1 // SPDX-License-Identifier: GPL-2.0
2 /* pci_sun4v.c: SUN4V specific PCI controller support.
4 * Copyright (C) 2006, 2007, 2008 David S. Miller (davem@davemloft.net)
7 #include <linux/kernel.h>
8 #include <linux/types.h>
10 #include <linux/init.h>
11 #include <linux/slab.h>
12 #include <linux/interrupt.h>
13 #include <linux/percpu.h>
14 #include <linux/irq.h>
15 #include <linux/msi.h>
16 #include <linux/export.h>
17 #include <linux/log2.h>
18 #include <linux/of_device.h>
19 #include <asm/iommu-common.h>
21 #include <asm/iommu.h>
23 #include <asm/hypervisor.h>
27 #include "iommu_common.h"
30 #include "pci_sun4v.h"
32 #define DRIVER_NAME "pci_sun4v"
33 #define PFX DRIVER_NAME ": "
35 static unsigned long vpci_major
;
36 static unsigned long vpci_minor
;
43 /* Ordered from largest major to lowest */
44 static struct vpci_version vpci_versions
[] = {
45 { .major
= 2, .minor
= 0 },
46 { .major
= 1, .minor
= 1 },
49 static unsigned long vatu_major
= 1;
50 static unsigned long vatu_minor
= 1;
52 #define PGLIST_NENTS (PAGE_SIZE / sizeof(u64))
55 struct device
*dev
; /* Device mapping is for. */
56 unsigned long prot
; /* IOMMU page protections */
57 unsigned long entry
; /* Index into IOTSB. */
58 u64
*pglist
; /* List of physical pages */
59 unsigned long npages
; /* Number of pages in list. */
62 static DEFINE_PER_CPU(struct iommu_batch
, iommu_batch
);
63 static int iommu_batch_initialized
;
65 /* Interrupts must be disabled. */
66 static inline void iommu_batch_start(struct device
*dev
, unsigned long prot
, unsigned long entry
)
68 struct iommu_batch
*p
= this_cpu_ptr(&iommu_batch
);
76 /* Interrupts must be disabled. */
77 static long iommu_batch_flush(struct iommu_batch
*p
, u64 mask
)
79 struct pci_pbm_info
*pbm
= p
->dev
->archdata
.host_controller
;
80 u64
*pglist
= p
->pglist
;
82 unsigned long devhandle
= pbm
->devhandle
;
83 unsigned long prot
= p
->prot
;
84 unsigned long entry
= p
->entry
;
85 unsigned long npages
= p
->npages
;
86 unsigned long iotsb_num
;
90 /* VPCI maj=1, min=[0,1] only supports read and write */
92 prot
&= (HV_PCI_MAP_ATTR_READ
| HV_PCI_MAP_ATTR_WRITE
);
95 if (mask
<= DMA_BIT_MASK(32) || !pbm
->iommu
->atu
) {
96 num
= pci_sun4v_iommu_map(devhandle
,
97 HV_PCI_TSBID(0, entry
),
101 if (unlikely(num
< 0)) {
102 pr_err_ratelimited("%s: IOMMU map of [%08lx:%08llx:%lx:%lx:%lx] failed with status %ld\n",
105 HV_PCI_TSBID(0, entry
),
106 npages
, prot
, __pa(pglist
),
111 index_count
= HV_PCI_IOTSB_INDEX_COUNT(npages
, entry
),
112 iotsb_num
= pbm
->iommu
->atu
->iotsb
->iotsb_num
;
113 ret
= pci_sun4v_iotsb_map(devhandle
,
119 if (unlikely(ret
!= HV_EOK
)) {
120 pr_err_ratelimited("%s: ATU map of [%08lx:%lx:%llx:%lx:%lx] failed with status %ld\n",
122 devhandle
, iotsb_num
,
139 static inline void iommu_batch_new_entry(unsigned long entry
, u64 mask
)
141 struct iommu_batch
*p
= this_cpu_ptr(&iommu_batch
);
143 if (p
->entry
+ p
->npages
== entry
)
145 if (p
->entry
!= ~0UL)
146 iommu_batch_flush(p
, mask
);
150 /* Interrupts must be disabled. */
151 static inline long iommu_batch_add(u64 phys_page
, u64 mask
)
153 struct iommu_batch
*p
= this_cpu_ptr(&iommu_batch
);
155 BUG_ON(p
->npages
>= PGLIST_NENTS
);
157 p
->pglist
[p
->npages
++] = phys_page
;
158 if (p
->npages
== PGLIST_NENTS
)
159 return iommu_batch_flush(p
, mask
);
164 /* Interrupts must be disabled. */
165 static inline long iommu_batch_end(u64 mask
)
167 struct iommu_batch
*p
= this_cpu_ptr(&iommu_batch
);
169 BUG_ON(p
->npages
>= PGLIST_NENTS
);
171 return iommu_batch_flush(p
, mask
);
174 static void *dma_4v_alloc_coherent(struct device
*dev
, size_t size
,
175 dma_addr_t
*dma_addrp
, gfp_t gfp
,
179 unsigned long flags
, order
, first_page
, npages
, n
;
180 unsigned long prot
= 0;
183 struct iommu_map_table
*tbl
;
189 size
= IO_PAGE_ALIGN(size
);
190 order
= get_order(size
);
191 if (unlikely(order
>= MAX_ORDER
))
194 npages
= size
>> IO_PAGE_SHIFT
;
196 if (attrs
& DMA_ATTR_WEAK_ORDERING
)
197 prot
= HV_PCI_MAP_ATTR_RELAXED_ORDER
;
199 nid
= dev
->archdata
.numa_node
;
200 page
= alloc_pages_node(nid
, gfp
, order
);
204 first_page
= (unsigned long) page_address(page
);
205 memset((char *)first_page
, 0, PAGE_SIZE
<< order
);
207 iommu
= dev
->archdata
.iommu
;
210 mask
= dev
->coherent_dma_mask
;
211 if (mask
<= DMA_BIT_MASK(32) || !atu
)
216 entry
= iommu_tbl_range_alloc(dev
, tbl
, npages
, NULL
,
217 (unsigned long)(-1), 0);
219 if (unlikely(entry
== IOMMU_ERROR_CODE
))
220 goto range_alloc_fail
;
222 *dma_addrp
= (tbl
->table_map_base
+ (entry
<< IO_PAGE_SHIFT
));
223 ret
= (void *) first_page
;
224 first_page
= __pa(first_page
);
226 local_irq_save(flags
);
228 iommu_batch_start(dev
,
229 (HV_PCI_MAP_ATTR_READ
| prot
|
230 HV_PCI_MAP_ATTR_WRITE
),
233 for (n
= 0; n
< npages
; n
++) {
234 long err
= iommu_batch_add(first_page
+ (n
* PAGE_SIZE
), mask
);
235 if (unlikely(err
< 0L))
239 if (unlikely(iommu_batch_end(mask
) < 0L))
242 local_irq_restore(flags
);
247 local_irq_restore(flags
);
248 iommu_tbl_range_free(tbl
, *dma_addrp
, npages
, IOMMU_ERROR_CODE
);
251 free_pages(first_page
, order
);
255 unsigned long dma_4v_iotsb_bind(unsigned long devhandle
,
256 unsigned long iotsb_num
,
257 struct pci_bus
*bus_dev
)
259 struct pci_dev
*pdev
;
265 list_for_each_entry(pdev
, &bus_dev
->devices
, bus_list
) {
266 if (pdev
->subordinate
) {
267 /* No need to bind pci bridge */
268 dma_4v_iotsb_bind(devhandle
, iotsb_num
,
271 bus
= bus_dev
->number
;
272 device
= PCI_SLOT(pdev
->devfn
);
273 fun
= PCI_FUNC(pdev
->devfn
);
274 err
= pci_sun4v_iotsb_bind(devhandle
, iotsb_num
,
275 HV_PCI_DEVICE_BUILD(bus
,
279 /* If bind fails for one device it is going to fail
280 * for rest of the devices because we are sharing
281 * IOTSB. So in case of failure simply return with
292 static void dma_4v_iommu_demap(struct device
*dev
, unsigned long devhandle
,
293 dma_addr_t dvma
, unsigned long iotsb_num
,
294 unsigned long entry
, unsigned long npages
)
296 unsigned long num
, flags
;
299 local_irq_save(flags
);
301 if (dvma
<= DMA_BIT_MASK(32)) {
302 num
= pci_sun4v_iommu_demap(devhandle
,
303 HV_PCI_TSBID(0, entry
),
306 ret
= pci_sun4v_iotsb_demap(devhandle
, iotsb_num
,
307 entry
, npages
, &num
);
308 if (unlikely(ret
!= HV_EOK
)) {
309 pr_err_ratelimited("pci_iotsb_demap() failed with error: %ld\n",
315 } while (npages
!= 0);
316 local_irq_restore(flags
);
319 static void dma_4v_free_coherent(struct device
*dev
, size_t size
, void *cpu
,
320 dma_addr_t dvma
, unsigned long attrs
)
322 struct pci_pbm_info
*pbm
;
325 struct iommu_map_table
*tbl
;
326 unsigned long order
, npages
, entry
;
327 unsigned long iotsb_num
;
330 npages
= IO_PAGE_ALIGN(size
) >> IO_PAGE_SHIFT
;
331 iommu
= dev
->archdata
.iommu
;
332 pbm
= dev
->archdata
.host_controller
;
334 devhandle
= pbm
->devhandle
;
336 if (dvma
<= DMA_BIT_MASK(32)) {
338 iotsb_num
= 0; /* we don't care for legacy iommu */
341 iotsb_num
= atu
->iotsb
->iotsb_num
;
343 entry
= ((dvma
- tbl
->table_map_base
) >> IO_PAGE_SHIFT
);
344 dma_4v_iommu_demap(dev
, devhandle
, dvma
, iotsb_num
, entry
, npages
);
345 iommu_tbl_range_free(tbl
, dvma
, npages
, IOMMU_ERROR_CODE
);
346 order
= get_order(size
);
348 free_pages((unsigned long)cpu
, order
);
351 static dma_addr_t
dma_4v_map_page(struct device
*dev
, struct page
*page
,
352 unsigned long offset
, size_t sz
,
353 enum dma_data_direction direction
,
358 struct iommu_map_table
*tbl
;
360 unsigned long flags
, npages
, oaddr
;
361 unsigned long i
, base_paddr
;
363 dma_addr_t bus_addr
, ret
;
366 iommu
= dev
->archdata
.iommu
;
369 if (unlikely(direction
== DMA_NONE
))
372 oaddr
= (unsigned long)(page_address(page
) + offset
);
373 npages
= IO_PAGE_ALIGN(oaddr
+ sz
) - (oaddr
& IO_PAGE_MASK
);
374 npages
>>= IO_PAGE_SHIFT
;
376 mask
= *dev
->dma_mask
;
377 if (mask
<= DMA_BIT_MASK(32))
382 entry
= iommu_tbl_range_alloc(dev
, tbl
, npages
, NULL
,
383 (unsigned long)(-1), 0);
385 if (unlikely(entry
== IOMMU_ERROR_CODE
))
388 bus_addr
= (tbl
->table_map_base
+ (entry
<< IO_PAGE_SHIFT
));
389 ret
= bus_addr
| (oaddr
& ~IO_PAGE_MASK
);
390 base_paddr
= __pa(oaddr
& IO_PAGE_MASK
);
391 prot
= HV_PCI_MAP_ATTR_READ
;
392 if (direction
!= DMA_TO_DEVICE
)
393 prot
|= HV_PCI_MAP_ATTR_WRITE
;
395 if (attrs
& DMA_ATTR_WEAK_ORDERING
)
396 prot
|= HV_PCI_MAP_ATTR_RELAXED_ORDER
;
398 local_irq_save(flags
);
400 iommu_batch_start(dev
, prot
, entry
);
402 for (i
= 0; i
< npages
; i
++, base_paddr
+= IO_PAGE_SIZE
) {
403 long err
= iommu_batch_add(base_paddr
, mask
);
404 if (unlikely(err
< 0L))
407 if (unlikely(iommu_batch_end(mask
) < 0L))
410 local_irq_restore(flags
);
415 if (printk_ratelimit())
417 return DMA_MAPPING_ERROR
;
420 local_irq_restore(flags
);
421 iommu_tbl_range_free(tbl
, bus_addr
, npages
, IOMMU_ERROR_CODE
);
422 return DMA_MAPPING_ERROR
;
425 static void dma_4v_unmap_page(struct device
*dev
, dma_addr_t bus_addr
,
426 size_t sz
, enum dma_data_direction direction
,
429 struct pci_pbm_info
*pbm
;
432 struct iommu_map_table
*tbl
;
433 unsigned long npages
;
434 unsigned long iotsb_num
;
438 if (unlikely(direction
== DMA_NONE
)) {
439 if (printk_ratelimit())
444 iommu
= dev
->archdata
.iommu
;
445 pbm
= dev
->archdata
.host_controller
;
447 devhandle
= pbm
->devhandle
;
449 npages
= IO_PAGE_ALIGN(bus_addr
+ sz
) - (bus_addr
& IO_PAGE_MASK
);
450 npages
>>= IO_PAGE_SHIFT
;
451 bus_addr
&= IO_PAGE_MASK
;
453 if (bus_addr
<= DMA_BIT_MASK(32)) {
454 iotsb_num
= 0; /* we don't care for legacy iommu */
457 iotsb_num
= atu
->iotsb
->iotsb_num
;
460 entry
= (bus_addr
- tbl
->table_map_base
) >> IO_PAGE_SHIFT
;
461 dma_4v_iommu_demap(dev
, devhandle
, bus_addr
, iotsb_num
, entry
, npages
);
462 iommu_tbl_range_free(tbl
, bus_addr
, npages
, IOMMU_ERROR_CODE
);
465 static int dma_4v_map_sg(struct device
*dev
, struct scatterlist
*sglist
,
466 int nelems
, enum dma_data_direction direction
,
469 struct scatterlist
*s
, *outs
, *segstart
;
470 unsigned long flags
, handle
, prot
;
471 dma_addr_t dma_next
= 0, dma_addr
;
472 unsigned int max_seg_size
;
473 unsigned long seg_boundary_size
;
474 int outcount
, incount
, i
;
477 struct iommu_map_table
*tbl
;
479 unsigned long base_shift
;
482 BUG_ON(direction
== DMA_NONE
);
484 iommu
= dev
->archdata
.iommu
;
485 if (nelems
== 0 || !iommu
)
489 prot
= HV_PCI_MAP_ATTR_READ
;
490 if (direction
!= DMA_TO_DEVICE
)
491 prot
|= HV_PCI_MAP_ATTR_WRITE
;
493 if (attrs
& DMA_ATTR_WEAK_ORDERING
)
494 prot
|= HV_PCI_MAP_ATTR_RELAXED_ORDER
;
496 outs
= s
= segstart
= &sglist
[0];
501 /* Init first segment length for backout at failure */
502 outs
->dma_length
= 0;
504 local_irq_save(flags
);
506 iommu_batch_start(dev
, prot
, ~0UL);
508 max_seg_size
= dma_get_max_seg_size(dev
);
509 seg_boundary_size
= ALIGN(dma_get_seg_boundary(dev
) + 1,
510 IO_PAGE_SIZE
) >> IO_PAGE_SHIFT
;
512 mask
= *dev
->dma_mask
;
513 if (mask
<= DMA_BIT_MASK(32))
518 base_shift
= tbl
->table_map_base
>> IO_PAGE_SHIFT
;
520 for_each_sg(sglist
, s
, nelems
, i
) {
521 unsigned long paddr
, npages
, entry
, out_entry
= 0, slen
;
529 /* Allocate iommu entries for that segment */
530 paddr
= (unsigned long) SG_ENT_PHYS_ADDRESS(s
);
531 npages
= iommu_num_pages(paddr
, slen
, IO_PAGE_SIZE
);
532 entry
= iommu_tbl_range_alloc(dev
, tbl
, npages
,
533 &handle
, (unsigned long)(-1), 0);
536 if (unlikely(entry
== IOMMU_ERROR_CODE
)) {
537 pr_err_ratelimited("iommu_alloc failed, iommu %p paddr %lx npages %lx\n",
539 goto iommu_map_failed
;
542 iommu_batch_new_entry(entry
, mask
);
544 /* Convert entry to a dma_addr_t */
545 dma_addr
= tbl
->table_map_base
+ (entry
<< IO_PAGE_SHIFT
);
546 dma_addr
|= (s
->offset
& ~IO_PAGE_MASK
);
548 /* Insert into HW table */
549 paddr
&= IO_PAGE_MASK
;
551 err
= iommu_batch_add(paddr
, mask
);
552 if (unlikely(err
< 0L))
553 goto iommu_map_failed
;
554 paddr
+= IO_PAGE_SIZE
;
557 /* If we are in an open segment, try merging */
559 /* We cannot merge if:
560 * - allocated dma_addr isn't contiguous to previous allocation
562 if ((dma_addr
!= dma_next
) ||
563 (outs
->dma_length
+ s
->length
> max_seg_size
) ||
564 (is_span_boundary(out_entry
, base_shift
,
565 seg_boundary_size
, outs
, s
))) {
566 /* Can't merge: create a new segment */
569 outs
= sg_next(outs
);
571 outs
->dma_length
+= s
->length
;
576 /* This is a new segment, fill entries */
577 outs
->dma_address
= dma_addr
;
578 outs
->dma_length
= slen
;
582 /* Calculate next page pointer for contiguous check */
583 dma_next
= dma_addr
+ slen
;
586 err
= iommu_batch_end(mask
);
588 if (unlikely(err
< 0L))
589 goto iommu_map_failed
;
591 local_irq_restore(flags
);
593 if (outcount
< incount
) {
594 outs
= sg_next(outs
);
595 outs
->dma_address
= DMA_MAPPING_ERROR
;
596 outs
->dma_length
= 0;
602 for_each_sg(sglist
, s
, nelems
, i
) {
603 if (s
->dma_length
!= 0) {
604 unsigned long vaddr
, npages
;
606 vaddr
= s
->dma_address
& IO_PAGE_MASK
;
607 npages
= iommu_num_pages(s
->dma_address
, s
->dma_length
,
609 iommu_tbl_range_free(tbl
, vaddr
, npages
,
612 s
->dma_address
= DMA_MAPPING_ERROR
;
618 local_irq_restore(flags
);
623 static void dma_4v_unmap_sg(struct device
*dev
, struct scatterlist
*sglist
,
624 int nelems
, enum dma_data_direction direction
,
627 struct pci_pbm_info
*pbm
;
628 struct scatterlist
*sg
;
631 unsigned long flags
, entry
;
632 unsigned long iotsb_num
;
635 BUG_ON(direction
== DMA_NONE
);
637 iommu
= dev
->archdata
.iommu
;
638 pbm
= dev
->archdata
.host_controller
;
640 devhandle
= pbm
->devhandle
;
642 local_irq_save(flags
);
646 dma_addr_t dma_handle
= sg
->dma_address
;
647 unsigned int len
= sg
->dma_length
;
648 unsigned long npages
;
649 struct iommu_map_table
*tbl
;
650 unsigned long shift
= IO_PAGE_SHIFT
;
654 npages
= iommu_num_pages(dma_handle
, len
, IO_PAGE_SIZE
);
656 if (dma_handle
<= DMA_BIT_MASK(32)) {
657 iotsb_num
= 0; /* we don't care for legacy iommu */
660 iotsb_num
= atu
->iotsb
->iotsb_num
;
663 entry
= ((dma_handle
- tbl
->table_map_base
) >> shift
);
664 dma_4v_iommu_demap(dev
, devhandle
, dma_handle
, iotsb_num
,
666 iommu_tbl_range_free(tbl
, dma_handle
, npages
,
671 local_irq_restore(flags
);
674 static int dma_4v_supported(struct device
*dev
, u64 device_mask
)
676 struct iommu
*iommu
= dev
->archdata
.iommu
;
678 if (ali_sound_dma_hack(dev
, device_mask
))
680 if (device_mask
< iommu
->dma_addr_mask
)
685 static const struct dma_map_ops sun4v_dma_ops
= {
686 .alloc
= dma_4v_alloc_coherent
,
687 .free
= dma_4v_free_coherent
,
688 .map_page
= dma_4v_map_page
,
689 .unmap_page
= dma_4v_unmap_page
,
690 .map_sg
= dma_4v_map_sg
,
691 .unmap_sg
= dma_4v_unmap_sg
,
692 .dma_supported
= dma_4v_supported
,
695 static void pci_sun4v_scan_bus(struct pci_pbm_info
*pbm
, struct device
*parent
)
697 struct property
*prop
;
698 struct device_node
*dp
;
700 dp
= pbm
->op
->dev
.of_node
;
701 prop
= of_find_property(dp
, "66mhz-capable", NULL
);
702 pbm
->is_66mhz_capable
= (prop
!= NULL
);
703 pbm
->pci_bus
= pci_scan_one_pbm(pbm
, parent
);
705 /* XXX register error interrupt handlers XXX */
708 static unsigned long probe_existing_entries(struct pci_pbm_info
*pbm
,
709 struct iommu_map_table
*iommu
)
711 struct iommu_pool
*pool
;
712 unsigned long i
, pool_nr
, cnt
= 0;
715 devhandle
= pbm
->devhandle
;
716 for (pool_nr
= 0; pool_nr
< iommu
->nr_pools
; pool_nr
++) {
717 pool
= &(iommu
->pools
[pool_nr
]);
718 for (i
= pool
->start
; i
<= pool
->end
; i
++) {
719 unsigned long ret
, io_attrs
, ra
;
721 ret
= pci_sun4v_iommu_getmap(devhandle
,
725 if (page_in_phys_avail(ra
)) {
726 pci_sun4v_iommu_demap(devhandle
,
731 __set_bit(i
, iommu
->map
);
739 static int pci_sun4v_atu_alloc_iotsb(struct pci_pbm_info
*pbm
)
741 struct atu
*atu
= pbm
->iommu
->atu
;
742 struct atu_iotsb
*iotsb
;
749 iotsb
= kzalloc(sizeof(*iotsb
), GFP_KERNEL
);
756 /* calculate size of IOTSB */
757 table_size
= (atu
->size
/ IO_PAGE_SIZE
) * 8;
758 order
= get_order(table_size
);
759 table
= (void *)__get_free_pages(GFP_KERNEL
| __GFP_ZERO
, order
);
764 iotsb
->table
= table
;
765 iotsb
->ra
= __pa(table
);
766 iotsb
->dvma_size
= atu
->size
;
767 iotsb
->dvma_base
= atu
->base
;
768 iotsb
->table_size
= table_size
;
769 iotsb
->page_size
= IO_PAGE_SIZE
;
771 /* configure and register IOTSB with HV */
772 err
= pci_sun4v_iotsb_conf(pbm
->devhandle
,
779 pr_err(PFX
"pci_iotsb_conf failed error: %ld\n", err
);
780 goto iotsb_conf_failed
;
782 iotsb
->iotsb_num
= iotsb_num
;
784 err
= dma_4v_iotsb_bind(pbm
->devhandle
, iotsb_num
, pbm
->pci_bus
);
786 pr_err(PFX
"pci_iotsb_bind failed error: %ld\n", err
);
787 goto iotsb_conf_failed
;
793 free_pages((unsigned long)table
, order
);
800 static int pci_sun4v_atu_init(struct pci_pbm_info
*pbm
)
802 struct atu
*atu
= pbm
->iommu
->atu
;
805 u64 map_size
, num_iotte
;
807 const u32
*page_size
;
810 ranges
= of_get_property(pbm
->op
->dev
.of_node
, "iommu-address-ranges",
813 pr_err(PFX
"No iommu-address-ranges\n");
817 page_size
= of_get_property(pbm
->op
->dev
.of_node
, "iommu-pagesizes",
820 pr_err(PFX
"No iommu-pagesizes\n");
824 /* There are 4 iommu-address-ranges supported. Each range is pair of
825 * {base, size}. The ranges[0] and ranges[1] are 32bit address space
826 * while ranges[2] and ranges[3] are 64bit space. We want to use 64bit
827 * address ranges to support 64bit addressing. Because 'size' for
828 * address ranges[2] and ranges[3] are same we can select either of
829 * ranges[2] or ranges[3] for mapping. However due to 'size' is too
830 * large for OS to allocate IOTSB we are using fix size 32G
831 * (ATU_64_SPACE_SIZE) which is more than enough for all PCIe devices
834 atu
->ranges
= (struct atu_ranges
*)ranges
;
835 atu
->base
= atu
->ranges
[3].base
;
836 atu
->size
= ATU_64_SPACE_SIZE
;
839 err
= pci_sun4v_atu_alloc_iotsb(pbm
);
841 pr_err(PFX
"Error creating ATU IOTSB\n");
845 /* Create ATU iommu map.
846 * One bit represents one iotte in IOTSB table.
848 dma_mask
= (roundup_pow_of_two(atu
->size
) - 1UL);
849 num_iotte
= atu
->size
/ IO_PAGE_SIZE
;
850 map_size
= num_iotte
/ 8;
851 atu
->tbl
.table_map_base
= atu
->base
;
852 atu
->dma_addr_mask
= dma_mask
;
853 atu
->tbl
.map
= kzalloc(map_size
, GFP_KERNEL
);
857 iommu_tbl_pool_init(&atu
->tbl
, num_iotte
, IO_PAGE_SHIFT
,
858 NULL
, false /* no large_pool */,
859 0 /* default npools */,
860 false /* want span boundary checking */);
865 static int pci_sun4v_iommu_init(struct pci_pbm_info
*pbm
)
867 static const u32 vdma_default
[] = { 0x80000000, 0x80000000 };
868 struct iommu
*iommu
= pbm
->iommu
;
869 unsigned long num_tsb_entries
, sz
;
870 u32 dma_mask
, dma_offset
;
873 vdma
= of_get_property(pbm
->op
->dev
.of_node
, "virtual-dma", NULL
);
877 if ((vdma
[0] | vdma
[1]) & ~IO_PAGE_MASK
) {
878 printk(KERN_ERR PFX
"Strange virtual-dma[%08x:%08x].\n",
883 dma_mask
= (roundup_pow_of_two(vdma
[1]) - 1UL);
884 num_tsb_entries
= vdma
[1] / IO_PAGE_SIZE
;
886 dma_offset
= vdma
[0];
888 /* Setup initial software IOMMU state. */
889 spin_lock_init(&iommu
->lock
);
890 iommu
->ctx_lowest_free
= 1;
891 iommu
->tbl
.table_map_base
= dma_offset
;
892 iommu
->dma_addr_mask
= dma_mask
;
894 /* Allocate and initialize the free area map. */
895 sz
= (num_tsb_entries
+ 7) / 8;
896 sz
= (sz
+ 7UL) & ~7UL;
897 iommu
->tbl
.map
= kzalloc(sz
, GFP_KERNEL
);
898 if (!iommu
->tbl
.map
) {
899 printk(KERN_ERR PFX
"Error, kmalloc(arena.map) failed.\n");
902 iommu_tbl_pool_init(&iommu
->tbl
, num_tsb_entries
, IO_PAGE_SHIFT
,
903 NULL
, false /* no large_pool */,
904 0 /* default npools */,
905 false /* want span boundary checking */);
906 sz
= probe_existing_entries(pbm
, &iommu
->tbl
);
908 printk("%s: Imported %lu TSB entries from OBP\n",
914 #ifdef CONFIG_PCI_MSI
915 struct pci_sun4v_msiq_entry
{
917 #define MSIQ_VERSION_MASK 0xffffffff00000000UL
918 #define MSIQ_VERSION_SHIFT 32
919 #define MSIQ_TYPE_MASK 0x00000000000000ffUL
920 #define MSIQ_TYPE_SHIFT 0
921 #define MSIQ_TYPE_NONE 0x00
922 #define MSIQ_TYPE_MSG 0x01
923 #define MSIQ_TYPE_MSI32 0x02
924 #define MSIQ_TYPE_MSI64 0x03
925 #define MSIQ_TYPE_INTX 0x08
926 #define MSIQ_TYPE_NONE2 0xff
931 u64 req_id
; /* bus/device/func */
932 #define MSIQ_REQID_BUS_MASK 0xff00UL
933 #define MSIQ_REQID_BUS_SHIFT 8
934 #define MSIQ_REQID_DEVICE_MASK 0x00f8UL
935 #define MSIQ_REQID_DEVICE_SHIFT 3
936 #define MSIQ_REQID_FUNC_MASK 0x0007UL
937 #define MSIQ_REQID_FUNC_SHIFT 0
941 /* The format of this value is message type dependent.
942 * For MSI bits 15:0 are the data from the MSI packet.
943 * For MSI-X bits 31:0 are the data from the MSI packet.
944 * For MSG, the message code and message routing code where:
945 * bits 39:32 is the bus/device/fn of the msg target-id
946 * bits 18:16 is the message routing code
947 * bits 7:0 is the message code
948 * For INTx the low order 2-bits are:
959 static int pci_sun4v_get_head(struct pci_pbm_info
*pbm
, unsigned long msiqid
,
962 unsigned long err
, limit
;
964 err
= pci_sun4v_msiq_gethead(pbm
->devhandle
, msiqid
, head
);
968 limit
= pbm
->msiq_ent_count
* sizeof(struct pci_sun4v_msiq_entry
);
969 if (unlikely(*head
>= limit
))
975 static int pci_sun4v_dequeue_msi(struct pci_pbm_info
*pbm
,
976 unsigned long msiqid
, unsigned long *head
,
979 struct pci_sun4v_msiq_entry
*ep
;
980 unsigned long err
, type
;
982 /* Note: void pointer arithmetic, 'head' is a byte offset */
983 ep
= (pbm
->msi_queues
+ ((msiqid
- pbm
->msiq_first
) *
984 (pbm
->msiq_ent_count
*
985 sizeof(struct pci_sun4v_msiq_entry
))) +
988 if ((ep
->version_type
& MSIQ_TYPE_MASK
) == 0)
991 type
= (ep
->version_type
& MSIQ_TYPE_MASK
) >> MSIQ_TYPE_SHIFT
;
992 if (unlikely(type
!= MSIQ_TYPE_MSI32
&&
993 type
!= MSIQ_TYPE_MSI64
))
998 err
= pci_sun4v_msi_setstate(pbm
->devhandle
,
999 ep
->msi_data
/* msi_num */,
1004 /* Clear the entry. */
1005 ep
->version_type
&= ~MSIQ_TYPE_MASK
;
1007 (*head
) += sizeof(struct pci_sun4v_msiq_entry
);
1009 (pbm
->msiq_ent_count
* sizeof(struct pci_sun4v_msiq_entry
)))
1015 static int pci_sun4v_set_head(struct pci_pbm_info
*pbm
, unsigned long msiqid
,
1020 err
= pci_sun4v_msiq_sethead(pbm
->devhandle
, msiqid
, head
);
1027 static int pci_sun4v_msi_setup(struct pci_pbm_info
*pbm
, unsigned long msiqid
,
1028 unsigned long msi
, int is_msi64
)
1030 if (pci_sun4v_msi_setmsiq(pbm
->devhandle
, msi
, msiqid
,
1032 HV_MSITYPE_MSI64
: HV_MSITYPE_MSI32
)))
1034 if (pci_sun4v_msi_setstate(pbm
->devhandle
, msi
, HV_MSISTATE_IDLE
))
1036 if (pci_sun4v_msi_setvalid(pbm
->devhandle
, msi
, HV_MSIVALID_VALID
))
1041 static int pci_sun4v_msi_teardown(struct pci_pbm_info
*pbm
, unsigned long msi
)
1043 unsigned long err
, msiqid
;
1045 err
= pci_sun4v_msi_getmsiq(pbm
->devhandle
, msi
, &msiqid
);
1049 pci_sun4v_msi_setvalid(pbm
->devhandle
, msi
, HV_MSIVALID_INVALID
);
1054 static int pci_sun4v_msiq_alloc(struct pci_pbm_info
*pbm
)
1056 unsigned long q_size
, alloc_size
, pages
, order
;
1059 q_size
= pbm
->msiq_ent_count
* sizeof(struct pci_sun4v_msiq_entry
);
1060 alloc_size
= (pbm
->msiq_num
* q_size
);
1061 order
= get_order(alloc_size
);
1062 pages
= __get_free_pages(GFP_KERNEL
| __GFP_COMP
, order
);
1064 printk(KERN_ERR
"MSI: Cannot allocate MSI queues (o=%lu).\n",
1068 memset((char *)pages
, 0, PAGE_SIZE
<< order
);
1069 pbm
->msi_queues
= (void *) pages
;
1071 for (i
= 0; i
< pbm
->msiq_num
; i
++) {
1072 unsigned long err
, base
= __pa(pages
+ (i
* q_size
));
1073 unsigned long ret1
, ret2
;
1075 err
= pci_sun4v_msiq_conf(pbm
->devhandle
,
1076 pbm
->msiq_first
+ i
,
1077 base
, pbm
->msiq_ent_count
);
1079 printk(KERN_ERR
"MSI: msiq register fails (err=%lu)\n",
1084 err
= pci_sun4v_msiq_info(pbm
->devhandle
,
1085 pbm
->msiq_first
+ i
,
1088 printk(KERN_ERR
"MSI: Cannot read msiq (err=%lu)\n",
1092 if (ret1
!= base
|| ret2
!= pbm
->msiq_ent_count
) {
1093 printk(KERN_ERR
"MSI: Bogus qconf "
1094 "expected[%lx:%x] got[%lx:%lx]\n",
1095 base
, pbm
->msiq_ent_count
,
1104 free_pages(pages
, order
);
1108 static void pci_sun4v_msiq_free(struct pci_pbm_info
*pbm
)
1110 unsigned long q_size
, alloc_size
, pages
, order
;
1113 for (i
= 0; i
< pbm
->msiq_num
; i
++) {
1114 unsigned long msiqid
= pbm
->msiq_first
+ i
;
1116 (void) pci_sun4v_msiq_conf(pbm
->devhandle
, msiqid
, 0UL, 0);
1119 q_size
= pbm
->msiq_ent_count
* sizeof(struct pci_sun4v_msiq_entry
);
1120 alloc_size
= (pbm
->msiq_num
* q_size
);
1121 order
= get_order(alloc_size
);
1123 pages
= (unsigned long) pbm
->msi_queues
;
1125 free_pages(pages
, order
);
1127 pbm
->msi_queues
= NULL
;
1130 static int pci_sun4v_msiq_build_irq(struct pci_pbm_info
*pbm
,
1131 unsigned long msiqid
,
1132 unsigned long devino
)
1134 unsigned int irq
= sun4v_build_irq(pbm
->devhandle
, devino
);
1139 if (pci_sun4v_msiq_setvalid(pbm
->devhandle
, msiqid
, HV_MSIQ_VALID
))
1141 if (pci_sun4v_msiq_setstate(pbm
->devhandle
, msiqid
, HV_MSIQSTATE_IDLE
))
1147 static const struct sparc64_msiq_ops pci_sun4v_msiq_ops
= {
1148 .get_head
= pci_sun4v_get_head
,
1149 .dequeue_msi
= pci_sun4v_dequeue_msi
,
1150 .set_head
= pci_sun4v_set_head
,
1151 .msi_setup
= pci_sun4v_msi_setup
,
1152 .msi_teardown
= pci_sun4v_msi_teardown
,
1153 .msiq_alloc
= pci_sun4v_msiq_alloc
,
1154 .msiq_free
= pci_sun4v_msiq_free
,
1155 .msiq_build_irq
= pci_sun4v_msiq_build_irq
,
1158 static void pci_sun4v_msi_init(struct pci_pbm_info
*pbm
)
1160 sparc64_pbm_msi_init(pbm
, &pci_sun4v_msiq_ops
);
1162 #else /* CONFIG_PCI_MSI */
1163 static void pci_sun4v_msi_init(struct pci_pbm_info
*pbm
)
1166 #endif /* !(CONFIG_PCI_MSI) */
1168 static int pci_sun4v_pbm_init(struct pci_pbm_info
*pbm
,
1169 struct platform_device
*op
, u32 devhandle
)
1171 struct device_node
*dp
= op
->dev
.of_node
;
1174 pbm
->numa_node
= of_node_to_nid(dp
);
1176 pbm
->pci_ops
= &sun4v_pci_ops
;
1177 pbm
->config_space_reg_bits
= 12;
1179 pbm
->index
= pci_num_pbms
++;
1183 pbm
->devhandle
= devhandle
;
1185 pbm
->name
= dp
->full_name
;
1187 printk("%s: SUN4V PCI Bus Module\n", pbm
->name
);
1188 printk("%s: On NUMA node %d\n", pbm
->name
, pbm
->numa_node
);
1190 pci_determine_mem_io_space(pbm
);
1192 pci_get_pbm_props(pbm
);
1194 err
= pci_sun4v_iommu_init(pbm
);
1198 pci_sun4v_msi_init(pbm
);
1200 pci_sun4v_scan_bus(pbm
, &op
->dev
);
1202 /* if atu_init fails its not complete failure.
1203 * we can still continue using legacy iommu.
1205 if (pbm
->iommu
->atu
) {
1206 err
= pci_sun4v_atu_init(pbm
);
1208 kfree(pbm
->iommu
->atu
);
1209 pbm
->iommu
->atu
= NULL
;
1210 pr_err(PFX
"ATU init failed, err=%d\n", err
);
1214 pbm
->next
= pci_pbm_root
;
1220 static int pci_sun4v_probe(struct platform_device
*op
)
1222 const struct linux_prom64_registers
*regs
;
1223 static int hvapi_negotiated
= 0;
1224 struct pci_pbm_info
*pbm
;
1225 struct device_node
*dp
;
1226 struct iommu
*iommu
;
1229 int i
, err
= -ENODEV
;
1230 static bool hv_atu
= true;
1232 dp
= op
->dev
.of_node
;
1234 if (!hvapi_negotiated
++) {
1235 for (i
= 0; i
< ARRAY_SIZE(vpci_versions
); i
++) {
1236 vpci_major
= vpci_versions
[i
].major
;
1237 vpci_minor
= vpci_versions
[i
].minor
;
1239 err
= sun4v_hvapi_register(HV_GRP_PCI
, vpci_major
,
1246 pr_err(PFX
"Could not register hvapi, err=%d\n", err
);
1249 pr_info(PFX
"Registered hvapi major[%lu] minor[%lu]\n",
1250 vpci_major
, vpci_minor
);
1252 err
= sun4v_hvapi_register(HV_GRP_ATU
, vatu_major
, &vatu_minor
);
1254 /* don't return an error if we fail to register the
1255 * ATU group, but ATU hcalls won't be available.
1259 pr_info(PFX
"Registered hvapi ATU major[%lu] minor[%lu]\n",
1260 vatu_major
, vatu_minor
);
1263 dma_ops
= &sun4v_dma_ops
;
1266 regs
= of_get_property(dp
, "reg", NULL
);
1269 printk(KERN_ERR PFX
"Could not find config registers\n");
1272 devhandle
= (regs
->phys_addr
>> 32UL) & 0x0fffffff;
1275 if (!iommu_batch_initialized
) {
1276 for_each_possible_cpu(i
) {
1277 unsigned long page
= get_zeroed_page(GFP_KERNEL
);
1282 per_cpu(iommu_batch
, i
).pglist
= (u64
*) page
;
1284 iommu_batch_initialized
= 1;
1287 pbm
= kzalloc(sizeof(*pbm
), GFP_KERNEL
);
1289 printk(KERN_ERR PFX
"Could not allocate pci_pbm_info\n");
1293 iommu
= kzalloc(sizeof(struct iommu
), GFP_KERNEL
);
1295 printk(KERN_ERR PFX
"Could not allocate pbm iommu\n");
1296 goto out_free_controller
;
1302 atu
= kzalloc(sizeof(*atu
), GFP_KERNEL
);
1304 pr_err(PFX
"Could not allocate atu\n");
1309 err
= pci_sun4v_pbm_init(pbm
, op
, devhandle
);
1311 goto out_free_iommu
;
1313 dev_set_drvdata(&op
->dev
, pbm
);
1321 out_free_controller
:
1328 static const struct of_device_id pci_sun4v_match
[] = {
1331 .compatible
= "SUNW,sun4v-pci",
1336 static struct platform_driver pci_sun4v_driver
= {
1338 .name
= DRIVER_NAME
,
1339 .of_match_table
= pci_sun4v_match
,
1341 .probe
= pci_sun4v_probe
,
1344 static int __init
pci_sun4v_init(void)
1346 return platform_driver_register(&pci_sun4v_driver
);
1349 subsys_initcall(pci_sun4v_init
);