vhost: adapt vhost_verify_ring_mappings() to virtio 1 ring layout
[qemu/ar7.git] / hw / ppc / spapr_iommu.c
blobae30bbe30f31050263367e16895f24ad93d7222c
1 /*
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/>.
19 #include "qemu/osdep.h"
20 #include "qemu/error-report.h"
21 #include "hw/hw.h"
22 #include "qemu/log.h"
23 #include "sysemu/kvm.h"
24 #include "hw/qdev.h"
25 #include "kvm_ppc.h"
26 #include "sysemu/dma.h"
27 #include "exec/address-spaces.h"
28 #include "trace.h"
30 #include "hw/ppc/spapr.h"
31 #include "hw/ppc/spapr_vio.h"
33 #include <libfdt.h>
35 enum sPAPRTCEAccess {
36 SPAPR_TCE_FAULT = 0,
37 SPAPR_TCE_RO = 1,
38 SPAPR_TCE_WO = 2,
39 SPAPR_TCE_RW = 3,
42 #define IOMMU_PAGE_SIZE(shift) (1ULL << (shift))
43 #define IOMMU_PAGE_MASK(shift) (~(IOMMU_PAGE_SIZE(shift) - 1))
45 static QLIST_HEAD(spapr_tce_tables, sPAPRTCETable) spapr_tce_tables;
47 sPAPRTCETable *spapr_tce_find_by_liobn(target_ulong liobn)
49 sPAPRTCETable *tcet;
51 if (liobn & 0xFFFFFFFF00000000ULL) {
52 hcall_dprintf("Request for out-of-bounds LIOBN 0x" TARGET_FMT_lx "\n",
53 liobn);
54 return NULL;
57 QLIST_FOREACH(tcet, &spapr_tce_tables, list) {
58 if (tcet->liobn == (uint32_t)liobn) {
59 return tcet;
63 return NULL;
66 static IOMMUAccessFlags spapr_tce_iommu_access_flags(uint64_t tce)
68 switch (tce & SPAPR_TCE_RW) {
69 case SPAPR_TCE_FAULT:
70 return IOMMU_NONE;
71 case SPAPR_TCE_RO:
72 return IOMMU_RO;
73 case SPAPR_TCE_WO:
74 return IOMMU_WO;
75 default: /* SPAPR_TCE_RW */
76 return IOMMU_RW;
80 static uint64_t *spapr_tce_alloc_table(uint32_t liobn,
81 uint32_t page_shift,
82 uint32_t nb_table,
83 int *fd,
84 bool need_vfio)
86 uint64_t *table = NULL;
87 uint64_t window_size = (uint64_t)nb_table << page_shift;
89 if (kvm_enabled() && !(window_size >> 32)) {
90 table = kvmppc_create_spapr_tce(liobn, window_size, fd, need_vfio);
93 if (!table) {
94 *fd = -1;
95 table = g_malloc0(nb_table * sizeof(uint64_t));
98 trace_spapr_iommu_new_table(liobn, table, *fd);
100 return table;
103 static void spapr_tce_free_table(uint64_t *table, int fd, uint32_t nb_table)
105 if (!kvm_enabled() ||
106 (kvmppc_remove_spapr_tce(table, fd, nb_table) != 0)) {
107 g_free(table);
111 /* Called from RCU critical section */
112 static IOMMUTLBEntry spapr_tce_translate_iommu(MemoryRegion *iommu, hwaddr addr,
113 bool is_write)
115 sPAPRTCETable *tcet = container_of(iommu, sPAPRTCETable, iommu);
116 uint64_t tce;
117 IOMMUTLBEntry ret = {
118 .target_as = &address_space_memory,
119 .iova = 0,
120 .translated_addr = 0,
121 .addr_mask = ~(hwaddr)0,
122 .perm = IOMMU_NONE,
125 if ((addr >> tcet->page_shift) < tcet->nb_table) {
126 /* Check if we are in bound */
127 hwaddr page_mask = IOMMU_PAGE_MASK(tcet->page_shift);
129 tce = tcet->table[addr >> tcet->page_shift];
130 ret.iova = addr & page_mask;
131 ret.translated_addr = tce & page_mask;
132 ret.addr_mask = ~page_mask;
133 ret.perm = spapr_tce_iommu_access_flags(tce);
135 trace_spapr_iommu_xlate(tcet->liobn, addr, ret.iova, ret.perm,
136 ret.addr_mask);
138 return ret;
141 static void spapr_tce_table_pre_save(void *opaque)
143 sPAPRTCETable *tcet = SPAPR_TCE_TABLE(opaque);
145 tcet->mig_table = tcet->table;
146 tcet->mig_nb_table = tcet->nb_table;
148 trace_spapr_iommu_pre_save(tcet->liobn, tcet->mig_nb_table,
149 tcet->bus_offset, tcet->page_shift);
152 static uint64_t spapr_tce_get_min_page_size(MemoryRegion *iommu)
154 sPAPRTCETable *tcet = container_of(iommu, sPAPRTCETable, iommu);
156 return 1ULL << tcet->page_shift;
159 static void spapr_tce_notify_flag_changed(MemoryRegion *iommu,
160 IOMMUNotifierFlag old,
161 IOMMUNotifierFlag new)
163 struct sPAPRTCETable *tbl = container_of(iommu, sPAPRTCETable, iommu);
165 if (old == IOMMU_NOTIFIER_NONE && new != IOMMU_NOTIFIER_NONE) {
166 spapr_tce_set_need_vfio(tbl, true);
167 } else if (old != IOMMU_NOTIFIER_NONE && new == IOMMU_NOTIFIER_NONE) {
168 spapr_tce_set_need_vfio(tbl, false);
172 static int spapr_tce_table_post_load(void *opaque, int version_id)
174 sPAPRTCETable *tcet = SPAPR_TCE_TABLE(opaque);
175 uint32_t old_nb_table = tcet->nb_table;
176 uint64_t old_bus_offset = tcet->bus_offset;
177 uint32_t old_page_shift = tcet->page_shift;
179 if (tcet->vdev) {
180 spapr_vio_set_bypass(tcet->vdev, tcet->bypass);
183 if (tcet->mig_nb_table != tcet->nb_table) {
184 spapr_tce_table_disable(tcet);
187 if (tcet->mig_nb_table) {
188 if (!tcet->nb_table) {
189 spapr_tce_table_enable(tcet, old_page_shift, old_bus_offset,
190 tcet->mig_nb_table);
193 memcpy(tcet->table, tcet->mig_table,
194 tcet->nb_table * sizeof(tcet->table[0]));
196 free(tcet->mig_table);
197 tcet->mig_table = NULL;
200 trace_spapr_iommu_post_load(tcet->liobn, old_nb_table, tcet->nb_table,
201 tcet->bus_offset, tcet->page_shift);
203 return 0;
206 static bool spapr_tce_table_ex_needed(void *opaque)
208 sPAPRTCETable *tcet = opaque;
210 return tcet->bus_offset || tcet->page_shift != 0xC;
213 static const VMStateDescription vmstate_spapr_tce_table_ex = {
214 .name = "spapr_iommu_ex",
215 .version_id = 1,
216 .minimum_version_id = 1,
217 .needed = spapr_tce_table_ex_needed,
218 .fields = (VMStateField[]) {
219 VMSTATE_UINT64(bus_offset, sPAPRTCETable),
220 VMSTATE_UINT32(page_shift, sPAPRTCETable),
221 VMSTATE_END_OF_LIST()
225 static const VMStateDescription vmstate_spapr_tce_table = {
226 .name = "spapr_iommu",
227 .version_id = 2,
228 .minimum_version_id = 2,
229 .pre_save = spapr_tce_table_pre_save,
230 .post_load = spapr_tce_table_post_load,
231 .fields = (VMStateField []) {
232 /* Sanity check */
233 VMSTATE_UINT32_EQUAL(liobn, sPAPRTCETable),
235 /* IOMMU state */
236 VMSTATE_UINT32(mig_nb_table, sPAPRTCETable),
237 VMSTATE_BOOL(bypass, sPAPRTCETable),
238 VMSTATE_VARRAY_UINT32_ALLOC(mig_table, sPAPRTCETable, mig_nb_table, 0,
239 vmstate_info_uint64, uint64_t),
241 VMSTATE_END_OF_LIST()
243 .subsections = (const VMStateDescription*[]) {
244 &vmstate_spapr_tce_table_ex,
245 NULL
249 static MemoryRegionIOMMUOps spapr_iommu_ops = {
250 .translate = spapr_tce_translate_iommu,
251 .get_min_page_size = spapr_tce_get_min_page_size,
252 .notify_flag_changed = spapr_tce_notify_flag_changed,
255 static int spapr_tce_table_realize(DeviceState *dev)
257 sPAPRTCETable *tcet = SPAPR_TCE_TABLE(dev);
258 Object *tcetobj = OBJECT(tcet);
259 char tmp[32];
261 tcet->fd = -1;
262 tcet->need_vfio = false;
263 snprintf(tmp, sizeof(tmp), "tce-root-%x", tcet->liobn);
264 memory_region_init(&tcet->root, tcetobj, tmp, UINT64_MAX);
266 snprintf(tmp, sizeof(tmp), "tce-iommu-%x", tcet->liobn);
267 memory_region_init_iommu(&tcet->iommu, tcetobj, &spapr_iommu_ops, tmp, 0);
269 QLIST_INSERT_HEAD(&spapr_tce_tables, tcet, list);
271 vmstate_register(DEVICE(tcet), tcet->liobn, &vmstate_spapr_tce_table,
272 tcet);
274 return 0;
277 void spapr_tce_set_need_vfio(sPAPRTCETable *tcet, bool need_vfio)
279 size_t table_size = tcet->nb_table * sizeof(uint64_t);
280 void *newtable;
282 if (need_vfio == tcet->need_vfio) {
283 /* Nothing to do */
284 return;
287 if (!need_vfio) {
288 /* FIXME: We don't support transition back to KVM accelerated
289 * TCEs yet */
290 return;
293 tcet->need_vfio = true;
295 if (tcet->fd < 0) {
296 /* Table is already in userspace, nothing to be do */
297 return;
300 newtable = g_malloc(table_size);
301 memcpy(newtable, tcet->table, table_size);
303 kvmppc_remove_spapr_tce(tcet->table, tcet->fd, tcet->nb_table);
305 tcet->fd = -1;
306 tcet->table = newtable;
309 sPAPRTCETable *spapr_tce_new_table(DeviceState *owner, uint32_t liobn)
311 sPAPRTCETable *tcet;
312 char tmp[32];
314 if (spapr_tce_find_by_liobn(liobn)) {
315 error_report("Attempted to create TCE table with duplicate"
316 " LIOBN 0x%x", liobn);
317 return NULL;
320 tcet = SPAPR_TCE_TABLE(object_new(TYPE_SPAPR_TCE_TABLE));
321 tcet->liobn = liobn;
323 snprintf(tmp, sizeof(tmp), "tce-table-%x", liobn);
324 object_property_add_child(OBJECT(owner), tmp, OBJECT(tcet), NULL);
326 object_property_set_bool(OBJECT(tcet), true, "realized", NULL);
328 return tcet;
331 void spapr_tce_table_enable(sPAPRTCETable *tcet,
332 uint32_t page_shift, uint64_t bus_offset,
333 uint32_t nb_table)
335 if (tcet->nb_table) {
336 error_report("Warning: trying to enable already enabled TCE table");
337 return;
340 tcet->bus_offset = bus_offset;
341 tcet->page_shift = page_shift;
342 tcet->nb_table = nb_table;
343 tcet->table = spapr_tce_alloc_table(tcet->liobn,
344 tcet->page_shift,
345 tcet->nb_table,
346 &tcet->fd,
347 tcet->need_vfio);
349 memory_region_set_size(&tcet->iommu,
350 (uint64_t)tcet->nb_table << tcet->page_shift);
351 memory_region_add_subregion(&tcet->root, tcet->bus_offset, &tcet->iommu);
354 void spapr_tce_table_disable(sPAPRTCETable *tcet)
356 if (!tcet->nb_table) {
357 return;
360 memory_region_del_subregion(&tcet->root, &tcet->iommu);
361 memory_region_set_size(&tcet->iommu, 0);
363 spapr_tce_free_table(tcet->table, tcet->fd, tcet->nb_table);
364 tcet->fd = -1;
365 tcet->table = NULL;
366 tcet->bus_offset = 0;
367 tcet->page_shift = 0;
368 tcet->nb_table = 0;
371 static void spapr_tce_table_unrealize(DeviceState *dev, Error **errp)
373 sPAPRTCETable *tcet = SPAPR_TCE_TABLE(dev);
375 QLIST_REMOVE(tcet, list);
377 spapr_tce_table_disable(tcet);
380 MemoryRegion *spapr_tce_get_iommu(sPAPRTCETable *tcet)
382 return &tcet->root;
385 static void spapr_tce_reset(DeviceState *dev)
387 sPAPRTCETable *tcet = SPAPR_TCE_TABLE(dev);
388 size_t table_size = tcet->nb_table * sizeof(uint64_t);
390 if (tcet->nb_table) {
391 memset(tcet->table, 0, table_size);
395 static target_ulong put_tce_emu(sPAPRTCETable *tcet, target_ulong ioba,
396 target_ulong tce)
398 IOMMUTLBEntry entry;
399 hwaddr page_mask = IOMMU_PAGE_MASK(tcet->page_shift);
400 unsigned long index = (ioba - tcet->bus_offset) >> tcet->page_shift;
402 if (index >= tcet->nb_table) {
403 hcall_dprintf("spapr_vio_put_tce on out-of-bounds IOBA 0x"
404 TARGET_FMT_lx "\n", ioba);
405 return H_PARAMETER;
408 tcet->table[index] = tce;
410 entry.target_as = &address_space_memory,
411 entry.iova = (ioba - tcet->bus_offset) & page_mask;
412 entry.translated_addr = tce & page_mask;
413 entry.addr_mask = ~page_mask;
414 entry.perm = spapr_tce_iommu_access_flags(tce);
415 memory_region_notify_iommu(&tcet->iommu, entry);
417 return H_SUCCESS;
420 static target_ulong h_put_tce_indirect(PowerPCCPU *cpu,
421 sPAPRMachineState *spapr,
422 target_ulong opcode, target_ulong *args)
424 int i;
425 target_ulong liobn = args[0];
426 target_ulong ioba = args[1];
427 target_ulong ioba1 = ioba;
428 target_ulong tce_list = args[2];
429 target_ulong npages = args[3];
430 target_ulong ret = H_PARAMETER, tce = 0;
431 sPAPRTCETable *tcet = spapr_tce_find_by_liobn(liobn);
432 CPUState *cs = CPU(cpu);
433 hwaddr page_mask, page_size;
435 if (!tcet) {
436 return H_PARAMETER;
439 if ((npages > 512) || (tce_list & SPAPR_TCE_PAGE_MASK)) {
440 return H_PARAMETER;
443 page_mask = IOMMU_PAGE_MASK(tcet->page_shift);
444 page_size = IOMMU_PAGE_SIZE(tcet->page_shift);
445 ioba &= page_mask;
447 for (i = 0; i < npages; ++i, ioba += page_size) {
448 tce = ldq_be_phys(cs->as, tce_list + i * sizeof(target_ulong));
450 ret = put_tce_emu(tcet, ioba, tce);
451 if (ret) {
452 break;
456 /* Trace last successful or the first problematic entry */
457 i = i ? (i - 1) : 0;
458 if (SPAPR_IS_PCI_LIOBN(liobn)) {
459 trace_spapr_iommu_pci_indirect(liobn, ioba1, tce_list, i, tce, ret);
460 } else {
461 trace_spapr_iommu_indirect(liobn, ioba1, tce_list, i, tce, ret);
463 return ret;
466 static target_ulong h_stuff_tce(PowerPCCPU *cpu, sPAPRMachineState *spapr,
467 target_ulong opcode, target_ulong *args)
469 int i;
470 target_ulong liobn = args[0];
471 target_ulong ioba = args[1];
472 target_ulong tce_value = args[2];
473 target_ulong npages = args[3];
474 target_ulong ret = H_PARAMETER;
475 sPAPRTCETable *tcet = spapr_tce_find_by_liobn(liobn);
476 hwaddr page_mask, page_size;
478 if (!tcet) {
479 return H_PARAMETER;
482 if (npages > tcet->nb_table) {
483 return H_PARAMETER;
486 page_mask = IOMMU_PAGE_MASK(tcet->page_shift);
487 page_size = IOMMU_PAGE_SIZE(tcet->page_shift);
488 ioba &= page_mask;
490 for (i = 0; i < npages; ++i, ioba += page_size) {
491 ret = put_tce_emu(tcet, ioba, tce_value);
492 if (ret) {
493 break;
496 if (SPAPR_IS_PCI_LIOBN(liobn)) {
497 trace_spapr_iommu_pci_stuff(liobn, ioba, tce_value, npages, ret);
498 } else {
499 trace_spapr_iommu_stuff(liobn, ioba, tce_value, npages, ret);
502 return ret;
505 static target_ulong h_put_tce(PowerPCCPU *cpu, sPAPRMachineState *spapr,
506 target_ulong opcode, target_ulong *args)
508 target_ulong liobn = args[0];
509 target_ulong ioba = args[1];
510 target_ulong tce = args[2];
511 target_ulong ret = H_PARAMETER;
512 sPAPRTCETable *tcet = spapr_tce_find_by_liobn(liobn);
514 if (tcet) {
515 hwaddr page_mask = IOMMU_PAGE_MASK(tcet->page_shift);
517 ioba &= page_mask;
519 ret = put_tce_emu(tcet, ioba, tce);
521 if (SPAPR_IS_PCI_LIOBN(liobn)) {
522 trace_spapr_iommu_pci_put(liobn, ioba, tce, ret);
523 } else {
524 trace_spapr_iommu_put(liobn, ioba, tce, ret);
527 return ret;
530 static target_ulong get_tce_emu(sPAPRTCETable *tcet, target_ulong ioba,
531 target_ulong *tce)
533 unsigned long index = (ioba - tcet->bus_offset) >> tcet->page_shift;
535 if (index >= tcet->nb_table) {
536 hcall_dprintf("spapr_iommu_get_tce on out-of-bounds IOBA 0x"
537 TARGET_FMT_lx "\n", ioba);
538 return H_PARAMETER;
541 *tce = tcet->table[index];
543 return H_SUCCESS;
546 static target_ulong h_get_tce(PowerPCCPU *cpu, sPAPRMachineState *spapr,
547 target_ulong opcode, target_ulong *args)
549 target_ulong liobn = args[0];
550 target_ulong ioba = args[1];
551 target_ulong tce = 0;
552 target_ulong ret = H_PARAMETER;
553 sPAPRTCETable *tcet = spapr_tce_find_by_liobn(liobn);
555 if (tcet) {
556 hwaddr page_mask = IOMMU_PAGE_MASK(tcet->page_shift);
558 ioba &= page_mask;
560 ret = get_tce_emu(tcet, ioba, &tce);
561 if (!ret) {
562 args[0] = tce;
565 if (SPAPR_IS_PCI_LIOBN(liobn)) {
566 trace_spapr_iommu_pci_get(liobn, ioba, ret, tce);
567 } else {
568 trace_spapr_iommu_get(liobn, ioba, ret, tce);
571 return ret;
574 int spapr_dma_dt(void *fdt, int node_off, const char *propname,
575 uint32_t liobn, uint64_t window, uint32_t size)
577 uint32_t dma_prop[5];
578 int ret;
580 dma_prop[0] = cpu_to_be32(liobn);
581 dma_prop[1] = cpu_to_be32(window >> 32);
582 dma_prop[2] = cpu_to_be32(window & 0xFFFFFFFF);
583 dma_prop[3] = 0; /* window size is 32 bits */
584 dma_prop[4] = cpu_to_be32(size);
586 ret = fdt_setprop_cell(fdt, node_off, "ibm,#dma-address-cells", 2);
587 if (ret < 0) {
588 return ret;
591 ret = fdt_setprop_cell(fdt, node_off, "ibm,#dma-size-cells", 2);
592 if (ret < 0) {
593 return ret;
596 ret = fdt_setprop(fdt, node_off, propname, dma_prop, sizeof(dma_prop));
597 if (ret < 0) {
598 return ret;
601 return 0;
604 int spapr_tcet_dma_dt(void *fdt, int node_off, const char *propname,
605 sPAPRTCETable *tcet)
607 if (!tcet) {
608 return 0;
611 return spapr_dma_dt(fdt, node_off, propname,
612 tcet->liobn, 0, tcet->nb_table << tcet->page_shift);
615 static void spapr_tce_table_class_init(ObjectClass *klass, void *data)
617 DeviceClass *dc = DEVICE_CLASS(klass);
618 dc->init = spapr_tce_table_realize;
619 dc->reset = spapr_tce_reset;
620 dc->unrealize = spapr_tce_table_unrealize;
622 QLIST_INIT(&spapr_tce_tables);
624 /* hcall-tce */
625 spapr_register_hypercall(H_PUT_TCE, h_put_tce);
626 spapr_register_hypercall(H_GET_TCE, h_get_tce);
627 spapr_register_hypercall(H_PUT_TCE_INDIRECT, h_put_tce_indirect);
628 spapr_register_hypercall(H_STUFF_TCE, h_stuff_tce);
631 static TypeInfo spapr_tce_table_info = {
632 .name = TYPE_SPAPR_TCE_TABLE,
633 .parent = TYPE_DEVICE,
634 .instance_size = sizeof(sPAPRTCETable),
635 .class_init = spapr_tce_table_class_init,
638 static void register_types(void)
640 type_register_static(&spapr_tce_table_info);
643 type_init(register_types);