -187,158 +202,106 @@ static inline void dmar_writeq(void __iomem *addr, u64 val)
187#define dma_frcd_source_id(c) (c & 0xffff) 202#define dma_frcd_source_id(c) (c & 0xffff)
188#define dma_frcd_page_addr(d) (d & (((u64)-1) << 12)) /* low 64 bit */ 203#define dma_frcd_page_addr(d) (d & (((u64)-1) << 12)) /* low 64 bit */
189 204
190/* 205#define DMAR_OPERATION_TIMEOUT ((cycles_t) tsc_khz*10*1000) /* 10sec */
191 * 0: Present 206
192 * 1-11: Reserved 207#define IOMMU_WAIT_OP(iommu, offset, op, cond, sts) \
193 * 12-63: Context Ptr (12 - (haw-1)) 208{\
194 * 64-127: Reserved 209 cycles_t start_time = get_cycles();\
195 */ 210 while (1) {\
196struct root_entry { 211 sts = op (iommu->reg + offset);\
197 u64 val; 212 if (cond)\
198 u64 rsvd1; 213 break;\
199}; 214 if (DMAR_OPERATION_TIMEOUT < (get_cycles() - start_time))\
200#define ROOT_ENTRY_NR (PAGE_SIZE_4K/sizeof(struct root_entry)) 215 panic("DMAR hardware is malfunctioning\n");\
201static inline bool root_present(struct root_entry *root) 216 cpu_relax();\
202{ 217 }\
203 return (root->val & 1);
204}
205static inline void set_root_present(struct root_entry *root)
206{
207 root->val |= 1;
208}
209static inline void set_root_value(struct root_entry *root, unsigned long value)
210{
211 root->val |= value & PAGE_MASK_4K;
212} 218}
213 219
214struct context_entry; 220#define QI_LENGTH 256 /* queue length */
215static inline struct context_entry *
216get_context_addr_from_root(struct root_entry *root)
217{
218 return (struct context_entry *)
219 (root_present(root)?phys_to_virt(
220 root->val & PAGE_MASK_4K):
221 NULL);
222}
223
224/*
225 * low 64 bits:
226 * 0: present
227 * 1: fault processing disable
228 * 2-3: translation type
229 * 12-63: address space root
230 * high 64 bits:
231 * 0-2: address width
232 * 3-6: aval
233 * 8-23: domain id
234 */
235struct context_entry {
236 u64 lo;
237 u64 hi;
238};
239#define context_present(c) ((c).lo & 1)
240#define context_fault_disable(c) (((c).lo >> 1) & 1)
241#define context_translation_type(c) (((c).lo >> 2) & 3)
242#define context_address_root(c) ((c).lo & PAGE_MASK_4K)
243#define context_address_width(c) ((c).hi & 7)
244#define context_domain_id(c) (((c).hi >> 8) & ((1 << 16) - 1))
245
246#define context_set_present(c) do {(c).lo |= 1;} while (0)
247#define context_set_fault_enable(c) \
248 do {(c).lo &= (((u64)-1) << 2) | 1;} while (0)
249#define context_set_translation_type(c, val) \
250 do { \
251 (c).lo &= (((u64)-1) << 4) | 3; \
252 (c).lo |= ((val) & 3) << 2; \
253 } while (0)
254#define CONTEXT_TT_MULTI_LEVEL 0
255#define context_set_address_root(c, val) \
256 do {(c).lo |= (val) & PAGE_MASK_4K;} while (0)
257#define context_set_address_width(c, val) do {(c).hi |= (val) & 7;} while (0)
258#define context_set_domain_id(c, val) \
259 do {(c).hi |= ((val) & ((1 << 16) - 1)) << 8;} while (0)
260#define context_clear_entry(c) do {(c).lo = 0; (c).hi = 0;} while (0)
261 221
262/* 222enum {
263 * 0: readable 223 QI_FREE,
264 * 1: writable 224 QI_IN_USE,
265 * 2-6: reserved 225 QI_DONE
266 * 7: super page
267 * 8-11: available
268 * 12-63: Host physcial address
269 */
270struct dma_pte {
271 u64 val;
272}; 226};
273#define dma_clear_pte(p) do {(p).val = 0;} while (0)
274
275#define DMA_PTE_READ (1)
276#define DMA_PTE_WRITE (2)
277 227
278#define dma_set_pte_readable(p) do {(p).val |= DMA_PTE_READ;} while (0) 228#define QI_CC_TYPE 0x1
279#define dma_set_pte_writable(p) do {(p).val |= DMA_PTE_WRITE;} while (0) 229#define QI_IOTLB_TYPE 0x2
280#define dma_set_pte_prot(p, prot) \ 230#define QI_DIOTLB_TYPE 0x3
281 do {(p).val = ((p).val & ~3) | ((prot) & 3); } while (0) 231#define QI_IEC_TYPE 0x4
282#define dma_pte_addr(p) ((p).val & PAGE_MASK_4K) 232#define QI_IWD_TYPE 0x5
283#define dma_set_pte_addr(p, addr) do {\
284 (p).val |= ((addr) & PAGE_MASK_4K); } while (0)
285#define dma_pte_present(p) (((p).val & 3) != 0)
286 233
287struct intel_iommu; 234#define QI_IEC_SELECTIVE (((u64)1) << 4)
235#define QI_IEC_IIDEX(idx) (((u64)(idx & 0xffff) << 32))
236#define QI_IEC_IM(m) (((u64)(m & 0x1f) << 27))
288 237
289struct dmar_domain { 238#define QI_IWD_STATUS_DATA(d) (((u64)d) << 32)
290 int id; /* domain id */ 239#define QI_IWD_STATUS_WRITE (((u64)1) << 5)
291 struct intel_iommu *iommu; /* back pointer to owning iommu */
292 240
293 struct list_head devices; /* all devices' list */ 241struct qi_desc {
294 struct iova_domain iovad; /* iova's that belong to this domain */ 242 u64 low, high;
243};
295 244
296 struct dma_pte *pgd; /* virtual address */ 245struct q_inval {
297 spinlock_t mapping_lock; /* page table lock */ 246 spinlock_t q_lock;
298 int gaw; /* max guest address width */ 247 struct qi_desc *desc; /* invalidation queue */
248 int *desc_status; /* desc status */
249 int free_head; /* first free entry */
250 int free_tail; /* last free entry */
251 int free_cnt;
252};
299 253
300 /* adjusted guest address width, 0 is level 2 30-bit */ 254#ifdef CONFIG_INTR_REMAP
301 int agaw; 255/* 1MB - maximum possible interrupt remapping table size */
256#define INTR_REMAP_PAGE_ORDER 8
257#define INTR_REMAP_TABLE_REG_SIZE 0xf
302 258
303#define DOMAIN_FLAG_MULTIPLE_DEVICES 1 259#define INTR_REMAP_TABLE_ENTRIES 65536
304 int flags;
305};
306 260
307/* PCI domain-device relationship */ 261struct ir_table {
308struct device_domain_info { 262 struct irte *base;
309 struct list_head link; /* link to domain siblings */
310 struct list_head global; /* link to global list */
311 u8 bus; /* PCI bus numer */
312 u8 devfn; /* PCI devfn number */
313 struct pci_dev *dev; /* it's NULL for PCIE-to-PCI bridge */
314 struct dmar_domain *domain; /* pointer to domain */
315}; 263};
316 264#endif
317extern int init_dmars(void);
318 265
319struct intel_iommu { 266struct intel_iommu {
320 void __iomem *reg; /* Pointer to hardware regs, virtual addr */ 267 void __iomem *reg; /* Pointer to hardware regs, virtual addr */
321 u64 cap; 268 u64 cap;
322 u64 ecap; 269 u64 ecap;
323 unsigned long *domain_ids; /* bitmap of domains */
324 struct dmar_domain **domains; /* ptr to domains */
325 int seg; 270 int seg;
326 u32 gcmd; /* Holds TE, EAFL. Don't need SRTP, SFL, WBF */ 271 u32 gcmd; /* Holds TE, EAFL. Don't need SRTP, SFL, WBF */
327 spinlock_t lock; /* protect context, domain ids */
328 spinlock_t register_lock; /* protect register handling */ 272 spinlock_t register_lock; /* protect register handling */
273 int seq_id; /* sequence id of the iommu */
274
275#ifdef CONFIG_DMAR
276 unsigned long *domain_ids; /* bitmap of domains */
277 struct dmar_domain **domains; /* ptr to domains */
278 spinlock_t lock; /* protect context, domain ids */
329 struct root_entry *root_entry; /* virtual address */ 279 struct root_entry *root_entry; /* virtual address */
330 280
331 unsigned int irq; 281 unsigned int irq;
332 unsigned char name[7]; /* Device Name */ 282 unsigned char name[7]; /* Device Name */
333 struct msi_msg saved_msg; 283 struct msi_msg saved_msg;
334 struct sys_device sysdev; 284 struct sys_device sysdev;
285#endif
286 struct q_inval *qi; /* Queued invalidation info */
287#ifdef CONFIG_INTR_REMAP
288 struct ir_table *ir_table; /* Interrupt remapping info */
289#endif
335}; 290};
336 291
337#ifndef CONFIG_DMAR_GFX_WA 292static inline void __iommu_flush_cache(
338static inline void iommu_prepare_gfx_mapping(void) 293 struct intel_iommu *iommu, void *addr, int size)
339{ 294{
340 return; 295 if (!ecap_coherent(iommu->ecap))
296 clflush_cache_range(addr, size);
341} 297}
342#endif /* !CONFIG_DMAR_GFX_WA */
343 298
299extern struct dmar_drhd_unit * dmar_find_matched_drhd_unit(struct pci_dev *dev);
300
301extern int alloc_iommu(struct dmar_drhd_unit *drhd);
302extern void free_iommu(struct intel_iommu *iommu);
303extern int dmar_enable_qi(struct intel_iommu *iommu);
304extern void qi_global_iec(struct intel_iommu *iommu);
305
306extern void qi_submit_sync(struct qi_desc *desc, struct intel_iommu *iommu);
344#endif 307#endif
diff --git a/drivers/pci/intr_remapping.c b/drivers/pci/intr_remapping.c
new file mode 100644
index 000000000000..bb642cc5e18c
--- /dev/null
+++ b/drivers/pci/intr_remapping.c
@@ -0,0 +1,471 @@
1#include <linux/dmar.h>
2#include <linux/spinlock.h>
3#include <linux/jiffies.h>
4#include <linux/pci.h>
5#include <linux/irq.h>
6#include <asm/io_apic.h>
7#include "intel-iommu.h"
8#include "intr_remapping.h"
9
10static struct ioapic_scope ir_ioapic[MAX_IO_APICS];
11static int ir_ioapic_num;
12int intr_remapping_enabled;
13
14static struct {
15 struct intel_iommu *iommu;
16 u16 irte_index;
17 u16 sub_handle;
18 u8 irte_mask;
19} irq_2_iommu[NR_IRQS];
20
21static DEFINE_SPINLOCK(irq_2_ir_lock);
22
23int irq_remapped(int irq)
24{
25 if (irq > NR_IRQS)
26 return 0;
27
28 if (!irq_2_iommu[irq].iommu)
29 return 0;
30
31 return 1;
32}
33
34int get_irte(int irq, struct irte *entry)
35{
36 int index;
37
38 if (!entry || irq > NR_IRQS)
39 return -1;
40
41 spin_lock(&irq_2_ir_lock);
42 if (!irq_2_iommu[irq].iommu) {
43 spin_unlock(&irq_2_ir_lock);
44 return -1;
45 }
46
47 index = irq_2_iommu[irq].irte_index + irq_2_iommu[irq].sub_handle;
48 *entry = *(irq_2_iommu[irq].iommu->ir_table->base + index);
49
50 spin_unlock(&irq_2_ir_lock);
51 return 0;
52}
53
54int alloc_irte(struct intel_iommu *iommu, int irq, u16 count)
55{
56 struct ir_table *table = iommu->ir_table;
57 u16 index, start_index;
58 unsigned int mask = 0;
59 int i;
60
61 if (!count)
62 return -1;
63
64 /*
65 * start the IRTE search from index 0.
66 */
67 index = start_index = 0;
68
69 if (count > 1) {
70 count = __roundup_pow_of_two(count);
71 mask = ilog2(count);
72 }
73
74 if (mask > ecap_max_handle_mask(iommu->ecap)) {
75 printk(KERN_ERR
76 "Requested mask %x exceeds the max invalidation handle"
77 " mask value %Lx\n", mask,
78 ecap_max_handle_mask(iommu->ecap));
79 return -1;
80 }
81
82 spin_lock(&irq_2_ir_lock);
83 do {
84 for (i = index; i < index + count; i++)
85 if (table->base[i].present)
86 break;
87 /* empty index found */
88 if (i == index + count)
89 break;
90
91 index = (index + count) % INTR_REMAP_TABLE_ENTRIES;
92
93 if (index == start_index) {
94 spin_unlock(&irq_2_ir_lock);
95 printk(KERN_ERR "can't allocate an IRTE\n");
96 return -1;
97 }
98 } while (1);
99
100 for (i = index; i < index + count; i++)
101 table->base[i].present = 1;
102
103 irq_2_iommu[irq].iommu = iommu;
104 irq_2_iommu[irq].irte_index = index;
105 irq_2_iommu[irq].sub_handle = 0;
106 irq_2_iommu[irq].irte_mask = mask;
107
108 spin_unlock(&irq_2_ir_lock);
109
110 return index;
111}
112
113static void qi_flush_iec(struct intel_iommu *iommu, int index, int mask)
114{
115 struct qi_desc desc;
116
117 desc.low = QI_IEC_IIDEX(index) | QI_IEC_TYPE | QI_IEC_IM(mask)
118 | QI_IEC_SELECTIVE;
119 desc.high = 0;
120
121 qi_submit_sync(&desc, iommu);
122}
123
124int map_irq_to_irte_handle(int irq, u16 *sub_handle)
125{
126 int index;
127
128 spin_lock(&irq_2_ir_lock);
129 if (irq >= NR_IRQS || !irq_2_iommu[irq].iommu) {
130 spin_unlock(&irq_2_ir_lock);
131 return -1;
132 }
133
134 *sub_handle = irq_2_iommu[irq].sub_handle;
135 index = irq_2_iommu[irq].irte_index;
136 spin_unlock(&irq_2_ir_lock);
137 return index;
138}
139
140int set_irte_irq(int irq, struct intel_iommu *iommu, u16 index, u16 subhandle)
141{
142 spin_lock(&irq_2_ir_lock);
143 if (irq >= NR_IRQS || irq_2_iommu[irq].iommu) {
144 spin_unlock(&irq_2_ir_lock);
145 return -1;
146 }
147
148 irq_2_iommu[irq].iommu = iommu;
149 irq_2_iommu[irq].irte_index = index;
150 irq_2_iommu[irq].sub_handle = subhandle;
151 irq_2_iommu[irq].irte_mask = 0;
152
153 spin_unlock(&irq_2_ir_lock);
154
155 return 0;
156}
157
158int clear_irte_irq(int irq, struct intel_iommu *iommu, u16 index)
159{
160 spin_lock(&irq_2_ir_lock);
161 if (irq >= NR_IRQS || !irq_2_iommu[irq].iommu) {
162 spin_unlock(&irq_2_ir_lock);
163 return -1;
164 }
165
166 irq_2_iommu[irq].iommu = NULL;
167 irq_2_iommu[irq].irte_index = 0;
168 irq_2_iommu[irq].sub_handle = 0;
169 irq_2_iommu[irq].irte_mask = 0;
170
171 spin_unlock(&irq_2_ir_lock);
172
173 return 0;
174}
175
176int modify_irte(int irq, struct irte *irte_modified)
177{
178 int index;
179 struct irte *irte;
180 struct intel_iommu *iommu;
181
182 spin_lock(&irq_2_ir_lock);
183 if (irq >= NR_IRQS || !irq_2_iommu[irq].iommu) {
184 spin_unlock(&irq_2_ir_lock);
185 return -1;
186 }
187
188 iommu = irq_2_iommu[irq].iommu;
189
190 index = irq_2_iommu[irq].irte_index + irq_2_iommu[irq].sub_handle;
191 irte = &iommu->ir_table->base[index];
192
193 set_64bit((unsigned long *)irte, irte_modified->low | (1 << 1));
194 __iommu_flush_cache(iommu, irte, sizeof(*irte));
195
196 qi_flush_iec(iommu, index, 0);
197
198 spin_unlock(&irq_2_ir_lock);
199 return 0;
200}
201
202int flush_irte(int irq)
203{
204 int index;
205 struct intel_iommu *iommu;
206
207 spin_lock(&irq_2_ir_lock);
208 if (irq >= NR_IRQS || !irq_2_iommu[irq].iommu) {
209 spin_unlock(&irq_2_ir_lock);
210 return -1;
211 }
212
213 iommu = irq_2_iommu[irq].iommu;
214
215 index = irq_2_iommu[irq].irte_index + irq_2_iommu[irq].sub_handle;
216
217 qi_flush_iec(iommu, index, irq_2_iommu[irq].irte_mask);
218 spin_unlock(&irq_2_ir_lock);
219
220 return 0;
221}
222
223struct intel_iommu *map_ioapic_to_ir(int apic)
224{
225 int i;
226
227 for (i = 0; i < MAX_IO_APICS; i++)
228 if (ir_ioapic[i].id == apic)
229 return ir_ioapic[i].iommu;
230 return NULL;
231}
232
233struct intel_iommu *map_dev_to_ir(struct pci_dev *dev)
234{
235 struct dmar_drhd_unit *drhd;
236
237 drhd = dmar_find_matched_drhd_unit(dev);
238 if (!drhd)
239 return NULL;
240
241 return drhd->iommu;
242}
243
244int free_irte(int irq)
245{
246 int index, i;
247 struct irte *irte;
248 struct intel_iommu *iommu;
249
250 spin_lock(&irq_2_ir_lock);
251 if (irq >= NR_IRQS || !irq_2_iommu[irq].iommu) {
252 spin_unlock(&irq_2_ir_lock);
253 return -1;
254 }
255
256 iommu = irq_2_iommu[irq].iommu;
257
258 index = irq_2_iommu[irq].irte_index + irq_2_iommu[irq].sub_handle;
259 irte = &iommu->ir_table->base[index];
260
261 if (!irq_2_iommu[irq].sub_handle) {
262 for (i = 0; i < (1 << irq_2_iommu[irq].irte_mask); i++)
263 set_64bit((unsigned long *)irte, 0);
264 qi_flush_iec(iommu, index, irq_2_iommu[irq].irte_mask);
265 }
266
267 irq_2_iommu[irq].iommu = NULL;
268 irq_2_iommu[irq].irte_index = 0;
269 irq_2_iommu[irq].sub_handle = 0;
270 irq_2_iommu[irq].irte_mask = 0;
271
272 spin_unlock(&irq_2_ir_lock);
273
274 return 0;
275}
276
277static void iommu_set_intr_remapping(struct intel_iommu *iommu, int mode)
278{
279 u64 addr;
280 u32 cmd, sts;
281 unsigned long flags;
282
283 addr = virt_to_phys((void *)iommu->ir_table->base);
284
285 spin_lock_irqsave(&iommu->register_lock, flags);
286
287 dmar_writeq(iommu->reg + DMAR_IRTA_REG,
288 (addr) | IR_X2APIC_MODE(mode) | INTR_REMAP_TABLE_REG_SIZE);
289
290 /* Set interrupt-remapping table pointer */
291 cmd = iommu->gcmd | DMA_GCMD_SIRTP;
292 writel(cmd, iommu->reg + DMAR_GCMD_REG);
293
294 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
295 readl, (sts & DMA_GSTS_IRTPS), sts);
296 spin_unlock_irqrestore(&iommu->register_lock, flags);
297
298 /*
299 * global invalidation of interrupt entry cache before enabling
300 * interrupt-remapping.
301 */
302 qi_global_iec(iommu);
303
304 spin_lock_irqsave(&iommu->register_lock, flags);
305
306 /* Enable interrupt-remapping */
307 cmd = iommu->gcmd | DMA_GCMD_IRE;
308 iommu->gcmd |= DMA_GCMD_IRE;
309 writel(cmd, iommu->reg + DMAR_GCMD_REG);
310
311 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
312 readl, (sts & DMA_GSTS_IRES), sts);
313
314 spin_unlock_irqrestore(&iommu->register_lock, flags);
315}
316
317
318static int setup_intr_remapping(struct intel_iommu *iommu, int mode)
319{
320 struct ir_table *ir_table;
321 struct page *pages;
322
323 ir_table = iommu->ir_table = kzalloc(sizeof(struct ir_table),
324 GFP_KERNEL);
325
326 if (!iommu->ir_table)
327 return -ENOMEM;
328
329 pages = alloc_pages(GFP_KERNEL | __GFP_ZERO, INTR_REMAP_PAGE_ORDER);
330
331 if (!pages) {
332 printk(KERN_ERR "failed to allocate pages of order %d\n",
333 INTR_REMAP_PAGE_ORDER);
334 kfree(iommu->ir_table);
335 return -ENOMEM;
336 }
337
338 ir_table->base = page_address(pages);
339
340 iommu_set_intr_remapping(iommu, mode);
341 return 0;
342}
343
344int __init enable_intr_remapping(int eim)
345{
346 struct dmar_drhd_unit *drhd;
347 int setup = 0;
348
349 /*
350 * check for the Interrupt-remapping support
351 */
352 for_each_drhd_unit(drhd) {
353 struct intel_iommu *iommu = drhd->iommu;
354
355 if (!ecap_ir_support(iommu->ecap))
356 continue;
357
358 if (eim && !ecap_eim_support(iommu->ecap)) {
359 printk(KERN_INFO "DRHD %Lx: EIM not supported by DRHD, "
360 " ecap %Lx\n", drhd->reg_base_addr, iommu->ecap);
361 return -1;
362 }
363 }
364
365 /*
366 * Enable queued invalidation for all the DRHD's.
367 */
368 for_each_drhd_unit(drhd) {
369 int ret;
370 struct intel_iommu *iommu = drhd->iommu;
371 ret = dmar_enable_qi(iommu);
372
373 if (ret) {
374 printk(KERN_ERR "DRHD %Lx: failed to enable queued, "
375 " invalidation, ecap %Lx, ret %d\n",
376 drhd->reg_base_addr, iommu->ecap, ret);
377 return -1;
378 }
379 }
380
381 /*
382 * Setup Interrupt-remapping for all the DRHD's now.
383 */
384 for_each_drhd_unit(drhd) {
385 struct intel_iommu *iommu = drhd->iommu;
386
387 if (!ecap_ir_support(iommu->ecap))
388 continue;
389
390 if (setup_intr_remapping(iommu, eim))
391 goto error;
392
393 setup = 1;
394 }
395
396 if (!setup)
397 goto error;
398
399 intr_remapping_enabled = 1;
400
401 return 0;
402
403error:
404 /*
405 * handle error condition gracefully here!
406 */
407 return -1;
408}
409
410static int ir_parse_ioapic_scope(struct acpi_dmar_header *header,
411 struct intel_iommu *iommu)
412{
413 struct acpi_dmar_hardware_unit *drhd;
414 struct acpi_dmar_device_scope *scope;
415 void *start, *end;
416
417 drhd = (struct acpi_dmar_hardware_unit *)header;
418
419 start = (void *)(drhd + 1);
420 end = ((void *)drhd) + header->length;
421
422 while (start < end) {
423 scope = start;
424 if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_IOAPIC) {
425 if (ir_ioapic_num == MAX_IO_APICS) {
426 printk(KERN_WARNING "Exceeded Max IO APICS\n");
427 return -1;
428 }
429
430 printk(KERN_INFO "IOAPIC id %d under DRHD base"
431 " 0x%Lx\n", scope->enumeration_id,
432 drhd->address);
433
434 ir_ioapic[ir_ioapic_num].iommu = iommu;
435 ir_ioapic[ir_ioapic_num].id = scope->enumeration_id;
436 ir_ioapic_num++;
437 }
438 start += scope->length;
439 }
440
441 return 0;
442}
443
444/*
445 * Finds the assocaition between IOAPIC's and its Interrupt-remapping
446 * hardware unit.
447 */
448int __init parse_ioapics_under_ir(void)
449{
450 struct dmar_drhd_unit *drhd;
451 int ir_supported = 0;
452
453 for_each_drhd_unit(drhd) {
454 struct intel_iommu *iommu = drhd->iommu;
455
456 if (ecap_ir_support(iommu->ecap)) {
457 if (ir_parse_ioapic_scope(drhd->hdr, iommu))
458 return -1;
459
460 ir_supported = 1;
461 }
462 }
463
464 if (ir_supported && ir_ioapic_num != nr_ioapics) {
465 printk(KERN_WARNING
466 "Not all IO-APIC's listed under remapping hardware\n");
467 return -1;
468 }
469
470 return ir_supported;
471}
diff --git a/drivers/pci/intr_remapping.h b/drivers/pci/intr_remapping.h
new file mode 100644
index 000000000000..05f2635bbe4e
--- /dev/null
+++ b/drivers/pci/intr_remapping.h
@@ -0,0 +1,8 @@
1#include "intel-iommu.h"
2
3struct ioapic_scope {
4 struct intel_iommu *iommu;
5 unsigned int id;
6};
7
8#define IR_X2APIC_MODE(mode) (mode ? (1 << 11) : 0)
diff --git a/drivers/pci/pci-sysfs.c b/drivers/pci/pci-sysfs.c
index 9c718583a237..77baff022f71 100644
--- a/drivers/pci/pci-sysfs.c
+++ b/drivers/pci/pci-sysfs.c
@@ -16,6 +16,7 @@
16 16
17 17
18#include <linux/kernel.h> 18#include <linux/kernel.h>
19#include <linux/sched.h>
19#include <linux/pci.h> 20#include <linux/pci.h>
20#include <linux/stat.h> 21#include <linux/stat.h>
21#include <linux/topology.h> 22#include <linux/topology.h>
@@ -484,6 +485,21 @@ pci_mmap_legacy_mem(struct kobject *kobj, struct bin_attribute *attr,
484#endif /* HAVE_PCI_LEGACY */ 485#endif /* HAVE_PCI_LEGACY */
485 486
486#ifdef HAVE_PCI_MMAP 487#ifdef HAVE_PCI_MMAP
488
489static int pci_mmap_fits(struct pci_dev *pdev, int resno, struct vm_area_struct *vma)
490{
491 unsigned long nr, start, size;
492
493 nr = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
494 start = vma->vm_pgoff;
495 size = pci_resource_len(pdev, resno) >> PAGE_SHIFT;
496 if (start < size && size - start >= nr)
497 return 1;
498 WARN(1, "process \"%s\" tried to map 0x%08lx-0x%08lx on %s BAR %d (size 0x%08lx)\n",
499 current->comm, start, start+nr, pci_name(pdev), resno, size);
500 return 0;
501}
502
487/** 503/**
488 * pci_mmap_resource - map a PCI resource into user memory space 504 * pci_mmap_resource - map a PCI resource into user memory space
489 * @kobj: kobject for mapping 505 * @kobj: kobject for mapping
@@ -510,6 +526,9 @@ pci_mmap_resource(struct kobject *kobj, struct bin_attribute *attr,
510 if (i >= PCI_ROM_RESOURCE) 526 if (i >= PCI_ROM_RESOURCE)
511 return -ENODEV; 527 return -ENODEV;
512 528
529 if (!pci_mmap_fits(pdev, i, vma))
530 return -EINVAL;
531
513 /* pci_mmap_page_range() expects the same kind of entry as coming 532 /* pci_mmap_page_range() expects the same kind of entry as coming
514 * from /proc/bus/pci/ which is a "user visible" value. If this is 533 * from /proc/bus/pci/ which is a "user visible" value. If this is
515 * different from the resource itself, arch will do necessary fixup. 534 * different from the resource itself, arch will do necessary fixup.
diff --git a/drivers/pci/pcie/aspm.c b/drivers/pci/pcie/aspm.c
index 9a7c9e1408a4..851f5b83cdbc 100644
--- a/drivers/pci/pcie/aspm.c
+++ b/drivers/pci/pcie/aspm.c
@@ -527,7 +527,7 @@ static int pcie_aspm_sanity_check(struct pci_dev *pdev)
527 */ 527 */
528 pci_read_config_dword(child_dev, child_pos + PCI_EXP_DEVCAP, 528 pci_read_config_dword(child_dev, child_pos + PCI_EXP_DEVCAP,
529 &reg32); 529 &reg32);
530 if (!(reg32 & PCI_EXP_DEVCAP_RBER && !aspm_force)) { 530 if (!(reg32 & PCI_EXP_DEVCAP_RBER) && !aspm_force) {
531 printk("Pre-1.1 PCIe device detected, " 531 printk("Pre-1.1 PCIe device detected, "
532 "disable ASPM for %s. It can be enabled forcedly" 532 "disable ASPM for %s. It can be enabled forcedly"
533 " with 'pcie_aspm=force'\n", pci_name(pdev)); 533 " with 'pcie_aspm=force'\n", pci_name(pdev));
diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
index cce2f4cb1fbf..36698e57b97f 100644
--- a/drivers/pci/probe.c
+++ b/drivers/pci/probe.c
@@ -304,6 +304,9 @@ static int __pci_read_base(struct pci_dev *dev, enum pci_bar_type type,
304 } else { 304 } else {
305 res->start = l64; 305 res->start = l64;
306 res->end = l64 + sz64; 306 res->end = l64 + sz64;
307 printk(KERN_DEBUG "PCI: %s reg %x 64bit mmio: [%llx, %llx]\n",
308 pci_name(dev), pos, (unsigned long long)res->start,
309 (unsigned long long)res->end);
307 } 310 }
308 } else { 311 } else {
309 sz = pci_size(l, sz, mask); 312 sz = pci_size(l, sz, mask);
@@ -313,6 +316,9 @@ static int __pci_read_base(struct pci_dev *dev, enum pci_bar_type type,
313 316
314 res->start = l; 317 res->start = l;
315 res->end = l + sz; 318 res->end = l + sz;
319 printk(KERN_DEBUG "PCI: %s reg %x %s: [%llx, %llx]\n", pci_name(dev),
320 pos, (res->flags & IORESOURCE_IO) ? "io port":"32bit mmio",
321 (unsigned long long)res->start, (unsigned long long)res->end);
316 } 322 }
317 323
318 out: 324 out:
@@ -383,7 +389,9 @@ void __devinit pci_read_bridge_bases(struct pci_bus *child)
383 res->start = base; 389 res->start = base;
384 if (!res->end) 390 if (!res->end)
385 res->end = limit + 0xfff; 391 res->end = limit + 0xfff;
386 printk(KERN_INFO "PCI: bridge %s io port: [%llx, %llx]\n", pci_name(dev), res->start, res->end); 392 printk(KERN_DEBUG "PCI: bridge %s io port: [%llx, %llx]\n",
393 pci_name(dev), (unsigned long long) res->start,
394 (unsigned long long) res->end);
387 } 395 }
388 396
389 res = child->resource[1]; 397 res = child->resource[1];
@@ -395,7 +403,9 @@ void __devinit pci_read_bridge_bases(struct pci_bus *child)
395 res->flags = (mem_base_lo & PCI_MEMORY_RANGE_TYPE_MASK) | IORESOURCE_MEM; 403 res->flags = (mem_base_lo & PCI_MEMORY_RANGE_TYPE_MASK) | IORESOURCE_MEM;
396 res->start = base; 404 res->start = base;
397 res->end = limit + 0xfffff; 405 res->end = limit + 0xfffff;
398 printk(KERN_INFO "PCI: bridge %s 32bit mmio: [%llx, %llx]\n", pci_name(dev), res->start, res->end); 406 printk(KERN_DEBUG "PCI: bridge %s 32bit mmio: [%llx, %llx]\n",
407 pci_name(dev), (unsigned long long) res->start,
408 (unsigned long long) res->end);
399 } 409 }
400 410
401 res = child->resource[2]; 411 res = child->resource[2];
@@ -431,7 +441,9 @@ void __devinit pci_read_bridge_bases(struct pci_bus *child)
431 res->flags = (mem_base_lo & PCI_MEMORY_RANGE_TYPE_MASK) | IORESOURCE_MEM | IORESOURCE_PREFETCH; 441 res->flags = (mem_base_lo & PCI_MEMORY_RANGE_TYPE_MASK) | IORESOURCE_MEM | IORESOURCE_PREFETCH;
432 res->start = base; 442 res->start = base;
433 res->end = limit + 0xfffff; 443 res->end = limit + 0xfffff;
434 printk(KERN_INFO "PCI: bridge %s %sbit mmio pref: [%llx, %llx]\n", pci_name(dev), (res->flags & PCI_PREF_RANGE_TYPE_64)?"64":"32",res->start, res->end); 444 printk(KERN_DEBUG "PCI: bridge %s %sbit mmio pref: [%llx, %llx]\n",
445 pci_name(dev), (res->flags & PCI_PREF_RANGE_TYPE_64) ? "64" : "32",
446 (unsigned long long) res->start, (unsigned long long) res->end);
435 } 447 }
436} 448}
437 449
diff --git a/drivers/pci/search.c b/drivers/pci/search.c
index 3b3b5f178797..4edfc4731bd4 100644
--- a/drivers/pci/search.c
+++ b/drivers/pci/search.c
@@ -162,7 +162,7 @@ EXPORT_SYMBOL(pci_find_slot);
162 * time. 162 * time.
163 */ 163 */
164struct pci_dev *pci_find_device(unsigned int vendor, unsigned int device, 164struct pci_dev *pci_find_device(unsigned int vendor, unsigned int device,
165 const struct pci_dev *from) 165 struct pci_dev *from)
166{ 166{
167 struct pci_dev *pdev; 167 struct pci_dev *pdev;
168 168
@@ -263,7 +263,7 @@ static int match_pci_dev_by_id(struct device *dev, void *data)
263 * this file. 263 * this file.
264 */ 264 */
265static struct pci_dev *pci_get_dev_by_id(const struct pci_device_id *id, 265static struct pci_dev *pci_get_dev_by_id(const struct pci_device_id *id,
266 const struct pci_dev *from) 266 struct pci_dev *from)
267{ 267{
268 struct device *dev; 268 struct device *dev;
269 struct device *dev_start = NULL; 269 struct device *dev_start = NULL;
@@ -303,7 +303,7 @@ static struct pci_dev *pci_get_dev_by_id(const struct pci_device_id *id,
303 */ 303 */
304struct pci_dev *pci_get_subsys(unsigned int vendor, unsigned int device, 304struct pci_dev *pci_get_subsys(unsigned int vendor, unsigned int device,
305 unsigned int ss_vendor, unsigned int ss_device, 305 unsigned int ss_vendor, unsigned int ss_device,
306 const struct pci_dev *from) 306 struct pci_dev *from)
307{ 307{
308 struct pci_dev *pdev; 308 struct pci_dev *pdev;
309 struct pci_device_id *id; 309 struct pci_device_id *id;
diff --git a/drivers/pci/setup-bus.c b/drivers/pci/setup-bus.c
index 1aad599816f7..3abbfad9ddab 100644
--- a/drivers/pci/setup-bus.c
+++ b/drivers/pci/setup-bus.c
@@ -540,7 +540,11 @@ static void pci_bus_dump_res(struct pci_bus *bus)
540 if (!res) 540 if (!res)
541 continue; 541 continue;
542 542
543 printk(KERN_INFO "bus: %02x index %x %s: [%llx, %llx]\n", bus->number, i, (res->flags & IORESOURCE_IO)? "io port":"mmio", res->start, res->end); 543 printk(KERN_INFO "bus: %02x index %x %s: [%llx, %llx]\n",
544 bus->number, i,
545 (res->flags & IORESOURCE_IO) ? "io port" : "mmio",
546 (unsigned long long) res->start,
547 (unsigned long long) res->end);
544 } 548 }
545} 549}
546 550
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-27 02:17:54 -0500