hw/s390x/s390-pci-vfio.c

   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 bool get_host_fh(S390PCIBusDevice *pbdev, struct vfio_device_info *info,
 128                         uint32_t *fh)
 129 {
 130     struct vfio_info_cap_header *hdr;
 131     struct vfio_device_info_cap_zpci_base *cap;
 132     VFIOPCIDevice *vpci = container_of(pbdev->pdev, VFIOPCIDevice, pdev);
 133
 134     hdr = vfio_get_device_info_cap(info, VFIO_DEVICE_INFO_CAP_ZPCI_BASE);
 135
 136     /* Can only get the host fh with version 2 or greater */
 137     if (hdr == NULL || hdr->version < 2) {
 138         trace_s390_pci_clp_cap(vpci->vbasedev.name,
 139                                VFIO_DEVICE_INFO_CAP_ZPCI_BASE);
 140         return false;
 141     }
 142     cap = (void *) hdr;
 143
 144     *fh = cap->fh;
 145     return true;
 146 }
 147
 148 static void s390_pci_read_group(S390PCIBusDevice *pbdev,
 149                                 struct vfio_device_info *info)
 150 {
 151     struct vfio_info_cap_header *hdr;
 152     struct vfio_device_info_cap_zpci_group *cap;
 153     S390pciState *s = s390_get_phb();
 154     ClpRspQueryPciGrp *resgrp;
 155     VFIOPCIDevice *vpci =  container_of(pbdev->pdev, VFIOPCIDevice, pdev);
 156     uint8_t start_gid = pbdev->zpci_fn.pfgid;
 157
 158     hdr = vfio_get_device_info_cap(info, VFIO_DEVICE_INFO_CAP_ZPCI_GROUP);
 159
 160     /*
 161      * If capability not provided or the underlying hostdev is simulated, just
 162      * use the default group.
 163      */
 164     if (hdr == NULL || pbdev->zpci_fn.pfgid >= ZPCI_SIM_GRP_START) {
 165         trace_s390_pci_clp_cap(vpci->vbasedev.name,
 166                                VFIO_DEVICE_INFO_CAP_ZPCI_GROUP);
 167         pbdev->zpci_fn.pfgid = ZPCI_DEFAULT_FN_GRP;
 168         pbdev->pci_group = s390_group_find(ZPCI_DEFAULT_FN_GRP);
 169         return;
 170     }
 171     cap = (void *) hdr;
 172
 173     /*
 174      * For an intercept device, let's use an existing simulated group if one
 175      * one was already created for other intercept devices in this group.
 176      * If not, create a new simulated group if any are still available.
 177      * If all else fails, just fall back on the default group.
 178      */
 179     if (!pbdev->interp) {
 180         pbdev->pci_group = s390_group_find_host_sim(pbdev->zpci_fn.pfgid);
 181         if (pbdev->pci_group) {
 182             /* Use existing simulated group */
 183             pbdev->zpci_fn.pfgid = pbdev->pci_group->id;
 184             return;
 185         } else {
 186             if (s->next_sim_grp == ZPCI_DEFAULT_FN_GRP) {
 187                 /* All out of simulated groups, use default */
 188                 trace_s390_pci_clp_cap(vpci->vbasedev.name,
 189                                        VFIO_DEVICE_INFO_CAP_ZPCI_GROUP);
 190                 pbdev->zpci_fn.pfgid = ZPCI_DEFAULT_FN_GRP;
 191                 pbdev->pci_group = s390_group_find(ZPCI_DEFAULT_FN_GRP);
 192                 return;
 193             } else {
 194                 /* We can assign a new simulated group */
 195                 pbdev->zpci_fn.pfgid = s->next_sim_grp;
 196                 s->next_sim_grp++;
 197                 /* Fall through to create the new sim group using CLP info */
 198             }
 199         }
 200     }
 201
 202     /* See if the PCI group is already defined, create if not */
 203     pbdev->pci_group = s390_group_find(pbdev->zpci_fn.pfgid);
 204
 205     if (!pbdev->pci_group) {
 206         pbdev->pci_group = s390_group_create(pbdev->zpci_fn.pfgid, start_gid);
 207
 208         resgrp = &pbdev->pci_group->zpci_group;
 209         if (cap->flags & VFIO_DEVICE_INFO_ZPCI_FLAG_REFRESH) {
 210             resgrp->fr = 1;
 211         }
 212         resgrp->dasm = cap->dasm;
 213         resgrp->msia = cap->msi_addr;
 214         resgrp->mui = cap->mui;
 215         resgrp->i = cap->noi;
 216         if (pbdev->interp && hdr->version >= 2) {
 217             resgrp->maxstbl = cap->imaxstbl;
 218         } else {
 219             resgrp->maxstbl = cap->maxstbl;
 220         }
 221         resgrp->version = cap->version;
 222         resgrp->dtsm = ZPCI_DTSM;
 223     }
 224 }
 225
 226 static void s390_pci_read_util(S390PCIBusDevice *pbdev,
 227                                struct vfio_device_info *info)
 228 {
 229     struct vfio_info_cap_header *hdr;
 230     struct vfio_device_info_cap_zpci_util *cap;
 231     VFIOPCIDevice *vpci =  container_of(pbdev->pdev, VFIOPCIDevice, pdev);
 232
 233     hdr = vfio_get_device_info_cap(info, VFIO_DEVICE_INFO_CAP_ZPCI_UTIL);
 234
 235     /* If capability not provided, just leave the defaults in place */
 236     if (hdr == NULL) {
 237         trace_s390_pci_clp_cap(vpci->vbasedev.name,
 238                                VFIO_DEVICE_INFO_CAP_ZPCI_UTIL);
 239         return;
 240     }
 241     cap = (void *) hdr;
 242
 243     if (cap->size > CLP_UTIL_STR_LEN) {
 244         trace_s390_pci_clp_cap_size(vpci->vbasedev.name, cap->size,
 245                                     VFIO_DEVICE_INFO_CAP_ZPCI_UTIL);
 246         return;
 247     }
 248
 249     pbdev->zpci_fn.flags |= CLP_RSP_QPCI_MASK_UTIL;
 250     memcpy(pbdev->zpci_fn.util_str, cap->util_str, CLP_UTIL_STR_LEN);
 251 }
 252
 253 static void s390_pci_read_pfip(S390PCIBusDevice *pbdev,
 254                                struct vfio_device_info *info)
 255 {
 256     struct vfio_info_cap_header *hdr;
 257     struct vfio_device_info_cap_zpci_pfip *cap;
 258     VFIOPCIDevice *vpci =  container_of(pbdev->pdev, VFIOPCIDevice, pdev);
 259
 260     hdr = vfio_get_device_info_cap(info, VFIO_DEVICE_INFO_CAP_ZPCI_PFIP);
 261
 262     /* If capability not provided, just leave the defaults in place */
 263     if (hdr == NULL) {
 264         trace_s390_pci_clp_cap(vpci->vbasedev.name,
 265                                VFIO_DEVICE_INFO_CAP_ZPCI_PFIP);
 266         return;
 267     }
 268     cap = (void *) hdr;
 269
 270     if (cap->size > CLP_PFIP_NR_SEGMENTS) {
 271         trace_s390_pci_clp_cap_size(vpci->vbasedev.name, cap->size,
 272                                     VFIO_DEVICE_INFO_CAP_ZPCI_PFIP);
 273         return;
 274     }
 275
 276     memcpy(pbdev->zpci_fn.pfip, cap->pfip, CLP_PFIP_NR_SEGMENTS);
 277 }
 278
 279 static struct vfio_device_info *get_device_info(S390PCIBusDevice *pbdev,
 280                                                 uint32_t argsz)
 281 {
 282     struct vfio_device_info *info = g_malloc0(argsz);
 283     VFIOPCIDevice *vfio_pci;
 284     int fd;
 285
 286     vfio_pci = container_of(pbdev->pdev, VFIOPCIDevice, pdev);
 287     fd = vfio_pci->vbasedev.fd;
 288
 289     /*
 290      * If the specified argsz is not large enough to contain all capabilities
 291      * it will be updated upon return from the ioctl.  Retry until we have
 292      * a big enough buffer to hold the entire capability chain.  On error,
 293      * just exit and rely on CLP defaults.
 294      */
 295 retry:
 296     info->argsz = argsz;
 297
 298     if (ioctl(fd, VFIO_DEVICE_GET_INFO, info)) {
 299         trace_s390_pci_clp_dev_info(vfio_pci->vbasedev.name);
 300         g_free(info);
 301         return NULL;
 302     }
 303
 304     if (info->argsz > argsz) {
 305         argsz = info->argsz;
 306         info = g_realloc(info, argsz);
 307         goto retry;
 308     }
 309
 310     return info;
 311 }
 312
 313 /*
 314  * Get the host function handle from the vfio CLP capabilities chain.  Returns
 315  * true if a fh value was placed into the provided buffer.  Returns false
 316  * if a fh could not be obtained (ioctl failed or capabilitiy version does
 317  * not include the fh)
 318  */
 319 bool s390_pci_get_host_fh(S390PCIBusDevice *pbdev, uint32_t *fh)
 320 {
 321     g_autofree struct vfio_device_info *info = NULL;
 322
 323     assert(fh);
 324
 325     info = get_device_info(pbdev, sizeof(*info));
 326     if (!info) {
 327         return false;
 328     }
 329
 330     return get_host_fh(pbdev, info, fh);
 331 }
 332
 333 /*
 334  * This function will issue the VFIO_DEVICE_GET_INFO ioctl and look for
 335  * capabilities that contain information about CLP features provided by the
 336  * underlying host.
 337  * On entry, defaults have already been placed into the guest CLP response
 338  * buffers.  On exit, defaults will have been overwritten for any CLP features
 339  * found in the capability chain; defaults will remain for any CLP features not
 340  * found in the chain.
 341  */
 342 void s390_pci_get_clp_info(S390PCIBusDevice *pbdev)
 343 {
 344     g_autofree struct vfio_device_info *info = NULL;
 345
 346     info = get_device_info(pbdev, sizeof(*info));
 347     if (!info) {
 348         return;
 349     }
 350
 351     /*
 352      * Find the CLP features provided and fill in the guest CLP responses.
 353      * Always call s390_pci_read_base first as information from this could
 354      * determine which function group is used in s390_pci_read_group.
 355      * For any feature not found, the default values will remain in the CLP
 356      * response.
 357      */
 358     s390_pci_read_base(pbdev, info);
 359     s390_pci_read_group(pbdev, info);
 360     s390_pci_read_util(pbdev, info);
 361     s390_pci_read_pfip(pbdev, info);
 362
 363     return;
 364 }