xref: /openbmc/qemu/hw/s390x/s390-pci-vfio.c (revision 9f54dc1c) 
1 /*
2  * s390 vfio-pci interfaces
3  *
4  * Copyright 2020 IBM Corp.
5  * Author(s): Matthew Rosato <mjrosato@linux.ibm.com>
6  *
7  * This work is licensed under the terms of the GNU GPL, version 2 or (at
8  * your option) any later version. See the COPYING file in the top-level
9  * directory.
10  */
11 
12 #include "qemu/osdep.h"
13 
14 #include <sys/ioctl.h>
15 #include <linux/vfio.h>
16 #include <linux/vfio_zdev.h>
17 
18 #include "trace.h"
19 #include "hw/s390x/s390-pci-bus.h"
20 #include "hw/s390x/s390-pci-clp.h"
21 #include "hw/s390x/s390-pci-vfio.h"
22 #include "hw/vfio/pci.h"
23 #include "hw/vfio/vfio-common.h"
24 
25 /*
26  * Get the current DMA available count from vfio.  Returns true if vfio is
27  * limiting DMA requests, false otherwise.  The current available count read
28  * from vfio is returned in avail.
29  */
30 bool s390_pci_update_dma_avail(int fd, unsigned int *avail)
31 {
32     uint32_t argsz = sizeof(struct vfio_iommu_type1_info);
33     g_autofree struct vfio_iommu_type1_info *info = g_malloc0(argsz);
34 
35     assert(avail);
36 
37     /*
38      * If the specified argsz is not large enough to contain all capabilities
39      * it will be updated upon return from the ioctl.  Retry until we have
40      * a big enough buffer to hold the entire capability chain.
41      */
42 retry:
43     info->argsz = argsz;
44 
45     if (ioctl(fd, VFIO_IOMMU_GET_INFO, info)) {
46         return false;
47     }
48 
49     if (info->argsz > argsz) {
50         argsz = info->argsz;
51         info = g_realloc(info, argsz);
52         goto retry;
53     }
54 
55     /* If the capability exists, update with the current value */
56     return vfio_get_info_dma_avail(info, avail);
57 }
58 
59 S390PCIDMACount *s390_pci_start_dma_count(S390pciState *s,
60                                           S390PCIBusDevice *pbdev)
61 {
62     S390PCIDMACount *cnt;
63     uint32_t avail;
64     VFIOPCIDevice *vpdev = container_of(pbdev->pdev, VFIOPCIDevice, pdev);
65     int id;
66 
67     assert(vpdev);
68 
69     id = vpdev->vbasedev.group->container->fd;
70 
71     if (!s390_pci_update_dma_avail(id, &avail)) {
72         return NULL;
73     }
74 
75     QTAILQ_FOREACH(cnt, &s->zpci_dma_limit, link) {
76         if (cnt->id  == id) {
77             cnt->users++;
78             return cnt;
79         }
80     }
81 
82     cnt = g_new0(S390PCIDMACount, 1);
83     cnt->id = id;
84     cnt->users = 1;
85     cnt->avail = avail;
86     QTAILQ_INSERT_TAIL(&s->zpci_dma_limit, cnt, link);
87     return cnt;
88 }
89 
90 void s390_pci_end_dma_count(S390pciState *s, S390PCIDMACount *cnt)
91 {
92     assert(cnt);
93 
94     cnt->users--;
95     if (cnt->users == 0) {
96         QTAILQ_REMOVE(&s->zpci_dma_limit, cnt, link);
97     }
98 }
99 
100 static void s390_pci_read_base(S390PCIBusDevice *pbdev,
101                                struct vfio_device_info *info)
102 {
103     struct vfio_info_cap_header *hdr;
104     struct vfio_device_info_cap_zpci_base *cap;
105     VFIOPCIDevice *vpci =  container_of(pbdev->pdev, VFIOPCIDevice, pdev);
106 
107     hdr = vfio_get_device_info_cap(info, VFIO_DEVICE_INFO_CAP_ZPCI_BASE);
108 
109     /* If capability not provided, just leave the defaults in place */
110     if (hdr == NULL) {
111         trace_s390_pci_clp_cap(vpci->vbasedev.name,
112                                VFIO_DEVICE_INFO_CAP_ZPCI_BASE);
113         return;
114     }
115     cap = (void *) hdr;
116 
117     pbdev->zpci_fn.sdma = cap->start_dma;
118     pbdev->zpci_fn.edma = cap->end_dma;
119     pbdev->zpci_fn.pchid = cap->pchid;
120     pbdev->zpci_fn.vfn = cap->vfn;
121     pbdev->zpci_fn.pfgid = cap->gid;
122     /* The following values remain 0 until we support other FMB formats */
123     pbdev->zpci_fn.fmbl = 0;
124     pbdev->zpci_fn.pft = 0;
125 }
126 
127 static void s390_pci_read_group(S390PCIBusDevice *pbdev,
128                                 struct vfio_device_info *info)
129 {
130     struct vfio_info_cap_header *hdr;
131     struct vfio_device_info_cap_zpci_group *cap;
132     ClpRspQueryPciGrp *resgrp;
133     VFIOPCIDevice *vpci =  container_of(pbdev->pdev, VFIOPCIDevice, pdev);
134 
135     hdr = vfio_get_device_info_cap(info, VFIO_DEVICE_INFO_CAP_ZPCI_GROUP);
136 
137     /* If capability not provided, just use the default group */
138     if (hdr == NULL) {
139         trace_s390_pci_clp_cap(vpci->vbasedev.name,
140                                VFIO_DEVICE_INFO_CAP_ZPCI_GROUP);
141         pbdev->zpci_fn.pfgid = ZPCI_DEFAULT_FN_GRP;
142         pbdev->pci_group = s390_group_find(ZPCI_DEFAULT_FN_GRP);
143         return;
144     }
145     cap = (void *) hdr;
146 
147     /* See if the PCI group is already defined, create if not */
148     pbdev->pci_group = s390_group_find(pbdev->zpci_fn.pfgid);
149 
150     if (!pbdev->pci_group) {
151         pbdev->pci_group = s390_group_create(pbdev->zpci_fn.pfgid);
152 
153         resgrp = &pbdev->pci_group->zpci_group;
154         if (cap->flags & VFIO_DEVICE_INFO_ZPCI_FLAG_REFRESH) {
155             resgrp->fr = 1;
156         }
157         resgrp->dasm = cap->dasm;
158         resgrp->msia = cap->msi_addr;
159         resgrp->mui = cap->mui;
160         resgrp->i = cap->noi;
161         resgrp->maxstbl = cap->maxstbl;
162         resgrp->version = cap->version;
163         resgrp->dtsm = ZPCI_DTSM;
164     }
165 }
166 
167 static void s390_pci_read_util(S390PCIBusDevice *pbdev,
168                                struct vfio_device_info *info)
169 {
170     struct vfio_info_cap_header *hdr;
171     struct vfio_device_info_cap_zpci_util *cap;
172     VFIOPCIDevice *vpci =  container_of(pbdev->pdev, VFIOPCIDevice, pdev);
173 
174     hdr = vfio_get_device_info_cap(info, VFIO_DEVICE_INFO_CAP_ZPCI_UTIL);
175 
176     /* If capability not provided, just leave the defaults in place */
177     if (hdr == NULL) {
178         trace_s390_pci_clp_cap(vpci->vbasedev.name,
179                                VFIO_DEVICE_INFO_CAP_ZPCI_UTIL);
180         return;
181     }
182     cap = (void *) hdr;
183 
184     if (cap->size > CLP_UTIL_STR_LEN) {
185         trace_s390_pci_clp_cap_size(vpci->vbasedev.name, cap->size,
186                                     VFIO_DEVICE_INFO_CAP_ZPCI_UTIL);
187         return;
188     }
189 
190     pbdev->zpci_fn.flags |= CLP_RSP_QPCI_MASK_UTIL;
191     memcpy(pbdev->zpci_fn.util_str, cap->util_str, CLP_UTIL_STR_LEN);
192 }
193 
194 static void s390_pci_read_pfip(S390PCIBusDevice *pbdev,
195                                struct vfio_device_info *info)
196 {
197     struct vfio_info_cap_header *hdr;
198     struct vfio_device_info_cap_zpci_pfip *cap;
199     VFIOPCIDevice *vpci =  container_of(pbdev->pdev, VFIOPCIDevice, pdev);
200 
201     hdr = vfio_get_device_info_cap(info, VFIO_DEVICE_INFO_CAP_ZPCI_PFIP);
202 
203     /* If capability not provided, just leave the defaults in place */
204     if (hdr == NULL) {
205         trace_s390_pci_clp_cap(vpci->vbasedev.name,
206                                VFIO_DEVICE_INFO_CAP_ZPCI_PFIP);
207         return;
208     }
209     cap = (void *) hdr;
210 
211     if (cap->size > CLP_PFIP_NR_SEGMENTS) {
212         trace_s390_pci_clp_cap_size(vpci->vbasedev.name, cap->size,
213                                     VFIO_DEVICE_INFO_CAP_ZPCI_PFIP);
214         return;
215     }
216 
217     memcpy(pbdev->zpci_fn.pfip, cap->pfip, CLP_PFIP_NR_SEGMENTS);
218 }
219 
220 /*
221  * This function will issue the VFIO_DEVICE_GET_INFO ioctl and look for
222  * capabilities that contain information about CLP features provided by the
223  * underlying host.
224  * On entry, defaults have already been placed into the guest CLP response
225  * buffers.  On exit, defaults will have been overwritten for any CLP features
226  * found in the capability chain; defaults will remain for any CLP features not
227  * found in the chain.
228  */
229 void s390_pci_get_clp_info(S390PCIBusDevice *pbdev)
230 {
231     g_autofree struct vfio_device_info *info = NULL;
232     VFIOPCIDevice *vfio_pci;
233     uint32_t argsz;
234     int fd;
235 
236     argsz = sizeof(*info);
237     info = g_malloc0(argsz);
238 
239     vfio_pci = container_of(pbdev->pdev, VFIOPCIDevice, pdev);
240     fd = vfio_pci->vbasedev.fd;
241 
242     /*
243      * If the specified argsz is not large enough to contain all capabilities
244      * it will be updated upon return from the ioctl.  Retry until we have
245      * a big enough buffer to hold the entire capability chain.  On error,
246      * just exit and rely on CLP defaults.
247      */
248 retry:
249     info->argsz = argsz;
250 
251     if (ioctl(fd, VFIO_DEVICE_GET_INFO, info)) {
252         trace_s390_pci_clp_dev_info(vfio_pci->vbasedev.name);
253         return;
254     }
255 
256     if (info->argsz > argsz) {
257         argsz = info->argsz;
258         info = g_realloc(info, argsz);
259         goto retry;
260     }
261 
262     /*
263      * Find the CLP features provided and fill in the guest CLP responses.
264      * Always call s390_pci_read_base first as information from this could
265      * determine which function group is used in s390_pci_read_group.
266      * For any feature not found, the default values will remain in the CLP
267      * response.
268      */
269     s390_pci_read_base(pbdev, info);
270     s390_pci_read_group(pbdev, info);
271     s390_pci_read_util(pbdev, info);
272     s390_pci_read_pfip(pbdev, info);
273 
274     return;
275 }
276