xref: /openbmc/qemu/hw/ppc/spapr_pci_vfio.c (revision 91bfcdb0)
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 "hw/ppc/spapr.h"
21 #include "hw/pci-host/spapr.h"
22 #include "hw/pci/msix.h"
23 #include "linux/vfio.h"
24 #include "hw/vfio/vfio.h"
25 
26 static Property spapr_phb_vfio_properties[] = {
27     DEFINE_PROP_INT32("iommu", sPAPRPHBVFIOState, iommugroupid, -1),
28     DEFINE_PROP_END_OF_LIST(),
29 };
30 
31 static void spapr_phb_vfio_finish_realize(sPAPRPHBState *sphb, Error **errp)
32 {
33     sPAPRPHBVFIOState *svphb = SPAPR_PCI_VFIO_HOST_BRIDGE(sphb);
34     struct vfio_iommu_spapr_tce_info info = { .argsz = sizeof(info) };
35     int ret;
36     sPAPRTCETable *tcet;
37     uint32_t liobn = svphb->phb.dma_liobn;
38 
39     if (svphb->iommugroupid == -1) {
40         error_setg(errp, "Wrong IOMMU group ID %d", svphb->iommugroupid);
41         return;
42     }
43 
44     ret = vfio_container_ioctl(&svphb->phb.iommu_as, svphb->iommugroupid,
45                                VFIO_CHECK_EXTENSION,
46                                (void *) VFIO_SPAPR_TCE_IOMMU);
47     if (ret != 1) {
48         error_setg_errno(errp, -ret,
49                          "spapr-vfio: SPAPR extension is not supported");
50         return;
51     }
52 
53     ret = vfio_container_ioctl(&svphb->phb.iommu_as, svphb->iommugroupid,
54                                VFIO_IOMMU_SPAPR_TCE_GET_INFO, &info);
55     if (ret) {
56         error_setg_errno(errp, -ret,
57                          "spapr-vfio: get info from container failed");
58         return;
59     }
60 
61     tcet = spapr_tce_new_table(DEVICE(sphb), liobn, info.dma32_window_start,
62                                SPAPR_TCE_PAGE_SHIFT,
63                                info.dma32_window_size >> SPAPR_TCE_PAGE_SHIFT,
64                                true);
65     if (!tcet) {
66         error_setg(errp, "spapr-vfio: failed to create VFIO TCE table");
67         return;
68     }
69 
70     /* Register default 32bit DMA window */
71     memory_region_add_subregion(&sphb->iommu_root, tcet->bus_offset,
72                                 spapr_tce_get_iommu(tcet));
73 }
74 
75 static void spapr_phb_vfio_eeh_reenable(sPAPRPHBVFIOState *svphb)
76 {
77     struct vfio_eeh_pe_op op = {
78         .argsz = sizeof(op),
79         .op    = VFIO_EEH_PE_ENABLE
80     };
81 
82     vfio_container_ioctl(&svphb->phb.iommu_as,
83                          svphb->iommugroupid, VFIO_EEH_PE_OP, &op);
84 }
85 
86 static void spapr_phb_vfio_reset(DeviceState *qdev)
87 {
88     /*
89      * The PE might be in frozen state. To reenable the EEH
90      * functionality on it will clean the frozen state, which
91      * ensures that the contained PCI devices will work properly
92      * after reboot.
93      */
94     spapr_phb_vfio_eeh_reenable(SPAPR_PCI_VFIO_HOST_BRIDGE(qdev));
95 }
96 
97 static int spapr_phb_vfio_eeh_set_option(sPAPRPHBState *sphb,
98                                          unsigned int addr, int option)
99 {
100     sPAPRPHBVFIOState *svphb = SPAPR_PCI_VFIO_HOST_BRIDGE(sphb);
101     struct vfio_eeh_pe_op op = { .argsz = sizeof(op) };
102     int ret;
103 
104     switch (option) {
105     case RTAS_EEH_DISABLE:
106         op.op = VFIO_EEH_PE_DISABLE;
107         break;
108     case RTAS_EEH_ENABLE: {
109         PCIHostState *phb;
110         PCIDevice *pdev;
111 
112         /*
113          * The EEH functionality is enabled on basis of PCI device,
114          * instead of PE. We need check the validity of the PCI
115          * device address.
116          */
117         phb = PCI_HOST_BRIDGE(sphb);
118         pdev = pci_find_device(phb->bus,
119                                (addr >> 16) & 0xFF, (addr >> 8) & 0xFF);
120         if (!pdev || !object_dynamic_cast(OBJECT(pdev), "vfio-pci")) {
121             return RTAS_OUT_PARAM_ERROR;
122         }
123 
124         op.op = VFIO_EEH_PE_ENABLE;
125         break;
126     }
127     case RTAS_EEH_THAW_IO:
128         op.op = VFIO_EEH_PE_UNFREEZE_IO;
129         break;
130     case RTAS_EEH_THAW_DMA:
131         op.op = VFIO_EEH_PE_UNFREEZE_DMA;
132         break;
133     default:
134         return RTAS_OUT_PARAM_ERROR;
135     }
136 
137     ret = vfio_container_ioctl(&svphb->phb.iommu_as, svphb->iommugroupid,
138                                VFIO_EEH_PE_OP, &op);
139     if (ret < 0) {
140         return RTAS_OUT_HW_ERROR;
141     }
142 
143     return RTAS_OUT_SUCCESS;
144 }
145 
146 static int spapr_phb_vfio_eeh_get_state(sPAPRPHBState *sphb, int *state)
147 {
148     sPAPRPHBVFIOState *svphb = SPAPR_PCI_VFIO_HOST_BRIDGE(sphb);
149     struct vfio_eeh_pe_op op = { .argsz = sizeof(op) };
150     int ret;
151 
152     op.op = VFIO_EEH_PE_GET_STATE;
153     ret = vfio_container_ioctl(&svphb->phb.iommu_as, svphb->iommugroupid,
154                                VFIO_EEH_PE_OP, &op);
155     if (ret < 0) {
156         return RTAS_OUT_PARAM_ERROR;
157     }
158 
159     *state = ret;
160     return RTAS_OUT_SUCCESS;
161 }
162 
163 static void spapr_phb_vfio_eeh_clear_dev_msix(PCIBus *bus,
164                                               PCIDevice *pdev,
165                                               void *opaque)
166 {
167     /* Check if the device is VFIO PCI device */
168     if (!object_dynamic_cast(OBJECT(pdev), "vfio-pci")) {
169         return;
170     }
171 
172     /*
173      * The MSIx table will be cleaned out by reset. We need
174      * disable it so that it can be reenabled properly. Also,
175      * the cached MSIx table should be cleared as it's not
176      * reflecting the contents in hardware.
177      */
178     if (msix_enabled(pdev)) {
179         uint16_t flags;
180 
181         flags = pci_host_config_read_common(pdev,
182                                             pdev->msix_cap + PCI_MSIX_FLAGS,
183                                             pci_config_size(pdev), 2);
184         flags &= ~PCI_MSIX_FLAGS_ENABLE;
185         pci_host_config_write_common(pdev,
186                                      pdev->msix_cap + PCI_MSIX_FLAGS,
187                                      pci_config_size(pdev), flags, 2);
188     }
189 
190     msix_reset(pdev);
191 }
192 
193 static void spapr_phb_vfio_eeh_clear_bus_msix(PCIBus *bus, void *opaque)
194 {
195        pci_for_each_device(bus, pci_bus_num(bus),
196                            spapr_phb_vfio_eeh_clear_dev_msix, NULL);
197 }
198 
199 static void spapr_phb_vfio_eeh_pre_reset(sPAPRPHBState *sphb)
200 {
201        PCIHostState *phb = PCI_HOST_BRIDGE(sphb);
202 
203        pci_for_each_bus(phb->bus, spapr_phb_vfio_eeh_clear_bus_msix, NULL);
204 }
205 
206 static int spapr_phb_vfio_eeh_reset(sPAPRPHBState *sphb, int option)
207 {
208     sPAPRPHBVFIOState *svphb = SPAPR_PCI_VFIO_HOST_BRIDGE(sphb);
209     struct vfio_eeh_pe_op op = { .argsz = sizeof(op) };
210     int ret;
211 
212     switch (option) {
213     case RTAS_SLOT_RESET_DEACTIVATE:
214         op.op = VFIO_EEH_PE_RESET_DEACTIVATE;
215         break;
216     case RTAS_SLOT_RESET_HOT:
217         spapr_phb_vfio_eeh_pre_reset(sphb);
218         op.op = VFIO_EEH_PE_RESET_HOT;
219         break;
220     case RTAS_SLOT_RESET_FUNDAMENTAL:
221         spapr_phb_vfio_eeh_pre_reset(sphb);
222         op.op = VFIO_EEH_PE_RESET_FUNDAMENTAL;
223         break;
224     default:
225         return RTAS_OUT_PARAM_ERROR;
226     }
227 
228     ret = vfio_container_ioctl(&svphb->phb.iommu_as, svphb->iommugroupid,
229                                VFIO_EEH_PE_OP, &op);
230     if (ret < 0) {
231         return RTAS_OUT_HW_ERROR;
232     }
233 
234     return RTAS_OUT_SUCCESS;
235 }
236 
237 static int spapr_phb_vfio_eeh_configure(sPAPRPHBState *sphb)
238 {
239     sPAPRPHBVFIOState *svphb = SPAPR_PCI_VFIO_HOST_BRIDGE(sphb);
240     struct vfio_eeh_pe_op op = { .argsz = sizeof(op) };
241     int ret;
242 
243     op.op = VFIO_EEH_PE_CONFIGURE;
244     ret = vfio_container_ioctl(&svphb->phb.iommu_as, svphb->iommugroupid,
245                                VFIO_EEH_PE_OP, &op);
246     if (ret < 0) {
247         return RTAS_OUT_PARAM_ERROR;
248     }
249 
250     return RTAS_OUT_SUCCESS;
251 }
252 
253 static void spapr_phb_vfio_class_init(ObjectClass *klass, void *data)
254 {
255     DeviceClass *dc = DEVICE_CLASS(klass);
256     sPAPRPHBClass *spc = SPAPR_PCI_HOST_BRIDGE_CLASS(klass);
257 
258     dc->props = spapr_phb_vfio_properties;
259     dc->reset = spapr_phb_vfio_reset;
260     spc->finish_realize = spapr_phb_vfio_finish_realize;
261     spc->eeh_set_option = spapr_phb_vfio_eeh_set_option;
262     spc->eeh_get_state = spapr_phb_vfio_eeh_get_state;
263     spc->eeh_reset = spapr_phb_vfio_eeh_reset;
264     spc->eeh_configure = spapr_phb_vfio_eeh_configure;
265 }
266 
267 static const TypeInfo spapr_phb_vfio_info = {
268     .name          = TYPE_SPAPR_PCI_VFIO_HOST_BRIDGE,
269     .parent        = TYPE_SPAPR_PCI_HOST_BRIDGE,
270     .instance_size = sizeof(sPAPRPHBVFIOState),
271     .class_init    = spapr_phb_vfio_class_init,
272     .class_size    = sizeof(sPAPRPHBClass),
273 };
274 
275 static void spapr_pci_vfio_register_types(void)
276 {
277     type_register_static(&spapr_phb_vfio_info);
278 }
279 
280 type_init(spapr_pci_vfio_register_types)
281