2 * QEMU sPAPR IOMMU (TCE) code
4 * Copyright (c) 2010 David Gibson, IBM Corporation <dwg@au1.ibm.com>
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
20 #include "qemu/osdep.h"
21 #include "qemu/error-report.h"
23 #include "qemu/module.h"
24 #include "sysemu/kvm.h"
26 #include "migration/vmstate.h"
27 #include "sysemu/dma.h"
28 #include "exec/address-spaces.h"
31 #include "hw/ppc/spapr.h"
32 #include "hw/ppc/spapr_vio.h"
43 #define IOMMU_PAGE_SIZE(shift) (1ULL << (shift))
44 #define IOMMU_PAGE_MASK(shift) (~(IOMMU_PAGE_SIZE(shift) - 1))
46 static QLIST_HEAD(, SpaprTceTable
) spapr_tce_tables
;
48 SpaprTceTable
*spapr_tce_find_by_liobn(target_ulong liobn
)
52 if (liobn
& 0xFFFFFFFF00000000ULL
) {
53 hcall_dprintf("Request for out-of-bounds LIOBN 0x" TARGET_FMT_lx
"\n",
58 QLIST_FOREACH(tcet
, &spapr_tce_tables
, list
) {
59 if (tcet
->liobn
== (uint32_t)liobn
) {
67 static IOMMUAccessFlags
spapr_tce_iommu_access_flags(uint64_t tce
)
69 switch (tce
& SPAPR_TCE_RW
) {
76 default: /* SPAPR_TCE_RW */
81 static uint64_t *spapr_tce_alloc_table(uint32_t liobn
,
88 uint64_t *table
= NULL
;
91 table
= kvmppc_create_spapr_tce(liobn
, page_shift
, bus_offset
, nb_table
,
97 table
= g_new0(uint64_t, nb_table
);
100 trace_spapr_iommu_new_table(liobn
, table
, *fd
);
105 static void spapr_tce_free_table(uint64_t *table
, int fd
, uint32_t nb_table
)
107 if (!kvm_enabled() ||
108 (kvmppc_remove_spapr_tce(table
, fd
, nb_table
) != 0)) {
113 /* Called from RCU critical section */
114 static IOMMUTLBEntry
spapr_tce_translate_iommu(IOMMUMemoryRegion
*iommu
,
116 IOMMUAccessFlags flag
,
119 SpaprTceTable
*tcet
= container_of(iommu
, SpaprTceTable
, iommu
);
121 IOMMUTLBEntry ret
= {
122 .target_as
= &address_space_memory
,
124 .translated_addr
= 0,
125 .addr_mask
= ~(hwaddr
)0,
129 if ((addr
>> tcet
->page_shift
) < tcet
->nb_table
) {
130 /* Check if we are in bound */
131 hwaddr page_mask
= IOMMU_PAGE_MASK(tcet
->page_shift
);
133 tce
= tcet
->table
[addr
>> tcet
->page_shift
];
134 ret
.iova
= addr
& page_mask
;
135 ret
.translated_addr
= tce
& page_mask
;
136 ret
.addr_mask
= ~page_mask
;
137 ret
.perm
= spapr_tce_iommu_access_flags(tce
);
139 trace_spapr_iommu_xlate(tcet
->liobn
, addr
, ret
.translated_addr
, ret
.perm
,
145 static void spapr_tce_replay(IOMMUMemoryRegion
*iommu_mr
, IOMMUNotifier
*n
)
147 MemoryRegion
*mr
= MEMORY_REGION(iommu_mr
);
148 IOMMUMemoryRegionClass
*imrc
= IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr
);
149 hwaddr addr
, granularity
;
151 SpaprTceTable
*tcet
= container_of(iommu_mr
, SpaprTceTable
, iommu
);
153 if (tcet
->skipping_replay
) {
157 granularity
= memory_region_iommu_get_min_page_size(iommu_mr
);
159 for (addr
= 0; addr
< memory_region_size(mr
); addr
+= granularity
) {
160 iotlb
= imrc
->translate(iommu_mr
, addr
, IOMMU_NONE
, n
->iommu_idx
);
161 if (iotlb
.perm
!= IOMMU_NONE
) {
162 n
->notify(n
, &iotlb
);
166 * if (2^64 - MR size) < granularity, it's possible to get an
167 * infinite loop here. This should catch such a wraparound.
169 if ((addr
+ granularity
) < addr
) {
175 static int spapr_tce_table_pre_save(void *opaque
)
177 SpaprTceTable
*tcet
= SPAPR_TCE_TABLE(opaque
);
179 tcet
->mig_table
= tcet
->table
;
180 tcet
->mig_nb_table
= tcet
->nb_table
;
182 trace_spapr_iommu_pre_save(tcet
->liobn
, tcet
->mig_nb_table
,
183 tcet
->bus_offset
, tcet
->page_shift
);
188 static uint64_t spapr_tce_get_min_page_size(IOMMUMemoryRegion
*iommu
)
190 SpaprTceTable
*tcet
= container_of(iommu
, SpaprTceTable
, iommu
);
192 return 1ULL << tcet
->page_shift
;
195 static int spapr_tce_get_attr(IOMMUMemoryRegion
*iommu
,
196 enum IOMMUMemoryRegionAttr attr
, void *data
)
198 SpaprTceTable
*tcet
= container_of(iommu
, SpaprTceTable
, iommu
);
200 if (attr
== IOMMU_ATTR_SPAPR_TCE_FD
&& kvmppc_has_cap_spapr_vfio()) {
201 *(int *) data
= tcet
->fd
;
208 static int spapr_tce_notify_flag_changed(IOMMUMemoryRegion
*iommu
,
209 IOMMUNotifierFlag old
,
210 IOMMUNotifierFlag
new,
213 struct SpaprTceTable
*tbl
= container_of(iommu
, SpaprTceTable
, iommu
);
215 if (old
== IOMMU_NOTIFIER_NONE
&& new != IOMMU_NOTIFIER_NONE
) {
216 spapr_tce_set_need_vfio(tbl
, true);
217 } else if (old
!= IOMMU_NOTIFIER_NONE
&& new == IOMMU_NOTIFIER_NONE
) {
218 spapr_tce_set_need_vfio(tbl
, false);
223 static int spapr_tce_table_post_load(void *opaque
, int version_id
)
225 SpaprTceTable
*tcet
= SPAPR_TCE_TABLE(opaque
);
226 uint32_t old_nb_table
= tcet
->nb_table
;
227 uint64_t old_bus_offset
= tcet
->bus_offset
;
228 uint32_t old_page_shift
= tcet
->page_shift
;
231 spapr_vio_set_bypass(tcet
->vdev
, tcet
->bypass
);
234 if (tcet
->mig_nb_table
!= tcet
->nb_table
) {
235 spapr_tce_table_disable(tcet
);
238 if (tcet
->mig_nb_table
) {
239 if (!tcet
->nb_table
) {
240 spapr_tce_table_enable(tcet
, old_page_shift
, old_bus_offset
,
244 memcpy(tcet
->table
, tcet
->mig_table
,
245 tcet
->nb_table
* sizeof(tcet
->table
[0]));
247 free(tcet
->mig_table
);
248 tcet
->mig_table
= NULL
;
251 trace_spapr_iommu_post_load(tcet
->liobn
, old_nb_table
, tcet
->nb_table
,
252 tcet
->bus_offset
, tcet
->page_shift
);
257 static bool spapr_tce_table_ex_needed(void *opaque
)
259 SpaprTceTable
*tcet
= opaque
;
261 return tcet
->bus_offset
|| tcet
->page_shift
!= 0xC;
264 static const VMStateDescription vmstate_spapr_tce_table_ex
= {
265 .name
= "spapr_iommu_ex",
267 .minimum_version_id
= 1,
268 .needed
= spapr_tce_table_ex_needed
,
269 .fields
= (VMStateField
[]) {
270 VMSTATE_UINT64(bus_offset
, SpaprTceTable
),
271 VMSTATE_UINT32(page_shift
, SpaprTceTable
),
272 VMSTATE_END_OF_LIST()
276 static const VMStateDescription vmstate_spapr_tce_table
= {
277 .name
= "spapr_iommu",
279 .minimum_version_id
= 2,
280 .pre_save
= spapr_tce_table_pre_save
,
281 .post_load
= spapr_tce_table_post_load
,
282 .fields
= (VMStateField
[]) {
284 VMSTATE_UINT32_EQUAL(liobn
, SpaprTceTable
, NULL
),
287 VMSTATE_UINT32(mig_nb_table
, SpaprTceTable
),
288 VMSTATE_BOOL(bypass
, SpaprTceTable
),
289 VMSTATE_VARRAY_UINT32_ALLOC(mig_table
, SpaprTceTable
, mig_nb_table
, 0,
290 vmstate_info_uint64
, uint64_t),
292 VMSTATE_END_OF_LIST()
294 .subsections
= (const VMStateDescription
*[]) {
295 &vmstate_spapr_tce_table_ex
,
300 static void spapr_tce_table_realize(DeviceState
*dev
, Error
**errp
)
302 SpaprTceTable
*tcet
= SPAPR_TCE_TABLE(dev
);
303 Object
*tcetobj
= OBJECT(tcet
);
307 tcet
->need_vfio
= false;
308 tmp
= g_strdup_printf("tce-root-%x", tcet
->liobn
);
309 memory_region_init(&tcet
->root
, tcetobj
, tmp
, UINT64_MAX
);
312 tmp
= g_strdup_printf("tce-iommu-%x", tcet
->liobn
);
313 memory_region_init_iommu(&tcet
->iommu
, sizeof(tcet
->iommu
),
314 TYPE_SPAPR_IOMMU_MEMORY_REGION
,
318 QLIST_INSERT_HEAD(&spapr_tce_tables
, tcet
, list
);
320 vmstate_register(VMSTATE_IF(tcet
), tcet
->liobn
, &vmstate_spapr_tce_table
,
324 void spapr_tce_set_need_vfio(SpaprTceTable
*tcet
, bool need_vfio
)
326 size_t table_size
= tcet
->nb_table
* sizeof(uint64_t);
330 g_assert(need_vfio
!= tcet
->need_vfio
);
332 tcet
->need_vfio
= need_vfio
;
334 if (!need_vfio
|| (tcet
->fd
!= -1 && kvmppc_has_cap_spapr_vfio())) {
338 oldtable
= tcet
->table
;
340 tcet
->table
= spapr_tce_alloc_table(tcet
->liobn
,
346 memcpy(tcet
->table
, oldtable
, table_size
);
348 spapr_tce_free_table(oldtable
, tcet
->fd
, tcet
->nb_table
);
353 SpaprTceTable
*spapr_tce_new_table(DeviceState
*owner
, uint32_t liobn
)
358 if (spapr_tce_find_by_liobn(liobn
)) {
359 error_report("Attempted to create TCE table with duplicate"
360 " LIOBN 0x%x", liobn
);
364 tcet
= SPAPR_TCE_TABLE(object_new(TYPE_SPAPR_TCE_TABLE
));
367 tmp
= g_strdup_printf("tce-table-%x", liobn
);
368 object_property_add_child(OBJECT(owner
), tmp
, OBJECT(tcet
), NULL
);
370 object_unref(OBJECT(tcet
));
372 object_property_set_bool(OBJECT(tcet
), true, "realized", NULL
);
377 void spapr_tce_table_enable(SpaprTceTable
*tcet
,
378 uint32_t page_shift
, uint64_t bus_offset
,
381 if (tcet
->nb_table
) {
382 warn_report("trying to enable already enabled TCE table");
386 tcet
->bus_offset
= bus_offset
;
387 tcet
->page_shift
= page_shift
;
388 tcet
->nb_table
= nb_table
;
389 tcet
->table
= spapr_tce_alloc_table(tcet
->liobn
,
396 memory_region_set_size(MEMORY_REGION(&tcet
->iommu
),
397 (uint64_t)tcet
->nb_table
<< tcet
->page_shift
);
398 memory_region_add_subregion(&tcet
->root
, tcet
->bus_offset
,
399 MEMORY_REGION(&tcet
->iommu
));
402 void spapr_tce_table_disable(SpaprTceTable
*tcet
)
404 if (!tcet
->nb_table
) {
408 memory_region_del_subregion(&tcet
->root
, MEMORY_REGION(&tcet
->iommu
));
409 memory_region_set_size(MEMORY_REGION(&tcet
->iommu
), 0);
411 spapr_tce_free_table(tcet
->table
, tcet
->fd
, tcet
->nb_table
);
414 tcet
->bus_offset
= 0;
415 tcet
->page_shift
= 0;
419 static void spapr_tce_table_unrealize(DeviceState
*dev
, Error
**errp
)
421 SpaprTceTable
*tcet
= SPAPR_TCE_TABLE(dev
);
423 vmstate_unregister(VMSTATE_IF(tcet
), &vmstate_spapr_tce_table
, tcet
);
425 QLIST_REMOVE(tcet
, list
);
427 spapr_tce_table_disable(tcet
);
430 MemoryRegion
*spapr_tce_get_iommu(SpaprTceTable
*tcet
)
435 static void spapr_tce_reset(DeviceState
*dev
)
437 SpaprTceTable
*tcet
= SPAPR_TCE_TABLE(dev
);
438 size_t table_size
= tcet
->nb_table
* sizeof(uint64_t);
440 if (tcet
->nb_table
) {
441 memset(tcet
->table
, 0, table_size
);
445 static target_ulong
put_tce_emu(SpaprTceTable
*tcet
, target_ulong ioba
,
449 hwaddr page_mask
= IOMMU_PAGE_MASK(tcet
->page_shift
);
450 unsigned long index
= (ioba
- tcet
->bus_offset
) >> tcet
->page_shift
;
452 if (index
>= tcet
->nb_table
) {
453 hcall_dprintf("spapr_vio_put_tce on out-of-bounds IOBA 0x"
454 TARGET_FMT_lx
"\n", ioba
);
458 tcet
->table
[index
] = tce
;
460 entry
.target_as
= &address_space_memory
,
461 entry
.iova
= (ioba
- tcet
->bus_offset
) & page_mask
;
462 entry
.translated_addr
= tce
& page_mask
;
463 entry
.addr_mask
= ~page_mask
;
464 entry
.perm
= spapr_tce_iommu_access_flags(tce
);
465 memory_region_notify_iommu(&tcet
->iommu
, 0, entry
);
470 static target_ulong
h_put_tce_indirect(PowerPCCPU
*cpu
,
471 SpaprMachineState
*spapr
,
472 target_ulong opcode
, target_ulong
*args
)
475 target_ulong liobn
= args
[0];
476 target_ulong ioba
= args
[1];
477 target_ulong ioba1
= ioba
;
478 target_ulong tce_list
= args
[2];
479 target_ulong npages
= args
[3];
480 target_ulong ret
= H_PARAMETER
, tce
= 0;
481 SpaprTceTable
*tcet
= spapr_tce_find_by_liobn(liobn
);
482 CPUState
*cs
= CPU(cpu
);
483 hwaddr page_mask
, page_size
;
489 if ((npages
> 512) || (tce_list
& SPAPR_TCE_PAGE_MASK
)) {
493 page_mask
= IOMMU_PAGE_MASK(tcet
->page_shift
);
494 page_size
= IOMMU_PAGE_SIZE(tcet
->page_shift
);
497 for (i
= 0; i
< npages
; ++i
, ioba
+= page_size
) {
498 tce
= ldq_be_phys(cs
->as
, tce_list
+ i
* sizeof(target_ulong
));
500 ret
= put_tce_emu(tcet
, ioba
, tce
);
506 /* Trace last successful or the first problematic entry */
508 if (SPAPR_IS_PCI_LIOBN(liobn
)) {
509 trace_spapr_iommu_pci_indirect(liobn
, ioba1
, tce_list
, i
, tce
, ret
);
511 trace_spapr_iommu_indirect(liobn
, ioba1
, tce_list
, i
, tce
, ret
);
516 static target_ulong
h_stuff_tce(PowerPCCPU
*cpu
, SpaprMachineState
*spapr
,
517 target_ulong opcode
, target_ulong
*args
)
520 target_ulong liobn
= args
[0];
521 target_ulong ioba
= args
[1];
522 target_ulong tce_value
= args
[2];
523 target_ulong npages
= args
[3];
524 target_ulong ret
= H_PARAMETER
;
525 SpaprTceTable
*tcet
= spapr_tce_find_by_liobn(liobn
);
526 hwaddr page_mask
, page_size
;
532 if (npages
> tcet
->nb_table
) {
536 page_mask
= IOMMU_PAGE_MASK(tcet
->page_shift
);
537 page_size
= IOMMU_PAGE_SIZE(tcet
->page_shift
);
540 for (i
= 0; i
< npages
; ++i
, ioba
+= page_size
) {
541 ret
= put_tce_emu(tcet
, ioba
, tce_value
);
546 if (SPAPR_IS_PCI_LIOBN(liobn
)) {
547 trace_spapr_iommu_pci_stuff(liobn
, ioba
, tce_value
, npages
, ret
);
549 trace_spapr_iommu_stuff(liobn
, ioba
, tce_value
, npages
, ret
);
555 static target_ulong
h_put_tce(PowerPCCPU
*cpu
, SpaprMachineState
*spapr
,
556 target_ulong opcode
, target_ulong
*args
)
558 target_ulong liobn
= args
[0];
559 target_ulong ioba
= args
[1];
560 target_ulong tce
= args
[2];
561 target_ulong ret
= H_PARAMETER
;
562 SpaprTceTable
*tcet
= spapr_tce_find_by_liobn(liobn
);
565 hwaddr page_mask
= IOMMU_PAGE_MASK(tcet
->page_shift
);
569 ret
= put_tce_emu(tcet
, ioba
, tce
);
571 if (SPAPR_IS_PCI_LIOBN(liobn
)) {
572 trace_spapr_iommu_pci_put(liobn
, ioba
, tce
, ret
);
574 trace_spapr_iommu_put(liobn
, ioba
, tce
, ret
);
580 static target_ulong
get_tce_emu(SpaprTceTable
*tcet
, target_ulong ioba
,
583 unsigned long index
= (ioba
- tcet
->bus_offset
) >> tcet
->page_shift
;
585 if (index
>= tcet
->nb_table
) {
586 hcall_dprintf("spapr_iommu_get_tce on out-of-bounds IOBA 0x"
587 TARGET_FMT_lx
"\n", ioba
);
591 *tce
= tcet
->table
[index
];
596 static target_ulong
h_get_tce(PowerPCCPU
*cpu
, SpaprMachineState
*spapr
,
597 target_ulong opcode
, target_ulong
*args
)
599 target_ulong liobn
= args
[0];
600 target_ulong ioba
= args
[1];
601 target_ulong tce
= 0;
602 target_ulong ret
= H_PARAMETER
;
603 SpaprTceTable
*tcet
= spapr_tce_find_by_liobn(liobn
);
606 hwaddr page_mask
= IOMMU_PAGE_MASK(tcet
->page_shift
);
610 ret
= get_tce_emu(tcet
, ioba
, &tce
);
615 if (SPAPR_IS_PCI_LIOBN(liobn
)) {
616 trace_spapr_iommu_pci_get(liobn
, ioba
, ret
, tce
);
618 trace_spapr_iommu_get(liobn
, ioba
, ret
, tce
);
624 int spapr_dma_dt(void *fdt
, int node_off
, const char *propname
,
625 uint32_t liobn
, uint64_t window
, uint32_t size
)
627 uint32_t dma_prop
[5];
630 dma_prop
[0] = cpu_to_be32(liobn
);
631 dma_prop
[1] = cpu_to_be32(window
>> 32);
632 dma_prop
[2] = cpu_to_be32(window
& 0xFFFFFFFF);
633 dma_prop
[3] = 0; /* window size is 32 bits */
634 dma_prop
[4] = cpu_to_be32(size
);
636 ret
= fdt_setprop_cell(fdt
, node_off
, "ibm,#dma-address-cells", 2);
641 ret
= fdt_setprop_cell(fdt
, node_off
, "ibm,#dma-size-cells", 2);
646 ret
= fdt_setprop(fdt
, node_off
, propname
, dma_prop
, sizeof(dma_prop
));
654 int spapr_tcet_dma_dt(void *fdt
, int node_off
, const char *propname
,
661 return spapr_dma_dt(fdt
, node_off
, propname
,
662 tcet
->liobn
, 0, tcet
->nb_table
<< tcet
->page_shift
);
665 static void spapr_tce_table_class_init(ObjectClass
*klass
, void *data
)
667 DeviceClass
*dc
= DEVICE_CLASS(klass
);
668 dc
->realize
= spapr_tce_table_realize
;
669 dc
->reset
= spapr_tce_reset
;
670 dc
->unrealize
= spapr_tce_table_unrealize
;
671 /* Reason: This is just an internal device for handling the hypercalls */
672 dc
->user_creatable
= false;
674 QLIST_INIT(&spapr_tce_tables
);
677 spapr_register_hypercall(H_PUT_TCE
, h_put_tce
);
678 spapr_register_hypercall(H_GET_TCE
, h_get_tce
);
679 spapr_register_hypercall(H_PUT_TCE_INDIRECT
, h_put_tce_indirect
);
680 spapr_register_hypercall(H_STUFF_TCE
, h_stuff_tce
);
683 static TypeInfo spapr_tce_table_info
= {
684 .name
= TYPE_SPAPR_TCE_TABLE
,
685 .parent
= TYPE_DEVICE
,
686 .instance_size
= sizeof(SpaprTceTable
),
687 .class_init
= spapr_tce_table_class_init
,
690 static void spapr_iommu_memory_region_class_init(ObjectClass
*klass
, void *data
)
692 IOMMUMemoryRegionClass
*imrc
= IOMMU_MEMORY_REGION_CLASS(klass
);
694 imrc
->translate
= spapr_tce_translate_iommu
;
695 imrc
->replay
= spapr_tce_replay
;
696 imrc
->get_min_page_size
= spapr_tce_get_min_page_size
;
697 imrc
->notify_flag_changed
= spapr_tce_notify_flag_changed
;
698 imrc
->get_attr
= spapr_tce_get_attr
;
701 static const TypeInfo spapr_iommu_memory_region_info
= {
702 .parent
= TYPE_IOMMU_MEMORY_REGION
,
703 .name
= TYPE_SPAPR_IOMMU_MEMORY_REGION
,
704 .class_init
= spapr_iommu_memory_region_class_init
,
707 static void register_types(void)
709 type_register_static(&spapr_tce_table_info
);
710 type_register_static(&spapr_iommu_memory_region_info
);
713 type_init(register_types
);