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