]> git.karo-electronics.de Git - karo-tx-linux.git/blob - drivers/pci/msi.c
[PATCH] PCI: per-platform IA64_{FIRST,LAST}_DEVICE_VECTOR definitions
[karo-tx-linux.git] / drivers / pci / msi.c
1 /*
2  * File:        msi.c
3  * Purpose:     PCI Message Signaled Interrupt (MSI)
4  *
5  * Copyright (C) 2003-2004 Intel
6  * Copyright (C) Tom Long Nguyen (tom.l.nguyen@intel.com)
7  */
8
9 #include <linux/mm.h>
10 #include <linux/irq.h>
11 #include <linux/interrupt.h>
12 #include <linux/init.h>
13 #include <linux/config.h>
14 #include <linux/ioport.h>
15 #include <linux/smp_lock.h>
16 #include <linux/pci.h>
17 #include <linux/proc_fs.h>
18
19 #include <asm/errno.h>
20 #include <asm/io.h>
21 #include <asm/smp.h>
22
23 #include "pci.h"
24 #include "msi.h"
25
26 static DEFINE_SPINLOCK(msi_lock);
27 static struct msi_desc* msi_desc[NR_IRQS] = { [0 ... NR_IRQS-1] = NULL };
28 static kmem_cache_t* msi_cachep;
29
30 static int pci_msi_enable = 1;
31 static int last_alloc_vector;
32 static int nr_released_vectors;
33 static int nr_reserved_vectors = NR_HP_RESERVED_VECTORS;
34 static int nr_msix_devices;
35
36 #ifndef CONFIG_X86_IO_APIC
37 int vector_irq[NR_VECTORS] = { [0 ... NR_VECTORS - 1] = -1};
38 u8 irq_vector[NR_IRQ_VECTORS];
39 #endif
40
41 static struct msi_ops *msi_ops;
42
43 int
44 msi_register(struct msi_ops *ops)
45 {
46         msi_ops = ops;
47         return 0;
48 }
49
50 static void msi_cache_ctor(void *p, kmem_cache_t *cache, unsigned long flags)
51 {
52         memset(p, 0, NR_IRQS * sizeof(struct msi_desc));
53 }
54
55 static int msi_cache_init(void)
56 {
57         msi_cachep = kmem_cache_create("msi_cache",
58                         NR_IRQS * sizeof(struct msi_desc),
59                         0, SLAB_HWCACHE_ALIGN, msi_cache_ctor, NULL);
60         if (!msi_cachep)
61                 return -ENOMEM;
62
63         return 0;
64 }
65
66 static void msi_set_mask_bit(unsigned int vector, int flag)
67 {
68         struct msi_desc *entry;
69
70         entry = (struct msi_desc *)msi_desc[vector];
71         if (!entry || !entry->dev || !entry->mask_base)
72                 return;
73         switch (entry->msi_attrib.type) {
74         case PCI_CAP_ID_MSI:
75         {
76                 int             pos;
77                 u32             mask_bits;
78
79                 pos = (long)entry->mask_base;
80                 pci_read_config_dword(entry->dev, pos, &mask_bits);
81                 mask_bits &= ~(1);
82                 mask_bits |= flag;
83                 pci_write_config_dword(entry->dev, pos, mask_bits);
84                 break;
85         }
86         case PCI_CAP_ID_MSIX:
87         {
88                 int offset = entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
89                         PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET;
90                 writel(flag, entry->mask_base + offset);
91                 break;
92         }
93         default:
94                 break;
95         }
96 }
97
98 #ifdef CONFIG_SMP
99 static void set_msi_affinity(unsigned int vector, cpumask_t cpu_mask)
100 {
101         struct msi_desc *entry;
102         u32 address_hi, address_lo;
103         unsigned int irq = vector;
104         unsigned int dest_cpu = first_cpu(cpu_mask);
105
106         entry = (struct msi_desc *)msi_desc[vector];
107         if (!entry || !entry->dev)
108                 return;
109
110         switch (entry->msi_attrib.type) {
111         case PCI_CAP_ID_MSI:
112         {
113                 int pos = pci_find_capability(entry->dev, PCI_CAP_ID_MSI);
114
115                 if (!pos)
116                         return;
117
118                 pci_read_config_dword(entry->dev, msi_upper_address_reg(pos),
119                         &address_hi);
120                 pci_read_config_dword(entry->dev, msi_lower_address_reg(pos),
121                         &address_lo);
122
123                 msi_ops->target(vector, dest_cpu, &address_hi, &address_lo);
124
125                 pci_write_config_dword(entry->dev, msi_upper_address_reg(pos),
126                         address_hi);
127                 pci_write_config_dword(entry->dev, msi_lower_address_reg(pos),
128                         address_lo);
129                 set_native_irq_info(irq, cpu_mask);
130                 break;
131         }
132         case PCI_CAP_ID_MSIX:
133         {
134                 int offset_hi =
135                         entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
136                                 PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET;
137                 int offset_lo =
138                         entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
139                                 PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET;
140
141                 address_hi = readl(entry->mask_base + offset_hi);
142                 address_lo = readl(entry->mask_base + offset_lo);
143
144                 msi_ops->target(vector, dest_cpu, &address_hi, &address_lo);
145
146                 writel(address_hi, entry->mask_base + offset_hi);
147                 writel(address_lo, entry->mask_base + offset_lo);
148                 set_native_irq_info(irq, cpu_mask);
149                 break;
150         }
151         default:
152                 break;
153         }
154 }
155 #else
156 #define set_msi_affinity NULL
157 #endif /* CONFIG_SMP */
158
159 static void mask_MSI_irq(unsigned int vector)
160 {
161         msi_set_mask_bit(vector, 1);
162 }
163
164 static void unmask_MSI_irq(unsigned int vector)
165 {
166         msi_set_mask_bit(vector, 0);
167 }
168
169 static unsigned int startup_msi_irq_wo_maskbit(unsigned int vector)
170 {
171         struct msi_desc *entry;
172         unsigned long flags;
173
174         spin_lock_irqsave(&msi_lock, flags);
175         entry = msi_desc[vector];
176         if (!entry || !entry->dev) {
177                 spin_unlock_irqrestore(&msi_lock, flags);
178                 return 0;
179         }
180         entry->msi_attrib.state = 1;    /* Mark it active */
181         spin_unlock_irqrestore(&msi_lock, flags);
182
183         return 0;       /* never anything pending */
184 }
185
186 static unsigned int startup_msi_irq_w_maskbit(unsigned int vector)
187 {
188         startup_msi_irq_wo_maskbit(vector);
189         unmask_MSI_irq(vector);
190         return 0;       /* never anything pending */
191 }
192
193 static void shutdown_msi_irq(unsigned int vector)
194 {
195         struct msi_desc *entry;
196         unsigned long flags;
197
198         spin_lock_irqsave(&msi_lock, flags);
199         entry = msi_desc[vector];
200         if (entry && entry->dev)
201                 entry->msi_attrib.state = 0;    /* Mark it not active */
202         spin_unlock_irqrestore(&msi_lock, flags);
203 }
204
205 static void end_msi_irq_wo_maskbit(unsigned int vector)
206 {
207         move_native_irq(vector);
208         ack_APIC_irq();
209 }
210
211 static void end_msi_irq_w_maskbit(unsigned int vector)
212 {
213         move_native_irq(vector);
214         unmask_MSI_irq(vector);
215         ack_APIC_irq();
216 }
217
218 static void do_nothing(unsigned int vector)
219 {
220 }
221
222 /*
223  * Interrupt Type for MSI-X PCI/PCI-X/PCI-Express Devices,
224  * which implement the MSI-X Capability Structure.
225  */
226 static struct hw_interrupt_type msix_irq_type = {
227         .typename       = "PCI-MSI-X",
228         .startup        = startup_msi_irq_w_maskbit,
229         .shutdown       = shutdown_msi_irq,
230         .enable         = unmask_MSI_irq,
231         .disable        = mask_MSI_irq,
232         .ack            = mask_MSI_irq,
233         .end            = end_msi_irq_w_maskbit,
234         .set_affinity   = set_msi_affinity
235 };
236
237 /*
238  * Interrupt Type for MSI PCI/PCI-X/PCI-Express Devices,
239  * which implement the MSI Capability Structure with
240  * Mask-and-Pending Bits.
241  */
242 static struct hw_interrupt_type msi_irq_w_maskbit_type = {
243         .typename       = "PCI-MSI",
244         .startup        = startup_msi_irq_w_maskbit,
245         .shutdown       = shutdown_msi_irq,
246         .enable         = unmask_MSI_irq,
247         .disable        = mask_MSI_irq,
248         .ack            = mask_MSI_irq,
249         .end            = end_msi_irq_w_maskbit,
250         .set_affinity   = set_msi_affinity
251 };
252
253 /*
254  * Interrupt Type for MSI PCI/PCI-X/PCI-Express Devices,
255  * which implement the MSI Capability Structure without
256  * Mask-and-Pending Bits.
257  */
258 static struct hw_interrupt_type msi_irq_wo_maskbit_type = {
259         .typename       = "PCI-MSI",
260         .startup        = startup_msi_irq_wo_maskbit,
261         .shutdown       = shutdown_msi_irq,
262         .enable         = do_nothing,
263         .disable        = do_nothing,
264         .ack            = do_nothing,
265         .end            = end_msi_irq_wo_maskbit,
266         .set_affinity   = set_msi_affinity
267 };
268
269 static int msi_free_vector(struct pci_dev* dev, int vector, int reassign);
270 static int assign_msi_vector(void)
271 {
272         static int new_vector_avail = 1;
273         int vector;
274         unsigned long flags;
275
276         /*
277          * msi_lock is provided to ensure that successful allocation of MSI
278          * vector is assigned unique among drivers.
279          */
280         spin_lock_irqsave(&msi_lock, flags);
281
282         if (!new_vector_avail) {
283                 int free_vector = 0;
284
285                 /*
286                  * vector_irq[] = -1 indicates that this specific vector is:
287                  * - assigned for MSI (since MSI have no associated IRQ) or
288                  * - assigned for legacy if less than 16, or
289                  * - having no corresponding 1:1 vector-to-IOxAPIC IRQ mapping
290                  * vector_irq[] = 0 indicates that this vector, previously
291                  * assigned for MSI, is freed by hotplug removed operations.
292                  * This vector will be reused for any subsequent hotplug added
293                  * operations.
294                  * vector_irq[] > 0 indicates that this vector is assigned for
295                  * IOxAPIC IRQs. This vector and its value provides a 1-to-1
296                  * vector-to-IOxAPIC IRQ mapping.
297                  */
298                 for (vector = FIRST_DEVICE_VECTOR; vector < NR_IRQS; vector++) {
299                         if (vector_irq[vector] != 0)
300                                 continue;
301                         free_vector = vector;
302                         if (!msi_desc[vector])
303                                 break;
304                         else
305                                 continue;
306                 }
307                 if (!free_vector) {
308                         spin_unlock_irqrestore(&msi_lock, flags);
309                         return -EBUSY;
310                 }
311                 vector_irq[free_vector] = -1;
312                 nr_released_vectors--;
313                 spin_unlock_irqrestore(&msi_lock, flags);
314                 if (msi_desc[free_vector] != NULL) {
315                         struct pci_dev *dev;
316                         int tail;
317
318                         /* free all linked vectors before re-assign */
319                         do {
320                                 spin_lock_irqsave(&msi_lock, flags);
321                                 dev = msi_desc[free_vector]->dev;
322                                 tail = msi_desc[free_vector]->link.tail;
323                                 spin_unlock_irqrestore(&msi_lock, flags);
324                                 msi_free_vector(dev, tail, 1);
325                         } while (free_vector != tail);
326                 }
327
328                 return free_vector;
329         }
330         vector = assign_irq_vector(AUTO_ASSIGN);
331         last_alloc_vector = vector;
332         if (vector  == LAST_DEVICE_VECTOR)
333                 new_vector_avail = 0;
334
335         spin_unlock_irqrestore(&msi_lock, flags);
336         return vector;
337 }
338
339 static int get_new_vector(void)
340 {
341         int vector = assign_msi_vector();
342
343         if (vector > 0)
344                 set_intr_gate(vector, interrupt[vector]);
345
346         return vector;
347 }
348
349 static int msi_init(void)
350 {
351         static int status = -ENOMEM;
352
353         if (!status)
354                 return status;
355
356         if (pci_msi_quirk) {
357                 pci_msi_enable = 0;
358                 printk(KERN_WARNING "PCI: MSI quirk detected. MSI disabled.\n");
359                 status = -EINVAL;
360                 return status;
361         }
362
363         status = msi_arch_init();
364         if (status < 0) {
365                 pci_msi_enable = 0;
366                 printk(KERN_WARNING
367                        "PCI: MSI arch init failed.  MSI disabled.\n");
368                 return status;
369         }
370
371         if (! msi_ops) {
372                 printk(KERN_WARNING
373                        "PCI: MSI ops not registered. MSI disabled.\n");
374                 status = -EINVAL;
375                 return status;
376         }
377
378         last_alloc_vector = assign_irq_vector(AUTO_ASSIGN);
379         status = msi_cache_init();
380         if (status < 0) {
381                 pci_msi_enable = 0;
382                 printk(KERN_WARNING "PCI: MSI cache init failed\n");
383                 return status;
384         }
385
386 #ifndef CONFIG_X86_IO_APIC
387         irq_vector[0] = FIRST_DEVICE_VECTOR;
388 #endif
389
390         if (last_alloc_vector < 0) {
391                 pci_msi_enable = 0;
392                 printk(KERN_WARNING "PCI: No interrupt vectors available for MSI\n");
393                 status = -EBUSY;
394                 return status;
395         }
396         vector_irq[last_alloc_vector] = 0;
397         nr_released_vectors++;
398
399         return status;
400 }
401
402 static int get_msi_vector(struct pci_dev *dev)
403 {
404         return get_new_vector();
405 }
406
407 static struct msi_desc* alloc_msi_entry(void)
408 {
409         struct msi_desc *entry;
410
411         entry = kmem_cache_alloc(msi_cachep, SLAB_KERNEL);
412         if (!entry)
413                 return NULL;
414
415         memset(entry, 0, sizeof(struct msi_desc));
416         entry->link.tail = entry->link.head = 0;        /* single message */
417         entry->dev = NULL;
418
419         return entry;
420 }
421
422 static void attach_msi_entry(struct msi_desc *entry, int vector)
423 {
424         unsigned long flags;
425
426         spin_lock_irqsave(&msi_lock, flags);
427         msi_desc[vector] = entry;
428         spin_unlock_irqrestore(&msi_lock, flags);
429 }
430
431 static void irq_handler_init(int cap_id, int pos, int mask)
432 {
433         unsigned long flags;
434
435         spin_lock_irqsave(&irq_desc[pos].lock, flags);
436         if (cap_id == PCI_CAP_ID_MSIX)
437                 irq_desc[pos].handler = &msix_irq_type;
438         else {
439                 if (!mask)
440                         irq_desc[pos].handler = &msi_irq_wo_maskbit_type;
441                 else
442                         irq_desc[pos].handler = &msi_irq_w_maskbit_type;
443         }
444         spin_unlock_irqrestore(&irq_desc[pos].lock, flags);
445 }
446
447 static void enable_msi_mode(struct pci_dev *dev, int pos, int type)
448 {
449         u16 control;
450
451         pci_read_config_word(dev, msi_control_reg(pos), &control);
452         if (type == PCI_CAP_ID_MSI) {
453                 /* Set enabled bits to single MSI & enable MSI_enable bit */
454                 msi_enable(control, 1);
455                 pci_write_config_word(dev, msi_control_reg(pos), control);
456         } else {
457                 msix_enable(control);
458                 pci_write_config_word(dev, msi_control_reg(pos), control);
459         }
460         if (pci_find_capability(dev, PCI_CAP_ID_EXP)) {
461                 /* PCI Express Endpoint device detected */
462                 pci_intx(dev, 0);  /* disable intx */
463         }
464 }
465
466 void disable_msi_mode(struct pci_dev *dev, int pos, int type)
467 {
468         u16 control;
469
470         pci_read_config_word(dev, msi_control_reg(pos), &control);
471         if (type == PCI_CAP_ID_MSI) {
472                 /* Set enabled bits to single MSI & enable MSI_enable bit */
473                 msi_disable(control);
474                 pci_write_config_word(dev, msi_control_reg(pos), control);
475         } else {
476                 msix_disable(control);
477                 pci_write_config_word(dev, msi_control_reg(pos), control);
478         }
479         if (pci_find_capability(dev, PCI_CAP_ID_EXP)) {
480                 /* PCI Express Endpoint device detected */
481                 pci_intx(dev, 1);  /* enable intx */
482         }
483 }
484
485 static int msi_lookup_vector(struct pci_dev *dev, int type)
486 {
487         int vector;
488         unsigned long flags;
489
490         spin_lock_irqsave(&msi_lock, flags);
491         for (vector = FIRST_DEVICE_VECTOR; vector < NR_IRQS; vector++) {
492                 if (!msi_desc[vector] || msi_desc[vector]->dev != dev ||
493                         msi_desc[vector]->msi_attrib.type != type ||
494                         msi_desc[vector]->msi_attrib.default_vector != dev->irq)
495                         continue;
496                 spin_unlock_irqrestore(&msi_lock, flags);
497                 /* This pre-assigned MSI vector for this device
498                    already exits. Override dev->irq with this vector */
499                 dev->irq = vector;
500                 return 0;
501         }
502         spin_unlock_irqrestore(&msi_lock, flags);
503
504         return -EACCES;
505 }
506
507 void pci_scan_msi_device(struct pci_dev *dev)
508 {
509         if (!dev)
510                 return;
511
512         if (pci_find_capability(dev, PCI_CAP_ID_MSIX) > 0)
513                 nr_msix_devices++;
514         else if (pci_find_capability(dev, PCI_CAP_ID_MSI) > 0)
515                 nr_reserved_vectors++;
516 }
517
518 #ifdef CONFIG_PM
519 int pci_save_msi_state(struct pci_dev *dev)
520 {
521         int pos, i = 0;
522         u16 control;
523         struct pci_cap_saved_state *save_state;
524         u32 *cap;
525
526         pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
527         if (pos <= 0 || dev->no_msi)
528                 return 0;
529
530         pci_read_config_word(dev, msi_control_reg(pos), &control);
531         if (!(control & PCI_MSI_FLAGS_ENABLE))
532                 return 0;
533
534         save_state = kzalloc(sizeof(struct pci_cap_saved_state) + sizeof(u32) * 5,
535                 GFP_KERNEL);
536         if (!save_state) {
537                 printk(KERN_ERR "Out of memory in pci_save_msi_state\n");
538                 return -ENOMEM;
539         }
540         cap = &save_state->data[0];
541
542         pci_read_config_dword(dev, pos, &cap[i++]);
543         control = cap[0] >> 16;
544         pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_LO, &cap[i++]);
545         if (control & PCI_MSI_FLAGS_64BIT) {
546                 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_HI, &cap[i++]);
547                 pci_read_config_dword(dev, pos + PCI_MSI_DATA_64, &cap[i++]);
548         } else
549                 pci_read_config_dword(dev, pos + PCI_MSI_DATA_32, &cap[i++]);
550         if (control & PCI_MSI_FLAGS_MASKBIT)
551                 pci_read_config_dword(dev, pos + PCI_MSI_MASK_BIT, &cap[i++]);
552         disable_msi_mode(dev, pos, PCI_CAP_ID_MSI);
553         save_state->cap_nr = PCI_CAP_ID_MSI;
554         pci_add_saved_cap(dev, save_state);
555         return 0;
556 }
557
558 void pci_restore_msi_state(struct pci_dev *dev)
559 {
560         int i = 0, pos;
561         u16 control;
562         struct pci_cap_saved_state *save_state;
563         u32 *cap;
564
565         save_state = pci_find_saved_cap(dev, PCI_CAP_ID_MSI);
566         pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
567         if (!save_state || pos <= 0)
568                 return;
569         cap = &save_state->data[0];
570
571         control = cap[i++] >> 16;
572         pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_LO, cap[i++]);
573         if (control & PCI_MSI_FLAGS_64BIT) {
574                 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_HI, cap[i++]);
575                 pci_write_config_dword(dev, pos + PCI_MSI_DATA_64, cap[i++]);
576         } else
577                 pci_write_config_dword(dev, pos + PCI_MSI_DATA_32, cap[i++]);
578         if (control & PCI_MSI_FLAGS_MASKBIT)
579                 pci_write_config_dword(dev, pos + PCI_MSI_MASK_BIT, cap[i++]);
580         pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
581         enable_msi_mode(dev, pos, PCI_CAP_ID_MSI);
582         pci_remove_saved_cap(save_state);
583         kfree(save_state);
584 }
585
586 int pci_save_msix_state(struct pci_dev *dev)
587 {
588         int pos;
589         int temp;
590         int vector, head, tail = 0;
591         u16 control;
592         struct pci_cap_saved_state *save_state;
593
594         pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
595         if (pos <= 0 || dev->no_msi)
596                 return 0;
597
598         /* save the capability */
599         pci_read_config_word(dev, msi_control_reg(pos), &control);
600         if (!(control & PCI_MSIX_FLAGS_ENABLE))
601                 return 0;
602         save_state = kzalloc(sizeof(struct pci_cap_saved_state) + sizeof(u16),
603                 GFP_KERNEL);
604         if (!save_state) {
605                 printk(KERN_ERR "Out of memory in pci_save_msix_state\n");
606                 return -ENOMEM;
607         }
608         *((u16 *)&save_state->data[0]) = control;
609
610         /* save the table */
611         temp = dev->irq;
612         if (msi_lookup_vector(dev, PCI_CAP_ID_MSIX)) {
613                 kfree(save_state);
614                 return -EINVAL;
615         }
616
617         vector = head = dev->irq;
618         while (head != tail) {
619                 int j;
620                 void __iomem *base;
621                 struct msi_desc *entry;
622
623                 entry = msi_desc[vector];
624                 base = entry->mask_base;
625                 j = entry->msi_attrib.entry_nr;
626
627                 entry->address_lo_save =
628                         readl(base + j * PCI_MSIX_ENTRY_SIZE +
629                               PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
630                 entry->address_hi_save =
631                         readl(base + j * PCI_MSIX_ENTRY_SIZE +
632                               PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
633                 entry->data_save =
634                         readl(base + j * PCI_MSIX_ENTRY_SIZE +
635                               PCI_MSIX_ENTRY_DATA_OFFSET);
636
637                 tail = msi_desc[vector]->link.tail;
638                 vector = tail;
639         }
640         dev->irq = temp;
641
642         disable_msi_mode(dev, pos, PCI_CAP_ID_MSIX);
643         save_state->cap_nr = PCI_CAP_ID_MSIX;
644         pci_add_saved_cap(dev, save_state);
645         return 0;
646 }
647
648 void pci_restore_msix_state(struct pci_dev *dev)
649 {
650         u16 save;
651         int pos;
652         int vector, head, tail = 0;
653         void __iomem *base;
654         int j;
655         struct msi_desc *entry;
656         int temp;
657         struct pci_cap_saved_state *save_state;
658
659         save_state = pci_find_saved_cap(dev, PCI_CAP_ID_MSIX);
660         if (!save_state)
661                 return;
662         save = *((u16 *)&save_state->data[0]);
663         pci_remove_saved_cap(save_state);
664         kfree(save_state);
665
666         pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
667         if (pos <= 0)
668                 return;
669
670         /* route the table */
671         temp = dev->irq;
672         if (msi_lookup_vector(dev, PCI_CAP_ID_MSIX))
673                 return;
674         vector = head = dev->irq;
675         while (head != tail) {
676                 entry = msi_desc[vector];
677                 base = entry->mask_base;
678                 j = entry->msi_attrib.entry_nr;
679
680                 writel(entry->address_lo_save,
681                         base + j * PCI_MSIX_ENTRY_SIZE +
682                         PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
683                 writel(entry->address_hi_save,
684                         base + j * PCI_MSIX_ENTRY_SIZE +
685                         PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
686                 writel(entry->data_save,
687                         base + j * PCI_MSIX_ENTRY_SIZE +
688                         PCI_MSIX_ENTRY_DATA_OFFSET);
689
690                 tail = msi_desc[vector]->link.tail;
691                 vector = tail;
692         }
693         dev->irq = temp;
694
695         pci_write_config_word(dev, msi_control_reg(pos), save);
696         enable_msi_mode(dev, pos, PCI_CAP_ID_MSIX);
697 }
698 #endif
699
700 static int msi_register_init(struct pci_dev *dev, struct msi_desc *entry)
701 {
702         int status;
703         u32 address_hi;
704         u32 address_lo;
705         u32 data;
706         int pos, vector = dev->irq;
707         u16 control;
708
709         pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
710         pci_read_config_word(dev, msi_control_reg(pos), &control);
711
712         /* Configure MSI capability structure */
713         status = msi_ops->setup(dev, vector, &address_hi, &address_lo, &data);
714         if (status < 0)
715                 return status;
716
717         pci_write_config_dword(dev, msi_lower_address_reg(pos), address_lo);
718         if (is_64bit_address(control)) {
719                 pci_write_config_dword(dev,
720                         msi_upper_address_reg(pos), address_hi);
721                 pci_write_config_word(dev,
722                         msi_data_reg(pos, 1), data);
723         } else
724                 pci_write_config_word(dev,
725                         msi_data_reg(pos, 0), data);
726         if (entry->msi_attrib.maskbit) {
727                 unsigned int maskbits, temp;
728                 /* All MSIs are unmasked by default, Mask them all */
729                 pci_read_config_dword(dev,
730                         msi_mask_bits_reg(pos, is_64bit_address(control)),
731                         &maskbits);
732                 temp = (1 << multi_msi_capable(control));
733                 temp = ((temp - 1) & ~temp);
734                 maskbits |= temp;
735                 pci_write_config_dword(dev,
736                         msi_mask_bits_reg(pos, is_64bit_address(control)),
737                         maskbits);
738         }
739
740         return 0;
741 }
742
743 /**
744  * msi_capability_init - configure device's MSI capability structure
745  * @dev: pointer to the pci_dev data structure of MSI device function
746  *
747  * Setup the MSI capability structure of device function with a single
748  * MSI vector, regardless of device function is capable of handling
749  * multiple messages. A return of zero indicates the successful setup
750  * of an entry zero with the new MSI vector or non-zero for otherwise.
751  **/
752 static int msi_capability_init(struct pci_dev *dev)
753 {
754         int status;
755         struct msi_desc *entry;
756         int pos, vector;
757         u16 control;
758
759         pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
760         pci_read_config_word(dev, msi_control_reg(pos), &control);
761         /* MSI Entry Initialization */
762         entry = alloc_msi_entry();
763         if (!entry)
764                 return -ENOMEM;
765
766         vector = get_msi_vector(dev);
767         if (vector < 0) {
768                 kmem_cache_free(msi_cachep, entry);
769                 return -EBUSY;
770         }
771         entry->link.head = vector;
772         entry->link.tail = vector;
773         entry->msi_attrib.type = PCI_CAP_ID_MSI;
774         entry->msi_attrib.state = 0;                    /* Mark it not active */
775         entry->msi_attrib.entry_nr = 0;
776         entry->msi_attrib.maskbit = is_mask_bit_support(control);
777         entry->msi_attrib.default_vector = dev->irq;    /* Save IOAPIC IRQ */
778         dev->irq = vector;
779         entry->dev = dev;
780         if (is_mask_bit_support(control)) {
781                 entry->mask_base = (void __iomem *)(long)msi_mask_bits_reg(pos,
782                                 is_64bit_address(control));
783         }
784         /* Replace with MSI handler */
785         irq_handler_init(PCI_CAP_ID_MSI, vector, entry->msi_attrib.maskbit);
786         /* Configure MSI capability structure */
787         status = msi_register_init(dev, entry);
788         if (status != 0) {
789                 dev->irq = entry->msi_attrib.default_vector;
790                 kmem_cache_free(msi_cachep, entry);
791                 return status;
792         }
793
794         attach_msi_entry(entry, vector);
795         /* Set MSI enabled bits  */
796         enable_msi_mode(dev, pos, PCI_CAP_ID_MSI);
797
798         return 0;
799 }
800
801 /**
802  * msix_capability_init - configure device's MSI-X capability
803  * @dev: pointer to the pci_dev data structure of MSI-X device function
804  * @entries: pointer to an array of struct msix_entry entries
805  * @nvec: number of @entries
806  *
807  * Setup the MSI-X capability structure of device function with a
808  * single MSI-X vector. A return of zero indicates the successful setup of
809  * requested MSI-X entries with allocated vectors or non-zero for otherwise.
810  **/
811 static int msix_capability_init(struct pci_dev *dev,
812                                 struct msix_entry *entries, int nvec)
813 {
814         struct msi_desc *head = NULL, *tail = NULL, *entry = NULL;
815         u32 address_hi;
816         u32 address_lo;
817         u32 data;
818         int status;
819         int vector, pos, i, j, nr_entries, temp = 0;
820         unsigned long phys_addr;
821         u32 table_offset;
822         u16 control;
823         u8 bir;
824         void __iomem *base;
825
826         pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
827         /* Request & Map MSI-X table region */
828         pci_read_config_word(dev, msi_control_reg(pos), &control);
829         nr_entries = multi_msix_capable(control);
830
831         pci_read_config_dword(dev, msix_table_offset_reg(pos), &table_offset);
832         bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
833         table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
834         phys_addr = pci_resource_start (dev, bir) + table_offset;
835         base = ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
836         if (base == NULL)
837                 return -ENOMEM;
838
839         /* MSI-X Table Initialization */
840         for (i = 0; i < nvec; i++) {
841                 entry = alloc_msi_entry();
842                 if (!entry)
843                         break;
844                 vector = get_msi_vector(dev);
845                 if (vector < 0) {
846                         kmem_cache_free(msi_cachep, entry);
847                         break;
848                 }
849
850                 j = entries[i].entry;
851                 entries[i].vector = vector;
852                 entry->msi_attrib.type = PCI_CAP_ID_MSIX;
853                 entry->msi_attrib.state = 0;            /* Mark it not active */
854                 entry->msi_attrib.entry_nr = j;
855                 entry->msi_attrib.maskbit = 1;
856                 entry->msi_attrib.default_vector = dev->irq;
857                 entry->dev = dev;
858                 entry->mask_base = base;
859                 if (!head) {
860                         entry->link.head = vector;
861                         entry->link.tail = vector;
862                         head = entry;
863                 } else {
864                         entry->link.head = temp;
865                         entry->link.tail = tail->link.tail;
866                         tail->link.tail = vector;
867                         head->link.head = vector;
868                 }
869                 temp = vector;
870                 tail = entry;
871                 /* Replace with MSI-X handler */
872                 irq_handler_init(PCI_CAP_ID_MSIX, vector, 1);
873                 /* Configure MSI-X capability structure */
874                 status = msi_ops->setup(dev, vector,
875                                         &address_hi,
876                                         &address_lo,
877                                         &data);
878                 if (status < 0)
879                         break;
880
881                 writel(address_lo,
882                         base + j * PCI_MSIX_ENTRY_SIZE +
883                         PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
884                 writel(address_hi,
885                         base + j * PCI_MSIX_ENTRY_SIZE +
886                         PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
887                 writel(data,
888                         base + j * PCI_MSIX_ENTRY_SIZE +
889                         PCI_MSIX_ENTRY_DATA_OFFSET);
890                 attach_msi_entry(entry, vector);
891         }
892         if (i != nvec) {
893                 i--;
894                 for (; i >= 0; i--) {
895                         vector = (entries + i)->vector;
896                         msi_free_vector(dev, vector, 0);
897                         (entries + i)->vector = 0;
898                 }
899                 return -EBUSY;
900         }
901         /* Set MSI-X enabled bits */
902         enable_msi_mode(dev, pos, PCI_CAP_ID_MSIX);
903
904         return 0;
905 }
906
907 /**
908  * pci_enable_msi - configure device's MSI capability structure
909  * @dev: pointer to the pci_dev data structure of MSI device function
910  *
911  * Setup the MSI capability structure of device function with
912  * a single MSI vector upon its software driver call to request for
913  * MSI mode enabled on its hardware device function. A return of zero
914  * indicates the successful setup of an entry zero with the new MSI
915  * vector or non-zero for otherwise.
916  **/
917 int pci_enable_msi(struct pci_dev* dev)
918 {
919         int pos, temp, status = -EINVAL;
920         u16 control;
921
922         if (!pci_msi_enable || !dev)
923                 return status;
924
925         if (dev->no_msi)
926                 return status;
927
928         if (dev->bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
929                 return -EINVAL;
930
931         temp = dev->irq;
932
933         status = msi_init();
934         if (status < 0)
935                 return status;
936
937         pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
938         if (!pos)
939                 return -EINVAL;
940
941         pci_read_config_word(dev, msi_control_reg(pos), &control);
942         if (control & PCI_MSI_FLAGS_ENABLE)
943                 return 0;                       /* Already in MSI mode */
944
945         if (!msi_lookup_vector(dev, PCI_CAP_ID_MSI)) {
946                 /* Lookup Sucess */
947                 unsigned long flags;
948
949                 spin_lock_irqsave(&msi_lock, flags);
950                 if (!vector_irq[dev->irq]) {
951                         msi_desc[dev->irq]->msi_attrib.state = 0;
952                         vector_irq[dev->irq] = -1;
953                         nr_released_vectors--;
954                         spin_unlock_irqrestore(&msi_lock, flags);
955                         status = msi_register_init(dev, msi_desc[dev->irq]);
956                         if (status == 0)
957                                 enable_msi_mode(dev, pos, PCI_CAP_ID_MSI);
958                         return status;
959                 }
960                 spin_unlock_irqrestore(&msi_lock, flags);
961                 dev->irq = temp;
962         }
963         /* Check whether driver already requested for MSI-X vectors */
964         pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
965         if (pos > 0 && !msi_lookup_vector(dev, PCI_CAP_ID_MSIX)) {
966                         printk(KERN_INFO "PCI: %s: Can't enable MSI.  "
967                                "Device already has MSI-X vectors assigned\n",
968                                pci_name(dev));
969                         dev->irq = temp;
970                         return -EINVAL;
971         }
972         status = msi_capability_init(dev);
973         if (!status) {
974                 if (!pos)
975                         nr_reserved_vectors--;  /* Only MSI capable */
976                 else if (nr_msix_devices > 0)
977                         nr_msix_devices--;      /* Both MSI and MSI-X capable,
978                                                    but choose enabling MSI */
979         }
980
981         return status;
982 }
983
984 void pci_disable_msi(struct pci_dev* dev)
985 {
986         struct msi_desc *entry;
987         int pos, default_vector;
988         u16 control;
989         unsigned long flags;
990
991         if (!pci_msi_enable)
992                 return;
993         if (!dev)
994                 return;
995
996         pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
997         if (!pos)
998                 return;
999
1000         pci_read_config_word(dev, msi_control_reg(pos), &control);
1001         if (!(control & PCI_MSI_FLAGS_ENABLE))
1002                 return;
1003
1004         spin_lock_irqsave(&msi_lock, flags);
1005         entry = msi_desc[dev->irq];
1006         if (!entry || !entry->dev || entry->msi_attrib.type != PCI_CAP_ID_MSI) {
1007                 spin_unlock_irqrestore(&msi_lock, flags);
1008                 return;
1009         }
1010         if (entry->msi_attrib.state) {
1011                 spin_unlock_irqrestore(&msi_lock, flags);
1012                 printk(KERN_WARNING "PCI: %s: pci_disable_msi() called without "
1013                        "free_irq() on MSI vector %d\n",
1014                        pci_name(dev), dev->irq);
1015                 BUG_ON(entry->msi_attrib.state > 0);
1016         } else {
1017                 vector_irq[dev->irq] = 0; /* free it */
1018                 nr_released_vectors++;
1019                 default_vector = entry->msi_attrib.default_vector;
1020                 spin_unlock_irqrestore(&msi_lock, flags);
1021                 /* Restore dev->irq to its default pin-assertion vector */
1022                 dev->irq = default_vector;
1023                 disable_msi_mode(dev, pci_find_capability(dev, PCI_CAP_ID_MSI),
1024                                         PCI_CAP_ID_MSI);
1025         }
1026 }
1027
1028 static int msi_free_vector(struct pci_dev* dev, int vector, int reassign)
1029 {
1030         struct msi_desc *entry;
1031         int head, entry_nr, type;
1032         void __iomem *base;
1033         unsigned long flags;
1034
1035         msi_ops->teardown(vector);
1036
1037         spin_lock_irqsave(&msi_lock, flags);
1038         entry = msi_desc[vector];
1039         if (!entry || entry->dev != dev) {
1040                 spin_unlock_irqrestore(&msi_lock, flags);
1041                 return -EINVAL;
1042         }
1043         type = entry->msi_attrib.type;
1044         entry_nr = entry->msi_attrib.entry_nr;
1045         head = entry->link.head;
1046         base = entry->mask_base;
1047         msi_desc[entry->link.head]->link.tail = entry->link.tail;
1048         msi_desc[entry->link.tail]->link.head = entry->link.head;
1049         entry->dev = NULL;
1050         if (!reassign) {
1051                 vector_irq[vector] = 0;
1052                 nr_released_vectors++;
1053         }
1054         msi_desc[vector] = NULL;
1055         spin_unlock_irqrestore(&msi_lock, flags);
1056
1057         kmem_cache_free(msi_cachep, entry);
1058
1059         if (type == PCI_CAP_ID_MSIX) {
1060                 if (!reassign)
1061                         writel(1, base +
1062                                 entry_nr * PCI_MSIX_ENTRY_SIZE +
1063                                 PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET);
1064
1065                 if (head == vector) {
1066                         /*
1067                          * Detect last MSI-X vector to be released.
1068                          * Release the MSI-X memory-mapped table.
1069                          */
1070 #if 0
1071                         int pos, nr_entries;
1072                         unsigned long phys_addr;
1073                         u32 table_offset;
1074                         u16 control;
1075                         u8 bir;
1076
1077                         pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
1078                         pci_read_config_word(dev, msi_control_reg(pos),
1079                                 &control);
1080                         nr_entries = multi_msix_capable(control);
1081                         pci_read_config_dword(dev, msix_table_offset_reg(pos),
1082                                 &table_offset);
1083                         bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
1084                         table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
1085                         phys_addr = pci_resource_start(dev, bir) + table_offset;
1086 /*
1087  * FIXME!  and what did you want to do with phys_addr?
1088  */
1089 #endif
1090                         iounmap(base);
1091                 }
1092         }
1093
1094         return 0;
1095 }
1096
1097 static int reroute_msix_table(int head, struct msix_entry *entries, int *nvec)
1098 {
1099         int vector = head, tail = 0;
1100         int i, j = 0, nr_entries = 0;
1101         void __iomem *base;
1102         unsigned long flags;
1103
1104         spin_lock_irqsave(&msi_lock, flags);
1105         while (head != tail) {
1106                 nr_entries++;
1107                 tail = msi_desc[vector]->link.tail;
1108                 if (entries[0].entry == msi_desc[vector]->msi_attrib.entry_nr)
1109                         j = vector;
1110                 vector = tail;
1111         }
1112         if (*nvec > nr_entries) {
1113                 spin_unlock_irqrestore(&msi_lock, flags);
1114                 *nvec = nr_entries;
1115                 return -EINVAL;
1116         }
1117         vector = ((j > 0) ? j : head);
1118         for (i = 0; i < *nvec; i++) {
1119                 j = msi_desc[vector]->msi_attrib.entry_nr;
1120                 msi_desc[vector]->msi_attrib.state = 0; /* Mark it not active */
1121                 vector_irq[vector] = -1;                /* Mark it busy */
1122                 nr_released_vectors--;
1123                 entries[i].vector = vector;
1124                 if (j != (entries + i)->entry) {
1125                         base = msi_desc[vector]->mask_base;
1126                         msi_desc[vector]->msi_attrib.entry_nr =
1127                                 (entries + i)->entry;
1128                         writel( readl(base + j * PCI_MSIX_ENTRY_SIZE +
1129                                 PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET), base +
1130                                 (entries + i)->entry * PCI_MSIX_ENTRY_SIZE +
1131                                 PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
1132                         writel( readl(base + j * PCI_MSIX_ENTRY_SIZE +
1133                                 PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET), base +
1134                                 (entries + i)->entry * PCI_MSIX_ENTRY_SIZE +
1135                                 PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
1136                         writel( (readl(base + j * PCI_MSIX_ENTRY_SIZE +
1137                                 PCI_MSIX_ENTRY_DATA_OFFSET) & 0xff00) | vector,
1138                                 base + (entries+i)->entry*PCI_MSIX_ENTRY_SIZE +
1139                                 PCI_MSIX_ENTRY_DATA_OFFSET);
1140                 }
1141                 vector = msi_desc[vector]->link.tail;
1142         }
1143         spin_unlock_irqrestore(&msi_lock, flags);
1144
1145         return 0;
1146 }
1147
1148 /**
1149  * pci_enable_msix - configure device's MSI-X capability structure
1150  * @dev: pointer to the pci_dev data structure of MSI-X device function
1151  * @entries: pointer to an array of MSI-X entries
1152  * @nvec: number of MSI-X vectors requested for allocation by device driver
1153  *
1154  * Setup the MSI-X capability structure of device function with the number
1155  * of requested vectors upon its software driver call to request for
1156  * MSI-X mode enabled on its hardware device function. A return of zero
1157  * indicates the successful configuration of MSI-X capability structure
1158  * with new allocated MSI-X vectors. A return of < 0 indicates a failure.
1159  * Or a return of > 0 indicates that driver request is exceeding the number
1160  * of vectors available. Driver should use the returned value to re-send
1161  * its request.
1162  **/
1163 int pci_enable_msix(struct pci_dev* dev, struct msix_entry *entries, int nvec)
1164 {
1165         int status, pos, nr_entries, free_vectors;
1166         int i, j, temp;
1167         u16 control;
1168         unsigned long flags;
1169
1170         if (!pci_msi_enable || !dev || !entries)
1171                 return -EINVAL;
1172
1173         status = msi_init();
1174         if (status < 0)
1175                 return status;
1176
1177         pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
1178         if (!pos)
1179                 return -EINVAL;
1180
1181         pci_read_config_word(dev, msi_control_reg(pos), &control);
1182         if (control & PCI_MSIX_FLAGS_ENABLE)
1183                 return -EINVAL;                 /* Already in MSI-X mode */
1184
1185         nr_entries = multi_msix_capable(control);
1186         if (nvec > nr_entries)
1187                 return -EINVAL;
1188
1189         /* Check for any invalid entries */
1190         for (i = 0; i < nvec; i++) {
1191                 if (entries[i].entry >= nr_entries)
1192                         return -EINVAL;         /* invalid entry */
1193                 for (j = i + 1; j < nvec; j++) {
1194                         if (entries[i].entry == entries[j].entry)
1195                                 return -EINVAL; /* duplicate entry */
1196                 }
1197         }
1198         temp = dev->irq;
1199         if (!msi_lookup_vector(dev, PCI_CAP_ID_MSIX)) {
1200                 /* Lookup Sucess */
1201                 nr_entries = nvec;
1202                 /* Reroute MSI-X table */
1203                 if (reroute_msix_table(dev->irq, entries, &nr_entries)) {
1204                         /* #requested > #previous-assigned */
1205                         dev->irq = temp;
1206                         return nr_entries;
1207                 }
1208                 dev->irq = temp;
1209                 enable_msi_mode(dev, pos, PCI_CAP_ID_MSIX);
1210                 return 0;
1211         }
1212         /* Check whether driver already requested for MSI vector */
1213         if (pci_find_capability(dev, PCI_CAP_ID_MSI) > 0 &&
1214                 !msi_lookup_vector(dev, PCI_CAP_ID_MSI)) {
1215                 printk(KERN_INFO "PCI: %s: Can't enable MSI-X.  "
1216                        "Device already has an MSI vector assigned\n",
1217                        pci_name(dev));
1218                 dev->irq = temp;
1219                 return -EINVAL;
1220         }
1221
1222         spin_lock_irqsave(&msi_lock, flags);
1223         /*
1224          * msi_lock is provided to ensure that enough vectors resources are
1225          * available before granting.
1226          */
1227         free_vectors = pci_vector_resources(last_alloc_vector,
1228                                 nr_released_vectors);
1229         /* Ensure that each MSI/MSI-X device has one vector reserved by
1230            default to avoid any MSI-X driver to take all available
1231            resources */
1232         free_vectors -= nr_reserved_vectors;
1233         /* Find the average of free vectors among MSI-X devices */
1234         if (nr_msix_devices > 0)
1235                 free_vectors /= nr_msix_devices;
1236         spin_unlock_irqrestore(&msi_lock, flags);
1237
1238         if (nvec > free_vectors) {
1239                 if (free_vectors > 0)
1240                         return free_vectors;
1241                 else
1242                         return -EBUSY;
1243         }
1244
1245         status = msix_capability_init(dev, entries, nvec);
1246         if (!status && nr_msix_devices > 0)
1247                 nr_msix_devices--;
1248
1249         return status;
1250 }
1251
1252 void pci_disable_msix(struct pci_dev* dev)
1253 {
1254         int pos, temp;
1255         u16 control;
1256
1257         if (!pci_msi_enable)
1258                 return;
1259         if (!dev)
1260                 return;
1261
1262         pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
1263         if (!pos)
1264                 return;
1265
1266         pci_read_config_word(dev, msi_control_reg(pos), &control);
1267         if (!(control & PCI_MSIX_FLAGS_ENABLE))
1268                 return;
1269
1270         temp = dev->irq;
1271         if (!msi_lookup_vector(dev, PCI_CAP_ID_MSIX)) {
1272                 int state, vector, head, tail = 0, warning = 0;
1273                 unsigned long flags;
1274
1275                 vector = head = dev->irq;
1276                 spin_lock_irqsave(&msi_lock, flags);
1277                 while (head != tail) {
1278                         state = msi_desc[vector]->msi_attrib.state;
1279                         if (state)
1280                                 warning = 1;
1281                         else {
1282                                 vector_irq[vector] = 0; /* free it */
1283                                 nr_released_vectors++;
1284                         }
1285                         tail = msi_desc[vector]->link.tail;
1286                         vector = tail;
1287                 }
1288                 spin_unlock_irqrestore(&msi_lock, flags);
1289                 if (warning) {
1290                         dev->irq = temp;
1291                         printk(KERN_WARNING "PCI: %s: pci_disable_msix() called without "
1292                                "free_irq() on all MSI-X vectors\n",
1293                                pci_name(dev));
1294                         BUG_ON(warning > 0);
1295                 } else {
1296                         dev->irq = temp;
1297                         disable_msi_mode(dev,
1298                                 pci_find_capability(dev, PCI_CAP_ID_MSIX),
1299                                 PCI_CAP_ID_MSIX);
1300
1301                 }
1302         }
1303 }
1304
1305 /**
1306  * msi_remove_pci_irq_vectors - reclaim MSI(X) vectors to unused state
1307  * @dev: pointer to the pci_dev data structure of MSI(X) device function
1308  *
1309  * Being called during hotplug remove, from which the device function
1310  * is hot-removed. All previous assigned MSI/MSI-X vectors, if
1311  * allocated for this device function, are reclaimed to unused state,
1312  * which may be used later on.
1313  **/
1314 void msi_remove_pci_irq_vectors(struct pci_dev* dev)
1315 {
1316         int state, pos, temp;
1317         unsigned long flags;
1318
1319         if (!pci_msi_enable || !dev)
1320                 return;
1321
1322         temp = dev->irq;                /* Save IOAPIC IRQ */
1323         pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
1324         if (pos > 0 && !msi_lookup_vector(dev, PCI_CAP_ID_MSI)) {
1325                 spin_lock_irqsave(&msi_lock, flags);
1326                 state = msi_desc[dev->irq]->msi_attrib.state;
1327                 spin_unlock_irqrestore(&msi_lock, flags);
1328                 if (state) {
1329                         printk(KERN_WARNING "PCI: %s: msi_remove_pci_irq_vectors() "
1330                                "called without free_irq() on MSI vector %d\n",
1331                                pci_name(dev), dev->irq);
1332                         BUG_ON(state > 0);
1333                 } else /* Release MSI vector assigned to this device */
1334                         msi_free_vector(dev, dev->irq, 0);
1335                 dev->irq = temp;                /* Restore IOAPIC IRQ */
1336         }
1337         pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
1338         if (pos > 0 && !msi_lookup_vector(dev, PCI_CAP_ID_MSIX)) {
1339                 int vector, head, tail = 0, warning = 0;
1340                 void __iomem *base = NULL;
1341
1342                 vector = head = dev->irq;
1343                 while (head != tail) {
1344                         spin_lock_irqsave(&msi_lock, flags);
1345                         state = msi_desc[vector]->msi_attrib.state;
1346                         tail = msi_desc[vector]->link.tail;
1347                         base = msi_desc[vector]->mask_base;
1348                         spin_unlock_irqrestore(&msi_lock, flags);
1349                         if (state)
1350                                 warning = 1;
1351                         else if (vector != head) /* Release MSI-X vector */
1352                                 msi_free_vector(dev, vector, 0);
1353                         vector = tail;
1354                 }
1355                 msi_free_vector(dev, vector, 0);
1356                 if (warning) {
1357                         /* Force to release the MSI-X memory-mapped table */
1358 #if 0
1359                         unsigned long phys_addr;
1360                         u32 table_offset;
1361                         u16 control;
1362                         u8 bir;
1363
1364                         pci_read_config_word(dev, msi_control_reg(pos),
1365                                 &control);
1366                         pci_read_config_dword(dev, msix_table_offset_reg(pos),
1367                                 &table_offset);
1368                         bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
1369                         table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
1370                         phys_addr = pci_resource_start(dev, bir) + table_offset;
1371 /*
1372  * FIXME! and what did you want to do with phys_addr?
1373  */
1374 #endif
1375                         iounmap(base);
1376                         printk(KERN_WARNING "PCI: %s: msi_remove_pci_irq_vectors() "
1377                                "called without free_irq() on all MSI-X vectors\n",
1378                                pci_name(dev));
1379                         BUG_ON(warning > 0);
1380                 }
1381                 dev->irq = temp;                /* Restore IOAPIC IRQ */
1382         }
1383 }
1384
1385 void pci_no_msi(void)
1386 {
1387         pci_msi_enable = 0;
1388 }
1389
1390 EXPORT_SYMBOL(pci_enable_msi);
1391 EXPORT_SYMBOL(pci_disable_msi);
1392 EXPORT_SYMBOL(pci_enable_msix);
1393 EXPORT_SYMBOL(pci_disable_msix);