block-backend: Decrease in_flight only after callback
[qemu/ar7.git] / hw / ppc / spapr_iommu.c
blob1b0880ac9edb39d1f9fb1a31aa25746e1bea7687
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 uint64_t bus_offset,
83 uint32_t nb_table,
84 int *fd,
85 bool need_vfio)
87 uint64_t *table = NULL;
89 if (kvm_enabled()) {
90 table = kvmppc_create_spapr_tce(liobn, page_shift, bus_offset, nb_table,
91 fd, need_vfio);
94 if (!table) {
95 *fd = -1;
96 table = g_malloc0(nb_table * sizeof(uint64_t));
99 trace_spapr_iommu_new_table(liobn, table, *fd);
101 return table;
104 static void spapr_tce_free_table(uint64_t *table, int fd, uint32_t nb_table)
106 if (!kvm_enabled() ||
107 (kvmppc_remove_spapr_tce(table, fd, nb_table) != 0)) {
108 g_free(table);
112 /* Called from RCU critical section */
113 static IOMMUTLBEntry spapr_tce_translate_iommu(IOMMUMemoryRegion *iommu,
114 hwaddr addr,
115 IOMMUAccessFlags flag,
116 int iommu_idx)
118 sPAPRTCETable *tcet = container_of(iommu, sPAPRTCETable, iommu);
119 uint64_t tce;
120 IOMMUTLBEntry ret = {
121 .target_as = &address_space_memory,
122 .iova = 0,
123 .translated_addr = 0,
124 .addr_mask = ~(hwaddr)0,
125 .perm = IOMMU_NONE,
128 if ((addr >> tcet->page_shift) < tcet->nb_table) {
129 /* Check if we are in bound */
130 hwaddr page_mask = IOMMU_PAGE_MASK(tcet->page_shift);
132 tce = tcet->table[addr >> tcet->page_shift];
133 ret.iova = addr & page_mask;
134 ret.translated_addr = tce & page_mask;
135 ret.addr_mask = ~page_mask;
136 ret.perm = spapr_tce_iommu_access_flags(tce);
138 trace_spapr_iommu_xlate(tcet->liobn, addr, ret.iova, ret.perm,
139 ret.addr_mask);
141 return ret;
144 static int spapr_tce_table_pre_save(void *opaque)
146 sPAPRTCETable *tcet = SPAPR_TCE_TABLE(opaque);
148 tcet->mig_table = tcet->table;
149 tcet->mig_nb_table = tcet->nb_table;
151 trace_spapr_iommu_pre_save(tcet->liobn, tcet->mig_nb_table,
152 tcet->bus_offset, tcet->page_shift);
154 return 0;
157 static uint64_t spapr_tce_get_min_page_size(IOMMUMemoryRegion *iommu)
159 sPAPRTCETable *tcet = container_of(iommu, sPAPRTCETable, iommu);
161 return 1ULL << tcet->page_shift;
164 static int spapr_tce_get_attr(IOMMUMemoryRegion *iommu,
165 enum IOMMUMemoryRegionAttr attr, void *data)
167 sPAPRTCETable *tcet = container_of(iommu, sPAPRTCETable, iommu);
169 if (attr == IOMMU_ATTR_SPAPR_TCE_FD && kvmppc_has_cap_spapr_vfio()) {
170 *(int *) data = tcet->fd;
171 return 0;
174 return -EINVAL;
177 static void spapr_tce_notify_flag_changed(IOMMUMemoryRegion *iommu,
178 IOMMUNotifierFlag old,
179 IOMMUNotifierFlag new)
181 struct sPAPRTCETable *tbl = container_of(iommu, sPAPRTCETable, iommu);
183 if (old == IOMMU_NOTIFIER_NONE && new != IOMMU_NOTIFIER_NONE) {
184 spapr_tce_set_need_vfio(tbl, true);
185 } else if (old != IOMMU_NOTIFIER_NONE && new == IOMMU_NOTIFIER_NONE) {
186 spapr_tce_set_need_vfio(tbl, false);
190 static int spapr_tce_table_post_load(void *opaque, int version_id)
192 sPAPRTCETable *tcet = SPAPR_TCE_TABLE(opaque);
193 uint32_t old_nb_table = tcet->nb_table;
194 uint64_t old_bus_offset = tcet->bus_offset;
195 uint32_t old_page_shift = tcet->page_shift;
197 if (tcet->vdev) {
198 spapr_vio_set_bypass(tcet->vdev, tcet->bypass);
201 if (tcet->mig_nb_table != tcet->nb_table) {
202 spapr_tce_table_disable(tcet);
205 if (tcet->mig_nb_table) {
206 if (!tcet->nb_table) {
207 spapr_tce_table_enable(tcet, old_page_shift, old_bus_offset,
208 tcet->mig_nb_table);
211 memcpy(tcet->table, tcet->mig_table,
212 tcet->nb_table * sizeof(tcet->table[0]));
214 free(tcet->mig_table);
215 tcet->mig_table = NULL;
218 trace_spapr_iommu_post_load(tcet->liobn, old_nb_table, tcet->nb_table,
219 tcet->bus_offset, tcet->page_shift);
221 return 0;
224 static bool spapr_tce_table_ex_needed(void *opaque)
226 sPAPRTCETable *tcet = opaque;
228 return tcet->bus_offset || tcet->page_shift != 0xC;
231 static const VMStateDescription vmstate_spapr_tce_table_ex = {
232 .name = "spapr_iommu_ex",
233 .version_id = 1,
234 .minimum_version_id = 1,
235 .needed = spapr_tce_table_ex_needed,
236 .fields = (VMStateField[]) {
237 VMSTATE_UINT64(bus_offset, sPAPRTCETable),
238 VMSTATE_UINT32(page_shift, sPAPRTCETable),
239 VMSTATE_END_OF_LIST()
243 static const VMStateDescription vmstate_spapr_tce_table = {
244 .name = "spapr_iommu",
245 .version_id = 2,
246 .minimum_version_id = 2,
247 .pre_save = spapr_tce_table_pre_save,
248 .post_load = spapr_tce_table_post_load,
249 .fields = (VMStateField []) {
250 /* Sanity check */
251 VMSTATE_UINT32_EQUAL(liobn, sPAPRTCETable, NULL),
253 /* IOMMU state */
254 VMSTATE_UINT32(mig_nb_table, sPAPRTCETable),
255 VMSTATE_BOOL(bypass, sPAPRTCETable),
256 VMSTATE_VARRAY_UINT32_ALLOC(mig_table, sPAPRTCETable, mig_nb_table, 0,
257 vmstate_info_uint64, uint64_t),
259 VMSTATE_END_OF_LIST()
261 .subsections = (const VMStateDescription*[]) {
262 &vmstate_spapr_tce_table_ex,
263 NULL
267 static void spapr_tce_table_realize(DeviceState *dev, Error **errp)
269 sPAPRTCETable *tcet = SPAPR_TCE_TABLE(dev);
270 Object *tcetobj = OBJECT(tcet);
271 gchar *tmp;
273 tcet->fd = -1;
274 tcet->need_vfio = false;
275 tmp = g_strdup_printf("tce-root-%x", tcet->liobn);
276 memory_region_init(&tcet->root, tcetobj, tmp, UINT64_MAX);
277 g_free(tmp);
279 tmp = g_strdup_printf("tce-iommu-%x", tcet->liobn);
280 memory_region_init_iommu(&tcet->iommu, sizeof(tcet->iommu),
281 TYPE_SPAPR_IOMMU_MEMORY_REGION,
282 tcetobj, tmp, 0);
283 g_free(tmp);
285 QLIST_INSERT_HEAD(&spapr_tce_tables, tcet, list);
287 vmstate_register(DEVICE(tcet), tcet->liobn, &vmstate_spapr_tce_table,
288 tcet);
291 void spapr_tce_set_need_vfio(sPAPRTCETable *tcet, bool need_vfio)
293 size_t table_size = tcet->nb_table * sizeof(uint64_t);
294 uint64_t *oldtable;
295 int newfd = -1;
297 g_assert(need_vfio != tcet->need_vfio);
299 tcet->need_vfio = need_vfio;
301 if (!need_vfio || (tcet->fd != -1 && kvmppc_has_cap_spapr_vfio())) {
302 return;
305 oldtable = tcet->table;
307 tcet->table = spapr_tce_alloc_table(tcet->liobn,
308 tcet->page_shift,
309 tcet->bus_offset,
310 tcet->nb_table,
311 &newfd,
312 need_vfio);
313 memcpy(tcet->table, oldtable, table_size);
315 spapr_tce_free_table(oldtable, tcet->fd, tcet->nb_table);
317 tcet->fd = newfd;
320 sPAPRTCETable *spapr_tce_new_table(DeviceState *owner, uint32_t liobn)
322 sPAPRTCETable *tcet;
323 gchar *tmp;
325 if (spapr_tce_find_by_liobn(liobn)) {
326 error_report("Attempted to create TCE table with duplicate"
327 " LIOBN 0x%x", liobn);
328 return NULL;
331 tcet = SPAPR_TCE_TABLE(object_new(TYPE_SPAPR_TCE_TABLE));
332 tcet->liobn = liobn;
334 tmp = g_strdup_printf("tce-table-%x", liobn);
335 object_property_add_child(OBJECT(owner), tmp, OBJECT(tcet), NULL);
336 g_free(tmp);
337 object_unref(OBJECT(tcet));
339 object_property_set_bool(OBJECT(tcet), true, "realized", NULL);
341 return tcet;
344 void spapr_tce_table_enable(sPAPRTCETable *tcet,
345 uint32_t page_shift, uint64_t bus_offset,
346 uint32_t nb_table)
348 if (tcet->nb_table) {
349 warn_report("trying to enable already enabled TCE table");
350 return;
353 tcet->bus_offset = bus_offset;
354 tcet->page_shift = page_shift;
355 tcet->nb_table = nb_table;
356 tcet->table = spapr_tce_alloc_table(tcet->liobn,
357 tcet->page_shift,
358 tcet->bus_offset,
359 tcet->nb_table,
360 &tcet->fd,
361 tcet->need_vfio);
363 memory_region_set_size(MEMORY_REGION(&tcet->iommu),
364 (uint64_t)tcet->nb_table << tcet->page_shift);
365 memory_region_add_subregion(&tcet->root, tcet->bus_offset,
366 MEMORY_REGION(&tcet->iommu));
369 void spapr_tce_table_disable(sPAPRTCETable *tcet)
371 if (!tcet->nb_table) {
372 return;
375 memory_region_del_subregion(&tcet->root, MEMORY_REGION(&tcet->iommu));
376 memory_region_set_size(MEMORY_REGION(&tcet->iommu), 0);
378 spapr_tce_free_table(tcet->table, tcet->fd, tcet->nb_table);
379 tcet->fd = -1;
380 tcet->table = NULL;
381 tcet->bus_offset = 0;
382 tcet->page_shift = 0;
383 tcet->nb_table = 0;
386 static void spapr_tce_table_unrealize(DeviceState *dev, Error **errp)
388 sPAPRTCETable *tcet = SPAPR_TCE_TABLE(dev);
390 vmstate_unregister(DEVICE(tcet), &vmstate_spapr_tce_table, tcet);
392 QLIST_REMOVE(tcet, list);
394 spapr_tce_table_disable(tcet);
397 MemoryRegion *spapr_tce_get_iommu(sPAPRTCETable *tcet)
399 return &tcet->root;
402 static void spapr_tce_reset(DeviceState *dev)
404 sPAPRTCETable *tcet = SPAPR_TCE_TABLE(dev);
405 size_t table_size = tcet->nb_table * sizeof(uint64_t);
407 if (tcet->nb_table) {
408 memset(tcet->table, 0, table_size);
412 static target_ulong put_tce_emu(sPAPRTCETable *tcet, target_ulong ioba,
413 target_ulong tce)
415 IOMMUTLBEntry entry;
416 hwaddr page_mask = IOMMU_PAGE_MASK(tcet->page_shift);
417 unsigned long index = (ioba - tcet->bus_offset) >> tcet->page_shift;
419 if (index >= tcet->nb_table) {
420 hcall_dprintf("spapr_vio_put_tce on out-of-bounds IOBA 0x"
421 TARGET_FMT_lx "\n", ioba);
422 return H_PARAMETER;
425 tcet->table[index] = tce;
427 entry.target_as = &address_space_memory,
428 entry.iova = (ioba - tcet->bus_offset) & page_mask;
429 entry.translated_addr = tce & page_mask;
430 entry.addr_mask = ~page_mask;
431 entry.perm = spapr_tce_iommu_access_flags(tce);
432 memory_region_notify_iommu(&tcet->iommu, 0, entry);
434 return H_SUCCESS;
437 static target_ulong h_put_tce_indirect(PowerPCCPU *cpu,
438 sPAPRMachineState *spapr,
439 target_ulong opcode, target_ulong *args)
441 int i;
442 target_ulong liobn = args[0];
443 target_ulong ioba = args[1];
444 target_ulong ioba1 = ioba;
445 target_ulong tce_list = args[2];
446 target_ulong npages = args[3];
447 target_ulong ret = H_PARAMETER, tce = 0;
448 sPAPRTCETable *tcet = spapr_tce_find_by_liobn(liobn);
449 CPUState *cs = CPU(cpu);
450 hwaddr page_mask, page_size;
452 if (!tcet) {
453 return H_PARAMETER;
456 if ((npages > 512) || (tce_list & SPAPR_TCE_PAGE_MASK)) {
457 return H_PARAMETER;
460 page_mask = IOMMU_PAGE_MASK(tcet->page_shift);
461 page_size = IOMMU_PAGE_SIZE(tcet->page_shift);
462 ioba &= page_mask;
464 for (i = 0; i < npages; ++i, ioba += page_size) {
465 tce = ldq_be_phys(cs->as, tce_list + i * sizeof(target_ulong));
467 ret = put_tce_emu(tcet, ioba, tce);
468 if (ret) {
469 break;
473 /* Trace last successful or the first problematic entry */
474 i = i ? (i - 1) : 0;
475 if (SPAPR_IS_PCI_LIOBN(liobn)) {
476 trace_spapr_iommu_pci_indirect(liobn, ioba1, tce_list, i, tce, ret);
477 } else {
478 trace_spapr_iommu_indirect(liobn, ioba1, tce_list, i, tce, ret);
480 return ret;
483 static target_ulong h_stuff_tce(PowerPCCPU *cpu, sPAPRMachineState *spapr,
484 target_ulong opcode, target_ulong *args)
486 int i;
487 target_ulong liobn = args[0];
488 target_ulong ioba = args[1];
489 target_ulong tce_value = args[2];
490 target_ulong npages = args[3];
491 target_ulong ret = H_PARAMETER;
492 sPAPRTCETable *tcet = spapr_tce_find_by_liobn(liobn);
493 hwaddr page_mask, page_size;
495 if (!tcet) {
496 return H_PARAMETER;
499 if (npages > tcet->nb_table) {
500 return H_PARAMETER;
503 page_mask = IOMMU_PAGE_MASK(tcet->page_shift);
504 page_size = IOMMU_PAGE_SIZE(tcet->page_shift);
505 ioba &= page_mask;
507 for (i = 0; i < npages; ++i, ioba += page_size) {
508 ret = put_tce_emu(tcet, ioba, tce_value);
509 if (ret) {
510 break;
513 if (SPAPR_IS_PCI_LIOBN(liobn)) {
514 trace_spapr_iommu_pci_stuff(liobn, ioba, tce_value, npages, ret);
515 } else {
516 trace_spapr_iommu_stuff(liobn, ioba, tce_value, npages, ret);
519 return ret;
522 static target_ulong h_put_tce(PowerPCCPU *cpu, sPAPRMachineState *spapr,
523 target_ulong opcode, target_ulong *args)
525 target_ulong liobn = args[0];
526 target_ulong ioba = args[1];
527 target_ulong tce = args[2];
528 target_ulong ret = H_PARAMETER;
529 sPAPRTCETable *tcet = spapr_tce_find_by_liobn(liobn);
531 if (tcet) {
532 hwaddr page_mask = IOMMU_PAGE_MASK(tcet->page_shift);
534 ioba &= page_mask;
536 ret = put_tce_emu(tcet, ioba, tce);
538 if (SPAPR_IS_PCI_LIOBN(liobn)) {
539 trace_spapr_iommu_pci_put(liobn, ioba, tce, ret);
540 } else {
541 trace_spapr_iommu_put(liobn, ioba, tce, ret);
544 return ret;
547 static target_ulong get_tce_emu(sPAPRTCETable *tcet, target_ulong ioba,
548 target_ulong *tce)
550 unsigned long index = (ioba - tcet->bus_offset) >> tcet->page_shift;
552 if (index >= tcet->nb_table) {
553 hcall_dprintf("spapr_iommu_get_tce on out-of-bounds IOBA 0x"
554 TARGET_FMT_lx "\n", ioba);
555 return H_PARAMETER;
558 *tce = tcet->table[index];
560 return H_SUCCESS;
563 static target_ulong h_get_tce(PowerPCCPU *cpu, sPAPRMachineState *spapr,
564 target_ulong opcode, target_ulong *args)
566 target_ulong liobn = args[0];
567 target_ulong ioba = args[1];
568 target_ulong tce = 0;
569 target_ulong ret = H_PARAMETER;
570 sPAPRTCETable *tcet = spapr_tce_find_by_liobn(liobn);
572 if (tcet) {
573 hwaddr page_mask = IOMMU_PAGE_MASK(tcet->page_shift);
575 ioba &= page_mask;
577 ret = get_tce_emu(tcet, ioba, &tce);
578 if (!ret) {
579 args[0] = tce;
582 if (SPAPR_IS_PCI_LIOBN(liobn)) {
583 trace_spapr_iommu_pci_get(liobn, ioba, ret, tce);
584 } else {
585 trace_spapr_iommu_get(liobn, ioba, ret, tce);
588 return ret;
591 int spapr_dma_dt(void *fdt, int node_off, const char *propname,
592 uint32_t liobn, uint64_t window, uint32_t size)
594 uint32_t dma_prop[5];
595 int ret;
597 dma_prop[0] = cpu_to_be32(liobn);
598 dma_prop[1] = cpu_to_be32(window >> 32);
599 dma_prop[2] = cpu_to_be32(window & 0xFFFFFFFF);
600 dma_prop[3] = 0; /* window size is 32 bits */
601 dma_prop[4] = cpu_to_be32(size);
603 ret = fdt_setprop_cell(fdt, node_off, "ibm,#dma-address-cells", 2);
604 if (ret < 0) {
605 return ret;
608 ret = fdt_setprop_cell(fdt, node_off, "ibm,#dma-size-cells", 2);
609 if (ret < 0) {
610 return ret;
613 ret = fdt_setprop(fdt, node_off, propname, dma_prop, sizeof(dma_prop));
614 if (ret < 0) {
615 return ret;
618 return 0;
621 int spapr_tcet_dma_dt(void *fdt, int node_off, const char *propname,
622 sPAPRTCETable *tcet)
624 if (!tcet) {
625 return 0;
628 return spapr_dma_dt(fdt, node_off, propname,
629 tcet->liobn, 0, tcet->nb_table << tcet->page_shift);
632 static void spapr_tce_table_class_init(ObjectClass *klass, void *data)
634 DeviceClass *dc = DEVICE_CLASS(klass);
635 dc->realize = spapr_tce_table_realize;
636 dc->reset = spapr_tce_reset;
637 dc->unrealize = spapr_tce_table_unrealize;
638 /* Reason: This is just an internal device for handling the hypercalls */
639 dc->user_creatable = false;
641 QLIST_INIT(&spapr_tce_tables);
643 /* hcall-tce */
644 spapr_register_hypercall(H_PUT_TCE, h_put_tce);
645 spapr_register_hypercall(H_GET_TCE, h_get_tce);
646 spapr_register_hypercall(H_PUT_TCE_INDIRECT, h_put_tce_indirect);
647 spapr_register_hypercall(H_STUFF_TCE, h_stuff_tce);
650 static TypeInfo spapr_tce_table_info = {
651 .name = TYPE_SPAPR_TCE_TABLE,
652 .parent = TYPE_DEVICE,
653 .instance_size = sizeof(sPAPRTCETable),
654 .class_init = spapr_tce_table_class_init,
657 static void spapr_iommu_memory_region_class_init(ObjectClass *klass, void *data)
659 IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_CLASS(klass);
661 imrc->translate = spapr_tce_translate_iommu;
662 imrc->get_min_page_size = spapr_tce_get_min_page_size;
663 imrc->notify_flag_changed = spapr_tce_notify_flag_changed;
664 imrc->get_attr = spapr_tce_get_attr;
667 static const TypeInfo spapr_iommu_memory_region_info = {
668 .parent = TYPE_IOMMU_MEMORY_REGION,
669 .name = TYPE_SPAPR_IOMMU_MEMORY_REGION,
670 .class_init = spapr_iommu_memory_region_class_init,
673 static void register_types(void)
675 type_register_static(&spapr_tce_table_info);
676 type_register_static(&spapr_iommu_memory_region_info);
679 type_init(register_types);