xref: /openbmc/qemu/hw/s390x/s390-pci-vfio.c (revision e9206163)
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     if (!vpdev->vbasedev.group) {
70         return NULL;
71     }
72 
73     id = vpdev->vbasedev.group->container->fd;
74 
75     if (!s390_pci_update_dma_avail(id, &avail)) {
76         return NULL;
77     }
78 
79     QTAILQ_FOREACH(cnt, &s->zpci_dma_limit, link) {
80         if (cnt->id  == id) {
81             cnt->users++;
82             return cnt;
83         }
84     }
85 
86     cnt = g_new0(S390PCIDMACount, 1);
87     cnt->id = id;
88     cnt->users = 1;
89     cnt->avail = avail;
90     QTAILQ_INSERT_TAIL(&s->zpci_dma_limit, cnt, link);
91     pbdev->iommu->max_dma_limit = avail;
92     return cnt;
93 }
94 
95 void s390_pci_end_dma_count(S390pciState *s, S390PCIDMACount *cnt)
96 {
97     assert(cnt);
98 
99     cnt->users--;
100     if (cnt->users == 0) {
101         QTAILQ_REMOVE(&s->zpci_dma_limit, cnt, link);
102     }
103 }
104 
105 static void s390_pci_read_base(S390PCIBusDevice *pbdev,
106                                struct vfio_device_info *info)
107 {
108     struct vfio_info_cap_header *hdr;
109     struct vfio_device_info_cap_zpci_base *cap;
110     VFIOPCIDevice *vpci =  container_of(pbdev->pdev, VFIOPCIDevice, pdev);
111     uint64_t vfio_size;
112 
113     hdr = vfio_get_device_info_cap(info, VFIO_DEVICE_INFO_CAP_ZPCI_BASE);
114 
115     /* If capability not provided, just leave the defaults in place */
116     if (hdr == NULL) {
117         trace_s390_pci_clp_cap(vpci->vbasedev.name,
118                                VFIO_DEVICE_INFO_CAP_ZPCI_BASE);
119         return;
120     }
121     cap = (void *) hdr;
122 
123     pbdev->zpci_fn.sdma = cap->start_dma;
124     pbdev->zpci_fn.edma = cap->end_dma;
125     pbdev->zpci_fn.pchid = cap->pchid;
126     pbdev->zpci_fn.vfn = cap->vfn;
127     pbdev->zpci_fn.pfgid = cap->gid;
128     /* The following values remain 0 until we support other FMB formats */
129     pbdev->zpci_fn.fmbl = 0;
130     pbdev->zpci_fn.pft = 0;
131     /* Store function type separately for type-specific behavior */
132     pbdev->pft = cap->pft;
133 
134     /*
135      * If appropriate, reduce the size of the supported DMA aperture reported
136      * to the guest based upon the vfio DMA limit.
137      */
138     vfio_size = pbdev->iommu->max_dma_limit << TARGET_PAGE_BITS;
139     if (vfio_size > 0 && vfio_size < cap->end_dma - cap->start_dma + 1) {
140         pbdev->zpci_fn.edma = cap->start_dma + vfio_size - 1;
141     }
142 }
143 
144 static bool get_host_fh(S390PCIBusDevice *pbdev, struct vfio_device_info *info,
145                         uint32_t *fh)
146 {
147     struct vfio_info_cap_header *hdr;
148     struct vfio_device_info_cap_zpci_base *cap;
149     VFIOPCIDevice *vpci = container_of(pbdev->pdev, VFIOPCIDevice, pdev);
150 
151     hdr = vfio_get_device_info_cap(info, VFIO_DEVICE_INFO_CAP_ZPCI_BASE);
152 
153     /* Can only get the host fh with version 2 or greater */
154     if (hdr == NULL || hdr->version < 2) {
155         trace_s390_pci_clp_cap(vpci->vbasedev.name,
156                                VFIO_DEVICE_INFO_CAP_ZPCI_BASE);
157         return false;
158     }
159     cap = (void *) hdr;
160 
161     *fh = cap->fh;
162     return true;
163 }
164 
165 static void s390_pci_read_group(S390PCIBusDevice *pbdev,
166                                 struct vfio_device_info *info)
167 {
168     struct vfio_info_cap_header *hdr;
169     struct vfio_device_info_cap_zpci_group *cap;
170     S390pciState *s = s390_get_phb();
171     ClpRspQueryPciGrp *resgrp;
172     VFIOPCIDevice *vpci =  container_of(pbdev->pdev, VFIOPCIDevice, pdev);
173     uint8_t start_gid = pbdev->zpci_fn.pfgid;
174 
175     hdr = vfio_get_device_info_cap(info, VFIO_DEVICE_INFO_CAP_ZPCI_GROUP);
176 
177     /*
178      * If capability not provided or the underlying hostdev is simulated, just
179      * use the default group.
180      */
181     if (hdr == NULL || pbdev->zpci_fn.pfgid >= ZPCI_SIM_GRP_START) {
182         trace_s390_pci_clp_cap(vpci->vbasedev.name,
183                                VFIO_DEVICE_INFO_CAP_ZPCI_GROUP);
184         pbdev->zpci_fn.pfgid = ZPCI_DEFAULT_FN_GRP;
185         pbdev->pci_group = s390_group_find(ZPCI_DEFAULT_FN_GRP);
186         return;
187     }
188     cap = (void *) hdr;
189 
190     /*
191      * For an intercept device, let's use an existing simulated group if one
192      * one was already created for other intercept devices in this group.
193      * If not, create a new simulated group if any are still available.
194      * If all else fails, just fall back on the default group.
195      */
196     if (!pbdev->interp) {
197         pbdev->pci_group = s390_group_find_host_sim(pbdev->zpci_fn.pfgid);
198         if (pbdev->pci_group) {
199             /* Use existing simulated group */
200             pbdev->zpci_fn.pfgid = pbdev->pci_group->id;
201             return;
202         } else {
203             if (s->next_sim_grp == ZPCI_DEFAULT_FN_GRP) {
204                 /* All out of simulated groups, use default */
205                 trace_s390_pci_clp_cap(vpci->vbasedev.name,
206                                        VFIO_DEVICE_INFO_CAP_ZPCI_GROUP);
207                 pbdev->zpci_fn.pfgid = ZPCI_DEFAULT_FN_GRP;
208                 pbdev->pci_group = s390_group_find(ZPCI_DEFAULT_FN_GRP);
209                 return;
210             } else {
211                 /* We can assign a new simulated group */
212                 pbdev->zpci_fn.pfgid = s->next_sim_grp;
213                 s->next_sim_grp++;
214                 /* Fall through to create the new sim group using CLP info */
215             }
216         }
217     }
218 
219     /* See if the PCI group is already defined, create if not */
220     pbdev->pci_group = s390_group_find(pbdev->zpci_fn.pfgid);
221 
222     if (!pbdev->pci_group) {
223         pbdev->pci_group = s390_group_create(pbdev->zpci_fn.pfgid, start_gid);
224 
225         resgrp = &pbdev->pci_group->zpci_group;
226         if (cap->flags & VFIO_DEVICE_INFO_ZPCI_FLAG_REFRESH) {
227             resgrp->fr = 1;
228         }
229         resgrp->dasm = cap->dasm;
230         resgrp->msia = cap->msi_addr;
231         resgrp->mui = cap->mui;
232         resgrp->i = cap->noi;
233         if (pbdev->interp && hdr->version >= 2) {
234             resgrp->maxstbl = cap->imaxstbl;
235         } else {
236             resgrp->maxstbl = cap->maxstbl;
237         }
238         resgrp->version = cap->version;
239         resgrp->dtsm = ZPCI_DTSM;
240     }
241 }
242 
243 static void s390_pci_read_util(S390PCIBusDevice *pbdev,
244                                struct vfio_device_info *info)
245 {
246     struct vfio_info_cap_header *hdr;
247     struct vfio_device_info_cap_zpci_util *cap;
248     VFIOPCIDevice *vpci =  container_of(pbdev->pdev, VFIOPCIDevice, pdev);
249 
250     hdr = vfio_get_device_info_cap(info, VFIO_DEVICE_INFO_CAP_ZPCI_UTIL);
251 
252     /* If capability not provided, just leave the defaults in place */
253     if (hdr == NULL) {
254         trace_s390_pci_clp_cap(vpci->vbasedev.name,
255                                VFIO_DEVICE_INFO_CAP_ZPCI_UTIL);
256         return;
257     }
258     cap = (void *) hdr;
259 
260     if (cap->size > CLP_UTIL_STR_LEN) {
261         trace_s390_pci_clp_cap_size(vpci->vbasedev.name, cap->size,
262                                     VFIO_DEVICE_INFO_CAP_ZPCI_UTIL);
263         return;
264     }
265 
266     pbdev->zpci_fn.flags |= CLP_RSP_QPCI_MASK_UTIL;
267     memcpy(pbdev->zpci_fn.util_str, cap->util_str, CLP_UTIL_STR_LEN);
268 }
269 
270 static void s390_pci_read_pfip(S390PCIBusDevice *pbdev,
271                                struct vfio_device_info *info)
272 {
273     struct vfio_info_cap_header *hdr;
274     struct vfio_device_info_cap_zpci_pfip *cap;
275     VFIOPCIDevice *vpci =  container_of(pbdev->pdev, VFIOPCIDevice, pdev);
276 
277     hdr = vfio_get_device_info_cap(info, VFIO_DEVICE_INFO_CAP_ZPCI_PFIP);
278 
279     /* If capability not provided, just leave the defaults in place */
280     if (hdr == NULL) {
281         trace_s390_pci_clp_cap(vpci->vbasedev.name,
282                                VFIO_DEVICE_INFO_CAP_ZPCI_PFIP);
283         return;
284     }
285     cap = (void *) hdr;
286 
287     if (cap->size > CLP_PFIP_NR_SEGMENTS) {
288         trace_s390_pci_clp_cap_size(vpci->vbasedev.name, cap->size,
289                                     VFIO_DEVICE_INFO_CAP_ZPCI_PFIP);
290         return;
291     }
292 
293     memcpy(pbdev->zpci_fn.pfip, cap->pfip, CLP_PFIP_NR_SEGMENTS);
294 }
295 
296 static struct vfio_device_info *get_device_info(S390PCIBusDevice *pbdev)
297 {
298     VFIOPCIDevice *vfio_pci = container_of(pbdev->pdev, VFIOPCIDevice, pdev);
299 
300     return vfio_get_device_info(vfio_pci->vbasedev.fd);
301 }
302 
303 /*
304  * Get the host function handle from the vfio CLP capabilities chain.  Returns
305  * true if a fh value was placed into the provided buffer.  Returns false
306  * if a fh could not be obtained (ioctl failed or capability version does
307  * not include the fh)
308  */
309 bool s390_pci_get_host_fh(S390PCIBusDevice *pbdev, uint32_t *fh)
310 {
311     g_autofree struct vfio_device_info *info = NULL;
312 
313     assert(fh);
314 
315     info = get_device_info(pbdev);
316     if (!info) {
317         return false;
318     }
319 
320     return get_host_fh(pbdev, info, fh);
321 }
322 
323 /*
324  * This function will issue the VFIO_DEVICE_GET_INFO ioctl and look for
325  * capabilities that contain information about CLP features provided by the
326  * underlying host.
327  * On entry, defaults have already been placed into the guest CLP response
328  * buffers.  On exit, defaults will have been overwritten for any CLP features
329  * found in the capability chain; defaults will remain for any CLP features not
330  * found in the chain.
331  */
332 void s390_pci_get_clp_info(S390PCIBusDevice *pbdev)
333 {
334     g_autofree struct vfio_device_info *info = NULL;
335 
336     info = get_device_info(pbdev);
337     if (!info) {
338         return;
339     }
340 
341     /*
342      * Find the CLP features provided and fill in the guest CLP responses.
343      * Always call s390_pci_read_base first as information from this could
344      * determine which function group is used in s390_pci_read_group.
345      * For any feature not found, the default values will remain in the CLP
346      * response.
347      */
348     s390_pci_read_base(pbdev, info);
349     s390_pci_read_group(pbdev, info);
350     s390_pci_read_util(pbdev, info);
351     s390_pci_read_pfip(pbdev, info);
352 
353     return;
354 }
355