xref: /openbmc/qemu/hw/ppc/spapr_pci_vfio.c (revision 61d9f32b)
1 /*
2  * QEMU sPAPR PCI host for VFIO
3  *
4  * Copyright (c) 2011-2014 Alexey Kardashevskiy, IBM Corporation.
5  *
6  *  This program is free software; you can redistribute it and/or modify
7  *  it under the terms of the GNU General Public License as published by
8  *  the Free Software Foundation; either version 2 of the License,
9  *  or (at your option) any later version.
10  *
11  *  This program 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
14  *  GNU General Public License for more details.
15  *
16  *  You should have received a copy of the GNU General Public License
17  *  along with this program; if not, see <http://www.gnu.org/licenses/>.
18  */
19 
20 #include "qemu/osdep.h"
21 #include "hw/ppc/spapr.h"
22 #include "hw/pci-host/spapr.h"
23 #include "hw/pci/msix.h"
24 #include "linux/vfio.h"
25 #include "hw/vfio/vfio.h"
26 
27 static Property spapr_phb_vfio_properties[] = {
28     DEFINE_PROP_INT32("iommu", sPAPRPHBVFIOState, iommugroupid, -1),
29     DEFINE_PROP_END_OF_LIST(),
30 };
31 
32 static void spapr_phb_vfio_finish_realize(sPAPRPHBState *sphb, Error **errp)
33 {
34     sPAPRPHBVFIOState *svphb = SPAPR_PCI_VFIO_HOST_BRIDGE(sphb);
35     struct vfio_iommu_spapr_tce_info info = { .argsz = sizeof(info) };
36     int ret;
37     sPAPRTCETable *tcet;
38     uint32_t liobn = svphb->phb.dma_liobn;
39 
40     if (svphb->iommugroupid == -1) {
41         error_setg(errp, "Wrong IOMMU group ID %d", svphb->iommugroupid);
42         return;
43     }
44 
45     ret = vfio_container_ioctl(&svphb->phb.iommu_as, svphb->iommugroupid,
46                                VFIO_CHECK_EXTENSION,
47                                (void *) VFIO_SPAPR_TCE_IOMMU);
48     if (ret != 1) {
49         error_setg_errno(errp, -ret,
50                          "spapr-vfio: SPAPR extension is not supported");
51         return;
52     }
53 
54     ret = vfio_container_ioctl(&svphb->phb.iommu_as, svphb->iommugroupid,
55                                VFIO_IOMMU_SPAPR_TCE_GET_INFO, &info);
56     if (ret) {
57         error_setg_errno(errp, -ret,
58                          "spapr-vfio: get info from container failed");
59         return;
60     }
61 
62     tcet = spapr_tce_new_table(DEVICE(sphb), liobn, info.dma32_window_start,
63                                SPAPR_TCE_PAGE_SHIFT,
64                                info.dma32_window_size >> SPAPR_TCE_PAGE_SHIFT,
65                                true);
66     if (!tcet) {
67         error_setg(errp, "spapr-vfio: failed to create VFIO TCE table");
68         return;
69     }
70 
71     /* Register default 32bit DMA window */
72     memory_region_add_subregion(&sphb->iommu_root, tcet->bus_offset,
73                                 spapr_tce_get_iommu(tcet));
74 }
75 
76 static void spapr_phb_vfio_eeh_reenable(sPAPRPHBVFIOState *svphb)
77 {
78     struct vfio_eeh_pe_op op = {
79         .argsz = sizeof(op),
80         .op    = VFIO_EEH_PE_ENABLE
81     };
82 
83     vfio_container_ioctl(&svphb->phb.iommu_as,
84                          svphb->iommugroupid, VFIO_EEH_PE_OP, &op);
85 }
86 
87 static void spapr_phb_vfio_reset(DeviceState *qdev)
88 {
89     /*
90      * The PE might be in frozen state. To reenable the EEH
91      * functionality on it will clean the frozen state, which
92      * ensures that the contained PCI devices will work properly
93      * after reboot.
94      */
95     spapr_phb_vfio_eeh_reenable(SPAPR_PCI_VFIO_HOST_BRIDGE(qdev));
96 }
97 
98 static int spapr_phb_vfio_eeh_set_option(sPAPRPHBState *sphb,
99                                          unsigned int addr, int option)
100 {
101     sPAPRPHBVFIOState *svphb = SPAPR_PCI_VFIO_HOST_BRIDGE(sphb);
102     struct vfio_eeh_pe_op op = { .argsz = sizeof(op) };
103     int ret;
104 
105     switch (option) {
106     case RTAS_EEH_DISABLE:
107         op.op = VFIO_EEH_PE_DISABLE;
108         break;
109     case RTAS_EEH_ENABLE: {
110         PCIHostState *phb;
111         PCIDevice *pdev;
112 
113         /*
114          * The EEH functionality is enabled on basis of PCI device,
115          * instead of PE. We need check the validity of the PCI
116          * device address.
117          */
118         phb = PCI_HOST_BRIDGE(sphb);
119         pdev = pci_find_device(phb->bus,
120                                (addr >> 16) & 0xFF, (addr >> 8) & 0xFF);
121         if (!pdev || !object_dynamic_cast(OBJECT(pdev), "vfio-pci")) {
122             return RTAS_OUT_PARAM_ERROR;
123         }
124 
125         op.op = VFIO_EEH_PE_ENABLE;
126         break;
127     }
128     case RTAS_EEH_THAW_IO:
129         op.op = VFIO_EEH_PE_UNFREEZE_IO;
130         break;
131     case RTAS_EEH_THAW_DMA:
132         op.op = VFIO_EEH_PE_UNFREEZE_DMA;
133         break;
134     default:
135         return RTAS_OUT_PARAM_ERROR;
136     }
137 
138     ret = vfio_container_ioctl(&svphb->phb.iommu_as, svphb->iommugroupid,
139                                VFIO_EEH_PE_OP, &op);
140     if (ret < 0) {
141         return RTAS_OUT_HW_ERROR;
142     }
143 
144     return RTAS_OUT_SUCCESS;
145 }
146 
147 static int spapr_phb_vfio_eeh_get_state(sPAPRPHBState *sphb, int *state)
148 {
149     sPAPRPHBVFIOState *svphb = SPAPR_PCI_VFIO_HOST_BRIDGE(sphb);
150     struct vfio_eeh_pe_op op = { .argsz = sizeof(op) };
151     int ret;
152 
153     op.op = VFIO_EEH_PE_GET_STATE;
154     ret = vfio_container_ioctl(&svphb->phb.iommu_as, svphb->iommugroupid,
155                                VFIO_EEH_PE_OP, &op);
156     if (ret < 0) {
157         return RTAS_OUT_PARAM_ERROR;
158     }
159 
160     *state = ret;
161     return RTAS_OUT_SUCCESS;
162 }
163 
164 static void spapr_phb_vfio_eeh_clear_dev_msix(PCIBus *bus,
165                                               PCIDevice *pdev,
166                                               void *opaque)
167 {
168     /* Check if the device is VFIO PCI device */
169     if (!object_dynamic_cast(OBJECT(pdev), "vfio-pci")) {
170         return;
171     }
172 
173     /*
174      * The MSIx table will be cleaned out by reset. We need
175      * disable it so that it can be reenabled properly. Also,
176      * the cached MSIx table should be cleared as it's not
177      * reflecting the contents in hardware.
178      */
179     if (msix_enabled(pdev)) {
180         uint16_t flags;
181 
182         flags = pci_host_config_read_common(pdev,
183                                             pdev->msix_cap + PCI_MSIX_FLAGS,
184                                             pci_config_size(pdev), 2);
185         flags &= ~PCI_MSIX_FLAGS_ENABLE;
186         pci_host_config_write_common(pdev,
187                                      pdev->msix_cap + PCI_MSIX_FLAGS,
188                                      pci_config_size(pdev), flags, 2);
189     }
190 
191     msix_reset(pdev);
192 }
193 
194 static void spapr_phb_vfio_eeh_clear_bus_msix(PCIBus *bus, void *opaque)
195 {
196        pci_for_each_device(bus, pci_bus_num(bus),
197                            spapr_phb_vfio_eeh_clear_dev_msix, NULL);
198 }
199 
200 static void spapr_phb_vfio_eeh_pre_reset(sPAPRPHBState *sphb)
201 {
202        PCIHostState *phb = PCI_HOST_BRIDGE(sphb);
203 
204        pci_for_each_bus(phb->bus, spapr_phb_vfio_eeh_clear_bus_msix, NULL);
205 }
206 
207 static int spapr_phb_vfio_eeh_reset(sPAPRPHBState *sphb, int option)
208 {
209     sPAPRPHBVFIOState *svphb = SPAPR_PCI_VFIO_HOST_BRIDGE(sphb);
210     struct vfio_eeh_pe_op op = { .argsz = sizeof(op) };
211     int ret;
212 
213     switch (option) {
214     case RTAS_SLOT_RESET_DEACTIVATE:
215         op.op = VFIO_EEH_PE_RESET_DEACTIVATE;
216         break;
217     case RTAS_SLOT_RESET_HOT:
218         spapr_phb_vfio_eeh_pre_reset(sphb);
219         op.op = VFIO_EEH_PE_RESET_HOT;
220         break;
221     case RTAS_SLOT_RESET_FUNDAMENTAL:
222         spapr_phb_vfio_eeh_pre_reset(sphb);
223         op.op = VFIO_EEH_PE_RESET_FUNDAMENTAL;
224         break;
225     default:
226         return RTAS_OUT_PARAM_ERROR;
227     }
228 
229     ret = vfio_container_ioctl(&svphb->phb.iommu_as, svphb->iommugroupid,
230                                VFIO_EEH_PE_OP, &op);
231     if (ret < 0) {
232         return RTAS_OUT_HW_ERROR;
233     }
234 
235     return RTAS_OUT_SUCCESS;
236 }
237 
238 static int spapr_phb_vfio_eeh_configure(sPAPRPHBState *sphb)
239 {
240     sPAPRPHBVFIOState *svphb = SPAPR_PCI_VFIO_HOST_BRIDGE(sphb);
241     struct vfio_eeh_pe_op op = { .argsz = sizeof(op) };
242     int ret;
243 
244     op.op = VFIO_EEH_PE_CONFIGURE;
245     ret = vfio_container_ioctl(&svphb->phb.iommu_as, svphb->iommugroupid,
246                                VFIO_EEH_PE_OP, &op);
247     if (ret < 0) {
248         return RTAS_OUT_PARAM_ERROR;
249     }
250 
251     return RTAS_OUT_SUCCESS;
252 }
253 
254 static void spapr_phb_vfio_class_init(ObjectClass *klass, void *data)
255 {
256     DeviceClass *dc = DEVICE_CLASS(klass);
257     sPAPRPHBClass *spc = SPAPR_PCI_HOST_BRIDGE_CLASS(klass);
258 
259     dc->props = spapr_phb_vfio_properties;
260     dc->reset = spapr_phb_vfio_reset;
261     spc->finish_realize = spapr_phb_vfio_finish_realize;
262     spc->eeh_set_option = spapr_phb_vfio_eeh_set_option;
263     spc->eeh_get_state = spapr_phb_vfio_eeh_get_state;
264     spc->eeh_reset = spapr_phb_vfio_eeh_reset;
265     spc->eeh_configure = spapr_phb_vfio_eeh_configure;
266 }
267 
268 static const TypeInfo spapr_phb_vfio_info = {
269     .name          = TYPE_SPAPR_PCI_VFIO_HOST_BRIDGE,
270     .parent        = TYPE_SPAPR_PCI_HOST_BRIDGE,
271     .instance_size = sizeof(sPAPRPHBVFIOState),
272     .class_init    = spapr_phb_vfio_class_init,
273     .class_size    = sizeof(sPAPRPHBClass),
274 };
275 
276 static void spapr_pci_vfio_register_types(void)
277 {
278     type_register_static(&spapr_phb_vfio_info);
279 }
280 
281 type_init(spapr_pci_vfio_register_types)
282